APPLICATION OF EARTH OBSERVATION AND RELATED TECHNOLOGY IN AGRO - HYDROLOGICAL MODELING By Matthew R yan Herman A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Biosystems Engi neering Doctor of Philosophy 201 8 ABSTRACT APPLICATION OF EARTH OBSERVATION AND RELATED TECHNOLOGY IN AGRO - HYDROLOGICAL MODELING By Matthew R yan Herman Freshwater is vital for life on Earth, and as the human population continues to grow so do e s th e demand for this limited resource. However, anthropogenic activities and climate change will continue to alter freshwater systems. Therefore, there is a need to understand how the hydrological cycle is changing across the landscape. Traditionally, th is ha s been done by single point monitoring stations; however, these stations do not have the spatial variability to capture different aspects of the hydrologic cycle required for detailed analysis. Therefore, hydrological models are traditionally calibrat ed an d validated against a single or a few monitoring stations. One solution to this issue is the incorporation of remote sensing data. However, the proper use of the se products ha s not been well documented in hydrological models. Furthermore, with a wide varie ty of different remote sensing datasets , it is challenging to know which datasets/products should be used when. To address these knowledge gaps , three studies we re conducted . The first study was performed to examine whether the incorporation of remote ly se nsed and spatially distributed datasets can improve the overall model performance. In this study, the applicability of two remote sensing actual evapotranspirati on (ETa) products (the Simplified Surface Energy Balance (SSEBop) and the Atmosphere - Land Excha nge Inverse (ALEXI)) were examined to improve the performance of a hydrologic model using two different calibration techniques (genetic algorithm and multi - varia ble). Results from this study showed that the inclusion of ETa remote sensing data along w ith t he multi - variable calibration technique c ould improve the overall performance of a hydrological model . The second study evaluate d the spatial and temporal perfor mance of eight ETa remote sensing products in a region that lacks observed data. The remot ely s ensed datasets were further compared with ETa results from a physically - based hydrologic model to examine the differences and describe discrepancy among them. All of these datasets were compared through t he use of the Generalized Least - Square estimati on wi th Autoregressive models that compared the ETa datasets on temporal (i.e., monthly and seasonal basis) and spatial (i.e., landuse ) scales at both watershed and subbasin levels. Results showed a lack of pa tterns among the datasets when evaluating the m onthl y ETa variations; however, the seasonal aggregated data presented a better pattern and fewer variances , and statistical difference at the 0.05 level during spring and summer compare d to fall and winter mo nths. Meanwhile, spatial analysis of the datase ts sh owed that the MOD16A2 500 m ETa product was the most versatile of the tested datasets, being able to differentiate between landuses during all seasons. Finally, the ETa output of the model was found to be similar to several of the ETa products (MOD16A 2 1 k m, NLDAS - 2: Noah, and NLDAS - 2: VIC) . The third study built upon the first study by expanding the use of remotely sensed E T a products from two to eight while examining a new calibration technique , which was the many - objective optimization . The results of th is analysis show that the multi - objective calibration still resulted in better performing models compared to the many - objective calibration. Furthermore, the ensemble of all of the ETa products produced the best performing model considering both strea mflow and evapotranspiration . Copyright by MATTHEW R. HERMAN 201 8 v This thesis is dedicated to my family for all the love and support they have given me. vi ACKNOWLEDGMENTS I would like to thank my major advisor Dr. Pouyan Nejadhashemi f o r bein best advisor by always being there to mentor and guide me on my path through graduate school. I am eternally grateful that you encouraged me to attend graduate school, and I know I could not have accomplished all I have without your su p p ort. You will forever be my role model and friend. I would also like to thank my committee members: Dr. Timothy Harrigan, Dr. Joseph Messina, and Dr. Amor Ines, for their support and guidance throughout my research. I would also like to thank Barb, Jamie L ynn, and Emily for not only helping me with all of the paperwork needed to navigate the administrative side of my degree but for also making the Biosystems Department feel like a family . I am truly grateful for all you have done! I would like to thank m y frien ds and lab mates; Sebastian Hernandez - Suarez and Ian Kroop, for without their incredible assistance this dissertation would have never have gotten as far as it has. Your assistance has been a blessing! In addition , I would also like to thank the res t of my friends and lab mates: Melissa Rojas - Downing, Fariborz Daneshvar , Umesh Adhikari , Babak Saravi , Sean Woznicki, Mohammad Abouali, Irwin Donis - Gonzalez, Ray Chen, Mahlet Garedew, Subhasis Giri, and Georgina Sanchez for all of the laughs, bar trivia , g ame ni ghts, and BBQs . You have made this whole journey an adventure with stories that will last a lifetime! Finally, I would like to e specially thank my family . To my parents, Mark and Christine, for their constant encourag ement throughout my graduate st u dies a nd for being there no mater the time. To my brothers, Michael and James, for being steadfast companions in both the hard and fun times and helping me find reasons to laugh every day. Thank you , my family, for all the l ove you have given me. vii TABLE O F CONT ENTS LIST OF TABLES ................................ ................................ ................................ .......................... x LIST OF FIGURES ................................ ................................ ................................ .................... xvii KEY TO ABBREVI ATIONS ................................ ................................ ................................ ...... xix 1. INTRODUCTION ................................ ................................ ................................ ................... 1 2. LITERATURE REVIEW ................................ ................................ ................................ ........ 4 2.1 Overvie w ................................ ................................ ................................ ............................... 4 2.2 Remote Sensing ................................ ................................ ................................ ..................... 4 2.2.1 T ypes of Remote Sensing Instruments ................................ ................................ ........... 6 2.2.2 Current Remote Sensing Projects ................................ ................................ ................... 8 2.3 The Hydrologic Cycle ................................ ................................ ................................ ......... 20 2.3.1 Evapotranspiration ................................ ................................ ................................ ........ 21 2.3.2 Groundwater ................................ ................................ ................................ ................. 21 2.3.3 Oceans ................................ ................................ ................................ .......................... 22 2.3.4 Precipitation ................................ ................................ ................................ .................. 22 2.3.5 Snow and Ice ................................ ................................ ................................ ................ 23 2.3.6 Soil Moisture ................................ ................................ ................................ ................ 24 2.3.7 Surface Water ................................ ................................ ................................ ............... 24 2.3.8 Water Vapor ................................ ................................ ................................ ................. 24 2.4 Monitoring Water Resources ................................ ................................ .............................. 25 2.4.1 MOD16 ................................ ................................ ................................ ......................... 26 2.4.2 ALEXI ................................ ................................ ................................ .......................... 28 2.4.3 SSEBop ................................ ................................ ................................ ......................... 29 2.5 Hydrological Modeling ................................ ................................ ................................ ....... 30 2.5.1 Soil and Water Assessment Tool ................................ ................................ .................. 31 2.5.2 Model Calibration ................................ ................................ ................................ ......... 43 2.5.3 Remote Sensing i n Hydro logical Modeling ................................ ................................ . 44 2.6 Modeling Uncertainty ................................ ................................ ................................ ......... 46 2.6.1 Data Uncertainty ................................ ................................ ................................ ........... 46 2.6.2 Model Structure Uncertainty ................................ ................................ ........................ 47 2.6.3 Parameter Uncertainty ................................ ................................ ................................ .. 48 2.7 Summary ................................ ................................ ................................ ............................. 49 3. INTRODUCTION TO METHODOLOGY AND RESULTS ................................ ............... 50 4. EVALUATING THE ROLE OF EVAPOTRANSPIRATION REMOTE SENSING DATA IN IMPROVING HYDROLOGICAL MODELING PREDICTA BILITY ................................ .. 53 4.2 Introduction ................................ ................................ ................................ ......................... 53 4.3 Materials and Methods ................................ ................................ ................................ ........ 55 4.3.1 Study Area ................................ ................................ ................................ .................... 55 viii 4.3.2 Data Collection ................................ ................................ ................................ ............. 56 4.3.3 Hydrological Model: SWAT ................................ ................................ ........................ 58 4.3.4 Calibration Appro aches ................................ ................................ ................................ 59 4.3.5 Statistical Analysis ................................ ................................ ................................ ....... 67 4.4 Results and Discussion ................................ ................................ ................................ ........ 67 4.4 .1 Initial Streamflow Calibration ................................ ................................ ...................... 67 4.4.2 Multi - variable Calibration ................................ ................................ ............................ 69 4.4.3 Genetic Algorithm Calibration ................................ ................................ ..................... 72 4.4.4 Statistical Sig nificance ................................ ................................ ................................ . 73 4.4.5 Comparison of the Multi - variable and Genetic Algorithm Calibrat ions ...................... 77 4.5 Conclusions ................................ ................................ ................................ ......................... 77 4.6 Acknowledgment ................................ ................................ ................................ ................ 78 5. EVALUATING THE SPATIAL AND TEMPO RAL VARIABILITY OF REMOTE SENSING AND HYDROLOGIC MODEL EVAPOTRANSPIRATION PRODUCTS ............. 80 5.1 Introduction ................................ ................................ ................................ ......................... 80 5.2 Materials and Methods ................................ ................................ ................................ ........ 82 5.2.1 Study Area ................................ ................................ ................................ .................... 82 5.2.2 Remote Sensing Evapotranspiration Products ................................ .............................. 86 5.2.3 Hydrological Model ................................ ................................ ................................ ...... 90 5.2.4 Remotely Sensed Actual Evapotranspiration Data Source and Conversion Procedure 92 5.2.5 Statistical Analysis ................................ ................................ ................................ ....... 93 5.3 Results and Discussion ................................ ................................ ................................ ........ 95 5.3.1 Temporal Statistical Analysis ................................ ................................ ....................... 95 5.3.2 Spatial Statistical Analysis ................................ ................................ ......................... 104 5.3.4 Subbasin - level Statistical Analysis ................................ ................................ ............. 116 5.4 Conclusions ................................ ................................ ................................ ....................... 120 5.5 Acknowledgment ................................ ................................ ................................ .............. 122 6. EVALUATION OF MULTI AND MANY - OBJECTIVE OPTIM IZATION TECHNIQUES TO IMPROVE THE PERFORMANCE OF A HYDROLOGIC MODEL USING EVAPOTRANSPIRATION REMOTE SENSING DATA ................................ ........................ 123 6 .1 Introduction ................................ ................................ ................................ ....................... 123 6.2 Metho dology ................................ ................................ ................................ ..................... 126 6.2.1 Study Area ................................ ................................ ................................ .................. 126 6.2.2 Hydrological Model ................................ ................................ ................................ .... 127 6.2 .4 Remote Sensing Actual Evapotranspiration Products ................................ ................ 129 6.2.5 Calibration Techniques ................................ ................................ ............................... 132 6.3 Results and Discussio n ................................ ................................ ................................ ...... 141 6.3.1 Evaluation of the Performance of the Different Multi - objective Calibrations ........... 141 6.3.2 Evaluation of the Performance of the Many - O bjective Calibration Technique ......... 149 6.3.3 Impact of Landuse Inputs on Remote Sensing Evapotranspiration Product Calibration Performance ................................ ................................ ................................ ......................... 155 6.4 Conclusions ................................ ................................ ................................ ....................... 160 6.5 Acknowledgment ................................ ................................ ................................ .............. 161 ix 7. CONCLUSIONS ................................ ................................ ................................ ................. 162 8. FUTURE RESEARCH RECOMMENDATIONS ................................ .............................. 165 APPENDIX ................................ ................................ ................................ ................................ . 167 REFERENCES ................................ ................................ ................................ ........................... 235 x LIST OF T ABL ES Table 2.1. List of datasets used to calculate MOD16 ET ................................ ............................. 27 Table 2.2. List of datasets used to calculate ALEXI ET ................................ ............................... 29 Table 2.3. List of datasets used to calculate SSEBop ET ................................ ............................. 30 Table 2.4. A list of the parameters used in SWAT surface runoff calculations ........................... 34 Table 2.5. A list of the parameters used in SWA T evapotranspiration calculations .................... 37 Table 2.6. A list of the parameters use d in SWAT soil water calculations ................................ .. 40 Table 2.7. A list of the parameters used in SWAT groundwater calculations .............................. 43 Table 4.1. S treamflow calibration parameters used in this study ................................ ................. 61 Table 4.2. Calibration and validation criteria ................................ ................................ ............... 68 Table 4.3. Statistica l criteria ETa when the results from base streamflow calibrated SWAT model was used ................................ ................................ ................................ ............................. 69 Table 4.4. Statistical criteria for optimal multi - variable cali bration models ................................ 72 Table 4.5. Statistical criteria for the optimal GA calibrated models ................................ ............ 73 Table 4.6. Mean differences and p - values from the mixed - effects model for com parison of the different streamflow datasets used in this study. Bolded values indicate significant difference at the 0.05 level ................................ ................................ ................................ ................................ . 76 Table 4.7. Mean differences and p - values from the mixed - eff ects model for comparison of the different ETa datasets used in this study. Bolded values indicate sig nificant difference at the 0.05 level ................................ ................................ ................................ ................................ ............... 76 Table 5.1. Summary of remotely sensed ETa datase ts used in this study ................................ .... 90 Table 5.2. Average monthly ETa values for each dataset for the entire watershed with clusters indicated by superscripts for each column ................................ ................................ .................... 99 T able 5.3. Average seasonal ETa values for each dataset for the entire watershed with clusters indicated by superscripts for each column ................................ ................................ .................. 101 xi Table 5.4. T able 4. Overall dataset averages for each major landuse category with clusters indicated by superscripts for each column ................................ ................................ .................. 107 Table 5.5. Table 5. Average seasonal values of the MOD16A2 1km dat aset for the entire watershed and each major landuse category for each column ................................ .................... 108 Table 5.6. Summary of landuse and season differentiation for all ETa products used in this study, mark co nditions that could be differentiated by the product ................................ ..... 113 Table 5.7. Overall summary of average ETa values for each dataset for the entire watershed and each major landuse category with cluste rs indicated by super scripts for each column .............. 116 Table 6.1. SWAT parameters considered during the model calibration and validation process 135 Table 6.2. Summary of multi - apotranspiration performance ................ 146 Ta ble 6.3. Results of the T - test comparison of streamflow performance of the Pareto frontiers wit h a 5% significance interval. Bold p - values show no difference at a significance value of 5% ................................ ................................ ................................ ................................ ..................... 146 Table 6.4. Results of the T - test comparison of ETa performance of the Pareto frontiers with a 5% significance interval. Bold p - values show no difference at a significance value of 5% ............. 147 Tab le 6.5. Results of the Wilcoxon comparison of streamflow performance of the Pareto frontiers with a 5% significance interval. Bold p - values no difference at a significance value of 5% ................................ ................................ ................................ ................................ ............... 147 Tab le 6.6. Results of the Wilcoxon comparison of ETa performance of the Pareto frontiers with a 5% significance interval. Bold p - values show no diffe rence at a significance value of 5% ...... 148 Table 6.7. Comparison of the SWAT model and MOD16 500 m ETa product landuse datasets, CDL 2 012 and MOD16, respectively ................................ ................................ ......................... 159 Table S5.1. Average monthly ETa values for each dataset for agricultural lands with clusters indica ted by superscripts for each column ................................ ................................ .................. 168 Table S5.2. Average monthly ETa values for each dataset for forest lands with clusters indicated by superscripts for each col umn ................................ ................................ ................................ . 169 Table S5.3. Average monthly ETa values for each dataset for urban lands with clusters indicated by superscripts for each column ................................ ................................ ................................ . 170 xii Table S5.4. Average monthly ETa values for each dataset for wetland lands with clusters indicated by superscripts for each column ................................ ................................ .................. 171 Table S5.5. Average monthly ETa values for each dataset for alfalfa (ALFA) regions with clusters indicated by superscripts for each column ................................ ................................ ..... 172 Table S5.6. Average mont hly ETa values for each dataset for corn (CORN) regions with cluste rs indicated by superscripts for each column ................................ ................................ .................. 173 Table S5.7. Ave rage monthly ETa values for each dataset for field peas (FPEA) regions with clusters indicated by superscripts for each column ................................ ................................ ..... 174 Table S5.8. Average monthly ETa values for each dataset for deciduous forest (FRSD) regions with clusters indicated by superscripts for each column ................................ ............................ 175 Table S5.9. Average monthly ETa values for each dataset for evergreen forest (FRSE) regions with clusters indicated by superscripts for each column ................................ ............................ 176 Table S5. 10. Average monthly ETa values for each dataset for hay (HAY) regions with clusters indicated by superscripts for each column ................................ ................................ .................. 177 Table S5.11. Average monthly ETa values for each dataset for pasture (PAST) regions with clusters indicated by superscripts for each column ................................ ................................ ..... 178 Table S5.12. Average monthly ETa values for each dataset for sugar beet (SGBT) regions with clusters indicat ed by superscripts for each colum n ................................ ................................ ..... 179 Table S5.13. Average monthly ETa values for each dataset for soybean (SOYB) regions with clusters indicated by superscripts f or each column ................................ ................................ ..... 180 Table S5.14. Average monthly ETa values for each dataset for urban low - density (URLD) regions with clusters ind icated by superscripts for each column ................................ ................ 181 Table S5.15. Average monthly ETa values for each dataset f or urban transportation (UTRN) regions with clusters indicated by superscripts for each column ................................ ................ 182 Table S5.16. A verage monthly ET a values for each dataset for woody wetlands (WETF) regions with clusters indicated by superscripts for each column ................................ ............................ 183 Table S5.17 . Average monthly ETa values for each datas et for winter wheat (WWHT) regions with clusters indicated by superscripts for each column ................................ ............................ 184 xiii Table S5.18. Average seasonal ETa values for each dataset for agricultural lands with clusters i ndicated by superscripts for each column ................................ ................................ .................. 185 Table S5.19. Average seasonal ETa values for each dataset for forest lands with clusters indicated by superscripts for each column ................................ ................................ .................. 186 Table S5.20. Average seasonal ETa values for each dataset for urban lands with clusters indicated by superscripts for each column ................................ ................................ .................. 187 Tab le S5.21. Average seasonal ETa values for each dataset for wetland lands with clusters indicated by superscripts for each column ................................ ................................ .................. 188 Table S5.22. Average seasonal ETa values for each dataset for alfalfa (ALFA) regions with clusters indicated by superscripts for each column ................................ ................................ ..... 189 Table S5.23. Average seasonal ETa values for each dataset for corn (CORN) regions with clusters indicate d by superscripts for each column ................................ ................................ ..... 190 Table S5. 24. Average seasonal ETa values for each dataset for field peas (FPEA) regions with clusters indicated by superscripts for each column ................................ ................................ ..... 191 Table S5.25. Average seasonal ETa values for each dataset for deciduous forest (FRSD) regions with clusters indicated by superscripts for each column ................................ ............................ 192 Table S5. 26. Average seasonal ETa values for each dataset for evergreen forest (FRSE) regions with clusters indicated by superscripts for each column ................................ ............................ 193 Table S5.27. A verage seasonal ETa values for each dataset for hay (HAY) regions with clusters indicated by superscripts for each column ................................ ................................ .................. 194 Table S5.28. Average seasonal ETa values for each dataset for pas ture (PAST) regions with clusters indicated by superscripts for each column ................................ ................................ ..... 195 Table S5.29. Average seasonal ETa values for each dataset for sugar beet (SGBT) regions with clusters indicated by superscripts for each column ................................ ................................ ..... 196 Table S5.30. Average seasonal ETa values for each dataset for soybean (SOYB) regions with clusters indicated by superscripts for each column ................................ ................................ ..... 197 Table S5.31. Average seasonal ETa values for each dataset for urban low - density (URLD) regions with clusters indicated by superscripts for each column ................................ ................ 198 xiv Table S5.32. Average seasonal ETa values for each dataset for urban transportation (UTRN) regions with clusters indicated by su perscripts for each column ................................ ................ 199 Table S5. 33. Average seasonal ETa values for each dataset for woody wetlands (WETF) regions with clusters indicated by superscripts for each column ................................ ............................ 200 Table S5.34. Average seasonal ETa values for each da taset for winter wheat (WWHT) regions with clusters indicated by superscripts for each column ................................ ............................ 201 Table S5.35. Average seasonal values of the MOD16A2 500 m dataset for each major landuse cate gory for each column ................................ ................................ ................................ ............ 202 Table S5.36. Average seasonal values of the SSEBop dataset for each major landuse category for each column ................................ ................................ ................................ ................................ 203 Table S5.37. Average seasonal values of the NLDAS - 2 Mosaic dataset for each major landuse category for each column ................................ ................................ ................................ ............ 204 Table S5.38. Average seasonal values of the NLDAS - 2 Noah datas et for each major landuse category for each column ................................ ................................ ................................ ............ 205 Table S5.39. Average seasonal values of the NLDAS - 2 VIC dataset for each major landuse category for each column ................................ ................................ ................................ ............ 206 Table S5.40. Average seasonal values of the TerraClimate dataset for each major landuse category for each column ................................ ................................ ................................ ............ 207 Table S5.41. Average seasonal values of the ALEXI dataset for each major landuse category for each column ................................ ................................ ................................ ................................ 208 Table S5.42. Average seasonal values of the SWAT model dataset for each major landuse category for each column ................................ ................................ ................................ ............ 209 Table S5.43. Average seasonal values of the Ensemble dataset for each major landuse category for each column ................................ ................................ ................................ ........................... 210 Table S5 .44. Average monthly values of the MOD16A2 1km dataset for each major landuse category for each column ................................ ................................ ................................ ............ 211 Table S5.45. Average monthly values of the MOD16A2 500 m dataset for each major land use category for each column ................................ ................................ ................................ ............ 212 xv Table S5.46. Average monthly values of the SSEBop dataset for each major landuse category for each column ................................ ................................ ................................ ................................ 213 Table S5.47. Average monthly values of the NLDAS - 2 Mosaic dataset for each major landuse category for each column ................................ ................................ ................................ ............ 214 Table S5.48. Average monthly values of the NLDAS - 2 Noah dataset for each major landuse category for each column ................................ ................................ ................................ ............ 215 Table S5.49. Average monthly values of the NLDAS - 2 VIC dataset for each major landuse category for each column ................................ ................................ ................................ ............ 216 Table S5.50. Average monthly values of the TerraClimate dataset for each major landuse category for each column ................................ ................................ ................................ ............ 217 Table S5.51. Average month ly values of the ALEXI dataset for each major landuse category for each column ................................ ................................ ................................ ................................ 218 Table S5.52. Average monthly values of the SWAT model dataset for each major landuse category for each colum n ................................ ................................ ................................ ............ 219 Table S5.53. Average monthly values of the Ensemble dataset for each major landuse category for each column ................................ ................................ ................................ ........................... 220 Table S5 .54. Average monthly values of the MOD16A2 1km dataset for each individual landuse with clusters indicated by superscripts for each column ................................ ............................ 221 Table S5. 55. Average monthly values of the MOD16 A2 500 m dataset for each individual landuse with clusters indicated by superscripts for each column ................................ ............... 222 Table S5.56. Average monthly values of the SSEBop dataset for each individual landuse with clusters indicated by superscripts for each column ................................ ................................ ..... 223 Table S5.57. Average monthly values of the NLDAS - 2 Mosaic dataset for each individual landuse with clusters indicated by superscrip ts for each column ................................ ............... 224 Table S5.58. Average monthly values of the NLDAS - 2 Noah dataset for each individual landuse with clusters indicated by superscripts for each co lumn ................................ ............................ 225 Table S5.59. Average monthly values of the NLDAS - 2 VIC dataset for each individual landuse with clusters indicated by sup erscripts for each column ................................ ............................ 226 xvi Table S5. 60. Average monthly values of the TerraClimate dataset for each individual landuse with clusters indicated by superscripts for each column ................................ ............................ 22 7 Table S5.61. Average monthly v alues of the ALEXI dataset for each individual landuse with clusters indicated by superscripts for each column ................................ ................................ ..... 228 Table S5.62. Average monthly values of the SWAT model dataset for each indiv idual landuse with clusters indicated by superscripts for each column ................................ ............................ 229 Table S5.63. Average monthly values of the Ensemble dataset for each individual landuse with clusters indicated by su perscripts for each column ................................ ................................ ..... 230 Table S5.64. Overall summary of average ETa values for each dataset for each individual landuse with clusters indicated by superscripts for each column ................................ ............... 231 Table S6.1. A summary of the remote sensing ETa products used in this study ........................ 234 xvii LIST OF FIGURES Fi gure 4.1. The study area (Honeyoey Creek - Pine Creek watershed) ................................ .......... 56 Figure 4.2. Comparison of observed and simulated daily streamflow ................................ ......... 68 Figure 4.3. Monte Carlo populations and Pareto frontiers for a) ALEXI and b) SSEBop datasets ................................ ................................ ................................ ................................ ....................... 70 Figure 4.4. Pareto frontiers and optimal Pareto population members for both ALEXI and SSEBop datasets ................................ ................................ ................................ ..................... 71 Figure 5.1. Map of the Honeyoey watershed and locations o f climatological stations within and near the region ................................ ................................ ................................ ............................... 85 Figure 5.2. Map of the individual (a) and major (b) landuse classes withi n the Honeyoey watershed based on the 30 m resolution map obtained from the Cropland Data Layer developed by the United States Department of Agriculture - National Agricultural St atistics Service .......... 86 Figure 5.3. Maps showing the mean difference between each ETa dataset and the SWAT model output. Maps correspond to a) MOD16A2 1 km, b) MOD16A2 500 m, c) SSEBop, d) NLDAS - 2: Mosaic, e) NLDA S - 2: Noah, f) NLDAS - 2: VIC, g) TerraClimate, and h) ALEXI ................... 118 Figure 5.4. Maps showing the mean difference between each ETa dataset and the Ensemble. Maps correspond to a) M OD16A2 1 km, b) MOD16A2 500 m, c) SSEBop, d) NLDAS - 2: Mosaic, e) NLDAS - 2: Noah, f) NLDAS - 2: VIC, g) TerraClimate, h) ALEXI, and i) SWAT model ................................ ................................ ................................ ................................ ........... 120 Figure 6.1. Map of the Honeyoey watershed ................................ ................................ .............. 127 Figure 6.2. Comparison of the Pareto frontiers of the nine multi - objective calibrated SWAT models ................................ ................................ ................................ ................................ ......... 143 Figure 6.3 Pairwise com parisons of the streamflow objective funciton and the ETa objective funcitons, for a) the first many - objective calibration (equal weights) and 2) the second many - objective calibration (balanced weights) ................................ ................................ .................... 151 for the first many - objective calibration runs (equal weigh ts). Red bold numbers indicate highly correlated objective functions ................................ ................................ ................................ ..... 152 xviii functions for the second many - objective calibration runs (balanced weights). Red bold numbers ind icate highly correlate d objective functions ................................ ................................ ............ 153 Figure 6.6. Comparison of the landuse products utilized by (a) the SWAT and (b) the MOD16 500 m ETa product ................................ ................................ ................................ ...................... 158 Figure S5.1. Maps showing regions of statistical difference and no difference between each ETa datase t and the SWAT model output. Maps correspond to a) MOD16A2 1 km, b) MOD16A2 500 m, c) SSEBop, d) NLDAS - 2:Mosaic, e) NLDAS - 2:Noah, f) NLDAS - 2:VI C, g) TerraClimate, and h) ALEXI ................................ ................................ ................................ ...... 232 Figure S5.2. Maps showing regions of statistical difference and no difference between each ETa dataset and the Ensemble. Maps correspond to a) MOD16A2 1 k m, b) MOD16A2 500 m, c) SSEBop, d) NLDAS - 2:Mosaic, e) NLDAS - 2:Noah, f) NLDAS - 2:VIC, g) TerraClimate, h) ALEXI, and i) SWAT model ................................ ................................ ................................ ...... 233 xix KEY TO ABBREVIATIONS ALEXI : Atmosphere - Land Exchange Inverse ALFA : Alfalfa ALPHA_BF : Baseflow recession constant BIOMIX : Biological mixing efficiency BMA : Bayesian Model Averaging CANMX : Maximum canopy storage CH_K2 : Effecti ve hydraulic conductivity o f channel CH_N2 : C N2 : Moisture condition II curve n umber CO2 : Carbon dioxide concentration CORN : C orn EnKF : Ensemble Kalman filte r EPA : Environmental Protection Agency EPCO : Plant uptake co mpensation factor ESCO : Soi l evaporation compensation coefficient ET : Evapotranspira tion ETa : Actual evapo transpirati on FPEA : Field peas FRGMAX : Fraction of maximu m stomatal conductance correspon ding to the second point on the stomatal conductance curve F RSD : Forest deciduous FRS E : Forest evergreen xx GA : Genetic algorithm GOES : Geosta tionary Operation al En vironmental Satellites GSI : Maximum stomatal conductance G W_DELAY : Delay time for aquifer recharge GW_REAP : Revap coefficient GWQMN : Threshold water l evel in the shallow aquifer for base flow HAY : Hay IPET : Potential evapotranspiratio n method MAX TEMP : Dai ly maximum temperature MIN TEMP : Daily minimum temperature MOD16A2 : MODIS Global Evapotran spiration Project MODIS : Moderate Resolution Imaging Spectro radiometer NASA : National A eronautics and Space Administration NASS : National Agricu ltural Statistics Serv ice NCDC : N ational Climatic Data Center NCEP : National Cen ters for Environmental Predictio n NED : National Elevation Dataset NHDPlus : National Hydrolog y Dataset plus NLDAS - 2 : Nor th American Land Data Assimilation Syste ms 2 Evapotranspi ration NOAA : National Oceanic and Atmospheric Administration NRCS : Natural Resou rces Conservation Service NSE : N ash - Sutcliffe efficiency NSGA - II : Nondominated Sorted Geneti c Algorithm II xxi OF : Objectiv e function PAST : Pasture PBIAS : Percent bias RCHRG_DP : A quifer percolation coe fficient RE VAPMN : Threshold water level in the shallow aqu ifer for revap RS : Remote Sensin g RSME : Root mean squared error RSR : Root mean squared error - observations standard devi ation ratio SGBT : Sugar beet SOL_AWC : Available water cap acity SOYB : Soybean SS EBop : Simpl ified Surface Energy Balance SURLAG : Surface ru noff lag coefficient SWAT : Soil and Water Assessment Tool URLD : Residential low density US DA : United States Departmen t of Agriculture USGS : United States Geological Survey UT RN : Urban transporta tion VPDFR : Vapor pressure deficit corresponding to the fr action given by FRGMAX WETF : Wet lands forested WND_SP : Daily wind speed WWHT : Winter wheat 1 1. INTRODUCTION As we adv ance into the 21 st century , the Earth and human civilizati on are faced with nume rous global challenges. One of the most pressing challeng es is global water security and the first step to address this challenge is to understand th e elements of the hydrologi cal cycle that directly or indirectly impacts global water security. Historicall y , streamfl ow was the only element of the hydrological cy cle that ha s been measured at l a rge scale s . T his has been done through the use of monitoring stations; in fact, the Uni ted States Geological Survey (USGS) operates over 1.5 mill ion monitoring sites a cross the U nited States (USGS, 2016a). However, these sta tions are often expensive to ins tall and maintain and often are too spread out across the la ndscape to provide high res olution data for stakeholders, policy makers, and decision makers (Wanders et al ., 2014). T his has led to the development of modeling tec hniques that are fast, inexpensi ve, and can estimate different elements of the hydrological cycle beyond the sites of s treamflow monitoring stations (Giri et al., 2016). However , since the hydrologic al cycle is complex with many linked processes, it is ver y challenging to accurately simu late all of the ir elements (Guerrero et al., 2013). Therefor e, the first step in model setup is to assure that those elements are accurat ely rep r ese nted by the model. This will b e done through the model calibration process i n which the model parameters are adjust ed to simulate better the natural systems they are tr ying to describe (Rajib et al., 2016). Typically, hydrological modeling calibration i s performed by only co nsidering s treamflow since it can be measured more accura tely than the other components ( Immerzeel and Droogers, 2008; Rajib et al., 2016). However, since streamflow is jus t on e component of the much larger, complex hydrological cycle , considering just str eamflow in model calibration could result in poor simulat ions of other hydrologic compone nts lowering the overall model performance (Wanders et al., 2014). One 2 solution to this would be to include additional hydrological components in the calibration proce ss (Crow et al., 2003). In this regard , evapotranspiratio n (ET) would be an important add ition to the calibration process since it accounts for two - t hirds of the water on e arth and plays a major role in the cycling of water from land and ocean surface sour ces into th e atmosphere (Hanson, 1991). However, very few studies explore the addition of ET to hydrological model calibration in addition to the tra ditional streamflow cal ibra tion. Remote sensing is defined as the science of identif ying, observing, and m easuring an object without physical contact (Graham, 1999 ). With the advancements in sate llite technology, remotely sensed satellite data has become a common source of cons iste nt monitoring for the entire globe, with applications rang ing from crop yields t o water res ources assessments (Graham, 1999; Long et al., 2014). Meanwhile, in the past f ew decades, many remotely sensed ET products have become ava ilable at different spa tial and temporal resolutions. However, it is important to not e that while remote se nsing data solves the issue of data quantity, the accurac y of this data is lower compared to on the ground monitoring stations and often has a higher level of uncertainty a ssoc iated with it ( Zhang et al., 2016 ) . The limitations associ ated with the remotely sensed dat a make the implantation of remotely sensed ET products in hydrological modelin g a challenging task . Therefore, this dissertation aims to a dvance understanding of the following knowledge gaps: Knowledge Gap 1: To understand the applicability of d ifferent ca libration techniques in a hydrologic model whe n both remotely sensed ET and st reamflow data are involved. Knowledge Gap 2: To exa mine the spatial and temporal sensit ivity of different ET products in regard to landuse/landco ver and seasonal clima te variabil ities 3 To address the knowledge gap 1 the foll owing objectives were developed: (1) determine the performance of a calibrated hydr ologic mo del in estimating ET agains t spatially distributed time series ET products obtained f rom remote sensing; (2 ) determine the impact of ET parameter calibration on str eamflow estimation; and (3) eval uate the performances of different calibration tech niques fo r streamflow and ET estimat ions . To address the knowledge gap 2 the following objecti ves were e x amined : ( 1 ) explore th e temporal performance of individual and an en semble remotely sensed ET datase ts; ( 2 ) evaluate the spatial performance of individual and a n ensemble remotely sensed ET datasets; ( 3 ) compare the performance of individual rem otely sensed ET datase ts to the e nsemble and hydrological . 4 2. LITERATURE REVIEW 2.1 Over view W ith the continued growth of the human population, the demand for freshwater has i ncreased exponentially, this increase has stressed freshwa ter resources and led to their de gradation (Walters et al., 2009: Young and Col lier, 2009; Dos Santos et al., 2 011; G iri et al., 2012; Pander and Geist, 2013). This degrad ation not only impacts the environment but also the humans who rely on these freshwat er systems. Furthermor e, as globa l temperatures rise and the climate changes, f urther stressors will impact fre shwate r resources, amplifying the demands and degradations o n these limited resources ( Meyer et al., 1999; Ridoutt and Pfister, 2010). In order t o mitigate the impacts of degrada tions and insure the sustainability of freshwa ter resources. However, freshwa ter is And in order to truly under stand what is happening within one part of this cycle , it is important to know h ow all the different components interact with each other. However , with 71% of the Earth covered in water (USGS, 2016 b ), it can be challenging to mon itor all parts of the hydro logical cycle. This is where the use of remote sensing can be beneficial. Remote sensing co llects data for the entire world, from the com position of the atmosphere to th (Graham, 1999). Another benefit of remote s ensing data is that it provides a time series that allows for the evaluation of patterns an d trends that occur over time. The goal of thi s review is to explore the appli cations of remote sensing in hydrology and identify knowledg e gaps within the field. 2. 2 Remote Sensing Back in 1946, V - 2 missiles carrying camer as were launched into the atmosph ere and captured the first photographs of the Earth from space (Reichhardt, 20 06). While the images 5 captured had a poor resolution; they o ffered scientists a chance to observe the Earth remotely from space. This was the daw n of remote sensing fr om space (G raham, 1999). However , it was not until the ad vent of satellites and the techn ological advancements made in this field that led to the exp losion of space - based remot e sensing. Today there are dozens of satellites orbiting t he Earth recording how and where the Earth is changing. From observing weather patterns to monitoring deforesta tion, remote sensing has become a vital link in understandin g how anthropogenic activat es shape the surface of the Earth. Remote sensing is defin ed as the science that identifies , observes, and measures an object without phy sical contact (Graham, 1999). Th is means that the earliest forms of remote sensing began wit h the development of camera s. However, in the modern age, remote sensing utilizes the entire electromagneti c spectrum and not just visible light used in photography (Graham, 1999). Everything with a temperature greater than absolute zero ( - 273ºC) constantl y reflects, absorbs, and em its energy or electromagnetic radiation (Graham, 1999). Wh ile individual composi tions influ ence how electromagnetic radiation interacts w ith the object, its temperature has the greatest influence on the emission of electromagneti c radiation. As the tempera ture increases, the wavelength of emitted electromagnetic radiation decreases; a nd vice ver sa (Graham, 1999). The entire range of electro magnetic wavelengths is known as the electromagnetic spectrum. Due to the wide range of wav elengths found within the e lectromagnetic spectrum, several intervals were defined ; t hese include gamma - ray s, x - rays, ultraviolet, visible, infra - red, microwaves, a nd radio waves (Graham, 1999). W ith gamma - rays having the smallest wavelength (measured in p icometers) and radio waves having the longest wavelength (measured in meters) (Graham , 1999). Of this entir e range, th e human eye can only detect wavelengths that f all within the visible category (NASA, 2010a). Another important characteristic of electroma gnetic waves 6 is their abili y (Graham, 1999). The transmissiv ity is dependent on the atmospheric compositio n since different gasses absorb different wavelengths. This creates a set of absorption band s and atmospheric windows t hat describe which forms of electromagnetic radiation can pass through the atmos phere and i nteract with the surface (Graham, 1999). By ob serving how these sources of rad iation interact with the atmosphere and the surface of the E arth it is possible to meas ure the levels of specific gasses or identif ies regions of vegetation. By taking into accou nt more than just the visible electromagnetic radiation, remote sensing is abl e to provide more detailed information about the Earth and h ow it is changing. This all ows us to surpass the limitations of the human eye and obs erve patterns from glo bal trends to changes within a single farm filed (Graham, 1999). Furthermore, by collecti ng repeated time series of images of the Earth, it is possib le to preform temporal anal ysis. This allows us to track how the Earth is changing ov er time and can be use d to develo p more accurate adaptation strategies. 2.2. 1 Types of Remote Sensing Instru ments As technology has advanced, a variety of instruments h ave been integrated in to re mote sensing. These instruments can be divided into two ca tegories: passive and active (Gra ham, 1999). Passive remote sensing instrument measure the electromagnetic rad iation reflected or 1999). There are a variety of different passive instruments used for remo te sensing i ncluding: radiometers, imaging ra diometers, spectrometers, and spectroradiomete rs (Graham, 1999). Radiometers, imaging radiometers, and spectroradiometers all measure the intensity of a specific ban d of electromagnetic radiation; 7 however, while a radiomete r only measures the in tensity, im aging radiometers have the ability to develop a two - dimensional array of pixel s that represent the electromagnetic radiation intensity of the surface it was observin g , and spectroradiometers measure the intensit y of multipl e wavelength bands (Gr aham, 1999) . A spectrometer observes the wavelengths give n off by particular surfaces to identify what they are; this is possible since all objects i nteract with electromagneti c radiation differently (NASA, 2010b). All of these instru ments are used to iden tify what i osphere. In contrast, active rem ote sensing instruments emit specific frequencies of electro magnetic radiation and then measure the electromagnetic radiation as it i s reflected back to the instrument (Graham, 1 999). There are a variety of different active instruments used for remote sens ing including: radar, scatterometers, Light Detection and Ra nging (Lidar), and laser al timeters (4). Radar utilizes the emission of r adio or micr owaves to determine ho w far away an object is (Graham, 1999); this can be used to observe the topography of the Earth as well as track how surface feature are changing. A scatterometer is similar to radar in the sense it uses emitted microwaves , but is des igned to measure backs catter radi ation and can be used to measure winds over th e oceans (Naderi et al., 1991; G raham, 1999). Lidar utilizes the emission of laser pulse s an d backscattering/reflection of the pulses to determine the location of di fferent obje cts such as aerosols a nd clouds ( Graham, 1999). A laser altimeter utilizers lid ar, however instead of determin i ng the compositions of what the laser passes through it dete rmines the height of the in 9). This is very similar to radar and is also used l as changes that occur such as the loss of glaciers. 8 2.2.2 Current Remote Sensing Projects With so many different type s of instruments that can b e used for remote s ensing, it i s no surprise that the re are also a great number of different remote sensing pr ojects. Each project has differe nt primary purposes that can range from tracking the composi tion on the atmosphere or m easuring the loss of glaciers and ice sheets. The followin g sections describe so me of the b etter - known remote sensing projects. It is imp ortant to note that for this dis sertation the remote sensing products are referred to any pr oducts that used remote sen sing in a direct or indirect manner to calcula te values su ch as potential evapot ranspiratio n. 2.2.2.1 Aqua The Aqua Earth - observing sat ellite mission, launched by the National Aeronautics and Space Administration (NASA) in 2002 , collects information on t he hydrological cycle of the Earth as well as radiative en ergy fluxes, aerosols, vegetation cover on the land, phytoplankton and dissolve d organic matter in the oceans, and air, land, and water temperatures (NASA, 2017b). In orde r to collect all of this in formation Aqua utilizes an array of six instru ments: the A tmospheric Infrared So under (AIRS ), the Advanced Microwave Sounding Unit (AMSU - A), the Humidity Sounder for Bra zil (HSB), the Advanced Microwave Scanning Radiometer for EO S (AMSR - E), the Moderate - Re solution Imaging Spectroradiometer (MODIS), an d the Clouds and the Earth's Radia nt Energy S ystem (CERES) (NASA, 2017j). The AIRS instrume nt is used to observe and map ai r and surface temperatures, water vapor, and cloud propertie s ( NASA, 2005b ). Furthermor e , AIRS can measure trace levels of greenhouse gasses in t he atmosphere ( NASA, 2 005b ). The AMSU - A instrument is used to not only to colle ct data on upper atmosphere temp eratures but also to collect data on atmospheric water ( NASA , 2005a ). The HSB instrumen t is used to collected humidity profiles throu ghout the at mosphere (NASA, 2017i) . By 9 combin ing the observations of the AIRS, AMSU - A, and HSB it is possible to collect hu midity profiles even when clouds are present (NASA, 2017i). The AMSR - E instrument is us ed to collect data on precipitation rates, clo ud water, wa ter vapor, sea surface winds, sea surface temperatures, ice, snow, and soil moi sture (NASA, 2017a). This was do ne by observing the intensity of emitted microwaves from the 7a). The MODIS instrument is used to collect p hysical prop erties of the atmosphe re, oceans, and land as well as biological properties of the oceans and land (NASA, 2017a a). The CERES instrument us used to collect information on t he electromagnetic radiatio surfa ce (NASA, 20 17f). This data can be used to ev aluate the thermal radiation budget of the Ear th. The combined observations of these instruments provide highly detailed information that is useful to policy makers since it provides maps of how the Earth is changing and he lps identify which reg ions requir e immediate mitigation projects. 2.2.2.2 Aqua rius The Aquarius Project provi ded worldwide data about ocean salinity ( NASA, 2017c ). This data was used by scientist s to advance our understanding of how changes in the salini ty of the ocean affect ed by the h ydrological cycle as well as ocean currents ( N ASA, 2017c ). Aquarius was launch ed on June 10 th , 2011, and remained in operation until June 8 th , 2015 ( NASA, 2017k ). T h roughout its time of operation, Aquarius produced a new sa linity map for the wor ld every se ven days ( NASA, 2017ad ). To evaluate the salin ity , three passive microwave rad iometers were used to detect minute changes in the ocean sur face emissions that corres p onded to the levels of salt within the water ( NASA, 2017c ) . Overall this mission was succes sful in the fact that it provided more data th an had been collected before and allowed for the advancement of our 10 understanding of how fre sh and salt water interact as well as how the ocean currents and circulations occur. 2.2.2.3 CBERS Series The CBERS or China Brazil Earth Resource Satellites are a series of satellites developed jointly between China and Brazil (INPE, 2011d). Currently , three satellites ( CBERS - 1, CBERS - 2, and CBERS - 2B) are in orbit capturing images of th have been used to track deforestation and monitor water resources and urban growth (INPE , 2011e). These satellites are equipped with high - resolution charge - coupled device cam e ras, an infra - red multispectral scanner (replaced in the C BERS - 2B with a high - re solution pa nchromatic camera), and a wide field imager (I NPE, 2011b). These instruments c from multiple spectral bands with resolutions ran g ing from 260 to 2.7 m 2 (INPE, 2011a). This allows for very precise measurements of the Eart policy makers . Given the success of these sate llites, two additional satellites ( CBERS - 3 and CBERS - 4) are secluded to be launched in the near future (INPE, 2011c). 2.2.2.4 CryoSat Series Th e mission of the CyroS at Satellit es is to monitor the thickness of the polar ic es sheets as well as identify re gions where the ice sheets are changing (ESA, 2017k). The Cr yoSat project was initiate d in 1999 by the European Space Agency (ESA) , and the first satellite was launche d in 2005 ( ESA, 2017k). However, this satellite was destr oyed during launch. Therefore, C ryoSat - 2 was built and successfully launched in 2010 (ESA, 2 017k). In order for this ne w satellite to collect the desired data , it must cover the distance between 88 d egrees nort h and 88 degrees south on every orbit. This is a v ery unique orbit and require d special consideration during the design process (ESA, 2017 d). The main payload for th e CryoSat - 2 is the Synthetic Aperture 11 Interferometric Rada r Altimeter, which was specially designed to detect changes in ice sheets (ESA, 201 7k). In fact, this instrumen t can measure changes in ice sheets at an accuracy of 1.5 cm /year over the open ocean ( ESA, 2017c). This provides researchers with detailed infor mation about how the E sphere is being affected by seasonal and clima te v ariabilities. 2.2.2.5 ENVI SAT Launched by the ESA in 2002, the Environmental Satellit e or ENVISAT was the succes sor to European Remote Sensing (ERS) satellites launched i n the ( ESA, 2017 v ). The main objective of this satellite w as to continue a nd e xpand the observations being collected by the ERS satellites ( ESA, 2017i ). This was done by expanding the range of observed parameters to allow for observations of not only but also i ts oceans, cryosphere, and atmosphere. This wo uld allow researchers to be bett monitor the Eart his objective , the satellite was designed and mounted with ten different sensors that allow it to collect environmental monitoring data f rom a wide range of spectral and spatial resolutions ( ESA, 2017g; ESA, 2017h ). These sensors include: the Advanced Alon g - Track Scanning Radiometer (AATSR), Advanced Synthetic Ap erture Radar (ASAR), D oppler Orbi tography and Radio - positioning Integrated by S atel lite (DORIS), Global Ozone M onitoring by Occultation of Stars (GOMOS), Laser Retro Refle ctor (LRR), Medium - Resoluti on Visible and Near - IR Spectrometer (MERIS), Michelson Int erferometer for Passiv e Atmospher ic Sounding (MIPAS), Microwave Radiometer (MWR ), R adar Altimeter 2 (RA - 2), and Scanning Imaging Absorption Spectrometer for Atmospheric Ca rtography (SCIAMACHY) ( ESA, 2017g ). 12 2.2.2.6 GEDI The Global Ecosystem Dynamics Inves tigation or GEDI will utilize lig ht detection and ranging (lidar) to produce hi gh - r esolution 3D images of the E These images will be used to h elp improve current underst anding and monitoring of major focus areas including fores t management and carbo n cycling, water resources, weather prediction, and topog raph y and surface deformation ( N ASA, 2016 ). In order to develop these 3D images, GEDI will f ire a total of 726 laser pu lses per second ( NASA, 2016 ). GEDI is expected to be laun ched in 2019 by NASA a nd will be attached to the International Space Station (N ASA, 2017g). 2.2.2.7 GOCE The Gravity field and steady - state Ocean Circulation Explorer sa tellite or GOCE, was launch ed in 2009 by the ESA to advance our understanding of the ( ESA, 2017l ). In order to itational field, GOCE was equipp ed with the Electrostatic Gravity Gradiometer (EGG), which w as composed of a set of six 3 - axis accelerometers ( ESA, 2017j ). This made it the most sensitive gradiometer ever flown in space and allowed GOCE to measure gravity gradients across the globe ( ESA, 2017e ). While the GOCE mission ended in 2013, the data coll ected by GOCE co ntinues to be utilized in a wide range of fields including oceanograp hy, solid Earth physic s, and geod esy and sea - level research ( ESA, 2017l ). 2.2.2 .8 GOSAT The Greenhouse Gases O Japa n Aerospace Expl oration Age ncy (JAXA) in 2009 with the sole focus of observing carbon dioxide and methane f rom space ( NIES, 2017b). This made it the first satellite to focus on greenhouse gas mapp ing. GOSAT utilizes a thermal and near infrared sensor to m easure 13 atmospher ic greenhou se gases, which is composed of two components: 1) a Fourie r Transform Spectromet er that tar gets O2, CO2, CH4, and H2O in the atmosphere a nd 2) a Cloud and Aerosol Imager targets clouds and aerosols in the atmosphere (NIES, 2017a) . The data colle cted by the se sensors have allowed researchers to map global distribu tions of carbon dioxid e and metha ne as well as identify how these concentration s change over time (NIES, 2017b) . 2.2.2.9 Jason Series Following in the steps of early ear th ocean topogra phy mission s the Jason series of satellites each focus on the continu ed monitoring of the t opography o providing scientists wit h detailed information about cha nges in the depths of the oceans. The first of the three Jas on satellites, J ason - 1, was launched in 2001 and continued to provide information abo ut ocean topography un til 2013 (N ASA, 2017x). Jason - 1 was used not only to moni tor the s oceans but also t o monitor the mass distributions of the E arth, which coul d be used t l). The next satellite was the OS TM/ Jason - 2 and was launched in 2008 (NASA, 201 7ab). The goals for this satelli te were to continue the data collection of the Jason - 1 (NASA , 2017ac). And f inally , the Jason - 3 satellite is planned for launch in 2015 and will continue the data coll ection of o cean topography like the Jason - 1 and OSTM/Jaso n - 2 (NASA, 2017m). Each of these satellites provide s data necessary to monitor how the ocean s are changing and can lead to forecasting of large - scale weather systems such as El Niño. 2.2.2.10 Landsat Series An other series of satellites launched by NASA, t he Landsat series consists of a string of eight sate llites (NASA, 2017h), with the first lau nched in 1972 (NASA, 2017n) and the most recent launched in 2013 (NASA, 2017u). The g oal and focus of these satellites ha ve been to 14 provide detailed records of how land cover changes across the gl obe (NASA, 2017v). L andsat 1 was launched in 1972 and was th e first Earth - observing sat ellite to focus solely on monitoring rface (NASA, 2017n). E quipped wit h a camera (Return Beam Vidicon (RBV)) and a m ultispectral scanner (MSS), Land sat 1 continued to f unction until 1978 and collected over 30 surface (NASA, 2017n). Landsat 2 was launched in 1975 and remained in service u ntil 1983 a nd was almost identical to Landsat 1 (NASA, 20 17o). Following the success of L andsat 1 and 2, Land sat 3 was launched in 1978 and remained in service until 1983 and m aintained the use of the RBV and MSS (NASA, 2017p). Howeve r, Landsat 3 had an im proved spat ial resolution that allowed for more accurate images of the ASA, 2017p). Landsat 4 was launched in 1982 and remained in orbit until 2001 (NASA, 201 7q). Unlike previous Landsat satellites, Landsat 4 did not use the RBV camera an d instead f ocused on expanding the spectral and spatial r esolutions through the use of th e Thematic Mapper (T M) and MSS (NASA, 2017q). Landsat 5 was launched in 1984 and remain ed operable until 2012 (NASA, 2017r). Landsat 5 was very s imilar to Landsat 4 an d even util ized the same sensors (MSS and TM) (NASA, 2017 r). Landsat 6 was planned to beg in use in 1993, howe ver , due to a disastrous launch, never m ade it to orbit (NASA, 2017 s). After the failure of Landsat 6, Landsat 7 was successf ully launched in 1999 and is stil l in operation today (NASA, 2017t). In continu ing with the trend on improving each successive sate llite, Landsat 7 again improved the spec tral and spatial resolution s of the collected data through the use of the Enhanced Th ematic Mapper Plus (ET M+), which replaced the TM used in previous satellites (N ASA, 2017t). Unfortunately, in 2 003 a hardware failu re on Landsat 7 resulted in gaps in the collected images that reduc e the usefulness of the collected data (NASA, 2017t). Lan dsat 8 was launched in 2013 and i s still functional today (NASA, 2017u). Given the advancements in 15 technology t hat have occurred, L andsat 8 is equipped with two new sensor s: 1) the Operational Land Imager (OLI) and 2) the Thermal Infrared Sensor (TIRS) (NA SA, 2017u). These sens ors still c over the spectral regions that were covered by the ETM+ on Landsat 7 but also improve the spectral resolution by adding two new spectral b ands and divide the ETM+ th ermal infrared band into two spate bands (NASA, 2017u). Co mbined the Landsat ser ies represe nts the longest lasting set of Earth observati ons, which makes this data vital to understanding ho w the planet has changed over the past 5 0 years (NASA, 2017v). 2.2 .2.11 METEOSAT Series The Meteosat satellites are geostat ionary meteorological satellites launched by the European Organization for the Exploitation of Meteorological S atellites (EUMETSAT) (EUMESAT, 2017b). These satellites are used to monitor weather con ditions across the globe and provide vital information for daily life as well as early warn ings of severe weather conditions (EUMESAT, 20 17b). Currently, EUMETSAT has fo ur Metosat satellite s in orbit ( Metosat - 8, Metosat - 9, Metosa t - 10, and Metosat - 11) . H owe ver , only Metosat - 8, Metosat - 9, and Metosat - 10 are current ly in use over Europe, Africa, an d the Indian Ocean (EUMESAT, 2017b). Each Meto sat satellite is equipped with t hree main components namely the Spinning Enhanced Visible an d Infrared Imager, the Geos tationary Earth Radiation Budget scanning radiometer, and the Mission Communicat ion Payload (EUMESAT, 2017a). These instruments allow the Metosat satellites to help dete ct and forecast a wi de range of weather and atmosphere condi tions including thunderstor ms, fog, dust storms, and volcanic ash clouds (EUMESAT, 20 17b). 2.2.2.12 METOP S eries The Meteorological Operational Satellite Programme (Metop) is a set of three satel lites (Metop - A, Meto p - B, and Metop_C) launched by the ESA to monitor meteorological var iables 16 across the globe, including temperature, moisture, and interactions withi n the atmos phere and between the atmosphere and the ocean (EUMESAT, 2017c; EUMESAT, 2017d ; EUMESAT, 2017e). I n order to observe all of these variable s, each Metop satellite is equipped with eleven scientific instruments including the Infrared Atmospheric S ounding Int erferometer, the Global Ozone Monitoring Exper iment - 2, the Advanced Very High Resolution Radiomete r/3, the Advanced Scatterometer, the Glo bal Navigation Satellite Sy stem Receiver for Atmospheric Sounding, the High Resolutio n Infrared Radiation S ounder/4, t he Advanced Microwave Sounding Unit A1 and A2, the Microwave Humidity Sounder, the Advanced Data C ollection System/2, the Search and Rescu e Satellite - Aided Tracking System, and the Space Environment Monitor (EUMESAT, 2017c) . The data collected b y these ins truments makes the Metop series of satellites a valuable resource for meteorol ogists and climatolo gist around the globe. 2.2.2.13 Sentin el Series Comprising of a s et of seven satellites (Sentinel - 1, Sentinel - 2, Sentinel - 3 , Sentinel - 4, Sentinel - 5, Sentine l - 5 Precursor, and Sentinel - 6), the Sentinel s atellite fleet launched by the E uropean Space Agency (ESA) focus on providing a variety of m surface, ranging from land cover identification to atmosph ere condition monitori ng (ESA, 20 17b). Sentinel - 1 utilizes an advanced radar in weather as well as m data collected by Sentinel - 1 can be used for a variety of applications including the monitoring of sea ice (ESA, 2017 q), the observation of changing land uses (ESA , 2017a), and the mapping of ter rains after natural disasters (ESA, 2017f). Sentinel - 2 utili zes a high - resolution multi (ESA, 2017n ). This supplies scien tists with which can be used for a variety of purposes, such a s monitoring plant health, changing land s, water 17 bodies, and natura l disaster (ESA, 2017n). Sentinel - 3 utilizes several instr uments to collect data on ocean t opography, surface temperatures, and surface c olors (ESA, 2017o). The instrume nts used by Sentinel - 3 include a Sea and Land Surface Temper ature Radiometer (SLSTR), a n Ocean and Land Colour Instrument (OLCI), and a Synthetic Aperture Radar Altime ter (SRAL) (ESA, 2017o). The Sentinel - 4, Sentinel - 5, and Sentinel - 5 Precursor missions f ocus on monitoring t he . The data collected throug h these satellites can be used to monitor changes in green house gasses well as m onitor chan ges in the ozone layers (ESA, 2017r). And fina lly, Sentinel - 6 focuses solely o n monitoring ocean t opography, producing new global images o f the oceans every ten days (ESA, 2017s). This data is vital to monitoring how the oc peeds, and wave height vary (ESA, 2017s). All of the data collected by the Sentinel Serie s provide scientist with a global view of how interconnected the Earth is as well as mo nitor how conditions are changing so policymakers can make informed decisions to implement mitigation strategies in the region that need the most help. 2.2.2.14 SMOS Th e Soil Moisture and Ocean Salinity (SMOS) mission was launch ed by the ESA in 2009, with two main objectives monitor the soil moisture of the land and the salinity of t he oceans ( ESA, 2017p ), both of which have major impacts on the hydrological cycle. The o utput of these obser vations are sets of global maps at 3 - day increments ( ESA, 2017t ); t his supplies scientist with a steady time series of data p oints that can be used to monitor changes in both salinity and soil moisture ov ertime . Furthermore, these sets of maps can be used and integrated with other hydrological c haracteristics to better un derstand how changes in soil moisture and salinity are con nected to the bigger h ydrological cycle. This can lead to more accurate weather predictions, better monitoring of the cryosphere, a nd improve water management projects ( ES A, 18 2017u ). To create these maps the SMOS utilizes a 2D interferometric radiometer ; th is is unique since it is currentl y the only satellite to utilize this instrumen t in a polar - orbiting alignment ( ESA, 2017p ). 2.2.2.15 SWOT The Surface Water Ocean Topogra phy or SWOT satellite is a joint project between ét udes Spatiales with a mission t o improve current understanding of global hydr ology (NASA, 2017ae). This will be a vital resource for monitoring and maintaining the Earth Curren tl y SWOT is expected to be launched within the next d ecade (NASA, 2017ae) . 2.2.2.16 Terra The Terra Earth - observing satellite mis sion, launched by NASA in 1999, collects w, ice, and energy budget ( NASA, 2017y). In order to collect all of this information Te rra utilizes an arra y of five i nstruments: the Advanced Spaceborne Thermal Em ission and Reflection Radiometer (ASTER), Clouds and Multi - angle Imaging Spectro radiometer (MISR), Moderate - resolution Imaging Spectroradi om eter (MODIS), and Me asurements of Pollution in the Troposphere (MOPITT) (NASA , 2017af). The ASTER instrument is used to observe and map land surface temperature, emissiv ity, reflectance, and eleva tion (NASA, 2017d). The CERES instrument us used to collec t information on the e lectromagne tic radiation reflected and emitted from the E used to measure the total radiation budget of the Earth (NA SA, 2017e).The MISR instrum ent is used to observer how electromagnetic radiation from t he sun interacts wit h the atmos phere (NASA, 2017w). This allows scientists to gather information about the co mposition of the atmosphere as well as what type of clouds a re 19 present and even landuse characteristics (NASA, 2017w). The MODIS instrument is us ed to collect physical properties of the atmosphere, oceans, and land as well a s biological properties of the o ceans and land (NASA, 2017z). The MOPITT instrument is used to observe how the lower at r focus placed on the movement of carbon monoxide (NASA, 2017ab). All of these instruments, like those in the A qua satellite, can provide scientists with highly detailed d ata and maps for monitoring how the Earth is changing. Furthermore, this data also al lo ws scientists to eva luate the r elationships between the different spheres (su ch as the atmosphere and biosphe re) of the Earth expanding our knowledge of how different pr ocesses respond to climate changes, enhancing future predictions of what can be expec te d . 2.2.2.17 TOPEX/Po seidon The TOPEX/Poseidon mission was launched by NASA i n 1992 and collected data until 2006 (NASA, 2017ag). During this time the TOPEX/Poseidon sat ellite collected data on th e topography of the oceans (NASA, 2017ag). This was the fi rs t satellite - based oc ean topogra phy mission and opened areas of research with respect to the interactions of o cean circulation and large - scale weather systems, such as El Niño (NASA, 2017ag). Ocean topography measurements observed were accurate to 4.2 cm (N ASA, 2017ag), this a llowed scie ntists to understand better how ocean circulat ion occurred and how it influenc es the rest of the Earth system processes, such as weather p atterns. 2.2.2.18 TRMM The Tropical Rainfall Measuring Mission or TRMM was a joint p ro ject between NASA an d the Japan Aerospace Exploration Agency that was launche d in 1997 and collected data unt il 2015 (NASA, 2017aj). The main goal of TRMM was to monitor precipitation for the trop ical and 20 sub - tropical regions of the Earth to determine th e distribution and var iability of precipitation across this region (NASA, 2017a i). TRMM accomplished this goal through the use of five instruments, namely the Visible Infr ared Radiometer, the TRMM M icrowave Imager, the Precipitation Radar, the Cloud and Ea rt h Radiant Energy Sen sor, and th e Lightning Imaging Sensor (NASA, 2017ah). The se instruments allowed TRMM to c ollect 3D images of storm systems that continue to be used t o improve our understanding of climatological events in the tropics. 2.3 The Hydrolo gi c Cycle We are surro unded by wa ter, from water vapor in the air to oceans and glaciers. In fact about 71% of the planet is covered in water (USGS, 2016 b ). However, we te nd to focus only on freshwa ter sources that are needed for drinking and agriculture a nd impact our lives da ily. Freshw ater is a very limited resource (USGS, 2016 c ); and with current population gro wth trends and changes brought on by climate change, it has become vital to insure the sustainability of these resources. The amount of freshwate r available is depende nt on how w ater is circulated through the atmosphere, acr oss the ground, through the crus t, and even through the biosphere in a process known as the water cycle or the hydrolog ical circle (USGS, 2017a). And the impacts th at occur in o ne sector of the cycle have casca de effects in other sectors ( Maxwell and Kolle t, 2008; Stampoulis et al., 2016 ). Therefore, in order to insure that the hydrological cycle continues to function, it is important to evaluate and monitor the chan ges within al l components of the hy drological cycle. However , with such a large amount of th e surface covered in water , this can be a daunting task. Furthermore, the process of collect ing data from monitoring st ations would only provide information at a fixed number of points making it diff icult to de termine how the 21 hydrological cycle is changing . Yet with the technological adv ancements in satellite technology , remote sensing data can h elp fill this data gap. The hydrological cycle can be broken down into the following components: evapotrans piration, g roundwater, oceans, precipitation, snow and ic e, soil moisture, surface water, and water vapor. Within each of the following sections, eac h component of the hydrolog ical cycle will be briefly explained . 2.3.1 Evapotranspir ation Evapotranspirat ion describ es the amount of water that is transferred fro m the surface to the atmosphere (USGS, 2016 d ). This includes both the water that just evapor ates from the ce as well as the water lo st from plants (transpiration) ( USGS, 2016 d ). This pro cess is res ponsible for weather patterns by supplying the water vapor needed to drive the weather systems that return water to the land (USGS, 2016 e ) . Therefore understanding t he levels and changes in e vapotranspiration for a region a llows us to monitor ho w much wate r loss occurs and can be used to figure out ho w much water remains. This is es pecially vital for agricultural lands where it can be used t o determine if there is eno ugh water to maintain crop yields or if irrigation is need ed. 2.3.2 Groundwater While gro undwater only accounts for about 0.8% of the w ater found on Earth, it represen ts about 30.1% of all the freshwater (USGS, 2016 f ). This mak es it a vital source of the limited freshwater, espec ially for regions where there is not enough rainfall o r surface w ater to supply the needs of anthropogenic acti vities. This has led to the inst allation and use of pumps and wells used to pull water up fr om the groundwater aquifers or reservoirs. However , t his is still a limited resource and can become deplete d if too mu ch is removed too quickly (USGS, 2016 g ). 22 This is easily evident in the shrinki ng of the Ogallala Aquifer in the great plains of the United States (Terrell et al., 20 02). 2.3.3 Oceans Oceans ce and account for abo ut 96.5% of all water on earth ( USGS, 2016 b ). Furthermore , all of the water in the oceans is called saltwater due to the significant levels of dissol ved salts found within it ( USGS, 2016 h ). This makes a ll the water in the oceans unusa ble for either drinkin g or agricu lture use without removing the salts. And whil e desalination processes that ca n purify saltwater exist, they are often expensive and requi re high energy inputs in or der to be useful to large populations ( USGS, 2016 i ). And w ith current efforts fo cusing on t he availability of freshwater, the oceans are often left out of consideration. However, while the water in the oceans is not easily access ible, it is estimated that 90% of all water vapor in the air comes from the oceans ( U SGS, 2016 h ). This show s that ocea ns , while seeming to only hold unusable water , have major impacts on weathers systems and drive much of the hydrological ( USGS, 2016 h ). Fu rthermore, the constant mov ement of water both through circulation in the water colum n and across the globe through oc ean currents alter the temperatures of the wat er ( USGS, 2016 h ). This , in turn, affects the evaporation rates across the globe and drives w eather cycles worldwide. Th erefore several diff erent remote sensing projects have foc used on monitoring the characteri stics of the oceans in order to determine how the oceans impact the rest of th e hydrological cycle. 2.3.4 Precipitation The process by w hich water vapor condenses and falls back to Ea as precipitatio n (USGS, 2016 j ). And w hile precip itation can have many forms depending on the c onditions of the atmosphere, it is the other main process (like evapotranspiration) that dri ves the 23 water cycle (USGS, 2016 j ). Therefore un derstanding how the rates of precipita tion change over the s urface of t he Earth allows us to determine which regions will have access to water or whe re water will be sparse. This is especially vital for agricu ltural lands where it can b e used to determine how much water is returning to the fie lds. When combined wit h evapotran spiration, it can be used to estimate how much water is present at farm fields , and help determine if pumps or irrigation systems are need ed to maintain crop yields . 2.3.5 Snow and Ice Snow and ice, also known as the cryos phere, represent anoth er source o f freshwater similar to groundwater. However , there is more than double the am ount of groundwater that can be found d ice reserves. Snow and ic e account for about 1.7% of all water and 68.7% of all fre shwater (USGS, 2016 k ). However, w hile this is a much larger source of freshwate r , it is harder access with most of it being found in glaciers and the ice sheets at the pol es. Yet, while most of this stored freshwater i s not accessible , it plays an importan t role in influencing climate ( USGS, 2016 k ). Due to the highly refl ective nature of snow and ice, m uch of the incoming electromagnetic radiation from the sun i s reflected back into space . This helps slow th e rate at which the Earth absorbs heat ; however with the rec ent rises i n global temperatures glaciers and ice sheets are rapidly disappearing, this , in turn, results in more energy and heat being absorbed by t he Earth and further meltin g of the snow and ic e ( USGS, 2016 k ). Furthermore , as this melting occurs , it alt ers other p arts of the hydrological cycle such as rising ocean levels (NSIDC, 2015). All of these factors ha ve made it vital to monitor the global ch anges in the cryosphere. 24 2. 3.6 Soil Moisture Soil moisture is similar to groundwater in the fact th at both groundwate r and soil moisture are measures of water in t he ground. However , unlike groun dwater, soil moisture describes the amount of water found wi thin the top layers of the This makes it vital to the a gricultural pro cess si nce this is the water that agricultural plants can draw f rom during their growing phase ( NASA, 1999; USGS, 2016 g ). Soil moisture is highly dependent on the temperature as well as evapotranspiration and precipitation (NIDIS, 2013). Wit h the need to m aintain or even in growing population, understandi ng how soil moisture levels vary across agricultural lands c an be used to estimate crop yields and lead to the implementation of mitigation measu res . 2.3.7 Surf ace Wat er Surface water is used to describe all other sources o f freshwater on the surf ace. This includes rivers, lakes, and swamps; and is the eas iest form of freshwater to access. However, surface water only accounts for about 0.2 9% of all fresh water o n the Earth (USGS, 2016 l ). And due to their ease of acces s, surface waters are often impa cted by anthropogenic activities (USGS, 2016 m ). This has led to an increase in the focu s put on these freshwater systems with the goals of mitiga ting anthropoge nic imp acts and in sure the sustainability of these systems for f uture generations (Walters et al ., 2009: Young and Collier, 2009; Dos Santos et al., 2011; G iri et al., 2012; Pander an d Geist, 2013). Therefore, it has become important to moni tor these syste ms . 2.3 .8 Water Va por When water evaporates , it becomes water v apor and enters the atmosphere. Once in the atmosphere , it interacts with electromagnetic ra diation; as the most abunda nt greenhouse gas, 25 water vapor traps the electromagnetic r adiation emitte d by Ea rth (NASA, 2008). This drives the warming trends seen in recent years. Furthermore, water vapor is vital to the weather of the world , wind currents m ove water vapor across the globe and as the temperature of the atmosphere changes wat er vapor conden ses to form clouds , the source of all precipitation ( USGS, 2016 n ). Therefore , by monitoring the water vapor levels in the atmosphere , it is possible to trac k the movement of water acr oss the globe as well as determine how much global tempera tures will incr ease. 2 .4 Monitori ng Water Resources Given the importance of wa ter resources and the increasing demand on these limited resources , it has become vital to e nsure their sustainability for future generations. However, given the complexity of t he hydrological cycle , this can b e challenging. Traditionally monitoring statio ns are used to measure different components ( e.g. , streamflow and ET) of the hydrological cy cle (Deser et al., 2000; NO AA, 2017 a ; USGS, 2017b ). In fact , when considering ET, the MSU Enviro - weather Pr ogram has 6 4 stations within the state of Michigan alone that provide valuable data for r esearchers (Bishop, 2010). However, compared to the size of Michigan that is roughly on e station every 3,914 km 2 . And si nce ET is a spatially dis tributed property, hav ing a resol ution like this would result in models that ar e unable to account for the vari ability in ET that exists in the landscape. This is true for other hydrological cycle c omponents as well, for which high er spatial resolutions ar e often needed by rese archers (Wa nders et al., 2014). At the same time, it is n ot feasible to install monitorin g stations every few hundred yards due to installation and m aintenance costs. One solut ion to this issue is the use of r emote sensing. This is ev en more evident given the vast nu mber of remote sensing projects that were disc ussed earlier in this review. In fact, remote sensing has even been used to develop spatial datasets for hydrological 26 c ycle components such as ET ( Kite and Droogers, 2000 ). The following sections des cribe a few on the more well - known remote sensing ET data sets and how they are calculated . 2.4.1 MOD16 MOD16 or MODIS Global Evapotranspiration Proj ect calculates 8 - day, month ly, and annual ET by using an alg orithm developed by Mu et al. (2011), which is based on th e Penman - Monteith equation. Below the Penman - M onteith equation is shown: ( 2. 1) w here is the latent heat flux; is the latent heat of e vaporation; s is the s lope of the curve relating saturated water vapor pressure ( e sat ) to temperature; A i s the available energy partitioned between sensible heat, latent heat and soil heat fluxes o n land surfaces; is the air density; C p is the specific heat capacity of air; r a is the a erodynamic resistance; r s is the surface resis tance; and is the psychro metric constant (Mu et al., 2011). This equation serves as the ba ckbone for ations. However, MOD16 divides the total ET into three mai n components as follow s: ( 2. 2) w here , wet _C is the evaporation from wet canopy surfaces; trans is the pl ant transpiration; and SO IL is the actual soil evaporation (Mu et al., 2011). This allows for the use of more specif ic equations to describe how water is lost fro m different surfaces. Equat ions 2. 3 through 2. 5 show the individual equations used for each component of the total ET ( Eq. 2. 2): wet_C : ( 2 . 3) 27 trans : ( 2 . 4) SOIL : ( 2. 5) wher e A C is the available energy partitioned between sensible heat, latent heat and soil heat fluxes allocated to the canopy; F C is the vegetation cover f raction; F wet is the water cover frac tion; P a is the atmospheric pressure; rvc is th e wet canopy resistanc e; is the emissivity of the atmosphere; A SOIL is the ava ilable energy partitioned between sensible heat, latent heat and soil heat fluxes allocated to the soil surface; VPD is the vapor pressure deficit; r as is the aerodynamic resis tance at the soil surface ; r tot i s the total aerodynamic resistance to vapor tra nsport; and RH is the rela tive humidity (Mu et al., 2011). From these equations, it is easy to see the influence of the Penman - Mo nteith equation on the MOD16 ET estimations. Ho wever, these equations do not ind icate what input data is required to calculate MOD16 ET. The following ta ble (Table 2.1) lists the datasets that were used to perform the a bove calculations: Table 2. 1. List of datasets used to calculate MOD16 ET Dataset Re motely Sensed GMAO meteo r ologica l data YES MODIS FPAR/LAI YES MODIS landcove r type 2 YES MODIS albedo YES 28 2.4.2 ALEXI ALEXI or the Atmosphere - Land Exchange Inverse M odel calculates daily ET by relating changes in morning surface temperatures to wate r loss (Anderson et al., 2007). T o do this, ALEXI utilizes a two - source energy b alance model that divides into two components, soil and canopy (Anderson et al., 2007). By doing this , it is possible to solve for the ET of each component before combining them aga in to de termine the overall ET. The first step is to ex tract the individual compo nent temperatures f rom the satellite data. This is done using the following equation: ( 2. 6) whe re, T RAD is the composite directional surface radiometric temperature; is the fractional cover; T S is the soil temperature; and T C is the canopy temperature (Anderson et al., 2007). After this, individual surface energy balance equations can be solved for both the soil (Eq. 2 . 7) and canopy (Eq. 2. 8) as follows: ( 2. 7) ( 2. 8) w here , RN is the net radiation; H is the s ensible heat; is the latent heat ; and G is the soil heat con duction flux. For these e quations , S C respectively (Anderson et al., 2007). In these equations, observed net radiation and surface temperature are used to solve f or ET. However, in order to determine the overa ll ET the individual comp onent ETs need to be summed as fol lows: ( 2. 9) 29 where is the ET of the soil and is the ET of the canopy (Anderson et al., 2007) . Similar to MOD16 a variety of input datasets are required to perform these calculations. Table 2. 2 presents these d ataset s: Table 2. 2. List of datasets used to calculate ALEXI ET Dataset Remotely Sensed ASOS/AWOS wind data NO GOES c loud co v er YES GOES net radiation YES GOES surface temperatures YES MODIS LAI YES Radiosonde lapse rate profile YES Radioso nde atm o spheric corrections YES STATSGO soil texture NO UMD global landcover YES 2.4.3 SSEBop SSEBop or the Operational Simp lified S urface Energy Balance Model calculates monthly and annual ET by combining ET fractions derived from remotely sensed MODI S therm a l imagery and reference ET (Senay et al., 2013). This is done by using the following equation: ( 2. 10) where ETf is the ET fraction; ETo is the grass reference ET for the location obtained from global weather dataset s; and k is a coefficient that scales the grass reference ET into the level of a maximum ET experienced by an aerodynamically ro ugher c r op (Senay et al., 2013). In order to calculate the ET fraction the following equation is used : ( 2. 11) where, Ts is the satellite - observed land surface temperature of the pixel whose ETf is being evaluated for a given time period ; Th is the estimated Ts condition of the pixel for a given time period ; and T c is th e estimated Ts at the idealized 30 time period . This makes the determination of Th and Tc key for estimating ET. In order to estimate Tc the following equation is used : ( 2. 12) where, Ta is the near - surface maximum air temperature for the given time period and c is a correction f actor that relates Ta to Ts for a well - watered, vegetation surface (Senay et al., 2013). Once Tc was determined, it was use d to s olve for Th as follows: ( 2. 13) where, R n is the net radiation; C p is the specific heat of air at constant pressure; a is the density of air; and r ah is the aerodynamic resistance to heat flow fro m a hy pothetical bare and dry surface (Senay et al., 2013). After determining these hot and cold temperatures, ET could be e stimated. Again several input datasets are required to perform these calculations. Table 2. 3 presents these datasets: Table 2. 3. Lis t of d atasets used to calculate SSEBop ET Dataset Remotely Sensed GDAS Reference ET NO MODIS albedo YES MODIS land surface temperature YES MODIS NDVI YES PRISM air temperature NO PRISM temperature correction coefficient NO SRTM elevation YES 2.5 H y drolo g ical Modeling While the advancements in remote sensing have improved our ability of monitor the and allowed for the development of datasets for individual components of the hydrological cycle, it is not yet possible to monitor the e n tire h ydrological model for any given region. Therefore, hydrological models are often used to simulate all components of the hydrological cycle. The use of the model is also an inexpensive, effective, and fast alternative to 31 extensive environmental monit o ring, which can be used to test as many scenarios as are desired by either researchers or policymakers . 2.5.1 Soil and Water Assessment Tool One of the more common hydrological models is the Soil and Water Assessment Tool or SWAT ( Neitsch et al., 2011 ). SWAT i s a semi - distributed physically based watershed scale model developed by the USDA Agricultural Research Servi ce and Texas A&M AgriLife Research that utilizes several layers of data, such as topography, soil characteristics, landcover , and climatolog i cal d a ta, to simulate the natural environment ( Neitsch et al., 2011 ). There have been many peer - reviewed publicatio ns that have used SWAT models to evaluate different components of the hydrological cycle (Sun et al., 2014; Markovic and Koch, 2015; Verma e t al., 2015; Cuceloglu et al., 2017; Saha et al., 2017). In order to simulate the hydrological cycle in a region, the SWAT model utilizes a water balance which can be seen below (Eq. 2. 1 4 ): ( 2. 1 4 ) where, SW t is the final soil water content, SW 0 is the initial soil water content on day i , t is the time in days, R day is the amount of precipitation on day i , Q surf is the amount of surface runoff on day i , E a is the amount of evapotranspiration on day i , w seep is the amount of water entering the val ose zone from the soil profile on day i , and Q gw is the amount of return flow on day i (Neitch et al., 2011). Each of these components is then either provided as in input or calculated based on various equations and relationships. The following sections de scribe the equations, models, and relationships utilized by the SWAT model concerning surface runoff, evapotra nspiration, soil water, and groundwater. 32 2.5.1.1 Surface Runoff Equations T he SWAT model can utilize two different techniques: 1) the Soil Conservation Service (SCS) curve number and 2) the Green and Ampt infiltration method (Neitch et al., 2011). The SCS curve number method is an empirical model that describes rainfall - runoff re lationships for a variety of different landuses and soils, and can be calculated with the following equation (Eq. 2.15 ): ( 2.15 ) where, Q surf is the runoff, R day is the daily rainfall, I a is the initial a bstractions such as surface storage, interception, and soil infiltration before runoff occurs and is often assumed to be 0.2 S , and S is the retention parame te r which is based on local characteristics such as soil properties, landuse, and slope and is calcu lated with Eq. 2.16 (Neitch et al., 2011). ( 2.16 ) where , CN is th e curve number which is dependent on the soil properties and can be adjusted by the user to better match local characteristics (Neitch et al., 2011). Meanwhile, the Green and Ampt infiltration method calculates surface runoff by first determining how much water infiltrated into the soil and then considering all rainfall over that amount to be runoff . The amount of infiltration that occurs is calculated with the following equation (Eq. 2.17 ): ( 2.17 ) wh ere, f inf is the infil tration rate for a given time t , K e is the effective hydraulic conductivity, is the wetting front matric potential, is the change in vol umetric moisture content across the wetting front, and F inf is the cumulative infiltration for a given time t ( Neitch et al., 2011 ). Here 33 again the curve number is used to adjust the equation for local characteristics by influencing the calculation of K e , which can be seen in Eq 2.18 . ( 2.18 ) where , K sat is the saturated hydraulic conductivity and CN is the curve number ( Neitch et al., 2011 ). In addition to these two techniques for calculating surface runoff, the SWAT model also calculates the peak runoff whi ch provides a measurement of how erosive runoff from a storm is to a region and takes into account time of concentration and rainfall intensity and is calculated by using the following equation: ( 2.19 ) where, q peak is the peak runoff rate, a tc is the fraction of daily rainfall that occurs during the time of concentration, Q surf is the surface runoff, Area is the area of the region, and t conc is the time of concentration for the region ( Neitch et al., 2011 ). Table 2.4 lists the parameters and their definitions within the SWAT model that affect the surface runoff calculations. 34 Table 2.4 . A list of the parameters used in SWAT surface r unoff calc ulations Parameter Definition CH_K(1) Effective hydraulic conducti v ity CH_L(1) Longest tributary channel length in subba si n CH_N(1) CH_S(1) The a verage slope of tributary channels CH_W(1) The a vera g e width of the tributary channel CLAY Percent clay content CN2 Moisture condition II curve number CNCOEF Weighting coefficient used to calculate the retention coefficient for daily curve number calculations dependent on plant evapotranspiration CNOP M o i s ture cond ition II curve number HRU_FR The f raction of total subba si n area contained in HRU HRU_SLP Aver ag e slope steepness ICN Daily curve number calculation method IDT Le n gth of the time step IEVENT Rainfall, runoff, routing option OV_N n value for the overland flow PRECIPITATION Precipitation during time step SAND Percent sand content SLSUBBSN Average slope length SOL_BD Moist bulk density SOL_K The s aturated hydraulic conductivity of the first layer SUB_KM Area of the subba si n i n k m 2 SURLA G Surface runoff lag coefficient 2.5.1.2 Evapotranspiration Equations In order to simulate evapotranspiration, the SWAT model has to take into account a variety of different factors including canopy storage, potential evapotranspiration, and a ct ual evapo transpiration (Neitch et al., 2011). Regarding canopy storage, or the amount of rai nfall technique was selected. If the SCS curve number is being used , cano p y storage i s considered as part of the initial abstractions; however, it the Green and Ampt te chnique is being used an additional calculation for canopy storage is needed (Eq. 2.20 ) (Neitch et al., 2011). ( 2.20 ) 35 where, can day is the amount of water trapped by the canopy, can mx is the amount of water that can be trapped when the canopy if fully developed, LAI is the leaf area index for a given day, a nd LAI mx is the maximum leaf area index for the given landuse (Neitch et al., 2 011). This value is important in calculating evapotranspiration, which regardless of the surface runoff technique the first step is calculating potential evapotranspiration. In the SWAT model, three different methods for calculating potential evapotranspir ation are available, namely the Penman - Monteith method, the Priestley - Taylor method, and the Hargreaves method (Neitch et al., 2011). Each of these techniques requires different inputs, with Penman - Monteith being the most complex requiring solar radiation, air temperature, relative humidity, and wind speed; Priestly - Taylor requiring solar radiation, air temperature, and relative humidity; and Hargreaves being the simplest requiri ng only air temperature (Neitch et al., 2011). Eqs 2.21 , 2.22 , and 2.23 are use d by SWAT to calculate potential evapotranspiration via the Penman - Monteith method, the Priestley - Taylor method, and the Hargreaves method, respectively. (2.21 ) where, is the latent heat flux density, E is the depth rate evaporation, is the slope of the saturation vapor pressure - temperature curve H net is the net radiation, G is the heat flux density to the ground, is the air density, c p is the specific heat a t constant pressure, is the saturation pressure of air at height z , e z is the water pressure of air at height z , is the psychrometric constant, r c is the plant canopy resistance, and r a is the diffusion resistance of the air layer ( Neitch et al., 2011 ). It is important to note that the SWAT model uses t he Penman - Monteith method by default, however, this can be cha nged by the user. ( 2.22 ) 36 where, is the latent heat of vaporization, E 0 is the potential evapotranspiration, is a coefficient, is the slope of the saturation vapor pressure - temperature c urve, is the psychrometric constant, H net is the net radiation, and G is the heat flux de nsity to the ground (Neitch et al., 2011). It is important to note that the Priestly - Taylor method assumes that advection is low, which makes it less ideal for semi arid or arid regions for which it will underestimate potential evapotranspiration ( Neitch et al., 2011). ( 2.23 ) where, is the latent heat of vaporization, E 0 is the potential eva potranspiration, H0 is the extraterrestrial radiation, Tmx is the maximum air temperature for a given day, Tmn is the minimum air temperature for a given day, and is the average temperature for a given day ( Neitch et al., 2011 ). After potential evapotranspiration is calculated , the SWAT model can then calculate actual evapotranspiration . This is done by taking into account the potential evapotranspiration method and value in addition to the evaporation of intercepted rainfall, transpiration, and sublimation and evaporation from the soil (Neitch et al., 2011). Evaporation of intercepted rainfall describe t he evaporation of water found in canopy storage and is dependent on the level of potential evapotranspiration possible and the amount of rainfal l for a given day. If potential evapotranspiration is less th an or equal to the initial water storage the actual evapotranspiration is equal to the potential evapotranspiration (Neitch et al., 2011). However, if the potential evapotranspiration is greater than the initial water storage, actu a l evapotranspiration exhausts the water held in the canopy before moving on to the plants and soil (Neitch et al., 2011). The transpiration calculation utilized by the SWAT model is dependent on the potential 37 evapotrans piration technique used. If the Penman - Monteith method is used , transpiration is already calculated ; however, if any other potential evapotranspiration technique is selected, transpiration is calculated as follows (Eq. 2.24 ) (Neitch et al., 2011): ( 2.24 ) where, E t is the maximum transpiration, is the potential evapotranspiration adjusted for evaporation of free water in the canopy, and LAI is the leaf area index. M eanwhile , sublimation and evaporation from the soil is calculated based on the following equation (Eq. 2.25 ): ( 2.25 ) where, E s is the maximum sublimation/soil evaporation for a specific day, is the potential evapotranspiration adjusted for evaporation of free water in the canopy, and cov sol is the soil cover index ( Neitch et al., 2011 ). Therefore, the final calculation of actual evapotranspiration is the sum of Eqs. 11 and 12. Table 2.5 l ists the parameters and their definitions within the SWAT model that affect the evapotranspiration calculations. Table 2.5 . A list of the parameters used in SWAT evapotranspiration calculations Parameter Definition CANMX Maximum canopy storage CO2 Carbo n dioxide concentration ESCO Soil evaporation compensation coefficient FRGMAX The f raction of maximum leaf conductance achieved at the vapor pressure deficit specified by VPDFR GSI Maximum leaf conductance IPET Potential evapotranspiration method MAX T EMP Daily maximum temperature MIN TEMP Daily minimum temperature VPDFR Vapor pressure deficit corresponding to value given for FRGMX WND_SP Daily wind speed 38 2.5.1.3 Soil Water Equations In order to simulate soil water or the movement of water though t he soil layers , the SWAT model has to take into account a variety of different factors including soil structure, percolation, bypass flow, perched water table, and lateral flow (Neitch et al., 2011). Soil properties are supplied to the SWAT model though u s er input from which the SWAT model is able to determine several characteristics such as density and soil composition. This allows the SWAT model to more accurately replicate soil water content and how water would move through the soils for the region of i n terest (Neitch et al., 2011). Meanwhile, percolation or the movement of water f rom one layer of soil to another, is determined through the use of a couple of equations. First , the volume of water available for percolation is calculated th r ough the followi n g set of equations : ( 2.26 ) where, is the drainable volume of water in the soil layer for a specific day, SW ly is the water content of the soil layer in question for a given day, and FC ly is the water content of the soil layer at field capacity ( Neitch et al., 2011 ). After determining the amount of water that is present the following equat i on is used to determine how much water actually transfers to the next layer of soil down: ( 2.27 ) where, is the amount of water percolating to the underlying soil layer for a given day, is the drainable volume of water in the soil layer for a specific day, is the length of the time step, and TT perc is the travel time for percolation ( Neitch et al., 2011 ). 39 Bypass flow is a condition caused by the swelling and shrinking of soils, most commonly Vertisols, which results in deep cracks in the s urface of the soil that can promote soil water movement ( Neitch et al., 2011 ). SWAT handles thee soils be calculating the volume of the crack within the soil and then using that volume as a componen t in surface storage calcu la tions . The equation used to determine this volume is as follows: ( 2.28 ) where, crk ly,i is the initial crack volume calculated for the soil layer on a given day expressed as a depth, crk max,ly is the maxim um crack volume possible for the soil layer, coef crk is an adjustment coefficient for crack flow, FC ly is the water content of the soil la yer at field capacity, and SW ly is the water content of the soil layer in question for a given day ( Neitch et al., 201 1 ). SWAT provides users the ability to define a perched water table, which happens in the region with a high seasonal water table. This r esults in ponding within the soil layers and affects the downward movement of water through the soil columns . To calcu late the height of the perched table, SWAT utilizes the following equation: ( 2.29 ) where, h wtbl is the height of the water table, SW is the water content of the soil profile, F C is the water content of the soil profile at field capacity, POR is the porosity of the soil profile, is the air - filled porosity expressed as a fraction, and depth imp is the depth to the impervious layer ( Neitch et al., 2011 ). The final compon ent of soil water calculations for the SWAT model is lateral flow, which describes the horizontal movement of wat er with in the soil column. SWAT utilizes and 40 kinematic storage model for subsurface flow to simulate this process which is shown in Eq 2.30 ( N eitch et al., 2011 ). ( 2.30 ) where, Q la t is the lateral flow, is the drainable volume of water in the soil layer for a specific day, K sat is the saturated hydr aulic conductivity, slp is the slope of the region, and L hill is the hill slope length ( Neitch et al., 2011 ). Table 2.6 lists the parameters and their definitions within the SWAT model that affect the soil water calculations. Table 2.6 . A list of the param eters used in SWAT soil water calculations Paramete r Definition CLAY Percent clay content DEP_IMP Depth to the impervious layer DEPIMP_BSN Depth to the impervious layer GDRAIN Drain tile lag time HRU_SLP The a verage slope on the subba si n IC RK Bypass flow code IWATABLE High water table code LAT_TTIME Lateral flow travel time SLSOIL Hillslope length SOL_AWC Available water capacity SOL_BD Bulk density SOL_CRK Potential crack volume for soil profile SOL_K Saturated hydraulic conductivity 2.5.1. 4 Groundwater Equations In order to simulate groundwater movement and storage, the SWAT model has to take into account shallow and deep aquifers (Neitch et al., 2011). Shallow aquifer s are groundwater systems that contribute water to the local riv ers and l a kes, while deep aquifers can contr ibute water t o regions outside of the subba si n or local area (Neitch et al., 2011). SWAT simulates shallow aquifers with the following water balance (Eq. 2.31 ): ( 2.31 ) 41 where, aq sh,i is the water stored in the shallow aquifer on day i , aq sh,i - 1 is the water stored in the shallow aquifer on the previous day, w rchrg,sh is recharge during day i , Q gw is the groundwater flow into the w revap is the amount of water moving up into the soil layers on day i , and w pump,sh is the amount of water pumped out of the shallow aquifer on day i (Neitch et al., 2011). Each of these components can be further described by additi o nal equations which are provide d below . T he recharge to the shallow aquifer or the water that enters the aquifer for any given day is calculated with the following equation (Eq. 2.32 ): ( 2.32 ) where, w rchrg,i is the amount of water recharge entering the aquifer on day i , is the delay time or drainage time of the overlaying geol ogic formations, w seep is the total amount of water exiting the soil layers and entering the aquifer, and w rchrg,i is the previo us days recharge (Neitch et al., 2011) . Groundwater flow or base flow, describes the water that leaves the shallow aquifer and reenters the main channel of the region, and in the SWAT model can be calculated for both steady - state (Eq. 2 .33 ) and non - steady - state (Eq. 2 .34 ) conditions: (2 .33 ) where, Q gw is the groundwater flow, K sat is the hydraulic conductivity of the aquifer, L gw is the distance from the ridge or subbasin divide for the g r oundwater system to the main channe l, and h wtbl is the water table height (Neitch et al., 2011). ( 2.34 ) 42 where, Q gw ,i is the groundwater flow on day i , Q gw,i - 1 is the groundwater flow on the previous day, is the baseflow recession constant, is the time step, w rchrg,sh is the amount of recharge occurring on day i , aq sh is the amount of water stored in the shallow aquifer at the beginning of day i , and aq shthr,q is the threshold water level on the shallo w aquifer for groundwater contribution to the main ch annel to occur ( Neitch et al., 2011 ). Revap describes the water in the shallow aquifer that moves upward into the so i l column to fill unsaturated zones, which for the SWAT model is modeled as a function of water demand for evapotranspiration and utilizes the following set of conditional equations (Eq. 2 .35 ): (2 .35 ) where, w revap , is the actual amount of water moving into the soil layers, is the revap coefficient, E o is the potential evapotranspiration, aq sh is the amount of water stored in the shallow aquifer at the beginning of the day, and aq shthr,rvp is the threshold water level in the shallow aquifer for revap to occur (Neitch et al., 2011). Regarding deep aquifers, SWAT simulates deep aquifers with the following water balance equation : (2 .36 ) where , aq dp,i i s the amount of water stored in the deep aquifer on day i , aq dp ,i - 1 is the previous seep aquifer, w deep is the amount of water percolating from the shallow aquifer to the deep aquifer, and w pump , dp is the amount of water being pu mped form the deep aquifer (Neitch et al., 2011). Of these terms w deep is calculated using the following equation (Eq. 2 .37 ): (2 .37 ) 43 where, w deep is the amount of water percolating from the shallow aquifer to the deep aquifer, is the aquifer percolation coefficient, and w rchrg is the amount of recharge entering both shallow and deep aquifers for a given day ( Neitch et al., 2011 ). Table 2.7 lists the parameters and their definitions within the SWAT model that affect the groundwater calculations. Table 2.7 . A list of the parameters used in SWAT groundwater calcul ations Parameter Definition GW_DELAY Delay time for aquifer recharge GWQMN Threshold water level in shallow aquifers for base flow ALPHA_BF Baseflow recession constant REVAPMN Threshold water level in shallow aquifers for revap GW_REVAP Revap coeffic ient R C HRG_DP Aquifer percolation coefficient GW_SPYLD Specific yield of the shallow aquifer 2.5. 2 Model Calibration While SWAT model applications are varied, one vital step in the model development process is calibration and validation. In fact, this is a n e eded step for all hydrological models since it insures that the model is able to capture local variabilities ( Santhi et al., 2001; White and Chaube y, 2005; Sahoo et al., 2006; Troy et al., 2008; Arnold et al., 2012). During this process , SWAT model output s are compared to collected observed data and the ability of the model to replicate the observed data is determined through the use of statistical c riteria. For SWAT models there are three main criteria that are recommended for use, namely Nash - Sutc liffe e f ficiency (NSE) which represented the ratio of residual variance to the actual data variance, percent bias (PBIAS) which measured the tendency of si mulated results to be larger or smaller than observed values, and the ratio of root - mean - square error to obs e rved standard deviation ratio (RSR). These statistical criteria were initially recommended by Moriasi et al. (2007) with the following ranges for s atisfactory model calibration and validation, NSE >0.5, PBIAS ±25%, and RSR <0.7. This goes to show t he SWAT model performance is limited by the availability of 44 reliable data . Which means that hydrological model development suffers from the same issues tha t monitoring water resources has. 2.5. 3 Remote Sensing in Hydrological Modeling One approach to addr essing t he issues of data availability and reliability for hydrological modeling is the use of remotely sensed data ( Schuurmans et al., 2003; Xu et al., 20 14). As discussed previously, remote sensing provides a source of continuous, spatially distributed d ata tha t can be used for regional analysis. This makes remote sensing data ideal for use in hydrological modeling. Nevertheless , there are still limitations to the use of remotely sensed data such as the spectral, spatial, and temporal resolutions of the c ollecte d images ( Lillesand et al., 2014 ). Howeve r, as long as these limitations are taken into account , it is possible to develop reliable datasets that can be incorporated into hydrological models (Xu et al., 2014). In fact in recent years several studies have l o oked at the use of remotely sensed ET da ta in the hydrological model calibration process (Immerzeel and Droogers, 2008; Schuurmans et al., 2011; Qin et al., 2013; Sousa et al., 2015; Mendiguren et al., 2017). In the study by Immerzeel and Droogers (2008) b i - weekly actual evapotranspiration (ETa) data, obtained from the Surface Energy Balance Algorithm (SEBAL), were integrated into the calibration of a SWAT model. This calibration process modified SWAT parameters that were related to land use soil ch aracter i stics, groundwater, and weather (Immerze el and Droogers, 2008). The results of this study showed that the incorporation of remotely sensed data could significantly improve the model calibration process and result in more accurate model ETa simulati ons (Im m erzeel and Droogers, 2008). In the study by Schuurmans et al. (2011) SEBAL ETa datasets derived from data collected by two different satellites (ASTER and MODIS) were integrated into a coupled groundwater and unsaturated zone model (MetaSWAP) to es timate s oil moisture. The result of 45 this study s howed that the inclusion of the remotely sensed ETa data was able to improve the spatial simulation of soil moisture levels (Schuurmans et al., 2011) This not only shows how remotely sensed data c ould improve the mo d eling process but also the interconnecte d nature of the hydrological cycle. In the study by Sousa et al. (2015) an ETa dataset based on MODIS imagery was developed and incorporated into a SWAT model. The results of this integration showed that by a dding t h e remotely sensed ETa, the SWAT model ha d improved streamflow estimates, especially in ungagged catchments (Sousa et al., 2015). This again shows that the addition of remotely sensed data in the model calibration process is quite beneficial. In the study b y Mendiguren et al. (2017) remotely sens ed ETa was used to improve the simulation of spatially distributed ETa. Results from this study indicated that the use of remotely sensed ETa was able to improve model simulations of the spatially distributed ETa fo r the region (Mendiguren et al.; 2017). T his again highlights the benefits of including remotely sensed data in hydrological model development. All of these studies show that the incorporation of remotely sensed data can improve the overall hydrolog ical mo d el performance. However, very few studie s consider a multi - objective calibration approach during the model calibration phase. Instead , most studies focus on a single component of the hydrological model during the calibration process (Immerzeel and Drooger s , 2008; Schuurmans et al., 2011; Sousa e t al., 2015; Mendiguren et al.; 2017). However, studies that have considered several hydrological components during the calibration process indicate that adding a multi - objective calibration can improve overa ll mode l performance and reduce the uncertainty associated with the final models (Crow et al., 2003; Rajib et al., 2016; Franco and Bonumá, 2017) However, no studies compare the applicability of different calibration techniques when performing a multi - obje ctive c a libration. This shows that there is a ne ed to perform further research in this area. 46 2.6 Modeling Uncertainty While hydrological models and remote sensing data allow for region - wide analysis and monitoring, it is important to note that these techni ques ha v e increased levels of error and uncertainty compared to monitoring stations. These errors and uncertainties are often grouped into three main categories, namely data uncertainty, model structure uncertainty, and parameter uncertainty (Jin et al., 2 010; Br i gode et al., 2012; Zhang et al., 2016). The following sections describe these categories in more detail. 2.6.1 Data Uncertainty Data uncertainty is a way to quantify the amount of noise within a dataset (Jin et al., 2010). This can be caused by a v ariety o f sources from environmental fac tors to the limitations of data collection equipment (Benz et al., 2004). This can have a major impact on models since they are dependent on the quantity and quality of input data. And any noise or uncertainty within the da t aset will be passed into the mod el outputs as the data is used in different calculations. This is of particular importance to remotely sensed data, which needs to account for noise from sources such as surface properties (topographic variability an d land s urface directional reflectance p roperties), atmospheric effects (spatial and temporal variations), and sensor design (spectral, spatial, and radiometric properties) (Kustas and Norman, 1996; Friedl et al., 2001; Long et al., 2014). For example, whe n consi d ering remotely sensed evapotrans piration datasets, uncertainty caused by variability surface properties (landcover type) c ould result in inaccurate evapotranspiration datasets, which would increase the uncertainty of any hydrological model that use s this e vapotranspiration dataset as an input (Long et al., 2014; Yang et al., 2015). One way to address this would be to perform accuracy assessments by comparing the evapotranspiration products to different land - based evapotranspiration station data for differe n t landcover types. In 47 fact, seve ral studies have focused solely on this task (Kim et al., 2012; Senay et al., 2014; Xia et al., 2015; Bhattarai et al., 2016). The results of these studies provide a look into the overall accuracy of different remote ly sens e d evapotranspiration datasets. T his allows researchers, policy makers , and stakeholders to make educated decisions about which datasets to use for further analysis based on their own ranges of acceptable uncertainty. 2.6.2 Model Structure Uncertai nty Mod e l structure uncertainty is a way (Brigode et al., 2012). Due to the complexity of natural systems, simplifications are used to streamline models. However, it is possible to oversimplify a model, which increas e s uncertainty associated with it by ignoring key factors and interconnected processes within the environment (Refsgaard et al., 2006; Qin et al., 2013). In fact, this has been identified by many studies as a major source of uncertainty (Usunoff et al., 19 9 2; Dubus et al., 2003; Linkov an d Burmistrov, 2003; Brigode et al., 2012). However, it is often challenging to reduce this uncertainty without developing a new model. Refsgaard et al. (2006) reviewed a variety of strategies for assessing model stru cture u n certainties and proposed a six - step protocol to examine conceptual uncertainty. These steps are: 1) formulate a conceptual model; 2) set up and calibrate the model; 3) repeat steps 1 and 2 until a sufficient number of conceptual models were develop ed ; 4) p erform validation tests and acc ept/reject models; 5) evaluate the tenability and completeness of remaining conceptual models; and 6) make model predictions and assess uncertainty (Refsgaard et al., 2006). This approach allows researchers to select the bes t model possible for each study and insure that the model used captures the necessary processes of the system being modeled . 48 2.6.3 Parameter Uncertainty Parameter uncertainty is used to describe how well model parameter values perform when simulat ing mod e l outputs (Brigode et al., 2012 ). However, minimizing this uncertainty is often challenging since hydrological models require a large number of parameters to simulate the complexity of hydrological systems. To address this , model calibration is the first s tep in model development in whi ch parameter values are altered in an attempt to better improve ability to represent the conditions in the area of study. The calibration process compares simulated model outputs to observed data and uses statist i cal analysis to determine how c lose the datasets are to each other (Immerzeel and Droogers, 2008; Golmohammadi et al., 2014). Within hydrological modeling, three statistical criteria are often used to determine if a model was successfully calibrate d, name l y Nash - Sutcliffe model efficien cy coefficient (NSE), root - mean - squared error - observations standard deviation ratio (RSR), and percent bias (Pbias) (Moriasi et al., 2007). However, while NSE, RSR, and Pbias can be used to determine if the calibratio n was s u ccessful; knowing which paramet ers need to be changed provides a unique challenge of its own. One way to address this would be to perform a sensitivity parameters. This can be done through different software packages such as SWAT - C U P, which allows modelers to per form sensitivity analysis, calibration, validation, and uncertainty analysis of SWAT models based on Sequential Uncertainty Fitting (SUFI2), Particle Swarm Optimization (PSO), Generalized Likelihood Uncertainty Estima tion (G L UE), Parameter Solution (ParaSo l), and Markov Chain Monte Carlo (MCMC) procedures (Abbaspour, 2007). By using SWAT - CUP, it is possible to identify which parameters should be altered as well ensure that the calibration process was successful at redu cing th e model output uncertainties. An other aspect of parameter uncertainty is equifinality , which describes the case in which a model calibration process 49 identifies multiple parameter sets that yield similar model performances (Lu et al., 2009; Jin et al ., 2010 ) . And while this is expected to occur within hydrological modeling calibration (Beven, 1996; Savenije, 2001), it can still impact a model usefulness. One approach that can help reduce the impact of equifinality within hydrological models is the c omplexi t y of the objective function, si nce as objective functions become more comprehensive the chance of having multiple calibrations performing the same is reduced (Abbaspour, 2007). By quantifying and minimizing parameter uncertainties, model performanc e can b e improved , which in turn result s in better model outputs for researchers, policymakers , and stakeholders. 2.7 Summary Overall, advancements in remote sensing technology have resulted in a wide variety of satellite - based sensors that have greatly i mprov ed recent years have seen an increase in the amount of research that utili zes remotely sensed data. In particular, the field of hydrological modeling can be greatly improved by the incorporation of satel lite da t a and the subsequently developed remotely sensed datasets. However, while studies have already shown the benefits of th e incorporation of this data in the area of model calibration; few studies have expanded the use of remotely sensed data to multi - object i ve model calibration. Furthermore, conducting studies that explore the impacts of remotely sensed data on different mul ti - objective hydrological model calibration techniques will advance the field of hydrological modeling and allow for the developm ent of m odels that more accurately simulate the real world. 50 3. INTRODUCTION TO METHODOLOGY AND RESULTS T his thesis is in the f Role of Evapotranspiration Remote Sensing Data in Improvi ng Hyd r ological Modeling modeling. As the global demands for the use of freshwater resources continue to rise, it has become increasingly important to e nsure the su staina b ility of this resource. This is accomplished through the use of management strategies that often utilize monitoring and the use of hydrological models. However, monitoring at large scales is not feasible and therefore model applications are becoming chall e nging, especially when spatially distributed datasets, such as evapotranspiration, are needed to understand the model performances. Due to these limitations, most of the hydrological models are only calibrated for data obtained from site/point obser vation s , such as streamflow. Therefore, the main fo cus of this paper is to examine whether the incorporation of remotely sensed and spatially distributed datasets can improve the overall performance of the model. In this study, actual evapotranspiration (E Ta) da t a was obtained from the two different sets o f satellite - based remote sensing data. One dataset estimates ETa based on the Simplified Surface Energy Balance (SSEBop) model while the other one estimates ETa based on the Atmosphere - Land Exchange Invers e (ALE X I) model. The hydrological model used in thi s study is the Soil and Water Assessment Tool (SWAT), which was calibrated against spatially distributed ETa and single point streamflow records for the Honeyoey Creek - Pine Creek Watershed, located in Mich igan, U SA. Two different techniques, multi - variable (NSGA - II) and genetic algorithm, were used to calibrate the SWAT model. Using the aforementioned datasets , the performance of the hydrological model was evaluated by calculating Nash - Sutcliffe 51 efficiency (NSE), percent bias (PBIAS), and root mean squared error - observations standard deviation ratio (RSR). Evaluating the Spatial and Temporal Variability of Remote Sensing and Hydrologic Model Evapotranspiration Products the sp a tial and temporal performance of eight ETa d atasets. Advances in satellite technology ha ve led to the availability of global remote sensing datasets that can be used to supplement gaps in observed hydrological data. However, it is often challenging to ide n tify the right dataset for different spatial and temporal scales. Therefore, the goal of this paper is to statistically explore the spatial and temporal performance of remotely sensed ETa datasets in a region that lacks observed data. The remotely s ensed d atasets were further compared with ETa resul ts from a physically - based hydrologic model to examine the differences and describe discrepancy among them. All of these datasets were compared through the use of Generalized Least - Square estimations that compar e d ETa datasets on temporal (i.e., monthly an d seasonal basis) and spatial (i.e., landuse) scales at both watershed and subbasin levels. a Many - Objective Optimization Technique to Improve the Performance o f a Hy d rologic Model Using Evapotranspiration Remote Sensing Data we combine streamflow and remotely sensed evapotranspiration data for hydrological model calibration with the goal of identifying the improvement level achieved by introducing spatially ex plicit data. This is simila r to the first study; however, while the first study was limited to just two objective functions (multi - objective) in the calibration process, this study selected an improved technique that allows many - objective (more than two ob jectiv e functions) calibrat ion. Furthermore, while the first study considered two evapotranspiration datasets (ALEXI and SSEBop), this study considers eight evapotranspiration datasets, namely: the USGS Simplified 52 Surface Energy Balance (SSEBop), the USDA/ NASA A t mosphere - Land Exchan ge Inverse (ALEXI), the MODIS Global Evapotranspiration Project (MOD16A2) 500m, the MOD16A2 1 km, the North American Land Data Assimilation Systems 2 Evapotranspiration (NLDAS - 2) Mosaic, the NLDAS - 2 Noah, the NLDAS - 2 VIC, and fin ally T e rraClimate. In addit ion to these datasets, an E nsemble was also developed and used. Regarding the calibration processes, the Non - dominated Sorting Genetic Algorithm, the Third Version (NSGA - III) was linked to SWA T ) to preform ten different calibrati ons. A total of 18 SWAT par ameters were considered during calibrations that impact t he model outputs in regard to both streamflow and evapotranspiration. The first eight calibrations utilized a multi - objective approach and used observed streamflow and an e vapotr a nspiration dataset a s the objective functions. The ninth calibration was another multi - objective calibration utilizing observed stream flow and the evapotranspiration E nsemble. And finally, the tenth calibration was a many - objective calibration util izing o bserved stream flow and all of the evapotranspiration datasets. Again, NSE, Pbias, and RSR were used as the statistical calibration criteria and a measure of the overall model performance . 53 4. EVALUATING THE ROLE OF EVAPOTRANSPIRATION REMOTE SENS ING DA T A IN IMPROVING HYDROLOGICAL MODELING PREDICTABILITY 4.2 Introduction As extreme climate conditions and anthropogenic activities continue to impact environmental systems, mitigation and restoration related projects have become common. Furthermore, en vironm e ntal systems, such as watersheds, are very complex with many relationships and interlocking processes ( Sivakumar and Singh, 2012; Guerrero et al., 2013 ). Therefore, it can be challenging to determine which management solution(s) should be selected a nd imp l ement ed (Herman et al., 2015; Sabbaghian et al., 2016). This has led to the development of many different modeling techniques that can simulate a variety of options and identify the best solution(s), based on the criteria put forth mostly by stakeho lders a nd po licy makers (Chen et al., 2012; Beven and Smith, 2014; Giri et al., 2016). Meanwhile, the first step in a model implementation is parameter calibration. Parameter calibration in model applications is used to adjust model performance to better simula t e the natural systems they are trying to describe ( Guerrero et al., 2013; Zhan et al., 2013; Rajib et al., 2016 ). While parameter calibration improves the ability of models to more accurately represent are still limit e d by the quality and quantity of input data and their availabilities (Nejadhashemi et al., 2011). Today, most hydrological studies rely on data collected at monitoring stations across the world. In fact, the United States Geological Survey (USGS) ha s abou t 1.5 million monitoring sites from which data can be obtained (USGS, 2016a). However, even with the existence of all these monitoring sites, there are times where higher spatial resolutions are needed by researchers, stakeholders , and policymakers t o more preci sely evaluate the hydrologic conditions and to determine the best place to implement 54 mitigation and restoration projects ( Wanders et al., 2014 ). One way to address this issue is the use of remotely sensed data. Remote sensing is defined as the scienc e of i dentifying, observing, and measuring an object without physical contact (Graham, 1999). With the advancements in satellite technology, remotely sensed satellite data has become a source of consistent monitoring for the entire globe, with applic ations rangi ng from crop yields to water resources assessments (Graham, 1999; Long et al., 2014 ). In order to model water resources more accurately, it is important to examin e different components of the hydrologic cycle, including water movement process es ( e. g . , ev aporation and streamflow) and water storage ( e.g. , soil moisture, water vapor, groundwater, and surface water bodies). While hydrological models simulate all components of the hydrological cycle, streamflow is often the only component that the model o utput s are compared against during the calibration process since it can be measured more accurately than the other components (Immerzeel and Droogers, 2008; Wanders et al., 2014; Rajib et al., 2016 ). This can result in poor simulations of other hydr ologic compo nents, which ultimately lowers the model performance ( Wanders et al., 2014; Rajib et al., 2016 ). Therefore, including additional hydrological components in the parameter calibration process could allow the model to better represent all process occurr i ng in the environment (Crow et al., 2003). In particular, evapotranspiration (ET) could be considered an important hydrological component added to the calibration process since it describes the moisture lost to the atmosphere from both biotic ( e.g. , plant s ) and abiotic ( e.g. , soils) sources ( Hanson, 1991; USGS, 2016 d ). Meanwhile, ET plays a major role in the cycling of water from land and ocean surface sources into the atmosphere, which in turn drives precipitation (Pan et al., 2015). Furthermore, Im merzee l and Droogers (2008) found that calibrating a hydrological mod el for ET significantly improved ET simulations; and that ET 55 simulation values were more sensitive to groundwater and meteorological parameters compared to soil and landuse parameters. T his in d icates that including additional parameters in a model calibra tion can improve the overall model performance. However, the applicably of different calibration techniques has not been explored when both remotely sensed ET and streamflow data are invo lved. I n addition , this study is unique in the sense that the perform ance of a hydrologic model for estimating streamflow was evaluated using different remotely sensed ET products. Therefore, the objectives for this paper are to (1) determine the performan ce of a calibrated hydrologic model in estimating ET against spatiall y distributed time series ET products obtained from remote sensing; (2) determine the impact of ET parameter calibration on streamflow estimation; and (3) evaluate the performances of dif ferent calibration techniques for streamflow and ET estimations. 4.3 Materials and Methods 4.3.1 Study Area The study area is the Honeyoey Creek - Pine Creek Watershed (Hydrologic Unit Code 0408020203), which is located within the Saginaw Bay Watershed in M ichiga n Peninsula (Figure 4. 1). The US Environmental Protecti on Agency (EPA) identified the Saginaw Bay Watershed as an area of concern due to the presence of contaminated soils and degradation of fisheries within the region (EPA, 2017). These cond itions were caused by the addition of both point and non - point source pollutants from a variety of sources such as industrial waste and agricultural runoff (EPA, 2016). The final outlet for this watershed is Lake Huron via the Saginaw River. Out of the app roxima t ely 1,100 km 2 within the Honeyoey Watershed, agriculture is th e dominant landuse (~52%) followed by forests (~23%), wetlands (~17%) and pasturelands (~5%). The remaining land is classified as urban (~3%). The Honeyoey Creek - Pine Creek 56 watershed has been s i gnificantly altered by anthropogenic activities as evidenced b y the landuse change (agricultural lands and urban area are dominant in the region), which in turn impacts the natural environment, especially water quality and quantity. Figure 4.1. T he stu d y area ( Honeyoey Creek - Pine Creek watershed) 4.3.2 Data Collection 4.3 .2.1 Physiographic Data Several spatial and temporal input datasets were needed to describe the study area in a hydrological model. These datasets describe characteristics such as topog r aphy, landuse, soil properties, climate, and crop manage ment practices. Data from the USGS were obtained to represent the topography of the region using their 30 m spatial resolution National Elevation Dataset (NED, 2014). Landuse information was ac quired from the 30 m spatial resolution Cropland Data Layer developed by the United States Department of Agriculture - National Agricultural Statistics Service (USDA - NASS) (NASS, 2012). The Natural Resources 57 Conservation Service (NRCS) Soil Survey Geographic (SSUR G O) Database was used to describe the soil properties for the region at a scale of 1:250,000 (NRCS, 2014). National Climatic Data Center (NCDC) weather stations (two precipitation stations and two temperature stations) were used to obtain daily preci pitati o n and temperature data for the time span of 2003 to 2014. A widely used stochastic weather generator called WXGEN was employed (Sharpley and Williams, 1990; Wallis and Griffiths, 1995), which is embedded in the Soil Water Assessment Tool (SWAT), to create climate time series for other climatological records ( e.g. relative humidity, solar radiation, and wind speed) that are required for SWAT to operate (Neitsch et al., 2011). Predefined crop management operations, schedules, and rotations were adopted from p revious studies performed in the same region (Love and Nejadhashemi, 2011; Giri et al., 2015). Due to the limitation of SWAT in simulating up to 250 different landuse, the subwatershed map that was provided by the National Hydrology Dataset P lus (NH DPlus) and the Michigan Institute for Fisheries Research at a scale of 1:24,000 were modified to accommodate this limitation (Einheuser et al., 2013) . 4.3 .2.2 Remote Sensing Data In order to evaluate the role of ET remote sensing data in improving a hyd rologi c model predictability, two satellite - based ET datasets were obtained for the period of 2003 to 2014 for the study area. One dataset was created based on the Simplified Surface Energy Balance (SSEBop) model while the other one was based on the Atmosp here - L a nd Exchange Inverse (ALEXI) model. The USGS dataset reported monthly actual evapotranspiration (ETa) using the SSEBop model (Senay et al., 2013). ETa is limited by the amount of water present at a site since it refers to the actual amount of water that i s lost through both evaporation and transpiration (NOAA, 58 2017 b ). This model utilizes ET fractions derived from 1 km Moderate Resolution Imaging Spectroradiometer (MODIS) thermal imagery collected every eight days to develop a 1 km monthly ETa dataset for t h e Conterminous U.S. (Senay et al., 2013; Velpuri et al ., 2013). Data were obtained from this dataset for each subwatershed in the study area. In order to provide an were a v eraged with respect to the area to generate the overall area weighted ETa average values for each month (USGS, 2016 o ). The second ETa dataset is created based on the ALEXI model, which was sponsored by the USDA and US National Aeronautics and Space Admin i stration (NASA). The ALEXI model utilizes remotely sensed morning land surface temperatures to determine ETa by relating the observed change in temperature to changes in surface moisture and ETa ( Anderson et al., 1997; Anderson et al., 2007 ). For th is stu d y, 4 km thermal images were obtained from Geostationary Operational Environmental Satellites (GOES) and used as to develop a daily 4 km ETa dataset for the Conterminous U.S. ( Hain et al., 2015 ). In order to make the second set of ETa data comparable to th e first set, the daily ETa values from the ALEXI model were averaged to create monthly ETa values. Next, these values were averaged for each subwatershed with respect to area. 4.3 .3 Hydrological Model: SWAT The ETa outputs of both the ALEXI and S SEB op m odels were used for the evaluation of SWAT models for the study region. SWAT is a widely used, continuous - time, semi - distributed, hydrological model that was developed by the USDA Agricultural Research Service (USDA - ARS) and Texas A&M AgriLife Resea rch (T e xas A&M University, 2017). By taking into account different spatiotemporal layers of information (Section 2.2.1), such as topography, 59 landuse, and climate, SWAT models are able to simulate a variety of hydrological proces se s , such as runoff, sedimen t t ran s port, and ET ( Gassman et al., 2007 ). This makes it a very useful tool for both researchers and policymakers . 4.3 .4 Calibration Approaches For this study, all of the collected physiographic data was incorporated into a SWAT model. However, there are ma ny d efault parameters in a SWAT model that represent an average or more probable condition that may or may not be true for the region of study (Arnold et al., 2012). Therefore, the SWAT model used in this study underwent a series of calibration and vali dat ion processes. To do this , all observed time series data were divided into calibration (2003 to 2008) and validation (2009 to 2014) periods. This process is simply referred to as calibration in the rest of the paper. Three different types of model calib rat ion were used in this study. The first was solely a streamflow calibration. In this approach, individual SWAT parameters that influence the streamflow calculations were tested to find their near - optimal value through the comparison of simulated streamfl ows to observed streamflows. Observed streamflow data was obtained from a USGS streamflow station on the Pine River at the outlet of the study area (USGS, 2016 p ). The next two calibration approaches, multi - variable and genetic algorithm, were used to impro ve the ETa estimation for the study region. For these sets of calibrations, SWAT parameters used in ETa calculations at the subwatershed level were altered to replicate the values obtained from the ALEXI and SSEBop ETa datasets. In order to examine the rol e o f t h ese remotely sensed data on the performance of SWAT for estimating ETa, the genetic algorithm approach was used since it is able to optimize the system for a single variable. Meanwhile, a multi - variable calibration approach was selected to determine th e i m pact of add ETa calibration on the SWAT model 60 performance for both ETa and streamflow estimation. Detailed descriptions of these calibration approaches are provided below. 4.3 .4.1 SWAT Parameters As mentioned above, during the SWAT model calibratio n, the SWAT parameter values were altered. The selection of these variables was done through the use of literature review and sensitivity analysis (Woznicki and Nejadhashemi, 2012). With respect to streamflow, 15 SWAT parameters were identified and altered durin g the calibration process including: baseflow recession constant (ALPHA_BF), biological mixing efficiency (BIOMIX), maximum canopy storage ( CANMX ), ef the main channel (CH_N2), m oistur e condition II curve number (CN2), plant uptake compensation factor (EPCO), soil evaporation compensation coefficient (ESCO), delay time for aquifer r echarge (GW_DELAY), revap coefficient (GW_REAP), threshold water level in shallow aquifer for base f low (G W QMN), aquifer percolation coefficient (RCHRG_DP), threshold water level in shallow aquifer for revap ( REVAPMN ), available water capacity (SOL_AWC), a nd surface runoff lag coefficient (SURLAG). These parameters were selected based on the information provid e d by the SWAT developer (Arnold et al., 2012). Table 4. 1 presents the minimum, maximum, default, and calibrated values for all of these parameters fo r the Honeyoey watershed. 61 Table 4.1 . Streamflow calibration parameters used in this study Parameter Mini m um Maximum Default Calibrated ALPHA_BF 0 1 0.048 0.55 BIOMIX 0 1 0.2 0.01 CANMX 0 100 0 1 CH_k2 - 0.01 500 0 65 CH_N2 - 0.01 0.3 0.014 0.025 CN2 - 25% 25% NA - 0.22% EPCO 0 1 1 0.37 ESCO 0 1 0.95 0.97 GW_DE L AY 0 500 31 9 GW_REVAP 0.02 0.2 0.02 0.055 GWQMN 0 5000 1000 1000 RCHRG_DP 0 1 0.05 0.35 REVAPMN 0 1000 750 900 SOL_AWC 0 1 NA 20% SURLAG 1 24 4 1 In regards to the ETa calibration, another set of 10 SWAT p a ramet e rs was identified as being influential to the ETa calculations ( Neitsch et al., 2011 ). These included: maximum canopy storage ( CANMX ), carbon dioxide concentration (CO2), soil evaporation compensation coefficient (ESCO), fraction of maximum stomatal condu c tance corresponding to the second point on the stomatal conductance curve (FRGMAX), maximum stomatal conductance (GSI), potenti al evapotranspiration method (IPET), daily maximum temperature (MAX TEMP), daily minimum temperature (MIN TEMP), vapor pre s sure d eficit corresponding to the fraction given by FRGMAX ( VPDFR ), and daily wind speed (WND_SP). However, some of these parameters could not be altered since they were provided by either observed data or the weather generator used in this study, includi n g MAX TEMP, MIN TEMP, and WND_SP. In addition , since climate chan ge was not a factor for this study, CO2 was also not altered. Furthermore, in an attempt to limit the impact of the ETa calibration on streamflow, any SWAT parameters already used in the str e amflo w calibration, CANMX and ESCO, were also not used during the ETa calibration process. This reduced the initial set of ETa parameters from 10 to four. Of this set of four 62 parameters, three are crop properties and have ranges of 0.001 to 0.1 for GSI, 0 to 1 f or FRGMAX, and 1.5 to 6 for VPDFR . The last parameter used in this study, IPET, indicates which method to use when calculating potential evapotranspiration (ETp). Within SWAT three different ETp methods are available: namely the Penman - Monteith meth o d, th e Priestley - Taylor method, and the Hargreaves method ( Neitsc h et al., 2011 ). All three methods were included in the ETa calibration process; however, it was found that the Penman - Monteith method produced the best results for the study area. 4.3 .4.2 I n itial Streamflow Calibration A streamflow calibration was perform ed to generate a base condition to which the ETa calibrations could be compared . In order to evaluate the performance of a hydrological model, three statistical criteria that were suggested b y Mor i asi et al. (2007), were used in this study. These criteria include: 1) Nash - Sutcliffe efficiency (NSE) representing the ratio of residual variance and observed data variance (Nash and Sutcliffe, 1970); 2) Percent bias (PBIAS) evaluating how much lar g er/sm a ller simulated data are than their corresponding observed d ata (Gupta et al., 1999); and 3) Root mean squared error (RMSE) - observations standard deviation ratio (RSR), reporting the ratio of RMSE and standard deviation of measured data (Legates and M cCabe , 1999). For evaluating the performance of a hydrologic mode l on simulating monthly streamflow values, NSE values above 0.5, PBIAS values within ±25%, and RSR values below 0.7 are considered as satisfactory (Moriasi et al., 2007). In addition , we als o repo r ted RMSE to examine the error associated with the simulated data in which lower values represent the better model performance. 4.3 .4.3 Multi - variable Calibration A multi - variable calibration procedure, based on Monte Carlo simulation and an evoluti o nary a lgorithm, was applied to the SWAT model using both remotely sensed ETa 63 datasets and observed streamflow from the study area. The procedure aimed to identify the Pa reto optimal frontier and the best trade - off solution. A solution is classified as Par e to op t imal (also known as non - dominated) when the value of any objective function cannot be improved without decreasing the performance of at least one other objective f unction (Chankong and Haimes, 1993; Tang et al., 2006). In multi - variable calibration, there is at least one objective function per observed variable. For this study, the minimization objective function (OF) for each variable ( i.e. ETa and streamflow) was based on the NSE. (4.1) The objective function fo r ETa w as computed using the area weighted average of the monthly simulated from the hydrologic model and satellite - based ETa time series for each subwatershed, which was determined as follows: (4.2) where, is th e average ETa for month ; is the total surface area of the watershed; is the surface area of subwatershed ; is the ETa for subwatershed and month ; and is the number of subwatersheds. Therefore, one pair of sim u lated - observed ETa series for the whole watershed was obtained to determine a unique NSE for this variable. This process was not employed for st reamflow since there is only one gauging station at the outlet of the study area (Figure 4. 1). The general outl i ne of the multi - variable calibration, which is further explained in the following sections, is as follows: A Monte Carlo simulation is performed to understand the SWAT model performance for ETa and streamflow with respect to the selected calibration 64 param e ters. Thus, 5,000 parameter sets were randomly generated via uniform sampling, which were then evaluated by executing the SWAT model for each ge nerated parameter set. The results were used to define, if possible, narrower calibration parameter ranges, and to ob t ain multi - objective scatter plots to identify preliminarily Pareto Optimal solutions. The next step consists of the application of a multi - objective evolutionary algorithm known as the Nondominated Sorted Genetic Algorithm II (NSGA - II) (Deb et al., 2 002) t o determine the optimal Pareto population. Finally, the decision - making method known as the Compromise Programming (Deb, 2001), using a Eu clidean distance metric, was employed to select the final optimal trade - off solution from the resulting Pareto O ptima l population. 4.3 .4.3.1 Monte Carlo Simulation A total of 5,000 runs for Monte Carlo simulation were performed using MATLAB®, with randomly generated corresponding parameter sets selected from uniform distributions. Ranges for calibration parameters w ere d e fined as follows: 0.001 to 0.1 for GSI, 0 to 1 for FRGMAX, and 1.5 to 6 for VPDFR . A SWAT model run was executed for each parameter combin ation, computing NSE for both ETa and streamflow. Dotty plots relating each OF with parameter values were obtai n ed to analyze parameter identifiability, and if possible, narrower calibration ranges to be explored with the NSGA - II algorithm. Likewise, multi - objective plots relating ETa and streamflow OF values were generated for preliminary Pareto frontiers identifi c ation . 4.3 .4.3.2 Multi - objective Evolutionary Algorithm: NSGA - II The NSGA - II is a multi - objective genetic algorithm that has been widely used in various disciplines and has been successfully implemented in other SWAT applications (Zhang et al., 2010; Lu e t al. , 2014; Zhang et al., 2016). The NSGA - II is a population - base d algorithm that is 65 comprised of a nondominated ranking process, a crowded distance calculation, an elitist selection method, and offspring reproduction operations (Deb, 2001). For this stu d y, a r eal - coded NSGA - II with simulated binary crossover (SBX) and polynomial mutation (Baskar et al., 2015) was applied, requiring the prior definition of distribution indexes for each operation (defined as 20 for crossover and mutation each). Other input param e ters include the population size (defined as 100), the maxim um number of generations as stopping criteria (defined as 50), and the mutation probability (defined as the reciprocal of the number of calibration parameters). 4.3 .4.3.3 Compromise Progra m m ing A p proach The compromise progra m ming approach using the met ric (which becomes the Euclidean distance metric) is used to select the optimal Pareto population member that is closest to a reference point (Deb, 2001). In this case, the ideal point, w hich i s unfeasible and is not located on the Pareto frontier, is sele cted as the reference point and it is comprised by the best objective function values (Deb, 2001). Before computing the distance between each Pareto point and the ideal point, the object i ve fu n ction values are normalized employing a Euclidian non - dimension alization (Sayyaadi and Mehrabipour, 2012): (4.3) where, i is the index for each point in the Pareto frontier, j is the index for each OF, m is the t o tal n u mber of the Pareto population, and n - dimensional The distance between each Pareto point and the ideal point , which is the metric, is calculated as follows: (4.4) where, N den o tes the total number of objective functions. 66 In the compromise programming approach, the point with the minimum distance metric value is chosen as the best trade - off solution. 4.3 .4.4 Genetic Algorithm Calibration The other approach used to calib r a te t h e SWAT models with respect to the ETa datasets was a genetic algorithm (GA). A GA is an optimization technique that imitates biological process to refine a population of potential solutions to identify the best final or set of final solutions (Goldb e r g, 1 9 89; Conn et al., 1991; Conn et al., 1997). For this study, a GA was used to guide ETa calibrations by changing the values of three parameters within the SWAT model, namely GSI, FRGMAX, and VPDFR . These are the same parameters that were modified in t h e mul t i - variable optimization approach, and thus the same ranges were used for this optimization. With each successive set of parameter values, a series of MATLAB ® codes were used to update and run the SWAT model (Abouali , 2017). First, the parameter valu e s wer e accepted by the code, which checked the values to the defined ranges and then applied the values to all subwatersheds within the region. After this was completed , the code executed the SWAT model and stored the out puts for further analysis. In summ a ry, t h e SWAT model was run 904,900 times. While executing these runs will not necessarily develop an ideal model, it will generate a landscape of how ET changes for each subwatershed based on the specified parameters. For each set of parameter values, the SWAT E Ta outputs were compared to the ALEXI and SSEBop datasets and NSE and RMSE were calculated for each subwatershed. The parameter set that had the largest NSE was considered to be the best and the lowest RMSE was used as the tiebreaker. This allowed f o r the identification of the best parameter values for each subwatershed, which then used to parametrize the best model that maximizes the ETa calibration. It should be noted that this is only possible based on the assumpt ion that the ETa calculation for o n e sub w atershed is not affected by the ETa calculation 67 for another subwatershed, otherwise it would not be possible to create the mosaic landscape of parameter values used in the best model, which to the best of our knowle dge has not been done in other SWA T stud i es. Furthermore, after the best parameters for each subwatershed were identified and applied within the SWAT models, the simulated ETa values were area averaged to produce a single ETa value for the entire watershed . This set of ETa values was then u sed t o calculate the NSE, PBIAS, RSR, and RSME for the entire region, just like was done in the multi - variable calibration. This was done to allow for a watershed level evaluation of the calibration approaches. 4.3 .5 Statistical Analysis To further evalu a te th e streamflow and ETa outputs from the calibrated models and ETa datasets, a mixed - effects model was used to compare the mean difference between each of the outputs (Kuznetsova et al., 2015). This process was perfo rmed twice, once for the streamflow d a taset s (observed, initial streamflow calibrated model, ALEXI multi - variable calibrated model, ALEXI genetic algorithm calibrated model, SSEBop multi - variable calibrated model, and SSEBop genetic algorithm calibrated mo del) and once for the ETa datasets (A L EXI, S SEBop, ALEXI multi - variable calibrated model, ALEXI genetic algorithm calibrated model, SSEBop multi - variable calibrated model, and SSEBop genetic algorithm calibrated model). This allowed for the determination o f significant mean differences betwee n the d atasets with a 95% confidence level. 4.4 Results and Discussion 4.4 .1 Initial Streamflow Calibration Daily streamflow was calibrated and validated for a 12 - year period (6 years calibration and 6 years validation) from 2003 to 2014 for the region. Ta b le 4. 2 shows the NSE, Pbias, RSR, and RSME values achieved for the calibrated model. As shown in the table, all criteria (NSE, PBIAS, 68 and RSR) are in their respective satisfactory ranges (Moriasi et al . , 2007) indicati ng that the model was successfully ca l ibrat e d and can be used to simulate streamflow values for the region. Furthermore, while the overall RSME was 6.522, the calibration period had a smaller RSME compared to the validation period, indicating a better mode l fit during the calibration period t h an th e validation period. The temporal variability of observed and simulated streamflow is also presented in Figure 4.2. Overall, the SWAT model represents the observed flow variations very accurately. Table 4.2. Calib ration and validation criteria NSE PBIAS (%) RSR RSME Overall (2003 - 2014) 0.612 - 0.965 0.623 6.522 Calibration (2003 - 2008) 0.611 4.303 0.624 5.996 Validation (2009 - 2014) 0.613 - 5.856 0.622 7.009 Figure 4.2. Comparison of observed and simulated daily streamflow T he re s ults of this section present the performance of the SWAT model in replicating the spatially distributed ETa data obtained from two remote sensing products (SSEBop and ALEXI datasets). Table 4. 3 shows the SWAT model performance for the overall, calib r ation , and 69 validation periods based on NSE, PBIAS, RSR, and RMSE of the ETa for the condition in which only the streamflow calibration was performed . These calculations followed the same procedure th at was discussed in the multi - variable and GA calibratio n sect i ons, in which ETa values were area averaged across the watershed and then used to calculate watershed level statistical criteria. When considering the entire time period , the streamflow calibra ted SWAT model was able to replicate the SSEBop ETa data s et mo r e accurately than the ALEXI ETa dataset. This can be seen by the fact that the statistical criteria for the SSEBop ETa were better than those for the ALEXI ETa. Similar results were see n for the calibration and validation periods. Overall, this show s that the SWAT model can better replicate the SSEBop ETa data compared to the ALEXI data. Table 4. 3 . Statistical crit eria ETa when the results from base streamflow calibrated SWAT model was used Period Variable/Dataset Statistical Measure NSE PBI AS ( %) RSR RMSE Overall (2003 - 2014) ALEXI ETa 0.62 27.82 0.62 21.79 SSEBop ETa 0.81 - 10.12 0.44 18.28 Calibration (2003 - 2008) ALEXI ETa 0.62 27.83 0.62 21.48 SSEBop ETa 0.81 - 8.46 0.44 18.13 Validation (2009 - 2014) ALEXI ETa 0 . 62 27.80 0.61 22.10 SSEBop ETa 0.80 - 11.78 0.44 18.42 4.4 .2 Multi - variable Calibration A combination of 5,000 Monte Carlo simulations and a n NSGA - II evolutionary algorithm were used to identify the Pareto frontiers for the SWAT model ca lib rat i ons for both the ALEXI and SSEBop ETa datasets. Figure 4. 3 shows both the entire Monte Carlo population as well as the Pareto frontiers identified by the NSGA - II evolutionary algorithm for each ETa dataset. This shows that Pareto frontiers were able to be identified from the Monte Carlo simulations run for each ETa datasets, which indicates the first phase of the multi - variable optimization was successful for both datasets. However, the SSEBop Pareto frontier was able to further minimize streamflow a nd ETa OFs compared to the ALEXI Pareto frontier. Therefore, 70 calibrating the SWAT model using the SSEBop ETa data was able to produce a more accurate model performance. This can be seen more clearly in Figure 4. 4, which shows the Pareto frontiers for both the SS E Bop and ALEXI datasets. This figure also highlights the optimal Pareto population member selected by the compromise progra m ming method, which shows the optimal m odel calibration for each dataset. This reinforces the conclusions that the SSEBop datas et per f ormed better than the ALEXI dataset and achieved a model calibration that was able to better simulate both streamflow and ET values for the entire region. In add ition , the results showed that the multi - variable calibration was able to identify a fin al cal i brated model for each dataset that improved both streamflow and ET simulations. Figure 4.3. Monte Carlo populations and Pareto frontiers for a) ALEXI and b) S SEBop datasets 71 Figure 4.4. Pareto frontiers and optimal Pareto population members f or bot h ALEXI and SSEBop datasets Table 4. 4 shows the NSE, PBIAS, RSR, and RMSE values achieved for both final calibrated models. All values presented in the table fal l within the satisfactory ranges and indicate that the models were successfully calibrat ed . Fu rt hermore, a comparison of these values with the base model simulations showed that with respect to ET there was an improvement in the statistical criteria. For e xample, when considering overall NSE the ALEXI calibrated model had a value of 0.73 comp are d t o the 0.62 for the base model and the SSEBop calibrated model had a value of 0.85 compared to the 0.81 for the base model. This indicates that the newly calibrate d models are better able to simulate ETa data. However, with respect to streamflow, all sta tis ti cal criteria remain within the satisfactory ranges and often similar to the base model statistical 72 criteria, suggesting that the streamflow simulations were not heavily impacted by the addition of the ET calibration . Overall, the results show that thi s c al ibration approach was successful at Table 4.4. Statistical criteria for optimal multi - variable calibration models Period ET Dataset Statistical Criteri a NSE PBIAS (%) RSR RMSE ET Streamflow ET Streamflow ET Streamflow ET Streamflow Overall (2003 - 2014) ALEXI 0.73 0.59 21.73 13.70 0.52 0.64 18.32 6.70 SSEBop 0.85 0.61 - 16.05 8.20 0.39 0.63 16.05 6.57 Calibration ( 2003 - 2 008 ) ALEXI 0.72 0.59 22.01 18.52 0.53 0.64 18.37 6.19 SSEBop 0.85 0.61 - 14.03 12.94 0.38 0.63 15.85 6.01 Validation (2009 - 2014) ALEXI 0.74 0.59 21.46 9.22 0.51 0.64 18.28 7.18 SSEBop 0.85 0.60 - 18.07 3.79 0.39 0.63 16. 25 7.09 4.4 .3 Genetic Algorithm Calibration In addition to the multi - variable approach, a GA optimization was also performed . Unlike the multi - variable approach, this approach focused on only improving the ETa estimations for two remotely sen s ed d at ase ts (ALEXI and SSEBop) without considering the st reamflow calibration. After hundreds of runs for each subwatershed, the GA was able to identify the optimal parameters values for each subwatershed and the ETa datasets. These final parameter values were u sed to develop SWAT models that represented the opt imal ETa calibration for each subwatershed. Table 4. 5 shows the NSE, PBIAS, RSR, and RMSE values achieved for both final calibrated models. All of the ETa statistical criteria values presented in th e tab le fa ll within the satisfactory ranges and indicate t hat the models were successfully calibrated with respect to ET. When compared to the base model, it can be seen that the ETa calibration performed here was able to improve the simulation of ETa valu e s fo r bot h the ALEXI and SSEBop datasets. For example whe n considering the overall NSE, the ALEXI calibrated model had a value of 0.75 compared to the 0.62 for the base model and the SSEBop calibrated model had a value of 0.84 73 compared to the 0.81 for the base m ode l. However, when considering the streamflow cali bration, most of the statistical values have fallen outside the satisfactory ranges (NSE > 0.5, PBIAS ±25%, and RSR < 0.7) for each criteri on . This indicates that while this process was able to impr o ve t he ET simulations, it was done at the cost of comprom ising streamflow simulations. This seems logical, knowing that this approach did not consider the streamflow calibration during the ETa calibration process. However, this does indicate that this app r oach w oul d be unsuitable for calibrating models that requ ire accurate streamflow values. Table 4.5. Statistical criteria for the optimal GA calibrated models Period ET Dataset Statistical Criteria NSE PBIAS (%) RSR RMSE ET Streamflow E T Str e amfl ow ET Streamflow ET Streamflow Overall (2003 - 2014) ALEXI 0.75 0.32 14.34 32.73 0.50 0.82 17.84 8.61 SSEBop 0.84 0.52 - 17.42 10.69 0.39 0.69 16/35 7.28 Calibration (2003 - 2008) ALEXI 0.74 0.22 14.89 39.24 0.51 0. 8 8 17 . 82 8.50 SSEBop 0.85 0.50 - 15.25 16.22 0.39 0.71 16.19 6.80 Validation (2009 - 2014) ALEXI 0.75 0.40 13.80 26.67 0.50 0.77 17.86 8.71 SSEBop 0.84 0.53 - 19.59 5.55 0.40 0.69 16.51 7.73 4.4 .4 Statistical Significance T h e re su lts of the statistical analysis of the mean difference between each of the datasets are presented for streamflow and ETa in Tables 4. 6 and 4. 7, respectively. Linear mixed - effects models were employed to account for the spatiotemporal effects that c a u se s am ple correlation violating the independence assumption for the usual paired t - test (Esfahanian et al., 2017). With regard to the streamflow datasets, all comparisons were found to be significantly different from each other except for the comparison o f the o bser ved dataset with the initial streamflow calibrated model. This indicates that the initial calibration was able to closely replicate the observed data to the point where statistically there is no difference between them. However, the significant d iffe re nce observed for all other models compared to the observed data indicates that 74 those models are not as accurate when simulating streamflow. This seems logical for the models calibrated via the genetic algorithm approach since there was a noticeable d ecre as e in the statistical criteria for the streamflow calibration in these models. However, we did not expect this for the models calibrated using the multi - variable approach, since these models showed little to no change in the calibration criteria for s trea mf low. These results indicate that even though the calibration process was able to satisfactorily calibrate streamflow , there exist more inconsistencies within the final simulated streamflow when compared to the observed data. When considering the co m p aris on of streamflow simulations between the initial model and the other four models tested, the significant difference makes sense and indicates that the addition of the ETa calibration influenced the streamflow calibration to an extent. Furthermore, si n c e al l of t hese the p - values were negative, the ETa calibrated models all underestimated the streamflow when compared to both the observed dataset and the initial streamflow model. This indicates that regardless of the calibration method used or the impac t seen o n th e statistical criteria, the ETa calibrated models produced lower streamflow values on average. Finally, the comparisons between the four ET calibrated models also showed a significant difference, which seems understandable given the use of diff e r ent ET dat asets and calibration process used in this study. With regards to the ETa datasets, almost all comparisons among datasets showed significant differences except for the SSEBop dataset versus the initial streamflow calibrated model and the SSEBop g enet ic alg orithm calibrated model versus the ALEXI multi - var iable calibrated model. These two cases are rather interesting since the first comparison (SSEBop versus the initial streamflow calibrated model) indicates that by only calibrating for streamflow it w as pos sible to simulate ETa so that it is not statistica lly different from the remotely sensed data. Meanwhile the second case (SSEBop genetic algorithm calibrated model versus ALEXI multi - 75 variable calibrated model) indicates that regardless of using d iffe re nt a pproaches and datasets, similar ETa simulations we re generated . Considering all of the other significant differences , the comparison between the ALEXI and SSEBop data made the most logical sense since different methodologies were used to calcula t e th es e da tasets. Furthermore, similar results to the stream flow were also seen when comparing the ETa calibrated models to the remotely sensed ETa datasets. These observations confirm that even though these models were able to satisfactorily simulate ETa valu es , th e SWAT simulated ETa was statistically different f rom the remotely sensed data used to calibrate them, and thus could not accurately replicate the remotely sensed data. However, while the streamflow comparisons showed that all of the ETa calibra t ed S WA T mo dels underestimated streamflow, here it can be see n that the SSEBop calibrated SWAT models overestimated ETa values while the ALEXI calibrated SWAT models underestimated the ETa values when compared to the SSEBop and ALEXI datasets, respectively . In ad diti on , similar to the streamflow comparisons, the fou r ETa calibrated models were significantly different from the initial streamflow calibrated model, which makes sense since all of the ETa calibrated models had an increase in the statistical crit e ria fo r ET a calibration compared to the initial streamflow c alibrated model. Finally, the comparisons between the four ETa calibrated models showed a significant difference from each other except for the case of the SSEBop genetic algorithm calibrated mod e l ve rs us t he ALEXI multi - variable calibrated model discussed previously. This is reasonable since different calibration approaches and ETa datasets were used . 76 Table 4.6. Mean differences and p - values from the mixed - effects model for comparison of the dif f eren t stre amflow datasets used in this study. Bolded values indicate significant difference at the 0.05 level Streamflow Datasets* Streamflow Datasets* A B C D E F A B 0.08 (0.75) C - 3.18 (0.00) - 3.26 (0.00) D - 1.0 4 (0.00) - 1.13 (0.00) 2.13 (0.00) E - 1.34 (0.00) - 1.42 (0.00) 1.84 (0.00) - 0.29 (0.01) F - 0.80 (0.00) - 0.89 (0.00) 2.37 (0.00) 0.24 (0.03) 0.53 (0.00) * A = Observed Streamflow, B = Initial Streamflow Calibrated Mode l , C = ALEXI Genetic Algorithm Calibrated Model, D = SSEBop Genetic Algorithm Calibrated Model, E = ALEXI Multi - Variable Calibrated Model, and F = SSEBop Multi - Variable Calibrated Model. Table 4.7. Mean differences and p - values from the mixed - effects mode l for c ompar ison of the different ETa datasets used in this study. Bolded values indicate significant difference at the 0.05 level ET Datasets* ET Datasets* A B C D E F G A B 20.10 (0.00) C 2.75 (0.09) - 17.35 ( 0.00) D 11.07 (0.00) - 9.03 (0.00) 8.32 (0.00) E 6.69 (0.00) - 13.41 (0.00) 3.94 (0.00) - 4.38 (0.00) F 6.97 (0.00) - 13.13 (0.00) 4.22 (0.00) - 4.10 (0.00) 0.28 (0.23) G 5.67 (0.00) - 14.43 (0.00) 2.92 (0.00) - 5.40 (0.00) - 1.02 (0.00) - 1.30 (0.00) * A = SSEBop, B = ALEXI, C = Initial Streamflow Calibrated Model, D = ALEXI Genetic Algorithm Calibrated Model, E = SSEBop Genetic Algorithm Calibrated Model, F = ALEXI Multi - Variable Calibrated Model, and G = S S EBop M ulti - V ariable Calibrated Model. 77 4.4 .5 Comparison of the Multi - variable and Genetic Algorithm Calibrations Based on the information provided in Tables 4. 4 and 4. 5, it can be concluded that the multi - variable approach used in this study was able to genera te bet ter overall SWAT models compared to the GA approach. However, if the goal of the model is to generate more accurate ETa data, the GA approach was able to outperform the multi - vari able approach. This shows that depending on the purpose of the m o del ap plicat ions, different calibration techniques should be used . Furthermore, it is to be noted that for both approaches the models built using the SSEBop data were able to achieve higher p erformances in simulating both streamflow and ETa data than the m odels made b ased on the ALEXI data. 4.5 Conclusions In this study, two different ETa calibration techniques were used to evaluate the impact of adding spatially distributed and remotely sen sed ETa datasets to the traditional streamflow calibration used i n hydr ologic al models. Both techniques, multi - variable and GA, were able to successfully improve the ETa calibration for the hydrological model using both remotely sensed ETa datasets. The GA technique was able to produce better ETa calibrations and thus b etter ETa si mulations; however, this was achieved at the cost of lowering the streamflow calibrations. Meanwhile , the multi - variable technique was able to improve the ETa calibration while ma intaining the streamflow calibration. Therefore, future use of t h ese ap proach es should be driven by the needs of the research. For example, if a study is focused solely on better ETa estimation, the GA approach is the better option; meanwhile , studies focu sed on better simulating the entire hydrological cycle for a reg i on sho uld us e the multi - variable approach. Concerning the ETa datasets used in this study, the calibrations performed with the SSEBop dataset resulted in better ETa estimations compared to th e calibrations based on the ALEXI 78 dataset for this study area. T h erefor e, it is recommended that future studies should perform this analysis in other regions to better understand how these datasets compare to each other as well as evaluating the impacts of different climate variabilities (e.g., snow cover). Statistica l analy sis of the streamflow and ETa showed that the remotely sensed ETa datasets were significantly different from each other, which was expected . Moreover, except for one exception, all of t he streamflow and ETa datasets produced by the ETa calibrated SW A T mode ls wer e also significantly different from each other. However, all four ETa calibrated models were also significantly different when compared to the remotely sensed data. This indicated that while the overall model calibration was successful it was u nable to clo sely replicate the remotely sensed data, showing that there still could be additional improvements in the both in the calibration process and the SWAT model simulations. It is to be noted that the ETa calibration processes used in this study o n ly alt ered t hree parameters within the SWAT model. This was due to temporal and computational limitations. However, the addition of other parameters to the calibration process, such as the so il evaporation compensation factor (ESCO), could result in even b etter model calibrations and thus better model outputs and should be the focus of future studies. In addition , while adding ETa calibration to the overall model calibration process was succes sful in this study, future studies should consider additional hy d rologi cal cy cle components, such as remotely sensed soil moisture datasets. This would allow for the development of even more realistic models and thus more accurate results for stakeholders and policymakers who rely on model outputs for managing freshwat e r reso urces. 4.6 Acknowledgment Authors would like to thank Dr. Wade Crow from USDA - ARS Hydrology and Remote 79 Sensing Laboratory at Beltsville, Maryland for his help with editing the paper. Th is work is supported by the USDA National Institute of Food and A gricul ture, Hatch project MICL02359. 80 5. EVALUATING THE SPATIAL AND TEMPORAL VARIABILITY OF REMOTE SENSING AND HYDROLOGIC MODEL EVAPOTRANSPIRATION PRODUCTS 5. 1 Introduction Freshwater is vital for life and therefore understanding how the hydrological c y cle ch anges ha s become a major focus of many researchers, especially given the increased demand for water across the globe ( Clark et al., 2015; Srinivasan et al., 2017 ). Tradit ionally, this has been accomplished using monitoring stations that record diffe r ent as pects of the hydrological cycle, such as streamflow and precipitation. However, these stations can be expensive to install, maintain, and operate and thus their coverage is often low and not enough to capture spatial and temporal variabilities of hy d rologi cal cy cle especially in large areas (Wanders et al., 2014). One solution to this is the use of remote sensing products. Remote sensing (RS) is the use of sensors and tool s to indirectly measure the characteristics of an object (Graham, 1999). And wi t h the advanc ement of satellite technology, remotely sensed has become a common approach for generating consistent global monitoring datasets such as different elements of hydro logical cycles ( Long, et al., 2014 ). In the hydrological cycle, evapotranspira t ion (E T) is an influential component since it is plants (USGS, 2016 d ). Which means th at ET supplies water vapor to the atmosphere driving weather patterns and preci p itatio n dist ributions (Pan et al., 2015; USGS, 2016 q ). Meanwhile, since ET measures the loss of moisture from plants and soil, its magnitude is dependent on the landscape. Ther efore, measuring ET is a large scale is difficult through traditional technique s (Wu e t al., 2008), but a prime hydrological component to be measured through remotely sensed techniques (Anderson et al., 2012). This has led to the development of a variety o f different ET 81 remotely sensed monitoring products, such as the Simplified Surf a ce Ene rgy Ba lance (SSEB) (Zhang et al., 2016), the Atmosphere - Land Exchange Inverse (ALEXI) (Anderson et al., 2007; Senay et al., 2013), the Moderate Resolution Imaging Spectro radiometer (MODIS) Global Evapotranspiration Project (MOD16) (Zhang et al., 201 6 ; NTSG , 2018 ), the Google Earth Engine Evapotranspiration Flux (Google, 2018), and the North American Land Data Assimilation Systems phase 2 (NLDAS - 2) (Xia et al., 2015). These products can be categorized based on the method they use to calculate ET with t he mos t comm on categories being Surface Energy Balance Methods, Penman - Monteith Methods, and Priestly - Taylor Methods (Bhattarai et al., 2016; Zhang et al., 2016). However, each of these methods ha ve different assumptions and inputs required to calculate E T while there is a higher level of uncertainty associated with the remotely sensed data compared to traditional ET monitoring techniques (van der Tol and Parodi, 2012; Zhang et al., 2016). All of these can make it challenging for researchers and policy mak e rs to know w hich ET product should be used considering landuse/landcover and a period of study. One technique to address the uncertainty within remotely sensed datasets is t he use of an ensemble of several different products (Duan et al., 2007). Creati n g an e nsembl e of datasets helps reducing the uncertainty of individual datasets by combining the benefits of each dataset while minimizing negative aspects like outliers ( Diett erich, 2000 ). This has led to the creation of a variety of ensemble techniques a nd app licati ons that have been applied to remotely sensed products ( Christensen and Lettenmaier, 2006; Fowler and Ekström, 2009; Lee et al., 2017; Wang et al., 2018 ). The compl exity of these techniques ranges from very simple calculations such as simple a v eragin g to v ery complex techniques such as ensemble Kalman filter (EnKF) (Giorgi and Mearns, 2003; Kim et al., 2015; Wang et al., 2018 ). However, the Bayesian Model Averaging ( BMA) is the most commonly used ensembling technique for ET remotely sensed 82 prod u cts (K im et al., 2015; Tian and Medina, 2017; Yao et al., 2017; Ma et al., 2018 ) that reduces overall dataset uncertainty by weighting ET products based on the observed data (K im et al., 2015). However, this technique is dependent on the availability of o b served data, which depending on the region can be difficult to obtain. In summary, the wide range of techniques can make it challenging to know which technique should be appli ed. Therefore, given the challenges associated with the selection and use of re m otely sensed ET products in the field of hydrology three objectives were identified for this study: 1) explore the temporal performance of individual and an ensemble remotely s ensed datasets; 2) evaluate the spatial performance of individual and an ensemb l e remo tely s ensed datasets; 3) compare the performance of individual remotely sensed datasets to the ensemble and 5. 2 Materials and Methods To accomplish the objectives of this study a variety of tasks were performed . First eight remote ly sensed ETa datasets along with an ETa E nsemble and ETa output of a hydrological model were obtained for a study area. Since each of these datasets h as different spatial and temporal resolutions, they were aggregated or disaggregated to crea t e a se ries o f comparable ETa datasets. In order to determine their performance in the study area, several forms of statistical analysis were performed to examine t he spatio temporal variabilities in addition to their fit to the E nsemble and hydrological mo d el out put. T he following sections provide additional information about all of the processes used in this study. 5. 2.1 Study Area The Honeyoey Creek - Pine Creek Wat ershed (Hydrologic Unit Code 0408020203) was selected for this study (Figure 5. 1). Located i n Michi Lower Peninsula, this watershed is 83 part of the Saginaw Bay Watershed, which is the largest watershed in Michigan with the final outlet at Lake Huron. F urthermore, this region has been identified as an area of concern by the US Environmental Pr o tectio n Agen cy due to the degradation of fisheries, the presence of contaminated sediments, and implementation of fish consumption advisories within the region (EPA, 2017). On average the region receives 81 cm of rainfall per year with higher rainfalls ob s erved during the months between April and November (US Climate D ata, 2018). Furthermore, the late fall and winter months (November through February) experience more clouds and shorter days, while the late spring and summer months experience fewer clouds a n d long er day s. Meanwhile, the air temperature in the region rang es from - 10 to 27 , with winter months (December through February) having colder temperatures and snow, while summer months (June through August) have hotter temperatures and more rainfall ( U S Clim ate Da ta, 2018). Soils in the area are dominated by mixtur es of loam and sand with low slopes (NRCS, 2018). Landuse in the Honeyoey watershed is dominated by agricultural land (~57%) followed by forests (~23%), wetlands (~17%), and urban areas (~3%) . Given the h eavy agricultural nature of the region is it importa nt to note that corn and soybean rotations are the most common crops; however, eight different cropping systems have been identified in the region including alfalfa, corn, field peas, hay, pa s ture, sugar beet, soybean, and winter wheat. In general, agricul tural operations like tillage and crop planning start in mid - spring (i.e., May) and crops are harvested mid - fall (i.e., October) (Love and Nejadhashemi, 2011). In cases where cover crops are u tilize d , pla nting begins post - harvest in the fall, which require s additional tillage and planting operations (Love and Nejadhashemi, 2011). Overall, 13 types of landuses were identified including: alfalfa (ALFA), corn (CORN), field peas (FPEA), forest d e ciduou s (FRS D), forest evergreen (FRSE), hay (HAY), pasture (P AST), sugar beet (SGBT), soybean (SOYB), residential low 84 density ( URLD ), urban transportation (UTRN), wetlands forested (WETF), and winter wheat (WWHT) (NASS, 2018). These individual la n duses were a lso combined into four major landuse categories of a griculture (ALFA, CORN, FPEA, HAY, PAST, SGBT, SOYB, and WWHT), forest (FRSD and FRSE), urban ( URLD and UTRN), and wetland (WETF) for additional analysis. Figure 5. 2 shows the spatial distrib u tion o f the major landuse categories throughout the Honeyoey wat ershed. Meanwhile , regarding hydrological and climatological monitoring in the area, st r eamflow is monitored by a United States Geological Survey (USGS) station located at the outlet of the r e gion ( Figure 5. 1). Furthermore, two precipitation and two temperature National Climatic Data Center (NCDC) stations are located within the Honeyoey watershed (NCDC, 2018) (Figure 5. 1). Automated airport weathe r stations are also located within and around t he Hon eyoey watershed and collect wind speed and direction, temperature, dew point, altimeter setting, density altitude, visibility, sky condition, cloud ceiling, precipitation, and precipitation type (FAA, 20 18). Additional weather stations from the MSU E nvirow eather system measure air and soil temperature, precipitation, relative humidity, wind speed and direction, solar radiation, leaf wetness, and potential ET (Enviroweathrer, 2018). However, none of the en viroweather stations are located within the st u dy reg ion. M eanwhile, there are several AmeriFlux stations located in Michigan that can be used to report ETa; however, the closest of these stations is 116 km from the Honeyoey watershed (AmeriFlux, 2018). 85 Figure 5. 1. Map of the Honeyoey watershed and locati ons of climatological stations within and near the region 86 Figure 5.2. Map of the individ ual (a) and major (b) landuse classes within the Honeyoey watershed based on the 30 m resolution map obtained from the Cropland Data Layer developed by the United S tates Department of Agriculture - National Agricultural Statistics Service 5. 2.2 Remote Sensing Evapotranspiration Products In order to examine the spatial and temporal performance of remotely sensed ET p roducts, eight actual ET (ETa) datasets were o b tained for th e study area. ETa describes the actual amount of water loss that occurs at a site via evaporation and transpiration and thus is limited by the actual amount of water present (NOAA, 2017 b ). The ETa datasets utilized for this study include 1) t h e USGS Simpli fied Surface Energy Balance (SSEBop), 2) the Atmosphere - Land Exchange Inverse (ALEXI), 3) the MODIS Global Evapotranspiration Project (MOD16A2) 500m, 4) the MOD16A2 1 km, 5) the North American Lan d Data Assimilation 87 Systems 2 Evapotranspirati o n (NLD AS - 2) M osaic, 6) the NLDAS - 2 Noah, 7) the NLDAS - 2 Variable Infiltration Capacity (VIC), and finally 8) TerraClimate. The first ETa dataset (SSEBop) was obtained from the USGS and calculates monthly ETa by using the simplified surface energy balance model (Senay et al., 2013). This technique utilizes 8 - day, 1 km MODIS thermal imagery to calculate ET fractions, which are then aggregated to develop monthly ETa values for the Contiguous United States (Senay et al., 2013; Velpuri et al., 2013). The secon d ETa d ataset (ALEXI) was developed as a joint project between the United States Department of Agriculture (USDA) and the National Aeronautics and Space Administration (NASA). In this dataset, the ETa was calcu lated by comparing changes in remotely sensed s urface temper atures, obtained from Geostationary Operational Environmental Satellites (GOES), and relating that difference to surface moisture loss (Anderson et al., 2007). This calculation is performed on a d aily basis , resulting in a spatial resolution o f 4 km ETa da taset for the Contiguous United States (Hain et al., 2015). The third and fourth ETa datasets (MOD16A2 500 m and 1 km) were developed as a joint project between NASA and the University of Montana Numerical Terradynamic Simulation Group (NTSG, 2018). This t echnique utilizes the improved ET algorithm based on the Penman - Monteith equation and takes into account additional information such as MODIS landcover , leaf area index (FPAR/LAI), and global surface meteorology (Mu et al. 2011; NASA, 2018a,b ) . The result is an 8 - day 500 m and 1 km global ETa datasets (NASA, 2014). The fifth through seventh ETa datasets (NLDAS - 2) are part of the North A merican Land Data Assimilation System (NLDAS) project, which was jointly worked on by the National Oceanic an d Atmos pheric Administration (NOAA) and the National Centers for Environmental Prediction (NCEP) Environmental Modeling Center, NASA's Goddard Spac 88 National Weather Service Offi c e of H ydrolog ical Development, and the NOAA/NCEP Climate Prediction Center (NASA, 2018c). NLDAS - 2 calculates ETa by coupling three different land surface models, namely the Mosaic model, the Noah model, and the VIC model (Xia et al., 2015). The use of the s e mode ls allo ws NLDAS - 2 to take into account a variety of physical processes/characteristics such as atmosphere interactions of water and energy, vegetation and soil moisture heterogeneity, water and energy budgets, and rainfall - runoff and water storage ( X ia et al., 20 15). The results are three ETa datasets that are calculated at both hourly and monthly time steps at a 1/8 degree spatial resolution (Long et al., 2014). And finally, the eighth ETa dataset (TerraClimate) was developed as a joint project betw e en the Univer sity of Idaho, the University of Montana, and the USDA Forest Service Rocky Mountain Research Station. The ETa product of TerraClim ate is produced based on the one - dimensional modified Thornthwaite - Mather climatic water - balance model (Abatz o glou e t al., 2018). This results in a monthly, global ETa dataset with a spatial resolution of 4 km (Abatzoglou et al., 2018). It should be noted that the TerraClimate did not report any ETa values for January and February. Table 5. 1 summarizes the spatia l and t emporal resolutions for each of the ETa datasets. Meanwhile, since the ETa products were obtained from remotely sensed, calibration and vali dation were necessary before the products were made available to the public. This was performed for all of th e afore mention ed ETa products and the levels of accuracy were also reported in Table 5.1. As can be seen in Table 5.1, the spatial accuracy of the ETa datasets varies between 3.65 mm/month to 30. 42 mm/month . However, given not every ETa product utilized th e same measure of accuracy (root mean squared e rror (RMSE) ; root - mean - square deviation ( RMSD ) ; mean absolute error ( MAE )), it is not appropriate to compare the accuracies between types of error measurement. However, among those ETa products that reported R MSE, t he most 89 accurate are the MOD16 products followed by SSEBop and then ALEXI. Meanwhile, among the products reporting RMSD, the most accurate i s NLDAS - 2: Noah, followed by NLDAS - 2: VIC and then NLDAS - 2: Mosiac. However, it is important to note that the s e erro rs are based on site - specific comparisons with observed data. This means that for any given location between two observed sites, the actual error associated with each dataset could flux. In addition , the accuracy level reported in Table 5.1 are not a bsolut e error s, which mean that they can change throughout the years and for different landuses. Given this, it is important to note that the goal of this study is not to perform revalidation for the selected datasets but to see how the ETa datasets perfo r m with in the study area. However, we are interested to assess how different spatial and temporal variations are represented by each dataset while identifying the possible sources of discrepancy among datasets . In addition , and as presented Section 2.1 of t his pa per, wh ile there are many monitoring sites within and around the study area, there is a lack of observed ETa datasets. Therefore, in order to help to account for the uncertainty within the datasets, an Ensemble dataset based on an av eraging techniqu e (Teba ldi and Knutti, 2007) was also created. It is important to note that, the use of a straight average for ensembling is not as robust as other techniques such as BMA (Krishnamurti et al., 2000); however, due to the lack of observed dat a in the region ( Figure 5.2) i t was considered as the most appropriate technique to use. 90 Table 5. 1. Summary of remotely sensed ETa datasets used in this study ETa Dataset Coverage Resolution Accuracy (mm/month) Spatial (km 2 ) Temporal Reference SS E Bop C onti guous United States 1.0 Monthly 27. 25 (RMSE) (Velpuri et al., 2013) ALEXI Contiguous United States 4.0 Daily 30.42 (RMSE) (Cammalleri et al., 2014) MOD16A2 1 km Global 0.5 8 - day 26.07 (RMSE) (Mu et al., 2011) MOD16A2 500m Globa l 1.0 8 - d ay 2 6.07 (RMSE) (Mu et al., 2011) NLDAS - 2: Mosaic North America 12.0 Hourly/Monthly 10.37 (RMSD) (Long et al., 2014) NLDAS - 2: Noah North America 12.0 Hourly/Monthly 3.6 5 (RMSD) (Long et al., 2014) NLDAS - 2: VIC North America 12.0 Hourl y/Mo nthly 6.66 (RMSD) (Long et al., 2014) TerraClimate Global 4.0 Monthly 4.75 (MAE) (Abatzoglou et al., 2018) *RMSE: Root Mean Squared Error; RMSD: root - mean - square deviation; MAE: mean absolute error 5. 2.3 Hydrological Model In additi o n to t he ei ght r emotely sensed ETa products and the ET a E nsemble, a hydrological model was used to estimate ETa for the region as well. Hydrological models are often used to simulate the hydrological cycle across the landscape, since they are an efficient and in expen sive alternative to monitoring (Giri et al., 2 012). They accomplish this, in general, by performing a water balance for the region, which utilizes various calculations describing water movement throughout the landscape as well as the interactio n s betw een w ater and biotic and abiotic characteristics ( M artinez - Martinez et al., 2014 ). ET is one of the major components of the water balance and as such plays a major role in hydrological models. In order to estimate ETa, hydrological models often firs t calcu late poten tial ET and then account for actual loss by determining the impacts of landcover and soil moisture (Kite and Droogers, 2000). 91 In this study, the hydrological model selected was the Soil and Water Assessment Tool or the SWAT model. SWAT is a semi - dist ribut ed, continuous - time hydrological model de veloped by the USDA Agriculture Research Service and Texas A&M AgriLife Research (Texas A&M University, 2018). This is the most widely used hydrologic model, which utilizes several different datas e ts, su ch as topo graphy, soil properties, landuse, and cli matological observations to simulate hydrological parameters such as streamflow and ET (Gassman et al., 2007). Regarding ETa estimation, the SWAT model first calculates potential evapotranspiration. This c an be done one of three ways, namely the Penman - Mon teith method, the Priestley - Taylor method, and the Hargreaves method, with the Penman - Monteith Method as the default (Neitsch et al., 2011). After this, SWAT takes into account the evaporation from r ainfal l int ercep ted by the canopy, maximum transpiration, maximum soil evaporation, and sublimation (during periods of snow cover) (Neitsch et al., 2011). These calculations are performed at the hydrologic response unit scale, which divided the region int o subba sins that have unique physiographical characteristi cs. For this study, 250 subba si ns were created in the study area. This number was selected due to limitations in the number of unique landuses that could be applied within the SWAT model. Ultimately , all o f the se ca lculations result in the creation of a da taset that reports monthly ETa at the subbasin level. To ensure that the hydrological model represented the study area , calibration and validation were performed for the period of 2003 to 2014, wit h 2003 - 2004 servi ng as a model warmup period, 2005 - 2009 us ed for calibration period, and 2010 to 2014 used for validation period. The hydrological cycle component used for this process was streamflow, with the daily model streamflow output being compared t o obser ved d aily streamflow at the watershed outlet. To evaluate this comparison , three statistical criteria were used, namely Nash - Sutcliffe efficiency 92 (NSE), percent bias (PBIAS), and root mean squared error - observations standard deviation ratio (RSR), w h ich we re id entif ied by Moriasi et al. (2007). Calibrati on and validation were successful for the developed model since the following ranges for each statistical criterion were met: NSE > 0.5, PBIAS ±25, and RSR < 0.7 (Herman et al. , 2015). 5. 2.4 Remotely Sensed Actu al Ev apotranspiration Data Source and Conver sion Procedure All the ETa datasets were obtained for the period 2003 - 2014 for the study area. This period was selected since all of the selected ETa datasets had data available during this period. T h e NLDA S - 2 d atase ts (from Mosaic, Noah and VIC models) w ere obtained using the NASA Goddard Earth Science Data and Information Services Center website (NASA/GSFG, 2018) . ETa values for each mo d el wer e ext racte d using the wgrib program developed by National Centers for Environmental Prediction (NOAA - NCEP, 2013) . Simil a rly, a verag e ETa values from MOD16A (8 - day values, 0.5 and 1 km resolutions) and TerraClimate (monthly values) datasets were obtained using the code editor of Google Earth Engine (Gorelick et al., 2017) . Missing 8 - day valu e s in M OD16A data sets were completed using multi - year av erages for either the respective week or month of the missing values (the latter when the average week value was not available) ( Mu et al. 2011 ). Meanwhile, averaged ETa values for the USGS SSEBop pro d uct an d the USDA - NASA ALEXI product, were obtained from a previous study (Herman et al., 2018) . However, to compare these ETa datasets, they need to be converted to similar spatial and temporal resolutions. The fi rst s tep w as to ensure that each ETa dataset was reported on a monthly basis. For datasets already reported on a monthly basis (SSEBop, NLDAS - 2, and Terra Climate) no processing was needed. However, for datasets that reported ETa on a daily (ALEXI) o r 8 - da y (MO D16) basis, values within each month were su mmed. The second step was to 93 ensure that each ETa dataset accounted for spatial variability within the landscape. To accomplish this the ETa datasets had to be averaged for each subbasin. This was don e by us ing w eight ed area averaging technique on all ETa datasets for each unique physiographic region within the Honeyoey watershed. The average ETa values were obtained by resampling the raster files to a cell size of 10 m and computing the mean value of t he cel ls wi thin each subbasin. Weighted area averaging was used since it was able to address the issues of multiple pixels and partial pixels occurring within unique physiographic regions and resulted in a single ETa value for each subbasin. By performing these two p roces ses, eight monthly ETa datasets at the subbasin level were created that can be used for further analysis. 5. 2.5 Statistical Analysis In order to compare the performance of the eight datasets, E nsemble, and SWAT model within the study regi o n, t hr ee di ffere nt statistical approaches were used . These analyses were performed to take into account different spatial scales (subbasin, watershed), landuses (major and individual), and temporal resolutions (overall, seasonally, and monthly). The first stat is tical appr oach evaluated the temporal variability of the different ETa datasets and utilized multi - pairwise comparisons to estimate the monthly mean differences between ETa datasets, for both the whole watershed and for specific landuse types. This w as d on e to deter mine if any patterns could be found among the ETa datasets throughout the year. Since this was done for both the entire watershed and for each landuse, a total of 18 area - weighted ETa monthly time - series were generated for each ETa dataset (13 in divid ual l anduses, 4 major landuses, and the entire watershed). To compare these datasets, two different models were used: 1) for overall comparisons the Generalized Least - Square (GLS) estimation with Autoregressive model, with a lag of 1, or AR (1) was us ed si nce c omplete time series was being 94 compared (Fox and Monette, 2002); 2) while for monthly and seasonal comparisons the GLS estimation with Continuous Autoregressive model with lag 1, or CAR (1) was used since irregularly - spaced time - series were bein g compa red ( Wang, 2013). In both cases, the difference between the two analyzed time - series was used as the response variable. Furthermore, a p - value less than 0.05 denoted datasets that were significantly different (Nejadhashemi et al., 2012). The seco nd stat istic al approach evaluated the spatial variability of the different ETa datasets and utilized multi - pairwise comparisons of different landuse types (including the watershed average) for each individual ETa dataset and across all ETa datasets. B y per fo rming both of these analyses, it is possible to evaluate the performance of individual ETa datasets in differentiating among landuse classes as well as determine if different ETa datasets perform in a similar manner for individual landuses. This aga i n ut il ized both GLS estimation with AR (1) and GLS CAR (1) in which area - weighted ETa monthly time - series obtained for the whole watershed while each landuse are pairwise compared. Again, the difference between the two analyzed time - series is used as the r espo ns e var iable . Finally, for the third approach, we computed the mean difference between the ETa datasets with respect to the E nsemble and SWAT model for each subbasin. In this case, the GLS - AR (1) regression method was used to perform an overall compa r ison , which repo rted the mean difference and p - value for each subbasin. As a result, a series of maps were created that represented the spatial variation of the mean differences with respect to the E nsemble and SWAT model ETa values. 95 5. 3 Results and Discu s sion 5 . 3.1 Tempo ral Statistical Analysis 5. 3.1.1 Monthly Analysis 5. 3.1.1.1 Overall Analysis Temporal cluster analysis was performed to determine if any of the ETa datasets produced similar results during specific times of the year. Table 5. 2 presents the mean m onthl y val ues of each ETa dataset for the entire Honeyoey watershed as well as any similarities between datasets with superscripted letters. When different datasets have the same superscript, it indicates that the mean difference of th e datasets is n ot sta tisti cally different from zero. Meanwhile, if two datasets have different superscripts, it indicates that while each dataset is similar to another dataset, the mean difference between them is statistically different from zero. And fina lly, if a data s et has no s uper script , it indicates that the mean difference of that dataset and all other datasets is statistically different from zero. As presented in Table 5.2 , similarities between datasets existed for all months, with TerraClimate (wi th mean monthl y value s ran ging from 1.40 mm to 110.67 mm) sharing similarity with the greatest number of other datasets overall. This indicates that the TerraClimate dataset serves as the middle ground between the different datasets, which could be due to the fact that T erraCl imate util izes a water - balance approach while the rest of the products utilize energy balances ( Abatzoglou et al., 2018 ). In addition , winter months, such as January and February, generally had fewer similarities between datasets, whil e summer month s , such as J une a nd July, had more similarities and more clusters. It is ETa datasets for which the mean difference is significantly not diffe rent from zero . This shows that there is a higher level of variability between the datasets in the winter months compared to 96 the summer months. This could be due to the challenges related to estimating ETa when snow cover and winter storms affect the regio n (Wang et al. , 2015) . Mea nwhil e, when considering the E nsemble dataset (with mean monthly values ranging from 9.41 mm to 115.59 mm), similarities with other datasets were seen for nine months out of the year (January, March, April, June, July, August, Sep tember, Octobe r , and Decem ber), with August showing the most similarity with five datasets identified as similar to the E nsemble. However, there was no consistent pattern for which datasets were found to be similar from month to month. This may be caused b y the variety o f ETa calcu latio n techniques used for the ETa products in the study, such as surface energy balances and water balances. Another interesting comparison is between the MOD16A2 1 km (with mean monthly values ranging from 15.12 mm to 100.06 mm) and MOD16A2 5 0 0 m (w ith m ean m onthly values ranging from 10.72 mm to 130.74 mm) datasets. While these two datasets are based on the same model, they were only found to be similar for only four months out of the year (January, February, March, and October) . This is like l y due to th e fac t that the 500 m dataset captures more of the landscapes spatial variability compared to the 1 km dataset. Regarding similarities between datasets throughout the year, no noticeable patterns were seen . This is likely due to t he fact that e a ch of the E Ta da tasets utilize different equations, approaches, and spatial resolutions. Another possible cause for the lack of patterns among the ETa datasets is the fact that this analysis is the summary over the entire watershed, and patt erns found wit h in spe cific land uses could be lost due to data aggregation at the watershed level. 5. 3.1.1.2 Landuse Analysis In order to determine if patterns among the ETa datasets were lost due to aggregation at the watershed scale, monthly analysis was also performe d for t he ma jor l anduses (agriculture, forest, urban, and wetland) as well as all of the individual landuses ( ALFA, CORN, FPEA, 97 FRSD, FRSE, HAY, PAST, SGBT, SOYB, URLD , UTRN, WETF, WWHT ). Tables S 5. 1, S 5. 2, S 5. 3, and S 5. 4 in the Appendix sho wing the mean mo nthly value s of each ETa dataset for agricultural, forest, urban, and wetland regions, respectively, with clusters identified with superscript letters. M eanwhile, Tables S 5. 5 to S 5. 17 in the Appendix show the same analysis fo r each individual landuse. Reg arding agri cultu ral regions (Table S 5. 1), similar results to the watershed scale analysis were obtained . Winter months had more unique dat asets and fewer clusters while summer months had more clusters and fewer unique dataset s. Furthermore, the TerraClimat e data set ( with mean monthly values ranging from 1.39 mm to 110.83 mm), on average, was found to be similar to more datasets. Meanwhile, t he E nsemble (with mean monthly values ranging from 9.34 mm to 115.56 mm) was found to b e similar to datasets more ofte n when cons ideri ng agricultural areas compared to the entire watershed, with eleven of the twelve months showing similarity to the dataset s. Among all of the datasets the E nsemble was found to be similar to the SWAT model out put the most with four months o ut of the y ear ( March, April, November, December), followed by a three - way tie between MOD16A2 500m (April, May, June), NLDAS - 2: VIC (June , July, August), and TerraClimate (March, August, September). SSEBop, ALEXI, and NLDAS - 2: Mosaic were similar to the E nsembl e for only two months each (July, August), (August and October), and (January, March) respectively; while NLDAS - 2: Noah was similar only during August. While these results do not produce a distinctive pattern for indivi dual ETa datasets, the general increa se in the number of similarities found in the months between March This ma y indicate that the presence of vegetation and fairer weather results in more agreement among the datasets. 98 When con sideri ng fo rest regions (Table S 5. 2), the pattern of fewer clusters and more unique ETa datasets in the winter and more clusters and fewer unique ETa datasets in the summer was not as apparent, though January and February sti ll had the most unique ETa data sets a nd fe west clusters. Meanwhile, all ETa datasets showed similarity with the E nsemble for at least one month, with June showing the gr eatest number of similarities with five of the eight datasets showing similarity. This is still aligned with the gener al pat tern seen at the watershed scale analysis. Furthermore, this pattern also matches the pattern seen for agricultural lands, in which the months during which canopy vegetation is present, in general, show more clustering and less variance. This may ind icate the p resen ce of vegetation improves ETa dataset convergence. While, weather conditions could still impact this finding, the fact tha t forest lands had more winter similarities than agricultural lands combined with the p resence of evergreen forests th at rem ain v egeta ted year - round makes this a possible conclusion. This is further supported by Tables S 5. 8 and S 5. 9 , deciduous and evergree n forests, respectively, for which in the winter months Table S 5. 9 (evergreen forests) showed more similarities than i n Tabl e S 5. 8 (de ciduous forests). Considering all of this , the importance of the presence of vegetation should be explored in further studies. Regarding urban regions (Table S 5. 3), again the pattern of a high number of clust ers during the growing season a nd few er du ring the winter months was observed; however, the pattern was less prominent. Given the observations that the p resence of vegetation plays a role in the number of clusters, this makes sense since urban regions tend to have fewer plants and more imperv ious surfa ces such as roads and buildings. However, when considering the number unique ETa datasets across the span of the year, more datasets showed similarity, especially in the winter months. This could be caused by the fact that urban regions exp erienc e les s sea sonal 99 variation. Clusters found with the E nsemble followed the trend of more similarity in summer months c ompared to winter months. Finally, regarding wetland regions (Table S 5. 4), similar results to the fores t regions was seen. However, th is mak es se nse s ince the wetland regions in this watershed are woody wetlands and thus there is a significant presence of trees. Overall, similar results to the overall analyses were seen, with winter months, such as January and February, having fewer clus ters a nd mo re un ique datasets, while summer months such as June and July, had more clusters and fewer unique datasets. How ever, analysis of the major landuse classifications showed that the presence of vegetation m ight result in less ETa dataset viability. Tabl e 5. 2 . Ave rage monthly ETa values for each dataset for the entire watershed with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. D ec. MOD16A2 1km 16.06 a 21.7 7 a 37 .24 a 37.9 9 59.82 a 83.85 a 100.06 a 83.16 a 44.02 25.29 a 20.22 a 15.12 a MOD16A2 500m 15.81 a 21.14 a 37.55 a 44.98 a 77.94 b 109.21 b 130.74 b,c 110.68 b 57.27 a,b 26.58 a 17.14 10.72 b SSEBop 0.03 0.01 10.3 8 b 26.82 b 50.02 c,d 92.37 a,c, d 117 .93 d 99.9 6 a,c 52.34 a,c 12.16 5.77 b 0.71 c NLDAS - 2:Mosaic 10.91 b,c 11.86 b 26.84 c,d 58.93 c 95.36 119.00 e 135.66 b 115.18 b 85.01 49.21 b 21.83 a 11.76 b NLDAS - 2:Noah 10.21 b 12.53 b 19.11 e 28.36 b 43.84 c 74.62 102.31 a 99.85 a,c 67.0 5 d 28 .46 10.55 7.36 d NLDAS - 2:VIC 7.61 9.77 10.19 b 15.40 48.11 d 89.21 a,c 116.70 d 97.60 a,c 50.54 c 16.37 6.00 b 7.14 d TerraClimate * * 18.00 b,c,d,e 81.94 101.65 110.67 b,e 97.78 a,e 87.19 a,c 65.24 a,b,c ,d 49.53 b 22.52 a 1.40 c ALE XI 22 .96 37.32 51.08 56.75 c 83.23 b 104.55 b,d 123.5 c 100.37 a,c 66.77 d 32.57 c,d 19.47 a 16.17 a SWAT 3.72 5.47 29.21 c 42.72 a 63.08 a 97.71 b,c,d 88.40 e 69.87 55.80 a,b,c 32.86 c 19.47 a 7.60 d,e Ensemble 11 .83 c 16.17 26.39 d 43.90 a 69 .99 9 7.94 d 115 .59 d 99.25 c 61.03 b 30.02 d 15.44 9.41 e *Note that no ETa values were provided for TerraClimate for the months of January and February. 5. 3.1.2 Seasonal Analysis 5. 3.1.2.1 Overall Analysis Temporal clus ter analysis was also performed on a season al b asis to determine if additional patterns among the ETa datasets could be identified . Table 5. 3 presents the mean 100 monthly seasonal values of each ETa dataset for the entire Honeyoey waters hed as well as any sim ilarities between datasets with super script ed l etters. During winter, the majority of the data sets were found to be unique with only two clusters identified, SSEBop (0.25 mm/month) and TerraClimate (1.09 mm/month) and Mosaic (11.51 m m/month) and the E nsem ble (12.47 mm/month), respectiv ely. R egardi ng s pring, only two datasets were found to be unique (MOD16A2 500 m (53.49 mm/month) and VIC (24.56 mm/month)) and four clusters were identified (MOD16A2 1 km (45.02 mm/month), SWAT (45.00 mm /month), and Ensemble (46.76 mm/month); SSEBop (29.07 mm/mo nth) a nd N oah (30.44 mm/month); Mosaic (60.38 mm/month) and ALEXI (63.68 mm/month); and TerraClimate (68.60 mm/month) and ALEXI (63.68 mm/month)). For summer, two datasets were identified as unique (Mosaic (123.28 mm/mo nth) and ALEXI (109.47 mm/month )) and four clus ters were found (MOD16A2 1km (89.02 mm/month), Noah (92.26 mm/month), TerraClimate (98.55 mm/month), and SWAT (85.33 mm/month); SSEBop (103.42 mm/month), VIC (101.17 mm/month), and TerraC limate (98.55 mm/month ); SSEBop (103.42 mm/month), VI C (101 .17 mm /mon th), TerraClimate (98.55 mm/month), and Ensemble (104.26 mm/month); and SSEBop (103.42 mm/month), TerraClimate (98.55 mm/month), and Ensemble (104.26 mm/month)). And finally, for fall, fo ur of the ten datasets were unique (MOD16A2 1km (29.8 4 mm/m onth), Mos aic (52.02 mm/month), TerraClimate (45.76 mm/month), and ALEXI (39.60 mm/month)) and three clusters were identified (MOD16A2 500m (33.66 mm/month), Noah (35.35 mm/month), and SWAT (36.04 mm/month); SSEBop (23. 42 mm/month) and VIC (24.30 mm/ month) ; Noah (35 .35 mm/month), SWAT (36.04 mm/month), and Ensemble (35.50 mm/month)). This is similar to the monthly analysis, in which winter and fall show fewer clusters and more unique datasets and sp ring and summer show m ore clusters and fewer unique d ataset s. Thi s is likely due to challenges such as 101 cloud cover and snow cover that occur during the fall and winter seasons (Wang et al., 2015). Meanwhile, unlike the monthly analysis, the E nsemble showed similarities for all seasons, though again no notice able p attern was seen in which datasets were found to be similar. Table 5. 3. Average seasonal ETa values for each dataset for the entire watershed with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1 km 17.65 45.02 a 89.02 a 29.84 MOD16A2 500m 15.89 53.49 116.88 33.66 a SSEBop 0.25 a 29.07 b 103.42 b,c,d 23.42 b NLDAS - 2:Mosaic 11.51 b 60.38 c 123.28 52.02 NLDAS - 2:Noah 10.03 30.44 b 92.26 a 35.35 a,c NLDAS - 2:VIC 8.17 24.56 101.17 b,c 24.30 b TerraClimate 1.09 a 68.6 d 98.55 a,b,c,d 45.76 ALEXI 25.48 63.68 c,d 109.47 39.60 SWAT 5.6 45.00 a 85.33 a 36.04 a,c Ensemble 12.47 b 46.76 a 104.26 c,d 35.50 c 5. 3.1.2.2 Landuse Analysis Si milar to the monthly analysis, major land us e and individual landuse seasonal analysis was performed to determine if patterns among the ETa datasets were lost due to aggregation at the watershed scale. Tables S 5. 18 through S 5. 21 show the mean seasonal val ue s of each ETa dataset for agric ultural, f or est, urban, and wetland regions, respectively, with clusters identified with superscript letters. Meanwhile, Tables S 5. 22 to S 5. 34 in the Appendix show the same analysis for each individual landuse. The results of the seasonal analysis for agri cultural l an ds (Table S 5. 18) shows more incidences of similarity compared to the overall analysis. In particular, this is true for the winter and fall seasons. In fact, for agricultural lands fall show ed the gr eatest number of clusters with four clusters a nd only on e unique ETa dataset (ALEXI (39.77 mm/month)). Meanwhile , winter, spring, and summer all had three clusters and winter showed the most unique ETa datas et s . All of this indicates that for agri cultural lands there is more agreement among the ETa da tasets at th e 102 seasonal level. However, while there is more agreement, there is still a lack of a pattern between the seasons. This is likely due to the various spatial resolutions (ranging from 0.5 km 2 to 12.0 km 2 ) and governing equations (e.g., Penman - Mon teith, ene rg y balance, water balance) used for the individual ETa products. When considering the forest regions within the Honeyoey watershed at the seasonal level, more unique ETa datasets were found compared to agricultural lands (Table S 5. 19). However, the numbe r of clusters for each season was identical to the agricultural lands. This is similar to the results found in the monthly analysis. However, the matching number of clusters per season with t he agricu ltural regions shows that, at the seasonal sca le, the fa ll season plays an important monthly section; but this could be explained by the fact that coniferous trees re main green year - round and that deciduous trees maintain ca nopy cover into the middle of the fall season. When considering the seasonal analysis for the urban areas (Table S 5. 20), even more unique ETa datasets were identified than for the forest and agri cultural lands. This indicates that there is m ore disagr ee ment between the different ETa datasets regarding the calculation of ETa in urban regions. This matches the pattern seen with the monthly analysis and supports the observation that the pres ence or l ack of vegetation plays a major role in ETa da taset agre ea nce. However, the spring and summer seasons still showed more similarities than the fall and winter seasons. This matches the pattern found in the monthly analysis, which indicates that sea sonal wea ther changes affect ETa dataset performance. I n addition , the summer season showed the greatest number of similarities among the ETa datasets and the E nsemble. This also matches the results observed in the monthly analysis. 103 Finally, when consider ing only wetlands regions with in the study area (Table S 5. 21), t he results seemed to be a mix of the agricultural and forest regions. The spring and fall seasons showed the fewest number of unique ETa datasets (NLDAS - 2: VIC (24.63 mm/month) and ALEXI (39. 20 mm/mon th) , respectively) followed by summer (MOD16A2 500m (124 .3 8 mm/month) , NLDAS - 2: Mosaic (117.63 mm/month) , and SWAT (77.87 mm/month) ) and then winter (MOD16A2 1km (18.65 mm/month) , MOD16A2 500m (15.58 mm/month) , NLDAS - 2: Noah (9.87 mm/month) , NLDAS - 2:VIC (8 .38 mm/month) , ALEXI (25.09 mm/month) , and SWA T (6.22 mm /m onth) ). This mixture of agricultural and forest results is interesting given the nature of the wetland regions, which are covered in both grasses and trees. Meanwhile, the s ummer season showed the g reatest number of similarities with the E nsemb le. This a ga in follows the pattern seen for the other landuse and monthly analysis. Overall, the seasonal analysis showed similar results to the monthly analysis. This confirmed that th e presence of vegetation plays a major role in the similarity between E Ta dataset s. Furthermore, spring and summer tended to show more similarities among the ETa datasets, while the fall and winter tended to have fewer similarities and more unique ETa data sets. This matches the we ather patterns found in the region and confirm s that clo ud cover and snow played a major role in ETa dataset variance. Meanwhile, there did not seem to be any noticeable patterns amongst ETa dataset similarities between the months and seasons. This is like ly due to the various accuracies and spatial r esolutions a ssociated with the individual ETa products. However, across the different landuses, ETa datasets tended to show similar patterns for specific months and seasons. This is lik ely due to similarities i n how the different ETa datasets were calculat ed, for ex am ple , Noah, Mosaic, SSEBop and ALEXI all utilize forms of 104 surface energy balances to calculate ETa while MOD16A2 and SWAT utilize Penman - Monteith techniques, and VIC and Terr aClimate utilize water ba lances. 5. 3.2 Spatial Statistical Analysis T he s patial s tatistical analysis was performed on the ETa datasets, E nsemble, and SWAT model outputs to determine how the different datasets performed across different landuses. The first step in this analysis w as to determine how the individual datasets pe rformed ac ro ss the different landuses found in the study area. This would highlight different landuses that generated similar ETa values as well as those landuses that produced unique ETa values for each ETa da taset. After this, a comparison among the diff erent data se ts was performed in order to determine if any of the dat asets showed similarities across different landuses. 5. 3.2.1 Landuse Distinction within each ETa Dataset 5. 3.2.1.1 Overview of Landuse Distin ction within each ETa Dataset Table 5. 4 prese nts the ov er all datasets averages with respect to the major landuse categories as well as the watershed scale average. Similarities between these regions are indicated with superscript letters. As can be seen, the MOD16A2 1 km dataset only showed similar E Ta values be tween forest (47.81 mm/month) and wet land (47.58 mm/month) areas, which makes sense given the nature of these landuses and could be explained by the accuracy (1.25mm/month) of the MOD16A2 1 km produ ct. However, when looking at the MOD16A2 500m dataset, n o similarities are seen. In fact, this is the only dataset to have this result. However, this is also the dataset with the highest resolution, which means it is better able to capture spatial variabil ity across the landscape. The SSEBop dataset s howed the sa me pattern as the MOD16A2 1 km datase t. This is interesting since both of these datasets reported ETa at 1 km resolutions, which could explain the similarity, especially since the SSEBop accuracy is 27.9 105 mm/month compared to the 1.25 mm/month f or the MOD 16 A2 1 km dataset. The NLDAS - 2 Mosaic d ataset showed similarity between agriculture (62.60 mm/month), forest (61.09 mm/month), and urban (63.28 mm/month) regions. This is interesting since these regio ns are considered to be quite different from e ach other es pecially in regard to vegetation cove r. This could be caused by either the aggregation of landscape data to the 12 km 2 scale used by the NLDAS - 2 datasets, the 6 mm/month accuracy, or the energy bala nce used to simulate ETa values for these land uses. When c onsidering the NLDAS - 2 Noah model, a clear distinction between agriculture (42.68 mm/month), urban (43.89 mm/month), and wetland (41.29 mm/month) areas is seen . However, forest (40.71 mm/month) regi ons were reported as similar to all of the oth er landuse s. This again could be due to issues wi th dataset resolution or it could indicate that despite the improvements made in the NLDAS - 2 Noah model (Xia et al., 2015) the dataset, in some regions, still fa ces challenges differentiating forest regions from other l anduses. The NLDAS - 2 VIC dataset had two clusters. The first showed the similarity between agriculture (40.14 mm/month) and wetland (40.37 mm/month) areas, which matches the pattern seen for the MOD 16A2 1km and SSEBop products; while the second cluster i nd icated similarity between urban (41.3 0 mm/month) and wetland (40.37 mm/month) regions. This again is interesting since urban areas and wetlands are considered to be different. However, this may be d ue to the spatial resolution of the NLDAS - 2 da tasets or th e 6.66 mm/month accuracy associated w ith the VIC product. Meanwhile, the TerraClimate product also showed two clusters. However, this time the first cluster included agricultural (66.72 mm/month), f orest (66.61 mm/month), and wetland (66.70 mm/ month) reg io ns, while the second cluster included forest (66.61 mm/month) and urban (67.01 mm/month) regions. While, the first set of similarities could be considered similar due to the high level of vegetation present and explained by the accuracy of 4.75 mm/month; t he 106 second set of similarities makes l ess sense. However, TerraClimate was the only ETa product that solely utilized a water balance approach with a spatial resolution of 4 km 2 , which could explain t he similarity seen here. Considering the ALEXI product, on e cluster of similarities was seen be tween agricultural (59.66 mm/month), forest (59.77 mm/month), and wetland (59.39 mm/month) regions. This is similar to the TerraClimate product and could be expl ained by the presence of vegetation. However, this may a ls o indicate that the 4 km 2 spatial res olution and 30.15 mm/month accuracy may prevent the ALEXI product from differentiating amongst different types of vegetation in this region. Meanwhile , the SWAT model output showed two sets of similar ETa va lues. The fi rst is for agricultural (43.50 mm/mon th) and forest (41.45 mm/month) regions, while the second was between agriculture (43.50 mm/month) and (44.68 mm/month) wetland areas. This matches other ETa pro ducts that found different vegetated landcover s to be si mi lar, which in the case of SWAT is lik ely the results of how the SWAT model calculates canopy cover and ETa from this surface ( Neitsch et al., 2011) . However, this could also be due to the aggregatio n of landuses at the subba si n level, since wet lands and fo rests are often surrounded by agricul tural lands in the region. Finally, when considering the dataset E nsemble, the same pattern as the MOD16A2 1 km and SSEBop products was seen, with forest (50.30 mm/month) and wetlands (50.39 mm/month) being considered a s similar. This could be explained by the fact that both landuses have tree cover. However, this was also a common paring amongst all of the datasets and given that the E nsemble is the average of al l of the datasets it makes sense that this sim ilarity wo ul d also be reported for this dataset. Overall, a number of similarities were identified . However, these could be caused by similarities in the landuses due to c hanges throughout the year. Therefore, the seasonal and monthly analysis was consider ed to furt he r examine these similarities and determine their cause. 107 Table 5. 4. Overall dataset averages for each major landuse category with clusters indicated b y superscripts for each column Landuse Dataset MOD16A2 1 km MOD16A2 500m SSEBo p NLDAS - 2: Mosaic NLD AS - 2:Noah NLDAS - 2:VIC TerraClimate ALEXI SWAT Ensemble Agriculture 43.92 52.60 37.89 62.60 a 42.68 a 40.14 a 66.72 a 59.66 a 43.50 a,b 49.39 Forest 47.81 a 59.30 40.89 a 61. 09 a,b 40.71 a,b,c 37.24 66.61 a,b 59.77 a 41 . 45 a 50.30 a Urban 41.68 49.48 35.40 63.28 a,b 43.89 b 41.30 b 67.01 b 56.83 34.25 48.46 Wetland 47.58 a 58.00 40.97 a 59.85 41.29 c 40.37 a,b 66.70 a 59.39 a 44.68 b 50.39 a 5. 3.2.1.2 Seasonal O verview of Landuse Distinction within each ETa Dataset T o determine if the similarities noticed in the overall analysis were related to specific times of the year, the seasonal analysis was performed for each dataset, with the results presented in Tables 5 .5 and S 5. 35 through S 5. 43. This analysis was also perfo rmed on a m onthly basis for the major landuse classes (Tables S 5. 44 to S 5. 53) and for individual landuses (Tables S 5. 54 to S 5. 63) with the results presented in the Appendix . Tables 5. 5 and S 5. 35 show the seas onal landuse comparisons for the MOD 1 6A2 produc ts. When co nsidering the 1 km dataset (Table 5. 5), forest (30.74 mm/month) and wetland (30.53 mm/month) regions were identified as similar in the fall season, while forest (18.02 mm/mo nth) and urban (17.98 mm/ month) regions were found to be simi l ar in the winter. The forest and wetland similarity match the overall comparison found in Table 5. 4 and could be explained by the fact that both forest and wetlands in this region have trees whi ch lose their leaves in t he fall as well as the accuracy of t h e dataset (1.25 mm/mo nth). However, the similarity between forest and urban regions in the summer was not seen in the overall analysis. This indicates that the distinction of winter clusters was lost at the overall year ly analysis. These clusters also ind i cate that the MOD16A2 1 km datasets should not be used to differentiate between forest and urban regions in the winter and forest and wetland regions in the fall with respect to ETa . Meanwhile , the spring and summer sea sons showed no similar ETa values am o ng the maj or landuse categories. This shows that the MOD16A2 1 km datasets is able to identify 108 between landuses for half of the year . On the other hand, when considering the 500 m dataset s (Tabl e S 5. 35), the same result seen in the overall analysis was re p orted, whi ch means th at no similarities between landuses was seen . This was the only dataset to show this pattern, which confirms that the higher resolution of this dataset was able to different iate between all the land uses for all times of the year. This makes this dataset id eal for isolating the ETa of individual landuses within the Honeyoey watershed. Table 5. 5. Average seasonal values of the MOD16A2 1km dataset for the entire waters hed and each major landuse category for each column Landuse Season Winter Spring Summ er Fall Agriculture 17.18 43.21 85.95 29.34 Forest 18.02 a 47.86 94.63 30.74 a Urban 17.98 a 44.17 76.14 28.43 Wetland 18.65 47.37 93.79 30.53 a Table S36 shows the seasonal landuse comparisons for t h e SSEBop dataset. For which, du ring the spring, summer, and fall seasons, forest (30.81 mm/month, 107.49 mm/month, and 24.80 mm/month, respectively) and wetland (30.99 mm/month, 107.57 mm/month, and 24.99 mm/month, respectively) regions were found to have similar E Ta values. This matche s the overall analysis results and is likely due to the similarities in the forest and wetland landuses. However, this also shows that this dataset is not ideal for differentiating between these landuses for the majority of t he year. Meanwhile, the spring season showed another set of similarities between agricultural (27.87 mm/month) and urban (27.55 mm/month) regions. This makes less sense given the nature of these landuses; however, it could be due to the dominance of agric u ltural l a nds in this region and the placement of agricultural lands near urban centers in the region as well as the accuracy of this dataset (27.90 mm/month). Another interesting note it the fact that the SSEBop dataset was able to differentiate between a l l of the landuses during the wi nter season. This is significant since winter offers the greatest challenges in calculating ETa 109 due to the lack of vegetation, increased cloud cover, and snow. Over all, the SSEBop dataset should not be used for most of the y e ar in th i s region, but it is id eal for winter landuse distinctions. Tables S 5. 37 through S 5. 39 show the seasonal landuse comparisons for the NLDAS - 2 products, Mosaic, Noah, and VIC, respectively . Interestingly each approach had its own strengths and weak n esses reg arding the different s easons. Overall the Mosaic dataset showed the most unique datasets through out the year, with differentiation of all landuse classes during both the spring and summe r (Table S 5. 37). While the winter and fall seasons only show e d similar ities between agricult ural (11.28 mm/month and 51.69 mm/month, respectively) and wetland (11.55 mm/month and 51.63 mm/month, respectively) regions. These results are different from what was seen at the overall analysis (Table 5. 4), which showed s i milaritie s between agricultural , forest, and urban regions. This shift at the seasonal level indicates th at distinctions seen at the seasonal level are lost when all the values are averaged . Over all, this shows that the Mosaic dataset performs better at t h e seasona l scale especially for spring and summer seasons. Following differentiating landuses, the Noah dataset was also able to fully differentiate between al l landuse classes during two seasons (winter and summer) (Ta b le S38). However, while the fal l season only showed one similarity (urban (34.52 mm/month) and wetland (3 4.47 mm/month)), the spring showed two similarities (agriculture (31.15 mm/month) and fore sts (29.82 mm/month) and forest (29.82 mm/month) and wetland (28.63 mm /month)). This was clo ser to the results seen in the overall analysis (Table 5. 4), which showed that the Noah product had difficulty distinguishing forest lands from other landuses. Fina lly, considering the VIC product, only one season (spring) s h owed full differentiation of th e landuses (Table S 5. 39). However, the remaining seasons each showed one c luster among the landuses, with 110 forest (8.52 mm/month) and wetland (8.38 mm/month) in the winter, agriculture (103.17 mm/month) and wetland (102.64 mm / month) in the summer, and agric ulture (24.04 mm/month), forest (23.87 mm/month), and urban (23.64 mm/mont h) in the fall. Again, these roughly match the findings from the overall analysis (Table 5 . 4 ), but also indicate that the VIC product performs best in the sprin g. Table S 5. 40 shows the seasonal landuse comparisons for the TerraClimate dataset. As can be seen, the only season for which each landuse class was successfully distinguished was spri ng. Meanwhile, each of the other seasons showed only one clu s ter of la nduse ETa values; with summer showing similarities between urban (97.91 mm/month) and wetland (9 7.99 mm/month) areas, fall showing similarities between agriculture (45.82 mm/month) and w etland (45.86 mm/month) regions, and winter showing similari t ies betwe en all landuse classes . However, the lack of uniqueness in the winter makes sense since the Terr aClimate dataset does not report any values for January of February. Meanwhile, the season al analysis included December, January, and February as the w inter mon ths. Therefore, the cl ustering of landuse classes in winter just reaffirms that the TerraClimate product should not be used for winter ETa values. However, it could be used successfully in the spring. Table S 5. 41 shows the seasonal landuse compa r isons for the ALEXI dataset, wh ich is the first ETa product to not have at least one season for which all landuse class could be distinguished. Instead each season showed one cluster among the la nduse classes; with winter showing similarities between fore s t (24.60 mm/month), urban (25.2 6 mm/month), and wetland (25.09 mm/month) regions; spring showing similari ties between agriculture (63.92 mm/month), forest (63.61 mm/month), and wetland (63.29 mm/ month) areas; summer showing similarities between forest (11 1 .09 mm/mo nth) and wetland (109. 96 mm/month) regions; and fall showing 111 similarities between agriculture (3 9.77 mm/month) and forest (39.76 mm/month) lands (Table S 5. 41). This is similar to the res ults found at the overall analysis for whic h agriculture, fo r est, and wetland regions were f ound to be similar for the ALEXI product. This could be due to issues with mixed pixels and a spatial resolution of 4 km 2 . However, it could also be influenced by t he fact that the ALEXI dataset also had the lowest accuracy a mong the ETa datasets, which co uld result in more error when trying to differentiate between landuses. Table S 5. 42 shows the seasonal landuse comparisons for the SWAT model dataset. The SWAT mo del output showed similar results to the Mosaic and MOD16A2 p roducts i n the fact that for bo th the spring and summer no similarities between landuse classes were seen . Meanwhile, the fall and winter seasons each showed one cluster of landuses, which in thi s case was agriculture (30.50 mm/month and 5.33 mm/month, re s pectively ) and urban (30.70 mm/ month and 5.27 mm/month, respectively) for both cases. This is different f rom the overall analysis reported in Table 5. 4, which showed urban as unique and agricultu re similar to both forest and wetland regions. The reduction in simila rities in the seasonal analysis indicates that aggregation to the overall analysis resulted in p oorer performance. Overall, the SWAT model is better able to distinguish among landuses in the spring and summer. This follows the pattern seen earlie r in the p aper, where spring and summer seasons showed more similarities, which indicates that once again the presence of vegetation plays a major role in determining ETa. Finally, Table S 5. 43 sh ows the seasonal landuse comparisons for the E nsemble. As ca n be seen , the E nsemble was able to successfully distinguish between landuses for winter, summer, and fall. Meanwhile spring shows similarities between agriculture (46.47 mm/month) and forest (47. 01 mm/month) and forest (47.01 mm/month), urban (47.19 mm/mo n th), and wetlan d (47.31 mm/mont h). However, the fact that the E nsemble was able to distinguish between landuses in 112 three of the seasons including winter shows that it was able to improve the over all ETa dataset performance in regard to landuse differentia t ion. Howe ver, t he landuse clust ers seen in the spring, indicate that the uniform weights applied in this study should be modified since the majority (5 out of 8) of the datasets were able to dist inguish between the landuses in the spring. A summary of th i s analysi s for all of the ETa p roducts can be seen in Table 5. 6. Overall, breaking the spatial analysis down to the seasonal scale improved the individual ETa product performances. This also allo wed for the identification of the best seasons to use each d a taset: MO D16A2 1 km: spring and summer; MOD16A2 500 m: winter, spring, summer, fall; SSEBop: winter; NLDAS - 2 Mosaic: spring and summer; NLDAS - 2 Noah: winter and summer; NLDAS - 2 VIC: spring; Terra Climate: spring; ALEXI: none; SWAT: spring and summer; Ensem b le: winte r, sum mer, and fall. T his shows that the best product for landuse distinction in the Honeyoey watershed was the MOD16A2 500 m, while the worst for the region was the ALEXI product. Furt hermore, analysis of the E nsemble showed that by averaging t h e ETa pro ducts it was possible to improve the performance and differentiate among the major landuses. 113 Table 5. 6. Summary of landuse and season differentiation for all ETa products used in this s tudy, X s mark conditions that could be differe ntiated by the product Datas et Spa tial Scale Temporal Scale Overall Winter Spring Summer Fall MOD16A2 1 km Agriculture X X X X X Forest X X Urban X X X X Wetland X X X MOD16A2 500 m Agriculture X X X X X F ores t X X X X X Ur ban X X X X X Wetland X X X X X SSEBop Agriculture X X X X Forest X Urban X X X X Wetland X NLDAS - 2: Mosaic Agricul ture X X Forest X X X X Urb an X X X X Wetland X X X NLDAS - 2: Noah Agriculture X X X Forest X X X Urban X X X Wetland X X NLDAS - 2: VIC Agriculture X X Forest X X X Urban X X X Wetland X X Terra Clim ate Agriculture X X Forest X X X Urban X X Wetland X ALEXI Agriculture X X Forest Urban X X X X Wetland X SWAT Agriculture X X Forest X X X X Urba n X X X Wetl and X X X X Ensemble Agriculture X X X X Forest X X X Urban X X X X Wetland X X X 114 5. 3.2.2 Landuse Similarities between ETa Data sets Statistical analysis comparing the ETa dat asets was also performed regarding t he entire watershed and the major landuse categories (Table 5. 7) and for all individual landuses (Table S 5. 64). This is similar to the temporal analysis and was performed to determine if any similarities existed between the ETa da tase ts when considering th e spatial distribution of landuses throughout the region. As can be seen in Table 5. 7, a variety of datasets cluster s were identified . However, unlike the temporal analysis, several of the clusters spanned multiple landu ses. For e xamp le, the MOD16A2 1 km ( 43.92 mm/month, 41.68 mm/month, and 47.59 mm/month, respectively), NLDAS - 2:Noah (42.68 mm/month, 43.89 mm/month, and 41.30 mm/month, respectively), NLDAS - 2:VIC (40.14 mm/mon th, 41.30 mm/month, and 40.38 mm/month, respect ively), an d SW AT model (43.50 mm/mon th, 34.25 mm/month, and 44.68 mm/month, respectively) datasets had similar ETa values across agricultural, urban, an d wetland regions. This is interesting since each of thes e datasets has different accuracies and spatial and tempo ral resolutions. However, the governing equations for each of these products is based on energy balances, which could explain why they produce d similar values across these landuses. Meanwhile, the AL EXI (59.57 mm/month, 59.77 mm/month, 56.84 mm/m onth, and 59.3 9 mm/month, respective ly) and TerraClimate (66.73 mm/month, 66.62 mm/month, 67.01 mm/month, and 66.70 mm/month, respectively) and the ALEX I (59.57 mm/month, 59.77 mm/month, 56.84 mm/month, and 59 .39 mm/month, respectively) and NLDAS - 2:Mosaic (62.61 mm/ mont h, 61.09 mm/month, 63. 28 mm/month, and 59.86 mm/month, respectively) ETa products were similar for all major landuses. The similarity betw een the ALEXI and Mosaic products can also be explained b y the fact that both use energy balances to cal culate ETa , wh ich is similar to the first cluster of ETa products. However, considering the TerraClimate/ALEXI 115 similarity, these products utilize differ ent governing equations (water balance vs . energy balance ) and have very different accuracies. However, they share the same spatial resoluti on. This in combination with the other clusters indicates that ETa product similarity can be obtained for datasets t hat utilize similar approaches; however, it is also possi ble to achieve this same result if datasets sha re a spati al r esolution. However, gi ven the spatial nature of ETa , this makes sense. On the other hand, at the watershed scale, there was a reduction in the number of datasets sets found to be similar, for exa mple at the watershed scale the MOD16A2 1 km da taset (45. 39 m m/month) was not consi dered similar to the NLDAS - 2: Noah (42.03 mm/month), NLDAS - 2: VIC (39.56 mm/month), and SWAT (43.00 mm/month) datase ts, which was seen when considering specific landuses. Th is shows that aggregation to the watershed leve l can resu lt i n the loss of similari ties among ETa products, which indicates that performing analysis for specific landuses improves overall product agr eement. Overall, the presence of recurring patterns among the ETa products when considering landuses sho ws that ag reem ent among the ETa prod ucts is possible and is influenced by spatially dependent variables such as landuse and governing equations. This ma kes sense given the spatial variability associated with E Ta; however, when combined with the lack of pat terns in t he t emporal analysis indic ates that landuse plays a more important role for ETa than seasonal variations, at least with respect to product agr eement. 116 Table 5. 7. Overall summary of average ETa values for each dataset for the entire watershed and e ach major land use category wit h clusters indicated by superscripts for each column Dataset Region Watershed Agricultural Forest Urban Wetlands MOD16A2 1km 45.39 a 43.92 a 47.82 a 41.68 a 47.59 a,b MOD16A2 500m 54.98 b 52.61 b 59.30 b,c 49.49 b 58.00 c,d S SEBop 39.05 c 3 7.89 40.90 d 35.40 c 40.98 a NLDAS - 2:Mosaic 61.80 d 62.61 c 61.09 b 63.28 d 59.86 c NLDAS - 2:Noah 42.03 a,c 42.68 a 40.72 d 43.89 a 41.30 a NLDAS - 2:VIC 39.56 c 40.14 a 37.24 e 41.30 a,c 40.38 a TerraClimate 66.71 b,d 6 6.73 b,c 6 6.62 c 67.01 e 66.70 d ALEXI 59.57 b,d 59.67 b,c 59.77 b,c 56.84 d,e 59.39 c,d SWAT 43.00 c 43.50 a 41.45 d,e 34.25 a,c 44.68 a,b Ensemble 49.75 49.39 50.31 a 48.46 b 50.39 b 5. 3.4 Subbasin - level Statistical Analysis 5. 3.4.1 SWAT Model Output Si mila r to the landuse analysis performed between all of the datasets, the spatial mean difference was also calculated between the eight remotely sensed ETa products and the SWAT model output at the su bbasin level. This analysis, presented graphica lly in F ig ure 5. 3, provides a spatial overview regarding how well the SWAT model was able to replicate the remotely sensed ETa datasets. As can be seen in Figure 5. 3, most of the subbasins in maps a), c), e), and f), MOD16A2 1 km, SSEBop, NLDAS - 2: Noah and NLDAS - 2 : VIC respectively, sh ow no difference in their ETa values with the SWAT model output. This matches the results seen in Section 3.2.2, which also showed that these datasets were more closely aligned wi th the SWAT model regarding individual landuse ETa valu es . Ho wever, while the previous analysis indicated that these datasets were similar in nature , the spatial subbasin analysis shows that the SWAT model output is over - or under - estimating different regi ons within the Honeyoey watershed. This shows t hat whil e the overall analysis showed agreement, it is important to take into account spatial variation within the landscape. On the other hand, when considering the other remotely sensed ETa datasets, most of the region shows that the SWAT model is undere stimatin g the 117 values reported by the MOD16A2 500 m, NLDAS - 2: Mosaic, TerraClimate, and ALEXI products. This also matches the earlier results of this study as well as the results of previous studies that showed that the SWAT model had a better fit with the SSEBop d at aset compared to the ALEXI dataset ( Herman et al., 2018 ). These results are further supported by subbasin level statistical difference/no difference presented in Figure S 5. 1 in the Appendix . 118 Figure 5. 3. Maps showing the mean difference between each ETa data set and the SW AT model output. Maps correspond to a) MOD16A2 1 km, b) MOD16A2 500 m, c) SSEBop , d) NLDAS - 2: Mosaic, e) NLDAS - 2: Noah, f) NLDAS - 2: VIC, g) TerraClimate , and h) ALEXI 5. 3.4.2 Ensemble The s ubbaisn level analys is was also perfor med compari ng the E ETa values to the eight remotely sensed ETa products and the SWAT model output. As can be seen in Figure 5.4 , the E nsemble was either under - or overestimating ETa va lues for all the datasets. 119 Interestingly the split as to which datasets w as under - or over estimated matched the split of those datasets that were either similar of different from the SWAT model output. With MOD16A2 1 km, SSEBop, NLDAS - 2: Noah and NLDAS - 2: VIC showing that the E nsemble overestimated Eta values, while comparisons to MOD1 6A2 500 m , NLDAS - 2: Mosaic, TerraClimate, and ALEXI products showed underestimation. However, this makes sense since the E nsemble was created by averaging all datasets used in this study. This would result in a dataset that fits the midd le ground b etw een a ll datase ts, which is the case here. This also explains why the E nsemble was found to be statistically different for the majority of subbasins for all datasets (Figure S 5. 2 in the Appendix ). 120 Figure 5. 4. Maps showing the mean differenc e bet ween e ach ETa dataset a nd the Ensemble. Maps correspond to a) MOD16A2 1 km, b) MOD16A2 500 m, c) SSEBop , d) NLDAS - 2: Mosaic, e) NLDAS - 2: Noah, f) NLDAS - 2: VIC, g) TerraClimate , h) ALEXI, and i ) SWAT model 5. 4 Conclusions Throughout the course of this study , stat is tical analysis was used to compare the performance of published remotely sensed ETa products in a region with no observed ETa data. Overall, temporal analysis of the datasets showed that there was no noticeable trend in similarities betw een s pecific data se ts at both mon thly and seasonal scales. However, a general 121 pattern was seen with summer and spring seasons showing more clusters among the datasets and fewer unique d atasets. Meanwhile, fall and winter seasons showed fewer clusters and m ore u nique ETa da ta sets. This is reflective of weather and vegetation cover trends within the region. Nevertheless, the lack of patterns among the datasets shows that temporal variation is less influential when compared to spatial variation . This likely du e to several fact or s such as spat ial resolutions. This was most clearly identified in the comparison of the two MOD16A2 products (1 km and 500 m). Despite both products utilizing the sa me approach and temporal resolution, the lack of similarity throughout the y ear can only b e attributed t o the impact of different spatial resolutions. Meanwhile, spatial analysis at both the watershed, landuse, and subbasin levels led to the identificatio n of two major clusters within the ETa datasets. With higher ETa values repo rted by MOD1 6A 2 500 m, NLDAS - 2: Mosaic, TerraClimate, and ALEXI; and lower ETa values reported by MOD16A2 1 km, SSEBop, NLDAS - 2: Noah, NLDAS - 2: VIC, and SWAT. These clusters were c onsistent across different landuses. This highlights two major points . F irst , there is l ot s of variance among the different remote sensing ETa products, which is driven by the use of different governing equations, spatial and temporal resolutions, and accu racies. However, the second point is that it is possible to find simila r ETa time series a cross differen t remote sensing ETa products, this is driven by the use of similar governing equations and spatial resolutions. However, it is important to note whic h datasets should be used when. Overall, the ETa product that was able to di fferentiate am ongst all of t he major landuses for all seasons was the MOD16A2 500 m dataset. However, all of the other datasets, except for ALEXI, were able to differentiate betwee n landuses for at least one season. Therefore, based on the analysis pe rform ed in this s tu dy the recomme nded seasons for each ETa product are: MOD16A2 1 km: 122 spring and summer; MOD16A2 500 m: winter, spring, summer, fall; SSEBop: winter; NLDAS - 2 Mosaic: spr ing and summer; NLDAS - 2 Noah: winter and summer; NLDAS - 2 VIC: spring; T erraC limate: spri ng ; ALEXI: none; SWAT: spring and summer; Ensemble: winter, summer, and fall. This can also help stakeholders, policy makers , and researchers select the best ETa datase t for different tasks such as monitoring of agricultural lands or track ing d eforestation . However, this study was performed for only one watershed ; future studies should be performed to expand this analysis to different climatological zones. This would hel p improve our understanding of how each ETa product performs across the glob al landscape a nd which one s hould or should not be recommended for a different time or landuse help e nsure that the correct ETa dataset is selected. Furthermore, other ensembling t echniques should be performed to identify the best for different region s. 5 . 5 Acknowled gm ent Authors wo uld like to thank Dr. Martha C. Anderson from USDA - ARS Hydrology and Remote Sensing Laboratory at Beltsville, Maryland and Dr. Christopher R. Hain from NASA Marshall Space Flight Center at Huntsville, AL for his help in providin g ALEXI data . This work is s upported by the USDA National Institute of Food and Agricultur e, Hatch project MICL02359. 123 6. EVALUATION OF MULTI AND MANY - OBJECTIVE OPTIMIZATION T ECHNIQUES TO IMPROVE THE PERFORMANCE OF A HYDROLOGIC MODEL USING EVAPOTRANSP IRATION REMO TE SEN SING DATA 6 .1 Introduction Unchecked anthropogenic activities have led to the degradation of natural systems that are vital to society and life as we know it. In freshwater supplies in combi nation with th e in creasing d emand for freshwater have made freshwater monitoring and water resources sustainability a major focus for researchers worldwide (Gleick, 1993; Srinivasa n et al., 2017; Haddeland et al., 2014). This requires the collection of dat a describing h ow different c omponents of the hydrological cycle change across space and time. Traditionally, this has been accomplished through the use of a variety of monitoring s tations, which are able to collect highly acc urate measurements of different components of the hydrologi cal cycle (USGS, 2018). However, monitoring stations are often expensive to install and maintain and are unable to provide the spatial resolution needed for large - scale analysis (Wanders et al., 20 14). This has led to the introd uction and u se of hydrologic al models (Einheuser et al., 2013). Hydrological models are fast, inexpensive, and versatile tools for researchers compared to monitoring stations. Howe ver, due to the fact that no model can perfec tly characterize all elements w ithin a wate rs hed , a level o f uncertainty is associated with all modeling practices (Kusre et al., 2010). One way to improve model performance and mitigate model uncertainty is th rough the use of model calibration and valida tion . Due to the complex nature of the hydr ol ogical cycle, hydrological models utilize hundreds of parameters to describe the natural world, each with a default value assigned by the model. However, the default value often does not represent the 124 real - world conditions; therefore, the par ameter value s need to be adj usted to improve model performance (Rajib et al., 2016). This is accomplished by modifying the parameter values and comparing the model output to observ ed data. In hydrological modeling , this is tr aditionally done by comparing s imulated and o bserved stream flows and using statistical criteria to test model performance (Wanders et al., 2014). However, since hydrological models are used to simulate other ele ments of the hydrological cycle, using just o ne element in the calibration p rocess could r esult in poor performances in other hydrological components, which reduces the overall model performance (Wanders et al., 2014; Rajib et al., 2016). Therefore, it is important to consider additional hydrological elements in the model calibrat ion process (C row et al., 20 03). When considering other hydrological element s, evapotranspiration (ET) stands out as an ideal addition to model calibration, since it describes the face , which in turn drives weat her patterns ( Pan et al., 20 15). In fact, the use of both ET and streamflow in hydrological model calibration has been the focus of recent research, which showed that global model performance was improved by the inclusion of ET (Herman et al., 2018). Howev er, while th is help s mitigat e model uncertainty, models are still dependent o n the quality and quantity of data available (Nejadhashemi et al., 2011). One solution to this is the u se of remotely sensed products. Remote sensing is the use of sensors and ima ging equipme nt to indirectly measure the characteristics of an object (Graham, 1999). Which when coupled with satellite technology has resulted in the development of many global mo nitoring datasets that can be used to measure element s of the hydrological c ycle ( Long e t al., 2014). In particular, remote sensing has become a source of monitoring data for actual evapotranspiration (ETa), which describes the actual loss of water from bo th evaporation and transpiration (USGS, 2016 d ). A variety of remotely sensed ETa product s have already b een 125 developed including the Simplified Surface Energy Balance (SSEB) (Zhang et al., 2016), the Atmosphere - Land Exchange Inv erse (ALEXI) (Anderson et al. , 2007; Senay et al., 2013), the Moderate Resolution Imaging Spectroradiomet er (MODIS) G lo bal Evapotrans piration Project (MOD16) (Zhang et al., 2016; NTSG, 2018), the Google Earth Engine Evapotranspiration Flux (Google, 2018), and the North American Land D ata Assimilation Systems phase 2 (NLDAS - 2) (Xia et al., 2015). Each of which ha s differe nt inputs and te mporal and spatial resolutions, and methodologies, which can make it challenging to know , which product to use when. Furthe rmore, it is important to not e that while remotely sensing helps the issue with improving data quantity a nd availabil it y, it does not directly solve the issue of data quality. However, one way to mitigate the uncertainty associated it remotely sensed produ cts is the use of ensemble te chniques, which aim to combine the benefits of each product while accounting for the ir l im itations (Diet terich, 2000; Duan et al., 2007). Here again, a variety of different techniques have been developed ranging from very simpl e calculations to complex mod eling approaches (Lee et al., 2017; Wang et al., 2018). Furthermore, some te chniques req ui re the use of accurate observed data to determine , which remotely sensed products are more accurate for the region of study (Kim et al., 2015). Overall, the wide range of techniques and remotely sensed products are available for hydrolog ic model cal ib ration; howeve r, there is lack of study on comparison among Eta remotely sensed products on the improvement of hydrologic model performance which is the goal of this study. Therefore, the objectives of this study are to 1) compare the perfor mance of ind iv idual remotely sensed ETa products and an ensemble through the use of a multi - objective calibration process and 2) explore the use of a many - objective calibration s th at takes into account multiple remotely sensed ETa products and streamflow. 126 6. 2 Methodol og y 6. 2.1 Study Area For this study , the Honeyoey Creek - Pine Creek Watershed (Hydrologic Unit Code 0408020203) located about the middle of the Lower Peninsula of Michig an (USA) was used to evaluate the applicability of remote sensing products t o improve th e overall perfor mance of hydrologic models (Figur e 6. 1). This watershed is a part of the Saginaw Bay Watershed and has a final outlet to Lake Huron. Covering approximat ely 1,100 km 2 , the region is predominantly used for agriculture, with about 52% of the l an d devoted for crop production. After agricultur e , the next major landuse is forests (~23%), wetlands (~17%), pasturelands (~5%), and finally urban (~3%). This region is ideal for testing the remote sensing products since there is a lack of s patial monit or ing data in th e area that can be used to setup a hydrologic model . However, streamflow is monitored on a daily basis at t he outlet of the watershed by United States G eological Survey (USGS) station (USGS, 2016d) and National Climatic Data Cen ter (NCDC) h as two stations in the region that measure daily precipitation and temperature (NCDC, 2018). However, there is no data available in the watershed or its surrounding are a for ETa. In fact, the closest source of observed ETa data are the AmeriFlu x stations l oc ated about 116 km aw ay from the Honeyoey watershed (AmeriFlux, 2018). All of this shows that remote sensing could serve a vital role in this region by providing consi stent datasets for monitoring of the hydrological conditions. 127 Figure 6. 1. Map of the Ho neyoey watersh ed 6. 2.2 Hydrological Model In order to evaluate the hydrological cycle in the Honeyoey watershed, the Soil and Water Assessment Tool or SWAT was select ed to be the hydrological model. SWAT is a commonly used hydrological model that is time c ontinuo us and semi - distributed and was developed by the USDA Agricultural Research Service and Texas A&M AgriLife Research (Texas A&M University, 2017). The SWAT mode l is able to simulate a variety of different hydrological process and scenar ios by takin g into ac count regional characteristics such as the climate, topography, soil properties, and landuse (Gassman et al., 2007). Relevant to this study is 128 the way in which SWAT simulates ETa. The first step is calculating potential evapotranspirat ion, which i n SWAT ca n be d one with three different techniques: 1) Penman - Monteith, 2) Priestley - Taylor, and 3) Hargreaves. The default method selected is the Penman - Monteith Metho d as the default (Neitsch et al., 2011). After calculating potential evapotr anspiration, t he SWAT model takes into account evaporation and transpiration from several sources including the evaporation from rainfall intercepted by the canopy, maximum transpi ration, maximum soil evap oration, and sublimation (during periods of snow co ver) (Neitsc h et al., 2011) . These are calculated at the hydrologic response unit scale, which in this study is the subbasin scale. Each subbasin has unique physiographical charact eristics , and for this st udy , the Honeyoey watershed was divided into 250 su bba si ns , due t o limitations in the number of unique landuses that could be applied within the SWAT model. Nevertheless, by calculating the potential evapotranspiration and taking i nto account the sources o f evaporation and transpiration at the subba si n lev el, the SWAT m odel is able to report monthly ETa values for the entire region . In order to develop the SWAT model for the Honeyoey watershed, several spatial and temporal datasets were used . This included topography, landuse, soil characteristics, climato logical cond it ions, and cro p management practices. For regional topography, the 30 m National Elevation Dataset from the USGS was used to calculate watershed slope (NED, 2014). Mea nwhile, 30 m landuse data was obtained from the 2012 Cropland Data Layer , wh ich was deve lo ped by the Un ited States Department of Agriculture (USDA) - National Agricultural Statistics Service (NASS, 2012). Regional soil characteristics were obtained on a scal e of 1:250,00 0 from the Natural Resources Conservation Service (NRCS) Soi l S urvey Geogra ph ic Database ( NRCS, 2014). Climatological conditions (precipitation and temperature) for the period of 2003 to 2014 were obtained from four National Climate Data Cente r stations (two temperature 129 and two precipitation stations ) (NCDC, 2018). Al l other clim at ological cond itions (e.g., wind speed, solar radiation, and relative humidity) that are required by the SWAT model were provided using a stochastic weather generator called WXGEN (Sharpley and Williams, 1990; Wallis and Griffiths, 1995; Ne its ch et al., 2 01 1). Crop mana gement practices, which included operations, schedules, and crop rotations, were adopted from studies the utilized the SWAT model in the same region to a ccount for local practices (Love and Nejadhashemi, 2011). A predefined su bba sin map with a scale of 1:2 4,000 was obtained from the National Hydrology Dataset P lus (NHDPlus) and the Michigan Institute for Fisheries Research and then modified to make a layer with 250 subba si ns . In order to perform the calibrations (which refers t o b oth calibrat io n and validat ion) in this study, observed streamflow data was obtained from the Pine River USGS station located at the outlet of the Honeyoey watershed for the period from 2003 to 2014 (USGS, 2016 p ). For this period the first two years (20 03 - 2004) were u se d for warm - up , the next five years (2005 - 2009) were used for model calibration, and the last five years (2010 - 2014) were used for model validation. 6. 2.4 Remote Sensi ng Actual Evapotranspiration Products Given the lack of observed ETa data in the region an d in order to calibrate the SWAT model, eight different remotely sensed ETa products were obtained : 1) the USGS Simplified Surface Energy Balance (SSEBop), 2) the Atm osphere - Land Exchange Inverse (ALEXI), 3) the MODIS Global Evapotranspiratio n Project (M OD 16A2) 500m, 4 ) the MOD16A2 1 km, 5) the North American Land Data Assimilation Systems 2 Evapotranspiration (NLDAS - 2) Mosaic, 6) the NLDAS - 2 Noah, 7) the NLDAS - 2 Varia ble Infiltration Capacity (VIC), and f inally 8) TerraClimate. Each of these products uti li zes different inputs and techniques the resulting products have different spatial and temporal resolutions. The following is a brief overview of 130 each ETa product as w ell as each spatial and temporal resol ution, while a summary of these datase ts is provid ed in Table S 6. 1 ( Appendix ) . The SSEBop ETa product was developed by the USGS and utilizes a simplified energy balance to calculate ETa on a monthly basi s for the Contiguous United States (Se nay et al., 2013). This is accomplished by calculatin g ET fractio ns from 8 - day, 1 km MODIS thermal imagery , which are then aggregated to a monthly scale (Senay et al., 2013; Velpuri et al., 2013). The next ETa product (ALEXI) was the product of a joint pr oject between the USDA and the National Aeronautics a nd Space Adm in istration (NA SA) and is also based on an energy balance. However, instead of ET fraction, ALEXI utilizes daily changes in surface temperature, obtained from Geostationary Operational Enviro nmental Satellites, and relates to surface water loss or ETa (And er son et al., 2 007). The resulting product reports ETa on a daily time step at a 4 km spatial resolution for the Contiguous United States (Hain et al., 2 015). The next two ETa products (MOD1 6A2 500m and MOD16A2 1 km) utilize the same methodolo gy but have di fferent spati al resolutions. These products were the result of a joint project between NASA and the University of Montana Numerical Terradynamic Simula tion Group (NTSG, 2018). ETa is calcul ated by using an ET algorithm that is based on the Pe nman - Monteit h equation and also requires MODIS landcover, the fraction of photosynthetically active radiation/leaf area index, and global surface meteorology (Mu et al. 2011; NASA, 2018a,b). The results are global 8 - day ETa products at 500 m or 1 km spatia l resolution s depending on the inputs used (NASA, 2014). The next three ETa products (Mosaic, Noah, and VIC) were developed as a joint project between National Ocean ic and Atmospheric Administration (NOAA) and the National Centers for Environmental Predict ion (NCEP) E nv ironmental Mo deling Center, NASA's Goddard Space 131 Weather Service Office of Hydrological Development, and the NOAA/NCEP Climate Prediction Center and are pa rt of the No rt h American La nd Data Assimilation System (N LDAS) project (NASA, 2018c). Each of the products utilize s a different land surface model to take into accou nt a variety of factors such as atmosphere interactions of water and energy, vegetation and soil moistu re heterogeneit y, water and energy budgets, a nd rainfall - runoff and water storage (Xia et al., 2015). The ETa products that result from these models repo rt ETa at both hourly and monthly time steps with a spatial resolution of 1/8 degree or 12 km for the e nt irety of Nort h America (Long et al., 2014). The last ETa product (TerraClimate) utilize d a water - balance model and was the result of a joint project be tween University of Idaho, the University of Montana, and the USDA Forest Service Rocky M ountain Rese ar ch Station. T he water - balance model utilize d by TerraClimate is based on the one - dimensional modified Thornthwaite - Mather climatic model (Abatzoglou et al., 2018 ), w hich resulted in a global ETa product that has a monthly time step and a 4 km spatial res ol ution (Abatzo glou et al., 2018). As discussed in the introduction, while remote sensing provides access to global spatially distributed da tasets, it also has more uncertainty associated with it. Therefore, during the development of each of t he ETa products used in this study, extensive calibration and validation w ere performed based on the observed data. The accuracies of these products can also be seen in Table S 6. 1 . However, it is important to note that these accuracies are base d on compari so ns to specifi c locations where obs erved data was available. And given the nature of ETa, these accuracies could flux across the landscape. This means t hat for each dataset it may perform better, equal, or worse in any other locations such as the Honeyoey w atershed. How ever, it is 132 important that the goal of this study is not to perform revalidation but to explore modeling applications of remote sensing ET a products to improve the performance of physically - based hydrologic models in regions lack ing observed d ata. This mea ns that while the model calibrations cannot confirm the best ETa product to use globally, it can highlight those that perform better than others in a region . This can be measured by comparing the level of improveme nt in the model predictabil it y of streamfl ow using different E T a products since the observed streamflow data are more available than observed ETa . Nevertheless, an ensemble of the ETa products is also u sed in this study for the model calibration to help re duce the u ncert ainty level as sociated with the ETa products. Concerning techniques for ETa product ensembleing Bayesian Model Averaging (BMA) is the commonly used technique (Kim et al., 2015; Tian and Medina, 2017; Yao et al., 2017; Ma et al., 2018). BMA reduces overall product unce rt ainty by dete rmining weights for each ETa product by comparing them to observed d ata (Kim et al., 2015). However, when the observed data is not availab le, as this is the case in this stu dy, an averaging technique can be used (Tebaldi and Knut ti, 2007). 6 . 2.5 Calibrat ion Techniques We implemented a multi - variable calibrati on approach in order to account for multiple sources of information de scribing both streamflow and actual evapotranspiration variables. The overall calibration p rocess consi st ed in 1) proc essing remote sensing products to obtain monthly ETa time series for the entire study area, 2) selecting model calibration parameters, 3) defining objective functions for each variable, 4) formulating the multi - objective optimiza tion problem t o solve, 5) s electing and implementing a suitable multi - and many - objective optim ization algorithm s and 6) selecting the best trade - off solution for an alysis purposes. 133 For this study, we implemented two different strategies in the formulation of the mult i - variable cali bration process. In the first st ra tegy , we implemented a multi - objec tive optimization approach in which the pr eformance of a hydrological model was evaluated using individ u al ETa remote sensing products while in the second strate gy, we used a many - objectiv e optimization algorithm in order to evaluate the performance of a h ydrological model when all ETa remote sensing products were simentensi nously considered used. For both strategies, we opted for implementing optimization approac h es , avoidin g the subjectiv e formulation of a single aggregated objective function. In addition , multi - and many - objective framework s provide a set of optimal soluti ons describing the trade - offs between streamflow and ETa performances while incorporating e xte rnal sour ce s of informat ion describing both variables. In this study, we employed the recently proposed evolutionary optimization algorithm Unified Non - dominate d Sorting Genetic Algorithm III (U - NSGA - III) (Seada and Deb, 2016) , which is ca pab le of sol vi ng different types of problems (i.e. , single , multi, and many - objective). Furthermore, in order to compare the resulting Pareto - optimal solutions, we s elected the best trade - off solution by employing the compromise programming approach. A det ail ed descri pt ion of the ca libration approach is presented below. 6.2.5.1 Data processing During the calibration process, the observed data consisted of eight raster - based remote sensing products for ETa, and a unique observed streamflow daily time series fro m a USGS ga uging station was comprised of a collection of images, each of those representing a snap shot of ETa over a specific time step (from daily to monthly, depending on the dataset). Fo r e ach image , we obtained t he average ETa value for each subbasin within the study area. This was done by u sing a w eighted area averaging 134 technique for each subba s in (Srinivasan and Arnold, 1954) . Weighted area averaging was used since it was able to acco unt for the mu ltiple pixels and partial pixels within each subbasin. To do this each E Ta product was resampled to a cell size 10 m and then area weighted averaging w as used for all the 10 m cell s within a s ubbasin ( Brown, 2014 ) . This resulted in a time se ries of aver ag e ETa values for each subbasin. Then since the ETa remote sensing products are also varied temporally, they w ere aggregated on a monthly time step for each subbasin . Finally, for each dataset, we computed a monthly area - weighted average time series for t he entire study area, as follows: (6.1) where , is the area - weighted ETa value of the i th monthly record, is the total watershed area, is the area of the subbasin w , is the ETa value of the i th monthly record for the subbasin w , and N is the total number of subbasins. The times series used for both str eamflow and ETa, were obtained for the period of 2003 2014. 6.2.5.2 Calibration parameters Sinc e the goal of the study is to simultaneou sly improve the SWAT model predictability concerning streamflow and ETa , relevant parameters affecting these elements of the hydrological cycle need to be identified and adjusted during the calibration process. Base d on the literature review, the SWAT mode l documentation, and sensitivity analysis; 18 parameters were selected for this study (Arnold et al., 2012; Woznicki a nd Nejadhashemi, 2012). These parameters are: baseflow recession constant (ALPHA_BF), biological mixing efficiency (BIOMIX), maximum canop y storage ( CANMX value for the main channel (CH_N2) , moisture condition II curve number (CN2), plant uptake compensation factor (EPCO), soil evapora tion compensation coefficient (ESCO), fra ction of 135 maximum stomatal conductance corresponding to the second point on the stomatal conductance curve (FRGMAX), ma ximum stomatal conductance at high solar radiation and low vapor pressure deficit (GSI), delay ti me for aquifer recharge (GW_DELAY), revap coefficient (GW_REAP), threshold water level in shallow aquifer for base flow (GWQMN), aquifer percolation coefficien t (RCHRG_DP), threshold water level in shallow aquifer for revap ( REVAPMN ), available water capac ity (SOL_AWC), surface runoff lag coeffic ient (SURLAG), and the vapor pressure deficit corresponding to the second point on the stomatal conductance curve ( VPD FR ). The minimum, maximum, and default values for all these parameters are presented in Table 6.1 . Table 6.1. SWAT parameters considered during the model calibration and validation process Parameter Minimum Value Maximum Value Default Value ALPHA_BF 0 1 0.048 BIOMIX 0 1 0.2 CANMX 0 100 0 CH_K2 - 0.01 500 0 CH_N2 - 0.01 0.3 0.014 CN2 - 25% 25% Various EPCO 0 1 1 ESCO 0 1 0.95 FRGMAX 0 1 Various GSI 0.001 0.05 Various GW_DELAY 0 500 31 GW_REVAP 0.02 0.2 0.02 GWQMN 0 5000 1000 RCHRG_DP 0 1 0.05 R EVAPMN 0 1000 750 SOL_AWC - 25% 25% Various SURLAG 1 24 4 VPDFR 1.5 6 Various 6.2.5.3 Objective functions For each variable, we formu l ated a minimization objective function ( f ) based on the Nash - Sutcliffe efficiency NSE , as follows: 136 (6.2) (6.3) where , is the i th observation of the considered variable (i. e. , s treamflow or ETa), is the average of the observed data, is the i th simula ted value of the considered variable, and n is the total number of observations. The range of the resulting objectiv e functions spans from zero to infinity, where zero represents a perfect fit between simulated and observed time series. I t is worth noting that we used daily time series f or streamflow, whereas for ETa, we used a monthly time step. The monthly ETa time se ries obtained for each remote sensing product were considered as the observed ETa data. For each simulation in the optimiz ation process, w e computed only one OF for streamflow using the available observed dataset at the outlet of the study area, and as man y OF for ETa as the number of ETa datasets. 6.2.5.4 Optimization strategies We implemented two c alibration strategies to e valuate the influe nce of the different ETa datasets in the prediction of daily streamflows. In the first strategy (multi - objective opt imization) , we formulated several multi - objective optimization problems to simultaneously minimize the difference between observed and simul ated time series for both streamflow and ETa. For each multi - objective problem, we used a different ETa dataset. Mor eover, in this strategy , we formulated an additional multi - objective optimization problem employing an ensemble ETa datase t. This ensemble d ataset was computed by averaging the monthly values of the ETa time series from each individual remote sensing product. As a result, for the first strategy we obtained as many Pareto - optimal fronts as the number of ETa datasets used in th is study. Each opt imization problem was formulated as follows: (6.4) 137 wher e , F is a vector composed of multiple objective functions, is a vector containing values for p model calibration parameter s, is a p - dimen sional parameter space limited by the calibration ranges for each model parameter, is the objective function evaluated for streamflow, and is the objective function evaluated for ETa. In the second strategy (many - object ive optimization) , we simultaneously minimized all the objective functions derived from each ETa datasets and for the streamflow variable . In this strategy , we did not include the ensemble dataset in order to avoid the addition of redundant information int o the overall opti mization process. Hence, this strategy results in one Pareto - optimal front. The many - objective optimization problem was formulated as follows: (6.5) where , is the objective function evaluated using the m th ETa dataset, and M is the total number of ETa datasets. 6.2.5.5 Multi - objective optimiz ation algorithm The U - NSGA - III algorithm is an extension of the recently proposed NSGA - III algorithm (Deb and Jain, 2014) . Th e original NSGA - III is a population - based, elitist procedure based on reference directions , which uses non - domination sorting and evolutionary operators ( i.e. , crossover and mutation) to move towards an optimal Pareto front. Reference directions are vector s that evenly fill the objective space. The algorithm uses these vectors to rank the diversity of individuals (Deb & Jain, 2014) . Moreover , these vectors are normalized by default in order t o achieve an equally diverse optimal Pareto front with respect to each objective function. NSGA - probabilities, and distributions indices associated with the cro ssover and mutation operations. 138 This algorithm has been found to reduce its performance when working with two or one objective functions. However, by incorporating an explicit selection procedure when scaling down to two and single - objective problem s, the U - NSGA - III algorithm is capable of solving single - , multi - (i.e. two to three objective functions) and many - objective (i.e. more than three objective functions) optimization problems without adding extra parameters (Seada and Deb, 2016) . It is worth noting that prioritizing streamflow and ETa for the many - objective optimization strategy described in the previous section, posed an interesting challenge. By default, U - NSGA - III equally prioritize all the objectives. Therefore, ETa calibration holds most of the total weight of the overall search for t his strategy since here we have eight ETa remote sensing products . Likewise, optimization under these default settings may result in a poor calibration performance for streamflow. To improve perfor mance , the balance of weights along each objective was modi fied by manipulating reference direction vectors in order to award the same amount of weight to the streamflow calibration objective function as all the ET a objective functions together. To modify the weights , the set of ( M +1) - dimensional reference direct ions is simplified into a two - dimensional reference direction set. The first dimension represents the weight given to all of the M ETa objectives, and the second dimension represents the weight giv en to the streamflow objective. The reference directions ar e created by the Das - Denis method that generat e s normalized reference direction vectors (Das & Dennis, 1998) . At this stage, the reference directions can be represented in the following matrix: 139 where r xy is the y th dimension fo r the x th reference direction, and d is the number of reference directions to create. Since the objectives are equally weighted according to the Das - Denis method, the following relationship must be attained : (6.6) Th en, we generate a d × ( M +1) reference direction matrix out of the d ×2 matrix by splitting the first dimension (representing ETa) into M different weights for each of the ET a objective functions: where , w z is z th weight for each of the ET a objectives, such that , . Since all ETa datasets will be equally weighed, i. e. w z = 1/ M z , and because , the sum of the elements of each row in the d × ( M +1) matrix showed above must be equal to 1: (6.7) Therefore, here we developed d ( M +1) - dimensional reference directions, where the first M dimensions together have the same weight as the ( M +1) th dimension. These reference directions then used in U - NSGA - III to provide the same weight for the M objective functions for ETa as for the stream flow objective function. The java code is implementing the U - NSGA - I II algorithms was provided by the Computational Optimization and Innovation (COIN) Laboratory at Michigan State University. This code was adapted for calibrating the SWAT model for this stu dy . To the best of our knowledge, this is the first time that the U - NSGA - III algorithm is used for hydrologic model calibration using both streamflow and ETa remotely sensed products . 140 6.2.5.6 Best trade - off solution In order to obtain the best trade - off calibration solution considering both streamflow and ETa, in this st udy, w e employed the compromise programming approach using the metric ( Zeleny & Cochrane, 1973) . This allows selection of an individual point from each Paret o - optimal front attained after implementing the multi - objective optimization strategies. The metric was computed for each member i of the Pareto front as fo llows: (6.8) where , j is an index identifying each objective function f ; N is the total number of Pareto - optimal points, is the vector containing the ideal point coordinates, which is an unfeasible solution located outside the Pareto front, representing the best e xpected objective function values (in this case, zero); and is the vector containing the nadir po int coordinates, which is comprised of the worst objective function values obtained for each dimension at the optimal Pareto front. The point with the minimum metric (i.e. , closest to t h e ideal point) is selected as the best trade - off solution. In addition to the best trade - off solutions of the multiple Pareto fronts, we also calculated the worst expected solutions from a common nadir point for all the Pareto - optimal solutions. For this purpose, we identified the individual solution ( i.e. , model si mulation) providing the minimum among all the Pareto fronts. Then, from this model so lution , w e computed the corresponding for ea ch of the ETa datasets (i.e. , we o btained the worst expected for each dataset). Thus, the nadir point was defined as the vector cont aining the maximum and the worst expected among all the Pareto - optimal solutions. 141 6. 2.5. 7 Calibration Eval uation In order to determine how well the SWAT model replicated either the spatially distributed remotely sensed ETa or p oint measurement of streamflow records at the outlet of the watershed , three statistical criteria were used, which were recommended by Moriasi et al. (2007). The three criteria are: 1) Nash - Sutcliffe efficiency (NSE); 2) percent bias (PBIAS); and 3) the ra tio of root - mean - square error (RMSE) to observed standard deviation ratio (RSR). NSE is a measure of the level of residual variance compared actual measured data variance (Nash and ndency to be larger or smaller than the observed data (Gupta et al., 1999). RSR is as its name implies a ratio between the RMSE and the observed standard deviation (Singh et al., 2005). For a model to be satisfactorily calibrated the following criteria nee ded to be met: NSE >0.5, PBIAS ±25%, and RSR <0.7 (Moriasi et al., 20 07). 6.3 Results and Discussion 6.3.1 Evaluation of the Performance of the Different Multi - objective Calibration s The first goal of this study was to evaluate the improvement in performan ce of a hydrological model in estimating the streamflow by comparing the potential benefit of using the eight different ETa products along with an E nsemble of all the datasets. To do this , nine SWAT models were calibrated by adjusting the 18 SWAT parameter s that affecting both streamflow and Eta estimation as it was discuss ed in Section 6.2.5.1. Each calibration had two objective functions, 1) streamflow with observed data collected from a USGS station at the watershed outlet and 2) ETa with each calibratio n using a different ETa product. This resulted in a total of 9 differ ent calibrations, each of which had 65,100 simulations, resulting in a total 585,900 simulations across all multi - objective calibrations. The results of the NSGA - III calibrations are 142 pres ented in Figure 6.2 in the form of optimal Pareto frontiers. As can b e seen, variability among the ETa products resulted in a wide range of performances . However, regarding individual model s , the E nsemble ETa product showed highest overall model performanc e while the TerraClimate ETa product showed the lowest overall model performance. This shows that the E nsemble, which was used minimize the uncertainty associated with the individual ETa products, was successful and outperformed all products. However, it i s important to note that none of the ETa products were compared to observed data, and thus the E nsemble cannot be labeled as the most accurate of the ETa products. Instead, the E nsemble was able to most closely replicate the SWAT model simulations. Meanwhi le , since SWAT is a phys ically based model, the fact that the E nsemble outperformed all the other ETa products indicates that at least for this region , the E nsemble product is more aligned with current knowledge of water movement is a watershed according t o the hydrological cycle . 143 Figure 6.2. Comparison of the Pareto frontiers of the nine multi - objective calibrated SWAT models To better understand Figure 6.2 and how each calibration performed, a summary of each set of Pareto optimal solutions for each m ulti - objective optimization, with respect to the NSE statistical criteria, is presented in Table 6. 2 . This includes the mean, standard deviation, coefficient of variation, minimum, maximum, the best trade - off solution, and the worst case ETa performance. T he m aximum and minimum columns show the best and worst NSE values for each objective func ti on . Here it is important to note that all cases fall within the satisfactory ranges for the NSE calibration criteria. This means that all of the potential solutions iden tified by the calibration process in the Pareto frontiers would be considered as acceptable models. However, the ETa model performance was higher than the streamflow performance, showing that the calibration process had a better fit in replicating the ETa products than the observed 144 streamflow. However, in this study , the best or optimal solutions to each calibration are - were calculated as the solution that had the smallest distance to the origin (0, 0 ) in Figure 6.2 for 1 - NSE and the furthest distance from the theoretical global worst - case model performance. This applied equal importance to streamflow and ETa. As can be seen, the trade - off solutions do not achieve the same levels of model performance r epor ted in the maximum column. For example, the maximum streamflow performance for the MOD16 1km product is an NSE of 0.78 while the tradeoff has a streamflow NSE of 0.77. However, all trade - off solutions are also better than the worst cases for each mode l . Ne vertheless, when comparing the individual calibrated model performances, the Ensemble had the best overall performance with a streamflow NSE of 0.79 and an ETa NSE of 0.95, while TerraClimate had the worst model performance with a streamflow NSE of 0. 7 5 an d an ETa NSE of 0.76. This shows that the Ensemble was able to outperform all of the individual ETa products, which shows that using the Ensemble was successful at improving the overall model performance. Meanwhile, b y considering the standard dev i atio n and coefficient of variation values, in general , ETa model performance had lower values than streamflow performance. However, when considering only streamflow model performance, the standard deviation ranged from 0.012 (ALEXI and NLDAS2 - Noah product s ) to 0.040 ( NLDAS2 - Mosaic product) , and the coefficient of variation ranged from 1.5% ( NLDAS2 - Noah product) to 5.2% ( NLDAS2 - Mosaic product) (Table 6. 2 ). This shows that during the calibration process the NLDAS2 - Mosaic product had the largest span in poten t ial solutions, while the NLDAS2 - Noah product had the smallest span when considering streamflow performance . Meanwhile, when considering only the ETa model performance, the standard deviation ranged from 0.001 (E nsemble product) to 0.039 145 (TerraClimate prod u ct), and the coefficient of variation ranged from 0.1% (Ensemble product) to 5.3% (TerraClimate product) (Table 6. 2 ). This shows that during the calibration process the TerraClimate product had the largest span in potential solutions, while the Ensemble p r oduct had the smallest span when considering ETa performance. In order to determine if any similarity existed between the Pareto frontiers of the different ETa products, the T - tests (parametric) ( Von Sto rch, 1999 ) and Wilcoxon rank sum tests (non - parame t ric) ( Wilcoxon, 1945) w ere performed for each objective function ( streamflow and ETa ) with a significance value of 5%. Tables 6. 3 , 6. 4 , 6. 5 , and 6. 6 show the results for the streamflow T - test, ETa T - test, streamflow Wilcoxon, and ETa Wilcoxon, respectivel y . As can be seen, in general , there only a few similarities found with more similarities found for the streamflow model performance (Tables 6. 3 and 6. 5 ). These tables provide an interesting insight in to the E nsemble ETa calibration. The E nsemble was desig n ed to reduce the uncertainties associated with each ETa product, and as seen above, the E nsemble calibration had the best model performance after ca libration (Figure 6.2 and Table 6. 2 ). However, this analysi s shows that while the E f ormance was similar to the MOD16 500m and Mosaic products when looking at the T - test and the MOD16 500m and Noah products when looking at the Wilcox on; neither test showed any similarity for ETa. This indicates that none of the published ETa products used in this study match the performance of the E nsemble, which further supports the idea that the E nsemble product has the be s ts fit with the SWAT model . Furthermore, due to the fact that both parametric and non - parametric tests show similar results increases the confidence in these results and the performance of the Ensemble ETa product . 146 Table 6. 2 . Summary of multi - objective calibration Pareto frontiers. actual evapotranspiration performance Datas et Mean St andard Deviation Coef ficient of Variation (%) Maximum Minimum Best T rade - off Solution Worst case NSE ET NSE Q NSE ET NSE Q NSE ET NSE Q NSE ET NSE Q NSE ET NSE Q NSE ET NSE Q NSE ET MODIS 1km 0.75 0.86 0.025 0.015 3.4 1.8 0.77 0.88 0.60 0.82 0.76 0.86 0.7 7 MODIS 500m 0.77 0.90 0.021 0.004 2.7 0.4 0.78 0.91 0.66 0.90 0.78 0.90 0.89 SSEBop 0.76 0.89 0.020 0.004 2.6 0.4 0.78 0.89 0.68 0.88 0.78 0.88 0.86 NLDAS2 - Mosaic 0.76 0.89 0.040 0.019 5.2 2.1 0.78 0.94 0.60 0.86 0.78 0.88 0.84 NLDAS2 - Noah 0.78 0.86 0 .0 12 0.014 1.5 1.6 0.78 0.90 0.68 0.84 0.77 0.88 0.82 NLDAS2 - VIC 0.73 0.88 0.030 0.002 4.0 0.2 0.76 0.88 0.65 0.87 0.76 0.87 0.82 TerraClimate 0.75 0.74 0.027 0.039 3.5 5.3 0.79 0.79 0.66 0.63 0.75 0.76 0.59 USDA - ALEXI 0.75 0.81 0.012 0.006 1.6 0.8 0.76 0 .82 0.70 0.80 0.76 0.81 0.77 Ensemble 0.77 0.95 0.021 0.001 2.7 0.1 0.79 0.95 0.70 0.95 0.79 0.95 0.95 Table 6. 3 . Results of the T - test comparison of streamflow performance of the Pareto frontiers with a 5% significance interval. Bold p - values show no d ifference at a significance value of 5% Dataset Ensemble MOD16A2 MOD16A2006 SSEBop NLDAS2 - Mosaic NLDAS2 - Noah NLDAS2 - VIC TerraClimate MOD16A2 0.000 MOD16A2006 0.197 0.000 SSEBop 0.000 0.484 0.000 NLDAS2 - Mosaic 0.064 0.120 0.0 0 7 0.252 NLDAS2 - Noah 0.001 0.000 0.065 0.000 0.000 NLDAS2 - VIC 0.000 0.000 0.000 0.000 0.000 0.000 TerraClimate 0.000 0.956 0.000 0.536 0.136 0.000 0.000 USDA - ALEXI 0.000 0.541 0.000 0.089 0.030 0.000 0.000 0.510 147 Table 6. 4 . Results of the T - t est comparison of ETa performance of the Pareto frontiers with a 5% significance interval. Bold p - values show no difference at a significance value of 5% Dataset Ensemble MOD16A2 MOD16A2006 SSEBop NLDAS2 - Mosaic NLDAS2 - Noah NLDAS2 - VIC TerraClimate MOD1 6A 2 0.000 MOD16A2006 0.000 0.000 SSEBop 0.000 0.000 0.000 NLDAS2 - Mosaic 0.000 0.000 0.000 0.475 NLDAS2 - Noah 0.000 0.044 0.000 0.000 0.000 NLDAS2 - VIC 0.000 0.000 0.000 0.000 0.000 0.000 TerraClimate 0.000 0.000 0.000 0. 00 0 0.000 0.000 0.000 USDA - ALEXI 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Table 6. 5 . Results of the Wilcoxon comparison of streamflow performance of the Pareto frontiers with a 5% significance interval. Bold p - values no difference at a signifi can c e value of 5% Dataset Ensemble MOD16A2 MOD16A2006 SSEBop NLDAS2 - Mosaic NLDAS2 - Noah NLDAS2 - VIC TerraClimate MOD16A2 0.000 MOD16A2006 0.917 0.000 SSEBop 0.000 0.126 0.000 NLDAS2 - Mosaic 0.012 0.000 0.000 0.000 NLDAS2 - Noa h 0 .1 58 0.000 0.000 0.000 0.000 NLDAS2 - VIC 0.000 0.000 0.000 0.000 0.000 0.000 TerraClimate 0.000 0.510 0.000 0.970 0.000 0.000 0.000 USDA - ALEXI 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.013 148 Table 6. 6 . Results of the Wilcoxon comparison of ETa performance of the Pareto frontiers with a 5% significance interval. Bold p - values show no difference at a significance value of 5% Dataset Ensemble MOD16A2 MOD16A2006 SSEBop NLDAS2 - Mosaic NLDAS2 - Noah NLDAS2 - VIC TerraClimate MOD16A2 0.000 M OD16A2006 0.000 0.000 SSEBop 0.000 0.000 0.000 NLDAS2 - Mosaic 0.000 0.000 0.000 0.023 NLDAS2 - Noah 0.000 0.255 0.000 0.000 0.000 NLDAS2 - VIC 0.000 0.000 0.000 0.000 0.000 0.000 TerraClimate 0.000 0.000 0.000 0.000 0.000 0.000 0. 000 USDA - ALEXI 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 149 6.3.2 Evaluation of the Performance of the Many - Objective Calibration Technique The second goal of this study was to explore the novel use of a many - objective calibration approach es fo r hy dr ological modeling. This was done by calibrating the SWAT model against all eight ETa products and streamflow. It is important to remember that this was done for two cases, one where the calibration had equal weights for all objective functions (eight ETa p roducts and one streamflow dataset) and one where the weights were balanced among objective function categori es ( ETa and streamflow). In total 688,600 simulations were run, 344,300 simulations for each calibration (equal and balanced weights), with t he c al ibration run for the balanced weights taking about twice as long as the equal weights calibration. The result s of the streamflow objective functions for both many - objective runs are presented in Figure 6.3. As can be seen , there is a vast difference betw ee n the two runs. For the equal weight calibration scenario (Figure 6.3, a) the range of the streamflow objecti ve function (1 - NSE) varies from approximately 0.5 to 7. When this is translated to NSE is outside the acceptable calibration range as describ ed b y Moriasi et al. (2007) , indicating that the calibration was not successful for streamflow. This shows the cali bration was bias ed towards the ETa remote sensing products . Meanwhile, the second many - objective calibration (Figure 6.3, b) had a much small er r an ge of objective function values (0.5 to 2). Which again shows a poor overall model calibration, but the effect of balancing the ETa objective functions weights shows a considerable improvement for the overall model pe rformances . Meanwhile, consideri ng t he ETa objective function performance, both the first and second many - objective calibrations, Figures 6.4 and 6.5, respectively, showed similar ranges, with values from 0.1 to 0.6. This shows that both calibration runs were able to achieve satisfactory mod el calibration for ETa simulation. Th ese results are interesting since they show 150 that balancing the weights of the ETa objective functions, which improv es the overall streamflow calibration , has little effect on the ETa calibration . In summary , while the ET a calibration performance is satisfactory , the low performance of the streamflow indicate s that many - objective calibration for the SWAT model is not as powerful as the multi - objective calibrations performed earlier, e sp eci ally when using the Ensemble pro du ct. 151 Figure 6.3 Pairwise comparisons of the streamflow objective funciton and the ETa objective funcitons, for a) the first many - objective calibration (equal weights) and 2) the second many - objective calibration (balanced weights) 152 Figure 6. 4 Pai rwis e c between the ETa objective func ti ons for the first many - objective calibration run s (equal weights) . Red bold numbers indicate highly correlated objective func ti ons 153 Figure 6. 5 . Pairwise compa ris ons and Pearson orre lations between the ETa objective func ti ons for the second many - objective calibration run s (balanced weights) . Red bold numbers indicate highly correlated objective func ti ons correlations among the ETa objective functions for both many - objective calibrations , similar patterns were found. In fact, objectives functions found to be highly correlated in the first run were also found to be highly correlated in the second run. T his can be explained since the calibration weights s houl d not have had an impact on the magnitude and pattern of the ETa products. Which is what the NSE objective function is a reflection of, since the model calibration attempted to replicate the pattern a nd magnitude of each ETa product. In fact, this furt hers supp ort the results found in the second study of this dissertation. 154 For example, Table 5.7 shows that there are similarities among the VIC, Noah, MOD16 1 km, and SSEBop ETa products, which is mirror e d in Figures 6.4 and 6.5. The results from Table 5.7 sho w tha t these datasets share similar magnitudes with watershed scale means of 39.56 mm /month for VIC, 42.03 mm/month for Noah, 45.39 mm/month for MOD16 1 km, and 39.05 mm/month for SSEBop . However, th e result presented in Figures 6.4 and 6.5 not only c onfi rm th at these products have similar magnitudes but also similar seasonal patterns. This is due to the fact that the calibration process aims to match both the pattern and magnitude of each ETa product and objective functions that are highly correlated show that as the model improved the fit for one ETa product it also improved the fit for the other ETa products. This also explains why the two MOD16 products are not highly correlated. While, these ETa p roducts used the same governing equations and thus s easo nal p attern they have different magnitudes (watershed scale means of 45.39 mm/month for the MOD16 1 km product and 54.98 mm/month for the MOD16 500 m product) , which is reflected by their lack of simi larity in Table 5.7. Therefore, when the calibration tri ed to replicate the MOD16 1 km dataset it pulled away from the values of the MOD16 500m product . All of this shows that it is important to consider both the seasonal pattern and the magnitude of the E Ta products to improve hydrological model performanc e th rough calibration. Another use for the correlations presented in Figures 6.4 and 6.5 is to determine if the calibration process contains redundant datasets and determine if a smaller set of ETa produ cts could be used form model calibration. For this t he h ighly correlated datasets should be considered. Correlation was determined by calculating between all in the upper triangle of Figu res 6.4 a nd 6.5. As can be seen, most of the objective functions have 155 low correlation values (r < 0.7). However, the re were two groups of ETa products that were highly correlated amongst each other; the first group includes the MOD16 1 km, SSEBop, Noah, an d VI C ETa products and the second group includes the MOD16 500 m, SSEBop, and Mosaic products (Figures 6.3 and 6.5). The high correlations found among these products were also echoed and noticeable in the pairwise regressions. The presence of these correla tion s ind icates that a smaller set of ETa products could be used for SWAT model calibration. However, in order to de termine which product to keep and which to remove, the ETa products with the highest correlations to the streamflow objective function neede d to be i dentified. The results of this show that from the set of the MOD16 1 km, SSEBop, Noah, and VIC ETa products , the MOD16 1 km 500 m, SSEBop, and Mosaic pro duct s, th correlation with a value of - 0.09. Based on this , the SSEBo p, Noah, and VIC products could be removed while keeping the MOD16 1 km ETa product and the ALEXI and Mosaic products could be removed while kee ping the MOD16 500 m ETa product. Which if these removals were done, the final ETa product set would include the MOD16 1 km, MOD16 500 m, and TerraClimate ETa products. Future studies should explore the use of this simplified ETa product set in hydrolo gica l mod el calibration. 6.3. 3 Impact of Landuse Inputs on Remote Sensing Evapotranspiration Product Calibration Pe rformance In order to examine the impacts of landuse on remotely sensed ETa products, a comparison was performed between the MOD16 500 m an d th e SWA T model. The reasons for this included that (1) not all ETa products utilize landuse files and (2) the MOD1 6 500 m products had the best remote sensing ETa product performance in the multi - objective calibration 156 (Study 3) as well as the highest sen siti vity to spatial and temporal variability (Study 2). In order to perform this analysis, the USDA - NASS Cropland Da ta Layer was obtained for 2012, used in the SWAT models developed in this dissertation , and the median landuse value of the MODIS Land Cover Typ e 1 p roduct for the period of study , used by the MOD16 500 m ETa product , were obtained . Figure 6.3 shows a visual comparison between these two datasets, which in general show a similar trend in the placement of landuse types (agricultural, forest, urb an, and w etland). However, due to the difference in resolution (30 m vs 500 m), the Croplan d Data Layer, is able to better capture the spatial variability across the landscape, especially regarding urban and wetland areas. In order to better understand the dif feren ce among these datasets, Table 6.7 summarizes each dataset and displays the percen tage of overlap s . As can be seen, from Figure 6.3, the MOD16 Landuse characteriz es the region as only having 1.3% urban and 0.4% as wetland; which contrasts the 7.9% and 14.0 % reported by the Cropland Data Layer for urban and wetland, respectively. This al so means that the MOD16 dataset classifies the region with more agricultural and forest lands compared to the Cropland Data Layer. However, to determine the ove rlap betw een t he dataset the intersection of the two layers were performed . The fourth column fr om the left in Table 6.7 reports the intersection value between the datasets in square kilometers, which represents the regions of the Honeyoey watershed for which b oth datas ets agree on the landuse. These values were then divided by the respective areas f or each landuse of each dataset to calculate the percent of intersection . For example, when looking at agricultural land, the intersection area was 603.3 km 2 , and fo r th e Cro pland Data Layer the total area of agricultural land was 647.7 km 2 , by dividing th e intersection by the total area the percentage of intersection was found to be 93.1%. This means that 93.1% of the agricultural land reported by the Cropland Data L ayer was shared with the MOD16 product. This analysis was done for the 157 entire region as wel l as each major landuse using both datasets as reference . As can be seen when looking at the entire region, 66.5% of the region was identified as being the same acro ss b oth l anduse datasets. This shows that the majority of the region was the same for both datasets. However, when looking at the individual landuses the lack of urban and wetland areas in the MOD16 dataset plays a major role in th e agreement among the two dat asets . Since the areas for both urban and wetland were small for the MOD16 dataset, onl y a small region of overlap was identified (8.4 km 2 and 0.9 km 2 , respectively). This translates to very small percentages of intersection for the Cropland Data Layer (10 .0% a nd 0.9%, respectively), since the overall areas for those landuses was much higher (84.2 km 2 and 148.6 km 2 , respectively) . On the other hand, the MOD16 dataset shows higher percentages of intersection for those landuses (59.9% and 23.9%, respectiv ely) , yet reported much smaller total areas for each landuses (14.1 km 2 and 3.8 km 2 , respec tively) . In fact, this trend of the dataset with the smaller area for a specific landuse report ing higher percentages of intersection and vice a versa held true for all landu ses . Overall, this shows that while the majority of the region is similar for both landuses datasets, there are still a number of differences that indicate that the similarity among the ETa performances is more likely linked to differences in gove rnin g equ ations and spatial resolutions than similarities among the landuse datasets utiliz ed. 158 Figure 6. 6 . Comparison of the landuse products utilized by (a) the SWAT and (b) the MOD16 500 m ETa product 159 Table 6. 7 . Comparison of the SWAT model and MOD 16 50 0 m E Ta product landuse datasets, CDL 2012 and MOD16, respectively Land cover Area (km 2 ) Regional Percentage (%) Percent of Intersection (%) CDL 2012 MOD16 Intersection CDL 2012 MOD16 CDL 2012 MOD16 Agriculture 647.7 835.7 603.3 60.8 78.5 93.1 72. 2 Ur ban 84 .2 14.1 8.4 7.9 1.3 10.0 59.9 Forest 184.2 211.0 95.1 17.3 19.8 51.6 45.1 Wetland 148.6 3.8 0.9 14.0 0.4 0.6 23.9 Total 1064.6 1064.6 707.7 100.0 100.0 66.5 66.5 160 6.4 Conclusions Throughout the course of this study , two different calibratio n tec hnique s were explored, 1) multi - objective and 2) many - objective. In general , the best model performances were obtained from the multi - objective calibrations. And considering all of the multi - objective calibrated models, the mo del with the best streamf low p erform ance was the E nsemble followed by Mos aic, SSEBop, MOD16 500m, Noah, ALEXI, VIC, MOD16 1 km, and finally TerraClimate. Meanwhile , when considering ETa performance, the SWAT model with the best performance again utilized t he E nsemble followed by t he MO D16 50 0m, Mosaic, SSEBop, Noah, VIC, MOD16 1km, ALEXI, and again finally TerraClimate. This shows that the E nsemble utilized in this study had the best fit with the SWAT model and outperformed the individual ETa products. Mean while, when considering t he ma ny - obj ective calibration, ETa performance was found to be satisfactory; however, the streamflow performance was not satisfactory. This shows that the many - objective calibration was not ideal for SWAT model calibration when con sidering both streamflow and E Ta sim ulation due to the fact that the search space is much larger than the multi - objective approach . However, correlations among the ETa objective functions show that a small er set of ETa products should be explored in future studies , namely the MOD1 6 1 k m, MOD 16 500 m, and TerraClimate ETa products. Another conclusion from the many - objective calibration is the importance of both the magnitude and the seasonal pattern in model calibration. However, it is important to no te that this study was performe d for only one watershed in Michigan; therefore, future studies should expand this work to regions with different physiographical and climatological zones. This would serve to confirm the robustness of the techniques implemen ted this study. This would help impr ove ou r understanding of how each ETa product performs in hydrological model calibration. Aligned with this is the fact that only the 161 SWAT model was used in this research, and different ETa products may fit better with d ifferent hydrological models, t heref ore fu ture studies should also explore the use of other widely used hydrological models . Furthermore, other ensembling techniques should be performed to identify the best for different regions. 6.5 Acknowledgment Authors would like to thank Dr. Marth a C. Anders on from USDA - ARS Hydrology and Remote Sensing Laboratory at Beltsville, Maryland and Dr. Christopher R. Hain from NASA Marshall Space Flight Center at Huntsville, AL for his help in providing A LEXI data. This work is supported by the USDA Nation al In stitut e of Food and Agriculture, Hatch project MICL02359. 162 7 . CONCLUSIONS Throughout this dissertation , two main topics were explored, 1) the integration of satellite - based remote sensing ETa products into the calibration and validation of hydrologica l mod els an d 2) the temporal and spatial performance of different satellite - based remotely sensed ETa produ cts. Th ese two topi cs w ere examined in the Honeyoey watershed in the state of Michigan, which is a region that lacks observed ETa data and is also co nside r ed as an area of concern in the Great Lakes basin . In the first study, the introduction of satellite - based remote sensing ETa products in hydrological model calibration was explored. In the second study, the performance of different remote sensing ET a pro ducts was compared on both temporal and spatial scales. And finally, in the third study, additional ca libration techniques were expanded building on the results of the first study. From these studies the major takeaways are as follows: Inclusion of ET a dat a in t he model calibration process improved the overall model performance. During the initial test of calibration techniques, the genetic algorithm technique show ed the greatest improvement of ETa simulation, but at the cost of lowering the streamflow simu lation . Meanwhile. the multi - variable technique was able to improve ETa simulation and maintain/improv e streamflow simulations . Thus, the use of the multi - variable technique was further explored. Statistical analysis of the calibration results for the firs t stud y showed that even in cases where calibration was satisfactory, there was still a significant difference between the SWAT model output s and the observed dataset s at the 5% level of significance . Considering seasonal analysis among the ETa pro duc ts, a n over all pattern of less variation in the spring and summer and more variation in the winter and fall was noticed. However, there were no noticeable patterns found between seasons regarding similarities 163 among ETa products. This indicates that temp ora l var iation is less influential when compared to spatial variation. Considering seasonal analysis with individual ETa products , the majority (MOD16 500 m, MOD16 1 km, ALEXI, TerraClimate, SWAT, and the Ensemble) were able to differentiate among all seas ons . How ever, for ETa products from SSEBop, NLDAS - 2: Mosaic, NLDAS - 2: Noah, and NLDAS - 2: VIC similarity among the spring and fall seasons were observed . Considering, spatial analysis of remotely sensed ETa products, two major clusters within the ETa produ cts were ident ified; datasets with higher ETa values reported by MOD16A2 500 m, NLDAS - 2: Mosaic, TerraClimate, and ALEXI; and lower ETa values reported by MOD16A2 1 km, SSEBop, NLDAS - 2: Noah, NLDAS - 2: VIC, and SWAT Among all of the ETa products tested, the MO D16A2 500 m product had the best spatial performance, being able to distinguish between all of the major landuses for all seasons. However, each products, except for ALEXI, were able to distinguish between the major landuses for at least one season: MOD 16A 2 1 k m: spr ing and summer; MOD16A2 500 m: winter, spring, summer, fall; SSEBop: winter; NLDAS - 2 Mosaic: spring and summer; NLDAS - 2 Noah: winter and summer; NLDAS - 2 VIC: spring; TerraClimate: spring; ALEXI: none; SWAT: spring and summer; Ensemble: winter , s ummer , and fall. Considering the use of different calibration techniques, the multi - objective calibrations resulted in better overall model performances than the many - objective calibration technique. Considering the performance of individual ETa produc ts in mo del ca libration, all products resulted in models that were satisfactorily calibrated for both streamflow and 164 ETa. However, use of the E nsemble in the multi - objective calibration resulted in a SWAT model with the be s t performance compared to all other ETa p roduct s. Meanwhile the introduction of the TerraClimate product resulted in the worst overall model performance. Regarding the many - objective calibration technique, analysis of the Pareto frontier showed that the ca libration was successful for all ETa prod ucts; however, the calibration was unable to achieve satisfactory results for streamflow performance. Results from the spatial and temporal sensitivity and many - objective calibration show ed that b oth the magnitude a nd seasonal pattern of the ETa produ cts p lay a major role in the agreement among the ETa products and their performance in the many - objective calibration process. This explained why the two MOD16 datasets that utilized the same technique were not highly co rrelated during the many - objective c alibr ation as well as their lack of agreement in the spatial and temporal analysis. Based on both the temporal and spatial performances as well as the final multi - objective calibration the MOD16 500 m ETa products has th e best performance for the Honeyoey water shed. After the MOD16 500 m , the ranking of the remaining remotely sensed ETa products from best to worst is: MOD16 1 km, NLDAS - 2: Mosaic, NLDAS - 2: Noah, SSEBop, NLDAS - 2: VIC, ALEXI, and finally TerraClimate . Compa risons between the landuse datasets used for th e SWAT model and the MOD16 500 m ETa products indicate d that while the majority of the study area is similar between both landuse datasets, the differences in ETa results is more likely originated from the app lication of different governing equa tions and s patial resolutions of the individual ETa products. 165 8 . FUTURE RESEARCH RECOMMENDATIONS This research explored the use of remote sensing ETa in hydrological model calibration and the performance of remote sens ing ETa products in a data scarce re gion. Howev er, this is not the definitive end for these knowledge gaps. Therefore, additional research should be performed to address the limitations of this work. Possible areas for future research are presented below: Our study was performed for a small water shed i n the state of Michigan. Therefore, the recommended calibration techniques s hould also be examined in other regions with different climatological and physiographical characteristics to improve our understanding of the linkage of hydrological mod eling and r emote sensing data. Results from the first study indicated that while the model calibration was successful, it was not able to replicate the ETa products. This indica tes that there is still room to improve the model calibration process to create even more realistic models that can provide stakeholders, decision makers, and policy makers with more accurate results . Due to the lack of observed data, a simplified ensembli ng technique was utilized . Therefore , in the future studies, it is recommended to ex amine the performance of different ensembling techniques on capturing spatial and temporal variabilities of ETa products. Identifying the best trade - off solution in the mo del calibration was done by giving streamflow and ETa equal importance . However , exp inputs should be considered to determine if this weighting assumption should be modified . Correlations among the ETa datasets, indicate that a smaller set of ETa products could be used in the many - objective calibration process and should be exp lored in fu ture studies. 166 This dissertation explored the use of ETa data in hydrological model calibration, future studies should explore the use of additional remotely sensed hydrological components such as soil moisture. 167 APPENDIX 168 APPENDIX Table S 5. 1. Average monthly ETa values for each dataset for agricultural lands with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 15.63 a 20.94 a 37.11 a 37.62 54. 89 a 76.62 a 98. 07 a 83.15 a 42.3 7 a 25.26 a 20.41 a 14.97 a MOD16A2 500m 16.29 a 21.39 a 37.69 a 44.50 a 70.02 b 98.24 b,c 126.96 b 108.35 53.02 b 26.39 a,b 17.40 b 10.95 b SSEBop 0.02 0.00 10.29 b 26.16 b 47.15 a,c 87.57 d 115.86 b,c 99.60 b 50.88 b 10.96 5.73 c 0.42 c NLDAS - 2:M osaic 10.67 b,c 11.39 b,c 27.43 c 61.21 97.41 121.73 e 138.43 116.14 84.49 48.54 c 22.03 a 11.79 b NLDAS - 2:Noah 9.58 b 12.02 b 18.73 d 28.52 b 46.18 c 78.30 a 105.06 d 101.68 b 67.06 c 27.73 b 10.19 7.05 d NLDAS - 2:VIC 7.37 9.68 c 10.08 b 15.29 50.63 a,c 93.34 b,d 119.23 b,c 96 .95 b 50.05 b 16 .29 5.78 c 6.95 c ,d TerraClimate * * 18.18 b,c,d 81.94 101.76 110.83 c,e 97.77 a,d 86.81 a,b,c 65.31 b,c,d 49.58 c 22.56 a 1.39 c,d ALEXI 23.37 38.17 51.98 57.51 82.28 102.76 c 123.09 b 101.13 b 66.95 c 32.60 d 19.75 a 16.35 a SWAT 3.71 5.40 28.54 c 42.24 a 62.54 106 .19 c 102.18 a,d 72. 81 c 48.28 a,b 26.33 a,b 16.87 b,d 6.88 d,e Ensemble 11.74 c 16.05 26.53 c 44.09 a 68.79 b 96.17 b 115.56 c 99.23 b 60.02 d 29.67 d 15.48 d 9.34 e *Note that no ETa values were provided for TerraClimate for the months of January and February . 169 Table S 5. 2 . Average month ly ETa values for each dataset for forest lands with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 16.47 22.63 37.46 a 38.79 a 67.33 95 .23 a 104.49 a 8 4.17 a 47.19 a 25 .36 19.66 a 14.97 a MOD16A2 500m 14.95 20.47 37.17 a 45.78 b 90.59 a,b 127.14 139.30 b 116.29 b 65.95 b,c 27.10 16.56 b 10.30 b,c,d SSEBop 0.02 0.01 10.02 b 28.15 c 54.26 c 99.60 a,b 122.22 c 100.66 c 54.51 d 14.15 a 5.75 c 1.37 e NLDAS - 2:Mosai c 11.60 a 13 .03 a 25.57 c 53.95 d 91.88 a 116.00 c 133.13 b 116.03 b 87.69 50.70 b 21.65 a,d 11.81 b NLDAS - 2:Noah 11.95 a 14.22 a 21.65 d 29.56 c 38.25 d 64.11 94.43 d 94.91 d 67.71 b 31.36 c 12.05 8.35 c,f NLDAS - 2:VIC 8.14 9.92 10.57 b 15.36 40.91 d 77.83 108.71 a 96.31 c,d 49. 27 a 15.83 a 6.5 1 c 7.49 c,f Ter raClimate * * 16.50 b,c,d 81.41 101.45 110.58 b,c 98.70 a,d 88.50 a,c,d 65.30 b,c,d 49.22 b 21.96 a,d 1.20 e ALEXI 22.10 35.72 49.88 55.66 d 85.29 b 108.6 c 125.04 c 99.64 c 67.11 b 33.04 c 19.12 a 15.99 a SWAT 3.53 5.32 28.90 e 40.11 a 61.12 c 84.92 68. 43 65.47 65.84 b,c 42.02 23.04 d 8.65 c,d,f Ensemble 12.06 a 16.40 26.21 c,e 43.58 b 71.25 99.89 a,b,c 115.75 99.56 c 63.09 c 30.84 c 15.41 b 9.59 d *Note that no ETa values were provided for TerraClimate for the months of January and February. 170 Table S 5. 3. Averag e m onthly ETa valu es for each dataset for urban lands with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 16.20 a 22.44 a 36.9 a 37.69 a 57.92 a 74.93 a 83.57 69.92 40.2 4 a 25.04 a 20.02 a,b 15.29 a MOD16A2 500m 16.53 a 22.12 a 37.02 a 44.60 b 73.39 b 96.76 b 107.32 a,b 92.38 a 50.04 b,c 26.06 a 17.01 a 10.57 b SSEBop 0.04 0.00 9.59 b 24.94 48.12 c 85.32 c 105.52 a,b 89.24 a 46.65 a,b,c 9.66 5.70 c 0.01 c NLDAS - 2:Mosaic 10.14 b,c 10 .55 b,c 28.5 3 c 65.19 99.41 d 12 4.34 d 141.60 116.60 82.09 d 47.06 b 22.12 b 11.71 b NLDAS - 2:Noah 8.64 b 11.44 b 17.9 d 29.01 50.43 a,c 84.32 c 109.72 a,b,c 105.00 b 67.49 e 26.50 a 9.58 6.62 d NLDAS - 2:VIC 6.81 9.48 c 9.65 b 15.26 56.05 a 101.77 b,d 124.04 95.09 a,b 49.26 b,c 1 6.29 5.36 c 6.5 4 c,d TerraClim ate * * 20.02 b,c,d 82.83 102.24 d 110.26 b,d 96.91 a 86.54 a 65.33 b,d,e,f,g 50.14 b 23.19 a,b 1.70 c,d ALEXI 22.87 37.45 49.36 55.35 80.66 98.24 b,d 115.04 c 95.18 a 63.36 f 30.50 a,c,d 18.53 a,b 15.45 a SWAT 3.44 5.18 27.81 c 40.74 a 55.5 a,c 72.35 a 57. 84 48.82 42.38 a ,c 31.51 c 18.20 a 7.19 d,e Ensemble 11.5 c 16.03 26.19 c 44.36 b 71.03 b 96.99 b 110.46 b 93.75 a 58.06 g 28.91 d 15.19 9.04 e *Note that no ETa values were provided for TerraClimate for the months of January and February. 171 Table S 5. 4 . Average m ont hly ETa values for each dataset for wetland lands with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 16.92 23.25 37.42 a 38.21 66.47 a 94.12 a 103.31 a 83.93 a 45.8 7 2 5.34 a 20.38 a 15 .77 a MOD16A2 500m 15.22 21.02 37.68 a 45.59 a,b 88.12 b 123.71 b 135.59 b 113.85 b 61.03 a,b 26.61 a,b 17.04 10.51 b,c SSEBop 0.04 0.06 11.24 b 27.57 c 54.16 99.80 c,d 121.07 c,d,e 101.85 c 55.17 c 13.86 5.94 b 0.91 d NLDAS - 2:Mosaic 10.91 a,b 12.10 a 26. 27 c 57.00 d 92.52 1 13.17 c,e 128.98 b,c 110.75 b 83.71 49.82 c 21.37 a,c 11.62 b NLDAS - 2:Noah 10.24 a 12.17 a 17.28 d 26.18 c 42.43 74.71 102.39 a 99.51 a,c 66.06 a,b 27.43 b 9.91 7.20 e NLDAS - 2:VIC 7.84 9.90 10.11 b 15.82 47.97 88.48 f 117.67 d,e 101.79 c 53.98 c 17.36 6.16 b 7 .39 e TerraClim ate * * 19.00 b,c,d 82.51 101.44 110.33 b,c,d,e 96.77 a 86.86 a,c 64.95 a,b,c 49.65 c 22.99 a,c 1.61 d ALEXI 22.73 36.62 49.93 55.87 d 84.05 b 106.14 c,d,e 124.14 b,c,d 99.60 a,c 66.29 a 32.18 d 19.14 a 15.93 a SWAT 4.05 5.95 32.01 47.95 a 6 8.48 a,c 90. 46 a ,f 73.93 69.21 69.37 a 42.51 23.53 c 8.65 c,e Ensemble 11.87 b 16.27 26.20 c 43.59 b 72.15 c 101.31 c,d 116.24 e 99.77 c 62.13 b 30.28 d 15.37 9.48 c *Note that no ETa values were provided for TerraClimate for the months of January and February. 172 Tabl e S 5. 5. Ave rag e monthly ETa v alues for each dataset for alfalfa (ALFA) regions with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 15.54 a 20.90 a 36.75 a 39.18 a 65.7 4 87.94 99. 00 a 80.57 a 45.81 a 24.95 a 18.95 a 14.23 a MOD16A2 500m 15.37 a 20.69 a 37.25 a 46.42 86.85 a,b 117.15 a 131.25 b,c 109.48 b 61.80 b,c 26.84 b 16.47 b 10.29 b,c SSEBop 0.02 0.01 8.33 b 28.62 b 54.69 c 98.78 b 121.28 b 100.50 c 55.05 d 13.86 c 5.54 c 1.11 d NLDAS - 2:Mo saic 11.57 b 12 .87 b 25.38 c 52. 79 90.39 a 114.05 a 131.68 c 114.43 b 86.39 e 49.47 d 21.43 a 11.85 b NLDAS - 2:Noah 12.05 b 14.37 b 21.94 d 29.52 b 37.18 61.85 92.72 d 93.47 d 67.04 b,c 31.39 e 12.14 8.47 c,e NLDAS - 2:VIC 8.20 10.09 10.78 e 15.42 41.24 78.09 c 109.40 e 96.95 c,d 49.24 a 15.8 2 c 6.51 c 7.45 c,e TerraClimate * * 16.02 b,c,d,e 81.14 101.36 110.39 a 98.84 a,d,e 88.44 a,c,d 65.53 b,c,d 49.20 d 21.55 a,b 1.13 d ALEXI 22.26 35.97 50.13 56.55 85.60 b 109.28 a 124.86 b,c 99.99 c 67.65 b 33.53 e 18.97 a 16.03 a SWAT 3.07 4.56 25.85 c 38.03 a 55.19 c 73 .62 c 96.64 a,d 92.5 1 c,d 72.84 b,c,e 29.44 a,b,e 14.77 d 5.39 e Ensemble 12.03 b 16.24 25.93 c 43.71 70.38 97.19 b 113.63 97.98 c,d 62.31 c 30.63 e 15.19 d 9.47 c *Note that no ETa values were provided for TerraClimate for the months of January and February . 173 Table S 5. 6. Average mont hly ETa values for each dataset for corn (CORN) regions with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 15.58 a 20.88 a 37.24 a 37.44 53.10 a,b 7 4.3 8 a 97.88 a 83.92 a 42.05 a 25.28 a 20.62 a 15.04 a MOD16A2 500m 16.42 a 21.45 a 37.74 a 44.17 a 67.45 c 95.16 b,c 126.77 b 109.36 b 52.27 b 26.30 a,b 17.54 b 11.01 b SSEBop 0.03 0.00 10.28 b 25.78 b 45.47 a 84.79 d 114.65 b,c,d 99.63 a,c 50.17 b 10.42 5.72 c 0.36 c NLDAS - 2:Mos aic 10.53 b,c 11.05 b,c 27.80 c 62.77 98.60 d 123.00 e 139.33 116.08 83.83 48.08 c 22.18 a 11.82 b NLDAS - 2:Noah 9.17 b 11.62 b 18.29 d 28.42 b 47.66 a,b 80.98 a,d 107.08 c 102.98 b,c 66.97 c 27.11 b 9.89 6.83 d NLDAS - 2:VIC 7.27 9.65 c 10.07 b 15.28 52.48 b 96.30 b,c 120.98 b,d 96. 71 a,c 49.97 b 16 .34 5.65 c 6.87 c,d TerraClimate * * 18.34 b,c,d 81.91 101.81 d 110.91 b,e 97.71 a,c 86.55 a,c 65.36 b,c,d 49.61 c 22.62 a 1.39 c,d ALEXI 23.41 38.52 52.37 57.76 81.73 101.84 b,c,e 122.91 b 101.52 c 67.00 c 32.55 d 19.84 a 16.39 a SWAT 3.73 5.42 28.61 c 4 2.65 a 63.62 c 11 5.34 b,e 108.57 c,d 71.70 a 42.54 a,b 25.83 a,b 16.63 b,d 6.92 d,e Ensemble 11.67 c 15.99 26.61 c 44.19 a 68.54 c 95.92 c 115.91 d 99.59 c 59.70 d 29.46 d 15.51 d 9.31 e *Note that no ETa values were provided for TerraClimate for the months of January an d F ebruary. 174 Tab le S 5. 7. Average monthly ETa values for each dataset for field peas (FPEA) regions with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 15.86 a 21. 59 a 37.02 a 38.28 a 63.00 a 90.61 97.33 a 72.87 a 43.22 25.34 a 19.76 a 14.84 a MOD16A2 500m 15.54 a 21.91 a 37.84 a 45.85 b 81.93 b 114.36 a 124.47 b 95.57 b,c 56.07 a 27.25 b 16.15 b 10.22 b,c SSEBop 0.00 0.00 10.94 b 27.72 c 56.83 a,c 104.65 b 125.90 b 102.54 b 56.5 9 a 15.50 c 5 .59 c 0.94 d NLDAS - 2:Mosaic 11.60 b 13.03 b 25.74 c 53.16 d 86.80 b 111.96 a 128.81 b 115.22 86.24 49.41 d 21.47 a 11.68 b NLDAS - 2:Noah 12.24 b 14.16 b 21.12 d 29.87 c 39.40 d 65.75 95.74 a 96.71 c 69.07 b 31.76 e 12.24 8.48 c,e NLDAS - 2:VIC 8.12 9.76 10.38 b 15.50 41.11 d 78.4 3 1 08.40 c 95.64 b,c 49.54 15.98 c 6.55 c 7.71 c,e TerraClimate * * 19.68 b,c,d 82.49 101.42 110.38 a,b 98.14 a,c 87.80 a,b,c 64.78 a,b,c 49.55 d 22.89 a 1.49 d ALEXI 23.09 36.70 50.07 56.59 d 87.85 b 109.90 a,b 125.50 b 101.39 b,c 67.29 b 32.52 e 19.08 a 15.91 a SWAT 3.70 5. 50 28.43 c 37.98 a 54.03 c 70.65 96.30 a 76.34 a 33.92 27.52 a,b 15.96 b 6.55 e Ensemble 12.24 b 16.56 26.77 c 43.68 b 69.79 98.26 113.04 95.97 b,c 61.60 c 30.92 e 15.47 b 9.53 c *Note that no ETa values were provided for TerraClimate for the months of Jan uary and Fe bru ary. 175 Table S 5. 8. Average monthly ETa values for each dataset for deciduous forest (FRSD) regions with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1 km 16.47 22 .63 37.46 a 38.79 a 67.33 95.22 a 104.48 a 84.16 a 47.19 a 25.36 19.67 a 14.97 a MOD16A2 500m 14.95 20.47 37.17 a 45.78 b 90.59 a,b 127.13 b 139.29 b 116.28 b 65.94 b,c 27.10 16.56 b 10.3 b,c,d SSEBop 0.02 0.01 10.02 b 28.15 c 54.26 c 99.60 a,c 122.22 c 100.66 c 54. 5 d 14.15 a 5 .75 c 1.37 e NLDAS - 2:Mosaic 11.60 a 13.03 a 25.57 c 53.95 d 91.88 a 115.99 d 133.13 b 116.03 b 87.69 50.70 b 21.65 a,d 11.81 b NLDAS - 2:Noah 11.95 a 14.22 a 21.64 d 29.56 c 38.25 d 64.12 94.44 d 94.91 d 67.71 b 31.36 c 12.05 8.35 c,f NLDAS - 2:VIC 8.14 9.92 10.57 b 15. 36 40.91 d 7 7.8 3 108.71 a 96.31 c,d 49.27 a 15.83 a 6.51 c 7.49 c,f TerraClimate * * 16.50 b,c,d 81.41 101.45 110.58 b,c,d 98.70 a,d 88.50 a,c,d 65.30 b,c,d 49.22 b 21.96 a,d 1.20 e ALEXI 22.10 35.72 49.88 55.66 d 85.29 b 108.6 d 125.04 c 99.65 c 67.11 b 33.04 c 19.12 a 15.9 9 a SWAT 3. 53 5.32 28.90 e 40. 12 a 61.12 c 84.92 68.42 65.47 65.84 b,c 42.02 23.04 d 8.65 c,d,f Ensemble 12.06 a 16.40 26.21 c,e 43.58 b 71.24 99.88 a,c 115.75 99.56 c 63.09 c 30.84 c 15.41 b 9.59 d *Note that no ETa values were provided for TerraClimate for the months of January an d February. 176 Table S 5. 9. Average monthly ETa values for each dataset for evergreen forest (FRSE) regions with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MO D16A2 1km 1 6.3 9 a 23.34 36.79 a 42.19 a 84.70 a 124.26 a 128.12 a 97.69 a 55.75 a,b,c 26.14 18.33 a 14.14 a MOD16A2 500m 15.44 a 21.85 35.09 a 48.54 103.15 b 156.60 160.06 130.24 b 75.08 d,e 28.85 a 14.90 b,c 10.06 b,c,d SSEBop 0.00 0.04 11.95 b 28.91 b 55.71 c 101.13 b,c 123 .46 a 104.63 c 5 8.59 a,b 15.38 b 5.99 d 2.79 e NLDAS - 2:Mosaic 11.61 b 13.26 a 25.64 c,d 54.58 c 96.89 d 124.84 a 144.84 125.07 b 94.29 53.12 c 22.05 e 11.80 b NLDAS - 2:Noah 12.26 b 15.17 a,b 24.39 c 32.34 b 38.06 e 60.88 92.32 b 94.28 a 69.22 d,e,f 33.63 d 13.10 b 8.76 c NLDAS - 2:V IC 7.55 9.5 6 1 0.11 b 15.04 39. 70 e 76.21 107.71 c 94.12 a 47.30 c 15.21 b 6.42 d 7.03 c TerraClimate * * 17.32 b,c,d 80.58 100.79 b,d 110.41 a,b,c 98.79 b,c 88.50 a 65.35 a,b,d,e,f,g 48.83 c 21.48 a,e 1.09 e ALEXI 23.53 36.53 49.26 53.96 c 84.18 a 107.98 b,c 123.90 a 95.25 a 65.64 e,f, g 3 2.17 a,d,e 18.32 a,c 15.91 a SWAT 3.35 5.22 27.95 c,d 36.22 59.88 c 96.06 b 72.68 64.08 45.68 b,c 28.85 a 16.07 a,c 6.36 c Ensemble 12.28 b 16.93 b 26.55 d 44.52 a 75.40 107.79 c 122.40 a 103.72 c 66.40 f,g 31.67 e 15.07 c 9.62 d *Note that no ETa values were provided fo r T erraClimate for the months of January and February. 177 Table S 5. 10. Average monthly ETa values for each dataset for hay (HAY) regions with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. A ug. Sep. Oc t. Nov. Dec. MOD1 6A2 1km 16.87 a 23.04a 37.52a 38.24a 67.22 95.36 a 102.43 a 83.90 a 45.38 25.54 a 21.15 a 16.37 a MOD16A2 500m 16.31 a 22.02 a 37.34 a 45.93 b 82.61 a 113.22 b,c 127.15 b,c 105.17 b 55.53 a,b 26.02 a 17.15 10.37 b,c SSEBop 0.01 0.05 10.68 b 27. 35 c 52.55 b 95. 37 a,d 115.52 d 9 8.06 a,c 51.82 a 11.54 5.77 b 0.39 d NLDAS - 2:Mosaic 10.18 b,c 10.33 b 27.47 c 55.38 87.34 c 111.00 b,c,e 132.52 b 111.70 d 78.95 c 41.56 b 20.22 a 11.63 b NLDAS - 2:Noah 8.78 b,c 10.41 b 14.69 d 23.99 c 45.02 78.97 f 104.95 a 98.75 a,b,c 63.19 d 24.7 6 a 8.64 6.4 4 e, f NLDAS - 2:VIC 7.22 b 9.92 b 10.13 b 15.93 55.46 b 99.97 a,d,e,g 130.23 b,c 108.20 b,c,d 55.36 a 17.98 5.81 b 6.95 d,e,f TerraClimate * * 19.93 b,c,d 82.84 102.31 110.16 b,c,d,e,g 96.42 a 86.50 a,c 65.21 a,b,c,d 49.87 23.19 a 1.62 d,e ALEXI 24.24 39.32 53 .37 59.79 8 4.3 6 a,c 104.24 c,d, e,g 123.50 c 103.80 b,c 67.76 d 33.3 20.71 a 16.81 a SWAT 3.70 5.60 28.37 c 37.96 a 54.45 b 73.88 f 69.19 64.10 56.22 a,b 38.93 b 20.64 a 8.00 c,e,f Ensemble 11.83 c 16.26 26.46 c 43.68 b 72.11 101.04 d,e,g 116.59 d 99.51 c 60.40 b 28.82 15.33 9 .42 b,c,f * Not e that no ETa v alues were provided for TerraClimate for the months of January and February. 178 Table S 5. 11. Average monthly ETa values for each dataset for pasture (PAST) regions with clusters indicated by superscripts for each column Dataset s Month J an. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 16.05 a 21.70 a 37.03 a 38.83 65.41 a 89.89 99.98 a 80.36 a 45.00 25.15 19.46 a 14.63 a MOD16A2 500m 14.90 a 20.54 a 37.48 a 45.91 a 84.84 b 116.82 a,b 131.93 b 107.29 b 58.76 a,b 26.50 16.6 2 b 10.55 b,c S SEBop 0.01 0.01 10.24 b 28.13 b 55.50 c 100.73 a,c 122.61 c 101.00 b,c 55.40 a 14.12 a 5.56 c 1.12 d NLDAS - 2:Mosaic 11.70 b 13.17 b 25.70 c 54.26 c 94.69 119.87 b 137.75 b 118.74 89.60 51.46 b 21.78 a 11.93 b NLDAS - 2:Noah 11.85 b 14.17 b 21.73 d 29.71 b 38.43 d 64 .11 94.42 a 94. 89 d 67.66 c 31.3 3 c 12.02 8.29 e NLDAS - 2:VIC 8.02 9.84 10.46 b 15.33 40.14 d 76.71 d 108.19 d 96.26 c,d 48.92 15.65 a 6.53 c 7.39 e TerraClimate * * 16.49 b,c,d 81.27 101.24 110.60 a,b,c 98.68 a,d 88.20 a,c,d 65.18 a,b,c 49.14 b 21.86 a 1.21 d ALEXI 22.60 35.93 50.29 56 .26 c 85.70 b 107 .90 a,b,c 124.16 c 99.37 c 67.40 c 33.15 c,d 19.22 a 16.06 a SWAT 3.62 5.45 30.27 43.52 a 62.09 a,c 77.28 d 66.53 62.48 54.76 a 36.79 d 20.86 a 7.50 c,e Ensemble 12.05 b 16.31 26.29 c 43.71 a 70.74 98.33 c 114.72 98.26 c,d 62.24 b 30.81 c 15.38 b 9.54 b,c,e *No te that no ETa values were provided for TerraClimate for the months of January and February. 179 Table S 5. 12. Average monthly ETa values for each dataset for sugar beet (SGBT) regions with clusters indicated by superscripts for each column Data sets Month J an. Feb. Mar. A pr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 15.90 21.23 a 38.44 a 36.77 46.30 a 65.78 a 95.42 a 83.65 a 40.79 25.89 a 22.18 a 15.59 a MOD16A2 500m 18.58 23.20 a 39.14 a 42.29 a 56.43 a,b 82.03 b 119.47 b,c 102.12 b,c,d 47.72 a 26.5 8 a 18.84 b,c 11 .48 b SSEBop 0. 34 0.04 10.87 b 25.38 b 37.31 72.20 a 108.52 d 98.90 b,c 48.34 a 9.30 5.98 d 0.08 c NLDAS - 2:Mosaic 9.94 a 10.10 b,c 28.29 c 66.56 101.82 c 126.91 c 141.16 114.66 d 81.01 b 46.61 b 22.26 a,b 11.66 b NLDAS - 2:Noah 8.64 a,b 11.30 b 18.11 d 29.56 b 51. 23 a,b 85.5 b 11 0.76 b,d 106.04 b ,d 67.15 c 26.49 a 9.62 6.65 d NLDAS - 2:VIC 6.97 b 9.69 c 10.20 b 15.22 58.12 b 105.09 d 125.03 c 94.49 a,b,c,d 49.51 a 16.34 5.04 d 6.55 c,d TerraClimate * * 19.09 b,c,d 81.71 102.04 c 111.65 c,d 97.20 a,d,e 85.19 a,c 65.47 b,c,d 49.76 b 22.81 a,b,c 1.51 c ,d ALEXI 24.2 40. 62 54.40 58.87 78.85 96.28 e 121.32 c 102.37 b,c,d 67.45 c 32.63 c 20.23 a,b,c 16.95 a SWAT 3.73 5.56 29.95 c 45.13 a 67.94 d 96.74 e 86.46 e 79.03 a 54.52 a,d 30.62 c,d 18.09 c 6.85 d Ensemble 11.97 16.41 27.41 c 44.54 a 66.51 d 93.18 e 114.86 b, c,d 98.43 b, c 5 8.43 d 29.20 d 15 .87 9.40 *Note that no ETa values were provided for TerraClimate for the months of January and February. 180 Table S 5. 13. Average monthly ETa values for each dataset for soybean (SOYB) regions with clusters indicated by supersc ripts for e ach column Dataset s Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 15.67 a 20.84 a 36.95 a 37.22 52.58 a,b 74.39 99.86 a 85.63 a 41.64 25.31 a 20.79 a 15.26 a MOD16A2 500m 16.95 a 21.91 a 37.81 a 44.41 a 66.07 c 93.67 a,b,c 12 6.20 b 108.5 8 b 50.50 a,b 26.24 a 17.71 b 11.22 b SSEBop 0.00 0.00 10.79 b 25.5b 46.39 a 87.14 a,b 115.88 b,c 100.56 c 51.20 a,b 10.28 5.87 c 0.13 c NLDAS - 2:Mosaic 10.36 b,c 11.16 b,c 27.61 c 62.12 97.31 120.41 d 137.01 114.34 82.92 c 48.35 b 21.82 a 11.61 b NLDAS - 2:Noah 8.9 2 b 11.43 b 1 7.3 7 d 27.56 b 47.86 a,b 81.78 a 107.66 d 103.26 b,c 66.68 d 26.46 a 9.49 6.62 d NLDAS - 2:VIC 7.18 9.66 c 9.78 b 15.29 52.70 b 96.55 b,c,d 121.99 b,c 98.52 a,b,c 51.47 a,b 16.65 5.61 c 6.86 c,d TerraClimate * * 19.11 b,c,d 82.61 101.94 110.81 c,d 97.24 a,d 86.27 a ,c 65.13 a,d ,e 49.85 b 23.04 a 1 .56 c,d ALEXI 23.70 38.55 51.89 57.48 81.03 100.68 b,c,d 122.28 b 101.41 c 66.55 d 32.04 c 19.81 a 16.32 a SWAT 3.71 5.51 28.36 c 41.82 a 60.15 91.57 a,b,c,d 97.47 a 83.41 a 69.58 b,c,d,e 21.39 16.53 b,d 6.81 d,e Ensemble 11.72 c 16.04 26.49 c 44.03 a 68 .23 c 95.68 b,c 116. 02 c 99.82 c 59.51 e 29.40 c 15.52 d 9.28 e *Note that no ETa values were provided for TerraClimate for the months of January and February. 181 Table S 5. 14. Average monthly ETa values for each dataset for urban low - density (URLD) reg ions with c lus ters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 16.21 a 22.76 a 36.78 a 38.13 58.52 a 73.79 80.97 a 67.57 a 40.12 25.04 a 19.83 a 15.19 a MOD16A2 500m 16.61 a 22.1 8 a 36.66 a 4 5.2 3 a 74.21 b 95.75 a,b 104.37 b,c 90.22 b 49.74 a 26.05 a 16.76 b 10.50 b,c SSEBop 0.03 0.00 9.25 b 24.62 47.24 c 83.60 c 106.69 b,c 89.65 b,c 46.21 a 9.49 5.32 c 0.00 d NLDAS - 2:Mosaic 10.23 b,c 10.78 b,c 28.59 c,d 64.08 97.66 122.43 d 141.45 117.43 82.50 47.09 b 21.94 a,d 1 1.7 2 a,b,c NLDAS - 2 :Noah 8.64 b 11.49 b 17.67 e 28.52 49.79 c 83.50 c 109.00 b,c 104.20 d 67.50 b,c 26.43 a 9.51 6.58 e NLDAS - 2:VIC 6.74 9.39 c 9.39 b 15.30 54.90 a 99.87 a,b,d 123.68 95.91 b,c,d 49.36 a 16.33 5.54 c 6.55 d,e TerraClimate * * 19.93 b,c,d,e 82.8 8 102.35 11 0.2 8 a,d 96.94 b 86. 72 a,b,c 65.26 a,b,c,d,e 50.15 b 23.17 a,d 1.71 d,e ALEXI 21.85 35.83 46.70 52.02 77.22 b 95.48 a,b,d 111.87 c 93.05 b,c 60.51 b,d,e 28.39 a,c 17.22 b 14.56a,b SWAT 3.56 5.30 29.05 c 45.84 a 66.81 d 90.46 b 77.07 a 72.28 a 66.22 b,c,d 41.10 22. 90 d 8.18 c,e ,f Ensemble 11.37 c 15.88 25.68 d 43.85 a 70.23 d 95.59 a,b 109.37 b,c 93.09 c 57.65 d,e 28.62 c 14.91 8.90 f *Note that no ETa values were provided for TerraClimate for the months of January and February. 182 Table S 5. 15. Average monthly ETa values for each datase t f or urban transp ortation (UTRN) regions with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 16.18 a 22.19 a 36.99 a 37.36 a 57.46 a 75.79 85.54 a 71.71 40.3 3 25.04 a 20 .16 a 15.38 a MOD16 A2 500m 16.47 a 22.08 a 37.30 a 44.12 b 72.76 b 97.53 a,b 109.56 b,c 94.02 a,b,c 50.27 a 26.08 a 17.21 10.62 b SSEBop 0.04 0.00 9.85 b 25.17 48.79 c 86.63 c 104.63 b 88.93 a,b 46.98 a 9.78 5.99 b 0.02 c NLDAS - 2:Mosaic 10.07 b,c 10.37 b,c 28.49 c 6 6.04 100.74 d 1 25.79 d 141.71 1 15.98 d 81.78 b 47.03 b 22.25 a 11.71 b NLDAS - 2:Noah 8.64 b 11.41 b 18.08 d 29.38 50.91 c 84.95 c 110.27 b,c 105.60 d 67.49 c 26.55 a 9.63 6.64 d NLDAS - 2:VIC 6.86 9.55 c 9.84 b 15.22 56.93 a 103.21 a,b,d 124.31 d 94.47 a,b,c,d 49.18 a 16.27 5.22 b 6.53 c,d Te rra Climate * * 2 0.10 b,c,d 82.80 102.16 d 110.25 a,d 96.89 a,b 86.41 a,b,c 65.39 a,b,c,d 50.13 b 23.20 a 1.70 c,d ALEXI 23.64 38.68 51.38 57.88 83.27 100.33 a,b,d 117.43 d 96.80 a,c 65.53 c 32.10 c 19.52 a 16.13 a SWAT 3.34 5.09 26.87 c 36.87 a 46.94 c 58.63 4 3.27 31.05 24. 32 24.24 a 14.65 c 6.44 d Ensemble 11.60 c 16.14 26.57 c 44.75 b 71.63 b 98.06 b 111.29 c 94.24 a,b,c 58.37 d 29.12 c 15.40 c 9.15 *Note that no ETa values were provided for TerraClimate for the months of January and February. 183 Table S 5. 16. Average mo nthly ETa v alu es for each dat aset for woody wetlands (WETF) regions with clusters indicated by superscripts for each column Datasets Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 16.92 23.25 37.42 a 38.21 66.47 a 94.12 a 103. 31 a 83.93 a 45. 87 25.34 a 20.38 a 15.77 a MOD16A2 500m 15.22 21.02 37.68 a 45.59 a,b 88.12 b 123.71 b 135.59 b 113.85 b 61.03 a,b 26.61 a,b 17.04 10.51 b,c SSEBop 0.04 0.06 11.24 b 27.57 c 54.16 99.80 c,d 121.07 c,d,e 101.85 c 55.17 c 13.86 5.94 b 0.91 d NLDAS - 2:Mosaic 10.9 1 a,b 12.10 a 26 .27 c 57.00 d 92. 52 113.17 c,e 128.98 b,c 110.75 b 83.71 49.82 c 21.37 a,c 11.62 b NLDAS - 2:Noah 10.24 a 12.17 a 17.28 d 26.18 c 42.43 74.71 102.39 a 99.51 a,c 66.06 a,b 27.43 b 9.91 7.20 e NLDAS - 2:VIC 7.84 9.90 10.11 b 15.82 47.97 88.48 f 117.67 d,e 101.79 c 53 .98 c 17.36 6.1 6 b 7.39 e Terra Climate * * 19.00 b,c,d 82.51 101.44 110.33 b,c,d,e 96.77 a 86.86 a,c 64.95 a,b,c 49.65 c 22.99 a,c 1.61 d ALEXI 22.73 36.62 49.93 55.87 d 84.05 b 106.14 c,d,e 124.14 b,c,d 99.60 a,c 66.29 a 32.18 d 19.14 a 15.93 a SWAT 4.05 5.95 32.01 47.9 5 a 68.48 a,c 90 .46 a,f 73.93 69 .21 69.37 a 42.51 23.53 c 8.65 c,e Ensemble 11.87 b 16.27 26.20 c 43.59 b 72.15 c 101.31 c,d 116.24 e 99.77 c 62.13 b 30.28 d 15.37 9.48 c *Note that no ETa values were provided for TerraClimate for the months of January and February. 184 Table S 5. 17 . A verage monthly ETa values for each dataset for winter wheat (WWHT) regions with clusters indicated by superscripts for each column Datasets Months Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. MOD16A2 1km 15.42 a 20.72 a 36.41 a 37.83 a 58.2 3 a 80.27 a 92.28 a 7 4.20 41.66 25.23 a 19.26 a 14.34 a MOD16A2 500m 15.69 a 21.01 a 37.35 a 45.03 b 76.28 104.87 b 121.46 b,c 98.00 a 53.81 a,b 27.21 b 16.88 b 10.64 b SSEBop 0.02 0.00 10.53 b 27.51 c 50.91 93.23 c 115.93 b,c 94.20 a,b 48.58 a 12.23 5.80 c 0.26 c NL DAS - 2:Mosai c 1 1.03 b 12.21 b 26 .71 c 57.31 d 93.38 119.10 d 137.87 119.03 87.49 49.58 c 21.76 a 11.65 b NLDAS - 2:Noah 10.85 b 13.35 b 20.74 d 29.79 c 42.59 b 70.87 99.49 d 98.37 a,b 68.29 c 30.37 d 11.46 7.82 d NLDAS - 2:VIC 7.43 9.54 9.96 b 15.21 45.93 b 86.00 a 114.75 b,c 96.4 5 a,b 49.25 a ,b 16.01 6.12 c 7.0 9 d TerraClimate * * 17.82 b,c,d 81.99 101.83 110.61 b,d 97.82 a,d 87.96 a,b 65.25 b,c,d 49.51 c 22.35 a 1.33 c ALEXI 23.95 37.88 51.36 57.09 d 84.43 106.75 b 124.53 b 99.22 a,b 66.17 c 33.15 d 19.89 a 16.52 a SWAT 4.01 5.45 27.61 c 39.83 a 62.54 a 108. 06 b 100.05 d 58.82 30.89 26.96 a,b 16.06 b,d 6.72 d Ensemble 11.94 b 16.21 26.45 c 43.97 b 69.20 96.46 c 113.02 c 95.93 b 60.06 d 30.41 d 15.44 d 9.32 *Note that no ETa values were provided for TerraClimate for the months of January and February. 185 Table S 5. 18. Ave rag e seasonal ETa values for each dataset for agricultural lands with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 17.18 a 43.21 a 85.95 29.34 a MOD16A2 500m 16.21 a 50.74 111.18 a 32.2 7 a,b,c SSE Bop 0.15 b 27.87 b 1 01.01 b,c 22.52 d NLDAS - 2:Mosaic 11.28 c 62.02 c 125.43 51.69 e NLDAS - 2:Noah 9.55 31.15 95.01 b 35.00 b,c NLDAS - 2:VIC 8.00 25.33 b 103.17 c 24.04 d TerraClimate 1.08 b 68.70 c 98.47 b,c 45.82 e ALEXI 25.96 63.92 c 108.99 a 39.77 SWAT 5.3 3 44.44 a 93 .72 b,c 30.50 a,b E nsemble 12.37 c 46.47 a 103.65 c 35.06 c 186 Table S 5. 19. Average seasonal ETa values for each dataset for forest lands with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1 km 18.02 47 .86 a 94.63 a 30.74 MOD16A2 500m 15.24 57.85 b 127.57 36.53 a SSEBop 0.47 a 30.81 c 107.49 b 24.80 b NLDAS - 2:Mosaic 12.15 b,c 57.14 b 121.72 53.35 NLDAS - 2:Noah 11.51 b 29.82 c 84.49 37.04 a NLDAS - 2:VIC 8.52 22.28 94.28 a 23.87 b TerraClimate 0.94 a 67.88 d 99.26 a,c 4 5.4 9 c,d ALEXI 24. 60 63.61 d 111.09 b 39.76 c SWAT 5.84 43.38 72.94 43.63 d Ensemble 12.68 c 47.01 a 105.07 c 36.45 a 187 Table S 5. 20. Average seasonal ETa values for each dataset for urban lands with clusters indicated by superscripts for each column Datasets S eas ons Winter Sp ring Summer Fall MOD16A2 1km 17.98 44.17 a 76.14 28.43 MOD16A2 500m 16.41 51.67 98.82 a 31.04 a SSEBop 0.01 a 27.55 b 93.36 b 20.67 NLDAS - 2:Mosaic 10.80 b 64.38 c 127.51 50.42 b NLDAS - 2:Noah 8.90 32.45 99.68 a 34.52 a,c NLDAS - 2:VIC 7.61 26.99 b 10 6.97 c 23.64 Te rraClimate 1.34 a 69.75 c 97.91 a,b 46.22 b ALEXI 25.26 61.79 c 102.82 a,c 37.46 SWAT 5.27 41.35 59.67 30.70 a Ensemble 12.19 b 47.19 a 100.40 a 34.05 c 188 Table S 5. 21. Average seasonal ETa values for each dataset for wetland lands wi th clusters in dicated by supe rscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 18.65 47.37 a 93.79 a 30.53 a MOD16A2 500m 15.58 57.13 b 124.38 34.89 b SSEBop 0.34 a 30.99 c 107.57 b,c 24.99 c NLDAS - 2:Mosaic 11.55 b 58.60 b,d 117.63 51.63 d NL DAS - 2:Noah 9.87 28 .63 c 92.20 a 34.47 a,b NLDAS - 2:VIC 8.38 24.63 102.64 a,b,d 25.84 c TerraClimate 1.26 a 69.04 d 97.99 a,d 45.86 d,e ALEXI 25.09 63.29 b,d 109.96 c 39.20 SWAT 6.22 49.48 a 77.87 45.13 e Ensemble 12.54 b 47.31 a 105.77 b,c,d 35.93 b 189 Ta ble S 5. 22. Ave rage seasonal E Ta values for each dataset for alfalfa (ALFA) regions with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 16.89 47.22 a 89.17 a,b 29.90 MOD16A2 500m 15.45 56.84 b 119. 29 c 35.04 a SS EBop 0.38 a 30.5 5 c 106.85 d 24.82 b NLDAS - 2:Mosaic 12.10 b,c 56.19 b 120.06 c 52.43 NLDAS - 2:Noah 11.63 b 29.55 c 82.68 a 36.85 a NLDAS - 2:VIC 8.58 22.48 94.82 e 23.86 b TerraClimate 0.89 a 67.61 d 99.22 d,e,f 45.43 c ALEXI 24.75 64.09 d 111.37 40.05 c SWA T 4.34 39.6 9 8 7.59 b 39.01 a,c Ensemble 12.58 c 46.67 a 102.93 f 36.05 a 190 Table S 5. 23. Average seasonal ETa values for each dataset for corn (CORN) regions with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1 km 17.17 a 42.59 a 8 5.39 29.32 a MOD16A2 500m 16.29 a 49.79 110.43 a 32.04 a,b,c SSEBop 0.13 b 27.17 b 99.69 b,c 22.10 d NLDAS - 2:Mosaic 11.13 c 63.06 c 126.14 51.36 e NLDAS - 2:Noah 9.21 31.46 97.01 b 34.66 b,c NLDAS - 2:VIC 7.93 25.94 b 104.66 a,b,c 23.99 a,d TerraClimat e 1 .08 b 68.75 c 98. 39 b,c 45.86 e ALEXI 26.11 63.95 c 108.75 a,c 39.80 SWAT 5.35 44.96 a 98.54 b,c 28.33 a,b,d Ensemble 12.32 c 46.44 a 103.81 c 34.89 c 191 Table S 5. 24. Average seasonal ETa values for each dataset for field peas (FPEA) regions with clus ters indica ted by superscript s for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 17.43 46.10 a 86.94 a 29.44 a MOD16A2 500m 15.89 55.21 b 111.46 b 33.15 SSEBop 0.31 a 31.83 c 111.03 b 25.90 a,b NLDAS - 2:Mosaic 12.10 b,c 55.24 b 118.66 52.37 NLDAS - 2:No ah 11.63 b 30.13 c 8 6.07 a 37.69 c,d NLDAS - 2:VIC 8.53 22.33 94.16 c 24.02 b TerraClimate 1.19 a 70.70 d 98.77 c,d 45.74 ALEXI 25.23 64.84 d 112.26 b 39.63 c SWAT 5.25 40.15 81.10 a 25.80 b Ensemble 12.78 c 46.75 a 102.42 d 35.99 d 192 Table S 5. 25. Average s easonal ETa va lues for each d ataset for deciduous forest (FRSD) regions with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 18.02 47.86 a 94.62 a 30.74 MOD16A2 500m 15.24 57.84 b 127.57 36.53 a SS EBop 0.47 a 30. 81 c 107.49 b 24. 80 b NLDAS - 2:Mosaic 12.15 b,c 57.14 b 121.72 53.35 NLDAS - 2:Noah 11.51 b 29.82 c 84.49 37.04 a NLDAS - 2:VIC 8.52 22.28 94.28 a 23.87 b TerraClimate 0.94 a 67.88 d 99.26 a,c 45.49 c,d ALEXI 24.60 63.61 d 111.10 b 39.76 c SWAT 5.84 43.38 72 .94 43.64 d En semble 12.68 c 4 7.01 a 105.07 c 36.45 a 193 Table S 5. 26. Average seasonal ETa values for each dataset for evergreen forest (FRSE) regions with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A 2 1km 17.96 54 .56 116.69 33.4 0 a MOD16A2 500m 15.78 62.26 a 148.97 39.61 b,c SSEBop 0.94 a 32.19 b 109.74 a 26.65 d NLDAS - 2:Mosaic 12.22 b,c 59.03 c 131.58 56.49 NLDAS - 2:Noah 12.06 b 31.60 b 82.49 b 38.65 b NLDAS - 2:VIC 8.05 21.62 92.68 c 22.98 TerraClimate 0.85 a 6 9.10 a 99.23 c 4 5.22 c ALEXI 25 .32 62.47 a,c 109.04 a 38.71 b SWAT 4.98 41.35 77.61 b 30.20 a,d Ensemble 12.94 c 48.82 111.30 a 37.71 b 194 Table S 5. 27. Average seasonal ETa values for each dataset for hay (HAY) regions with clusters indicated by superscripts for each colum n D atasets Seasons Winter Spring Summer Fall MOD16A2 1km 18.76 47.66 a 93.90 a 30.69 a MOD16A2 500m 16.23 55.29 b 115.18 b,c 32.90 a SSEBop 0.15 a 30.19 c 102.98 d 23.04 NLDAS - 2:Mosaic 10.71 56.73 b 118.41 b 46.91 b NLDAS - 2:Noah 8.54 b 27.90 c,d 94.22 a 32.20 a NL DAS - 2:VIC 8.03 b 27 .17 d 112.8 b,c 26.38 TerraClimate 1.29 a 69.74 e 97.69 a,d 46.09 b,c ALEXI 26.79 65.84 e 110.51 c 40.59 b,c SWAT 5.77 40.26 69.05 38.60 c Ensemble 12.50 47.42 a 105.71 34.85 195 Table S 5. 28. Average seasonal ETa values for each datas et for past ure (PAST) regions with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 17.46 47.09 a 90.08 a,b 29.87 MOD16A2 500m 15.33 56.08 b 118.68 33.96 SSEBop 0.38 a 31.29 c 108.11 c 25.03 a NLDAS - 2 :Mosaic 12. 26 b ,c 58.22 b 125.4 5 54.28 NLDAS - 2:Noah 11.44 b 29.96 c 84.47 a 37.00 b NLDAS - 2:VIC 8.42 21.98 93.72 a,b,d 23.70 a TerraClimate 0.95 a 67.76 d 99.16 b,d 45.39 c ALEXI 24.87 64.08 d 110.48 c 39.92 b,c SWAT 5.53 45.30 a 68.77 37.47 b,c Ensemble 12.63 c 46.91 a 103.77 d 3 6.1 4 b 196 Table S 5. 29. Average seasonal ETa values for each dataset for sugar beet (SGBT) regions with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 17.58 a 40.50 81.62 a 29.62 a MOD1 6A2 500m 17 .75 a 45.95 a 101.21 b,c,d 31.05 a SSEBop 0.15 b 24.52 b 93.20 a,b,e 21.21 b NLDAS - 2:Mosaic 10.57 65.56 c 127.58 49.96 c NLDAS - 2:Noah 8.86 32.97 100.77 b,c 34.42 a,d NLDAS - 2:VIC 7.74 27.85 b 108.20 c,d,f 23.63 b TerraClimate 1.17 b 69.00 c 98.02 b,c,e,f 46.0 1 c,e ALEXI 27 .26 64.04 c 106. 65 b,d,f 40.11 e SWAT 5.38 47.67 a 87.41 a,e 34.41 d Ensemble 12.59 46.16 a 102.16 b,c 34.50 d 197 Table S 5. 30. Average seasonal ETa values for each dataset for soybean (SOYB) regions with clusters indicated by superscripts for each column Data set s Seasons Win ter Spring Summer Fall MOD16A2 1km 17.26 a 42.25 a,b 86.63 a 29.25 a MOD16A2 500m 16.69 a 49.43 109.49 b 31.48 b SSEBop 0.04 b 27.56 c 101.19 c 22.45 c NLDAS - 2:Mosaic 11.04 c 62.35 d 123.92 51.03 d NLDAS - 2:Noah 8.99 30.93 97.57 d 34.21 a, b,e NLDAS - 2:V IC 7.90 25.92 c 105.69 b,c 24.58 c TerraClimate 1.22 b 69.28 d 98.11 c,d 46.01 d ALEXI 26.19 63.47 d 108.13 b 39.47 f SWAT 5.34 43.44 a 90.82 a,d 35.84 a,b,e,f Ensemble 12.35 c 46.25 b 103.84 c 34.81 e 198 Table S 5. 31. Average seasonal ETa values for eac h dataset f or urban low - densi ty (URLD) regions with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 18.05 44.47 a 74.11 28.33 MOD16A2 500m 16.43 52.03 96.78 a,b 30.85 a SSEBop 0.01 a 27.04 b 93.31 a 20.34 NLDA S - 2 :Mosaic 10.91 b 63.44 c,d 127.10 50.51 b NLDAS - 2:Noah 8.90 31.99 98.90 a,b 34.48 a,c NLDAS - 2:VIC 7.56 26.53 b 106.49 c 23.74 TerraClimate 1.34 a 69.77 c 97.98 a,b,c 46.19 b,d ALEXI 24.08 58.65 d 100.13 a,b 35.37 c SWAT 5.68 47.23 a 79.94 43.40 d Ensemb le 12.05 b 4 6.5 9 a 99.35 b 33.73 c 199 Table S 5. 32. Average seasonal ETa values for each dataset for urban transportation (UTRN) regions with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 17.92 43. 94 a 77.68 2 8.5 1 MOD16A2 500m 16.39 51.39 100.37 a 31.18 a SSEBop 0.02 a 27.94 b 93.40 b 20.92 b NLDAS - 2:Mosaic 10.71 b 65.09 c 127.82 50.35 c NLDAS - 2:Noah 8.90 32.79 100.27 a,c 34.56 a,d NLDAS - 2:VIC 7.65 27.33 b 107.33 c 23.56 e TerraClimate 1.33 a 69.73 c 97.85 a,b, c 46.24 c A LEX I 26.15 64.18 c 104.85 a,c 39.05 SWAT 4.95 36.90 44.31 21.07 b,e Ensemble 12.30 b 47.65 a 101.20 a 34.30 d 200 Table S 5. 33. Average seasonal ETa values for each dataset for woody wetlands (WETF) regions with clusters indicated by superscripts for each colum n D atasets Seasons Winter Spring Summer Fall MOD16A2 1km 18.65 47.37 a 93.79 a 30.53 a MOD16A2 500m 15.58 57.13 b 124.38 34.89 b SSEBop 0.34 a 30.99 c 107.57 b,c 24.99 c NLDAS - 2:Mosaic 11.55 b 58.60 b,d 117.63 51.63 d NLDAS - 2:Noah 9.87 28.63 c 92.20 a 34.47 a,b N LDA S - 2:VIC 8.38 24 .63 102.64 a,b,d 25.84 c TerraClimate 1.26 a 69.04 d 97.99 a,d 45.86 d,e ALEXI 25.09 63.29 b,d 109.96 c 39.20 SWAT 6.22 49.48 a 77.87 45.13 e Ensemble 12.54 b 47.31 a 105.77 b,c,d 35.93 b 201 Table S 5. 34. Average seasonal ETa values for each datase t f or winter wheat (WWHT) regions with clusters indicated by superscripts for each column Datasets Seasons Winter Spring Summer Fall MOD16A2 1km 16.83 a 44.16 a,b 82.25 a 28.72 MOD16A2 500m 15.78 a 52.89 108.11 b 32.63 SSEBop 0.09 b 29.65 c 101.12 c 22.20 a N LDA S - 2:Mosaic 11.6 3 c,d 59.14 d 125.33 52.94 NLDAS - 2:Noah 10.67 c 31.04 c 89.58 a,c,d 36.71 b NLDAS - 2:VIC 8.02 23.70 99.07 c 23.79 a TerraClimate 1.04 b 68.63 e 98.80 c,d 45.71 c ALEXI 26.12 64.29 d,e 110.16 b 39.74 c SWAT 5.39 43.33 a 88.98 a,d 24.63 a Ens emble 12.49 d 4 6.54 b 101.80 c 3 5.30 b 202 Table S 5. 35. Average seasonal values of the MOD16A2 500 m dataset for each major landuse category for each column Landuse Season Winter Spring Summer Fall Agriculture 16.21 50.74 111.18 32.27 Forest 15.24 57.85 12 7.57 36.53 Ur ban 16.41 51.67 98.82 31.04 Wetland 15.58 57.13 124.38 34.89 203 Table S 5. 36. Average seasonal values of the SSEBop dataset for each major landuse category for each column Landuse Season Winter Spring Summer Fall Agriculture 0.15 27.87 a 1 01.01 22.52 F orest 0.47 30.8 1 b 107.49 a 24.80 a Urban 0.01 27.55 a 93.36 20.67 Wetland 0.34 30.99 b 107.57 a 24.99 a 204 Table S 5. 37. Average seasonal values of the NLDAS - 2 Mosaic dataset for each major landuse category for each column Landuse Season Winter Spring Sum mer Fall Agricult ure 11.28 a 62.02 125.43 51.69 a Forest 12.15 57.14 121.72 53.35 Urban 10.80 64.38 127.51 50.42 Wetland 11.55 a 58.60 117.63 51.63 a 205 Table S 5. 38. Average seasonal values of the NLDAS - 2 Noah dataset for each major landuse cat egory for e ach column Landuse Season Winter Spring Summer Fall Agriculture 9.55 31.15 a 95.01 35.00 Forest 11.51 29.82 a,b 84.49 37.04 Urban 8.90 32.45 99.68 34.52 a Wetland 9.87 28.63 b 92.20 34.47 a 206 Table S 5. 39. Average seasonal values of the NLDAS - 2 VIC datas et for each major landuse category for each column Landuse Season Winter Spring Summer Fall Agriculture 8.00 25.33 103.17 a 24.04 a Forest 8.52 a 22.28 94.28 23.87 a,b Urban 7.61 26.99 106.97 23.64 a,b Wetland 8.38 a 24.63 102.64 a 25.84 207 Tabl e S 5. 40. Av era ge seasonal val ues of the TerraClimate dataset for each major landuse category for each column Landuse Season Winter Spring Summer Fall Agriculture 1.08 a 68.70 98.47 45.82 a Forest 0.94 a 67.88 99.26 45.49 Urban 1.34 a 69.75 97.91 a 46.22 W etland 1.26 a 6 9.04 97.99 a 45. 86 a 208 Table S 5. 41. Average seasonal values of the ALEXI dataset for each major landuse category for each column Landuse Season Winter Spring Summer Fall Agriculture 25.96 63.92 a,b 108.99 39.77 a Forest 24.60 a 63.61 a,b 111. 09 a 39.76 a Ur ban 25.26 a 61.7 9 102.82 37.46 Wetland 25.09 a 63.29 b 109.96 a 39.20 209 Table S 5. 42. Average seasonal values of the SWAT model dataset for each major landuse category for each column Landuse Season Winter Spring Summer Fall Agriculture 5.33 a 44.44 93. 72 30.50 a Forest 5.84 43.38 72.94 43.63 Urban 5.27 a 41.35 59.67 30.70 a Wetland 6.22 49.48 77.87 45.13 210 Table S 5. 43. Average seasonal values of the Ensemble dataset for each major landuse category for each column Landuse Season Winter Spr ing Summer Fal l Agriculture 12.37 46.47 a 103.65 35.06 Forest 12.68 47.01 a,b 105.07 36.45 Urban 12.19 47.19 b 100.40 34.05 Wetland 12.54 47.31 b 105.77 35.93 211 Table S 5. 44. Average monthly values of the MOD16A2 1km dataset for each major landuse catego ry for each co lumn Landuse Mo nth Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Agriculture 15.63 a 20.94 37.11 a,b 37.62 a 54.89 76.62 a 98.07 83.15 a,b 42.37 25.26 a 20.41 a 14.97 a Forest 16.47 b 22.63 a 37.46 a 38.79 67.33 95.23 104.49 84.17 a 47.1 9 25.36 a 19 .66 b 14.97 a Urban 16.20 a,b 22.44 a 36.90 b 37.69 a 57.92 74.93 a 83.57 69.92 40.24 25.04 20.02 b,c 15.29 a Wetland 16.92 23.25 37.42 a 38.21 66.47 94.12 103.31 83.93 a,b 45.87 25.34 a 20.38 a,c 15.77 212 Table S 5. 45. Average monthly values of the MOD16A 2 500 m dat ase t for each majo r landuse category for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Agriculture 16.29 21.39 37.69 a,b,c 44.50 a 70.02 98.24 a 126.96 108.35 53.02 26.39 17.4 10.95 Forest 14.95 20.47 37.1 7 a,b,d 45.7 8 b 90.59 127.14 13 9.30 116.29 a 65.95 27.10 16.56 10.30 a Urban 16.53 22.12 37.02 a,c,d 44.60 a 73.39 96.76 a 107.32 92.38 50.04 26.06 17.01 a 10.57 a,b Wetland 15.22 21.02 37.68 b,c,d 45.59 b 88.12 123.71 135.59 113.85 a 61.03 26.61 17.04 a 10.51 b 213 T able S 5. 46. Av erage monthly v alues of the SSEBop dataset for each major landuse category for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Agriculture 0.02 a,b,c 0.00 10.29 a 26.16 a 47.15 a 87.57 115.86 99.60 a 50.88 1 0.96 5.73 a 0.4 2 Forest 0.02 a 0.01 10.02 a,b 28.15 b 54.26 b 99.60 a 122.22 a 100.66 a,b 54.51 a 14.15 a 5.75 a 1.37 Urban 0.04 a,b,c 0.00 9.59 a 24.94 a 48.12 a 85.32 105.52 89.24 46.65 9.66 5.70 a 0.01 Wetland 0.04 c 0.06 11.24 b 27.57 b 54.16 b 99.80 a 121.07 a 101.85 b 5 5.17 a 13.86 a 5 .94 a 0.91 214 Table S 5. 47. Average monthly values of the NLDAS - 2 Mosaic dataset for each major landuse category for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Agriculture 10.67 a 11.39 a 27.43 a 61. 21 97.41 12 1.7 3 138.43 116.14 a 84.49 a,b 48.54 22.03 a 11.79 a Forest 11.60 b 13.03 25.57 a 53.95 91.88 a 116.00 133.13 116.03 a 87.69 50.70 a 21.65 a,b,c 11.81 a Urban 10.14 a,b 10.55 b 28.53 a 65.19 99.41 124.34 141.60 116.60 a,b 82.09 a 47.06 22.12 a,b,c 11.71 a Wetl and 10.91 a 12. 10 a,b 26.27 a 57 .00 92.52 a 113.17 128.98 110.75 b 83.71 a,b 49.82 a 21.37 c 11.62 a 215 Table S 5. 48. Average monthly values of the NLDAS - 2 Noah dataset for each major landuse category for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul . Aug. Sep. Oc t. Nov. Dec. A griculture 9.58 12.02 a 18.73 28.52 46.18 78.30 105.06 101.68 67.06 a 27.73 10.19 7.05 a Forest 11.95 14.22 21.65 29.56 a 38.25 64.11 94.43 94.91 67.71 a 31.36 12.05 8.35 Urban 8.64 11.44 17.90 29.01 a 50.43 84.32 109.72 105.00 67. 49 a 26.50 9 .58 a 6.62 Wetland 10.24 12.17 a 17.28 26.18 42.43 74.71 102.39 99.51 66.06 27.43 9.91 a 7.20 a 216 Table S 5. 49. Average monthly values of the NLDAS - 2 VIC dataset for each major landuse category for each column Landuse Month Jan. Feb. Mar. Apr. M ay Jun. Jul . A ug. Sep. Oct. N ov. Dec. Agriculture 7.37 9.68 a 10.08 a 15.29 a 50.63 93.34 119.23 a 96.95 a 50.05 a 16.29 a 5.78 6.95 a Forest 8.14 a 9.92 a 10.57 a 15.36 a 40.91 77.83 108.71 96.31 a 49.27 a 15.83 a 6.51 a 7.49 a,b Urban 6.81 9.48 a 9.65 b 15.26 a 56.05 101 .77 124.04 95. 09 a,b 49.26 a 16 .29 a,b 5.36 6.54 a Wetland 7.84 a 9.90 a 10.11 a,b 15.82 47.97 88.48 117.67 a 101.79 b 53.98 17.36 b 6.16 a 7.39 b 217 Table S 5. 50. Average monthly values of the TerraClimate dataset for each major landuse category for each column La nduse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Agriculture * * 18.18 81.94 a 101.76 a 110.83 a 97.77 86.81 a 65.31 a 49.58 a 22.56 1.39 a Forest * * 16.50 81.41 a,b 101.45 a,b,c 110.58 a,b 98.70 88.50 65.30 a,b 49.22 21.96 1.20 a Urban * * 20.0 2 8 2.83 a,b 102.24 b 110.26 b 96.91 b 86.54 a,b 65.33 a,b 50.14 23.19 1.70 a Wetland * * 19.00 82.51 b 101.44 c 110.33 b 96.77 b 86.86 a,b 64.95 b 49.65 a 22.99 1.61 a *Note that no ETa values were provided for TerraClimate for the months of January and Fe bruary. 218 Tab le S 5. 51. Avera ge monthly values of the ALEXI dataset for each major landuse category for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Agriculture 23.37 a 38.17 a 51.98 57.51 82.28 102.76 123.09 a 101.1 3 a 66.95 a,b 32 .60 a,b 19.75 16 .35 a Forest 22.10 a 35.72 49.88 a 55.66 a 85.29 a 108.60 a 125.04 a 99.64 a,b 67.11 a 33.04 a 19.12 a 15.99 a,b,c Urban 22.87 a 37.45 a,b 49.36 a 55.35 a 80.66 98.24 115.04 95.18 63.36 30.50 18.53 a 15.45 b Wetland 22.73 a 36.62 b 49.93 a 55.87 a 84.05 a 10 6.1 4 a 124.14 a 99.6 0 b 66.29 b 32.18 b 19.14 a 15.93 b,c 219 Table S 5. 52. Average monthly values of the SWAT model dataset for each major landuse category for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Agri culture 3.7 1 5 .40 a 28.54 a 42. 24 a 62.54 a 106.19 102.18 72.81 a 48.28 26.33 16.87 6.88 a Forest 3.53 a 5.32 a 28.90 a 40.11 b 61.12 a 84.92 68.43 65.47 65.84 42.02 a 23.04 8.65 b Urban 3.44 a 5.18 a 27.81 a 40.74 a,b 55.50 72.35 57.84 48.82 42.38 31.51 18.20 7.19 a Wet land 4.05 5 .95 32.01 47.95 68 .48 90.46 73.93 69.21 a 69.37 42.51 a 23.53 8.65 b 220 Table S 5. 53. Average monthly values of the Ensemble dataset for each major landuse category for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Agr iculture 11.74 a 16.05 a 26.53 a 44.09 a 68.79 96.17 a 115.56 a 99.23 a 60.02 29.67 15.48 a 9.34 a Forest 12.06 b 16.40 a,b 26.21 a,b 43.58 b 71.25 a 99.89 115.75 a 99.56 a 63.09 30.84 15.41 a,b 9.59 a Urban 11.50 16.03 a,b 26.19 a,b 44.36 a 71.03 a 96.99 a 110.4 6 93.75 58. 06 28.91 15.19 b 9. 04 Wetland 11.87 a,b 16.27 b 26.20 b 43.59 b 72.15 101.31 116.24 a 99.77 a 62.13 30.28 15.37 a,b 9.48 a 221 Table S 5. 54. Average monthly values of the MOD16A2 1km dataset for each individual landuse with clusters indicated by supersc rip ts for each col umn Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. ALFA 15.54 a 20.9 a 36.75 a,b,c 39.18 a 65.74 a,b 87.94 a 99.00 a,b 80.57 a 45.81 a 24.95 a,b,c,d 18.95 a 14.23 a,b CORN 15.58 a,b 20.88 a,b 37.24 a,b,d 37.44 b,c ,d 53.10 74 .38 b,c 97.88 a 83.9 2 b,c 42.05 b 25.28 a,b,c,e,f 20.62 b 15.04 c,d,e,f FPEA 15.86 a,b,c 21.59 a,b,c 37.02 a,b,c,d 38.28 b,e 63.00 90.61 a,d 97.33 a,b,c 72.87 d,e 43.22 b,c,d 25.34 a,b,c,d,e,f,g 19.76 c,d,e,f 14.84 c,d,e,f,g FRSD 16.47 c,d,e 22.63 c,d,e 37.46 a,c, d 38.79 a 67 .33 c 95.22 e 104.48 d 84.16 b,c,f 47.19 25.36 a,d,e,f,g 19.67 c,d 14.97 c,d,e,g FRSE 16.39 a,b,c,d,e 23.34 c,d,f 36.79 a,b,c,d 42.19 84.70 124.26 128.12 97.69 55.75 26.14 g,h 18.33 14.14 a,b,c,f,g HAY 16.87 d,e 23.04 d,e,f 37.52 a,c,d 38.24 a,b,e 67.22 a,c 95 .36 e,f 102. 43 b ,d,e 83.90 b,c,f 45.38 a,c 25.54 b,e,f,g,h 21.15 g 16.37 PAST 16.05 a,b,c 21.7 b,c 37.03 a,b,c 38.83 a 65.41 b 89.89 d 99.98 a,b 80.36 a 45.00 c 25.15 a,b,c,d,e 19.46 e 14.63 a,c,d,f,g SGBT 15.90 a,b,c,d,e 21.23 a,b,c,e 38.44 d 36.77 b,c,f 46.30 65.78 95.42 a,b ,c 83.65 a,b ,c, f 40.79 b,d 25.8 9 f,g,h 22.18 15.59 d,e,f,g,h SOYB 15.67 a,b 20.84 a,b 36.95 b,c,d 37.22 c,f 52.58 74.39 b,c 99.86 a,b 85.63 b,f 41.64 b,d 25.31 a,b,c,e,f,h 20.79 g 15.26 d,e,f,g,h URLD 16.21 a,b,c,d 22.76 d,e,f 36.78 a,b,c 38.13 b,e 58.52 d 73.79 b 80.97 67.5 7 40.12 d 25 .04 a,b,c,d 19.83 c, d 15.19 c,d,e,f,g UTRN 16.18 b,c,d 22.19 b,c,e,f 36.99 a,b,c,d 37.36 c,d,f 57.46 e 75.79 c 85.54 71.71 d 40.33 d 25.04 b,c,d,f 20.16 f 15.38 c,d,e,g,h WETF 16.92 e 23.25 d,f 37.42 a,c,d 38.21 e 66.47 a,b,c 94.12 f 103.31 e 83.93 b,c,f 45.87 a 25. 34 a,b,e,f 2 0.3 8 b,f 15.77 g,h WWHT 15.42 a 20.72 a 36.41 b 37.83 b,d,e,f 58.23 d,e 80.27 92.28 c 74.20 e 41.66 b,d 25.23 a,b,c,d,e,f 19.26 a,d,e 14.34 a,b,f 222 Table S 5. 55. Average monthly values of the MOD16A2 500 m dataset for each individual landuse with clusters indicated b y s uperscripts for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. ALFA 15.37 a,b 20.69 a 37.25 a,b,c,d,e 46.42 a 86.85 a 117.15 a 131.25 a,b 109.48 a,b 61.80 a 26.84 a,b 16.47 a 10.29 a,b CORN 16.42 c,d 21.45 b 37.74 a ,b,c 44.17 b 67 .45 95.16 b,c 12 6.77 a 109.36 a,b,c 52.27 b,c 26.30 c,d,e 17.54 11.01 FPEA 15.54 a,b,e 21.91 b,c 37.84 a,b,c,d,e,f 45.85 a,c,d 81.93 a,b,c 114.36 a,d 124.47 a,b,c 95.57 d,e,f,g 56.07 b,c,d,e 27.25 a,b,c,d,e,f 16.15 10.22 a,b,c FRSD 14.95 f 20.47 d 37.17 a,b ,d ,e 45.78 a,c ,d 90.59 127.13 13 9.29 116.28 h 65.94 27.10 a 16.56 a,b 10.30 a,b FRSE 15.44 a,b,e,f 21.85 b,c 35.09 d 48.54 103.15 156.60 160.06 130.24 75.08 28.85 f 14.90 10.06 a,b,c HAY 16.31 c 22.02 c 37.34 a,b,c,d,e 45.93 a,c,e 82.61 b 113.22 d 127.15 a,b,c 105.17 a,b,c, d 55.53 b,d 26. 02 c 17.15 c,d,e 10.37 a,b,c,d PAST 14.90 f 20.54 a,d 37.48 a,b,c,e 45.91 a,c,d 84.84 c 116.82 a,d 131.93 b 107.29 a,c 58.76 e 26.50 b,d,e 16.62 a,b 10.55 c,d SGBT 18.58 23.20 39.14 f 42.29 56.43 82.03 119.47 c 102.12 d,e 47.72 f 26.58 a,b ,c,d,e 18.84 11.48 e SOYB 16.95 21. 91 c 37.81 a,b,c, e 44.41 b,e 66.07 93.67 b 126.20 a 108.58 a,b,c 50.50 g 26.24 b,c,d,e 17.71 11.22 e URLD 16.61 d 22.18 c 36.66 a,d 45.23 c,d,e 74.21 95.75 b,c 104.37 90.22 f 49.74 f,g 26.05 c,d,e 16.76 a,b,c,d 10.50 a,b,c,d UTRN 16.47 c,d 22.08 c 37.30 a,c,e 44 .12 b 72.76 97. 53 c 109.56 94.0 2 g 50.27 g 26.08 c,d 17.21 c,e 10.62 a,b,c,d WETF 15.22 a,b 21.02 e 37.68 b,c,e 45.59 a,c,d 88.12 a 123.71 135.59 113.85 h 61.03 a 26.61 a,b,e 17.04 b,c,e 10.51 b,c,d WWHT 15.69 b,e 21.01 e 37.35 a,b,c,d,e 45.03 a,b,d,e 76.28 104.87 121.46 a,c 98.00 d,e 53 .81 b,c,d 27.21 a,b 16.88 b,d,e 10.64 c,d 223 Table S 5. 56. Average monthly values of the SSEBop dataset for each individual landuse with clusters indicated by superscripts for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. O ct. Nov. De c. ALFA 0.02 a,b 0 .01 a,b,c,d,e 8.33 a,b 28.62 a 54.69 a,b 98.78 a 121.28 a 100.50 a,b,c 55.05 a,b 13.86 a,b 5.54 a,b,c,d 1.11 a CORN 0.03 a,b 0.00 a,b,c,d,e 10.28 a,c,d,e,f 25.78 b 45.47 c 84.79 b,c 114.65 b 99.63 a,b 50.17 c,d,e 10.42 c,d,e 5.72 a,b,c,d 0.36 b FPE A 0.00 c 0.0 0 a, b,c 10.94 a,c,d, e,f 27.72 a,b 56.83 a 104.65 d 125.90 c 102.54 a,c,d 56.59 a,b,f 15.50 a 5.59 a,b,c,d 0.94 c FRSD 0.02 a 0.01 a 10.02 a,c,d,e 28.15 a 54.26 b 99.60 a 122.22 a,c 100.66 a,b,c 54.50 a 14.15 a,b 5.75 a,b,c 1.37 FRSE 0.00 d 0.04 a,b,c,d,e,f 11.95 a,c,d ,f 28.91 a 5 5.7 1 a,b 101.13 a,d, e 123.46 a,c 104.63 c,d 58.59 f 15.38 a,b 5.99 a,b,c,d 2.79 HAY 0.01 a,b,c,d,e 0.05 a,b,c,d,e,f 10.68 a,c,d,e,f 27.35 a,b 52.55 a,b,d 95.37 f 115.52 b,d 98.06 a,b 51.82 c,d 11.54 c,d,f 5.77 a,b,c,d 0.39 b,d PAST 0.01 a 0.01 a,b,c,d,e 10.24 a,c,d ,e 28.13 a 5 5.5 0 a,b 100.73 e 12 2.61 a,c 101.00 a,b,c 55.40 b 14.12 a,b 5.56 a,b,d 1.12 a SGBT 0.34 0.04 a,b,c,d,e,f 10.87 a,c,d,e,f 25.38 a,b 37.31 72.20 108.52 e 98.90 a,b,c 48.34 c,e 9.30 c,e,f 5.98 a,b,c,d 0.08 d,e,f SOYB 0.00 e 0.00 a,b,d,e 10.79 c,e,f 25.5 b 46.39 c 87.1 4 b 115.88 d 100 .56 a,b 51.20 d 1 0.28 c,d,e,f 5.87 a,b,c,d 0.13 e URLD 0.03 a,b 0.00 a,c,d 9.25 a,b,d,e,f 24.62 b 47.24 c 83.60 c 106.69 e,f 89.65 e 46.21 e 9.49 d,e,f 5.32 a,c,d 0.00 f UTRN 0.04 a,b 0.00 a,c,e 9.85 b,c,d,e,f 25.17 b 48.79 86.63 b,c 104.63 f 88.93 e 46.98 e 9.78 d, e,f 5.99 b,c ,d 0.02 f WETF 0.0 4 b 0.06 f 11.24 c,d,e,f 27.57 a 54.16 a,b 99.80 a,e 121.07 a,c 101.85 c,d 55.17 a,b 13.86 b 5.94 a,b,c,d 0.91 c WWHT 0.02 a,e 0.00 b,c,d,e 10.53 a,c,d,e 27.51 a 50.91 d 93.23 f 115.93 b,d 94.20 48.58 e 12.23 c 5.80 a,b,c,d 0.26 b,d 224 Table S 5. 57. Average mo nth ly values of th e NLDAS - 2 Mosaic dataset for each individual landuse with clusters indicated by superscripts for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. ALFA 11.57 a,b 12.87 a,b,c 25.38 a,b 52.79 a 9 0.39 114.05 a 1 31.68 a 114.43 a, b,c,d,e,f 86.39 a,b,c 49.47 a 21.43 a 11.85 a,b,c,d CORN 10.53 c,d,e,f 11.05 d 27.8 a,b,c,d,e 62.77 98.60 a 123.00 b 139.33 b 116.08 a,b,c,d,e,g 83.83 a,b,d 48.08 b 22.18 b,c,d 11.82 a,b,c FPEA 11.60 a,b,g 13.03 a,b,e 25.74 a,b,c 53.16 a 86.80 b 111.96 a,c 128 .81 c 115.22 a,b, c,d,f 86.24 a,b,c 49.41 a,b 21.47 a,b 11.68 a,b,d FRSD 11.60 a,b 13.03 a,b,e 25.57 a,c 53.95 b 91.88 c 115.99 133.13 d 116.03 a,b,c,f 87.69 c 50.70 c 21.65 a,b,c 11.81 a,b,d FRSE 11.61 a,b,c,d,e,g,h 13.26 a,e 25.64 a,b,c,d,e 54.58 c 96.89 a,d 12 4.84 b,d,e 1 44. 84 b,e,f 125.07 94.29 53.12 22.05 c,d 11.80 a,b,d HAY 10.18 c,d,f,h,i 10.33 f,g 27.47 d,e 55.38 a,b,c,d 87.34 b 111.00 c 132.52 a,d 111.70 a,b,d,e,f 78.95 e,f,g 41.56 20.22 11.63 a,b,c,d PAST 11.70 a,g 13.17 e 25.7 a,c,d,e 54.26 c 94.69 119.87 b,f 137.75 b,e, g 118.74 a,b ,f, g 89.60 51.46 2 1.78 a,b,c,d 11.93 a,c,d SGBT 9.94 e,f,h,i 10.10 f 28.29 a,b,c,d,e 66.56 101.82 126.91 d 141.16 b,f 114.66 a,c,d,e,f,g 81.01 d,e,f 46.61 22.26 a,b,c,d 11.66 a,b,c,d SOYB 10.36 c,f,h,i 11.16 h 27.61 a,b,c,d,e 62.12 97.31 d 120.41 f 137.01 g 11 4.34 a,b,c,d ,e, f,g 82.92 d,g 48 .35 a,b 21.82 a,b,d 11.61 b,c,d URLD 10.23 b,c,d,e,f,i 10.78 d,h,i 28.59 d 64.08 97.66 d 122.43 b,f 141.45 b,e,f 117.43 b,c,d,e,g 82.50 d,g 47.09 d 21.94 a,b,d 11.72 a,b,c,d UTRN 10.07 c,d,e,f,i 10.37 g 28.49 a,b,c,d,e 66.04 100.74 125.79 e 14 1.71 e,f 115 .98 b,c,d,e,f,g 81. 78 d,e,g 47.03 d 22.25 a,b,c,d 11.71 a,b,c,d WETF 10.91 c,d,e,f,h,i 12.10 c,i 26.27 a,b,c,e 57.00 d 92.52 c,e 113.17 a,c 128.98 c 110.75 a,b,e,f 83.71 a,d,f,g 49.82 a,c 21.37 a,b,d 11.62 a,b,d WWHT 11.03 c,d,e,h 12.21 b,c,i 26.71 d,e 57.31 d 93. 38 e 119.1 f 137 .87 b,g 119.03 a, b,g 87.49 b,c 49.58 a 21.76 a,b,d 11.65 a,b,c,d 225 Table S 5. 58. Average monthly values of the NLDAS - 2 Noah dataset for each individual landuse with clusters indicated by superscripts for each column Landuse Month Jan. Feb. Mar. Apr. May Ju n. Jul. Aug. Sep. Oct. Nov. Dec. ALFA 12.05 a 14.37 a 21.94 a 29.52 a,b,c,d 37.18 61.85 a 92.72 a 93.47 67.04 a,b,c,d 31.39 a 12.14 a 8.47 a,b CORN 9.17 b 11.62 b,c 18.29 b 28.42 a 47.66 a 80.98 107.08 102.98 a 66.97 a,b,c,d,e 27.11 b 9.89 b 6.83 c,d,e FPEA 12.2 4 a,c,d 14.1 6 a, d 21.12 c,d 29.8 7 b,c,e 39.40 b 65.75 95.74 96.71 b 69.07 a,f 31.76 12.24 a,c 8.48 a,b,f FRSD 11.95 c 14.22 d 21.64 c 29.56 a,b,c,d 38.25 c 64.12 b 94.44 b 94.91 c 67.71 a,b,c,e,g 31.36 a 12.05 a,c 8.35 a FRSE 12.26 a 15.17 24.39 32.34 38.06 b,c,d 60.88 a 92.32 a 94.28 c 69. 22 a ,f,g 33.63 13.1 0 8.76 b HAY 8.78 b,e,f 10.41 14.69 23.99 45.02 78.97 104.95 98.75 b,d,e 63.19 24.76 8.64 6.44 c,d,g PAST 11.85 d 14.17 d 21.73 a 29.71 b,c,e 38.43 d 64.11 b 94.42 b 94.89 c 67.66 a,b,c,e 31.33 a 12.02 c 8.29 f SGBT 8.64 e 11.30 e,f 18.11 b 29 .56 b,d,e 51 .23 85.50 110.76 1 06.04 67.15 b,c,d,e,f,g 26.49 c 9.62 b,d,e 6.65 c,d,e,g SOYB 8.92 f 11.43 b,e,f 17.37 e 27.56 47.86 a 81.78 107.66 103.26 a 66.68 b,c,d,e 26.46 c 9.49 d,e 6.62 c,g URLD 8.64 e 11.49 b,c,e,f 17.67 f 28.52 a,d 49.79 83.50 109.00 104.20 67.50 a,b ,d,e,f,g 26 .43 c 9.51 d 6.58 c,g UTRN 8.64 e 11.41 b,c,f 18.08 b 29.38 c,d,e 50.91 84.95 110.27 105.60 67.49 a,b,d,e,f,g 26.55 c 9.63 b,d,e 6.64 c,e,g WETF 10.24 12.17 17.28 e,f 26.18 42.43 e 74.71 102.39 99.51 d 66.06 c,d 27.43 b 9.91 b,e 7.20 d,e,g WWHT 10.85 13.35 20. 74 d 29.79 b, c,d ,e 42.59 e 70.87 99.49 98.37 e 68.29 a,e,f,g 30.37 11.46 7.82 226 Table S 5. 59. Average monthly values of the NLDAS - 2 VIC dataset for each individual landuse with clusters indicated by superscripts for each column Landuse Month Jan. Feb. Mar. A pr. May Jun . J ul. Aug. Sep. O ct. Nov. Dec. ALFA 8.20 a,b,c 10.09 a,b,c,d,e 10.78 a 15.42 a,b,c,d 41.24 a 78.09 a 109.40 96.95 a,b,c 49.24 a,b 15.82 a,b,c 6.51 a,b,c 7.45 a,b,c,d,e,f,g,h CORN 7.27 a,d,e,f 9.65 a,b,c,d,e,f 10.07 a,b,c,d 15.28 a,b,c,e,f 52.48 96.30 120.98 a 96.71 a,b, d,e ,f 49.97 a,b 16. 34 a,b,d,e 5.65 a,d 6.87 a,b,c,d,e,f,g FPEA 8.12 b,c,d 9.76 a,b,c,d,e,f 10.38 a,b,c,e 15.50 a,b,c,d,e 41.11 a,b 78.43 a 108.4 b,c 95.64 a,b,c,d,e 49.54 a,b 15.98 a,b,c 6.55 b,c 7.71 a,b,c,d,e,f,h FRSD 8.14 a,b,c 9.92 a,b,c,d,e 10.57 a,b,c 15.3 6 a,b,d,e 40 .91 b 77.83 108.71 b 96.31 a,c,d 49.27 a,b 15.83 a,b,c 6.51 b,c 7.49 a,b,c,d,e,g,h FRSE 7.55 a,b,e,f,g 9.56 a,b,c,d,f 10.11 a,b,c,d,e 15.04 a,d,f 39.70 76.21 107.71 c 94.12 a,c,e 47.30 a 15.21 d 6.42 a,b ,c,d,e,f,g 7.03 a,b,c,g,h HAY 7.22 a,c,d,e,f 9.92 a,b,c,e,f 10.13 a,b,c ,e 15.93 b,c,d,e,f 55.46 99.97 b 130.23 d,e,f 108.20 55.36 c 17.98 f 5.81 b,c,d,e 6.95 a,b,d,e,f,g,h PAST 8.02 a,b,d,e 9.84 a,b,c,d,f 10.46 a,b,c,e 15.33 a,b,d,e 40.14 76.71 108.19 c 96.26 a,b,c,d 48.92 a,b 15.65 a,c,e 6.53 b,c 7.39 a,b,c,d,f,g,h SGBT 6.97 a,c, d,e,f,g 9.6 9 a, b,c,d,e,f 10.20 a,b,c,d,e 15.22 a,b,c,e,f 58.12 105.09 125.03 d,e,g,h 94.49 a,b,c,d,e,f 49.51 a,b 16.34 a,b,c,d,e,f 5.04 f 6.55 a,c,d,e,f,g SOYB 7.18 a,d,e,f 9.66 a,b,c,d,e,f 9.78 b,d,e 15.29 a,b,c,e,f 52.70 96.55 121.99 d,f,g 98.52 b,c,d,e,f 51.47 b 16.65 b,c,e 5.61 a ,d, e 6.86 a,b,c,d,e ,f,g URLD 6.74 f,g 9.39 a,b,d,e,f 9.39 c,d,e 15.3 a,b,c,e,f 54.90 99.87 b 123.68 d,f,h 95.91 a,b,c,d,e,f 49.36 a,b 16.33 a,b,c,d,e 5.54 a,b,d,e 6.55 a,c ,d,e,f,g,h UTRN 6.86 a,d,e,f,g 9.55 a,b,c,d,e,f 9.84 a,b,c,d,e 15.22 a,b,c,e,f 56.93 103 .21 124.31 e ,f, g,h 94.47 a,b,c, d,e,f 49.18 a,b 16.27 a,b,c,d,e,f 5.22 g 6.53 c,d,e,f,g WETF 7.84 b,c,d 9.90 a,c,d,e,f 10.11 a,b,c,e 15.82 b,c,d 47.97 88.48 117.67 a 101.79 a,c,f 53.98 c 17.36 f 6.16 b,c,e 7.39 b,d,e,f,g,h WWHT 7.43 a,e,f,g 9.54 b,c,d,e,f 9.96 b,c,e 15.21 a, c,e,f 45.93 86 .00 114.75 96.4 5 a,b,c,d,e 49.25 a,b 16.01 a,b,e 6.12 a,c,d,e 7.09 a,b,c,e,f,g,h 227 Table S 5. 60. Average monthly values of the TerraClimate dataset for each individual landuse with clusters indicated by superscripts for each column Landuse Month Jan. Feb. Mar . Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. ALFA * * 16.02 a 81.14 a,b,c 101.36 a,b,c,d 110.39 a,b,c,d,e,f 98.84 a 88.44 a,b,c 65.53 a,b,c 49.20 a,b 21.55 1.13 a CORN * * 18.34 81.91 a,b,c,d,e 101.81 a,b,e 110.91 a,b,c,d,e,g 97.71 b 86.55 a,d,e,f 6 5.36 a,b,d,e 49 .61 a,c,d 22.62 1.39 a FPEA * * 19.68 b 82.49 f,g,h 101.42 a,c,d,e 110.38 a,b,c,d,g 98.14 a,b 87.80 b,c 64.78 c,d 49.55 c,d 22.89 a,b 1.49 a FRSD * * 16.50 c 81.41 a,d 101.45 a,b,c,d 110.58 a,b,c,d,e,f,g 98.7 a 88.50 b 65.30 a,c,e 49.22 a 21.96 1.20 a FRSE * * 17.3 2 a 80.58 i 100.79 c 110.41 a,b,c,d,e,f,g 98.79 a,b,c 88.50 a,b,c,d 65.35 a,b,c,d,e 48.83 21.48 1.09 a HAY * * 19.93 d,e 82.84 f,g 102.31 b,f 110.16 a,b,c,d,e,g 96.42 b,c,d 86.50 a,c,d,e,f 65.21 a,b,c,d,e 49.87 c,e,f 23.19 c 1.62 a PAST * * 16.49 c 81.27 a,b, c 101.24 b,c ,d, e 110.60 a,b,c,e ,f,g 98.68 a 88.20 a,b,c 65.18 a,c,d 49.14 b 21.86 1.21 a SGBT * * 19.09 f 81.71 a,b,c,d,e,i 102.04 a,b,d,e,f 111.65 d,f,g 97.20 b,c,d 85.19 65.47 a,b,c,d,e 49.76 a,c,d,e 22.81 a 1.51 a SOYB * * 19.11 f 82.61 e,f,h 101.94 a,b,e 110.81 a,b, d,e,g 97.24 c,d 86.27 a,d,e 65. 13 a,b,c,d,e 49.85 e 23.04 b 1.56 a URLD * * 19.93 d 82.88 f,g 102.35 f 110.28 a,c,d,e,g 96.94 c,d 86.72 a,c,e,f 65.26 a,b,c,d,e 50.15 f 23.17 c 1.71 a UTRN * * 20.10 e 82.80 c,d,e,f,g,h 102.16 a,b,e,f 110.25 a,c,d,e,g 96.89 c,d 86.41 d,e,f 65.39 a,b,c, d,e 50.13 f 23.20 c 1.70 a WETF * * 19.00 f 82.51 e,f,h 101.44 a,c,d,e 110.33 a,c,d,e,g 96.77 d 86.86 a,d,f 64.95 b,c,d,e 49.65 c,d 22.99 b 1.61 a WWHT * * 17.82 b 81.99 a,e,h 101.83 b,f 110.61 b,c,d,e,g 97.82 b 87.96 b,c 65.25 a,b,d,e 49.51 c,d 22.35 1.33 a * Note that n o E Ta values were provided for TerraClimate for the months of January and February. 228 Table S 5. 61. Average monthly values of the ALEXI dataset for each individual landuse with clusters indicated by superscripts for each column Landuse Month Ja n. Feb. Mar . A pr. May Jun. Ju l. Aug. Sep. Oct. Nov. Dec. ALFA 22.26 a,b,c,d 35.97 a,b 50.13 a,b 56.55 a,b,c 85.6 a,b,c 109.28 a,b,c 124.86 a,b,c,d 99.99 a,b 67.65 a,b 33.53 a,b 18.97 a,b,c 16.03 a,b,c,d CORN 23.41 a,b,c,d,e 38.52 c,d,e 52.37 c 57.76 a 81.73 d 101.84 d 122 .91 a,b,c 10 1.5 2 a 67.00 a,b,c 3 2.55 a,b,c 19.84 a,b,d,e 16.39 a,b,c,d FPEA 23.09 a,b,c,d,e 36.70 a,b,c 50.07 a,b 56.59 a,b,c 87.85 a 109.90 a,b,c,e 125.50 a,b,c,d 101.39 a,b,c 67.29 a,b,c,d,e 32.52 a,b,c 19.08 a,b,c,d 15.91 a,b,c FRSD 22.10 a,b ,c,d,e 35.72 a 49.88 a 55.66 b 85.29 b,c 10 8.6 0 a,b,c,f 125.04 a,b,d 99.65 a,b 67.11 a,b,d 33.04 a,b 19.12 a,b,c 15.99 a,b,c FRSE 23.53 a,b,c,d,e 36.53 a,b,d 49.26 a,b 53.96 d 84.18 b,c,d 107.98 a,b,c,e,f,g 123.90 a,b,c,d 95.25 b,d 65.64 c,d,e 32.17 a,c 18.32 a,c 15.91 a,b ,c,d,e HAY 24.24 a,e 39.32 c,d,e 5 3.37 c,d 59. 79 84.36 b,c 104.24 g 123.50 a,b,c,d 103.80 c 67.76 a,b,c 33.30 a,b 20.71 e 16.81 a,b,c,d PAST 22.60 a,b,c,d,e 35.93 a,b 50.29 a,b,e 56.26 b,c 85.70 b 107.9 a,b,c,e,f,g 124.16 a,c,d 99.37 a,b 67.40 a,b,d,e 33.15 a,b 19.22 a,b,c 16.06 a,b,c,d SGBT 24.20 a,b,c,e 40. 62 e 54.40 d 58. 87 78.85 e 96.28 h 121.32 a,b,c,d 102.37 a,c 67.45 a,b,c,e 32.63 a,b,c 20.23 d,e 16.95 a,b,d SOYB 23.70 a,b,d,e 38.55 c,d 51.89 e 57.48 a,c 81.03 d 100.68 i 122.28 a,c 101.41 a 66.55 a,c,d,e 32.04 c 19.81 a,b,d,e 16.32 a,b ,c,d URLD 21.85 c,d 35.83 a,b 46.70 52.0 2 d 77.22 e 9 5.4 8 h 111.87 93.05 d 60.51 28.39 17.22 c 14.56 e UTRN 23.64 a,b,c,d,e 38.68 c,d,e 51.38 b,e 57.88 a 83.27 b,c,d 100.33 d,i 117.43 96.80 b,d 65.53 a,c,d,e 32.10 b,c 19.52 a,b,d 16.13 a,c,d WETF 22.73 b,c,d,e 36.62 b 49.93 a,b 55.87 b,c 84.05 c 106.14 b,e,f 124.14 a ,b,c,d 99.6 0 b 66.29 b,c,d,e 32 .18 c 19.14 a,b,c 15.93 a,c WWHT 23.95 a,b,c,d,e 37.88 c,d,e 51.36 e 57.09 a,c 84.43 b,c 106.75 c,e,f 124.53 b,c,d 99.22 b 66.17 c,d,e 33.15 a,b 19.89 b,d,e 16.52 b,c,d 229 Table S 5. 62. Average monthly values of the SWAT model dataset for ea ch individu al landuse with cl usters indicated by superscripts for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. ALFA 3.07 a 4.56 a 25.85 a 38.03 a,b 55.19 a 73.62 a 96.64 a 92.51 a 72.84 a,b 29.44 a,b,c,d 14.77 a 5.39 CORN 3 .73 b,c 5.42 a,b ,c,d,e,f 28.61 b ,c,d,e,f 42.65 63.62 b,c,d 115.34 b 108.57 71.70 b,c,d,e,f 42.54 c,d,e 25.83 a,b,c 16.63 b 6.92 a FPEA 3.70 b,c,d 5.50 b,c,d,e,f 28.43 a,b,c,d,e,f,g,h 37.98 a 54.03 e 70.65 c 96.30 a 76.34 b,c,d 33.92 c,d 27.52 a,b 15.96 b,c 6.55 b FRSD 3.53 d,e 5.32 a,b,c 28. 90 b,c,d,e,g 40. 12 c 61.12 b,c,f,g 84.92 d 68.42 b 65.47 e,g 65.84 a 42.02 e,f 23.04 d 8.65 c FRSE 3.35 a 5.22 a,b 27.95 a,b,c,d,f,h 36.22 b 59.88 b,d,f,g 96.06 e 72.68 c,d 64.08 e,f,g 45.68 c 28.85 d 16.07 b,c 6.36 d HAY 3.70 b,c,d 5.60 d,e,f 28.37 a,b,c,d,f,h 37. 96 a 54.45 a, e 7 3.88 a,c 69.19 b, e 64.10 e,f,g 56.22 e 38.93 g 20.64 e 8.00 PAST 3.62 b,d,e 5.45 b,c,d 30.27 b,g,i 43.52 62.09 b,c,f 77.28 f 66.53 e 62.48 f,g 54.76 e 36.79 g 20.86 e 7.50 SGBT 3.73 b,c 5.56 d,e,f 29.95 e,g,i 45.13 d 67.94 h,i 96.74 e 86.46 79.03 b 54.52 c,e 30.62 d 18.09 6.8 5 a SOYB 3.71 b,c,e 5.51 c,d,e,f 28.36 a,b,c,d,e,f 41.82 60.15 b,c,g 91.57 b,d,e,f,g 97.47 a 83.41 a,b,c 69.58 a,b,e 21.39 16.53 b,c 6.81 a,b URLD 3.56 c,d,e 5.30 a,b,c 29.05 b,c,d,e,f,h,i 45.84 d 66.81 h 90.46 g 77.07 c 72.28 c 66.22 a 41.10 e 22.90 d 8.18 UTRN 3 .34 a,d,e 5. 09 a ,b,c,d,e 26.87 a ,d,f,h 36.87 a,b 46.94 58.63 43.27 31.05 24.32 24.24 a,c 14.65 a 6.44 a,b,d WETF 4.05 f 5.95 32.01 47.95 68.48 i 90.46 g 73.93 d 69.21 d 69.37 b 42.51 f 23.53 8.65 c WWHT 4.01 f 5.45 a,b,c,d,e,f 27.61 a,b,f,h 39.83 c 62.54 c,d,f,g 108.06 b 100 .05 a 58.82 e ,f, g 30.89 d 26.96 a ,b 16.06 c 6.72 a 230 Table S 5. 63. Average monthly values of the Ensemble dataset for each individual landuse with clusters indicated by superscripts for each column Landuse Month Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. AL FA 12.03 a,b,c 1 6.24 a,b,c,d,e,f,g 25.93 a,b,c,d 43.71 a,b,c,d 70.38 a 97.19 a 113.63 a 97.98 a,b 62.31 a 30.63 a,b,c 15.19 a,b,c 9.47 a,b ,c,d,e CORN 11.67 d 15.99 a,b,c 26.61 a,b,e 44.19 e,f,g 68.54 95.92 b 115.91 b 99.59 a,b,c 59.70 b,c 29.46 d 15.51 a,b,d 9.31 a,b,c,d,f,g F PEA 12.24 a,b 16 .56 a,d,e,f 26.77 a,c,e,f 43.68 a,b,c,d,e 69.79 a,b 98.26 a,c 113.04 a,c 95.97 d 61.60 a,b 30.92 a,b 15.47 a,b,d 9.53 a,b,c,d,e,f FRSD 12.06 a,b,c 16.40 a,d,e,f,g 26.21 a,c,d,e 43.58 a,b 71.24 a,c,d 99.88 d 115.75 b 99.56 a,c 63.09 30.84 a 15.41 a ,b,d 9.59 a, b,c ,f FRSE 12.28 a 16.93 d 26.55 a,b,c,d,e,f 44.52 c,e,f,g 75.40 107.79 122.40 103.72 66.40 31.67 15.07 a,c 9.62 a,b ,c,d,e,f,g HAY 11.83 b,c,d 16.26 a,d,e,f,g 26.46 a,b,c,e 43.68 a,b,c,d,f 72.11 c 101.04 d,e 116.59 b 99.51 a,b,c 60.40 a,b,c 28.82 e,f 15.33 a,c ,d 9.42 a,b, d,f ,g PAST 12.05 a ,b,c 16.31 a,b,c,d,e,g 26.29 a,c,d,e 43.71 a,c,d 70.74 a,c,e 98.33 c 114.72 d 98.26 a,b 62.24 a 30.81 a 15.38 a,b,d 9.54 a,b,c,e,f SGBT 11.97 a,b,c 16.41 d,e,f,g 27.41 f 44.54 c,e,f,g 66.51 93.18 114.86 a,b,d 98.43 a,b,c,d 58.43 d 29.20 d,e,f 15 .87 9.40 a,b ,c, d,f,g SOYB 11. 72 c 16.04 a,b,c,f,g 26.49 a,b,c,e 44.03 c,d,f,g 68.23 95.68 b 116.02 b,d 99.82 a,b,c 59.51 c 29.40 d,e 15.52 b,d 9.28 a,c,d,f,g URLD 11.37 15.88 b,c 25.68 b,d,e 43.85 a,b,c,d,e,g 70.23 a,d,e 95.59 b 109.37 93.09 57.65 28.62 f 14.91 c 8.90 UTR N 11.60 d 16 .14 a,c,e,f,g 26.57 a,b,c,e 44.75 f,g 71.63 c,d,e 98.06 a,c 111.29 c 94.24 d 58.37 d 29.12 e 15.40 a,b ,c,d 9.15 d,e,f,g WETF 11.87 b,c 16.27 a,d,e,f,g 26.20 b,c,e 43.59 a,b,d 72.15 c 101.31 e 116.24 b 99.77 a,c 62.13 a 30.28 c 15.37 a,c,d 9.48 a,b,c,d,f WWHT 11.94 a, b,c 16.21 a, d,e ,f,g 26.45 a,c,e 43.97 c,d,e,f,g 69.20 b 96.46 b 113.02 a 95.93 d 60.06 b,c 30.41 b,c 15.44 a,b ,d 9.32 b,c,e,g 231 Table S 5. 64. Overall summary of average ETa values for each dataset for each individual landuse with clusters indicated by superscripts f or each col umn Dataset Landus e ALFA CORN FPEA FRSD FRSE HAY PAST SGBT SOYB URLD UTRN WETF WWHT MOD16 1km 45.80 a 43.62 a 44.98 a 47.81 a 55.65 a 47.75 a,b 46.12 a,b 42.33 a 43.85 a 41.24 a,b 42.01 a 47.58 a,b 42.99 a,b MOD16 500m 56.66 b 52.14 b 53.93 b 59.30 b,c 66.65 b 54.90 c 56 .01 c 48.99 b 51.77 b 49.02 c 49.83 b 58.00 c,d 52.35 SSEBop 40.65 a,c 37.27 42.26 a,c 40.89 d 42.38 c 39.09 d 41.20 a,b 34.77 37.81 35.17 a 35.57 c 40.97 a 38.27 a,b NLDAS: Mosaic 60.19 d 62.92 c 59.59 d 61.09 b 64.83 b 58.19 c,e 62.55 d 63.42 c 62.08 c 62.99 63.50 d 59.85 c 62.2 6 c NLDAS: Noah 40 .18 a,c 43.08 a 41.38 a,c 40.71 d 41.20 c 40.71 a,d 40.72 a 44.25 a 42.92 a 43.57 b 44.13 a 41.29 a 42.00 a NLDAS: VIC 37.43 c 40.63 a 37.26 c 37.24 e 36.33 d 43.60 a,b,d 36.95 e 41.85 a 41.02 a 41.08 b 41.47 a 40.37 a 38.65 b TerraClimate 66.49 b,d 66 .73 b,c 67.3 9 b, d 66.61 c 66.87 a ,b,e 66.90 c,e 66.51 c,d 66.74 d 66.85 b,c 67.03 66.99 e 66.70 d 66.76 d ALEXI 60.07 b,d 59.65 b,c 60.49 d 59.77 b,c 58.89 b 60.93 e 59.84 c,d 59.51 c,d 59.31 b,c 54.56 58.56 d,e 59.39 c,d 60.08 c,d SWAT 42.66 a,c 44.30 a 38.07 a,c 41.45 d,e 38.53 c ,d 38.42 a,d 39 .26 a,e 43.72 a 4 3.86 a 44.06 b 26.81 c 44.68 a,b 40.58 a,b Ensemble 49.56 49.37 49.49 50.30 a 52.70 e 50.12 b 49.87 b 48.85 b 49.31 47.93 c 48.86 b 50.39 b 49.03 232 Figure S 5. 1. Maps showing regions of statistical difference and no difference between ea ch ETa data set and the SWAT m odel output. Maps correspond to a) MOD16A2 1 km, b) MOD16A2 500 m, c) SSEBop, d) NLDAS - 2:Mosaic, e) NLDAS - 2:Noah, f) NLDAS - 2:VIC, g) TerraClimate, and h) ALEXI 233 Figure S 5. 2. Maps showing regions of statistical difference and no differenc e between each ET a dataset and the Ensemble. Maps correspond to a) MOD16A2 1 km, b) MOD16A2 500 m, c) SSEBop, d) NLDAS - 2:Mosaic, e) NLDAS - 2:Noah, f) NLDAS - 2:VIC, g) TerraClimate, h) ALEXI, and i) SWAT model 234 Table S6.1. A summary of the remot e sensing ETa products used i n this study ETa Product Base Equation Resolution Accuracy (mm/day) Reference SSEBop Simplified Surface Energy Balance Monthly 1.0 km 2 for the Contiguous United States 0. 89 6 (RMSE) (Velpuri et al., 2013) ALEXI Surface Ener gy Balance Da i ly 4.0 km 2 for the Contiguous United States 1.00 (RMSE) (Cammalleri et al., 2014) MOD16A2 1 km Penman - Monteith 8 - day 1.0 km 2 for the entire globe 0.857 (RMSE) (Mu et al., 2011) MOD16A2 500m Penman - Monteith 8 - day 0.5 km 2 for the entire gl obe 0.857 (RM S E) (Mu et al., 2011) NLDAS - 2: Mosaic Mosaic Land Surface Model Hourly/Monthly 12.0 km 2 for North America 0.341 (RMSD) (Long et al., 2014) NLDAS - 2: Noah Noah Land Surface Model Hourly/Monthly 12.0 km 2 for North America 0.120 (RMSD) (Long e t al., 2014) NLDAS - 2: VIC V ariable Infiltration Capacity Land Surface Model Hourly/Monthly 12.0 km 2 for North America 0.219 (RMSD) (Long et al., 2014) TerraClimate One - dimensional Modified Thornthwaite - Mather Water Balance Monthly 4.0 km 2 for the entir e globe 0.156 ( MAE) (Abatzogl ou et al., 2018) *RMSE : Root Mean Squared Error; RMSD: root - mean - square deviation; MAE : mean absolute error 235 REFERENCES 236 REFERENCES Abatzoglou, J. T., Dobrowski , S. Z., Parks, S. A., & Hegewisch, K. C. (2018) . TerraClimat e, a high - resolution global d ataset of month ly climate and climatic water balance from 1958 2015. Scientific data, 5, 170191. Abbaspour, K. C. (2007). User manual for SWAT - CUP, SWAT calibration a nd uncertainty analysis programs. Swiss Federal Institute of Aquatic Science and Technolog y, Eawag, Duebe ndorf, Switzerland. Abouali M., (2017) SWATUtilities. (accessed 4.26.17). Ameri Flux. (2018). AmeriFlux. http://ameriflux.lbl.go v/ (Last acce ss: 9/22/18) Anderson, M. C., Allen, R. G., Morse, A., & Kustas, W. P. (2012). Use of Landsat thermal imagery in monitoring evapotranspiration and managing water resources. Remote Sensing of Envi ronment, 122, 50 - 65. Anderson, M. C., Norman, J. M., Mecikals ki, J. R., Otkin, J. A., & Ku stas, W. P. (20 07). A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 1. Model formulation. Journal of Geophysical Research: Atmo spheres, 112( D10). Anderson, M.C., Norman, J.M., Diak, G. R., Kustas, W.P., Mecikalski, J.R. ( 1997 ) . A two - source time - integrated model for estimating surface fluxes using thermal infrared remote sensi ng. Remote Sens. Environ. 60, 195 216. Anderson, M.C., Norman, J.M., Mecikalski, J.R., Otkin, J.A ., Kustas, W.P. ( 2007 ) . A climatological study of evapotranspiration and moisture stress across the continental United States based on thermal remote sensing: 1. Model formulation. J. Geophys. Res. 112, D10117. Arn old, J.G., Mo riasi, D.N., Gassman, P.W., A bbaspour, K.C., White, M.J., Srinivasan, R., Santhi, C., Harmel, R.D., Van Griensven, A., Van Liew, M.W. and Kannan, N. (2012). SWAT: Model use, calibration, and validation. Transactions of the ASABE, 55(4), pp.149 1 - 1508. Baska r, S., Tamilselvi, S. Varshin i, P.R., 2015. MATLAB code for Constrained NSGA II - Dr.S.Baskar, S. Tamilselvi and P.R.Varshini - File Exchange - MATLAB Central. ( accessed 4.26.17). Bekele, E.G., Nicklow, J.W. ( 2007 ) . Multi - objective automatic calibration of SWAT using NSGA - II. J. Hydrol. 341, 165 176. 237 B enz, U. C., Hofmann, P., Willhauck, G., Lingenfelder, I. , & Heynen, M . (2004). Multi - resolution, o bject - oriented fuzzy analysis of remote sensing data for GIS - ready information. ISPRS Journal of photogrammetry and remote sensing, 58(3), 239 - 258. Beven, K. (1996). 12 Equifinality and Uncertainty in Geomorphologi cal Modelling . In The Scientific Nature of Geomorphology: Proceedings of the 27th Binghamton Symposium in Geomorphology, Held 27 - 29 September, 1996 (Vol. 27, p. 289). John Wiley & Sons. Beven, K., Smi th, P. ( 2014 ) . Concepts of information content and likel ihood in para meter calibration for hydrolo gical simulatio n models. J. Hydrol. Eng. 20, A4014010. Bhattarai, N., Shaw, S. B., Quackenbush, L. J., Im, J., & Niraula, R. (2016). Evaluating five remote se nsing based single - source surface energy balance models for estimatin g daily evapotranspiration in a humid subtro pical climate. International Journal of Applied Earth Observation and Geoinformation, 49, 75 - 86. Bishop, B. (2010). Weather Data from Enviro - we ather Program Retrieved from Michigan Dairy Review: http s://msu.edu/~ mdr/vol15no3/weather.html Bri gode, P., Oudin , L., & Perrin, C. (2013). Hydrological model parameter instability: A source of additional uncertainty in estimating the hydrological impacts of climate change?. Journal of Hydrology, 476, 410 - 425. Brown, J. L. (2014). SDM toolbox: a python based GIS tool kit for landscape genetic, biogeographic and species distribution model analyses. Methods in Ecology and Evolution , 5 (7), 694 - 700. Cammalleri, C., Anderson, M. C., Gao, F., Hain, C. R., & Kustas, W. P. (2014). Mapping daily evapotranspiratio n at field scal es over rainfed and irrigated agricultural areas using remote sensing data fusion. Agricultural and forest meteorology, 186, 1 - 11. Chankong, V. , Haimes, Y.Y. ( 1993 ) . Multi - objective optimization: Pareto optimality. Concise Encyclopedia of En vironmental Sys tems. Pergamon Press, UK 387 396. Chen, H., Xu, C. - Y., Guo, S. ( 2012 ) . Comparison and evaluation of multiple GCMs, statistical downscaling and hydrological models in the study of climate change impacts on runoff. J. Hydrol. 434, 36 45. Chris tensen, N., & L ettenmaier, D. P. (2006). A multimodel ensemble approach to assessment of climate change impacts on the hydrology and water resources of the Colorado River Basin. Hydrology and Earth System Sciences Discussions, 3(6), 3727 - 3770. Clark, M. P. , Fan, Y., Lawr ence, D. M., Adam, J. C., Bolster, D., Gochis, D. J., Hooper, R. P., Kumar, M., Leung, L. R., Mackay, D. S., & Maxwell, R. M. (2015). Improving the representation of hydrologic processes in Earth System Models. Water Resources Research, 51(8 ), 5929 - 5956. 238 C onfesor, R.B., Whittaker, G.W. ( 2007 ) . Automatic Calibration of Hydrologic Models with Multi - Objective Evolutionary Algorithm and Pareto Optimization1. JAWRA J. Am. Water Resour. As. 43, 981 989. Conn, A., Gould, N., Toint, P. ( 1997 ) . A glob ally convergent Lagrangian barrier algorithm for opti mization with general inequality constraints and simple bounds. Math. Comput. Am. Math. Soc. 66, 261 288. Conn, A.R., Gould, N.I., Toint, P. ( 1991 ) . A globally convergent augmented Lagrangian algorithm f or optimization with general constraints and simple bounds. SIAM J. Numer. Anal. 28, 545 572. Crow, W. T., Wood, E. F., & Pan, M. (2003). Multiobjective calibration of land surface model evapotranspiration predictions using streamflow observati ons and spac eborne surface radiometric temperature retrievals. Journal of Geophysical Research: Atmospheres, 108(D23). Cuceloglu, G., Abbaspour, K. C., & Ozturk, I. (2017). Assessing the Water - Resources Potential of Istanbul by Using a Soil and Water Asses sment Tool ( SWAT) Hydrologi cal Model. Water, 9(10), 814. Deb, K. ( 2001 ) . Multi - objective optimization using evolutionary algorithms, 2001. Chicheter, John - Wiley. Das, I., & Dennis, J. E. (1998). Normal - Boundary Intersection: A New Method for Generating the Pareto Surf ace in Nonlinea r Multicriteria Optimization Problems, 8(3), 631 657. Deb, K., & Jain, H. (2014). An Evolutionary Many - Objective Optimization Algorithm Using Reference - Point - Based Nondominated Sorting Approach, Part I: Solving Problems With Box Constraints. IEEE Transacti ons on Evolutionary Computation , 18(4), 577 601. Deb, K., Pratap, A., Agarwal, S., Meyarivan, T. ( 2002 ) . A fast and elitist multiobjective genetic algorithm: NSGA - II. IEEE T. Evolut. Comput. 6, 182 197. Deser, C., Walsh, J. E., & Timlin, M. S. (2000). Arc tic sea ice variability in the context of recent atmospheric circulation trends. Journal of Climate, 13(3), 617 - 633. (20) Dietterich, T. G. (2000). Ensemble methods in machine learning. In International workshop on multiple class ifier system s (pp. 1 - 15). S pringer, Berlin, Heidelberg. Duan, Q., Ajami, N. K., Gao, X., & Sorooshian, S. (2007). Multi - model ensemble hydrologic prediction using Bayesian model averaging. Advances in Water Resources, 30(5), 1371 - 1386. Dubus, I. G., Brown, C. D., & Be ulke, S. (2003) . Sources of uncertainty in pesticide fate modelling. Science of the total environment, 317(1), 53 - 72. 239 Einheuser, M. D., Nejadhashemi, A. P., & Woznicki, S. A. (2013). Simulating stream health sensitivity to landscape changes due to bioenerg y crops expansi on. Biomass and bioenergy , 58, 198 - 209. Enviroweather. (2018). Enviroweather: Weather - based pest, natural resources, and production management tools. https://www.enviroweather.msu.edu/ (Last access: 9/22/18) EPA, ( 2016 ) . About Saginaw River and Bay AOC. < https://www.epa.gov/saginaw - river - bay - aoc/about - saginaw - river - and - bay - aoc> (accessed 4.26.17). EPA, ( 2017 ) . Saginaw River and Bay Area of Concern. (accessed 4.26.17). Erwin, M. L. and Hamilton, P. A. (2005). Moni toring Our Rivers and Streams. USGS. ESA. (2017a). Changing Lands. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/Copernicus/Sentinel - 1/Changing_lands ESA. (2017b). Copernicus Overview. Retri eved from ESA: http://w ww.esa.int/Our_ Activities/Observing_the_Earth/Copernicus/Overview4 ESA. (2017c). Cryosat Facts and Figures. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/CryoSat/Facts_and_figures ESA. (2017d). Cryosat Sat ellite. Retrieved from ESA: http://m.e sa.int/Our_Activities/Observing_the_Earth/CryoSat/Satellite ESA. (2017e). Data Products. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/GOCE/Data_products ESA. (2017f). Emergency Response. Re trieved from ESA: http: //www.esa.int/O ur_Activities/Observing_the_Earth/Copernicus/Sentinel - 1/Emergency_response ESA. (2017g). Envisat Instruments Retrieved from ESA: https://earth.esa.int/web/guest/missions/esa - operational - eo - missions/envisat/instruments ESA. (2017h). Envisat M ission Summary Retrieved from ESA: https://earth.esa.int/web/guest/missions/esa - eo - missions/envisat/mission - summary ESA. (2017i). Envisat Objectives. Retrieved from ESA: https://earth.esa.int/web/guest/missions/esa - operational - eo - mis sions/envisat/objective s ESA. (2017j). http://www.esa.int/Our_Activities/Observing_the_Earth/GOCE/Instruments 240 ESA. (2017k). Introducing Cryosat. Retrieved from ESA: http://m.esa.int/Our_Activities/Observing_the_Earth/Cry oSat/Introducing_CryoSa t ESA. (2017l). Introducing GOCE. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/GOCE/Introducing_GOCE ESA. (2017m). Introducing Sentinel - 1. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_t he_Earth/Copernicus/Sen tinel - 1/Introdu cing_Sentinel - 1 ESA. (2017n). Introducing Sentinel - 2. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/Copernicus/Sentinel - 2/Introducing_Sentinel - 2 ESA. (2017o). Introducing Sentinel - 3. Retriev ed from ESA: http://www .esa.int/Our_Ac tivities/Observing_the_Earth/Copernicus/Sentinel - 3/Introducing_Sentinel - 3 ESA. (2017p). Introducing SMOS. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/SMOS/Introducing_SMOS ESA. (2017q). Oc eans and Ice. Retrieved from ESA: http ://www.esa.int/Our_Activities/Observing_the_Earth/Copernicus/Sentinel - 1/Oceans_and_ice ESA. (2017r). Sentinel - 4 and - 5. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/Copernicus/Sentinel - 4_an d_ - 5 ESA. (2017s). Sent inel - 6. Retriev ed from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/Copernicus/Sentinel - 6 ESA. (2017t). SMOS Facts and Figures. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/SMOS/Facts_and_fi gures ESA. (2017u). SMO S Novel Technol ogy. Retrieved from ESA: http://www.esa.int/Our_Activities/Observing_the_Earth/SMOS/Novel_technology ESA. (2017v). What is Envisat? Retrieved from ESA: https://earth.esa.int/web/guest/missions/esa - operational - eo - missio ns/envisat Esfahanian, E., Nejadhashem i, A.P., Abouali, M., Adhikari, U., Zhang, Z., Daneshvar, F., Herman, M.R. ( 2017 ) . Development and evaluation of a comprehensive drought index. J. Environ. Manage. 185, 31 43. EUMETSAT. (2017a). METEOSAT Design. Retrieved from EUMETSAT: http s://www.eumetsa t.int/website/home/Satellites/CurrentSatellites/Meteosat/MeteosatDe sign/index.html 241 EUMETSAT. (2017b). METEOSAT. Retrieved from EUMETSAT: https://www.eumetsat.int/website/home/Satellites/CurrentSatellites/Meteosat /index.html EUMETSAT. (2017c) . METOP Design. Retrieved from EUMETSAT: https://www.eumetsat.int/website/home/Satellites/CurrentSatellites/Metop/MetopDesign/i ndex.html EUMETSAT. (2017d). METOP Services. Retrieved from EUMETSAT: https://www.eumetsat.int/websi te/home/Satellites/CurrentSat ellites/Metop/M etopServices /index.html EUMETSAT. (2017e). METOP. Retrieved from EUMETSAT: https://www.eumetsat.int/website/home/Satellites/CurrentSatellites/Metop/index.html FAA. (2018). Surface Weather Observation Stations: AS OS/AWOS. https://www.faa.gov/ air_traffic/wea ther/asos/?state=MI (Last access: 9/22/18) Fowler, H. J., & Ekström, M. (2009). Multi model ensemble estimates of climate change impacts on UK seasonal precipitation extremes. International Journal of Climatology , 29(3), 385 - 416. Fox, J., & Monette, G. (20 02). An R and S - Plus companion to applied r egression. Sage. Franco, A. C. L., & Bonumá, N. B. (2017). Multi - variable SWAT model calibration with remotely sensed evapotranspiration and observed flow. RBRH, 22. Fri edl, M. A., Mcgwire, K. C., & Mciver, D. K. (2001). An overview of uncertainty in optic al remotely sensed data for ecological applications. In Spatial Uncertainty in Ecology (pp. 258 - 283). Springer New York. Gassman, P. W., Reyes, M. R., Green, C. H., & Ar nold, J. G. (2007). The soil and water asses sment tool: historical development, applica tions, and future research directions. Transactions of the ASABE, 50(4), 1211 - 1250. Giorgi, F., & Mearns, L. O. (2003). Probability of regional climate change based on the Reliability Ensemble Avera ging (REA) meth od. Geophysical research letters, 30(12). G iri, S., Nejadhashemi, A. P., & Woznicki, S. A. (2012). Evaluation of targeting methods for implementation of best management practices in the Saginaw River Watershed. Journal of environmental manag ement, 103, 24 - 40. Giri, S., Nejadhashemi, A.P., Woznicki, S.A. ( 2016 ) in a framework for bioenergy development. Land Use Policy 59, 143 153. Giri, S., Nejadhashemi, A.P., Zhang, Z., Woznicki, S.A. ( 2015 ) . Integr ating statistic al and hydrological models to identify impl ementation sites for agricultural conservation practices. Environ. Model. Softw. 72, 327 340. 242 Gleick, P. H. (1993). Water in crisis: a guide to the worlds fresh water r esources. Goldberg, D.E. ( 1989 ) . Genetic algo rithms in search, optimization, and machine learning, addison - wesley, reading, ma, 1989. Reading, MA. Golmohammadi, G., Prasher, S., Madani, A., & Rudra, R. (2014). Evaluating three hydrological distributed wate rshed models: MIKE - SHE, APEX, SWAT. Hydrology , 1(1), 20 - 39. Google. (2018). EEFlux. http s://eeflux - level1.appspot.com/ (Last access: 9/22/18) Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R. (2017). Google Earth Engine: Planetary - scale geospatial analysis for everyone. Remo te Sens. Environ. 202, 18 27. Graham, S. ( 1999 ) . Remote Sensing : Feature Articles. (accessed 4.26.17). Guerrero, J. - L., Westerberg, I.K., Halldin, S. , Lundin, L. - C., Xu, C. - Y. ( 20 13 ) . Exploring the hydrological robustness of model - parame ter values with alpha shapes. Water Resour. Res. 49, 6700 6715. Gupta, H. V., Sorooshian, S., & Yapo, P. O. (1999). Status of automatic calibration for hydrologic model s: Comparison with multilevel expert calibrat ion. Journal of Hydrologic Engineering , 4(2 ), 135 - 143. Haddeland, I., Heinke, J., Biemans, H., Eisner, S., Flörke, M., Hanasaki, N., Konzmann, M., Ludwig, F., Masaki, Y., Schewe, J. and Stacke, T. ( 2014 ) . Global water resources affected by h uman interventi ons and climate change. Proceedings of the National Academy of Sciences , 111(9), pp.3251 - 3256. Hain, C. R., Crow, W. T., Anderson, M. C., & Yilmaz, M. T. (2015). Diagnosing Neglected Soil Moisture Source Sink Processes via a Thermal Infrared Based Two - Sour ce Energy Balance Model. Journal of Hydrome teorology, 16(3), 1070 - 1086. Hanson, R. L. ( 1991 ) . Evapotranspiration and droughts. Geological Survey Water - Supply Paper, 2375, 99 - 104. Herman, M.R., Nejadhashe mi, A.P., Abouali, M., Hernandez - Suare z, J.S., Danesh var, F., Zhang, Z., Anderson, M.C., Sadeghi , A.M., Hain, C.R., Sharifi, A. (2018). Evaluating the Role of Evapotranspiration Remote Sensing Data in Improving Hydrological Modeling Predictability. J. Hydr ol. 556, 39 49. Herman, M.R., Nejadha shemi, A.P., Da neshvar, F., Ross, D.M., Woznicki, S.A., Zh ang, Z., Esfahanian, A. - H. ( 2015 ) . Optimization of conservation practice implementation strategies in the context of stream health. Ecol. Eng. 84, 1 12. Immerze el, W. W., & Droogers, P. (2008). Cali bration of a di stributed hydrological model based on satel lite evapotranspiration. Journal of hydrology, 349(3), 411 - 424. 243 INPE. (2011a). CBERS 1, 2 and 2B Cameras. Retrieved from the National Institute for Space Resear ch: http://www.cbers.inpe.br/ingles/sa tellites/camera s_cbers1_2_2b.php INPE. (2011b). CBERS 1, 2 and 2B Description. Retrieved from the National Institute for Space Research: http://www.cbers.inpe.br/ingles/satellites/description_cbers1_2_2b.php INPE. (201 1c). CBERS 3 and 4 Launching. Retrieve d from the Nati onal Institute for Space Research: http://w ww.cbers.inpe.br/ingles/satellites/launching_cbers3_4.php INPE. (2011d). CBERS History. Retrieved from the National Institute for Space Research: http://www.cbe rs.inpe.br/ingles/satellites/history.p hp INPE. (2011e ). CBERS. Retrieved from the National Insti tute for Space Research: http://www.cbers.inpe.br/ingles/ Jin, X., Xu, C. Y., Zhang, Q., & Singh, V. P. (2010). Parameter and modeling uncertainty simulated by GLUE and a formal Bayesian method for a conceptual hy drological model. Journal of Hydrology, 383 (3), 147 - 155. Kim, H. W., Hwang, K., Mu, Q., Lee, S. O., & Choi, M. (2012). Validation of MODIS 16 global terrestrial evapotranspiration products in various cli mates and land cover types in Asia. KS CE Journal of C ivil Engineering, 16(2), 229 - 238. Kim, J., Mohanty, B. P., & Shin, Y. (2015). Effective soil moisture estimate and its uncertainty using multimodel simulation based on Bayesian Model Averaging. Journal of Geophysical Research: Atmospheres, 1 20(16), 8023 - 80 42. Kite, G. W., & Droogers, P. (2 000). Com paring evapotranspiration estimates from satellites, hydrological models and field data. Journal of Hydrology, 229(1), 3 - 18. Krishnamurti, T. N., Kishtawal, C. M., Zhang, Z., LaRow, T., Bachiochi, D ., Williford, E ., Gadgil, S. and Surendran, S. (2 000). Mul timodel ensemble forecasts for weather and seasonal climate. Journal of Climate, 13(23), pp.4196 - 4216. Kusre, B. C., Baruah, D. C., Bordoloi, P. K., & Patra, S. C. (2010). Assessment of hydropower p otential using GIS and hydrological modeling tech nique in Kopili River basin in Assam (India). Applied Energy , 87(1), 298 - 309. Kustas, W. P., & Norman, J. M. (1996). Use of remote sensing for evapotranspiration monitoring over land surfaces. Hydrological S ciences Journal , 41(4), 495 - 516. Kuznetsova, A., Brockhoff version 2. Lee, H.S., Liu, Y., Ward, J., Brown, J., Maestre, A., Herr, H., Fresch, M.A., Wells, E., Reed, S.M. and Jones, E. (2017). Nationwide vali dation of ensemble streamflow fore casts fro m the Hydrologic Ensemble Forecast Service (HEFS) of the US National Weather Service. In AGU Fall Meeting Abstracts. 244 Legates, D.R., McCabe, G.J. ( 1999 ) - of - res in hydrolog ic and hydroclimatic model validat ion. Wate r Resour. Res. 35, 233 241. Lillesand, T., Kiefer, R. W., & Chipman, J. (2014). Remote sensing and image interpretation. John Wiley & Sons. Linkov, I., & Burmistrov, D. (2003). Model uncertainty and choices made b y modelers: Lessons learned from t he intern ational atomic energy agency model intercomparisons. Risk Analysis, 23(6), 1297 - 1308. Long, D., Longuevergne, L., & Scanlon, B. R. (2014). Uncertainty in evapotranspiration from land surface modelin g, remote sensi ng, and GRACE satellites. Water Re sources R esearch, 50(2), 1131 - 1151. Love, B., & Nejadhashemi, A. P. (2011). Environmental impact analysis of biofuel crops expansion in the Saginaw River watershed. Journal of biobased materials and bioenerg y, 5(1), 30 - 54. Lu, L., Jun, X., Chong - Yu, X., Jianjing, C ., & Rui, W. (2009). Analysis of the sources of equifinality in hydrological models using GLUE methodology. IAHS publication, 331, 130. Lu, S., Kayastha, N., Thodsen, H., Van Griensven, A., Andersen , H.E. ( 2014 ) . M ultiobjective calibration for comparing ch annel sediment routing models in the soil and water assessment tool. J. Environ. Qual. 43, 110 120. Ma, Y., Hong, Y., Chen, Y., Yang, Y., Tang, G., Yao, Y., Long, D., Li, C., Han, Z. and Liu, R. (20 18) Performance of optimally merged multisatellite precipi tation products using the dynamic Bayesian model averaging scheme over the Tibetan Plateau. Journal of Geophysical Research: Atmospheres, 123(2), pp.814 - 834. Markovic, D., & Koch, M. (2015). Stream response to pre c ipitation variability: a spectral view bas ed on analysis and modelling of hydrological cycle components. Hydrological processes, 29(7), 1806 - 1816. Martinez - Martinez, E., Nejadhashemi, A. P., Woznicki, S. A., & Love, B. J. (2014). Modeling t he hydrological significance of wetland restoration scenar ios. Journal of environmental management, 133, 121 - 134. Maxwell, R. M., & Kollet, S. J. (2008). Interdependence of groundwater dynamics and land - energy feedbacks under climate change. Nature Geoscie nce, 1(10), 665 - 669. Mendiguren, G., Koch, J., & Stisen, S . (2017). Spatial pattern evaluation of a calibrated national hydrological model a remote - sensing - based diagnostic approach. Hydrology and Earth System Sciences, 21(12), 5987. 245 Moriasi, D. N., Arnold , J. G., Van Li e w, M. W., Bingner, R. L., Harmel, R. D., & Veith, T. L. (2007). Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Transactions of the ASABE, 50(3), 885 - 900. Mu, Q., Zhao, M., & Running, S. W. (2 011). Improveme n ts to a MODIS global terrestrial evapotran spiration algorithm. Remote Sensing of Environment, 115(8), 1781 - 1800. Naderi, F. M., Freilich, M. H., & Long, D. G. (1991). Spaceborne radar measurement of wind velocity over the ocean - an overview of the NSCAT sc a tterometer system. Proceedings of the IEEE , 79(6), 850 - 866. NASA. (1999). Soil Moisture. Retrieved from Soil Moisture: https://weather.msfc.nasa.gov/landprocess/ NASA. (2005a). The AMSU - A Instrument. Retrieved from the NASA AIRS Mission & I nstrument: https ://airs.jpl.nasa.gov/mission_and_instrumen t/instrument/amsu - instrument - detail NASA. (2005b). The Atmospheric Infrared Sounder on NASA's Aqua Satellite. Retrieved from the NASA AIRS Mission & Instrument: https://airs.jpl.nasa.gov/mission_and _instrument/miss ion NASA. (2008). Water Vapor Confirmed as Major Player in Climate Change. Retrieved from the NASA News and Features: https://www.nasa.gov/topics/earth/features/v apor_warming.html NASA. (2010a). Introduction to the Electromagnetic Spectrum. Retrieved from the NASA Science website: http://science.nasa.gov/ems/01_intro NASA. (2010b). Wave Behaviors. Retrieved from NASA Science website: http://science.nasa.gov/ems/03_ behaviors NASA. (2 014). MOD16A2: MODIS/Terra Net Evapotranspiration 8 - Day L4 Gl obal 500 m SIN Grid V006. https://lpdaac.usgs.gov/dataset_discovery/modis/modis_products_table/mod16a2_v006 (Last access: 9/22/18) NASA. (2016). GEDI. Retrieved from NASA Scienc e Beta: https://sc ience.nasa.gov/missions/gedi NASA. (2017a). AMSR - E. Retrieved from the NASA Aqua Project Science: https://aqua.nasa.gov/content/amsr - e NASA. (2017aa). MODIS. Retrieved from the NASA Aqua Project Science: https://aqua.nasa.gov/modis NASA. (2017ab). MOPITT. Retrieved from NASA TERRA: https://terra.nasa.gov/about/terra - instruments/mopitt 246 NASA. (2017ac). OSTM/Jason - 2. Retrieved from NASA Jet Propulsion Laboratory: https://sealevel.jpl.nasa.gov/missions/ostmjason2/ NASA. (2017ad). Science. Retr ieved from Aquariu s: https://aquarius.umaine.edu/cgi/science.htm NASA. (2017ae) . Surface Water and Ocean Topography Mission. Retrieved from NASA Jet Propulsion Laboratory: https://swot.jpl.nasa.gov/mission/ NASA. (2017af). Terra Instruments. Retrieved from NASA TERRA: https ://terra.nasa.gov/about/terra - instruments NASA. (2017ag). TOP EX/Poseidon. Retrieved from NASA Jet Propulsion Laboratory: https://sealevel.jpl.nasa.gov/missions/topex/ NASA. (2017ah). TRMM Instruments. Retrieved from NASA Precipitation Mea surement Missions: https://pmm.nasa.gov/TRMM/trmm - instruments NASA. (2017ai). T RMM Mission Overview. Retrieved from NASA Precipitation Measurement https://pmm.nasa.gov/TRMM/mission - overview NASA. (2017aj). TRMM. Retrieved from NASA Precipitation Measuremen t Missions: https: //pmm.nasa.gov/TRMM NASA. (2017b). Aqua Earth - observing satel lite mission. Retrieved from the NASA Aqua Project Science: https://aqua.nasa.gov/ NASA. (2017c). Aquarius Mission Overview. Retrieved from NASA Spacecraft and Instruments: http s://www.nasa.gov/m ission_pages/aquarius/overview/index.html NASA. (2017d). ASTE R. Retrieved from NASA TERRA: https://terra.nasa.gov/about/terra - instruments/aster NASA. (2017e). CERES. Retrieved from NASA TERRA: https://terra.nasa.gov/about/terra - instrument s/ceres NASA. (201 7f). CERES. Retrieved from the NASA Aqua Project Science: htt ps://aqua.nasa.gov/ceres NASA. (2017g). Global Ecosystem Dynamics Investigation Lidar (GEDI on ISS). Retrieved ns/global - ecosyste m - dynamics - investigation - lidar NASA. (2017h). History. Retrie ved from NASA Landsat Science: https://landsat.gsfc.nasa.gov/about/history/ NASA. (2017i). HSB. Retrieved from the NASA Aqua Project Science: https://aqua.nasa.gov/content/hsb 247 N ASA. (2017j). Inst ruments. Retrieved from the NASA Aqua Project Science: https: //aqua.nasa.gov/content/instruments NASA. (2017k). International Spacecraft Carrying NASA's Aquarius Instrument Ends Operations. Retrieved from Aquarius News: https://aquarius.u maine.edu/cgi/news _more.htm?id=51 NASA. (2017l). Jason - 1. Retrieved from NASA J et Propulsion Laboratory: https://sealevel.jpl.nasa.gov/missions/jason1/ NASA. (2017m). Jason - 3. Retrieved from NASA Jet Propulsion Laboratory: https://sealevel.jpl.nasa.gov/mis sions/jason3/ NASA . (2017n). Landsat 1. Retrieved from NASA Landsat Science: ht tps://landsat.gsfc.nasa.gov/landsat - 1/ NASA. (2017o). Landsat 2. Retrieved from NASA Landsat Science: https://landsat.gsfc.nasa.gov/landsat - 2/ NASA. (2017p). Landsat 3. Retrieve d from NASA Landsa t Science: https://landsat.gsfc.nasa.gov/l andsat - 3/ NASA. (20 17q). Landsat 4. Retrieved from NASA Landsat Science: https://landsat.gsfc.nasa.gov/landsat - 4/ NASA. (2017r). Landsat 5. Retrieved from NASA Landsat Science: https://landsat.gsf c.nasa.gov/landsat - 5/ NASA. (2017s). Landsat 6. Retrieved fr om NASA Landsat Sci ence: https://landsat.gsfc.nasa.gov/landsat - 6/ NASA. (2017t). Landsat 7. Retrieved from NASA Landsat Science: https://landsat.gsfc.nasa.gov/landsat - 7/ NASA. (2017u). Landsat 8. Retrieved from NAS A Landsat Science: https://landsat.gsfc.na sa.gov/landsat - data - continuity - mission/ NASA. (2017v). Landsat Overview. Retrieved from NASA Landsat Science: https://www.nasa.gov/mission_pages/landsat/overview/index.html NASA. (2017w). MISR. Re trieved from NASA TERRA: https://terra.nasa.gov/about/terra - instruments/misr NA SA. (2017x). Mission Basics. Retrieved from NASA Jet Propulsion Laboratory: https://sealevel.jpl.nasa.gov/missions/ NASA. (2017y). Mission. Retrieved from NASA TERRA: https://te rra.nasa.gov/about /mission 248 NASA. (2017z). MODIS. Retrieved f rom NASA TERRA: htt ps://terra.nasa.gov/about/terra - instruments/modis NASA. (2018a). MOD 16 Evapotranspiration. https://modis.gsfc.nasa.gov/data/dataprod/dataproducts.php?MOD_NUMBER=16 (Last acce ss: 9/22/18) NASA. (2018b). MOD16A2 - MODIS/Terra Evapotrans piration 8 - Day Leve l - 4 Global 500m SIN Grid https://ladsweb.modaps.eosdis.nasa.gov/missions - and - measurements/products/evapotranspiration/MOD16A2/ (Last access: 9/22/18) NASA. (2018c). NLDAS Concep t/Goals. https://l das.gsfc.nasa.gov/nldas/NLDASgoals.php (La st access: 9/22/18) NASA/GSFG. (2018). Goddard Earth Science Data and Information Services Center (GES DISC). https://disc.gsfc.nasa.gov/ (Last access: 9/22/18) Nash, J. E., & Sutcliffe, J. V. (19 70). River flow fo recasting through conceptual models part I A discussion of pr inciples. Journal of Hydrology , 10(3), 282 - 290. NASS . ( 2012 ) . CropScape - Cropland Data Layer. (accessed 4.26.17). NASS. (2018). CropSca pe Cropland Data Layer. https://nassgeodata.gmu.edu/CropScape/ (Last access: 9/22/18) NCDC. (2018). Data Tools: Local Climatological Data (LCD). https://www.ncdc.noaa.gov/cdo - web/datatools/lcd (Last access: 9/22/18) N ED . ( 2014 ) . The National Map: Elevation. < https://nationalmap .gov/elevation.html> (accessed 4.26.17). Neitsch, S.L., Arnold, J.G., Kiniry, J.R., and Williams, J.R. (2011) Soil and water assessment tool theoretical documentation, version 2 009. Temple, Tex.: USDA - ARS Grassland, Soil and Water Research Laboratory; http ://swat.tamu.edu/media/99192/swat2009 - theory.pdf. Nejadhashemi, A. P., Wardynski, B. J., & Munoz, J. D. (2012). Large - scale hydrologic modeling of the Michigan and Wisconsin agr icultural regions to study impacts of land use changes . Transactions of the ASA BE, 55(3), 821 - 838. Nejadhashemi, A.P., Woznicki, S.A., Douglas - Mankin, K.R. ( 2011 ) . Comparison of Four Models (STEPL, PLOAD, L - THIA, and SWAT) in Simulating Sediment, Nitrogen, and Phosphorus Lo ads and Pollutant Source Areas. T. ASABE 54, 875 890. NIDIS. (2013). Soil Moisture. Retrieved from the National Integrated Drought Information System Drought Portal: https://www.drought.gov/drought/data - maps - tools/soil - moisture 249 NIES. (201 7a). About GOSAT. Retrieved from Gosat Proj ect: http://www.gosat.nies.go.jp/en/ about_%ef%bc%92_observe.html NIES. (2017b). GOSAT. Retrieved from Gosat Project: http://www.gosat.nies.go.jp/en/ NOAA. (2017a). Land - Based Datasets and Products. Retrieved from NOAA National Cent ers for Environmental Inf ormation: https://www.ncdc.noaa.gov/ data - access/land - based - station - data/land - based - datasets NOAA. (2017b). Potential Evapotranspiration https://www.ncdc.noaa.gov/monitoring - references/dyk/potential - evapotranspirat ion (Last access: 9/22/18) NOAA - NCEP. (2013 ). Climate Prediction Center - wgrib home page. http://www.cpc.ncep.noaa.gov/products/wesley/wgrib.html (Last access: 9/22/18) NRCS, ( 2014 ) . Web Soil Survey. (accessed 4.26.17). NRCS. (2018). Web Soil Survey. https://websoilsurv ey.sc.egov.usda.gov/App/WebSoilSurvey.aspx (Last access: 9/22/18) NSIDC. (2015). SOTC: Contribution of the Cryosphere to Changes in Sea Level. Retrieved from the State of the Cr yosphere: http://n sidc.org/cryosphere/sotc/ sea_level.html NTSG. (2018). MODIS G lobal Evapotranspiration Project (MOD16). http://www.ntsg.umt.edu/project/modis/mod16.php (Last access: 9/22/18) Pan, S., Tian, H., Dangal, S. R., Yang, Q., Yang, J., Lu, C., Ta o, B., Ren, W., & Ouyang, Z. (2015). Respon ses of global terrestrial evapotrans piration to climate change and increasing atmospheric CO2 in the 21st century. Earth's Future, 3(1), 15 - 35. Qin, H., Cao, G., Kristensen, M., Refsgaard, J.C., Rasmussen, M.O., H e, X., Liu, J., Sh u, Y. and Zheng, C. (2013 ). Integrated hydrological modeling of the North China Plain and implications for sustainable water management. Hydrology and Earth System Sciences, 17(10), pp.3759 - 3778. Rajib, M. A., Merwade, V., & Yu, Z. (2016) . Multi - objective calibration of a hydrolog ic model using spatially distributed remotely sensed/in - situ soil moisture. Journal of Hydrology, 536, 192 - 207. Refsgaard, J. C., Van der Sluijs, J. P., Brown, J., & Van der Keur, P. (2006). A framework for dealin g with uncertainty due to model structure e rror. Advances in Water Resources, 2 9(11), 1586 - 1597. Reichhardt, T. (2006). The first photo from space. Air & Space magazine. Sabbaghian, R.J., Zarghami, M., Nejadhashemi, A.P., Sharifi, M.B., Herman, M.R., Dane shvar, F. ( 2016 ) . Application of risk - based multiple criteria decision analysis for selection of the 250 best agricultural scenario for effective watershed management. J. Environ. Manage. 168, 260 272. Saha, A., Thakur, P. K., & Chouksey, A. (2017). Hydrologic al Simulation usin g Process Based and Empir ical Models for Flood Peak Estimatio n. International Journal of Advanced Remote Sensing and GIS, pp - 2253. Sahoo, G. B., Ray, C., & De Carlo, E. H. (2006). Calibration and validation of a physically distributed hyd rological model, M IKE SHE, to predict strea mflow at high frequency in a flashy mountainous Hawaii stream. Journal of Hydrology, 327(1), 94 - 109. Santhi, C., Arnold, J. G., Williams, J. R., Dugas, W. A., Srinivasan, R., & Hauck, L. M. (2001). Validation of t he swat model on a large rwer basin with po int and nonpoint sources. JAWRA Jour nal of the American Water Resources Association, 37(5), 1169 - 1188. Savenije, H. H. (2001). Equifinality, a blessing in disguise?. Hydrological processes, 15(14), 2835 - 2838. Sayy aadi, H., Mehrabip our, R. ( 2012 ) . Efficienc y enhancement of a gas turbine cycle using an optimized tubular recuperative heat exchanger. Energy 38, 362 375. Schuurmans, J. M., Troch, P. A., Veldhuizen, A. A., Bastiaanssen, W. G. M., & Bierkens, M. F. P. (20 03). Assimilation of remotely sensed latent heat flux in a distributed hydrolog ical model. Advances in Water Resources, 26(2), 151 - 159. Schuurmans, J. M., Van Geer, F. C., & Bierkens, M. F. P. (2011). Remotely sensed latent heat fluxes for model error diag nosis: a case stud y. Hydrology and ea rth system sciences, 15(3), 759 - 769. Seada , H., & Deb, K. (2016). A Unified Evolutionary Optimization Procedure for Single, Multiple, and Many Objectives. IEEE Transactions on Evolutionary Computation , 20(3), 358 369. Senay, G. B., Bohm s , S., Singh, R. K., Gowda, P. H., Velpuri, N. M., Alemu, H., & Verdin, J. P. (2013). Operational evapotranspiration mapping using remote sensing and weather datasets: A new parameterization for the SSEB approach. JAWRA Journal of the Ame rican Water Resour c es Association, 49(3), 577 - 591. Senay, G. B., Gowda, P. H., Bohms, S., Howell, T. A., Friedrichs, M., Marek, T. H., & Verdin, J. P. (2014). Evaluating the SSEBop approach for evapotranspiration mapping with landsat data using lysimetric observations in th e semi - arid Texas High Plains. Hydrology and Earth System Sci ences Discussions, 11(1), 723 - 756. Sharpley, A.N., Williams, J.R. ( 1990 ) . EPIC, Erosion/Productivity Impact Calculator. Technical bulletin (USA). 251 Singh, J., Knapp, H. V., & Demi ssie, M. (2004). H y drologic modeling of the Iroquois River watershed using HSPF and SWAT. ISWS CR 2004 08. Illinois State Water Survey, Champaign. Sivakumar, B., Singh, V.P. ( 2012 ) . Hydrologic system complexity and nonlinear dynamic concepts for a catchmen t classification f r amework. Hydrol. Earth Sys. Sc. 16, 4119. Sousa, A. M. L. D. , Vitorino, M. I., Castro, N. M. D. R., Botelho, M. D. N., & Souza, P. J. O. P. D. (2015). Evapotranspiration from Remote Sensing to Improve the Swat Model in Eastern Amazonia. Floresta e Ambient e , 22(4), 456 - 464. Srinivasan, R., & Arnold, J. G. (1994). IN TEGRATION OF A BASIN SCALE WATER QUALITY MODEL WITH GIS 1. JAWRA Journal of the American Water Resources Association , 30 (3), 453 - 462. Srinivasan, V., Konar, M., & Sivapalan, M. (2017). A dynamic framework for water security. Water Security, 1, 12 - 20. Stamp oulis, D., Andreadis, K.M., Granger, S.L., Fisher, J.B., Turk, F.J., Behrangi, A., Ines, A.V. and Das, N.N. (2016). Assessing hydro - ecological vulnerability using microwave radi ometric me asuremen ts from WindSat. Remote Sensing of Environment, 184, pp.58 - 72 . Sun, S., Chen, H., Ju, W., Hua, W., Yu, M., & Yin, Y. (2014). Assessing the future hydrological cycle in the Xinjiang Basin, China, using a multi model ensemble and SWAT model . Internat ional Journal of Climatology, 34(9), 2972 - 2987. Tang, Y., Reed, P., Wag ener, T. ( 2005 ) . How effective and efficient are multiobjective evolutionary algorithms at hydrologic model calibration? Hydrol. Earth Sys. Sc. 2, 2465 2520. Tebaldi, C., & Knutti, R. ( 2007). The use of the multi - model ensemble in probabilistic climate pro jections. Philosophical Transactions of the Royal Society of London A: Mathematical, Physical and Engineering Sciences, 365(1857), 2053 - 2075. Terrell, B. L., Johnson, P. N., & Segarra, E. (2002). Ogallala aquifer depletion: economic impact on the Texas hig h plains. Water Policy, 4(1), 33 - 46. Texas A&M University . ( 2017 ) . SWAT: Soil and Water Assessment Tool. (accessed 4.26.17). Texas A&M University . ( 2018 ) . SWAT: Soil and Water Assessment Tool. http://swat.tamu.edu/ (Last access: 9/22/18) Tian, D., & Medina, H. (2017). Comparative assessment of several post - processing methods for correcting evapotranspiration fore casts derived from TIGGE datasets. In AGU Fall Meeting Abstracts. 252 Troy, T. J., Wood, E. F., & Sheffield, J. (2008). An efficient calibration method for continental scale land surface modeling. Water Resources Research, 44(9). US Climate Data. (2018). Clima te Mount Pleasant Michigan. Retrieved from the U.S. Climate Data: htt ps://www.usclimatedata.com/climate/mount - pleasant/michigan/united - states/usmi0577/2018/1 (Last access: 9/22/18) USGS . (2016a). USGS Water Data for the Nation. (accessed 4.26.17). USGS. (2016b). How much water is there on, in, and above the Earth? Retrieved from the USGS Water Science School: https://water.usgs.gov/edu/earthhowmuch.html U SGS. (2016c). The Water Cycle: Freshwater Storage Retrieved from the USGS Water Science School: https://water.usgs.gov/edu/watercyclefreshs torage.html USGS. (2016d). Evapotranspiration - The Water Cycle. Retrieved from the USGS Water Science School: https: //water.usgs.gov/edu/watercycleevapotranspiration.html (Last access: 9/22/18) USGS. (2016e). The Water Cycle: Evaporation. Retrieved from t he USGS Water Science School: https://water.usgs.gov/edu/watercycleevaporation.html USGS. (2016f). Groundwater Storag e - The Water Cycle. Retrieved from the USGS Water Science School: https://water.usgs.gov/edu/watercyclegwstorage.html USGS. (2016g). A Met hod of Determining Soil Water Content from Remotely Sensed Data. Retrieved from the Center of Excellence for Geospati al Information Science (CEGIS): https://cegis.usgs.gov/soil_moisture/ USGS. (2016h). The Water Cycle: The Oceans. Retrieved from the USGS W ater Science School: https://water.usgs.gov/edu/watercycleoceans.html USGS. (2016i). Saline water: Desalination. Retrieved from the USGS Water Science School: https://water.usgs.gov/edu/drinkseawater.html USGS. (2016j). Precipitation: The Water Cycle. Retr ieved from the USGS Water Science School: https://water.usgs.gov/edu/watercycleprecipitation.html U SGS. (2016k). Ice, Snow, and Glaciers: The Water Cycle. Retrieved from the USGS Water Science School: https://water.usgs.gov/edu/watercycleice.html USGS. (20 https://water.usgs.gov/edu/e arthwherewater.html USGS. (2016m). Runoff (surface water runoff). Retrieved from the USGS Water Science School: https://water.usgs.gov/edu/runoff.html 253 USGS. (2016n). The Water Cycle: Water Storage in the Atmosphere. Retrieved from the USGS Water Science Sc hool: https://water.usgs.gov/edu/watercycleatmosphere.html USGS. (2016o). USGS Geo Data Portal. Retrieved from Simplified Surface Energy Balance Actual Evapo transpiration data for the Conterminous U.S.: http://cida.usgs.gov/gdp/client/#!catalog/gdp/dataset /54dd5d21e4b08de9379b38b6 (Last access: 9/22/18) USGS. (2016p). USGS Current Conditions for USGS 04155500 PINE RIVER NEAR MIDLAND, MI. (accessed 4.26.17). USGS. (2016q). The Water Cycle: Evaporation. Retrieved from the USGS Water Science School: https://water.usgs.gov/edu/watercycleevaporation.html (Last access: 9/22/18) USGS. (2017a). The Water Cycle. Retrieved from the USGS Water Science School: https://water.usgs.gov/edu/watercyc le.html USGS. (2017b). USGS Surface - Water Data for the Nation. Retrieved from the USGS National Water Information System: Web Interface: https://waterdata.us gs.gov/nwis/sw USGS. (2018) National Water Information System: Web Interface, https://waterdata.usg s.gov/usa/nwis/sw Usunoff, E., Carrera, J., & Mousavi, S. F. (1992). An approach to the design of experiments for discriminating among alternative conceptual models. Advances in Water Resources, 15(3), 199 - 214. van der Tol, C., & Parodi, G. N. (2012). Guid elines for remote sensing of evapotranspiration. In Evapotranspiration - Remote Sensing and Modeling. InTech. Velpuri, N. M., Senay, G. B., Singh, R. K., Bohms , S., & Verdin, J. P. (2013). A comprehensive evaluation of two MODIS evapotranspiration products o ver the conterminous United States: Using point and gridded FLUXNET and water balance ET. Remote Sensing of Environment , 139, 35 - 49. Verma, S., Bhattarai, R. , Bosch, N. S., Cooke, R. C., Kalita, P. K., & Markus, M. (2015). Climate change impacts on flow, sediment and nutrient export in a Great Lakes watershed using SWAT. CLEAN Soil, Air, Water, 43(11), 1464 - 1474. Von Storch, H. (1999). Misuses of statistical a nalysis in climate research. In Analysis of Climate Variability (pp. 11 - 26). Springer, Berli n, Heidelberg. Wallis, T.W., Griffiths, J.F. ( 1995 ) . An assessment of the weather generator (WXGEN) used in the erosion/productivity impact calculator (EPIC). Agr. Forest Meteorol. 73, 115 133. 254 Wanders, N., Bierkens, M. F., de Jong, S. M., de Roo, A., & Ka rssenberg, D. (2014). The benefits of using remotely sensed soil moisture in parameter identification of large scale hydrological models. Water resources research , 50 (8), 6874 - 6891. Wang, S., Ancell, B. C., Huang, G. H., & Baetz, B. W. (2018). Improving Ro bustness of Hydrologic Ensemble Predictions Through Probabilistic Pre and Post Processing in Sequential Data Assimilation. Water Resources Research , 54 (3), 2129 - 2151. Wang, S., Pan, M., Mu, Q., Shi, X., Mao, J., Brümmer, C., Jassal, R.S., Krishnan, P., Li, J. and Black, T.A. (2015). Comparing evapotranspiration from eddy covariance measurements, water budgets, remote sensing, and land surface models over Canada. Journal of Hydrometeorology , 16(4), 1540 - 1560. Wang, Z. (2013). cts : an R package for continuous time autoregressive models via Kalman filter. Journal of Statistical Software , 53(5), 1 - 19. White, K. L., & Chaubey, I. (2005). Sensitivity analysis, calibration, and validations for a multisite and multivariable SWAT model. JAWRA Journal of the American Water Resources Association, 41(5), 1077 - 1089. Wilcoxon, F. (1945). Individual comparisons by ranking methods. Biometrics bulletin , 1 (6), 80 - 83. Woznicki, S.A., Nejadhashemi, A.P. ( 2012 ) . Sensitivity analysis of best ma nagement practices under climate chan ge scenarios. J. Am. Water Resour. As. 48, 90 112. Wu, B. F., Xiong, J., Yan, N. N., Yang, L. D., & Du, X. (2008). ETWatch for monitoring regional evapotranspiration with remote sensing. Advances in Water Science , 19 (5), 671 - 678. Xia, Y., Hobbins, M. T., M u, Q., & Ek, M. B. (2015). Evaluation of NLDAS 2 evapotranspiration against tower flux site observations. Hydrological processes, 29(7), 1757 - 1771. Xu, X., Li, J., & Tolson, B. A. (2014). Progress in integrating remo te sensing data and hydrologic modeling. Progress in Physical Geography, 38(4), 464 - 498. Yang, Y., Long, D., Guan, H., Liang, W., Simmons, C., & Batelaan, O. (2015). Comparison of three dual source remote sensing evapotranspiration models during the MUSOEX E 12 campaign: Revisit of model physics. Water Resources Research, 51(5), 3145 - 3 165. Yao, Y., Liang, S., Li, X., Chen, J., Liu, S., Jia, K., Zhang, X., Xiao, Z., Fisher, J.B., Mu, Q. and Pan, M. (2017). Improving global terrestrial evapotranspiration estim ation using support vector machine by in tegrating three process - based algorithms. Agricultural and forest meteorology, 242, pp.55 - 74. Zeleny, M., & Cochrane, J. L. (1973). Multiple criteria decision making . University of South Carolina Press. 255 Zhan, C. - S., Song, X. - M., Xia, J., Tong, C. ( 2013 ) . A n efficient integrated approach for global sensitivity analysis of hydrological model parameters. Environ. Model. Softw. 41, 39 52. Zhang, K., Kimball, J. S., & Running, S. W. (2016). A review of remote sensing based actual evapotranspiration estimation. W iley Interdisciplinary Reviews: Water, 3(6), 834 - 853. Zhang, X., Srinivasan, R., Liew, M.V. ( 2010 ) . On the use of multi - algorithm, genetically adaptive multi - objective method for multi - site calibration of the SWAT mo del. Hydrol. Process. 24, 955 969. Zhang , Y., Shao, Q., Taylor, J.A. ( 2016 ) . A balanced calibration of water quantity and quality by multi - objective optimization for integrated water system model. J. Hydrol. 538, 802 816.