AGRONOMIC MANAGEMENT OF CORN USING SEASONAL CLIMATE PREDICTIONS, REMOTE SENSING AND CROP SIMULATION MODELS By Prakash Kumar Jha A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Crop and Soil Sciences - Doctor of Philosophy 2019 ABSTRACT AGRONOMIC MANAGEMENT OF CORN USING SEASONAL CLIMATE PREDICTIONS, REMOTE SENSING AND CROP SIMULATION MODELS By Prakash Kumar Jha Management decisions in corn (Zea mays mays L ) production are usually based on specific growth stages . However, because of climate and weather variability, phenological stages vary from season to season across geographic locations. This variability in growth and phenology entails risks and quantifying it will help in managing climate related risks. Crop simulation models can play a significant role in minimizing these risks through designing management strategies ; however, they are not al ways accurate . R emote sensing observations and climate predictions can improve the accuracy in managing time bound climate - sensitive decisions at larger spatiotemporal scale . However, there is also a disconnect between climate forecasts and crop model s . Th e unavailability of downscal ing tool that can downscale rainfall and temperature forecasts simultaneously make this task more challenging. To address these knowledge gaps, this dissertation consists of three studies focused on interdisciplinary approaches to agronomic management of corn. In the first study, we calibrated and validated genetic coefficients of CERES - Maize using field data from the Michigan corn performance trials. Multiple methods of estimating genetic coefficients GENCALC ( Genotype Coeffici ent Calculator), GLUE (Generalized Likelihood Uncertainty Estimate), and NMCGA (Noisy Monte Carlo Genetic Algorithm) were evaluate d and ensembled to estimate more reliable genetic coefficients. The calibrations were done under irrigated conditions and vali dation under rainfed conditions . The results suggested that ensembled genetic coefficients performed best among all, with d - index of 0.94 and 0.96 in calibrati on and validation for anthesis and maturity date s, and yield . In the second study, simulated gro wth stages from the calibrated crop model were used to develop site - specific crop coefficients (k c ) using ensembled ET and reference ET from the nearest weather station. ET from multiple models were ensembled and validated with the measured ET from eddy - covariance flux towers for 2010 2017 . R esults suggest that the ensembled ET performed best among all ET models used, with highest d - index of 0.94. Likewise, t he performance of the newly derived kc - curve was compared with FAO - kc curve using a soil wa ter balance model. Then, the derived region - specific K c - curve was used to design irrigation scheduling and results suggest that it performed better than FAO Kc - curve in minimizing the amount irrigation while maintaining a prescribed allowable water stress . The third study used the calibrated crop model to simulate anthesis using downscaled seasonal climate forecasts . The predicted anthesis and downscaled seasonal climate forecasts w ere used to develop risk analysis model for ear rot disease management in corn. In this study an innovative downscaling tool , called FResamplerPT , was introduced to downscale rainfall and temperature simultaneously . The results suggest that temperature and re lative humidity are better predictor s (combined) as compared to temperature and rainfall (combined) . With this risk analysis model, growers can evaluate and assess the future climatic conditions in the season before planting the crops. The seasonal climate information with the lead - time of 3 months can help growers to prepare integrated management strategies for ear rot disease management in maize . Copyright by PRAKASH KUMAR JHA 2019 v This thesis is dedicated to the supreme personality of Godhead, my lovely family and friends for all the support and kindness they have given me. vi ACKNOWLEDGMENTS I acknowledge the great power, who has paved the way on which I have walked so far. As a prelude to my thanksgivin g , at first I wish to thank and bow my head before Lord Almighty, for blessing me with all his grace to successfully complete my studies and research endeavor at Michigan State University. At this moment of accomplishment, it is my great privilege to express a deep sense of gratitude to my advisor Dr. Amor V. M. Ines. As the chair of my advisory committee, he guided me as a teacher, helped me like a friend and cared me as a guardian. I am also thankful for every bit of his valuable guidance, personal a ttention, constructive criticism, immense patience, untiring help, wholehearted support and encouragement throughout the period of my study and preparation of this manuscript. I am sincerely thankful to Dr. Amor Ines for the funding support during my resea rch program. I specifically acknowledge the funding support from IFPRI, Columbia University, NASA - SERVIR, Chubu University, USAID - Phillipines, ListenField, MSU AgBioResearch and Corn Marketing Program of Michigan (CMPM) for the entire study. I also acknow ledge College of Agricultural and Natural Resources (CANR), Michigan State University for the fellowships during my course, which helped me in completing my work smoothly. I am also thankful to South Asia Studies centre at Michigan State University for off - koo I express my sincere thanks and gratefulness to Dr. Maninder Singh, for encouraging me to undertake this study and providing me with his precious advice, ne cessary guidance and help rendered during my field studies. I express my sincere and heartiest thanks to Dr. Bruno Basso, Dr. Jeff Andresen and Dr. Pouyan Nejadhashemi for, ever - ready support and guidance all the vii time during my research work as the members of my advisory committee and specially for their most valuable help rendered in relation to planning, execution and analysis of this work. I owe a great deal of gratitude towards my teachers and scientists in the department, for their valuable suggestions and constant encouragement during my course of research. Specially, I would like to acknowledge Dr. Jiquan Chen, Dr. Michael Abraha, Dr. Joshua Fisher, Dr. Younsuk Dong, Steve Miller, and Lyndon Kelley for their invaluable contributions in shaping my diss ertation. My sincere thanks go to Dr. Chirag K. Vyas, Mr. Suyog Chaudhari, Mr. Eeswaran Rasu and Mr. Abhijit Abhishek for their timely help and encouragement. No words can describe the unending love and constant moral support from my parents and lovely wif e, Mrs. Supriya Jha. I affectionately cherish the blessings and good wishes of my beloved parents, my loving siblings, and all my relatives who were the constant sources of inspiration to me from miles apart. viii TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ .......................... x LIST OF FIGURES ................................ ................................ ................................ ...................... xii KEY TO ABBREVIAT IONS ................................ ................................ ................................ ...... xvi 1. INTRODUCTION ................................ ................................ ................................ ................... 1 2. INTRODUCTION TO METHODOLOGY AND RESULTS ................................ ................. 6 3. ESTIMATING GENETIC COEFFICIENTS OF CERES - MAIZE TO SIMULATE PHENOLOGY AND YIELD OF MAIZE IN MICHIGAN ................................ ........................... 9 3.1 Introduction ................................ ................................ ................................ ........................... 9 3.2 Materials and Methods ................................ ................................ ................................ ........ 14 3.2.1 Study area ................................ ................................ ................................ ..................... 14 3.2.2 CERES - Maize model ................................ ................................ ................................ .... 16 3.2.3 Inpu t data for CERES - Maize ................................ ................................ ........................ 17 3.2.4 Calibration methods ................................ ................................ ................................ ..... 21 3.2.5. Ensembling approach ................................ ................................ ................................ .. 28 3.2.6 Validation and statistical analysis ................................ ................................ ................ 29 3.3. Results and Discussion ................................ ................................ ................................ ....... 30 3.3.1 Calibration ................................ ................................ ................................ .................... 30 3.3.2 Potential yields from calibrated genetic coefficients ................................ ................... 38 3.3.3 Validation ................................ ................................ ................................ ..................... 39 3.4 Summary and Conclusions ................................ ................................ ................................ .. 48 3.5 Acknowledgment ................................ ................................ ................................ ................ 49 4. ESTIMATION AND VALIDATION OF REMOTELY SENSED EVAPOTRANSPIRATION FOR THE DEVELOPMENT OF CROP COEFFICIENTS OF MAIZE AND IRRIGATION SCHEDULING ................................ ................................ ................................ ............................. 50 4.1 Introduction ................................ ................................ ................................ ......................... 50 4.2 Materials and Methods ................................ ................................ ................................ ........ 53 4.2.1 Study Area ................................ ................................ ................................ ..................... 53 4.2.2 Data Collection ................................ ................................ ................................ ............. 55 4.2.3 ET mod els descriptions ................................ ................................ ................................ . 60 4.2.4 Ensembling approach ................................ ................................ ................................ ... 63 4.2.5 Kc Curve and Irrigation Scheduling ................................ ................................ ............ 65 4.2.6 Development of Kc - curve ................................ ................................ ............................ 66 4.2.7 Validation of Kc - curve ................................ ................................ ................................ .. 67 4.2.8 Statistical evaluation of model performance ................................ ................................ 68 4.3 Results and Discussion ................................ ................................ ................................ ........ 69 4.3.1 ET Estimation and comparison ................................ ................................ .................... 69 4.3.2 Ensembling ET estimates ................................ ................................ .............................. 79 ix 4.3.3 Development of K c Curve ................................ ................................ .......................... 85 4.3.4 Validation of Kc Curve ................................ ................................ .............................. 90 4.3.5 Irrigation Scheduling of Maize: a comparison of standard FAO - K c and the derived K c ................................ ................................ ................................ ................................ ............... 95 4.4 Summar y and Conclusions ................................ ................................ ................................ .. 96 4.5 Acknowledgements ................................ ................................ ................................ ............. 98 5. DOWNSCALI NG SEASONAL RAINFALL AND TEMPERATURE FORECASTS TO DEVELOP A RISK ANALYSIS MODEL FOR EAR ROT DISEASE MANAGEMENT IN MAIZE ................................ ................................ ................................ ................................ .......... 99 5.1 Introduction ................................ ................................ ................................ ......................... 99 5.2 Materials an d Method ................................ ................................ ................................ ........ 103 5.2.1 Study Area ................................ ................................ ................................ ................... 103 5.2.2 Data colle ction ................................ ................................ ................................ ........... 104 5.2.3 Downscaling rainfall and temperature: software description ................................ ... 109 5.2.4 Crop model setup ................................ ................................ ................................ ........ 111 5.2.5 Deve loping risk indicators ................................ ................................ .......................... 111 5.2.6 Statistical analysis ................................ ................................ ................................ ...... 112 5.3 Results and Discussions ................................ ................................ ................................ .... 113 5.3.1 Predicting phenology ................................ ................................ ................................ .. 113 5.3.2 Risk Probabilities using Forecast data ................................ ................................ ...... 115 5.3.3 Validation of risk analysis model ................................ ................................ ............... 127 5.4 Summary and Conclusions ................................ ................................ ................................ 131 5.5 Acknowledgement ................................ ................................ ................................ ............. 132 6. CONCLUSIO NS ................................ ................................ ................................ ................. 133 7. FUTURE RESEARCH RECOMMENDATIONS ................................ .............................. 136 REFERENCES ................................ ................................ ................................ ........................... 138 x LIST OF TABLES Table 3. 1. Location of the maize fields under Michigan maize Performance Trials (MMPT). .. 15 Table 3. 2. Summary of weather data (2017 and 2018) during the crop season (May - October) for selected locations. ................................ ................................ ................................ ......................... 17 Table 3. 3. Descriptions of field observations used in the study. ................................ ................. 19 Table 3. 4. Descriptions of genetic coefficients in CERES - Maize. ................................ .............. 22 Table 3. 5. Representations of phenology and growth parameters in NMCGA. .......................... 28 Table 3. 6. Genetic coeffi cients estimated by GENCALC, GLUE, NMCGA and Ensembling approach in 2017. ................................ ................................ ................................ .......................... 31 Table 3. 7. Performance of CERES - Maize using estimated p arameters by GENCALC, GLUE, NMCGA and ensembling approach during calibration in 2017. ................................ .................. 32 Table 3. 8. Genetic coefficients estimated by GENCALC, GLUE, NMCGA_SD, NMCGA_NO_SD and Ensembling approach in 2017 and 2018. ................................ ................ 33 Table 3. 9. Performance of CERES - Maize using parameters estimated by GENCALC, GLUE, NMCGA and ensembling approach during calibration in 2017 and 2018. ................................ .. 34 Table 3. 10. Per formance of CERES - Maize using parameters estimated by GENCALC, GLUE, NMCGA and ensembling approach during validation at rainfed locations with soil profiles used in calibration. ................................ ................................ ................................ ................................ 39 Table 3. 11. Soil root growth distribution factor (SRGF) of the sandy loam soil used in calibration and validation. ................................ ................................ ................................ ............. 41 Table 3. 12. Soil root growth distribution factor (SRGF) of the loam soil used in calibration and validation. ................................ ................................ ................................ ................................ ...... 41 Table 3. 13. Performance of CERES - Maize using parameters estimated by GENCALC, GLUE, NMCGA and ensembling approach during validation at rainfed locations with the revised soil prof iles. ................................ ................................ ................................ ................................ ......... 45 Table 4. 1. Statistical evaluation of seasonal ET estimates for 2010 - 2017 among the models and flux tower during the growing season (May - October) at Marshall and KBS maize farms. ........ 72 Table 4. 2. DSSAT based simulated and up dated length of crop development stages used for ensembling purposes ................................ ................................ ................................ ..................... 79 xi Table 4. 3. Crop coefficients as an input for soil water balance model ................................ ........ 90 Table 4. 4. Statistical evaluation of soil water balance model perform ance under three different rainfall inputs by comparing observed and simulated total available water in root zone (mm). .. 91 Table 4. 5 . Available water holding capacity of soil and percent capacity filled at different depth for the sandy clay loam (Jagtap et al., 2004) ................................ ................................ ................ 95 Table 5. 1. Monthly averages of weather variables (1981 - 2010) in the study area. 105 Table 5. 2. Field observation and U2U estimation of silking stage at Saginaw, Huron and Montcalm in 2018. ................................ ................................ ................................ ...................... 109 Table 5. 3. Statistical significance of forecast risk probabilities compared to climatological risks. ................................ ................................ ................................ ................................ ..................... 126 xii LIST OF FIGURES Figure 2. 1. The general flowchart for the methodology of this dissertation. ................................ . 7 Figure 3. 1. Locations of maize performance trials in Michigan. ................................ ................. 14 Figure 3. 2. Weather data (Rainfall, solar radiation, maximum and minimum temperature)for cropping season (May - October) for 2017 (Source: MSU Enviroweather) ................................ ... 20 Figure 3. 3. Weather data (Rainfall, solar radiation, maximum and minimum temperature) for cropping season (May - October) for 2018. (Source: MSU Enviroweather) ................................ .. 20 Figure 3. 4. GENCALC - CERES - Maize working flowchart (Adnan et al., 2019). ...................... 23 Figure 3. 5. CERES - Maize performance after calibration of GENCALC, GLUE, NMCGA_SD, NMCGA_NO_SD, arithmetic and weighted averaging for 2017 and 2018. ................................ 37 Figure 3. 6. Potential yields and observed yield for 2017 and 2018 for all maize trial locations using weighted averaging of genetic coefficients. ................................ ................................ ........ 38 Figure 3. 7. Model performance ((root depth, water and nitrogen stress andyield (a)) and (root length density (b,c))) in 2017 at W ashtenaw ( rainfed ) under old (b) and revised soil (c) profiles. ................................ ................................ ................................ ................................ ....................... 43 Figure 3. 8. Model performance ((Root depth, water and nitrogen st ress, yield (a)) and (root length density (b,c))) in 2017 at Cass ( irrigated ) under old (b) and revised soil (c) profiles. ...... 45 Figure 3. 9. CERES - Maize performance for validation of calibrated coefficients of GENCALC, GLUE, NMCGA_SD, NMCGA_NO_SD, arithmetic and weighted averaging for 2017 and 2018. ................................ ................................ ................................ ................................ ....................... 47 Figure 4. 1. Eddy covariance towers at the maize fields ................................ .............................. 54 Figure 4. 2. Mean reference ET (2010 - 2017) from the nearest weather station (Hasting, Michigan) in the study area. ................................ ................................ ................................ .......... 56 Figure 4. 3. Rain gauge (a) and soil moisture sensors (b) at the maize field in Manchester, Michigan. ................................ ................................ ................................ ................................ ...... 57 Figure 4. 4. Crop coefficient curve from FAO and the curve created from CERES - Maize growth stages. ................................ ................................ ................................ ................................ ............ 65 Figure 4. 5. Conceptualization of developing satellite derived crop coefficient curve. .............. 66 Figure 4. 6. Bounding box for remote sensing analysis in the study area ................................ .... 69 xiii Figure 4. 7. ET map (SEBAL, PT - JPL and MOD16A2) for a representative d ay during growing season (June 26, 2012) ................................ ................................ ................................ .................. 70 Figure 4. 8. Comparison of estimated ET (a) MOD16A2, (b) SEBAL, (c) PT - JPL, (d) FAO with the eddy covariance measurements at the KBS maize field for 2012. ................................ .......... 71 Figure 4. 9. Temporal variation of monthly ET among the me thods (b) FAO (c) SEBAL, (d) PT - JPL, (e) MOD16A2 and ground observation of flux tower (a) during growing season (May - October) at KBS maize farm ................................ ................................ ................................ ......... 75 Figure 4. 10. Temporal variation of seasonal ET among the models and ground observations from flux towers during growing season (May - October) at KBS (a) and Marshall (b) .............. 76 Figure 4. 11. Inter model comparison of seasonal ET among the models and ground observations from the flux tower during growing season (May - October) at KBS ................................ ........... 77 Figure 4. 12. Inter model comparison of seasonal ET among the models and ground observations from the flux tower during growing se ason (May - October) at Marshall ................................ ..... 78 Figure 4. 13. Ensemble ET estimates by stage based Inverse Distance Weighting (IDW) for 2012 with a reference of ground observations from the flux tower during growing season (May - October) at KBS. ................................ ................................ ................................ ........................... 80 Figure 4. 14. Variation in time series of seasonal ET (monthly average) estimates among the models and ground observations from flux tower during growing season (May - October) at KBS (ET - Average is from ensemble method) ................................ ................................ ....................... 81 Figure 4. 15. Variation in time series of seasonal ET (monthly average) estimates among the models and ground observations from flux tower during growing season (May - October) at M arshall (ET - Average is from ensemble method) Moreover, the interannual varaition in ET .... 82 Figure 4. 16. Variation in ensembled daily E T estimates for crop season (May - October) for 2010 - 2017 at KBS ................................ ................................ ................................ ......................... 83 Figure 4. 17. Variation in ensembled daily ET estimates for crop season (May - October) for 2010 - 2017 at Marshall ................................ ................................ ................................ .................. 84 Figure 4. 18. Temporal variation in crop coefficient curve of Maize derived by ensemble ET for 2010 - 2017 at KBS ................................ ................................ ................................ ......................... 86 Figure 4. 19. Temporal variation in crop coefficient curve of Maize derived by ensemble ET for 2010 - 2017 at Marshall ................................ ................................ ................................ ................. 87 Figure 4. 20. Smooth fitted crop coefficient curve of Maize derived by ensemble ET for 2010 - 2017 at KBS and Marshall ................................ ................................ ................................ ............ 88 xiv Figure 4. 21. Averaged (2010 - 2017) crop coefficient curve of Maize at KBS and Marshall ...... 89 Figure 4. 22. Observed rainfall and irrigation amount with (a) volumetric soil moisture at different depth (15, 30, 45, 60 and 90cm); (b) total available wate r in root zone (90 cm) ........... 92 Figure 4. 23. Observed and simulated total available water (TAW) in root zone with three rainfall input (a) rain gauge (b) weather station (c) reanalyzed product from NASA - POWER, and irrigation value reported by grower at Manchester maize field for four Kc - curve (FAO, KBS, Marshall, and average) in 2019 maize growing season. ................................ ............................... 93 Figure 4. 24. Simulated total available water in root zone using water balance model for two crop coefficients (a) standard FAO - K c (b) K c derived in 2019 m aize growing season with 60 % set capacity of available water. ................................ ................................ ................................ ........... 94 Figure 4. 25. Simulated total available water in root zone using water balance model for standard FAO - K c capacity of available water. ................................ ................................ ................................ ........... 96 Figure 5. 1. Study area having high ear rot disease severity in 2018. ................................ ........ 103 Figure 5. 2. Observed weather variables for 2017 and 2018 during crop growing season in the study area. ................................ ................................ ................................ ................................ ... 104 Figure 5. 3. Illustration of (a) historical and climate prediction utility for management decisions (b) prediction horizon of seasonal climate forecast and usability for farm related d ecisions. .... 107 Figure 5. 4. Framework for downscaling probabilistic seasonal rainfall (P) & temperature (T) forecasts that preserves forecast probabilities of rainfall p (P) and temperature p (T) (Ines et al., 2018) ................................ ................................ ................................ ................................ ........... 110 Figure 5. 5. Predicted anthesis for all three locations for 2018 forecast (a), warm and humid forecast (b), warm and dry forecast (c), cool and wet forecast (d) , and cool and dry forecast (e). ................................ ................................ ................................ ................................ ..................... 114 Figure 5. 6. Risk probabilities for temperature (a - c), relative humidity (RH) (d - f) and rain (g - i) at Saginaw (a,d,g), Huron (b,e,h), and Montcalm (c,f,i) for 2018 forecast and climatology (1981 - 2010). The red arrow indicates predicted mean anthesis days ................................ .................... 117 Figure 5. 7. Risk probabilities for combination of temperature and relative humidity and combination of temperature and rain at Saginaw (a), Huron (b), and Montcalm (c) for 2018 forecast and climatology (1981 - 2010). The red ar row indicates predicted mean anthesis days 118 Figure 5. 8. Risk probabilities for temperature at Saginaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) cool and wet, (d) cool and dry forecast scenarios and xv climatology (1981 - 2010) for all three locations. The red arrow indicates pred icted mean anthesis days. ................................ ................................ ................................ ................................ ............ 121 Figure 5. 9. Risk probabilities for relative humidity (RH) at Saginaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) cool and wet, (d) cool and dry forecast scenarios and climatology (1981 - 2010) for all three locations. The red arrow indicates predicted mean anthesis days. ................................ ................................ ................................ ................................ ............ 122 Figure 5. 10. Risk probabilities for rain at Saginaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) cool and wet, (d) cool and dry forecast scenarios and climatology (1981 - 201 0) for all three locations. The red arrow indicates predicted mean anthesis days. ..... 123 Figure 5. 11. Risk probabilities for combination of temperature and relative humidity (T&RH) at Saginaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) cool and wet, (d) cool and dry forecast scenarios and climatology (1981 - 2010) for all three locations. The red arrow indica tes predicted mean anthesis days. ................................ ................................ ........... 124 Figure 5. 12. Risk probabilities for combination of rain and temperature and rain (R&T) at Sag inaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) cool and wet, (d) cool and dry forecast scenarios and climatology (1981 - 2010) for all three locations. The red arrow indicates predicted mean anthesis days. ................................ ................................ ........... 125 Figure 5. 13. DON concertation at each location in 2017 and 2018; Source: Blaine et al., (2018) ................................ ................................ ................................ ................................ ..................... 127 Figure 5. 14. Risk factor of weather variable(s) in predicting ear rot disease for Saginaw, Huron and Montcalm (a) relative humidity (RH), (b) temperature (T), (c) rainfall (R), (d) RH & T, (e) R & T in 2017 and 2018. ................................ ................................ ................................ ................ 130 xvi KEY TO ABBREVIATIONS ADAP: Anthesis Days After Planting AN: Above Normal APSIM: A gricultural P roduction S ystems s IM ulator ALEXI: Atmosphere - Land Exchange Inverse AVHRR: Advanced Very High Resolution Spectroradiometer Solution BN: Below Normal CCP: Cloud Climatology Project CERES: Crop Environment Resource Synthesis CRM: Comparative Relative Maturity CRP: Conservation Reserve Program CSM: Cropping System Model DSSAT: Decision Support System for Agro - technology Transfer DAP: Days after p lanting DON: Deoxynivalenol .dat: data file ERI: Ear Rot Incidence E: Environment EC: eddy covariance ECOSTRESS: ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station ENVI: Environment for Visualizing Images xvii ET: Evapotranspiration ET a : Actual evapotranspiration ET crop : Actual crop evapotranspiration ET ref : Reference Evapotranspiration ET RS : Evapotranspiration from remote sensing FAO: Food and Agriculture Organization of the United Nations FResamplerPT: Forecast Resampler Precipitation and Temperature FYM: Farm Yard Manure G: Genoty pe GA: Genetic algorithm GC - MS: Gas Chromatography - Mass Spectrometry GDD: Growing Degree Days GDD F : Growing Degree Days (in F) GDD C : Growing Degree Days (in C) GENCALC: Genotype Coefficient Calculator GHCN: Global Historical Climatology Network GLUE: Generalized Likelihood Uncertainty Estimate GMAO: Global Modeling and Assimilation Office GPCP: Global Precipitation Climate Project GREEEN: Generating Research and Extension to meet Economic and Environmental Needs GRIB: General Regularly - distributed Information in Binary form Ha: hectare HDF: Hierarchical Data format xviii . hdr: Header file HWAM: Yield at harvest maturity IDW: Inverse Distance Weightage IPCC: Intergovernmental Panel on Climate Change ISLSCP: International Satellite Land - Surface Climatolo gy Project IR: Infrared JPL: Jet Propulsion Laboratory K: Potassium K c : Crop coefficients K cr : Crop coefficients derived through reflectance Kg: Kilogram M: Management MBE: Mean Bias Error MDAP: Maturity Date Days After Planting MDARD: Michigan Department of Agriculture and Rural Development METRIC: Mapping EvapoTranspiration at high Resolution with Internalized Calibration MMPT: Michigan Maize Performance Trial locations MOD09GA: Daily surface reflectance product MOD11A1: Land surface temperature & emissivity product MOD16A2: MODIS Global Evapotranspiration Project MODIS: Moderate Resolution Imaging Spectroradiometer MRT: MODIS Reprojection Tool xix MCTK: MODIS conversion toolkit MSU: Michigan State University N: Nitr ogen NN: Near - Normal NASA: National Aeronautics and Space Administration NASS: National Agricultural Statistics Service NCDC: National Climatic Data Center NCEP: National Centers for Environmental Prediction NDVI: Normalized Difference Vegetation Index NDF D: National Digital Forecast Database NH 4+ : Ammonium ion NIR: Near Infrared NOAA: National Oceanic and Atmospheric Administration NO 3 - : Nitrate ion NO_SD: without standard deviation NMCGA: Noisy Monte Carlo Genetic Algorithm P: Phosphorus P: Precipitation PM: Penman - Monteith POWER: Prediction Of Worldwide Energy Resources PT: Preistley - Taylor R: Coefficient of correlation R 2 : Coefficient of determinations xx R: Rainfall RH: Relative humidity RS: Remote Sensing RSME: Root mean squared error SD: Standard Deviation SCF: Seasonal Climate Forecast S2S: Sub seasonal to Seasonal SEBAL: Surface Energy Balance Algorithm for Land SEBS: Surface Energy Balance System SAVI: Soil Adjusted Vegetation Index SSEBop: Simplified Surface Energy Balance operation al SRGF: Soil Root Growth Factor V: Vegetative stages of maize VPD: Vapor Pressure Deficit S - IDW: Stage based - Inverse Distance Weightage SSURGO: Soil Survey Geographic Database SWAP: Soil, Water, Atmosphere and Plant T: Temperature T max : Maximum temperature T min : Minimum temperature TIR: Thermal Infrared TAW: Total available water US: United States xxi USGS: United States Geological Survey U2U: Useful to Usable U.S.A: United States of America WFO: Weather Forecast Offices WMO: World Meteorological or ganization WUE: Water Use Efficiency WSS: Web Soil Survey 1 1. INTRODUCTI ON Maize ( Zea mays L.) is a major crop grown in the Midwestern US. Most of the maize grown in Michigan are rainfed and used for biofuel and animal feeds. However, there are also irrigated system in the southwest region. To meet the growing demand on maize for feed, food and fuel, genetic performance of hybrid maize must be improved and agronomic practices, optimized. Management decisions in maize are associated with critical crop growth stages and controlled by climatic variability. Moreover, these growth stages are closely related to crop productivity and has economic implications. Therefore, investigating phenological changes in maize associated with variable climate is important to improve management of resources and to achieve better yields. With a goal of achieving high yields, plant breeders and agronomists have studied interactions of genotype, environment and management through field experiments (Elias et al., 2016). However, field experiments are constrained by time and resources. Crop simulation models can facilitat e the process by extrapolation of field experiments at larger spatiotemporal scales (Lobell et al., 2009). Crop simulation models have been used in assessing model performances and predicting yields (Rosenzweig et al., 2013). However, uncertainty in crop m odel output may arise due to the choice of the model (structure, parameter and assumption) or to erroneous input data (Zhang et al., 2019). Here, cultivar specific factors, so - called genetic coefficients, are important parameters used by the crop model for predicting crop daily growth and development response to weather, soil and management practices. Estimation of crop genetic coefficients for specific agro climatic conditions is a first step to be done before a crop model can be applied (Wallach et al., 2 001; Jones et al., 2011). This can be done through extensive field experiments or multiple methods, each has its own advantages and 2 limitations leading to uncertainty in parameter estimates, and hence prediction. A multi - model ensemble approach reduces the uncertainty and improves prediction. The strength of using an ensemble is that, it averages out biases, reduces the variance, and is unlikely to overfit. However, in the crop modelling community, there is a lack of consensus over ensembling parameters of the models instead of ensembling model itself (Challinor et al., 2009) unlike in hydrology field. To date, there is no documented study yet, which performs ensembling of crop genetic coefficients to simulate phenology and yield of maize ( Zea mays L. ). Knowledge Gap 1: Lack of understanding on the applicability of ensemble genetic coefficients in a crop model. This dissertation will advance our understanding on the impact of ensembling of genetic coefficients in a maize model. To address knowledge gap 1 , the following objectives were developed: (1) to calibrate genetic coefficients of a maize hybrid in Michigan using Genotype Coefficient Calculator (GENCALC), G eneralized Likelihood Uncertainty Estimation ( GLUE), Noisy Monte Carlo Genetic Algorithm (NMCGA), and an ensembling approach; and (2) to validate estimated genetic coefficients in simulating maize phenology and yields in Michigan. In order to manage resources at a large spatiotemporal scale, remote sensing (RS) can play a significant role in observing and estimating change in biophysical parameters at the surface (Lobell et al., 2018). Moreover, RS when integrated with ground measurements and crop simulation models can significantly improve management decisions. However, translating r emote sensing data into field applications has its own limitations and uncertainties (Zhang et al., 2016). Management decisions like irrigation during critical crop growth stages require accurate 3 estimation of evapotranspiration (ET). Traditionally, the st andard FAO 56 - crop coefficient ( K c ) is used to estimate ET (as a product of K c and reference ET (ET r ) based on Penman - Monteith approach). However, the validity of these k c values may not be general in every agro - climatic condition . Therefore, region specific K c - curves that reflect the genetic characteristics of a maize hybrid are needed to be estimated. The K c values can be estimated ( K c = ET c /ET r ) from in - situ methods or RS - based ET models. However, ET method/model choice and input data can lead to uncertainty and biases in the outputs. As with the genetic coefficient study above, a multi - model ensemble approach can improve estimates by reducing t he uncertainty associated with individual models. Knowledge Gap 2: Lack of utility of ensembled Evapotranspiration (ET) products in developing crop coefficients (Kc) using predicted growth stage for irrigation scheduling in maize . T here is a limited rese arch in estimating crop coefficients from an ensemble of remote sensing ET (RS - ET) models and its application in irrigation scheduling of maize. This dissertation will advance our understanding on the applicability of crop coefficients developed from RS fo r irrigation scheduling of maize. To address knowledge gap 2 , the following objectives were developed: (1) to estimate evapotranspiration (ET) using ensemble of models (RS - based SEBAL, Priestley - Taylor - JPL, MODIS Penman - Monteith, and ground - based empirical model, FAO - k c ); (2) to derive crop coefficients of maize based on the ensembled ET and reference ET; and (3) to validate derived kc - curves using a soil water balance model and design irrigation scheduling. 4 Moreover, management decis ions in maize production are usually coordinated with specific growth stages, which are very sensitive to climate and weather variability (Egli, 2008). Rainfed maize in Michigan are sensitive to in - season climatic variability leading to year - to - year variab ility of phenological stages. This variability phenology entails risks and quantifying it will help in managing climate related risks. In maize, weather related disease risks are prevalent in the last few years. Due to continuous maize rotation with minimu m tillage, conditions become more favourable for fungal diseases outbreaks (Wise and Mueller, 2011). In Michigan, Gibberella and Fusarium , which produce mycotoxins, are the primarily causal organisms for the development of ear rot disease in maize (Chilver s, 2018). The pathogen spreads through infecting maize ears during early silk stage, approximately 6 to 8 days after silk emergence. Ear rot disease infection coincides with the anthesis period. Hence, it is important to have an accurate prediction of this phenological stage in order to manage this disease. The calibrated crop model can be used to predict anthesis using seasonal climate forecast (SCF) (Apipattanavis et al., 2010). However, climate forecasts produced by NOAA are in tercile probabilities, whi ch are needed to be downscaled on a daily basis if they are to be linked with a crop model. While methods exist for downscaling rainfall probabilities (Han et al., 2017), downscaling temperature probabilities from SCF are not readily available. Temperature is critical for predicting phenology. Growing Degree Days (GDD), a temperature - based index of thermal time, is frequently used to estimate crop growth and development rates. The U2U GDD Tool predicts silking and maturity dates based on GDD and is used by Midwest maize growers (Prokopy et al., 2017). It provides a projection of GDD 30 days in advance at the seasonal horizon (3 months). However, the tool only uses historical and forecast temperature (30 - days). Moreover, it does not 5 use seasonal climate fore cast (3 - month) and hence cannot predict silking with a lead - time of more than one month. Knowledge Gap 3: Lack of downscaling tool for seasonal rainfall and temperature forecast to predict anthesis and risk assessment model to manage ear rot disease in ma ize. To address knowledge gap 3 , the following objectives were developed: (1) to predict phenology using the calibrated CERES - Maize and downscaled seasonal climate forecast using FResamplePT tool; and (2) to develop and validate the risk analysis model for ear rot disease management in maize using seasonal climate forecast. This dissertation will integrate seasonal climate predictions, remote sensing and crop models to address, advance our understanding and bridge lack of the three knowledge gaps m entioned above. 6 2. INTRODUCTION TO METHODOLOGY AND RESULTS This dissertation is a compendium of three research studies that are co - dependent to address some pressing issues on the agronomic management of corn in Michigan under irrigated and rainfed conditions. The general methodology revolves around the prediction of phenology using a calibrated crop model (CERES - Maize) (Fig. 2.1). The calibrated crop model in the first study was used to develop region specific crop coefficients in the second study for improving irrigation management in corn. Innovative evapotransp iration (ET) estimation and ensembling techniques were used and developed in the process. In the third study, the calibrated crop model was used to predict anthesis using seasonal climate forecasts (SCF). Predicted anthesis from the calibrated crop model w as used to drilldown on the critical period for identifying ear rot disease risks. - Maize to simulate ficients of CERES - Maize and evaluated their performance in simulating phenology and yields. The calibration methods include GENCALC, GLUE and NMCGA. More robust crop genetic coefficients were estimated based on ensembling the parameters derived from the i ndividual methods. Validation of the calibrated genetic coefficients however revealed that soil model structure under rainfed/stressed conditions must be adjusted to allow the root system of hybrid maize to explore available water and nutrient in the soil profile. 7 Figure 2. 1 . The general flowchart for the methodology of this dissertation. evapotranspiration for the development of crop coefficients of mai evaluated multiple evapotranspiration (ET) models from satellite - based model (SEBAL, Priestley - Taylor - JPL and MODIS Penman - Monteith) to ground - based empirical model (FAO - kc) using eddy - covariance flux towers. The calibrated CERES - Maize in the first study was used to simulate crop growth stages, which were used as bases of ensembling ET from different methods using a stage - based Inverse Distance Weightage (IDW) approach. Ensembled ET was used to derive region specific crop coe fficients for maize irrigation scheduling in Southwest Michigan. 8 - based seasonal c limate forecasts (SCF, rainfall and temperature) from NOAA and calibrated CERES - Maize model to predict anthesis to inform when would be the critical period for evaluating ear rot risks in maize. An innovative SCF downscaling technique for rainfall (P) and temperature (T) was introduced, which was used to develop risk analysis model for ear rot disease management in maize. The risk analysis model showed an enhanced ear rot risks in 2018 relative to climatological risks. 9 3. ESTIMATING GENETIC COEFFICIENTS OF CERES - MAIZE TO SIMULATE PHENOLOGY AND YIELD OF MAIZE IN MICHIGAN 3.1 Introduction Maize ( Zea mays L .) is a major crop grown in the Midwestern US. The economic incentives associated with the ethanol mandate and the biotechnological revolutio n in developing transgenic crops encouraged the expansion of maize acreage (Fausti, 2015). Michigan maize farmers harvested a total production of 6.5 million tons with 10 tons ha - 1 average yield on 6.63% compared to 11 tons ha - 1 nationa l average yield. (USDA - NASS , 2017). Most of the maize grown in the state is rainfed and used for biofuel and animal feeds. However, the lower part of southwest Michigan, having fertile and san dy soils, grow hybrid seed under intensive irrigation. To meet the growing demand, the genetic performance of hybrid maize must be improved and agronomic practices, optimized. Varietal improvements and development however require a lot of time and resources. Since the mid - 20 th cent ury, yield increments in maize have been attributed ~50 - 60% to improved genetics, and ~40 - 50 % to management practices (Duvick, 2005; Kucharik and Ramankutty, 2005; Lee and Tollenaar, 2007; Egli, 2008; and Sacks and Kucharik, 2011). The impacts of climate change like, longer growing period and more summer rainfall in the last few decades, benefited maize production through greater accumulation of photosynthates (Andresen et al., 2001, Lobell and Asner, 2003, Twine and Kucharik, 2009). Plant breeders account for these physiological gains due to climate change in the process of developing modern hybrids. Prior to the release of advanced cultivars, these are grown over several years at various locations to evaluate the Genotype x Environment interactions (G x E ; Elias et al., 2016; Kleinknecht et al., 2016; Paderewski et al., 2016; van Eeuwijk et al., 2016 and Yan, 2016). Although plant breeders 10 have leveraged these interactions, within seasonal variability of weather plays a challenging role. Sound agronomic ma nagement within the season can help overcome these challenges and minimize yield gap (i.e., difference of potential and actual yields). However, improving management and cultivar development by traditional agronomic research methods are constrained by time (Vilayvong et al., 2015). Crop simulation models can facilitate the cultivar development process by virtual extrapolations of field experiments at multiple locations and seasons (Lobell et al., 2009). Crop simulation models have been used in studying grow th and development of crops under different environments (White and Hoogenboom, 2010; Asseng et al., 2013; MacCarthy et al., 2017 and Jha et al., 2018). Crop growth models integrate knowledge in soil science, plant physiology, micrometeorology and agronomy to simulate crop performance (van Ittersum et al., 2003; Yin et al., 2004; Löffler et al., 2005 and Cooper et al., 2009). Crop cultivar specific traits control the interactions of environmental factors (e.g., temperature, daylength, light) and plant proce (White and Hoogenboom, 1996; Boote et al., 2003; Hoogenboom et al., 2004). The first step in the applications of crop models is the estimation of these genetic coefficients (Wallach et al., 2001; Jones et al., 2011). They can be estimated by extensive and exhaustive field and laboratory experiments (Du Toit, 2002; Suriharn et al., 2 007). However, this process can be hastened and simplified by using carefully designed model calibration and validation, and was the focus of this study. The Decision Support System for Agro - technology Transfer (DSSAT) is a globally accepted decision suppo rt tool. (Tsuji, 1998; Jones et al., 2003 and Hoogenboom et al., 2013). 11 The Cropping System Model (CSM) in DSSAT is a suite of models, which incorporates more than 40 crop models, integrates soil, weather and management for simulating growth and developmen t of crops (Hoogenboom et al., 2013). The CERES - Maize (Jones and Kiniry, 1986) in CSM is the most widely used maize model for the assessment of management practices on maize production. Due to advances in crop physiology, climatology and agronomy, CERES - Ma ize has evolved over time. It was adopted in APSIM (Keating et al., 2003) by revising the soil organic pool. Lizaso et al. (2011) have developed CSM - IXIM by revising codes of CERES - Maize model for carbon assimilation and partitioning. CERES - Maize has also been used to study nitrogen cycle in long term maize production (Liu et al., 2011; Li et al. , 2015 ). CERES - Maize cultivar parameters have been calibrated to optimize irrigation practices in Pakistan (Mubeen et al., 2016), to evaluate management practices i n Ghana and Tanzania (Mourice et al., 2014; MacCarthy et al., 2017), to evaluate climate resilient technologies in Bangladesh (Ahmed et al., 2017), and to design management strategies under data - scarce environments in Nigeria (Adnan et al., 2019), among ot hers. Several approaches have been adopted to estimate crop model parameters. Grimm et al. (1993) used downhill simplex to estimate phenological parameters of soybean cultivars. Simulated annealing was used to estimate soil and root parameters of a soybean model (Calomn et al., 1999; Mavromatis et al., 2002). Hunt et al. (1993) developed GENCALC (Genotype Coefficient Calculator) based on sequential search method. GENCALC estimates the coefficients by multiple iterations of approximate coefficients in a pre - set sequence and compares outputs based on the difference between simulated and observed values (e.g., anthesis and maturity dates, yields). The genotype coefficients are altered until a good model fit is found. (Hunt et al., 1993). The genetic coefficient s of groundnut (Arachis hypogea L.) (Anothai et al., 2008), 12 soybean (Glycine max L.) (Bao et al., 2015), wheat (Triticum aestivum L.) (Ibrahim et al., 2016) and maize (Román - Paoli et al., 2000; Hassanien et al., 2007; Yang et al., 2009; Bao et al., 2017 an d Adnan et al., 2019) have been estimated using GENCALC. However, GENCALC does not estimate uncertainties of the derived parameters (He et al., 2010). A detailed review of methods for calibrating model parameters, and discussions about the past, present an d future of model calibrations can be found in Siedel et al. (2018). Model outputs are prone to errors due to uncertainties in data inputs and model parameters. It is unrealistic to conclude that one set of parameters represents the model behaviour; rathe r it is better to assess likelihood weights of the parameters, which can better predict the model behaviour (Beven and Binley, 1992). A Bayesian framework that assesses uncertainty of parameters using Monte Carlo technique, called Generalized Likelihood Un Mertens et al., 2004; Candela et al., 2005; He et al., 2009). It uses observed data to develop prior parameter distributions (He, 2008). The posterior distribution is comput theorem (Makowski et al., 2006). GLUE has been used in the field of hydrology and crop sciences for parameter estimation (e.g., Yan et al., 2017; Beven, 2018; Sikorska and Seibert, 2018; He et al., 2009; He et al., 2010; López - Cruz, 2016 ; Sun et al., 2016). Several works have compared GENCALC and GLUE in wheat, rice and maize calibrations but did not find any significant differences in performance between the methods (e.g., Ibrahim et al., 2016; Budhhaboon et al., 2018; Adnan et al., 2019 ). Genetic Algorithm (GA; Goldberg, 1989) also has been used in crop model calibration, sometimes outperforming other gradients or Bayesian - based optimization methods because of its ability to search through vast search spaces (Miller and Goldberg, 1996; Smalley et al., 2000; 13 Gopalakrishnan et al., 2003; Wu et al., 2006). Pabico et al. (1999) used GA to determine genetic coefficients of cultivars, mapping coefficients as a single chromosome. The Noisy Monte Carlo Genetic Algorithm (NMCGA; Ines and Mohanty, 2008a) evaluates realizations of strings (analogous to chromosome) of model parameters with given distributions using Monte Carlo approach (Wang, 1991; Ines and Droogers, 2002; Pabico, 2007; Ines and Mohanty, 2008b; Dai et al., 2009). In each generation, fitter chromosomes (least difference between observed and predicted) are selected; undergo crossover and mutation until a solution is achieved. Shin et al. (2013) used NMCGA to estimate soil hydraulic parameters of an agro - hydrological model, SWAP (Van Dam ., 2000). Estimating parameters using individual methods always possess some levels of uncertainty due to errors in initial conditions, data and model structure. Ensemble - based parameter estimation methods can improve accuracy and account for different so urces of uncertainties in model calibr ation (Chen et al., 2015) and provide more robust estimates of model parameters ( Vrugt and Robinson, 2007). Parameters estimated by one method can better predict phenology, others can better predict growth. Ensembling parameters derived from different methods can account for the biases associated with the methods, and can better predict phenology and yields (Jha et al., 2019, in review). In order to achieve robust genetic coefficients, which can simulate phenology and y ield of maize in Michigan, we designed this study to ensemble genetic coefficients from multiple methods and evaluate the model performance. The specific objectives of the study are: (i) to calibrate genetic coefficients of a maize hybrid in Michigan usi ng multiple methods, and (ii) to validate CERES - Maize model for simulating maize hybrid phenology and yields in Michigan. In this study, we used three methods for estimating genetic coefficients of CERES - Maize namely, GENCALC, GLUE and NMCGA, 14 employed an e nsembling approach, and assessed their performance. We used maize performance trial locations in Michigan in the calibration and validation. Irrigated locations were used for calibration and rainfed locations for validation. 3.2 Materials and Methods 3.2.1 Study area Data for this study were collected from the field experiments conducted at Michigan Maize Performance Trial locations (MMPT, Singh et al., 2018). Performances of commercia l maize hybrids are evaluated at MMPT locations annually , which lie s in different zones ranging from south to north across Michigan (Fig. 3.1). Climatic conditions are similar within each zone and consisted of three trial locations in each zone. Figure 3. 1 . Locations of maize performance tri als in Michigan. These zones were established based on long - term accumulated growing degree - days (GDD). The 30 - year (1981 - 2010) normal accumulated GDD F from May 1 to October 31 were 2557 o o o GDD for maize growth and development are calculated by deducting the base temperature for maize growth (50 F/10 C) (Cross and Zuber, 1972; Stewart et al., 1998) from the average air 15 temperature in a 24 - hour period , starting from the emergence date (Abendroth et al., 2010; Angel et al., 2017). The upper threshold for optimum growth in maize is considered (86 F/30 C) for calculating average temperature. It means that whenever air temperature goes beyond 86 F / 30 C, the daily maximum temperature has to be set equal to 86 F/30 C. Based on heat accumulation, seed companies provide information of relative maturity to understand the crop maturity period (planting to physiological maturity) with specified GDD n umbers from planting to anthesis and to maturity, respectively. The GDD estimates are based on F for the hybrids in the US and for this study. However, genetic coefficients which controls growth and development are based on C. GDD F can be converted into GDD C by dividing the final value of GDD F by 1.8 (Abendroth et al., 2010). The zones, coordinates, major soil types and management practices of the field trials used in this study are given in Table 3.1. Table 3. 1 . Location of t h e maize fields under Michigan M aize Performance Trials (MMPT). Name of County (Town) Zone Coordinates Major Soil Type a Management Practice (Rainfed/Irrigated) Washtenaw (Milan) 1 42.15 N, 83.56 W Sandy Loam Rainfed Branch (Coldwater) 1 41.97 N, 85.08 W Sandy Loam Irrigated Cass (Vandalia) 1 41.86 N, 85.88 W Loam Irrigated Allegan (Martin) 2 42.57 N, 85.63 W Loam Rainfed Ingham (Williamston) 2 42.71 N, 84.47 W Loam Rainfed Saginaw (New Lothrop) 2 43.13 N, 83.97 W Loam Rainfed Huron (Bad Axe) 3 43.83 N, 82.98 W Sandy Loam Rainfed Montcalm (Greenville) 3 43.22 N, 85.21 W Sandy Loam Rainfed Mason (Scottville) 3 43.98 N, 86.15 W Loam Irrigated a Source: WSS - SSURGO Database . 16 3.2.2 CERES - Maize mode l CERES - Maize is a crop module embedded within the suite of Cropping System Model (CSM) in DSSAT (Jones and Kiniry, 1986; Boote et al., 2010; Hoogenboom et al., 2013). DSSAT facilitates the assessment and evaluation of different management practices on growth and development of c rops with a goal of enhancing current knowledge of Genotype X Environment X Management interactions (Boote et al., 2010; Elias et al., 2016). CERES - Maize, a Fortran based process - oriented model, utilizes daily weather data to simulate crop growth stages on a daily basis, integrating soil water and nitrogen balance associated with maize growth. Therefore, CSM in DSSAT integrates the interaction and effects of climate, soil and other management practices, which can be used to predict/assess their impacts on c rop growth and development in the past, present and future (Lobell et al., 2009). Based on heat accumulation and photoperiod, the model assumes that the rate of development increases linearly above a base temperature (10 C) until 34 C and decreases linear ly from 34 to 44 C , which are governed by genetic coefficients (P1, P2, P5, PHINT; Table 3.3). The phenological development also accounts for the process of morphological development of leaves, stems and roots resulting in biomass accumulation and partitio ning. CERES - Maize growth progresses through phenological stages. The simulated growth stages are sequential, starting from emergence to end of juvenile phase, followed by peak vegetative growth, which culminates at tassel initiation. There is a small trans ition phase from vegetative to reproductive stage, which starts from tassel initiation and ends at anthesis. The last two stages in the crop season are anthesis to start of grain filling, followed by effective grain filling period. These stages coincide wi th high water and nutrient demands, which translate biomass partitioning into grain filling. 17 temperature and daylength ( Kiniry and Bonhomme, 1991). 3.2.3 Input data for CERES - Maize 3.2.3.1 . Weather data Weather data were collected from MSU Enviro - weather Network, a weather - based information system that helps growers and stakeholders in making farm related decisions in Michigan. Daily weather data (rainfall, minimum and maximum temperature, solar radiation) for the 2017 and 2018 cropping seasons (May - October) are shown in Fig. 3.2 and 3.3 respectively. All the weather data were ingested in the weather database of DSSAT using WeatherMan, which converts data to DSSAT weather format (Pickering et al., 1994). Average seasonal values of total rainfall, solar radiation, maximum and minimum temperatures are given in Table 3.2. Except for Huron, all the stations had greater total solar radiation in 2017 growing season than 2018, which could have benef ited the crops to better accumulate their potential biomass productions in 2017. The rainfall observation also suggests that 2018 growing season was wetter than 2017. Table 3. 2 . Summary of weath er data (2017 and 2018) during the crop season (May - October) for selected locations. Locations Zones Average Tmax ( C) Average Tmin ( C) Total Rainfall (mm) Average Solar Radiation (MJ m - 2 day - 1 ) 2017 2018 2017 2018 2017 2018 2017 2018 Washtenaw 1 25.04 25.51 11.86 13.03 360.4 406.0 18.03 16.72 Branch 1 23.39 25.35 9.82 13.23 428.2 573.6 16.36 14.25 Cass 1 23.92 24.80 12.40 13.90 585.8 572.2 16.47 14.99 Allegan 2 24.20 24.72 11.67 13.20 593.6 679.7 18.69 17.03 Ingham 2 23.45 23.93 11.48 12.11 389.2 464.8 16.96 15.78 Saginaw 2 24.01 24.62 12.20 13.15 466.9 493.1 17.54 17.09 Huron 3 22.76 23.27 11.53 11.81 477.4 348.7 16.36 16.67 Montcalm 3 23.18 23.59 10.35 11.37 485.5 656.6 16.22 15.63 Mason 3 21.94 22.59 10.72 11.40 528.6 539.4 18.38 17.68 18 3.2.3 .2 So il data and agronomic management All nine locations were categorized under two major soil types i.e., loam and sandy loam, based on major soil type according to WSS - SSURGO database (Table 3.1). Initial conditions for sandy loam soil (volumetric water = 0.22 cm 3 /cm 3 ; Soil - N (NH 4+ ) = 3 g(N) / Mg (Soil) and Soil - N (NO 3 - ) = 5 g(N) / Mg (Soil)) and for loam soil (volumetric water = 0.40 cm 3 /cm 3 ; Soil - N (NH 4+ ) = 5 g(N) / Mg (Soil) and Soil - N (NO 3 - ) = 6 g(N) / Mg (Soil)) were kept in the medium range (Rutan and Steinke, 2017). Field data like fertilizer applications, planting and anthesis (75% of silking) dates and yields are given in Table 3.3. However, due to limited data availability on maturity (blac k layer), dates they were estimated based on degree - days accumulation for 2500 GDD F using U2U tool (Angel et al., 2017) for 2017 and 2018 for all locations (Table 3.1). The split doses of nitrogen were applied uni formly at each location, first 2 5% of N (as Urea) was applied during planting and 75% of remaining N was applied 30 - 40 days after planting as urea ammonium nitrate solution during V6 to V8 stages. Farm Yard Manure (FYM) was applied one week before planting at Allegan, Ingham, Huron and Mason sites. The harvested yields were estimated from the centre two rows from the p lots size of four rows with 6.7 m length and a row spacing of 0.76 m. Those rows were harvested with a Kincaid 8 - XP plot combine after physiological maturity to collect data on grain y ield and moisture content. The final yields were estimated on dry basis (adjusted to 0% moisture) . 19 Table 3. 3 . Descriptions of field observations used in the study. *Estimated from U2U tool ( https://mrcc.illinois.edu/U2U/gdd/ ); FYM - Farm Yard Manure Name of Location Year Fertilizer & Manure Application (N - P - K; Kg.ha - 1 ) Planting Date Anthesis (DAP) Maturity * (DAP) Yield (Kg.ha - 1 ) Washtenaw 2017 206 - 10 - 3 15 - May 72 132 10984 2018 206 - 10 - 3 12 - June 60 120 9723 Branch 2017 248 - 10 - 3 29 - May 72 139 12477 2018 213 - 10 - 3 29 - May 66 116 8698 Cass 2017 269 - 10 - 3 14 - May 76 138 12688 2018 274 - 10 - 3 09 - May 67 129 9041 Allegan 2017 122 - 10 - 3 + 20 ton FYM 12 - May 70 130 13391 2018 179 - 10 - 3 + 10 ton FYM 18 - May 62 114 9769 Ingham 2017 181 - 10 - 3 + 10 ton FYM 18 - May 69 138 11010 2018 179 - 10 - 3 + 10 ton FYM 08 - May 71 123 8197 Saginaw 2017 172 - 10 - 3 29 - May 66 131 10451 2018 179 - 10 - 3 30 - May 65 116 9297 Huron 2017 142 - 10 - 3 + 15 tons FYM 17 - May 78 145 11034 2018 179 - 10 - 3 + 10 tons FYM 16 - May 72 128 8245 Montcalm 2017 172 - 10 - 3 16 - May 79 155 10820 2018 179 - 10 - 3 18 - May 64 124 8973 Mason 2017 122 - 10 - 3 + 20 tons FYM 10 - May 73 141 13578 2018 179 - 10 - 3 + 10 tons FYM 23 - May 70 147 8951 20 Figure 3. 2 . Weather data (Rainfall, solar radiation, maximum and minimum temperature)for cropping season (May - Octob er) for 2017 (Source: MSU Enviroweather) Figure 3. 3 . Weather data (Rainfall, solar radiation, maximum and minimum temperature) for cropping season (May - October) for 2018. (Source: MSU Enviroweather) 21 3.2.4 Calibration methods 3.2.4 .1. Genetic coefficients Maize hybrid of Comparative Relative Maturity (CRM) group 103 (Lauer, 1998) that requires around 2500 GDD from planting to physiological maturity was selected for the calibration. However, these specificat ions of GDD requirements for physiological maturity changes with different hybrid seed companies. Calibrations were conducted using all three irrigated locations (Branch, Cass and Mason) out of total nine varietal trial locations during 2017 and 2018 growi ng seasons. Maize grown at irrigated locations were selected in the calibration, as the best conditions (water and nutrients) are required for the calibration process (Lobell et al., 2009). The starting cultivar selected for GENCALC was PC0001 (2500 - 2600 G DD) from the DSSAT database (DSSAT V4.6: Hoogenboom et al., 2013), which is suitable for Michigan based on GDD accumulation. The descriptions of genetic coefficients in CERES - Maize are listed in Table 3.4. P1 is the thermal time needed for phase change f rom germination to end of juvenile, when photoperiod does not have any significant role in phasic development, expressed in degree - days (Jones and Kiniry, 1986). P2 is expressed in days for delay when photoperiod is increased each hour above the longest ph otoperiod (12.5 hr) having maximum development rate (Hanks and Ritchie, 1991). Tassel initiation is controlled by both P1 and P2 (Román - Paoli et al., 2000). Tassel initiation starts approximately 4 days after P1 completion (Jones and Kiniry, 1986). P5 repr esents thermal time from anthesis to physiological maturity and expressed in degree - days. G2 is the maximum number of kernel a plant can grow potentially. Du Toit (2002) explained how fitted coefficients (estimated) have differences in model predictions co mpared with determined coefficients from the field after rigorous multi - year experiments. G3 controls the kernel - filling rate during post 22 anthesis to maturity phase and expressed in mg day - 1 . P1 and P2 determine anthesis, and P2 and P5 determine maturity d ates, while P5, G2, G3 and PHINT control the yield and its components e.g., dry matter, grain size and canopy weight. PHINT controls phenology and growth as well, through determination of the timing to leaf tip appearance (Hammad et al., 2018). PHINT was k ept fixed in the calibration, while the other five parameters described above were calibrated. It is recommended not to change PHINT unless sufficient field data for leaf numbers are available (DSSAT v 4.6; Hoogenboom et al., 2013). These parameters deter mine the phasic development of a maize cultivar, and these developments are attributed to genetic variations among the cultivars, hence called genetic coefficients. Table 3. 4 . Descriptions of genetic coefficients in CERES - Maize. Genetic coefficient Description PC0001 (2500 - 2600 GDD) (Starting cultivar for GENCALC) Phenology Coefficient Pl Juvenile phase coefficient, °C - d 168.9 P2 Photoperiod sensitivity coefficient, days 0.734 P5 Grain - filling duration coefficient, °C - d 780.0 Growth Coefficient G2 Potential kernel number coefficient 750.0 G3 Kernel filling rate, mg/day 8.50 PHINT Phylochron interval, °C - d 32.33 (fixed in calibration) 3.2.4 .2. GENCALC: Genetic Calculator GENCALC is a built - in software in DSSAT, which estimates genetic coefficients using a gradient search method (Fig. 3.4; Hunt et al., 1993; Adman, 2019). With the pre - defined set of experiments and starting cultivar coefficients of a selected maize variety, it runs CERES - Maize iteratively to search for the best parameter estimates. With an in i tial value for each parameter, it adjusts genetic coefficients until it fits to the provided observed value within the range of their 23 physiological characterist ics i.e., flowering date, maturity date, daylenth, kernel size, etc. (Hunt et al., 1993). The algorithm in the software exploits a known point in the search space that depends on a starting point (i.e., coefficients of starting cultivar) and based on the d ifferences between simulated and observed values, it adjust the coefficients. Figure 3. 4 . GENCALC - CERES - Maize working flowchart (Adnan et al., 2019). It minimizes the error between simulated and observed values in each run (Hunt et al., 1993). Because of a small sampling area of the search space, the final coefficients cannot be optimized for large ranges of physiological characteristics (Pabico et al., 1999). Pre - defined govern the sequence of genotype coefficient calculation), govern the search until the best fit to each observation is found. After multiple iterations, the parameter set that gives the best fit with observed anthesis date (ADAP), maturity date (MDAP) and yield (HWAM) is chosen. GENCALC estimates genetic coefficients 24 in two steps, optimizing the phenology parameters (P1, P2, and P5 ) first, and then the growth parameters (G2, G3). The phenological development depends on degree - day accumulation and growth dep ends on phenology (Fig. 3.4). Hence, it is logical that phenology parameter is derived first and then growth parameters. GENCALC does not estimate uncertainties of parameters. 3.2.4 .3. Generalized Likelihood Uncertainty Estimate: GLUE GLUE estimates parameters using a Bayesian approach. GLUE first develops the prior parameter distributions using genetic coefficients from the DSSAT database (Hoogenboom et al., 2013) by fitting them to a multivariate normal distribution, and then estimates the posterior , (1) constant. To calculate likelihood values, random parameter sets i are generated from the prior distributions. The more the number of parameters set realizations, the more stable results can be obtai ned (He et al., 2010). For stability in results, we selected 30,000 runs for GLUE. A likelihood value L [ i |O] for each observation (anthesis date, maturity date and yield) is estimated based on Gaussian likelihood function (Eq. 2) (He et al., 2010). (2) i is the i th parameter set, M is the number of observations, O j is the j th o 2 is i ) is the output of the model. In addition, Eq. 3 calculates the probability of the parameter set , 25 (3) i ) is the probability or likelihood weight of the i th i i |O) is the i , given observations O (He et al., 2010) The empirical posterior distributions were constructed from the pairs of parameter set and i i calculated as in Eqs. 4, 5 (He et al., 2010) ; (4) (5) where µ post 2 post i ) is the probability of the i th parameter set. The parameter estimation in GLUE follows a similar step as in GENCALC i.e., estimate first the phenology parameters (P1, P2, P5) and then the growth parameters (G2, G3). The parameter set that gives the maximum likelihood value is selected. 3.2.4 .4. Noisy Mont e Carlo Genetic Algorithm: NMCGA There are multiple methods of parameter estimation, each has its own advantages and disadvantages. Ensemble - based parameter estimation methods tend to improve model accuracy and accounts for different sources of uncertainty, thus provide more robust estimates of model parameters ( Vrugt and Robinson, 2007). A long with GENCALC and GLUE, we also employed t he Noisy Monte Carlo Genetic Algorithm (NMCGA; Ines and Mohanty, 2008a) to estimate maize genetic coefficients. Here, genetic algorithm (GA) estimates combination of parameters (i.e., means and standard deviations) and evaluate their fitness. Based on a - pr iori distributions and a - priori range of parameter values from DSSAT cultivar database, parameter set are 26 evaluated using Monte Carlo resampling. Resampled parameters sets are then passed to CERES - Maize to evaluate the fitness of that parameter set. The f itness of the parameters are tested by evaluating the difference between simulated and observed values. NMCGA, being a noisy GA, evaluates the fitness of a parameter set under a noisy fitness space (Wu et al., 2006), thus an overall fitness of a parameter set is evaluated from the average fitness of several ensemble runs from parameter set realizations. The fittest parameters are selected and allowed to reproduce for multiple generations undergoing crossovers and mutations until an optimal solution is achie ved. For consistency, we also employed a two - step parameter estimation technique like in GENCALC and GLUE i.e., estimating phenology parameters first, then growth parameters. The coefficients (first and second moments) of phenology (P1, P2 and P5) and grow th (G2 and G3) were arranged as a set of genes in a chromosome during those steps, respectively. The objective function of the parameter set for the i th ensemble is formulated as Eq. 6; (6) where, K r is set of K parameters combinations with r realizations generated from Monte Carlo resampling and N resample is the total number of realizations for simulated (Sim (K r )) and observed varia bles (Obs t ), t i is running index for time T (Ines and Mohanty, 2008a). Noisy fitness is calculated using the inverse of the modified - penalty approach of Hilton and Culver (2000) (Eqs. 7, 8); (7) (8) where, p* is the chromosome, and fitness(p*) is the noisy fitness of that chromosome sampled from each ensemble i of the Monte Carlo resampling. A chromosome realization is penalized 27 (Penalty (K)) if its predicted variables violate some preset rules agains t the goodness - of - fit evaluation (Ines and Mohanty, 2008a). Sampling fitness is calculated based on Eq. 9 to reduce the noise in fitness; (9) where, R is total number of ensemble i. The arrays of pa rameters set (chromosome) of means and standard deviations, undergo through the search process until the best chromosome is generated. When calibrating for phenology, ADAP and MDAP were given the same weights while HWAM was not used in the objective funct ion (Eq. 6). When calibrating for growth, HWAM was used in the objective function while ADAP and MDAP were not. Moreover, we ran NMCGA in two ways, one by estimating only the means of parameters (NMCGA_NO_SD), and other by estimating both the means and sta ndard deviations of the parameters (NMCGA_SD). The representations of p* as used in this study are given in Table 3.5. 28 Table 3. 5 . Representations of phenology and growth parameters in NMCGA. NMCGA_SD NMCGA_NO_SD Paramet er Minimu m Values Maximu m Values Number of Bits (L) 2 L Minimu m Values Maximu m Values Number of Bits (L) 2 L Genetic Algorithm Variables µ (P1) 5 458 8 25 6 110 458 8 25 6 (P1) 0 67.56 8 25 6 0 0 2 4 µ (P2) 0 2 8 25 6 0 2 8 25 6 (P2) 0 0.28 8 25 6 0 0 2 4 µ (P5) 390 1035 8 25 6 390 1035 8 25 6 (P5) 0 117.85 8 25 6 0 0 2 4 µ (G2) 248 1170 8 25 6 248 1170 8 25 6 (G2) 0 161.63 8 25 6 0 0 2 4 µ (G3) 4.8 16.5 8 25 6 4.8 16.5 8 25 6 (G3) 0 1.76 8 25 6 0 0 2 4 Monte Carlo Variables Pl 110 458 110 458 P2 0 2 0 2 P5 390 1035 390 1035 G2 248 1170 248 1170 G3 4.8 16.5 4.8 16.5 3.2.5 . Ensembling approach Along with the comparisons of GENCALC, GLUE and two variants of NMCGA, we evaluated an ensembling approach of estimating crop model parameters. The purpose of ensembling was to integrate the strengths of the four methods with a goal of achieving a more robust set of gene tic coefficients. The general framework is shown in Eqs. 10 - 12; (10) 29 (11) (12) where P j is the ensembled value of a phenology or growth parameter, W i is the weight of a parameter from method i, p ij is the parameter value of a phenology or growth parameter from method i, j is an index of a phenology or growth parameter. w i is the inverse squa red - distance between the predicted (A pij,k ) and observed (A o,k ) variable(s) (e.g., ADAP, MDAP or HWAM) substantially impacted by that phenology or growth parameter. N is the number of methods (here, N=4), k is an index for predicted or observed variable(s) , K is the number of variable(s) impacted by p ij and is the weight for that variable, k ( Jha et al., 2019, in review). However, since HWAM and MDAP have different units, Eq. 12 is transformed to Eq. 13. (13) For arithmetic average, W i 3.2.6 Validation and statistical analysis Before validation, the calibrated model was used to estimate potential productions (no water and nutrient stresses) for all locations to analyze the genetic potentials of the calibrated cultivar in those locations. All rainfed locations (Washtenaw, Allegan, Ingham, Saginaw, Huron and Montcalm) were used for validation (Table 3.1). Genetic coefficients derived from GENCALC, GLUE, two variant s of NMCGA, and the ensembling approach were used to validate CERES - Maize under rainfed conditions. Validations were done for both growing seasons (2017 and 2018). We compared predicted and observed ADAP, MDAP and HWAM and used the coefficient of determin ation (R 2 ; Eq. 14), Mean Bias Error (MBE; Eq. 15), Root Mean Square 30 Error (RMSE; Eq. 16) and Index of Agreement (d - index; Eq. 17) (Willmott, 1982) to measure the performances of the calibration methods, (14) (15) (16) (17) where M and O are simulated and observed variables (e.g., ADAP MDAP or HWAM), respectively. 3.3. Results and Discussion 3.3.1 Calibration Initially, we calibrated CERES - Maize using the irrigated locations (Branch, Cass and Mason) only in 2017. Well - watered and fertilized conditions are suggested for calibration purposes (Grassini et al., 2015). According to GDD requirement, the maize hybrid (PC0001 - 2500 - 2600 GDD) (Cultivar code for CERES - Maize; Ho ogenboom et al., 2013) selected for starting the search of parameters (GENCALC only) is suitable for the study area (Prokopy et al., 2017). Table 3.6 shows the calibrated genetic coefficients using only one cropping season (2017) in the calibration. Based on Tables 3.4 and 3.6, GENCALC only changed the values of P1 and P2, respectively. Other coefficients from PC0001 - 2500 - 2600 GDD did not change. In GLUE, the calibrated value of P1 was reduced compared to the selected maize hybrid, which suggests that the j uvenile stage should reach three to four days earlier and hence anthesis. 31 However, NMCGA (SD and NO_SD) estimated an increased P1 value, which suggests that the juvenile stage needs more degree - days to complete hence anthesis is delayed. However, a lower P2 value was estimated, which signifies that any delay in anthesis is compensated as there is little delay in the developmental process if photoperiod is increased above the maximum physiological limit of 12.5 hrs (Jones and Kiniry, 1986; Hanks and Ritchie , 1991 and Román - Paoli et al., 2000). For all methods, it was observed that there are low estimates of P5, which signify that the calibrated cultivar requires lesser thermal time from anthesis to physiological maturity compared to PC0001 ( 2500 - 2600 GDD ) . G 2 values are relatively low especially for NMCGA_NO_SD. Modern hybrids should have a maximum kernel number as close to 800 (Du Toit, 2002). For a more objective comparison, we evaluated CERES - Maize performance using the calibrated coefficients. Table 3. 6 . Genetic coefficients estimated by GENCALC, GLUE, NMCGA and Ensembling approach in 2017. Methods P1 P2 P5 G2 G3 GENCALC 168.3 0.539 780.0 750.0 8.50 GLUE 147.2 (33.796) 1.011 (0.638) 638.4 (33.142) 701.73 (188.74) 12.28 (2.749) NMCGA_SD 199.1 (9.651) 0.347 (0.238) 574.2 (84.179) 774.9 (23.090) 11.49 (0.251) NMCGA_NO_SD 199.4 0.289 666.5 579.7 16.50 Arithmetic Average 178.5 0.546 664.8 701.6 12.19 Weighted Average 185.2 0.560 650.9 719.3 12.07 Note: values in parentheses are standard de viations of parameter estimates 32 Table 3. 7 . Performance of CERES - Maize using estimated parameters by GENCALC, GLUE, NMCGA and ensembling approach during calibration in 2017. Table 3.7 shows that all methods performed well in simulating ADAP). However, most of the methods did not perform well in simulating MDAP and yield. This modest performance could be attributed to the length of the data used in the calibration. Development of genetic coefficients of a cultivar for a specific agro - climatic environment requires adequate data from multiple locations and cropping seasons to include environmental variability to the genotype, environment and management interactions (Kersebaum et a l., 2015; He et al., 2017). In order to obtain representative parameter estimates, it is better to use multiple years of calibration data (Bulatewicz et al., 2009; Confalonieri et al., 2016; Seidel et al., 2018). In order to improve the calibration, parame ter uncertainty has to be minimized, which can be done by reducing uncertainty in input like soil properties, initialization variables and management practices (Wallach et al., 2012; Dzotsi et al., 2015; Varella et al., 2012; Roux et al., 2014; Waha et al ., 2015). We then re - run the calibrations using 2017 and 2018 data from the irrigated locations. The genetic coefficients calibrated from the two - year datasets are given in Table 3.8. CERES - Maize per formance is shown in Table 3.9. ADAP (Day) MDAP (Day) HWAM (Kg.ha - 1 ) MBE RMSE d - index MBE RMSE d - index MBE RMSE d - index GENCALC - 0.33 2.89 0.78 6.67 11.27 0.37 360.33 552.17 0.72 GLUE - 1.33 3.87 0.89 5.83 11.38 0.39 961.00 1514.28 0.32 NMCGA_SD - 1.17 4.06 0.86 4.00 8.74 0.40 29.00 361.31 0.70 NMCGA_NO_SD - 1.17 4.06 0.86 - 1.17 4.10 0.72 492.50 935.61 0.48 Arithmetic Average - 1.00 3.51 0.80 3.00 6.24 0.47 803.17 1246.14 0.40 Weighted Average - 1.00 3.51 0.80 3.00 6.24 0.47 424.50 635.70 0.68 33 Table 3. 8 . Genetic coefficients estimated by GENCALC, GLUE, NMCGA_SD, NMCGA_NO_SD and Ensembling approach in 2017 and 2018. Methods P1 P2 P5 G2 G3 GENCALC 143.8 0.539 780.0 750.0 8.50 GLUE 133.2 (30.673) 1.690 (0.453) 767.5 (40.102) 762.2 (176.24) 12.58 (2.663) NMCGA_SD 134.4 (8.356) 1.714 (0.202) 758.6 (73.134) 779.7 (20.574) 13.15 (0.191) NMCGA_NO_SD 134.9 1.714 666.4 806.6 14.83 Arithmetic Average 136.6 1.414 743.1 774.6 12.27 Weighted Average 137.5 1.414 749.0 777.4 11.37 Note: values in ( ) are standard deviation of parameter estimates 3.3.1.1. Anthesis and physiological maturity: calibration results Overall, CERES - Maize performance in simulating phenology have improved after calibration using two years of data (see Tables 3.7, 3.9). Due to lower P1 values in the new calibration (see Tables 3.6, 3.8), anthesis dates (ADAP) were slightly under - predicted in all the methods, which is reflected by the negative MBE (Table 3.7, 3.9). GENCALC, GLUE, NMCGA_SD and NMCGA_NO_SD under - predicted anthesis. Except for NMCG A _NO_SD, all the methods over - predicted maturity (Table 3.9). After re - calibration however, d - index for ADAP and MDAP have improved substantially for all the methods. Poor optimization of parameters P1, P2 and P5 po ssibly have caused the deviations in predicted phenology. Especially for P5, which determines the growth of the cultivar after anthesis and hence maturity, might not be optimized (see Hanks and Ritchie, 1991; Rom á n - Paoli et al., 2000). P5 represents therma l time from anthesis to physiological maturity and it varies from 700 - 1000 GDD for modern cultivars ( Rom á n - Paoli et al., 2000) . Phenological impacts however are the results of the combined effects of all phenological parameters, hence improper calibration of the phenological parameters will create a slight difference in GDD that can affect anthesis and maturity ( Rom á n - Paoli et al., 2000). Coefficient 34 of determinations (R 2 ) between predicted and observed anthesis dates were found moderate for all the method s (Fig. 3.5). Overall, however, all the methods under - predicted anthesis in 2017 and 2018 at Branch and Cass, but over - predicted at Mason for both years (Table 3.9). This result reflects the inter - annual variability in predicting anthesis. Variability in g enetic expressions depends greatly on the variations in weather parameters, especially solar radiation during the growing season (Lee et al., 2016). Mason has higher amount of accumulated solar radiation than Branch and Cass (Table 3.2). Coefficient of det erminations (R 2 ) for maturity dates were found higher for all methods (Fig. 3.5). Table 3. 9 . Performance of CERES - Maize using parameters estimated by GENCALC, GLUE, NMCGA and ensembling approach during calibration in 2017 and 2 018. ADAP (DOY) MDAP (DOY) HWAM (Kg.ha - 1 ) MBE RMSE d - index MBE RMSE d - index MBE RMSE d - index GENCALC - 0.83 3.98 0.93 5.50 12.19 0.84 603.67 783.64 0.96 GLUE - 1.83 4.74 0.89 4.00 12.42 0.84 1065.1 1093.2 0.93 NMCGA_SD - 1.67 4.90 0.88 2.17 10.06 0.87 80.67 665.09 0.97 NMCGA_NO_SD - 1.67 4.90 0.88 - 7.83 10.86 0.84 594.00 833.42 0.95 Arithmetic Avg. - 2.17 4.45 0.91 - 0.50 8.75 0.90 860.83 1106.8 0.90 Weighted Avg. - 2.17 4.45 0.91 - 0.50 8.75 0.90 801.00 936.96 0.94 3.3.1.2. Yield calibration results The growth parameters G2 and G3 directly control yield whereas P5 and P2 control it indirectly. The reproductive growth stages and yield components (e.g., dry matter, grain size and canopy weight) are controlled by P5, G2, G3 and PHINT and their interactions ( Hammad et al., 2018) . However, G2 is the most critical parameter in predicting yield (Ritchie and Wei, 2000; Du Toit, 2002; Ritchie and Alagarswamy, 2003; Lizaso et al., 2007), which is dependent on the sowing date and associated we ather variability (Zhou et al., 2017). G2 values increased during re - calibration and its impact was manifested in yield improvements in all the methods (Table 3.8, 35 3.9). GENCALC predicted yield with d - index of 0.96 and RMSE of 783.64 kg ha - 1 while GLUE predicted yield with d - index of 0.93 and RMSE of 1093.2 kg ha - 1 . NMCGA_SD outperformed all the methods, with predicted yield d - index of 0.97 and RMSE of 665.09 kg ha - 1 . (Table 3.9; Fig 3.5). Coefficient of determinations (R 2 ) of yield were high showing a strong confidence of the model in the simulation of yields (Fig. 3.5). Yearly variations of ADAP, MDAP and yield were attributed to the variability in weather and other management practices (e.g., Confalonieri et al., 2016; Waha et al., 2015). 3 .3.1.3. Calibration results from ensemble of methods As shown above, the individual methods performed differently in predicting phenology and yield. Some are better in predicting phenology and some are better in predicting yield. However, we wanted to cali brate genetic coefficients that are robust and resil ient. As Vrugt and Robin son (2007) noted the advantage of ensembling models, we performed en sembling of the coefficients, ra n them in CERES - Maize, and compared their performance with the individual method s. We employed weighted and arithmetic averaging to ensemble the coefficients (Section 2.3.5). In the weighted averaging method, weights were assigned to the parameters based on the distance between the predicted and observed variables that they mostly in fluenced e.g., ADAP, MDAP and HWAM. In the initial calibration (i.e., using only 2017 data), ensembled coefficients performed poorly, especially for MDAP and yield, as individual methods. With the re - calibration using 2017 and 2018 data, their performance improved significantly, outperforming some of the methods (Table 3.7, 3.9). Overall, NMCGA_SD performed best in predicting yield (d - index = 0.97) (Table 3.8). However, weighted averaging performed relatively better in predicting phenology (anthesis and mat urity) and comparable in predicting yield (Table 3.8). Arithmetic 36 averaging of the coefficients also performed well. Our previous study calibrating rice varieties in the Philippines suggested a better performance of arithmetic averaging (Jha et al., 2019, in review). Nevertheless, these results corroborate the value of using multiple methods in crop model calibration and ensembling the derived parameters for better performance. 37 Figure 3. 5 . CERES - Maize performance after calibration of GENCALC, GLUE, NMCGA_SD, NMCGA_NO_SD, arithmetic and weighted averaging for 2017 and 2018. 38 3.3.2 Potential yields from calibrated genetic coefficients The potential yields indicate that the crop model can sim ulate the genotype expression without stress. The potential yield of a crop cultivar is determined by climatic parameters (temperature, solar radiation) and genetic characteristics only (Andrea et al., 2018). Potential yields can be achieved in a given cli mate under non - stressed conditions (water, nutrients, biotic). However, this is hard to achieve in real world conditions that is why agronomists and breeders always aim to optimize agronomic management to reduce the gap between actual (observed) and potent ial yields (Lobell et al., 2009). Fig. 3.6 shows that the genetic potential can attain the levels of observed yields in the rainfed locations (Washtenaw, Allegan, Ingham, Saginaw, Huron and Montcalm). Note that the rainfed (validation) locations were not i ncluded in the calibration. Figure 3. 6 . Potential yields and observed yield for 2017 and 2018 for all maize trial locations using weighted averaging of genetic coefficients. 39 3.3.3 Validation The calibrated crop coefficients were validated at the six - rainfed locations (Washtenaw, Allegan, Ingh am, Saginaw, Huron and Montcalm; see Table 3.1) for two years (2017 and 2018). The representative soil characteristics used in calibration were also used in validation. The validation results for phenology (anthesis and maturity) showed that the model coul d perform well, with d - index varying from 0.84 to 0.95 for all methods (Table 3.10). However, yields were substantially under - predicted, with high MBE (negative) and RMSE, and low d - index (Table 3.10). Table 3. 10 . Performance o f CERES - Maize using parameters estimated by GENCALC, GLUE, NMCGA and ensembling approach during validation at rainfed locations with soil profiles used in calibration. ADAP (DOY) MDAP (DOY) HWAM (Kg.ha - 1 ) MBE RMSE d - index MBE RMSE d - index MBE RMSE d - index GENCALC 0.33 2.89 0.84 6.33 7.96 0.76 - 4446.3 4924.03 0.19 GLUE - 1.00 3.21 0.87 3.33 5.13 0.91 - 3315.5 3877.22 0.26 NMCGA_SD - 0.50 2.42 0.92 3.50 4.60 0.93 - 4400.0 4843.11 0.21 NMCGA_NO_SD - 0.50 2.42 0.92 - 5.83 6.96 0.81 - 3593.0 4053.51 0.26 Ar ithmetic Avg. - 1.17 3.08 0.85 1.00 3.37 0.95 - 3576.3 4186.99 0.25 Weighted Avg. - 1.00 2.89 0.87 1.50 3.76 0.94 - 3838.8 4494.99 0.23 3.3.3.1. Soil Root Growth Factor (SRGF) adjustment In Section 3.1.4, we showed that the potential yields simulated from the rainfed locations (Fig 3.6; see Table 3.1) using the calibrated genetic coefficients are capable of attaining the levels of observed yields. However, the model structure used in irrig ated conditions and imposed on water stressed environments could restrict the crop from exploring available resources from 40 the soil. Note that the soil structure model in DSSAT is static. Varella et al (2012) explained that uncertainties in soil input para meters can influence model performances. Water and nutrient availability in the root zone have a substantial impact on root geometry, dynamics and physiology, therefore influencing plant water and nutrient uptakes and yields (Ritchie et al., 1998; Ma et al ., 2006). Maize root grows deeper in the soil to extract more water per unit length of root in rainfed than irrigated conditions (Sharp and Davies, 1985). These root dynamics are more prevalent in the event of stress during critical periods e.g., tasselli ng to grain filling in maize (Lorens et al., 1987; Wan et al., 2000; Vamerali et al., 2003, Hund et al., 2009; Garcia et al., 2009). Lenka et al. (2009), Panda et al. (2004) and Djaman and Irmak (2012) noted that in well - irrigated conditions, root water ex traction mostly takes place from the top soil layers. Based on our literature review and multiple model iterations we suggest that the root growth factors in the soil model as used in irrigated conditions may be restricting the roots to go deeper/wider wh en applied under water stressed conditions (e.g., López - Cedrón et al., 2008; Yang et al., 2009). The soil database in DSSAT soil module has a root growth factor ( SRGF) parameter, which controls the maximum rooting depth and root mass distribution in the s oil profile (Jones and Kiniry, 1986). The root growth distribution function in CERES - Maize has been calibrated for different soil types as root hospitality factor, which gives flexibility to root growth in the model according to soil water availability and structure (Friasse et al., 2001). In DSSAT v4.6, SRGF with a value of 1.0 allows the root to grow equally in the soil layer and gradually decreases to zero in tapered form through the deeper layer (Yang et al., 2017). Table 3.11 and 41 3.12 show the SRGF ver tical distributions of the sandy loam and loam soils used in calibration (irrigated locations) and validation (rainfed locations). Table 3. 11 . Soil root growth distribution factor (SRGF) of the sandy loam soil used in calibrat ion and validation. Soil Layer Depth (cm) Montcalm Sandy Loam SRGF (Calibration) Irrigated SRGF (Validation) Rainfed 0 - 10 1.0 1.0 10 - 25 1.0 1.0 25 - 40 0.5 1.0 40 - 65 0.1 0.8 65 - 90 0.1 0.6 90 - 115 0.1 0.4 115 - 140 0.1 0.2 140 - 165 0.1 0.0 165 - 190 0.1 0.0 190 - 215 0.1 0.0 Table 3. 12 . Soil root growth distribution factor (SRGF) of the loam soil used in calibration and validation. Soil Layer Depth (cm) Kalamazoo Loam SRGF (Calibration) Irrigated SRGF (Validation) Rainfed 0 - 10 1.0 1.0 10 - 22 1.0 1.0 22 - 31 1.0 1.0 31 - 41 0.9 1.0 41 - 51 0.7 1.0 51 - 61 0.5 1.0 61 - 75 0.3 1.0 75 - 89 0.1 1.0 89 - 102 0.1 0.8 102 - 120 0.1 0.6 120 - 140 0.1 0.4 140 - 160 0.1 0.2 160 - 180 0.1 0.0 180 - 200 0.1 0.0 42 Based on a loam soil profile developed for Iowa, U.S.A. (see DSSATv4.6 SOIL.SOL by Ritchie for Iowa; Hoogenboom et al., 2003), roots were allowed to fully grow by keeping SRGF=1.0 until the soil depth of 90 cm, then slowly tapering down until the depth of 190 cm, whe re SRGF=0. For our sandy loam soil, we revised the SRGF parameters by relaxing SRGF to 1.0 until 40 cm depth then slowly tapering down to 0.2 until 140 cm (Table 3.11). For the loam soil, SRGF values were relaxed to 1.0 until 89 cm then slowly tapering dow n to 0.2 until 160 cm, except for Ingham (Table 3.1, 3.12) where we allowed the root to grow deep until 200 cm. These revised soil profiles were used in validation and CERES - Maize predictions of yields were substantially improved (Table 3.13). We analysed the new validation results to evaluate our hypothesis about the adaptive capacity of the plant under stressed environments that static model parameters may restrict. Fig. 3.7a suggests yield improvement from 4,058 kg ha - 1 with the old soil profile to 11,13 7 kg ha - 1 with the revised one, which is close to the 10,984 kg/ha observed yield. The improvement in yield prediction was an indication that some structural components in the model set up need to be adjusted when applied under rainfed conditions. The plant - stressed environment, e.g. root expansion and lengthening, might not be explicitly accounted for in the model, particularly on the flexibility of the soil capacity to allow roots to wander deeper and wider in t he root zone . Currently, this is strictly restricted by the soil profile set up a - priory to simulations. The observed yields are our to adjust a static soil struct ural property in order to get a better fit of the data. 43 Figure 3. 7 . Model performance ((root depth, water and nitrogen stress andyield (a)) and (root length density (b,c))) in 2017 at Washtenaw ( rainfed ) under old (b) and revised soil (c) profiles. Physiologically, maximum root growth occurs during pre - anthesis period to provide enough water for crop growth and development (Liu et al., 2017; Yang et al., 2017). In the case of the revised soil profile, root growth ceased after anthesis (78 DAP), which is typically 5 - 7 days after tasselling (Fig. 3.7a, 3.7c), which helps in the reduction of water stress during tassel initiation to early grain filling period (e.g., Lizaso et al., 2018). The root length density of each soil layer is uniform in the case of the revised soil profile (Fig. 3.7b, 3.7c). The root layer density sugge sts that the deeper root layers (8 and 9) are able to extract water and nutrients during the peak demand period of crop growth (Fig 3.7c). This is restricted in the old soil profile configuration (Fig 3.7b) as roots in soil layers 7 and 8 grew only after t asselling and anthesis and 44 did not grow deep enough, hence were not able to support the peak demands for water and nutrient in the growing season (Fig. 3.7a, 3.7b). We also performed simulations under irrigation with the revised soil profile to evaluate i f relaxing SRGF contributed to yield improvement when water is readily available. We found that there was no substantial difference in predicted yields under irrigated conditions before and after relaxing SRGF values (Fig 3.8a). Root length density in the deeper layers however became more uniform after tasselling after relaxing SRGF values (Fig 3.8c), while the deeper roots in layers 9 and 10 continued growing after anthesis with the older soil profile, but that did not contribute much to the yield dynamics as sufficient water was already provided by irrigation (Fig. 3.8a, 3.8b). The revised soil profiles were used in the validation of the derived genetic coefficients for all rainfed locations in 2017 and 2018. Sections 3.2.2 and 3.2.3 show the performance of phenology and yield predictions with the revised soil profile configurations. 45 Figure 3. 8 . Model performance ((Root depth, water and nitrogen stress, yield (a)) and (root length density (b,c))) in 2017 at Cass ( irrigated ) under old (b) and revised soil (c) profiles. Table 3. 13 . Performance of CERES - Maize using parameters estimated by GENCALC, GLUE, NMCGA and ensembling approach during validation at rainfed locations with the revised soil profil es. ADAP (DOY) MDAP (DOY) HWAM (Kg.ha - 1 ) MBE RMSE d - index MBE RMSE d - index MBE RMSE d - index GENCALC 0.17 2.74 0.96 6.75 8.49 0.87 332.25 653.96 0.93 GLUE - 1.50 3.03 0.96 3.17 4.97 0.96 645.58 774.23 0.92 NMCGA_SD - 0.50 2.65 0.97 3.67 5.16 0.95 323.67 536.84 0.96 NMCGA_NO_SD - 0.50 2.65 0.97 - 4.42 5.63 0.94 391.58 636.98 0.94 Arithmetic Avg. - 1.25 2.84 0.96 0.58 2.99 0.98 561.83 732.40 0.92 Weighted Avg. - 1.25 2.84 0.96 1.08 3.12 0.98 419.92 577.99 0.96 46 3.3.3.2 Anthesis and maturity validation results Predictions of anthesis (ADAP) and maturity dates (MDAP) did not change substantially when using the old (Table 3.10) or new soil profiles (Table 3.13) in the validation. For all the methods, anthesis dates were slightly underpredicted b ut with high d - index (Table 3.13). It suggests that the calibrated P1 and P2 coefficients performed well in predicting anthesis at the rainfed locations. Overpredictions of maturity dates by 3 days (GLUE and NMCG_SD) to a week (GENCALC), and underpredictio n by 4 days (NMCGA_NO_SD) were also observed (Table 3.13). Ensembling methods performed better than individual methods in predictin g maturity dates (Table 3.13). Coefficient of determinations (R 2 ) in Fig. 3.9 also show that the model could predict anthesis with high accuracy. Overlapping phenological data in some locations can be observed (e.g., Saginaw 2018 and Huron 2017) and is due mainly to the inter - annual variability of crop response to planting dates and weather patterns. Ensembling methods h ave the highest R 2 values for MDAP prediction (Fig. 3.9). 3.3.3.3 Yield validation results Tables 3.10 and 3.13 suggest that majority of the impacts from enhancing root dynamics is accounted for by the improvements in yields and not on phenology (see Sect ion 3.2.2). The yield d - index for all methods improved from <0.3 (Table 10) to >0.9 (Table 13) and R 2 > 0.8 (Fig. 3.9). Yields were still over predicted with RMSE ranging from 536 to 774 kg/ha (Table 3.13). Allowing the root system to explore more resource s allowed CERES - Maize to match better the observed yields under rainfed conditions, which is our gold standard data for fitting the model. This result suggests the intrinsic limitation of a rigid soil - based root growth factor (SRGF) (i.e., not dynamic) if one assumes it blindly and does not account for the adaptive capacity of the crops when setting up a soil profile data under water stress conditions. 47 Figure 3. 9 . CERES - Maize performance for validation of calibrated coefficients of GENCALC, GLUE, NMCGA_SD, NMCGA_NO_SD, arithmetic and weighted averaging for 2017 and 2018. 48 3.4 Summary and Conclusions In this study, we employed existing (GENCALC, GLUE) and new methods (NMCGA_SD and NMCGA_NO_SD) of crop model calibrations, ensembled them to get a robust estimates of a hybrid maize genetic coefficients. The genetic coefficients of CERES - Maize were calibra ted using three irrigated locations (Branch, Cass and Mason) in Michigan during 2017 and 2018 growing seasons. Results suggest that using two years (2017 - 2018) of data for calibration gave better results than using only one - year data (2017). The phenologic al parameters were mostly affected by the inclusion of 2018 data in the calibration. The calibrated coefficients were used to evaluate CERES - Maize performance across the six - rainfed locations (Washtenaw, Allegan, Ingham, Saginaw, Huron and Montcalm) during 2017 and 2018 growing seasons. The estimated coefficients were able to produce potential yields that are higher that observed yields, suggesting that they can generate those levels of yields if applied in real - world conditions. However, under stressed en vironments, validation results suggested that there are mechanisms that are not accounted for or permitted by the irrigated model setup used in calibration on the adaptive capacity of a crop grown under rainfed conditions, particularly root dynamics. Based on literature reviews and this above hypothesis, we relaxed the capacity of the soil to allow root geometry to change from the irrigated setup, which resulted in a better model fit. Using similar soil model structure in 2018, the calibrated parameters per formed well under rainfed conditions. The revised soil model structure was cross validated under irrigated conditions with nearly similar results as the old soil model structure suggesting that the root system did not change substantially between the two soil model models when the crops were irrigated, possibly due to the abundance of water in the upper root zone. 49 We found that ensembling coefficients by weighted averaging, based on the inverse squared distances between the predicted and observed variable(s) (e.g., ADAP, MDAP or HWAM), outperformed most of the calibration models. The purpose of ensembling genetic coefficients from several methods was to integrate the strengths of the methods with a goal of achieving a set of parameters (i.e., genetic coefficients) that is robust and resilient. Our results suggest that there is value in using mul tiple methods in calibrating crop models and extracting the best information from them. 3.5 Acknowledgment This work is partly funded by the Corn Marketing Program of Michigan (CMPM) and MSU AgBioResearch. We acknowledge Chubu University, ListenField, U SAID - Philippines and NASA - acknowledge Bill Widdicombe for the Michigan Maize Performance Trial (MMPT) experiments and Katlin Fusilier for collecting specific field data. 50 4. ESTIMAT ION AND VALIDATION OF REMOTELY SENSED EVAPOTRANSPIRATION FOR THE DEVELOPMENT OF CROP COEFFICIENTS OF MAIZE AND IRRIGATION SCHEDULING 4.1 Introduction Crop production consumes the largest proportion of fresh water through irrigation (Döll et al., 2014; Wada and Bierkens, 2014; Haddeland et al., 2014). Increasing use of irrigation and associated water withdrawals for agriculture have led to unsustainable water use and a number of environmental consequences (Rosa et al., 2018) and in the near future, irrigatio n water scarcity is expected to increase (IPCC, 2014; Lu and Kueppers, 2015; Porter et al., 2017). Irrigation during critical crop growth stages helps in meeting the crop water demand and minimizing the yield volatility (Irmak et al., 2016; Williams et al. , 2016; Jha et al., 2018; Comas et al., 2019). Simulation studies have shown that if irrigation were eliminated, the global cereal production would decrease by ~20 % (Siebert and Döll, 2010; Siebert et al., 2015; AQUASTAT, 2018). Seasonal variation in prec ipitation especially in the peak - growing season in humid to sub - humid regions makes water management challenging (Siebert and Doll, 2010). On the other hand, irrigation influences the local crop micrometeorology through energy fluxes in and over the crop canopy (Siebert et al., 2014; Thiery et al., 2017; Sridhar and Anderson, 2017; Zhang et al., 2019) and the regional climate conditions through water and irrigatio n contributes 418 - 1233 km 3 year - 1 of water to the atmosphere through evapotranspiration (ET), the largest sink of irrigation water, most of which returns to the irrigated areas as recycled precipitation (Thiery et al., 2017). In the temperate climate of Midwest US, two - thirds of precipita tion returns to atmosphere through ET and the majority of 51 that occurs during the crop - growing season (Abraha et al., 2015). ET uniquely links water, carbon, and the energy cycle and is considered an important predictor of crop yield (Abtew and Melesse, 201 3; Li et al., 2014; Fisher et al., 2017; Yang et al., 2018) and in determining water use efficiency (WUE) (Ito and Inatomi, 2012; Knauer et al., 2018). To increase WUE, an improved irrigation management strategy with a reliable ET estimate is necessary. B ecause of its importance in water management, ET has been studied extensively. ET can be measured by lysimeter, eddy covariance (EC), or the Bowen - ratio energy methods at the field scale (Bowen, 1926; Tomlinson, 1996). Due to scaling and interactions of la ndscape features with climate, ET at the regional scale is more difficult to quantify accurately using in - situ methods (Abraha et al., 2015; 2016). With advances in thermal satellite remote sensing (RS), ET patterns have been observed over time and space ( Hamada et al., 2015; Anderson et al., 2018). ET either can be estimated with RS through the surface energy balance method (EB) or with reflectance based crop coefficients ( K cr ). The EB uses remotely sensed surface reflectance in the visible (VIS) and near infrared (NIR), and surface temperature in the infrared (IR) regions of the electromagnetic spectrum (Holmes et al., 2018). In contrast, K cr is derived through vegetation indices (Glenn et al., 2011). Widely used EB based ET models include the Atmosphere - L and Exchange Inverse (ALEXI) model (Anderson et al., 1997); the Surface Energy Balance Algorithm for Land (SEBAL) model (Bastiaanssen et al., 1998), the Surface Energy Balance System (SEBS) model (Su, 2002), the Mapping EvapoTranspiration at high Resolutio n with Internalized Calibration (METRIC) model (Allen et al., 2007), the Simplified SS) (Dhungel et al., 2019). 52 These ET estimation methods have their own uncertainties associated with data inputs, model structures and abstractions (Zhu et al., 2016). In order to minimize associated uncertainties from individual methods, the use of multi - model ensembles have gained popularity in the past (e.g., Vrugt and Robinson, 2007). Ensembling crop models helped in better predicting yield and on the sensitivity and uncertainty analyses of input parameters (Bassu et al., 2014; Huang et al., 2017; Basso et al., 2018; Rosenzweig et al., 2018; Kimball et al., 2019; Müller et al., 2019). Hydrological models (e.g., Velázquez et al., 2011; Wagena et al., 2019) and climate models (e.g., Rötter et al., 2011; Chen et al., 2019) have been ensembled to minimize th e uncertainty associated with individual models. Different ET models have been ensembled to assess their predictability over a wide range of climatic conditions and landscapes and have been used for irrigation scheduling (Garcia et al., 2013; Ershadi et al ., 2014; Zhu et al., 2014, 2016). Scheduling of irrigation is done by estimating a ctual crop ET (ET crop ) and available water in the root zone. Traditionally, ET crop is estimated by multiplying a crop coefficient ( k c ) with a reference ET (Jensen et al., 1 990; Allen et al., 1998). Crop coefficients are typically calculated based on crop development stages under optimum management conditions using lysimeter studies. These specific measurements do not describe the spatial variation in crop coefficients at the regional scale due to varying management practices (e.g., planting density, irrigation, tillage and others) and crop conditions (Marin et al., 2016). Characterization of k c at the larger scale using remote sensing can be helpful for better water managemen t. Vegetation indices - based k c can be estimated through remote sensing and incorporated in irrigation scheduling (Hunsaker et al., 2003; Mokhtari et al., 2019). K c can also be estimated inversely by estimating ET, since K c is the ratio of actual ET to the reference ET (Allen et al., 2005). Crop 53 coefficients have been derived inversely by estimating ET through the SEBAL energy balance model (Michael and Bastiaanssen, 2000; Tasumi et al., 2005; Samani et al., 2008), in - situ ET measurements through the Bowen r atio method (Sobenko et al., 2019), lysimeter measurements (Martínez - Cob, 2008 Piccinni et al., 2009) and eddy covariance measurements (Facchi et al., 2013; Migliaccio et al., 2014; Corbari et al., 2017). However, all of these approaches of estimating crop coefficients were used to derive K c for a single location or from a single ET model. The main goal of this study is to derive reliable estimates of maize crop coefficients (Kc) in the irrigated region of Southwest Michigan by using an ensemble of ET mode ls. Specific objectives are as follows: (i) to estimate actual ET of maize fields using RS based methods (SEBAL, PT - JPL and MOD16), FAO - Kc and an ensembling approach, (ii) to derive crop coefficients (Kc) of maize from the remotely sensed ET, (iii) to vali date the region specific dervived Kc using a soil water balance model, and (iv) to evaluate the region specific derived Kc to design irrigation scheduling. 4.2 Materials and Methods 4.2.1 Study Area The study sites are located in Southwest Michigan, which is the heartland of hybrid seed production for maize in the northeastern part of the US Midwest Maize Belt (MSU extension, 2014). The two maize fields selected for this study are equipped with eddy c ovariance towers (sites in the AmeriFlux network; Cammalleri et al., 2014; Abraha et al., 2019 ) and are under a continuous rainfed maize crop rotation (Fig. 4.1). The first field (US - KL1, KBS Lux Arbor Reserve Corn) was managed as maize - soybean crop rotati on for more than 50 years, before conversion in 2009 to a no - till continuous maize rotation. It is located at the Long - term 54 Ecological Research site of W. K. Kellogg Biological Station, Kalamazoo, Michigan (42.49° N, 85.44° W, 288 m asl). The second field (US - KM1, KBS Marshall Farms Corn; 42.44° N, 85.33° W, 286 m asl) was under the Conservation Reserve Program (CRP) Grasslands of smooth brome grass ( Bromus inermis L.) for 22 years before conversion in 2009 to a no - till continuous maize rotation. These two sites with different land - use history offer a special opportunity for ET analysis (Senay et al., 2019). Both fields are located in a humid continental temperate climate (Koppen, 1900) six miles apart, with a mean total annual precipitation of 1027 mm, out of which more than half (523 mm) occurs during maize growing season (May - September) (Abraha et al., 2019). The dominant soil types are a well drained Kalamazoo (loam) and an Osthemo (sandy loam) (Robertson and Hamilton, 2015; Luehmann et al., 2016). Fig ure 4. 1 . Eddy covariance towers at the maize fields 55 4.2.2 Data Collection 4.2.2.1 Weather Data The weather data for this study were obtained primarily from the Enviro - weather network, a sustainable weather - based information system that helps growers and stakeholders in making farm related decisions in Michigan (Andresen et al., 2012). The reference evapotranspiration ( ET ref ) for estimating ET crop were obtained from the nearest weather station, Pierce Cedar Creek Institute, Hasting (42.53 N, 85.30 W, 283 m asl). Means and standard deviations of daily ET ref (2010 - 2017) for the region are shown in Fig 4.2 showing mean ETref peaks in July. Rainfall data were not used directly in ET estimation. However, being an impor tant driver of ET dynamics at the spatiotemporal scale of the study, they were used to assess the reliability of the estimated ET. Rainfall data from the Leslie station site (42.47 N, 84.46 W; 291 m asl) nearest to our validation site Manchester, Michigan (42.19 N, 84.04 W) were used for simulating the water balance. For sensitivity analysis of the water balance model, rainfall input from the Hastings Enviroweather station site, a rain gauge at the field site and reanalysed rainfall estimates from NASA - POWE R, version 2.1 Global Precipitation Climate Project (GPCP 1DD), were also collected (Bolvin et al., 2009). 56 Figure 4. 2 . Mean reference ET (2010 - 2017) from the nearest weather station (Hasting, Michigan) in the study area. 4.2.2.2 Soil moisture and irrigation data Validation of the derived crop coefficients were done using observed soil moisture in the maize field at Manchester (42.19 N, 84.04 W). The soil texture at this field was sandy clay loam. Rainfall was measured by a n ATMOS 41 weather station (Meter Group, Pullman, WA) (Fig 4.3a) along with air temperature, relative humidity, vapour pressure, barometric pressure, wind speed and direction, solar radiation, and lightning strike frequency. The rain gauge was adjusted at two week intervals to ensure that it remained over the canopy all the times during the growing season. The measured rainfall from the rain gauge installed at the field was used for the sensitivity analysis of the water balance model. Actual irrigation am ounts applied were obtained from grower observations at the field site, from July 1 to August 16, 2019, which coincided with tasseling to milking stage in maize (Abendroth et al., 2010). Volumetric soil moisture, soil temperature, and electrical conductivi ty were monitored with five Teros 12 57 probes (Meter Group, Pullman, WA) at 15, 30, 45, 60, and 90 cm depths. Sensors were installed by digging a shallow trench and inserting the sensors horizontally into the soil, then backfilling the trench (Fig 4.3b). The resolution and accuracy of the volumetric moisture sensor are 0.001 m 3 m - 3 and 0.02 m 3 m - 3 , respectively. The measurements of the soil moisture sensors and the weather parameters were recorded hourly using an EM 60G datalogger (Meter Group, Pullman, WA; Fig 4.3a), and converted into daily variables. Figure 4. 3 . Rain gauge (a) and soil moisture sensors (b) at the maize field in Manchester, Michigan. 4.2.2.3 Eddy covariance (EC) data EC sensors were placed at 2.0 m above the average canopy height at the centre of the each of the two maize fields (Abraha et al., 2 019). The locations of the two towers are shown in (a) ( b ) 58 Fig 4.1. ET data were computed as 30 minutes averages using EdiRe software (University of Edinburgh, v 1.5.0.32, 2012) (Clement, 1999). The processed ET data were passed through quality checks and controls , and gap - filled using REddyProc (Wutzleret al., 2018), which follows the algorithm of Reichstein et al., (2005). The estimated daily ET data from the flux towers were used for validation and as a reference for ensembling ET methods using an Inverse Distan ce Weighting (IDW) method, which will be described later. 4.2.2.4 Remote sensing data Due to the complexity of hydrological process and heterogeneity of natural environments, it is difficult to estimate ET over large areas with in situ measurements. Remote sensing - based ET estimates for this study were generated using SEBAL (Bastiaanssen et al., 1998), PT - JPL (Fisher et al., 2008) and PM - MODIS (ET) (Mu et al., 2011). 4.2.2.5 SEBAL Based on the coordinates of the two maize fields, the til e grid (sinusoidal projection) of MODIS (Moderate Resolution Imaging Spectroradiometer) data products were obtained using the MODLAND tile calculator. The spatial domain of the study area lies within horizontal 11 and vertical 04 sinusoidal tiles. MOD09GA version 6 (Daily surface reflectance products L2G; Band 1 to 7 at 500 m resolution) and MOD11A1 version 6 (Land surface temperature & emissivity product at 1 km resolution) for the years (2010 - 2017) were downloaded from buted Active Archive Center) ( https://lpdaac.usgs.gov/ ). The downloaded data in hierarchical data format (.hdf) were geo - registered and downscaled at 500 m resolution using MODIS Reprojection Tool (MRT) under MODIS conversion toolkit (MCTK) in ENVI Classic 5.3 software. Daily wind speed data were collected from the nearest weather station, Pierce Cedar Creek Institute, Hasting. We used these 59 data to estimate ET using our C - program implementation of the Surface Energy Algorithm for Land (SEBAL). 4.2.2.6 PT - JPL The modified Priestley - Taylor ET is a Level - 3 (L - 3) product of the ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) mission, which is a combination of the land surface temperature p roduct and ancillary data products (Hulley, 2015; Fisher et al., 2011; 2017). The NASA - JPL group provided daily ET estimates (2010 - 2017) in GeoTiff format for validation and application in this study. It estimated actual ET with inclusion of plant - and soi l - based eco - physiological stress factors. Surface latent heat flux is partitioned into soil evaporation, canopy transpiration and evaporation from intercepted moisture. These products are validated using FLUXNET and are ready - to - use product, hence, no cali bration or ground data was required (Fisher et al., 2008). 4.2.2.7 PM - MODIS (MOD16A2 - ET) The MOD16A2 ET version 6 (500 m resolution) datasets were downloaded from the data repository of the USGS - EarthExplorer website. These downloaded data in hierarchica l data format (.hdf) were geo - registered using the MODIS Reprojection Tool (MRT) under MODIS conversion toolkit (MCTK) in ENVI Classic 5.3 software and then converted into ENVI standard data (.dat) with header (.hdr) files. Annual time series of ET estimat es were extracted for further analysis. The MOD16 ET estimates are based on Penman - Monteith equation (Monteith, 1965, Mu et al., 2011). ET are available as 8 - day composites (Mu et al., 2011). 60 4.2.3 ET models descriptions 4.2.3.1 SEBAL ET is driven by the flux of energy at the surface. The Surface Energy Balance Algorithm for Land (SEBAL) developed by Bastiaanssen et al (1998, 2000, 2005) estimates instantaneous ET for each pixel of the processed image as a residual from the energy balance equation (Eq. 1) considering latent heat as a proxy of ET: (1) where, LE is the latent heat flux (W m - 2 ), Rn is the net radiation flux (W m - 2 ), G is the soil heat flux (W m - 2 ), and H is the sensible heat flux to the air (W m - 2 ). All the components in Eq. 1 are n index (NDVI), and leaf area index (LAI). Net radiation is the difference of incoming and outgoing fluxes computed using the surface radiation balance equation described by Zhou et al. (2017). Sensible heat (H) is computed through air and vapor momentum resistances (surface roughness) and the vertical temperature gradient (dT), which are functions of the wind speed and air temperature, respectively, known as the aerodynamic function (Allen et al., 2007). The air temperature is estimated as a proxy of surf ace temperature at the cold pixel (Bastiaanssen et al, 2005; Senkondo et al., 2019) and wind speed (at 2 m) obtained from the nearest station were used to compute heat and vapor transport for sensible heat (H) computation. In order to estimate dT , the mode l has to select anchor (hot and cold) pixels, either manually or automatically. Manual selection is subjected to user - based uncertainty of choices and selection; therefore, we employed SEBAL for automatic selection of the pixel. 61 The latent heat flux is di vided by the latent heat of vaporization (~2.45 MJ Kg - 1 ) to estimate instantaneous ET. The instantaneous ET is scaled up to 24 hours to get daily ET estimates assuming constant evaporative fraction throughout the day. This fraction expresses the ratio of a ctual to potential evaporative demand and hence is used in estimating daily ET (Farah et al., 2004). The SEBAL technique has been used and applied successfully in a number of previous studies related to ET estimation and water management (Bastiaanssen et a l., 2005; Allen et al., 2011; Singh and Senay, 2015; Grosso et al., 2018; Gobbo et al., 2019). Tang et al. (2013), for example, found daily estimates of ET derived from SEBAL have an accuracy of 85%, with 95% for seasonal estimates, and 96% for annual est imates as compared to measured ET from eddy covariance methods. 4.2.3.2 PT - JPL A form of the Modified Priestley - Taylor method ( Priestley and Taylor, 1972 ), popularly known as PT - JPL and developed by the NASA Jet Propulsion Laboratory estimates actual ET through the introduction of plant - and soil - based eco - physiological stress factors (Fisher et al, 2008; 2017). The idea of introducing these stress factors is to scale down the potential ET to actual ET in data scarce regions. It follows the two source s of evaporation (Soil and Crop) and combines plant moisture and temperature constraints, which helps in estimating daily variability in ET due to plant stress. It is a ready - made product generated by combining datasets from the International Satellite Lan d - Surface Climatology Project (ISLSCP; Los et al., 2000) and the Advanced Very High Resolution Spectroradiometer (AVHRR; Houborg and Soegaard, 2004), along with the original PT method with Priestley constant at 1.26 (Fisher et al., 2 008). The model has been validated through FLUXNET measured ET data at different ecological environments (Baldocchi, 2019) and does not require 62 site calibration. The daily ET estimates from PT - JPL is in TIFF format and hence there is no need to reproject o r georeferenced the data before processing them in ENVI. It has an advantage over SEBAL for having fewer cloud days as cloud corrections were performed using parameters from Cloud Climatology Project (CCP) data (Rossow et al., 1996) 4.2.3.3 MOD16ET The MO D16, global 8 - day ET (MOD16A2) datasets at 500 m resolution are based on the Penman - Monteith relationship for ET (Monteith, 1965), which uses reanalyzed daily meteorological data from et al. 19 93) , and vegetation dynamics every 8 - days from MODIS (Mu et al., 2007, 2011). The model assumes that biome - specific parameters do not vary spatiotemporally (Running et al., 2017). Based on the algorithm, MOD16A2 estimates ET at 8 - day intervals which has been validated by NASA with measured ET at multiple global eddy covariance towers (Mu et al., 2011). 4.2.3.4 FAO - 56 Kc approach The Food and Agriculture Organization (FAO) method (Allen et al., 1998) has been used traditionally for ET estimates by multiplyin g a crop coefficient ( K c ) by reference ET. ET ref is estimated with the Penman - Monteith equation (Monteith, 1965) for a hypothetical flat, unshaded grass - covered surface with height 0.12 m, albedo of 0.23 and surface resistance of 70 s m - 1 at the daily int erval. ET ref (2010 - 2017) were collected from the nearest Enviroweather site, Hastings. Actual daily crop ET ( ET crop ) values were calculated by multiplying ET ref and the tabulated daily values of the crop coefficients ( K c ) for maize ( FAO - 56 ; Allen et al., 1 998), based on FAO tabulated crop growth stages. 63 4.2.4 Ensembling approach As accurate and reliable estimation of ET is of paramount importance for irrigation scheduling (Zhu et al., 2014). A large number of models have been developed for ET estimation. However, their fidelity depends on the choice of model and associated parameters. Scaling of the potential ET to actual ET requires complex model parameterization and hence the associated uncertainty arises. In this study, we have used the SEBA L energy balance approach (SEBAL), the Priestley - Taylor approach (PT - JPL), and the Penman Monteith approach (MOD16 and FAO - Kc ) to estimate a more robust actual ET. All of them vary in parameterizations and data requirements. Hence, the actual ET values es timated from these methods are expected to differ. Moreover, uncertainties in input variables either weaken the model performance or create biased estimation (Ershadi et al., 2014). In order to get the best estimates, multiple models have been evaluated an d compared at different locations and time ( Xystrakis and Matzarakis, 2010; Fisher et al., 2011, Ershadi et al., 2014; Li et al., 2018; Olioso et al., 2019; Zhang et al., 2020 ). The ensembling of multiple model outputs not only minimizes the uncertainties in model structures and input but also improves accuracy in output (Vrugt and Robinson, 2007). Based on ET measurements from the flux towers, all the methods were ensembled through weighted averaging using Inverse Distance Weighting (IDW) method. A weighte d average ensemble provides a specific weight to the individual model in a proportion to observed parameter in order to predict that parameter. The ensembling was done with a purpose of achieving more reliable estimates of ET. The general framework is show n in Eqs. 2 - 4; (2) (3) 64 (4) where P j is the ensembled value of ET, W i is the weight of a ET from an individual method i, p ij is ET estimates from method i, j is an index of ET estimat es. w i is the inverse squared - distance between the estimated (A pij,k ) and observed (A o,k ) ET. N is the number of models selected for ensembling (here, N=4), k is an index for predicted or observed variable(s), K is the number of. variable(s) impacted by p i j and is the weight for that variable, k ( Jha et al., 2019, in review). Furthermore, we also performed a crop stage - based IDW, where ET were averaged at different crop growth stages (Early, developmental, mid and late stages) similar to growth stagin g in FAO - 56 (Allen, 2005). The stage - based averaged ET was used to generate inverse distance Weighting and ensembled further after providing weight to individual method. The calibrated CERES - Maize model in Chapter 4 (Hoogenboom et al., 2013; Jha et al., 20 19, in review) was used for designating growth stages and mapped them with FAO - 56 categorization. The early growth stage was designated from the germination of maize to floral initiation (~30 days after planting) when crop growth is minimal. The developmen tal stage was designated from the floral initiation to 75% silk. The mid stage was designated from the 75% silk to mid of beginning to end of the grain filling period. The late growing stage was designated from the mid of beginning to end of the grain fill ing period to maturity (Fig 4.4). The end of grain filling is marked by senescence of leaves and formation of black layer in the maize kernels followed by harvest at the moisture level of 15 to 16 % depending on the field conditions and other ancillary su pports ( Abendroth et al., 2010). 65 Figure 4. 4 . Crop coefficient curve from FAO and the curve created from CERES - Maize growth stages. 4.2.5 Kc Curve and Irrigation Scheduling The uncertainties associated with ET estimation through an individual method can also be passed on to the derivation of the crop coefficient. In order to obtain reliable estimates, we ensembled ET estimates from SEBAL, PT - JPL, MOD16 and FAO method. The ensembled ET estimates were used to deri ve crop coefficient using reference evapotranspiration from the nearest weather station, Hastings. The approach and workflow of deriving the crop coefficient in this study is shown in Fig 4.5, where multiple models for ET estimates were ensembled to get ET RS and later divided by ET ref from the nearest weather station to derive the crop coefficient ( K c ). 66 Figure 4. 5 . Conceptualization of developing satellite derived crop coefficient curve. 4.2.6 Development of Kc - curve The daily ensemble ET estimates were used to calculate the crop coefficient ( k c ) as Eq. 5, (5) where ET ref = reference ET were collected from the nearest weather station which was calculated using Penman - Monteith method with a standard grass (FAO 56, Allen et al., 1998). The ET estimates from all the methods were ensembled based on crop staging fr om CERES - 67 Maize for the study location for eight years (2010 - 2017). The stage - based K c curves were generated for each year and for both locations. The curves were fitted based on each growth stages (early, developmental, mid and late) following the FAO - 56 a pproach. Out of the derived daily K c cloud, the wet days with 5 mm or more rainfall were chosen as a threshold to train the model for picking the candidate - wet days. The best - fit lines were drawn to get K c - curves for the crop development stages. Eventuall y all the staged K c - curves were combined to form a seasonal K c - curves for the crop and location. In order to arrive at the best representation of crop coefficients, we averaged k c for all years (2010 - 2017) and locations (KBS and Marshall Maize farm). 4.2 .7 Validation of Kc - curve The soil water balance model (Martin et al., 1990) was used to validate the derived K c curves, which was later used for irrigation scheduling. The soil water balance model is a macro - based automated irrigation scheduling tool, w hich allows user to approximate growing season length, by inputing planting date, crop and soil type, initial and desired soil available water capacity, and weather input (rainfall, potential ET) (Martin et al., 1990). Based on user selection, it estimates available water in root zone at daily interval, which indicates crop water demand and irrigation requirement. The performance of the soil water balance with standard FAO K c - curve and remote sensing derived K c - curve were compared. Comparisons of simulate d available water in the root zone were made with measurements using soil moisture sensors at the Manchester field. The observed volumetric soil moisture were converted as total available water and scaled with root depth factor to derive total available wa ter in root zone using soil water parameters for this texture (Jagtap et al., 2004). We run the soil water balance using three 68 different rainfall datasets, rain gauge in the field, nearest weather station and reanalyzed NASA - POWER data. 4.2.8 Statistical evaluation of model performance Model performances were evaluated statistically using coefficient of correlation (R; Eq. 6), Mean Bias Error (MBE; Eq. 7), Root Mean Square Error (RMSE; Eq. 8) and Index of agreement (d - index; Eq. 9) (Willmott, 1982) to measure the models performances, (6) (7) (8) (9) where M and O are estimated/simulated and observed variables (ET, Total available water in root zone) respectively. These metrics were used to evaluate the performance of ET models and the soil water balance model under standard FAO kc - curve and the derive d region specific kc curve. 69 4.3 Results and Discussion 4.3.1 ET Estimation and comparison The spatial domain for the remote sensing analysis within the study area covered a large portion of southwestern Lower Michigan (Fig 4.6.) where, a large portion of maize production is irrigated (MSU Extension, 2014). Historically, i rrigation transformed the high risk, low water holding capacity soils in this area into productive and profitable croplands. However, over and under irrigation is common a mong Michigan irrigators, both with economic and ecological consequences. With a goal of improving water use efficiency, we estimated ET from the maize fields in the study area with four methods (SEBAL, PT - JPL, MOD16A2 and FAO - Kc approach) and ensembled th em into estimates of field scale ET. Sample ET maps of the early growing season in 2012 based on SEBAL, PT - JPL and MODIS model show the spatiotemporal variation of ET during the early part of the growing season (Fig 4.7). Figure 4. 6 . Bounding box for remote sensing analysis in the study area 70 Figure 4. 7 . ET map (SEBAL, PT - JPL and MOD16A2) for a representative day during growing season (June 26, 2012) The evapotranspiration for the two maize fie lds, where flux towers are located, was estimated from eight years (2010 - 2017) remote sensing data. ET estimates from the individual methods were compared and validated with the measured ET from the eddy covariance towers located in the maize fields. For s implicity, we showed here ET comparison with eddy covariance measurement for one year (2012) at the KBS maize field (Fig 4.8). The ET estimates from all the models were compared for all years (2010 - 2017) and for both locations. ET estimates based on Penma n Monteith equation (8 - day composite MOD16A2) (Fig 4.8a) were found to have the highest statistical agreement among the individual models (Table 4.1). However, the temporal resolution of 8 - day makes it less desirable for decision - making process at the fi eld scale. Some studies have shown that the performance of MOD16A2 remains poor in arid and semiarid regions with significant underestimation of ET (Ramoelo et al., 2014; Hu et al., 2015). In our study area, under humid continental temperate type climate ( Koppen, 1900), the performance was found to be good (Table 4.1), which is in agreement with previous studies (Hu et al., 2015; Yang et al., 2016). 71 Figure 4. 8 . Comparison of estimated ET (a) MOD16A2, (b) SEBAL, (c) PT - JPL, (d) FAO with the eddy covariance measurements at the KBS maize field for 2012. 72 Table 4. 1 . Statistical evaluation of seasonal ET estimates for 2010 - 2017 among the models and flux tower during the growing season (M ay - October) at Marshall and KBS maize farms. Note: Values in parenthesis are standard deviation for 8 years (2010 - 2017). ET estimates FAO SEBAL PT - JPL MODIS Average Maize Farms KBS Marshall KBS Marshall KBS Marshall KBS Marshall KBS Marshall MBE 0.31 (0.28) 0.30 (0.32) 1.00 (0.43) 0.84 (0.30) - 0.35 (0.40) - 0.54 (0.41) - 0.01 (0.15) 0.01 (0.17) 0.24 (0.24) 0.15 (0.23) RMSE 0.81 (0.68) 0.68 (0.56) 1.57 (.99) 1.30 (0.55) 0.45 (0.29) 0.72 (0.46) 0.22 (0.13) 0.23 (0.14) 0.20 (0.14) 0.18 (0.13) d - index 0.80 (0.15) 0.87 (0.09) 0.62 (0.15) 0.66 (0.14) 0.80 (0.08) 0.72 (0.11) 0.90 (0.05) 0.91 (0.03) 0.91 (0.07) 0.94 (0.05) Correlation (r) 0.79 (0.3) 0.81 (0.31) 0.72 (0.28) 0.68 (0.32) 0.75 (0.30) 0.71 (0.28) 0.79 (0.30) 0.78 (0.30) 0.85 (0.32) 0.83 (0.31) 73 ET estimates from SEBAL were too high in the early and mid - growing season. Moreover, in the growing season, ET from SEBAL were overestimated with MBE of 1 mm day - 1 and 0.84 mm day - 1 fo r KBS and Marshall Farm respectively compared with measured ET from the eddy covariance (Table 4.1). The most plausible reason behind overestimation lies in its model approach, which estimates ET as residual energy from the energy balance equation (Eq, 1) and neglects energy stored as body heat in vegetation and the energy required for photosynthesis (Bastiaanssen et al., 1998, 2005). In addition, due to close proximity of the Great Lakes, residual moisture content at the hot pixels remains high most of th e early to mid - growing season which might cause underestimation of the sensible heat flux and hence overestimation of ET (Fig 4.8 b) (Singh and Irmak, 2011; Singh and Senay, 2016). Early in the growing season, when the canopy is sparse, the heat transfer t o soil (G; Eq. 1) becomes part of net radiation if soils are dry (Bastiaanssen et al., 1998). The wet saturated soil in the study area led to underestimation of soil heat flux (G) and hence ET might be overestimated The energy balance models in previous st udies have been found to overestimate ET in semi - humid to humid environments with small VPD (Basso and Ritchie, 2018; Yang et al., 2016). The modified Priestley - Taylor based ET estimates (PT - JPL) were found to have smaller biases and a higher d - index tha n SEBAL across the growing season (Table 4.1; Fig 4.8c). Although model performance was good, there are multiple days where considerable deviation from the measured value were observed (Fig 4.8c). The PT - JPL, which generally underestimated ET with MBE of - 0.35 mm day - 1 and - 0.54 mm day - 1 for KBS and Marshall Farm and R 2 of 0.75 (KBS) and 0.71(Marshall) (Table 4.1) when evaluated with eddy covariance data, which is in agreement with previous studies under similar climate (Chen et al., 2013; Ershadi et al., 2013). Moreover, Garcia et al. (2013) compared PT - JPL performance over 74 semi - arid and semi - humid climates and concluded that PT - JPL has lower performance in semi - arid environments than semi - humid. Traditionally, the FAO - 56 method (Allen et al., 1998) is used for calculating crop water demand and irrigation scheduling using ET ref and the crop coefficient (K c ). We performed a single source crop coefficient approach to estimate ET using ET ref from the nearest weather station. The overestimation of ET throug h the FAO method was due to the use of crop coefficient, which did not represent crop and soil moisture conditions at the fields (Er - Raki et al., 2009). ET estimates from FAO showed high correlation of 0.79 (KBS) and 0.81 (Marshall Farm) and d - index of 0. 80 and 0.87 with ET from flux tower. ET estimates were slightly overestimated with MBE of 0.31 mm day - 1 (KBS) and 0.30 mm day - 1 (Marshall) (Table 4.1). There is a general recommendation that the crop coefficients should be adjusted for specific agro - climat ic conditions to better represent stress conditions at the field (Katerji & Rana, 2006). The daily ET in the months of the maize growing season for all models and inter model comparison with measured ET are shown in Fig. 4.9. The monthly variation in esti mating ET within the growing season (May October) is mostly attributed to the stage of crop growth, development, and crop water demand. Early in the season, crop water requirements remain low and can be met by the combination of precipitation and stored soil water. As the season progresses, increase in VPD regulates water demand (Jha et al., 2018) and irrigation meets this water demand during the peak of vegetative growth stages (June, July and August). 75 The FAO method has the highest monthly variation (Fig 4.9b) as it utilized an unadjusted, single source crop coefficient (Allen et al., 2016). Higher variation in ET estimates from SEBAL in early season (May) might be due to wet soils and underestimation o f H and G (Eq. 1) (Fig 4.9c). PT - JPL underestimated ET (Fig 4.9a) as compared to measured ET from the eddy covariance data for all months in the growing season (Fig 4.9a). However, all model captured the seasonal ET trends for maize. Figure 4. 9 . Temporal variation of monthly ET among the methods (b) FAO (c) SEBAL, (d) PT - JPL, (e) MOD16A2 and ground observation of flux tower (a) during growing season (May - October) at KBS maize farm 76 Figure 4. 10 . Temporal variation of seasonal ET among the models and ground observations from flux towers during growing season (May - October) at KBS (a) and Marshall (b) (b) (a) 77 Figure 4. 11 . Inter model comparison of seasonal ET among the models and ground observations from the flux tower during growing season (May - October) at KBS 78 Figure 4. 12 . Inter model comparison of seasonal ET among the models and ground observations from the flux tower during growing season (May - October) at Marshall 79 Moreover, the interannual varaition in ET for both the locations and all models are shown in Fig. 4.10. Intermodel comparison acro ss the years (2010 - 2017) (Fig. 4.11, 4.12), show that MODIS ET data follow the pattern of the measured ET from the eddy covariance. Seasonal ET estimates from PT - JPL were lowest and from SEBAL were highest among all the models and across all the years (Fi g 4.11 and 4.12). The range of seasonal ET estimates were highest in the FAO approach. Each of the model were found to have their own advantages and limitations leading to uncertainties and biases in ET estimation. 4.3.2 Ensembling ET estimates The stage based ensemble ET (S - IDW) were estimated for all years and locations. The length of crop development stages simulated by CERES - Maize model are shown in Table 4.2. These ET from all the individual methods were averaged during these stages to estimate the weighted average ET estimates. The representative ensemble ET (2012 - KBS) is shown in Fig 4.13. The comparison of monthly average of ensemble ET with flux - tower data (2010 - 2017) showed RMSE of 0.2 and 0.18 mm day - 1 , and 0.91 and 0.94 d - index for KBS a nd Marshall Farm respectively (Table 4.1) Table 4. 2 . DSSAT based simulated and updated length of crop development stages used for ensembling purposes Note: DAP - Day after planting Length of crop development stages Year Early (DAP) Developmental (DAP) Mid (DAP) Late (DAP) 2010 0 - 31 32 - 69 70 - 105 106 - 136 2011 0 - 32 33 - 68 69 - 108 109 - 138 2012 0 - 33 34 - 64 65 - 97 98 - 132 2013 0 - 37 0 - 80 81 - 112 113 - 142 2014 0 - 33 34 - 79 80 - 109 110 - 138 2015 0 - 34 35 - 71 72 - 107 108 - 140 2016 0 - 34 35 - 71 73 - 104 105 - 133 2017 0 - 35 36 - 76 77 - 114 115 - 148 80 Figure 4. 13 . Ensemble ET estimates by stage based Inverse Distance Weighting (IDW) for 2012 with a reference of ground observations from the flux tower during growing season (May - October) at KBS. Mean of monthly ET estimates from the individual and ensemble me thods were compared with ET from the flux towers at both the locations (Fig. 4.14, 4.15). ET estimates from FAO were found to be the highest followed by SEBAL ET across the years (2010 - 2017). ET MODIS were found to the closest with the observed one through out this study period. ET from PT - JPL were found to be underestimated. However, ensemble ET (average) (Fig 4. 14, 4.15) performance was best among all the methods with d - index of 0.91 and 0.94 at KBS and Marshall field (Table 4.1). The daily ET estimate fr om ensemble method (2010 - 2017) for both locations are shown in Fig. 4.16 and 4.17. The median value of daily box plots for eight years could capture the seasonal ET trend from the measured ET (Fig 4.17, 4.18 and 9a). The ensembled ET in the season were in agreement with maize seasonal crop water demand as physiologically simulated by 29 maize models (Kimball et al., 2019) and for the humid temperate climate conditions similar to study area (Abraha et al., 2016; 2019). 81 Figure 4. 14 . Variation in time series of seasonal ET (monthly average) estimates among the models and ground observations from flux tower during growing season (May - October) at KBS (ET - Average is from ensemble method) 82 Figu re 4. 15 . Variation in time series of seasonal ET (monthly average) estimates among the models and ground observations from flux tower during growing season (May - October) at Marshall (ET - Average is from ensemble method) . 83 Figure 4. 16 . Variation in ensembled daily ET estimates for crop season (May - October) for 2010 - 2017 at KBS 84 Figure 4. 17 . Variation in ensembled daily ET estimates for crop season (May - October) for 2010 - 2017 at Marshall 85 4.3.3 Development of K c Curve The distribution of K c derived from the ensemble ET estimates are shown in Fig 4.19 and 4.20 for the respective KBS and Marshall maize fields for eight years (2010 - 2017). The temporal variation in K c during the early season was due to the varying soil wetness and soil evaporati on. As canopy cover increased during the peak growing season, K c was mostly influenced by the transpiration. In 2012 (a dry year), the K c values for both the locations were lower than the standard FAO - K c , throughout the season. The negative deviation from the standard FAO - K c might be due to scant rainfall during development stage, as ET in that case is limited by the energy availability (Tasumi et al., 2005). Except for the early season, the derived crop coefficients were lower than standard FAO K c - curve in the growing season. The positive deviation from the standard FAO - K c can be attributed to the rainfall events. In the irrigated field, a more positive deviation can be observed during irrigation events (Tasumi et al., 2005). These deviated K c - curves we re fitted according to different development stages (early, developmental, mid and late) similar to the FAO standard. The candidate - wet days were chosen K c values for the wet days were fitted to a t rend line.The fitted curves were generated for all the years and for both locations by aggregating individual fit stages into a seasonal curve (Fig 4.21). The reason for temporal variation in Kc - curve is attributed to our approach of derivation, where we used calibrated CERES - Maize for defining the length of the growth stages (Fig 4.4; Table 4.2). The study area contained two different maize fields with different land - use histories (Section 2.1). 86 Figure 4. 18 . Temporal variation in crop coefficient curve of Maize derived by ensemble ET for 2010 - 2017 at KBS 87 Figure 4. 19 . Temporal variation in crop coefficient curve of Maize derived by ensemble ET for 2010 - 2017 at Marshall 88 Figure 4. 20 . Smooth fitted crop coefficient curve of Maize derived by ensemble ET for 2010 - 2017 at KBS and Marshall 89 The K c curves were averaged temporally (2010 - 2017) for both the locations, with a goal to nullify the effect of temporal variation in the Kc (i.e. interannual variablility) and resulting into the mean state of the Kc (i.e. intraseasonal pattern). Furthermore, th e average K c - curve from the two locations were combined. This was done to introduce the effect of two different replicates of the maize fields representing two different land - use histories. Figure 4. 21 . Averaged (2010 - 2017) crop coefficient curve of Maize at KBS and Marshall Additionally, we compared the standard FAO K c curve with the consolidated K c K c curves were validated using soil water balance model. Across the year, KBS maize farm had the lowest crop coefficient curves among all the curves. This field was under 50 - year corn - soybean rotation before 2010 and hence due to loose soil structure and p oor water storage capability in the root zone, might be one of the reason for the low K c (Abraha et al, 2016). Moreover, Marshall was under grassland for 50 years before 2010 and likely has a higher water 90 storage capacity in the root zone. However, the tem poral stability of soil moisture across long - term agricultural system experiments is a current area of investigation (Abraha et al., 2016; Fry et al., 2018; Zhao et al., 2018) and out of the scope of this study. 4.3.4 Validation of Kc Curve The volumetri c soil moistures were converted to total available water in root zone using soil water parameters (Jagtap et al., 2004) (Fig 4.23b). The point values of crop coefficient curve were fed to the model at the different percent of growth, and model interpolated K c curve from the point values. Table 4. 3 . Crop coefficients as an input for soil water balance model Percent of Growth (%) Root Depth (cm) Maize Crop Coefficients ( K c ) FAO KBS Marshall Average 0 15.2 0.30 0.45 0.43 0.44 10 32.2 0.30 0.45 0.43 0.44 20 49.1 0.30 0.45 0.43 0.44 30 66.0 0.50 0.54 0.56 0.55 40 83.0 0.85 0.70 0.78 0.74 50 91.4 1.20 0.89 1.04 0.97 60 91.4 1.20 0.89 1.04 0.97 70 91.4 1.02 0.89 1.04 0.97 80 91.4 1.08 0.89 1.04 0.97 90 91.4 0.77 0.63 0.68 0.66 100 91.4 0.46 0.38 0.46 0.42 The grower recorded the amount of irrigation applied. We found that the rain gauge in the field had some erroneous results, particularly during heavy rainfall events, due to clogging by insects and dust particles. Hence, with the three rainfall datasets ( Rain gauge, Enviroweather and NASA - POWER), the soil water balance model was run to estimate the total available water (TAW) in root zone (Fig 4.24 a, 4.24b, 4.24c). We compared simulated TAW with observed TAW to evaluate the model performances. 91 Table 4. 4 . Statistical evaluation of soil water balance model performance under three different rainfall inputs by comparing observed and simulated total available water in root zone (mm). The derived Kc performed better than the standard FAO in all cases. However, overall performance of the model was best when reanalysed rainfall values from NASA - POWER were used ( K c - average; d - index - 0.77; R - 0.61) (Table 4.4). We used the validated water balance model for practice of irrigation (Fig 4.25 and 4.26). Rainfall Input K c - type MBE RMSE d - index Correlation (R) Rain Gauge (At field) FAO - K c - 4.21 19.11 0.54 0.33 K c (KBS) - 0.35 14.75 0.66 0.45 K c (Marshall) - 2.56 17.00 0.60 0.39 K c (Average) - 1.48 15.85 0.63 0.42 Weather Station (Enviroweather) FAO - K c - 2.15 13.65 0.67 0.46 K c (KBS) 2.46 12.29 0.70 0.48 K c (Marshall) - 0.65 11.87 0.74 0.55 K c (Average) 0.64 11.55 0.74 0.55 NASA - POWER FAO - K c 2.22 10.80 0.76 0.58 K c (KBS) 3.90 11.20 0.77 0.62 K c (Marshall) 2.86 10.93 0.77 0.61 K c (Average) 3.37 11.04 0.77 0.61 92 Figure 4. 22 . Observed rainfall and irrigation amount with (a) volumetric soil moisture at different depth (15, 30, 45, 60 and 90cm); (b) total available water in root zone (90 cm) (a) (b) 93 Figure 4. 23 . Observed and simulated total available w ater (TAW) in root zone with three rainfall input (a) rain gauge (b) weather station (c) reanalyzed product from NASA - POWER, and irrigation value reported by grower at Manchester maize field for four Kc - curve (FAO, KBS, Marshall, and average) in 2019 maize growing season. 94 Figure 4. 24 . Simulated total available water in root zone using water balance model for two crop coefficients (a) standard FAO - K c (b) K c derived in 2019 maize growing season with 60 % set capacity of available water. 95 4.3.5 Irrigation Scheduling of Maize: a comparison of standard FAO - K c and the derived K c Irrigation scheduling tools, using the concept of reference evapotranspiration and the crop coefficients were initiated by Jensen (1969) and have been continuously evolving. The irrigation - scheduling tool (Martin et al., 1990) used in this study uses the variables: ET ref and K c , along with observed rainfall to estimate total available water in the root zone and allows user to irrigate when the available soil water in root zone rea ches the threshold of percentage capacity filled. We set up 60% threshold for our study and irrigated on the days when available soil water in root zone crosses/nearly - crosses the 60% capacity (Fig 4.25a, 4.25b). The available water holding capacity of the soil (cm water/cm of soil) for our simulation is listed in table 4.5. Table 4. 5 . Available water holding capacity of soil and percent capacity filled at different depth for the sandy clay loam (Jagtap et al., 2004) We compared two crop coefficients (FAO K c and derived K c (average)) to simulate our irrigation scheduling output and found that total irrigation added in the season using FAO - K c was 214 mm whereas using our derived K c , total irrigation applied was 153 mm in a growing season of 2019. We also compared the result s with the irrigation values that farmer had reported and ran our irrigation scheduling tools to investigate his amount and timing of irrigation. We found that although the farmer applied exactly the same amount of 214 mm as simulated by our tool using FAO Kc but the timing was more frequent during post silking and early grain filling periods (Fig. 4.26) and most of them went to drainage. It can be noted that the field also experienced water stress during late grain filling and maturity period after August 27, 2019 (Fig. Depth Range (cm) AW (cm/cm) Capacity filled (%) 0 - 15 0.2286 60 15 - 30 0.2286 80 30 - 45 0.2286 95 45 - 60 0.2286 95 60 - 90 0.2286 95 96 4.26). The results of our comparisons suggest that farmers can minimize the irrigation applications and maximize profitability, if they use our derived crop coefficients. When the FAO K c and derived K c were compared using the irrigation - sche duling tool, the difference in irrigation amount was found to be 60 mm (2.4 inch). And, for 100 - hectare (247 acre) farm, it is equivalent to 70 million litre (16.1 million gallon) excess water in a season, leading to economic and environmental losses. Figure 4. 25 . Simulated total available water in root zone using water balance model for standard FAO - K c % set capacity of available water. 4.4 Summary and Conclusions Improved water management for a crop requires an understanding of daily to seasonal water dynamics in soil - plant - atmospheric continuum. Crop evapotranspiration plays a significant role in the demand side of this flux and accurate es timation of crop evapotranspiration for water management or irrigation scheduling is key. Traditionally, irrigation scheduling has been done 97 by estimating crop water demand from the crop coefficients and reference evapotranspiration. However, spatiotempora l variability in soil and crop type, management practices and other physiographic factors influence the crop coefficients. Characterization of the crop coefficients through remotely sensed vegetation indices or inversely through ET estimate from models hav e gained momentum in the past few decades. Each individual method has its own advantages and limitations related to uncertainty in ET estimation. In this study, we have ensembled multiple ET estimation methods to get reliable ET estimates through a physiol ogically crop stage - based (derived by crop model) Inverse Distance Weighting (IDW). SEBAL tends to overestimate ET across the entire growing season, whereas PT - JPL tends to underestimate. MOD16 ET estimates were closest to the observed variation of ET. We found that ensembled ET method outperformed all individual methods with up to 94% agreement (d - index) and 85% correlation (r) with the measured ET. We used ensembled ET to derive crop coefficient K c - curve and validated those using the water balance model . Kc derived from our study performed better than the standard FAO Kc. Using the derived K c , we designed irrigation scheduling for maize field at Manchester, Michigan where we have installed soil moisture at different depth. Our simulated soil water avai lability in the root zone (90 cm) were in agreement (60 %) with the measured soil water balance in root zone. Based on the results of this study we can draw the following conclusions: The accurate ET estimate from ground based sensors like, eddy covariance aids in water management. Remote sensing aids in ET estimation across spatio - temporal scales in data scarce environments. 98 The energy balance model, SEBAL overestimates ET most days in the season. However, MODIS and PT - JPL underestimates, creating uncertai nty in ET estimates. Ensembling of ET estimates minimizes the uncertainty associated with individual model. Crop models can be used in simulating the growth stages which guides stage based ensembling using Inverse Distance Weighting method based on ET meas urements from the eddy covariance. Ensembled ET derives crop coefficients that are more site specific than standard FAO kc. Derived region/site - specific kc used in soil water balance for irrigation scheduling can save water thus improving profitability an d environmental stewardship. 4.5 Acknowledgements This work is partly funded by project GREEEN (Generating Research and Extension to meet Economic and Environmental Needs) with AgBioResearch, MSU Extension, and the Michigan Department of Agriculture and Rural Development (MDARD). We acknowledge Chubu University, ListenField, USAID - Philippines and NA SA - SERVIR for partly funding 99 5. DOWNSCALING SEASONAL RAINFALL AND TEMPERATURE FORECASTS TO DEVELOP A RISK ANALYSIS MODEL FOR EAR ROT DISEASE MANAGEMENT IN MAIZE 5.1 Introduction The global crop production system remains vulnerable to climate variability despite technological and varietal improvements. Specifically, the variability in climatic extremes has increased over the past few decades and made food production systems more sensit ive and vulnerable to them, especially those under rainfed conditions (Lobell et al., 2008; Hatfield et al., 2011; Kistner et al., 2018; Ogutu et al., 2018). In order to meet the challenges posed by climatic variability, tools for informing in advance asso ciated decisions and preparedness can help growers optimize agronomic management and use of their resources. The availability of climate information within the growing season is relevant in agricultural risk management (Hansen et al., 2006). Yield predict ion under variable and extreme events can provide an opportunity to policy makers to prepare and assist stakeholders develop preemptive actions (Shelia et al., 2019). However, the usefulness of such information among farming communities varies with the tim ing In the Midwest US, maize ( Zea mays L .) is a major crop, producing 14.6 billion bushels from an acreage of 90.2 million acres (USDA - NASS, 2017). Michigan is in the northeast part of the Corn Belt, growing maize mostly under rainfed conditions, with an average yield of 161 bushels per acre as compared to national average of 176 bushel per acre (USDA - NASS, 2017). With a goal of achieving higher yield, there have been continuous improvements in genetics and management in the last few decades (Sacks and Kucharik, 2011). In spite of these improveme nts, maize is sensitive to seasonal climatic variability at different stages of development. These 100 sensitivities ultimately have an effect on yields. Due to highly incentivized maize prices, ther than disease resistant varieties. Combined with continuous production of maize with minimum tillage and erratic weather patterns, conditions became more suitable for fungal diseases outbreaks (Wise and Mueller, 2011). In Michigan, Gibberella and Fusar ium , which produce mycotoxins, are the primarily causal organisms for the development of ear rot disease in maize (Chilvers, 2018). The pathogen spreads through infecting maize ears during early silking stage, approximately 6 to 8 days after silk emergence (Reid et al., 1992; Warfield and Davis, 1996; Munkvold, 2003; Schmale III and Bergstrom, 2004). The ear rot in maize causes quantitative and qualitative losses through producing mycotoxins, like deoxynivalenol (DON; Pitt and Miller, 2016; Mueller et al., 2016). Weather, being an important driver of disease development (Chilvers, 2019), favorable condition during anthesis, helps in its spread. However, there is a debate among maize growers and the scientific community whether infection occurs under environ mental conditions varying from cool and wet (Blaine and Singh, 2018; Wise et al., 2016; Durst and Singh, 2019), with moderate temperature and wet conditions (Chilvers, 2019) to hot and dry conditions (Kaatz, 2018. MSU extension; Crop protection network, 20 19). Specifically, favorable weather C (Sutton, 1982; Miller, 1994; Vigier et al., 2001; Munkvold, 2003; Mansfield et al., 2005). Ambient temperature and moisture are suggested to be the two important indicators of risk for ear rot disease. Vigier et al. (2001) also conducted regression analysis of DON and humidity and Mansfield et al. (2005) evaluated the correlation of weather variable(s) (daily average 101 temperature and humidity) with disease development during tasseling to maturity and found that DON levels and average daily temperatures were positively correlated during tassel ing, silking and milk stages, and negatively correlated with daily precipitation during the blister stage. Studies have also found that cool and wet conditions during grain filling and maturity phase promote the development of DON. The correlations of DON and weather variables are bound to the critical crop growth stages, which are controlled by temperature during the growing season (Ghamghami et al., 2019). Variability in temperature and rainfall during the growing season makes management decisions more ch allenging due to associated variability in the crop growth stages. Although an expensive option, the incidence of ear rot can be minimized by applying fungicide (Chilvers, 2018). However, if any preemptive actions are supplemented by predictions of climat e/weather and associated crop growth stages during silking to maturity, it can better inform growers to improve the efficacy of their fungicide applications or divest from using fungicide depending on the levels of predicted risks (Anderson et al., 2017; R osburg and Menapace, 2018). These critical growth stages and associated management strategies can be predicted and evaluated using process - based crop models (Ritchie and Nesmith, 1991; Rodríguez et al., 2019). Advanced climate information plays a signific ant role in the process of decision making at the earlier crop development stages and lets growers prepare for preemptive actions (Hansen et al., 2006). Unlike deterministic (weather scale) predictions, probabilistic seasonal climate predictions usually ha ve low - moderate skills but have an advantage in terms of greater lead - times. This lead - time advantage can help growers in managing farm related decisions effectively and prepare farm management logistics in advance (Meinke et al., 2004). However, because o f 102 the disparity between the temporal scale of data produced (i.e., three months) and the scale of always straightforward (Hansen et al., 2006). In addition , climate service providers provide SCF in the form of tercile probabilities e.g., probabilities of above normal (AN), near normal (NN) and below normal (BN). With advances in climate applications, this forecast information can be downscaled at the daily b asis to inform process - based crop models simulate phenology and yield (e.g., Hansen and Indeje, 2004; Hansen et al., 2006; Apipattanavis et al., 2010; Shafiee - Jood et al., 2014; Kim et al., 2016; Han et al., 2017). There are several downscaling tools, whi ch can downscale rainfall from the SCF and link the outputs with crop models (e.g., Hansen and Indeje, 2004; Semenov and Doblas - Reyes, 2007; Capa - Morocho et al., 2016; Han and Ines, 2017; Ines et al., 2018; Han et al., 2019, among others). However, there i s a lack of downscaling methods, which can downscale probabilistic rainfall and temperature forecasts from the SCF and link with the crop models (Ines et al., 2018). For maize disease management, temperature is critical to better predict phenology. Current ly, Midwest maize growers use U2U Corn GDD tool to track crop growth and make important decisions prior and during the growing season (Angel et al., 2017; Prokopy et al., 2017). However, the tool only uses historical and forecast temperature (30 - days) and hence cannot predict silking with a lead - time of more than one month. Downscaled temperature and rainfall forecast from SCF when linked with process based physiological crop models can predict phenology with a lead - time of more than a month. The overall goal of this study is to evaluate the performance of downscaling rainfall and temperature in the development of a risk analysis model for ear rot management of maize in Michigan. Specific objectives are: i) to downscale seasonal probabilistic rainfall (P) and 103 temperature (T) forecasts for predicting phenology, ii) to develop a risk analysis model for ear rot management in maize, and iii) to validate the risk analysis model at selected locations in Michigan. 5.2 Materials and Method 5.2.1 Study Area The stu dy area consists of field trial locations from Michigan Maize Performance Trial program (MMPT; Fig. 5.1). The county (city) locations selected for this study were Saginaw (New Lothrop; 43.13 N, 83.97 W), Huron (Bad Axe; 43.83 N, 82.98 W) and Montcalm ( Greenville; 43.22 N, 85.21 W). The risk analysis was conducted in 2017 and 2018 based on disease severity recorded at those three locations (Blaine and Singh, 2018). Figure 5. 1 . Study area having high ear rot disease severity in 2018. 104 5.2.2 Data collection 5.2.2.1 Weather and climate data 5.2.2.1.1 Observed weather data, 2017 - 2018 The weather data for this study (Saginaw, Huron and Montcalm) were obtained from Enviroweather netwo rk, a sustainable weather - based information system that helps growers and stakeholders in making farm related decisions in Michigan. The nearest weather station to Saginaw is Flint at Applewood estate (43.02 N, 83.68 W, 229 m asl), to Huron is Kinde at Mea de piling ground (43.92 N, 83.01 W, 216 m asl) and to Montcalm is Entrica at MSU Montcalm Research farm (43.35 N, 85.18 W, 290 m asl). Figure 5. 2 . Observed weather variables for 2017 and 2018 during crop growing season in th e study area. 105 The daily rainfall, average relative humidity and average temperature data were collected for 2017 and 2018 (Fig. 5.2). Across all locations, the average temperature were higher in July and Augusts in 2018 than 2017 and rainfall was higher (up to 20 mm in a day) in late August and early September in 2018 as compared to 2017 (up to 10 mm in a day), which might contributed to the DON development (Fig. 5.2). 5.2.2.1.2 Historical weather data The historical climate data (1981 - 2010) were obtained from the neare st weather stations of Enviroweather network (https://enviroweather.msu.edu/), and Global Historical Climatology Network (GHCN) Daily Database (Menne et al., 2012). The missing weather data (especially relative humidity) were collected from the reanalysed data from the NASA - POWER, which is available at 0.5 X 0.5 degree grid. The climatology (1981 - 2010) of relative humidity, average temperature and total rainfall for the months of interest for all three locations are shown in Table 5.1. Table 5. 1 . Monthly averages of weather variables (1981 - 2010) in the study area. Month Rainfall (mm) Average Temperature ( C) Relative Humidity (%) S H M S H M S H M May 87.9 79.9 92.9 13.2 10.1 13.1 72.1 78.7 73.4 Jun 87.3 82.0 88.5 18.4 15.9 18.3 73.2 78.4 74.2 Jul 83.1 78.3 78.4 21.0 20.1 20.8 72.1 76.5 72.6 Aug 85.7 81.4 88.9 20.6 20.3 20.4 72.6 76.1 72.4 Sep 78.1 78.9 79.7 17.0 17.0 16.6 71.9 75.4 71.9 Oct 80.2 81.0 96.2 10.1 10.7 9.8 74.8 77.1 75.9 Note: S - Saginaw; H - Huron; M - Montcalm 5.2.2.1.3 Seasonal climate forecast (SCF) data forecast data, which was stored in GRIdded Binary or General Regularly - distributed Information in Binary for m (GRIB2) format at National Digital Forecast Database (NDFD) server (Glahn and 106 Ruth 2003, Ruth et al., 2009). These digital forecasts are prepared and produced in collaboration with Weather Forecast Offices (WFOs), National Centers for Environmental Pred iction (NCEP) and the Central Quality Control unit at National Weather Services (NWS) and verified through a prototype for point - to - point comparison of temperature and rainfall forecast to the observation (Dagostaro et al., 2004; Ruth et al., 2009). The da ta files are organized by headers designed by World Meteorological organization (WMO) and categorized by issuance by time of the day. The study area is in the grid of Central Great Lakes region according to the database of NDFD. The forecast data for 2018 were decoded using GRIB2 decoder (Zeng, 2018). In this study, we used forecasts for June - July - August (JJA) since anthesis and associated ear rot disease incidence coincide mostly during the last week of June to 4 th week of July, depending on planting date. The overarching goal of using SCF is to provide the users a time horizon where they can decide how to incorporate the forecast meaningfully into farm related decisions (Fig. 5.3). The historical datasets were used to predict the climatological ranges of phenology to compare with the observed timing of phenology. Once the crop model (CERES - Maize) is calibrated, the historical observed data can be utilized for regression - based decision making involving the past data. Moreover, the growers can use SCF for f arm related decisions, if these tailored climate - June - July (MJJ) can be used in April to decide the planting date based on the predicted weather conditions. The forecast data for Jun e - July - August (JJA) is important for tactical (in - season) decision making at the farm, which involves decision related to irrigation, fertilizer and pesticides application (fungicides applications in this study). The forecast for the later part of the seas on can be used for post - harvest processing and storage decisions. For example, in ear rot disease management, designing grain moisture management strategies for minimizing the risk of DON can be done by 107 drying seeds after harvest and in storage, as DON con centration is directly proportional to moisture content in kernels (Munkvold, 2014). Figure 5. 3 . Illustration of (a) historical and climate prediction utility for management decisions (b) prediction horizon of seasonal climate forecast and usability for farm related decisions. 5.2.2.2 Ear rot incidence and mycotoxin data Data on ear rot disease inci dence and severity in 2017 and 2018 were collected at each location in Fig. 5.1. The severity of ear rot was calculated as the number of damaged kernels on each damaged ear. An index for ear rot (ERI) was calculated as the product of incidence and severity (Groth et al., 1999). DON data at each location were analyzed from ground samples through 1 mm screen using gas chromatography - mass spectrometry (GC - MS) at the Mycotoxin (b ) (a) 108 Diagnostic Laboratory, University of Minnesota. The Risk Analysis Model evaluates on ly the ear rot disease risk in Saginaw, Huron and Montcalm (Fig. 5.1). 5.2.2.3 Anthesis data Field observation data on anthesis (75% of silk) were collected at each location during period of fungicide (proline) application. These data were compared with estimated data on silking based on accumulation of growing degree days (GDD) using Useful to Usable tool (Angel et al., 2017) (Table 5.2). The GDD are calculated using standard formula (Eq. 1). , (1) where T max is the daily maximum temperature up to 86 F (30 C) and T min is the daily minimum temperature up to 50 F (10 C) and T base is the lowest temperature for maize optimum g rowth 50 F (10 C). The range is based on the optimum temperature range for maize growth, which is 50 F to 86 F (10 C to 30 C). The projection of Maize GDD for next 30 days are based on operational ensemble forecast from National Weather Service (NWS ) (Saha et al., 2014) and the progression of phenological stages are counted based on GDD accumulation explained by Abendroth et al. (2011). Silking can be estimated based on a simple regression (Eq. 2), which varies with the different hybrid seed companie s; (2) where, GDD = Growing Degree Days ( F) CRM = Comparative Relative Maturity rating reported on hybrid maize bags (in days) 109 Table 5. 2 . Field observati on and U2U estimation of silking stage at Saginaw, Huron and Montcalm in 2018. 5.2.3 Downscaling rainfall and temperature: software description Recently, temporal downscaling techniques were developed to disaggregate probabilistic seasonal precipitation (P) outlooks using resampling (FResampler1) and conditional weather generators (predict WTD) (Han and Ines, 2017; Han et al., 2017). The FResampler1 is based on forecasts. This method randomly samples a block of daily time - series of weather data fo r a target season from historical observations conditioned on those tercile probabilities. Sampling is done with replacement. FResampler1 preserves the covariance between rainfall and other weather variables, e.g., minimum and maximum temperature and solar radiation on a particular day (Han and Ines, 2017). The tercile probabilities of rainfall (P) and temperature (T) predictions from SCF were 2018). The goal of do whose union probabilities will preserve their forecast probabilities (Fig. 5.4). It aims to find ch that when Locations Planting date Anthesis (75% silking) Field observation (DAP) Silking (U2U) (DAP) Saginaw May 30 65 65 Huron May 16 72 75 Montcalm May 18 64 67 110 2018). Since we are developing a risk analysis model for ear rot management, we included relative humidity (RH) in the pool of weather varia bles when P&T are being downscaled. Figure 5. 4 . Framework for downscaling probabilistic seasonal rainfall (P) & temperature (T) forecasts that preserves forecast probabilities of rainfall p (P) and temperature p (T) (Ines et al., 2018) Using this downscaling tool, the NOAA three month SCF of June - July - August (JJA) in 2018 were downscaled to generate daily weather variables (i.e., rainfall, minimum and maximum temperature, solar radiation and relative humidity). The long term historical weather data (1981 - 2010) were used as climatology for analysis. The 2018 probabilities for rainfall (42 AN : 34 NN : 24 BN) and temperature (33 AN : 33 NN : 34 BN) for the Central Great Lakes regions were used for this study. The purpose of using JJA in this study was to match the anthesis and ear rot incidence period in maize growth and development. For a normal planting window of May 1 15, tasseling to silking phase could vary from June 25 - July 15 depending on weather conditions and hybrid selection (Abendroth et al., 2010). Moreover, we performed downscaling of four hypothetical extreme forecast scenarios to compare behaviours of the risk analysis tool based on 111 possible weather combinations favourable for disease develeopment. The scenarios were; warm and humid forecast (50 AN : 33 AN: 17 BN for rainfall and temperatu re), the warm and dry forecast (7 AN : 33 AN: 60 BN for rainfall and 60 AN : 33 AN: 7 BN for temperature), the cool and wet forecast (60 AN : 33 AN: 7 BN for rainfall and 7 AN : 33 AN: 60 BN for temperature), and the cool and dry forecast (7 AN : 33 AN: 50 BN for rainfall and temperature). 5.2.4 Crop model setup We used the calibrated DSSAT CERES - Maize in Chapter 1. The SCF was disaggregated into seasonal long term weather data (.WTD) files and converted to yearly weather (.WTH) files as input weather files for DSSAT. We generated 200 realizations (100 P and 100 T) for each forecast scenarios, i.e., 200 different weat her files for each location. Experimental files with 200 treatments and fields were created for each of location (Saginaw, Huron and Montcalm), which used the 200 different weather realizations (each one is assumed to be a weather file for each assigned fi eld in that location). Outputs of the 200 treatments were used to predict anthesis/silking, which was used as a basis in developing the time horizon of the ear rot risk analysis model. 5.2.5 Developing risk indicators Based on previous studies of similar climatic zones (Mansfield et al., 2005; Munkvold, 2014), the follwing thresholds of weather variables were selected: Temperature 1 mm; Relative Humidity 80% during July and August. The tasseling, silking and milking stages are highly correlated with ear rot incidence that fall in the months of July and August (approximately based on May/June planting period). The daily weather variables were assigned a unique flag to identify the days when it crossed the threshold using a simple logic model (e.g., 112 - eet the set conditions, and that day is counted as a risky day. We developed risk indicators for temperature (T), rainfall (R) and relative and T and R with boolean - one can hone in more on detailed risks. The daily risk factors were converted into risk probabilities based on the 200 realizations for the forecast, and from 30 years observations for the climatology. The daily risk probabilities were averaged at five - day intervals in the time series to better match the requiurements of disease progression monitoring (Mansfield et al., 2005). We analyzed and validated risk indicators using disease severity in 2018. Data from 2017 was also used to analyze the reason for lower DON levels during that growing season. 5.2.6 Statistical analysis The 5 - day average risk probabilities led to 12 sets of different risk probabilities for climatology and forecast , respectively. The 12 sets of climatological of risk probabilities were compared with the 12 sets of forecasted risk probabilities for each variable (T/P/RH) and the combinations (T&RH; T&P). In order to test the significance, t - tests were conducted by bo otstrapping the data until 500 replicates (see Hall and Wilson, 1991), using T - Score Calculator ( https://www.socscistatistics.com/pvalues/tdistribution.aspx ) . The p - values were estimated for each variable and their combinations, at 5% and 1% significance levels. 113 5.3 Results and Discussions 5.3.1 Predicting phenology The forecast for 2018 and hypothetical extreme forecast scenarios were downscaled using FResamplerPT and 200 realizations were used to simulate crop growth stages using CERES - Maize model in the DSSAT. The predicted anthesis (box - plots) for all five - forecas t categories (2018; warm and humid; warm and dry; cool and wet; and cool and dry) are shown in Fig. 5.5. The purpose of predicting phenology was to evaluate risk indicators during anthesis from the risk analysis model. At Saginaw, the mean value of predic ted anthesis using NOAA forecast is the same as the observed anthesis in the field (Fig. 5.5a). However, the anthesis were over predicted by four and three days at Huron and Montcalm, respectively (Fig. 5.5.a). Physiologically, the more heat accumulation i n plants due to warm conditions will accelerate the phenological development compared to cooler conditions (e.g., Zhu and Troy, 2018). This observation was found to be true for all forecast scenarios (including actual 2018 forecast) at each location. In Sa ginaw, the mean value of anthesis among 200 realizations were 63 days after planting (DAP) for both the warm and humid, and warm and dry forecast, which were two days earlier than the observed anthesis (65 DAP) (Fig. 5.5b). However, cool conditions with sl ower heat accumulation showed over prediction of anthesis (Fig. 5.5d - e). Specifically, the mean values of predicted anthesis were one day higher (66 DAP) than observed anthesis (65 DAP) at Saginaw, four days higher (76 DAP) than observed (72 DAP) at Huron, and four days higher (68 DAP) than observed (64 DAP) at Montcalm. A similar trend was observed between Huron and Montcalm. Due to its vicinity from the Great Lakes, changes in rainfall and temperature might have significant contributions to the variations in predicting anthesis at Huron, in all forecast scenarios (Fig. 5.5d, 5.5e, 5.5f). 114 Figure 5. 5 . Predicted anthesis for all three locations for 2018 forecast (a), warm and humid forecast (b), warm and dry forecast (c ), cool and wet forecast (d), and cool and dry forecast (e). 115 5.3.2 Risk Probabilities using Forecast data The developed risk probabilities for all the individual variables, and combinations of variables for all forecast scenarios are shown i n Fig. 5.6 - 5.12. The risk probabilities of exceeding the temperature threshold at Saginaw indicated a favorable weather condition for disease development during 3 rd week of July and it might coincide with anthesis, if maize is planted during the normal planting window of May 1 - 15. However, due to the delay in planting (May 30), anthesis were observed on August 3 (65 DAP). The downscaled weather from forecast of 2018 captured the trend of risk probabilities for temperature and was slightly higher than climatology (Fig. 5.6a). The risk probabilities exceeding the temperature threshold for Huron was higher than the climatology during the entire study period and especiall y during post anthesis period (Fig 5.6b). The risk probabilities at Montcalm were higher during anthesis but later in August, risk probabilities were lower (Fig. 5.6c). Statistically, the risk probabilities exceeding the temperature threshold were signific antly higher than climatology for Huron (p = 0.00492) and Montcalm (p = 0.00394) at 99 % confidence level respectively, and for Saginaw (p = 0.04916) at 95 % confidence level (Table 5.3). The risk probabilities exceeding the RH threshold at all locations f ollowed the climatological trend although it was found to be slightly higher during anthesis (Fig. 5.6d - 5.6f) Huron had high climatological and forecast risks probabilities exceeding the RH threshold due to vicinity of the Great Lakes (Fig. 5.7e). At Mon tcalm, post anthesis risk probabilities exceeding RH threshold were found to be sustained through the milking stage (Fig 5.6f). The risk probabilities exceeding the RH threshold were significantly different from climatology across all locations, with 99 % confidence level. 116 Furthermore, the risks probabilities exceeding the rainfall threshold were found to be significantly higher than the climatology during anthesis for all locations (Fig. 5.6g - 5.6i). The risk probabilities exceeding the rainfall threshold in 2018 forecast were found to be highest at Saginaw, especially during anthesis (Fig. 5.6g). In case of rainfall, the risk probabilities exceeding the threshold were found to be no significant difference at Huron (p = 0.33516) compared to climatology (Ta ble 5.3). However, it was found to be significantly higher than climatology for Saginaw (p = 0.00395) and Montcalm (p = 0.00025) at 99% confidence level. In case of combined risks probabilities exceeding the T and RH thresholds, the forecast risks probabi lities were higher at all locations (Fig. 5.7a 5.7c). Specifically, the risk was high in the 3 rd week of July and lower through 1 st week of August followed by the gradual increase in 2 nd and 3 rd week of August. Huron and Montcalm had sustained risk proba bilities during anthesis and post anthesis period (Fig. 5.7b, c). The combined risk probabilities exceeding the threshold of T and RH were found to be significantly different from climatology across all locations at 99 % confidence level. However, the comb ined risk probabilities exceeding the threshold of T and R were non - consistent during the period at Saginaw and Montcalm (Fig. 5.7d - 5.7f) and showed a non - significant difference than climatological risk (Table 5.3). It suggests that the combination of rai nfall and temperature are not a good predictor of ear rot risks. 117 Figure 5. 6 . Risk probabilities for temperature (a - c), relative humidity (RH) (d - f) and rain (g - i) at Saginaw (a,d,g), Huron (b,e,h), and Montcalm (c,f,i ) for 2018 forecast and climatology (1981 - 2010). The red arrow indicates predicted mean anthesis days 118 Figure 5. 7 . Risk probabilities for combination of temperature and relative humidity and combination of temperature and rain at Saginaw (a), Huron (b), and Montcalm (c) for 2018 forecast and climatology (1981 - 2010). The red arrow indicates predicted mean anthesis days 119 For the four extreme scenarios (warm and humid, warm and dry, cool and wet, a nd cool and dry), the case of warm and humid scenario in Huron and Montcalm showed the greater difference in risk probabilities between forecast and climatology than Saginaw (for temperature risk) (Fig. 5.8a). The risk probabilities exceeding the temperatu re threshold were highest in case of warm and dry scenario at all locations (Fig. 5.8b). Warm forecasts had higher risk probabilities exceeding the temperature thresholds, than cool forecasts across all locations, for all variables (Fig. 5.8a - 5.8d), whi ch suggests that warm conditions are more favorable for disease development then cool conditions, corroborating Kaatz (2018) and Chilvers (2019). However, some studies reported cool conditions as more favorable for ear rot disease development (Blaine and S ingh, 2018; Wise et al., 2016; Durst and Singh, 2019). For all the studied forecast extreme scenarios, the risk probabilities exceeding the temperature were significantly higher (warm scenarios) and lower (cool scenarios) than climatology at 99% confidence level, except for cool and dry forecast at Montcalm, where it was non - significant (Table 5.3). The risk probabilities exceeding the RH thresholds were found to be the highest for the extreme forecast scenario of warm and humid (Fig. 5.9a). Warm and dry fo recast showed significantly lower risk than climatology for the risk probabilities exceeding the RH threshold (Fig. 5.9b). The risks were higher than climatology in humid and wet conditions (Fig. 5.9a,c) as compared to dry conditions (Fig. 5.9b,d) during t he season (July - August). It corroborates the findings in several studies (Blaine and Singh, 2018; Wise et al., 2016; Durst and Singh, 2019, Chilvers, 2019) and contradicts the finding by Kaatz (2018). It suggests that humid or wet conditions are favorable for the disease development. For all the forecast scenarios, the risk 120 probabilities exceeding the temperature were significantly higher (Wet/humid scenarios) and lower (dry scenarios) than climatology at 99% confidence level (Table 5.3). The risk probabili ties exceeding the rainfall threshold were lower than the climatological risk in dry scenarios (Fig. 5.10b, d) as compared to humid/wet scenarios (Fig. 5.10a, c). For all the forecast scenarios, the risk probabilities exceeding the rainfall were significan tly higher (Wet/humid scenarios) and lower (dry scenarios) than climatology at 99% confidence level (Table 5.3), except for cool and wet forecasts at Montcalm, which was non - significant (Table 5.3). Based on the individual variable, it is difficult to diag nose the risk of disease as the disease development and progression is influenced by combination of weather variables. In order to analyze the effect of combination of variables, we developed risk probabilities for combined variable of T&RH and T&R. The risk probabilities exceeding the combination of T & RH thresholds were significantly higher than climatological risks in case of warmer scenarios during anthesis period (Fig. 5.11a, b), and lower in case of cooler scenarios (Fig. 5.11c, d). Statistica lly, the forecast risks were significantly different from climatological risks at 99% confidence level across the location (Table 5.3). However, in case of T & R, these were non - significant difference than climatology for all locations except for warm and dry forecast scenario at Saginaw (Table 5.3). Specifically, cooler scenarios (Fig. 5.12c, d) showed lower risk probabilities exceeding the threshold of T & R combined than warmer scenarios (Fig. 5.12a, b). The results of suggests that combined T & RH is be tter predictor/indictor than combined T & R to predict the risk analysis for disease development. 121 Figure 5. 8 . Risk probabilities for temperature at Saginaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) co ol and wet, (d) cool and dry forecast scenarios and climatology (1981 - 2010) for all three locations. The red arrow indicates predicted mean anthesis days. 122 Figure 5. 9 . Risk probabilities for relative humidity (RH) at Saginaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) cool and wet, (d) cool and dry forecast scenarios and climatology (1981 - 2010) for all three locations. The red arrow indicates predi cted mean anthesis days. 123 Figure 5. 10 . Risk probabilities for rain at Saginaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) cool and wet, (d) cool and dry forecast scenarios and climatology (1981 - 2010) f or all three locations. The red arrow indicates predicted mean anthesis days. 124 Figure 5. 11 . Risk probabilities for combination of temperature and relative humidity (T&RH) at Saginaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) cool and wet, (d) cool and dry forecast scenarios and climatology (1981 - 2010) for all three locatio ns. The red arrow indicates predicted mean anthesis days. 125 Figure 5. 12 . Risk probabilities for combination of rain and temperature and rain (R&T) at Saginaw, Huron and Montcalm for (a) warm and humid, (b) warm and dry, (c) cool and wet, (d) cool and dry forecast scenarios and climatology (1981 - 2010) for all three locations. The red arrow indicates predicted mean anthesis days. 126 Table 5. 3 . Statistical significance of forecast risk probabilitie s compared to climatological risks. Location Weather Variables for risk analysis 2018 - Forecast Warm and Humid Forecast Warm and Dry Forecast Cool and Wet Forecast Cool and Dry Forecast Saginaw Temperature 0.04916 * 0.00534 ** 0.00375 ** 0.00736 ** 0.00329 ** RH 0.00429 ** 0.00387 ** 0.00695 ** 0.00915 ** 0.00726 ** Rainfall 0.00395 ** 0.00036 ** 0.00067 ** 0.00062 * 0.00473 ** Temp & RH 0.00384 ** 0.00628 ** 0.00478 ** 0.00052 * 0.00154 ** Temp & Rainfall 0.06891 ns 0.05789 ns 0.04697 * 0.08453 ns 0.06439 ns Huron Temperature 0.00492 ** 0.00316 ** 0.00537 ** 0.00331 ** 0.00310 ** RH 0.00363 ** 0.00128 ** 0.00925 ** 0.00557 ** 0.00289 ** Rainfall 0.33516 ns 0.00023 ** 0.00817 ** 0.00385 ** 0.00516 ** Temp & RH 0.00451 ** 0.00429 ** 0.00358 ** 0.03486 * 0.00159 ** Temp & Rainfall 0.06876 ns 0.06595 ns 0.06697 ns 0.08474 ns 0.06923 ns Montcalm Temperature 0.00394 ** 0.00458 ** 0.00607 ** 0.00639 ** 0.37025 ns RH 0.00416 ** 0.00457 ** 0.00495 ** 0.00377 ** 0.00439 ** Rainfall 0.00025 ** 0.00032 ** 0.00407 ** 0.09204 ns 0.00028 ** Temp & RH 0.00034 ** 0.00482 ** 0.00658 ** 0.00565 ** 0.01948 * Temp & Rainfall 0.07129 ns 0.10017 ns 0.06754 ns 0.07519 ns 0.06193 ns * p value at p < 0.05; ** p value at p < 0.01; ns Non - significant; RH is daily average relative humidity and temp is daily average temperature. 127 5.3.3 Validation of risk analysis model The developed risk analysis model was validated using observed weather conditions in 2018 and 2017. The DON concentrations were found to be elevated in 2018 than 2017 (Fig. 5.11; Blaine et al., 2018). The DON concentration > 2 µg g 1 was considered as a threshold for disease severity in this study. The high DON concentration at Washtenaw in 2017 (Fig. 5.11) could be attributed by insect infestation and is out of the scope of this study (Parker et al., 2017; Blaine et al., 2018). Figure 5. 13 . DON concertation at each location in 2017 and 2018; Source: Blaine et al., (2018) The observed weather data (rainfall, temperature and RH) in 2018 were used to validate the developed risk analysis model. We also used 2017 to compare and contrast the weather conditions that elevated the risks in 2018 and low disease incidence in 2017. Simila r to risk 128 probabilities, the observed weather variables from July 1 to August 31 were used for validation, as this period coincides with the pre - and post - anthesis phase and disease developement (Munkvold, 2014; Mansfield et al., 2005). The in - vitro risk a nalysis to study the effects on environmental conditions on mycotoxin production by Marin et al (1999) suggested that warm and moist conditions supported by rain (Munkvold, 2014) during anthesis help in disease spread. Our results of risk indicator were in agreement with the findings (Fig. 5.12). For the risk analysis, 0 (no risk) to 1 (high risk) risk factor was used. In Saginaw, the frequency of risk factor due to RH in 2017 was greater than 2018, however, sustained risk of high RH supplemented by sustai ned high risk due to rainfall in the first week of August made environmental conditions more favorable for ear rot incidence in 2018 (Fig. 5.12a, 5.12c) (e.g., Mansfield et al., 2005). The high temperature risk and the combination of T and RH as well as T and R showed high risk during the same periods (Fig. 5.12d, 5.12e). The early grain filling period in the 2 nd and 3 rd week of August received high rainfall that might had exacerbated the disease severity and hence DON concentration (Munkvold, 2014). During anthesis, Huron and Montcalm had experienced sustained higher risks due to RH and rainfall in 2018 as compared to 2017 (Fig. 5.12a, 5.12c). However, the combination of T and RH, as well as T and R, during last week of August may have influenced DON concen trations (Vieger et al., 2001). Across all locations, the risks due to combined rainfall and temperature were high during early grain filling period in 2018 and no risk at all in 2017 in case of combined risk of T and RH and T and R. (Fig. 5.12d, 5.12e) (e .g., Vigier et al., 2001; Munkvold, 2014). Rainfall occurence during anthesis followed by dry periods helps in accumulation of pathogens producing mycotoxins (Miller et al., 2007). Moreover, late rainfall after milking (late August to early September) migh t be a plausible 129 reason for high mycotoxin concentrations in 2018 at all locations (Fig. 5.2) (Munkvold, 2014; Battilani and Logrieco, 2014) In the last two decades, there have been some development in prototype models for predicting DON in maize. However, none of them could validate the threshold and duration of g to harvesting were used to predict ear rot in maize but they can not be validated (Battilani et al., 2003; Maiorano et al., 2009). Schaafsma and Hooker (2007) developed DONcast model for wheat and maize to predict ear rot and correlated temperature and r ainfall in the period of 10 days prior and 14 days post silking, and found that rainfall > 2mm is favorable for disease development. However, there is a general agreement that it is difficult to predict the disease spread based on the use of only weather variables, as ears are exposed to insect damage too (Parker et al., 2017). It is difficult to predict the most accurate combinations of variable and their thresholds, which are favorable to the disease development. The developed risk analysis model can inform the growers to optimize their management strategies during critical cro p growth stages using forecast information at a lead - time of 3 months and more. The associated management decisions can be optimized through prediction of anthesis using the calibrated crop growth model (CERES - Maize) and seasonal climate forecast. The mana gement strategies include, the selection of hybrids, crop rotation (Mabuza et al., 2018), tillage practice (Craven and Nel., 2017), planting date and density (Blandino et al., 2008) can be informed using forecast and risk analysis model for risk aversion of ear rot disease. In - season tactical decision include irrigation (Gxasheka et al., 2015), and insect control can also inform using advance climate information and risk analysis model (Munkvold, 2014; Owour et al., 2015). 130 Figure 5. 14 . Risk factor of weather variable(s) in predicting ear rot disease for Saginaw, Huron and Montcalm (a) relative humidity (RH), (b) temperature (T), (c) rainfall (R), (d) RH & T, (e) R & T in 2017 and 2018. (a) (b) (c) (d ) (e) 131 5.4 Summary and Conclusions Climate - sensitive decisions are bound to the horizon of available climate information and hence the usefulness varies with growers and their decisions. The management decisions in maize depend on crop growth stages, based on heat accumulation during the gr owing season. Prior information on rainfall and temperature in a season helps in predicting growth stages. In this study, we used the calibrated crop model, CERES - Maize to simulate anthesis using downscaled seasonal climate predictions to evaluate ear rot disease management in corn. The tercile - based seasonal climate forecasts (rainfall and temperature) from NOAA were downscaled at the daily time scale to feed the crop model. An innovative downscaling tool, FResamplerPT, was used in this study, which downsc ales rainfall and temperature simultaneously. Using the predicted phenology, thresholds of weather variables (rainfall, temperature and humidity), and observed data on disease occurrence, a risk analysis model was developed for ear rot management in corn. Risk analysis model was validated for 2018 observed data and we found that risk of disease were more in 2018 due to combined effect of temperature and RH during and post anthesis period. The rainfall during late milking period also had influence on the dis ease development. The results from the extreme forecasts suggested that warm forecast scenarios had higher risk than cool forecast scenarios across the location for all the variables. Based on the results of this study we can draw the following conclusions : The calibrated crop model (CERES - Maize) can predict phenology of maize using rainfall and temperature data from the forecast using a new downscaling tool, FResamplerPT. The FresamplerPT could downscale the realization of the extreme and normal probabilit ies of rainfall and temperature data. 132 The risk analysis model was developed based on threshold of weather variables August, which coincides with the disease onset during anthe sis. For all the studied forecast extreme scenarios, the risk probabilities exceeding the temperature were significantly higher (warm scenarios) and lower (cool scenarios) than climatology at 99% confidence level, except for cool and dry forecast at Montc alm, where it was non - significant. The temperature and relative humidity are the best indicators for risk analysis. 5.5 Acknowledgement Authors would like to thank Corn Marketing Program of Michigan (CMPM) and MSU AgBioResearch for funding this work. We ac knowledge Chubu University, ListenField, USAID - Philippines and NASA - 133 6. CONCLUSION S Throughout this dissertation, three main topics were explored. The first topic explored different methods to est imate genetic coefficients of CERES - Maize and evaluate their performance simulating phenology and yield. The second topic explored the multiple model ensemble of ET and derivation of crop coefficient s for maize irrigation scheduling. The third topic explor ed the development of risk analysis model using predicted phenology of the above calibrated crop model and seasonal climate forecast. The first and third topics were examined at different locations from the Michigan Maize Performance Trial program, and the second topic was examined at two fields under continuous maize rotation with eddy covariance flux towers, situated in Kalamazoo, Michigan. The resulting region - specific kc - curves were evaluated and In t he first study, the new method, Noisy Monte Carlo Genetic Algorithm (NMCGA) was introduced for estimating genetic coefficients, and compared with existing methods, Genotype Coefficient Calculator (GENCALC) and G eneralized Likelihood Uncertainty Estimation ( GLUE). The coefficients from the ensemble of multiple methods and individual methods were evaluated for prediction of phenology and yield. From this study, the major conclusions drawn are as follows: The multi - model ensemble of genetic coeffici ents minimized the biases associated with individual methods. The ensembled genetic coefficients improved predictions of phenology and yields. Under water stressed conditions, soil root growth factor needs to be adjusted to allow hybrid maize to explore a vailable water and nutrients in the soil profile. 134 In the second study, the multi - model ensemble of ET estimates (RS - based SEBAL, Priestley - Taylor - JPL, MODIS Penman - Monteith and a ground - based empirical model, FAO - kc) were used to derive crop coefficients a t two locations in Michigan for water management. This study also evaluated , if standard FAO - Kc values are generally applicable in all situations. The derived crop coefficients were compared with standard FAO - Kc approach for irrigation scheduling. From thi s study, the major conclusions drawn are as follows: The calibrated crop model simulated crop development stages can be used to map crop coefficients ( K c ) to reflect the genetic characteristics of the maize hybrid. The multi - model ensemble of ET estimates reduced the uncertainty and biases associated with individual methods, hence, gave better information in the development of crop coefficients ( K c ). The standard FAO based K c curve for maize cannot be generally applicable in all situations hence, it must be derived for specific locations. The derived crop coefficients can improve irrigation scheduling, minimizing losses due to over irrigation. In the third study, the calibra ted model was used to predict anthesis using seasonal climate forecast. The innovative downscaling tool, FResamplePT, which downscales probabilities of rainfall and temperature from seasonal climate forecast was used in this study. Based on the forecast in formation, risk probabilities exceeding the thresholds of weather variables (temperature, rainfall, relative humidity and combination of these) were developed, which indicates favorable conditions to ear rot infection in maize. The risk analysis model deve loped on this basis was validated in 2018. From this study, the major conclusions drawn are as follows: 135 The calibrated crop model can predict phenology using rainfall and temperature forecasts dow n scaled by FResamplerPT. Forecast - based ear rot risk probabi rot in maize. Temperature and relative humidity are the best indictors for ear rot disease risk assessments. 136 7. FUTURE RESEARCH RECOMMENDATION S Possible avenues for future research are as follows: Field experiments to study water stress at different soil depths should be performed to validate the veracity of our hypothesis on the adjustment of root growth factor under water - stressed conditions. The NMCGA can be integrated in DSSAT to provid e users more multi - method comparison and ways for ensembling genetic coefficients. Open - source ET ensemble framework for irrigation planning can be developed. In depth study understanding why SEBAL overestimated ET and PT - JPL underestimated ET. Using ECOST RESS to estimate site - specific kc - curves. Linking sub - seasonal to seasonal (S2S) with the ear rot risk assessment model. Develop a software tool that integrates the crop model, climate forecast and ear rot disease assessment model for decision support to growers. 137 REFERENCES 138 REFERENCE S Abas, F., Ismail, I. S., & Lajis, N. H. (2018). Generalized Likelihood Uncertainty Estimation (GLUE) methodology for optimization of extraction in natural products. Food Chemistry , 250: 37 - 45. Abendroth, L. J., Elmore, R. W., Boyer, M. J., & Marlay, S. K. (2010). Understanding corn development: A key for successful crop management. rates for rainfed corn vs. perennial bioenergy crops in a mesic landscape. Hydrolog ical Processes . Abraha, M., Gelfand, I., Hamilton, S. K., Shao, C., Su, Y. J., Robertson, G. P., & Chen, J. (2016). Ecosystem water - use efficiency of annual corn and perennial grasslands: contributions from land - use history and species composition. Ecosyst ems , 19 (6), 1001 - 1012. Abtew, W., & Melesse, A. (2013). Crop Yield Estimation Using Remote Sensing and Surface Energy Flux Model. In Evaporation and Evapotranspiration (pp. 161 - 175). Springer, Dordrecht. Adnan, A.A., Diels, J., Jibrin, J.M., Kamara, A.Y., Craufurd, P., Shaibu, A.S., Mohammed, I.B. and Tonnang, Z.E.H. (2019). Options for calibrating CERES - maize genotype specific parameters under data - scarce environments. PloS one , 14 (2), p.e0200118. Ahmed, F., Choudhury, A.K., Akhter, S., Aziz, M.A., Biswas, J.C., Maniruzzaman, M., Miah, M.M.U., Rahman, M.M., Jahan, M.A.H.S., Ahmed, I.M. and Sen, R. (2017). Calibration and Validation of Decision Support System for Agro - Technology Transfer Model for Simulating Growth and Yield of Maize in Bangladesh. Am erican Journal of Plant Sciences , 8 (07), p.1632. Allen, R. G. (2000). Using the FAO - 56 dual crop coefficient method over an irrigated region as part of an evapotranspiration intercomparison study. Journal of Hydrology , 229 (1 - 2), 27 - 41. Allen, R. G., Pereir FAO Irrigation and Drainage Paper 56 , Food and Agricultural Organization of the United Nations, Rome. Allen, R. G., Tasumi, M., & Trezza, R. (2007). Satellite - based energy balance for map ping evapotranspiration with internalized calibration (METRIC) Model. Journal of irrigation and drainage engineering , 133 (4), 380 - 394. Allen, R., Irmak, A., Trezza, R., Hendrickx, J. M., Bastiaanssen, W., & Kjaersgaard, J. (2011). tion in agriculture using SEBAL and METRIC. Hydrological Processes , 25 (26), 4011 - 4027. 139 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 in frared remote sensing. Remote sensing of environment , 60 (2), 195 - 216. Anderson, M., Gao, F., Knipper, K., Hain, C., Dulaney, W., Baldocchi, D., Eichelmann, E., Hemes, K., Yang, Y., Medellin - Azuara, J. and Kustas, W. (2018). Field - scale assessment of land a nd water use change over the California Delta using remote sensing. Remote Sensing , 10 (6), p.889. Anderson, N. R., Romero Luna, M. P., Ravellette, J. D., & Wise, K. A. (2017). Impact of Foliar Fungicides on Gibberella Ear Rot and Deoxynivalenol Levels in I ndiana Corn. Plant health progress , 18 (3), 186 - 191 Andrea, M. C. D. S., Boote, K. J., Sentelhas, P. C., & Romanelli, T. L. (2018). Variability and limitations of maize production in Brazil: Potential yield, water - limited yield and yield gaps. Agricultural Systems , 165, 264 - 273. Andresen, J. A., Alagarswamy, G., Rotz, C. A., Ritchie, J. T., & LeBaron, A. W. (2001). Weather impacts on maize, soybean, and alfalfa production in the Great Lakes region, 1895 1996. Agronomy Journal , 93(5), 1059 - 1070. Andresen, J., L. Olsen, T. Aichele, B. Bishop, J. Brown, J. Landis, S. Marquie, and A. Pollyea (2012). Enviro - weather: A Weather - based pest and crop management information system for Michigan. Proc. 7th International Integrated Pest Management Symposium, Memphis, TN , 2 7 - 29 March 2012. Angel, J. R., Widhalm, M., Todey, D., Massey, R., & Biehl, L. (2017). The U2U corn growing degree day tool: Tracking corn growth across the US Corn Belt. Climate Risk Management , 15, 73 - 81. Anothai, J., Patanothai, A., Jogloy, S., Pannangp etch, K., Boote, K. J., & Hoogenboom, G. (2008). A sequential approach for determining the cultivar coefficients of peanut lines using end - of - season data of crop performance trials. Field Crops Research , 108(2), 169 - 178. Apipattanavis, S., Bert, F., Podestá, G., & Rajagopalan, B. (2010). Linking weather generators and crop models for assessment of climate forecast outcomes. Agricultural and forest meteorology , 150 (2), 166 - 174. AQUASTAT (2019) Food and Agriculture Organization of the United Nations (FA O). AQUASTAT. Database http://www.fao.org/nr/water/aquastat/main/index.stm Asseng, S., Ewert, F., Rosenzweig, C., Jones, J.W., Hatfield, J.L., Ruane, A.C., Boote, K.J., Thorburn, P.J., Röt ter, R.P., Cammarano, D. and Brisson, N., (2013). Uncertainty in simulating wheat yields under climate change. Nature Climate Change , 3(9), p.827. Balboa, G.R., Archontoulis, S.V., Salvagiotti, F., Garcia, F.O., Stewart, W.M., Francisco, E., Prasad, P.V. a nd Ciampitti, I.A. (2019). A systems - level yield gap assessment of maize - soybean rotation under high - and low - management inputs in the Western US Corn Belt using APSIM. Agricultural Systems , 174 , pp.145 - 154. 140 Baldocchi, D. D. (2019). How eddy covariance flux measurements have contributed to our understanding of Global Change Biology. Global change biology . Bao, Y., Hoogenboom, G., McClendon, R. W., & Paz, J. O. (2015). Potential adaptation strategies for rainfed soybean production in the southeastern USA unde r climate change based on the CSM - CROPGRO - Soybean model. The Journal of Agricultural Science , 153(5), 798 - 824. Bao, Y., Hoogenboom, G., McClendon, R., & Vellidis, G. (2017). A comparison of the performance of the CSM - CERES - Maize and EPIC models using maize cultivar trial data. Agricultural Systems , 150, 109 - 119. Basso, B., & Ritchie, J. T. (2018). Evapotranspiration in high - yielding maize and under increased vapor pressure deficit in the US Midwest. Agricultural & Environmental Letters , 3 (1). Basso, B., (20 19). Methods for mapping temporal and spatial stability and sustainability of a cropping system . U.S. Patent Application 16/343,933. Basso, B., Dumont, B., Maestrini, B., Shcherbak, I., Robertson, G.P., Porter, J.R., Smith, P., Paustian, K., Grace, P.R., Asseng, S. and Bassu, S. (2018). Soil organic carbon and nitrogen feedbacks on crop yields under climate change. Agricultural & Environmental Letters , 3 (1). Bassu, S., Brisson, N., Durand, J.L., Boote, K., Lizaso, J., Jones, J.W., Rosenzweig, C., Rua ne, A.C., Adam, M., Baron, C. and Basso, B. (2014). How do various maize crop models vary in their responses to climate change factors?. Global change biology , 20 (7), pp.2301 - 2320. Bastiaanssen, W. G. M., Noordman, E. J. M., Pelgrum, H., Davids, G., Thores on, B. P., & Allen, R. G. (2005). SEBAL model with remotely sensed data to improve water - resources management under actual field conditions. Journal of irrigation and drainage engineering , 131 (1), 85 - 93. Bastiaanssen, W. G., Menenti, M., Feddes, R. A., & H oltslag, A. A. M. (1998). A remote sensing surface energy balance algorithm for land (SEBAL). 1. Formulation. Journal of hydrology , 212 , 198 - 212. Bastiaanssen, W. G., Molden, D. J., & Makin, I. W. (2000). Remote sensing for irrigated agriculture: examples from research and possible applications. Agricultural water management , 46 (2), 137 - 155. Battilani, P., & Logrieco, A. F. (2014). 22 Global Risk Maps for Mycotoxins in Wheat and Maize. Mycotoxin Reduction in Grain Chains , 309. Battilani, P., Rossi, V., & Pi etri, A. (2003). Modelling Fusarium verticillioides infection and fumonisin synthesis in maize ears. Beven, K. J. (2018). On hypothesis testing in hydrology: Why falsification of models is still a really good idea. Wiley Interdisciplinary Reviews: Water . D OI: 10.1002/wat2.1278. 141 Beven, K., & Binley, A. (1992). The future of distributed models: model calibration and uncertainty prediction. Hydrological Processes , 6(3), 279 - 298. Blandino, M., Reyneri, A., & Vanara, F. (2008). Effect of plant density on toxigen ic fungal infection and mycotoxin contamination of maize kernels. Field Crops Research , 106 (3), 234 - 241. Bolvin, D. T., Adler, R. F., Huffman, G. J., Nelkin, E. J., & Poutiainen, J. P. (2009). Comparison of GPCP monthly and daily precipitation estimates wi th high - latitude gauge observations. Journal of Applied Meteorology and Climatology , 48 (9), 1843 - 1857. Boote, K. J., Jones, J. W., Batchelor, W. D., Nafziger, E. D., & Myers, O. (2003). Genetic coefficients in the CROPGRO soybean model. Agronomy Journal , 9 5(1), 32 - 51. Boote, K. J., Jones, J. W., Hoogenboom, G., & White, J. W. (2010). The role of crop systems simulation in agriculture and environment. International Journal of Agricultural and Environmental Information Systems (IJAEIS) , 1(1), 41 - 54. Bowen, I. S. (1926). The ratio of heat losses by conduction and by evaporation from any water surface. Physical review , 27 (6), 779. Buddhaboon, C., Jintrawet, A., & Hoogenboom, G. (2018). Methodology to estimate rice genetic coefficients for the CSM - CERES - Rice mode l using GENCALC and GLUE genetic coefficient estimators. The Journal of Agricultural Science , 156(4), 482 - 492. Bulatewicz, T., Jin, W., Staggenborg, S., Lauwo, S., Miller, M., Das, S., Andresen, D., Peterson, J., Steward, D.R. and Welch, S.M., 2009. Calibr ation of a crop model to irrigated water use using a genetic algorithm. Hydrology and Earth System Sciences , 13(8), pp.1467 - 1483. Calmon, M. A., Jones, J. W., Shinde, D., & Specht, J. E. (1999). Estimating parameters for soil water balance models using ada ptive simulated annealing. Applied Engineering in Agriculture , 15(6), 703. Cammalleri, C., Anderson, M. C., & Kustas, W. P. (2014). Upscaling of evapotranspiration fluxes from instantaneous to daytime scales for thermal remote sensing applications. Hydrology and Earth System Sciences , 18(5), 1885 - 1894. Campbell, E. P., Fox, D. R. , & Bates, B. C. (1999). A Bayesian approach to parameter estimation and pooling in nonlinear flood event models. Water Resources Research , 35(1), 211 - 220 Candela, A. N. G. E. L. A., Noto, L. V., & Aronica, G. (2005). Influence of surface roughness in hydr ological response of semiarid catchments. Journal of Hydrology , 313(3), 119 - 131. Capa - Morocho, M., Ines, A. V., Baethgen, W. E., Rodríguez - Fonseca, B., Han, E., & Ruiz - Ramos, M. (2016). Crop yield outlooks in the Iberian Peninsula: Connecting seasonal clim ate forecasts with crop simulation models. Agricultural systems , 149 , 75 - 87. Carberry, P. S., Muchow, R. C., & McCown, R. L. (1989). Testing the CERES - Maize simulation model in a semi - arid tropical environment. Field Crops Research , 20(4), 297 - 315. 142 Chen, J ., Brissette, F. P., Zhang, X. J., Chen, H., Guo, S., & Zhao, Y. (2019). Bias correcting climate model multi - member ensembles to assess climate change impacts on hydrology. Climatic Change , 153 (3), 361 - 377. Chen, W., Huang, C., Shen, H., & Li, X. (2015). C omparison of ensemble - based state and parameter estimation methods for soil moisture data assimilation. Advances in Water Resources , 86, 425 - 438. Chen, Y., Xia, J., Liang, S., Feng, J., Fisher, J.B., Li, X., Li, X., Liu, S., Ma, Z., Miyata, A. and Mu, Q. ( 2014). Comparison of satellite - based evapotranspiration models over terrestrial ecosystems in China. Remote Sensing of Environment , 140 , pp.279 - 293. Chilvers, (2019). Michigan Farm News, Accessed at: https://www.michiganfarmnews.com/develop - a - management - strategy - for - corn - ear - rots - and - mycotoxins Clement, R. (1999). EdiRe data software. Edinburgh , Scotland: The University of Edinburgh . Comas, L. H., Trout, T. J., DeJonge, K. C., Zhang, H., & Gleason, S. M. (2019). Water productivity under strategic growth stage - based deficit irrigation in maize. Agricultural water management , 212 , 433 - 440. Confalo nieri, R., Bregaglio, S., & Acutis, M. (2016). Quantifying uncertainty in crop model predictions due to the uncertainty in the observations used for calibration. Ecological Modelling , 328, 72 - 77. Cooper, M., van Eeuwijk, F. A., Hammer, G. L., Podlich, D. W ., & Messina, C. (2009). Modeling QTL for complex traits: detection and context for plant breeding. Current Opinion in Plant Biology , 12(2), 231 - 240. Corbari, C., Ravazzani, G., Galvagno, M., Cremonese, E., & Mancini, M. (2017). Assessing crop coefficients for natural vegetated areas using satellite data and eddy covariance stations. Sensors , 17 (11), 2664. Craven, M., & Nel, A. A. (2017). Effect of conservation agriculture associated crop rotation systems on root and crown rot severity and respective soil - b orne pathogens of maize (Zea mays L.) in the Highveld area of South Africa. South African Journal of Plant and Soil , 34 (2), 87 - 95. Crop Protection Network, (2019), Accessed at ( https://cropprotectionnetwork.org/resources/articles/diseases/gibberella - ear - rot - of - corn Cross, H. Z., & Zuber, M. S. (1972). Prediction of Flowering Dates in Maize Based on Different Methods of Estimating Thermal Units 1. Agronom y Journal , 64(3), 351 - 355. Dagostaro, V.J., Shaffer, W.A., Schenk, M.J., Gorline, J.L. and Taylor, A.A., 2004, January. A prototype verification system for examining NDFD forecasts. In Preprints, 17th Conf. on Probability and Statistics in the Atmospheric Sciences, Seattle, WA, Amer. Meteor. Soc., CD - ROM (Vol. 2). Dai, C., Yao, M., Xie, Z., Chen, C., & Liu, J. (2009). Parameter optimization for growth model of greenhouse crop using genetic algorithms. Applied Soft Computing , 9(1), 13 - 19. 143 DeJonge, K. C., & T horp, K. R. (2017). Implementing standardized reference evapotranspiration and dual crop coefficient approach in the DSSAT cropping system model. Transactions of the ASABE , 60 (6), 1965 - 1981. Djaman, K., & Irmak, S. (2012). Actual crop evapotranspiration an d alfalfa - and grass - reference crop coefficients of maize under full and limited irrigation and rainfed conditions. Journal of Irrigation and Drainage Engineering , 139(6), 433 - 446. D'Odorico, P., Davis, K.F., Rosa, L., Carr, J.A., Chiarelli, D., Dell'Angelo , J., Gephart, J., Reviews of geophysics , 56 (3), pp.456 - 531. assessment of groundwater depletion and related groundwater abstractions: Combining hydrological modeling with information from well observations and GRACE satellites. Water Resources Research , 50 (7), 5698 - 5720. Du Toit, A. S. (2002). Comparisons of using fitted (calculated) and determined (measured) genetic coefficient G2 in CERES - Maize. South African Journal of Plant and Soil , 19(4), 208 - 210. Durst and Singh, (2019), MSU Extension. Accessed at https://www.canr.msu.edu/news/farmers - invited - to - submit - corn - silage - samples - for - mycotoxin - study Duvick, D. N. (2005). The contribution of breeding to yield advances in maize (Zea mays L.). Advances in Agron omy , 86, 83 - 145. Dzotsi, K. A., Basso, B., & Jones, J. W. (2015). Parameter and uncertainty estimation for maize, peanut and cotton using the SALUS crop model. Agricultural Systems , 135, 31 - 47. Egli, D. B. (2008). Comparison of corn and soybean yields in the United States: Historical trends and future prospects. Agronomy Journal , 100(Supplement_3), S - 79. Elias, A. A., Robbins, K. R., Doerge, R. W., & Tuinstra, M. R. (2016). Half a century of stu dying genotype× environment interactions in plant breeding experiments. Crop Science , 5 Er - Raki, S. A. L. A. H., Chehbouni, A., Guemouria, N., Ezzahar, J., Khabba, S., Boulet, G., & Hanich, L. (2009). Citrus orchard evapotranspiration: comparison between e ddy covariance measurements and the FAO - 56 approach estimates. Plant Biosystems , 143 (1), 201 - 208. Ershadi, A., McCabe, M. F., Evans, J. P., Chaney, N. W., & Wood, E. F. (2014). Multi - site evaluation of terrestrial evaporation models using FLUXNET data. Agricultural and Forest Meteorology , 187 , 46 - 61. Ershadi, A., McCabe, M. F., Evans, J. P., Chaney, N. W., & Wood, E. F. (2014). Multi - site evaluation of terrestrial evaporation models using FLUXNET data. Agricultural and Forest Meteorology , 187 , 46 - 6 1. 144 Facchi, A., Gharsallah, O., Corbari, C., Masseroni, D., Mancini, M., & Gandolfi, C. (2013). Determination of maize crop coefficients in humid climate regime using the eddy covariance technique. Agricultural Water Management , 130 , 131 - 141. FAOSTAT (2019) Food and Agriculture Organization of the United Nations (FAO). FAOSTAT. Database. http://faostat.fao.org/site/291/default.aspx Farah, H. O., Bastiaanssen, W. G. M., & Feddes, R. A. (2004). Eval uation of the temporal variability of the evaporative fraction in a tropical watershed. International Journal of Applied Earth Observation and Geoinformation , 5 (2), 129 - 140. Fausti, Scott W. "The causes and unintended consequences of a paradigm shift in co rn production practices." Environmental Science & Policy 52 (2015): 41 - 50. Fisher, J. B., Tu, K. P., & Baldocchi, D. D. (2008). Global estimates of the land atmosphere water flux based on monthly AVHRR and ISLSCP - II data, validated at 16 FLUXNET sites. Rem ote Sensing of Environment , 112 (3), 901 - 919. Fisher, J. B., Whittaker, R. J., & Malhi, Y. (2011). ET come home: potential evapotranspiration in geographical ecology. Global Ecology and Biogeography , 20 (1), 1 - 18. Fisher, J.B., Melton, F., Middleton, E., Hai n, C., Anderson, M., Allen, R., McCabe, M.F., Hook, S., Baldocchi, D., Townsend, P.A. and Kilic, A., (2017). The future of evapotranspiration: Global requirements for ecosystem functioning, carbon and climate feedbacks, agricultural management, and water r esources. Water Resources Research , 53(4), pp.2618 - 2626. Foley, J.A., Ramankutty, N., Brauman, K.A., Cassidy, E.S., Gerber, J.S., Johnston, M., Mueller, cultivated planet. Na ture , 478(7369), p.337. Fraisse, C. W., Sudduth, K. A., & Kitchen, N. R. (2001). Calibration of the CERES MAIZE model for simulating site specific crop development and yield on claypan soils. Applied Engineering in Agriculture , 17(4), 547. Fry, J., Guber, A., & Kravchenko, A. N. (2018, December). The Role of Surface Topography in Temporal Stability of Field - Scale Patterns of Soil Water Content. In AGU Fall Meeting Abstracts . García, M., Sandholt, I., Ceccato, P., Ridler, M., Mougin, E., Kergoat, L., Morilla s, L., Timouk, F., Fensholt, R. and Domingo, F. (2013). Actual evapotranspiration in drylands derived from in - situ and satellite data: Assessing biophysical constraints. Remote Sensing of Environment , 131 , pp.103 - 118. Glahn, H. R., & Ruth, D. P. (2003). Th e new digital forecast database of the National Weather Service. Bulletin of the American Meteorological Society , 84 (2), 195 - 202. coefficients to estimate evap otranspiration by remote sensing in agricultural and natural ecosystems. Hydrological Processes , 25(26), 4050 - 4062. 145 Gobbo, S., Lo Presti, S., Martello, M., Panunzi, L., Berti, A., & Morari, F. (2019). Integrating SEBAL with in - Field Crop Water Status Measu rement for Precision Irrigation Applications A Case Study. Remote Sensing , 11 (17), 2069. Goldberg, D. E. (1989). Optimization and Machine Learning. Genetic algorithms in Search . Gopalakrishnan, G., Minsker, B. S., & Goldberg, D. E. (2003). Optimal sampling in a noisy genetic algorithm for risk - based remediation design. Journal of Hydroinformatics , 5(1), 11 - 25. Grassini, P., van Bussel, L.G., Van Wart, J., Wolf, J., Claessens, L., Yang, H., Boogaard, H., de Groot, H., van Ittersum, M.K. and Cassman, K.G., 2015. How good is good enough? Data requirements for reliable crop yield simulations and yield - gap analysis. Field Crops Research, 177, pp.49 - 63. Grimm, S. S., Jones, J. W., Boote, K. J., & Hesketh, J. D. (1993). Parameter estimation for predicting flowering date of soybean cultivars. Crop Science , 33(1), 137 - 144. Grosso, C., Manoli, G., Martello, M., Chemin, Y., Pons, D., Teatini, P., Piccoli, I. and Morari, F. (2018). Mapping maize evapotranspiration at field scale using SEBAL: A comparison with th e FAO method and soil - plant model simulations. Remote Sensing , 10 (9), p.1452. Groth, J. V., Ozmon, E. A., & Busch, R. H. (1999). Repeatability and relationship of incidence and severity measures of scab of wheat caused by Fusarium graminearum in inoculated nurseries. Plant disease , 83 (11), 1033 - 1038. Gxasheka, M., Wang, J., Tyasi, T. L., & Gao, J. (2015). Scientific understanding and effects on ear rot diseases in maize production: a review. International Journal of Soil and Crop Sciences , 3 (4), 077 - 084. 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 human interventions and climate change. Proceedings of the National Aca demy of Sciences , 111 (9), pp.3251 - 3256. Hall, P., & Wilson, S. R. (1991). Two guidelines for bootstrap hypothesis testing. Biometrics , 757 - 762. Hammad, H. M., Abbas, F., Ahmad, A., Farhad, W., Anothai, J., & Hoogenboom, G. (2018). Predicting water and nitr ogen requirements for maize under semi - arid conditions using the CSM - CERES - Maize model. European Journal of Agronomy , 100, 56 - 66. Hammer, G.L., Dong, Z., McLean, G., Doherty, A., Messina, C., Schussler, J., Zinselmeier, C., Paszkiewicz, S. and Cooper, M., 2009. Can changes in canopy and/or root system architecture explain historical maize yield trends in the US Corn Belt? Crop Science , 49(1), pp.299 - 312. Han, E., & Ines, A. V. (2017). Downscaling probabilistic seasonal climate forecasts for decision support in agriculture: A comparison of parametric and non - parametric approach. Climate Risk Management , 18 , 51 - 65. 146 Han, E., Baethgen, W.E., Ines, A.V., Mer, F., Souza, J.S., Berterretche, M., Atunez, G. and Barreira, C., 2019. SIMAGRI: An agro - climate decision s upport tool. Computers and Electronics in Agriculture , 161 , pp.241 - 251. Han, E., Ines, A. V., & Baethgen, W. E. (2017). Climate - Agriculture - Modeling and Decision Tool (CAMDT): A software framework for climate risk management in agriculture. Environmental m odelling & software , 95 , 102 - 114. Han, S., Tang, Q., Xu, D., & Wang, S. (2014). Irrigation - induced changes in potential evaporation: more attention is needed. Hydrological Processes , 28 (4), 2717 - 2720. Hanks, J., & Ritchie, J. T. (1991). Modeling plant and soil systems. Soil Science Society of America. Hansen, J. W., Challinor, A., Ines, A., Wheeler, T., & Moron, V. (2006). Translating climate forecasts into agricultural terms: advances and challenges. Climate research , 33 (1), 27 - 41. Hassanien, M. K., & Medany, M. A. (2007). The impact of climate change on production of maize (Zea mays L.). In Proc. of the international conference on climate change and their impacts on costal zones and River Deltas, Alexandria - Egypt (pp. 23 - 25). Hatfie ld, J.L., Boote, K.J., Kimball, B.A., Ziska, L.H., Izaurralde, R.C., Ort, D., Thomson, A.M. and Wolfe, D. (2011). Climate impacts on agriculture: implications for crop production. Agronomy journal , 103 (2), pp.351 - 370. He, D., Wang, E., Wang, J., & Robertso n, M. J. (2017). Data requirement for effective calibration of process - based crop models. Agricultural and Forest Meteorology , 234, 136 - 148. He, J., Dukes, M. D., Jones, J. W., Graham, W. D., & Judge, J. (2009). Applying GLUE for estimating CERES - Maize gen etic and soil parameters for sweet corn production. Transactions of the ASABE, 52(6), 1907 - 1921. He, J., Jones, J. W., Graham, W. D., & Dukes, M. D. (2010). Influence of likelihood function choice for estimating crop model parameters using the generalized likelihood uncertainty estimation method. Agricultural Systems , 103(5), 256 - 264. Holmes, T. R., Hain, C., Crow, W. T., Anderson, M. C., & Kustas, W. P. (2018). Microwave implementation of two - source energy balance approach for estimating evapotranspiration . Hydrology and earth system sciences , 22 (2), 1351 - 1369. Hoogenboom, G., Jones, J. W., Wilkens, P. W., Porter, C. H., Boote, K. J., Hunt, L. A., Singh, U., Lizaso, J.L., White, J.W., Uryasev, O & Royce, F. S. (2013). Decision Support System for Agrotechnol ogy Transfer (DSSAT) Version 4.6. 0.016. Hoogenboom, G., White, J. W., & Messina, C. D. (2004). From genome to crop: integration through simulation modeling. Field Crops Research , 90(1), 145 - 163. Houborg, R. M., & Soegaard, H. (2004). Regional simulation o f ecosystem CO2 and water vapor exchange for agricultural land using NOAA AVHRR and Terra MODIS satellite data. Application to Zealand, Denmark. Remote Sensing of Environment , 93 (1 - 2), 150 - 167. 147 Hu, G., Jia, L., & Menenti, M. (2015). Comparison of MOD16 and LSA - SAF MSG evapotranspiration products over Europe for 2011. Remote Sensing of Environment , 156 , 510 - 526. Huang, X., Huang, G., Yu, C., Ni, S., & Yu, L. (2017). A multiple crop model ensemble for improving broad - scale yield prediction using Bayesian mode l averaging. Field Crops Research , 211 , 114 - 124. Hund, A., Ruta, N., & Liedgens, M. (2009). Rooting depth and water use efficiency of tropical maize inbred lines, differing in drought tolerance. Plant and Soil , 318(1 - 2), 311 - 325. Hunt, L. A., Pararajasingh am, S., Jones, J. W., Hoogenboom, G., Imamura, D. T., & Ogoshi, R. M. (1993). GENCALC: software to facilitate the use of crop models for analyzing field experiments. Agronomy Journal , 85(5), 1090 - 1094. Ibrahim, O. M., Gaafar, A. A., Wali, A. M., Tawfik, M. M., & El - Nahas, M. M. (2016). Estimating cultivar coefficients of a spring wheat using GENCALC and GLUE in DSSAT. Journal of Agronomy , 15(3), 130 - 135. Ines, A. V. M., & Droogers, P. (2002). Inverse modelling in estimating soil hydraulic functions: a Genet ic Algorithm approach. Hydrology and Earth System Sciences , 6(1), 49 - 66. Ines, A. V. M., & Mohanty, B. P. (2008a). Parameter conditioning with a noisy Monte Carlo genetic algorithm for estimating effective soil hydraulic properties from space. Water Resources Research , 44(8). Ines, A. V. M., & Mohanty, B. P. (2008b). Near - surfa ce soil moisture assimilation for quantifying effective soil hydraulic properties using genetic algorithm: 1. Conceptual modeling. Water Resources Research , 44(6). Ines, A. V. M., Baule, W. J., & Jha, P. K. (2018, December). Quantifying the Impacts of Earl y and Late Growing Season Precipitation on Midwestern US Corn Production: A Downscaling and Modeling Approach. In AGU Fall Meeting Abstracts . Ines, A. V., Honda, K., Han, E., Barras, C., & Chinnachodteeranun, R. (2018, December). Downscaling Probabilistic Seasonal Rainfall and Temperature Forecasts for Climate Risk Management in Agriculture. In AGU Fall Meeting Abstracts . IPCC (2014): Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Gr oup II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change [Field, C.B., V.R. Barros, D.J. Dokken, K.J. Mach, M.D. Mastrandrea, T.E. Bilir, M. Chatterjee, K.L. Ebi, Y.O. Estrada, R.C. Genova, B. Girma, E.S. Kissel, A.N. Levy, S. MacCracken, P.R. Mastrandrea, and L.L. White (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 1132 pp. Irmak, S., Djaman, K., & Rudnick, D. R. (2016). Effect of full and limited irrigation amount and frequency on subsu rface drip - irrigated maize evapotranspiration, yield, water use efficiency and yield response factors. Irrigation Science , 34 (4), 271 - 286. 148 Ito, A., & Inatomi, M. (2012). Water - use efficiency of the terrestrial biosphere: a model analysis focusing on intera ctions between the global carbon and water cycles. Journal of Hydrometeorology , 13 (2), 681 - 694. Jackson, R. D., Pinter Jr, P. J., & Reginato, R. J. (1985). Net radiation calculated from remote multispectral and ground station meteorological data. Agricultu ral and forest Meteorology , 35 (1 - 4), 153 - 164. Jagtap, S. S., Lall, U., Jones, J. W., Gijsman, A. J., & Ritchie, J. T. (2004). Dynamic nearest - neighbor method for estimating soil water parameters. Transactions of the ASAE , 47 (5), 1437. Jensen ME. (1968). Wa ter consumption by agricultural plants. In Water deficits and plant growth, vol.2, Chapter 1 , Kozlowski TT (ed). Academic Press: Salt Lake City, Utah, 1 22 pp. Jensen, M. E. (1973). Consumptive use of water and irrigation water requirements. ASCE. New York . Jensen, M. E., Burman, R. D., & Allen, R. G. (1990). Evapotranspiration and irrigation water requirements. ASCE. Jha, P. K., Kumar, S. N., & Ines, A. V. (2018). Responses of soybean to water stress and supplemental irrigation in upper Indo - Gangetic p lain: Field experiment and modeling approach. Field Crops Research , 219, 76 - 86. Jones, C. A., & Kiniry, J. R. (1986). CERES - Maize; A simulation model of maize growth and development . eds (No. 633.153 JON. CIMMYT.). Jones, J. W., He, J., Boote, K. J., Wilke ns, P., Porter, C. H., & Hu, Z. (2011). Estimating DSSAT cropping system cultivar - specific parameters using Bayesian techniques. Methods of Introducing System Models into Agricultural Research , p365 - 394. Jones, J. W., Hoogenboom, G., Porter, C. H., Boote, K. J., Batchelor, W. D., Hunt, L. A., Wilkens, P.W., Singh, U., Gijsman, A. J., & Ritchie, J. T. (2003). The DSSAT cropping system model. European Journal of Agronomy , 18(3), 235 - 265. Kang, M. S. (2002). 15 Genotype Environment Interaction: Progress and Pr ospects. Quantitative Genetics, Genomics, and Plant Breeding , 219. Katerji, N., & Rana, G. (2006). Modelling evapotranspiration of six irrigated crops under Mediterranean climate conditions. Agricultural and Forest Meteorology , 138 (1 - 4), 142 - 155. Katz, P. (2018) MSU Extension. Accessed at : https://www.canr.msu.edu/news/are - there - molds - and - mycotoxins - in - your - 2018 - corn - crop Kersebaum, K.C., Boote, K.J., Jorgen son, J.S., Nendel, C., Bindi, M., Frühauf, C., Gaiser, T., Hoogenboom, G., Kollas, C., Olesen, J.E. and Rötter, R.P., 2015. Analysis and classification of data sets for calibration and validation of agro - ecosystem models. Environmental Modelling & Software , 72, pp.402 - 417. 149 Keune, J., Sulis, M., Kollet, S., Siebert, S., & Wada, Y. (2018). Human water use impacts on the strength of the continental sink for atmospheric water. Geophysical Research Letters , 45 (9), 4068 - 4076. Kimball, B.A., Boote, K.J., Hatfield, J.L., Ahuja, L.R., Stockle, C., Archontoulis, S., Baron, C., Basso, B., Bertuzzi, P., Constantin, J. and Deryng, D. (2019). Simulation of maize evapotranspiration: An inter - comparison among 29 maize models. Agricultural and Forest Meteorology , 271 , pp.264 - 284. Kiniry, J. R., & Bonhomme, R. (1991). Predicting maize phenology. Predicting Crop Phenology , 11, 5 - 131. Kistner, E., Kellner, O., Andresen, J., Todey, D., & Morton, L. W. (2018). Vulnerability of specialty crops to short - term climatic variability and adaptation strategies in the Midwestern USA. Climatic change , 146 (1 - 2), 145 - 158. Kleinknecht, K., Möhring, J., Laidig, F., Meyer, U., & Piepho, H. P. (2016). A simulation - based approach for evaluating the efficiency of multienvironment trial designs. Crop Science , 56(5), 2237 - 2250. Knauer, J., Zaehle, S., Medlyn, B.E., Reichstein, M., Williams, C.A., Migliavacca, M., De Kauwe, M.G., Werner, C., Keitel, C., Kolari, P. and Limousin, J.M. (2018). Towards data. Global Change Biology , 24 (2), pp.694 - 710. Kondo, M., Murty, M. V., & Aragones, D. V. (2000). Characteristics of root growth and water uptake from soil in upland rice and maize under water stress. Soil Science and Plant Nutrition , 46(3), 721 - 732. Koppen, W. (1900). Attempt to classify the climates, preferabl y according to their relationship to the plant world. Geographical Journal , 6 (11th H), 593 - 611. Kucharik, C. J., & Ramankutty, N. (2005). Trends and variability in US maize yields over the twentieth century. Earth Interactions , 9(1), 1 - 29. Kumudini, S., A ndrade, F.H., Boote, K.J., Brown, G.A., Dzotsi, K.A., Edmeades, G.O., Gocken, T., Goodwin, M., Halter, A.L., Hammer, G.L. and Hatfield, J.L. (2014). Predicting maize phenology: intercomparison of functions for developmental response to temperature. Agronom y Journal , 106 (6), pp.2087 - 2097. Lauer, J. (1998). The Wisconsin comparative relative maturity (CRM) system for corn. Lee, E. A., & Tollenaar, M. (2007). Physiological basis of successful breeding strategies for maize grain yield. Crop Science , 47(Suppleme nt_3), S - 202. Lee, E. A., Deen, W., Hooyer, M. E., Chambers, A., Parkin, G., Gordon, R., & Singh, A. K. (2016). Involvement of year - to - year variation in thermal time, solar radiation and soil available moisture in genotype - by - environment effects in maize. Crop Science , 56(5), 2180 - 2192. 150 Li, S., Wang, G., Sun, S., Chen, H., Bai, P., Zhou, S., Huang, Y., Wang, J. and Deng, P., 2018. Assessment of Multi - Source Evapotranspiration Products over China Using Eddy Covariance Observations. Remote Sensing , 10 (11), p. 1692. Li, Z. T., Yang, J. Y., Drury, C. F., & Hoogenboom, G. (2015). Evaluation of the DSSAT - CSM for simulating yield and soil organic C and N of a long - term maize and wheat rotation experiment in the Loess Plateau of Northwestern China. Agricultural Syste ms , 135, 90 - 104. Liu, H.L., Yang, J.Y., Drury, C.A., Reynolds, W.D., Tan, C.S., Bai, Y.L., He, P., Jin, J. and Hoogenboom, G. (2011). Using the DSSAT - CERES - Maize model to simulate crop yield and nitrogen cycling in fields under long - term continuous maize p roduction. Nutrient cycling in agroecosystems , 89 (3), pp.313 - 328. Liu, Z., Zhu, K., Dong, S., Liu, P., Zhao, B., & Zhang, J. (2017). Effects of integrated agronomic practices management on root growth and development of summer maize. European Journal of Agronomy , 84, 140 - 151. Lizaso, J. I., Boote, K. J., Jones, J. W., Porter, C. H., Echarte, L., Westgate, M. E., & Sonohat, G. (2011). CSM - IXIM: A new maize simulation model for DSSAT Version 4.5. Agronomy Journal , 103(3), 766 - 779. Lizaso, J. I., Fonseca, A. E., & Westgate, M. E. (2007). Simulating source - limited and sink - limited kernel set with CERES - Maize. Crop science , 47(5), 2078 - 2088. Lizaso, J.I., Ruiz - Ramos, M., Rodríguez, L., Gabaldon - Leal, C., Oliveira, J.A., Lorite, I.J., Sánchez, D., Gar cía, E. and Rodríguez, A. (2018). Impact of high temperatures in maize: Phenology and yield components. Field crops research , 216 , pp.129 - 140. Lobell, D. B., & Asner, G. P. (2003). Climate and management contributions to recent trends in US agricultural yi elds. Science , 299(5609), 1032 - 1032. Lobell, D. B., Azzari, G., Burke, M., Gourlay, S., Jin, Z., Kilic, T., & Murray, S. (2018). Eyes in the sky, boots on the ground: assessing satellite - and ground - based approaches to crop yield measurement and analysis in Uganda . The World Bank. Lobell, D. B., Burke, M. B., Tebaldi, C., Mastrandrea, M. D., Falcon, W. P., & Naylor, R. L. (2008). Prioritizing climate change adaptation needs for food security in 2030. Science , 319(5863), 607 - 610. Lobell, D. B., Cassman, K. G. , & Field, C. B. (2009). Crop yield gaps: their importance, magnitudes, and causes. Annual Review of Environment and Resources , 34, 179 - 204. Lobell, D. B., Roberts, M. J., Schlenker, W., Braun, N., Little, B. B., Rejesus, R. M., & Hammer, G. L. (2014). Gre ater sensitivity to drought accompanies maize yield increase in the US Midwest. Science , 344 (6183), 516 - 519. Löffler, C.M., Wei, J., Fast, T., Gogerty, J., Langton, S., Bergman, M., Merrill, B. and Cooper, M. (2005). Classification of maize environments us ing crop simulation and geographic information systems. Crop Science , 45 (5), pp.1708 - 1716. 151 López - Cedrón, F. X., Boote, K. J., Piñeiro, J., & Sau, F. (2008). Improving the CERES - Maize model ability to simulate water deficit impact on maize production and yi eld components. Agronomy Journal , 100(2), 296 - 307. López - Cruz, I. L., Ruiz - García, A., Fitz - Rodríguez, E., Salazar - Moreno, R., & Rojano - Aguilar, A. (2016, September). A comparison of Bayesian and classical methods for parameter estimation in greenhouse cro p models. In V International Symposium on Models for Plant Growth, Environment Control and Farming Management in Protected Cultivation 1182 (pp. 241 - 248). Los, S.O., Pollack, N.H., Parris, M.T., Collatz, G.J., Tucker, C.J., Sellers, P.J., Malmström, C.M., DeFries, R.S., Bounoua, L. and Dazlich, D.A (2000). A global 9 - yr biophysical land surface dataset from NOAA AVHRR data. Journal of Hydrometeorology , 1 (2), pp.183 - 199. Lu, Y., & Kueppers, L. (2015). Increased heat waves with loss of irrigation in the Unite d States. Environmental Research Letters , 10 (6), 064010. Ma, L., Hoogenboom, G., Ahuja, L. R., Ascough Ii, J. C., & Saseendran, S. A. (2006). Evaluation of the RZWQM - CERES - Maize hybrid model for maize production. Agricultural Systems , 87(3), 274 - 295. Mabuza, L. M., van Rensburg, B. J., Flett, B. C., & Rose, L. J. (2018). Accumulation of toxigenic Fusarium species and Stenocarpella maydis in maize grain grown under different cropping systems. European journal of plant pathology , 152 (2), 297 - 308 . MacCarthy, D. S., Adiku, S. G., Freduah, B. S., & Gbefo, F. (2017). Using CERES - Maize and ENSO as decision support tools to evaluate climate - sensitive farm management practices for maize production in the northern regions of Ghana. Frontiers in Plant Sci ence , 8, 31. Maiorano, A., Reyneri, A., Sacco, D., Magni, A., & Ramponi, C. (2009). A dynamic risk assessment model (FUMAgrain) of fumonisin synthesis by Fusarium verticillioides in maize grain in Italy. Crop Protection , 28 (3), 243 - 256. Makowski, D., Hilli er, J., Wallach, D., Andrieu, B., & Jeuffroy, M. H. (2006). Parameter estimation for crop models. Working with dynamic crop models. Elsevier, Amsterdam , 101 - 149. Mansfield, M. A., De Wolf, E. D., & Kuldau, G. A. (2005). Relationships between weather condit ions, agronomic practices, and fermentation characteristics with deoxynivalenol content in fresh and ensiled maize. Plant disease , 89 (11), 1151 - 1157. Marin, F. R., Angelocci, L. R., Nassif, D. S., Costa, L. G., Vianna, M. S., & Carvalho, K. S. (2016). Crop coefficient changes with reference evapotranspiration for highly canopy - atmosphere coupled crops. Agricultural Water Management , 163 , 139 - 145. Martin E.C. , Shayya W.H. , Bralts V.F. , Loudon T.L. , Johnson J.A. (1990) Irrigation scheduling. A statewide program in Michigan . ASAE Publication, (4 - 90), pp . 700 - 706. Martínez - Cob, A. (2008). Use of thermal units to estimate corn crop coefficients under semiarid climatic conditions. Irrigation Science , 26 (4), 335 - 345. 152 Mavromatis, T., Boote, K. J., Jones, J. W., Wilkerson, G. G., & Hoogenboom, G. (2002). Repea tability of model genetic coefficients derived from soybean performance trials across different states. Crop Science , 42(1), 76 - 89. Meinke, H., Donald, L., DeVoil, P., Power, B., Baethgen, W., Howden, M., Allan, R. and Bates, B. (2004). How predictable is the climate and how can we use it in managing cropping (CD - ROM, and www.cropscience.org.au . Menne, M. J., Durre, I., Vose, R. S., Gleason, B. E., & Houston, T. G. (2012). An overview of the global historical climatology network - daily database. Journal of Atmospheric and Oceanic Technology , 29 (7), 897 - 910. Mertens, J., Madsen, H., Feyen, L., Jacques, D., & Feyen, J. (2004). Inc luding prior information in the estimation of effective soil parameters in unsaturated zone modelling. Journal of Hydrology , 294(4), 251 - 269. Miao, R., Khanna, M., & Huang, H. (2015). Responsiveness of crop yield and acreage to prices and climate. American Journal of Agricultural Economics , 98 (1), 191 - 211. Michael, M. G., & Bastiaanssen, W. G. (2000). A new simple method to determine crop coefficients for water allocation planning from satellites: results from Kenya. Irrigation and Drainage Systems , 14 (3), 237 - 256. Migliaccio, K. W., & Barclay Shoemaker, W. (2014). Estimation of urban subtropical bahiagrass (Paspalum notatum) evapotranspiration using crop coefficients and the eddy covariance method. Hydrological processes , 28 (15), 4487 - 4495. Miller, B. L., & Goldberg, D. E. (1996, November). Optimal sampling for genetic algorithms. In conference (Vol. 6, pp. 291 - 297). Miller, J. D. (1994). Epidemiology of Fusarium ear diseases of cer eals. Mycotoxins in grain: compounds other than aflatoxin , 19 - 36. Miller, S. S., Reid, L. M., & Harris, L. J. (2007). Colonization of maize silks by Fusarium graminearum, the causative organism of gibberella ear rot. Botany , 85 (4), 369 - 376. Mokhtari, A., N oory, H., Vazifedoust, M., Palouj, M., Bakhtiari, A., Barikani, E., Afrooz, R.A.Z., Fereydooni, F., Naeni, A.S., Pourshakouri, F. and Badiehneshin, A. (2019). Evaluation of single crop coefficient curves derived from Landsat satellite images for major crop s in Iran. Agricultural Water Management , 218 , pp.234 - 249. Molden, D. (2007). Comprehensive assessment of water management in agriculture. Water for food, water for life: a comprehensive assessment of water management in agriculture. London: Earthscan and Colombo: International Water Management Institute . Monteith, J. L. (1965). Evaporation and the environment. Symposium of the Society of Exploratory Biology, 153 Mourice, S. K., Rweyemamu, C. L., Tumbo, S. D., & Amuri, N. (2014). Maize cultivar spe cific parameters for decision support system for agrotechnology transfer (DSSAT) application in Tanzania. MSU Extension, (2014). Accessed online at November 12, 2019 at https://www.canr.msu.edu/uploads/235/67987/resources/SWMichiganValueOfIrrigation9 - 23 - 14.pdf Mu, Q., Heinsch, F. A., Zhao, M., and Running, S. W. (2007). Development of a global evapotranspiration algorithm based on MODIS and global met eorology data. Remote Sensing of Environment , 111 (4), 519 - 536. Mu, Q., Zhao, M., and Running, S. W. (2011). Improvements to a MODIS global terrestrial evapotranspiration algorithm. Remote Sensing of Environment , 115 (8), 1781 - 1800. Mubeen, M., Ahmad, A., Wajid, A., Khaliq, T., Hammad, H.M., Sultana, S.R., Ahmad, S., Fahad, S. and Nasim, W. (2016). Application of CSM - CERES - Maize model in optimizing irrigated conditions. Outlook on Agriculture , 45 (3), pp.173 - 184. Mueller, D. S., Sisson, A., and Wise, K. A. ( 2016). Corn disease loss estimates from the United States and Ontario, Canada, 2015. CPN - 2007 - 14 - W. Crop Protection Network . Mueller, N. D., Gerber, J. S., Johnston, M., Ray, D. K., Ramankutty, N., & Foley, J. A. (2012). Closing yield gaps through nutrient and water management. Nature , 490 (7419), 254. Müller, C., Elliott, J., Kelly, D., Arneth, A., Balkovic, J., Ciais, P., Deryng, D., Folberth, C., Hoek, S., Izaurralde, R.C. and Jones, C.D., 2019. The Global Gridded Crop Model Intercomparison phase 1 simula tion dataset. Scientific data , 6 (1), p.50. Munkvold, G. P. (2003). Epidemiology of Fusarium diseases and their mycotoxins in maize ears. European Journal of Plant Pathology , 109 (7), 705 - 713. Munkvold, G. P. (2014). Crop management practices to minimize the risk of mycotoxins contamination in temperate - zone maize. Mycotoxin Reduction in Grain Chains , 59 - 75. Ogutu, G. E., Franssen, W. H., Supit, I., Omondi, P., & Hutjes, R. W. (2018). Probabilistic maize yield prediction over East Africa using dyn amic ensemble seasonal climate forecasts. Agricultural and forest meteorology , 250, 243 - 261. Olioso, A., Allies, A., Boulet, G., Delogu, E., Demarty, J., Elvira, B.G., Mira, M., Marloie, O., Chauvelon, P., Boutron, O. and Buis, S., 2018, July. Monitoring E vapotranspiration with Remote Sensing Data and Ground Data Using Ensemble Model Averaging. In IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium (pp. 7656 - 7659). IEEE. Owour, M. J., Midega, C. A., Khan, Z. R., & Obonyo, M. (2015). Impact of climate - smart push - pull technology on maize ear rots and mycotoxins: Incidence and severity. Pabico, J. P. (2007). 'Breeding' genetic coefficients in plant growth simulation models. Philippine Journal of Crop Science (Philippines) . 154 Pabico, J. P., Hoogenboom, G., & McClendon, R. W. (1999). Determination of cultivar coefficients of crop models using a genetic algorithm: a conceptual framework. Transactions of the ASAE , 42(1), 223. Paderewski, J., Gauche, H. G., Mary, W., & Gacke, E. (2016). AMMI ana lysis of four - way genotype× location× management× year data from a wheat trial in Poland. Crop Science , 56(5), 2157 - 2164. Parker, N. S., Anderson, N. R., Richmond, D. S., Long, E. Y., Wise, K. A., & Krupke, C. H. (2017). Larval western bean cutworm feeding damage encourages the development of Gibberella ear rot on field corn. Pest management science , 73 (3), 546 - 553. Piccinni, G., Ko, J., Marek, T., & Howell, T. (2009). Determination of growth - stage - specific crop coefficients (KC) of maize and sorghum. Agric ultural water management , 96 (12), 1698 - 1704. Pickering, N. B., Hansen, J. W., Jones, J. W., Wells, C. M., Chan, V. K., & Godwin, D. C. (1994). Weatherman: a utility for managing and generating daily weather data. Agronomy Journal , 86(2), 332 - 337. Pitt, J. I., & Miller, J. D. (2016). A concise history of mycotoxin research. Journal of agricultural and food chemistry , 65 (33), 7021 - 7033. Porter, J.R., Xie, L., Challinor, A.J., Cochrane, K., Howden, S.M., Iqbal, M.M., Travasso, M.I., Barros, V.R., Field, C.B., Dokken, D.J. and Mastrandrea, M.D. (2017). Food security and food production systems. Priestley, C. H. B., & Taylor, R. J. (1972). On the assessment of surface heat flux and evaporation using large - scale parameters. Monthly weather review , 100 (2), 81 - 92. P rokopy, L. S., Carlton, J. S., Haigh, T., Lemos, M. C., Mase, A. S., & Widhalm, M. (2017). Useful to usable: Developing usable climate science for agriculture. Climate Risk Management , 15 , 1 - 7. Ramoelo, A., Majozi, N., Mathieu, R., Jovanovic, N., Nickless, A., & Dzikiti, S. (2014). Validation of global evapotranspiration product (MOD16) using flux tower data in the African savanna, South Africa. Remote Sensing , 6 (8), 7406 - 7423. Reichstein, M., Falge, E., Baldocchi, D., Papale, D., Aubinet, M., Berbigier, P. , Bernhofer, C., Buchmann, N., Gilmanov, T., Granier, A. and Grünwald, T (2005). On the separation of net ecosystem exchange into assimilation and ecosystem respiration: review and improved algorithm. Global Change Biology , 11 (9), pp.1424 - 1439. Reid, L. M. , Bolton, A. T., Hamilton, R. I., Woldemariam, T., & Mather, D. E. (1992). Effect of silk age on resistance of maize to Fusarium graminearum. Canadian Journal of Plant Pathology , 14 (4), 293 - 298. Ritchie, J. T., & Alagarswamy, G. (2003). Model concepts to e xpress genetic differences in maize yield components. Agronomy Journal , 95(1), 4 - 9. Ritchie, J. T., & Nesmith, D. S. (1991). Temperature and crop development. Modeling plant and soil systems , (modelingplantan), 5 - 29. 155 Ritchie, J. T., & Wei, J. (2000). Model s of kernel number in maize. Physiology and modeling kernel set in maize , (physiologyandmo), 75 - 88. Ritchie, J. T., Singh, U., Godwin, D. C., & Bowen, W. T. (1998). Cereal growth, development and yield. In Understanding options for agricultural production (pp. 79 - 98). Springer, Dordrecht. Rodríguez, A., Ruiz - Ramos, M., Palosuo, T., Carter, T.R., Fronzek, S., Lorite, I.J., Ferrise, R., Pirttioja, N., Bindi, M., Baranowski, P. and Buis, S. (2019). Implications of crop model ensemble size and composition for e stimates of adaptation effects and agreement of recommendations. Agricultural and Forest Meteorology , 264, pp.351 - 362. Román - Paoli, E., Welch, S. M., & Vanderlip, R. L. (2000). Comparing genetic coefficient estimation methods using the CERES - Maize model. Agricultural Systems , 65(1), 29 - 41. (2018). Closing the yield gap while ensuring water sustainability. Environmental Research Letters , 13 (10), 104002. Rosburg, A., & Menapace, L. (2018). Factors influencing corn fungicide treatment decisions. Journal of Agricultural and Resource Economics , 43 (1835 - 2018 - 2976), 151 - 171. Rosenzweig, C., Jones, J.W., Hatfield, J.L., Ruane, A.C., Boote, K.J., Thorburn, P., Antle, J.M., Nelson, G.C., Porter, C., Janssen, S. and Asseng, S., 2013. The agricultural model intercomparison and improvement pro ject (AgMIP): protocols and pilot studies. Agricultural and Forest Meteorology , 170 , pp.166 - 182. Rosenzweig, C., Ruane, A.C., Antle, J., Elliott, J., Ashfaq, M., Chatta, A.A., Ewert, F., Folberth, C., Hathie, I., Havlik, P. and Hoogenboom, G. (2018). Coord inating AgMIP data and models across global and regional scales for 1.5 C and 2.0 C assessments. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , 376 (2119), p.20160455. Rossow, W. B., Walker, A. W., Beusch el, D. E., & Roiter, M. D. (1996). International Satellite Cloud Climatology Project (ISCCP): Documentation of new cloud data sets. World Meteorological Organization , 115. Rötter, R. P., Carter, T. R., Olesen, J. E., & Porter, J. R. (2011). Crop climate mo dels need an overhaul. Nature climate change , 1 (4), 175. Roux, S., Brun, F., & Wallach, D. (2014). Combining input uncertainty and residual error in crop model predictions: A case study on vineyards. European Journal of Agronomy , 52, 191 - 197. Running, S. W ., Mu, Q., Zhao, M., & Moreno, A. (2017). Modis Global Terrestrial Evapotranspiration (ET) Product (NASA MOD16A2/A3) NASA Earth Observing System Modis Land Algorithm. NASA: Washington, DC, USA . Rutan, J., & Steinke, K. (2017). Determining corn nitrogen rat es using multiple prediction models. Journal of Crop Improvement , 31(6), 780 - 800.6(5), 2090 - 2105. 156 Ruth, D. P., Glahn, B., Dagostaro, V., & Gilbert, K. (2009). The performance of MOS in the digital age. Weather and Forecasting , 24 (2), 504 - 519. Saab, M. T. A ., Todorovic, M., & Albrizio, R. (2015). Comparing AquaCrop and CropSyst models in simulating barley growth and yield under different water and nitrogen regimes. Does calibration year influence the performance of crop growth models?. Agricultural Water Man agement , 147, 21 - 33. Sacks, W. J., & Kucharik, C. J. (2011). Crop management and phenology trends in the US Corn Belt: Impacts on yields, evapotranspiration and energy balance. Agricultural and Forest Meteorology , 151(7), 882 - 894. Saha, S., Moorthi, S., Wu , X., Wang, J., Nadiga, S., Tripp, P., Behringer, D., Hou, Y.T., Chuang, H.Y., Iredell, M. and Ek, M., 2014. The NCEP climate forecast system version 2. Journal of Climate , 27 (6), pp.2185 - 2208. Samani, Z., Bawazir, A. S., Bleiweiss, M., Skaggs, R., Longwor th, J., Tran, V. D., & Pinon, A. (2009). Using remote sensing to evaluate the spatial variability of evapotranspiration and crop coefficient in the lower Rio Grande Valley, New Mexico. Schaafsma, A. W., & Hooker, D. C. (2007). Climatic models to predict oc currence of Fusarium toxins in wheat and maize. International journal of food microbiology , 119 (1 - 2), 116 - 125. Schmale III, D. G., & Bergstrom, G. C. (2004). Spore deposition of the ear rot pathogen, Gibberella zeae, inside corn canopies. Canadian journal of plant pathology , 26 (4), 591 - 595. Schubert, S. D., Rood, R. B., and Pfaendtner, J. (1993). An assimilated dataset for earth science applications. Bulletin of the American meteorological Society , 74 (12), 2331 - 2342. Seidel, S. J., Palosuo, T., Thorburn, P. , & Wallach, D. (2018). Towards improved calibration of crop models Where are we now and where should we go? European Journal of Agronomy , 94, 25 - 35. Semenov, M. A., & Doblas - Reyes, F. J. (2007). Utility of dynamical seasonal forecasts in predicting crop y ield. Climate Research, 34(1), 71 - 81. Senay, G.B., Schauer, M., Velpuri, N.M., Singh, R.K., Kagone, S., Friedrichs, M., Litvak, M.E. and Douglas - Mankin, K.R. (2019). Long - term (1986 2015) crop water use characterization over the Upper Rio Grande Basin of United States and Mexico using Landsat - based evapotranspiration. Remote Sensing , 11 (13), p.1587. Senkondo, W., Munishi, S. E., Tumbo, M., Nobert, J., & Lyon, S. W. (2019). Comparing Remotely - Sensed Surface Energy Balance Evapotranspiration Estimates in Het erogeneous and Data - Valley. Remote Sensing , 11 (11), 1289. Shafiee - Jood, M., Cai, X., Chen, L., Liang, X.Z., Kumar, P., 2014. Assessing the value of seasonal climate forecast information through an end - t o - end forecasting framework: application to US 2012 drought in central Illinois. Water Resources Research . 50 (8), 6592 6609. 157 Shah, D. A., Paul, P. A., De Wolf, E. D., & Madden, L. V. (2019). Predicting plant disease epidemics from functionally represent ed weather series. Philosophical Transactions of the Royal Society B , 374 (1775), 20180273. Shaibu, A. S., Yusuf, S. I., & Adnan, A. A. (2017). Phenotyping and prediction of Maize (Zea mays L.) yield using physiological traits. Journal of Dryland Agricultur e , 3(1), 28 - 35. Sharp, R. E., & Davies, W. J. (1985). Root growth and water uptake by maize plants in drying soil. Journal of Experimental Botany , 36(9), 1441 - 1456. Shelia, V., Hansen, J., Sharda, V., Porter, C., Aggarwal, P., Wilkerson, C. J., & Hoogenboo m, G. (2019). A multi - scale and multi - model gridded framework for forecasting crop production, risk analysis, and climate change impact studies. Environmental Modelling & Software , 115 , 144 - 154. Shin, Y., Mohanty, B. P., & Ines, A. V. (2013). Estimating effective soil hydraulic properties using spatially distributed soil moisture and evapotranspiration. Vadose Zone Journal , 12(3). Siebert, S., & Döll, P. (2010). Quantifying blue and green virtual water contents in global crop production as well as potential production losses without irrigation. Journal of Hydrology , 384 (3 - 4), 198 - 217. Siebert, S., Ewert, F., Rezaei, E. E., Kage, H., & Graß, R. (2014). Impact of heat stress on crop yield on the impo rtance of considering canopy temperature. Environmental Research Letters , 9 (4), 044012. Siebert, S., Kummu, M., Porkka, M., Döll, P., Ramankutty, N., & Scanlon, B. R. (2015). A global data set of the extent of irrigated land from 1900 to 2005. Hydrology an d Earth System Sciences , 19 (3), 1521 - 1545. Sikorska, A. E., & Seibert, J. (2018). Value of different precipitation data for flood prediction in an alpine catchment: A Bayesian approach. Journal of Hydrology , 556, 961 - 971. Singh, M.P., W.D. Widdicombe, and L.A. Williams. (2018). 2018 Michigan Corn Hybrids compared. Michigan State University Extension bulletin E - 431. Available at https://varietytrials.msu.edu/corn. (Accessed on September 4, 2019). Singh, R. K., & Irmak, A. (2011). Treatment of anchor pixels in the METRIC model for improved estimation of sensible and latent heat fluxes. Hydrological sciences journal , 56 (5), 895 - 906. Singh, R., & Senay, G. (2016). Comparison of four different energy balance models for estimating evapotranspiration in the Midwes tern United States. Water , 8 (1), 9. design using a noisy genetic algorithm. Water Resources Research , 36(10), 3043 - 3052. Sobenko, L. R., Souza, T. T., Goncalves, A. O., Bianchini, V. J., SILVA, E. H., Souza, L. T., & Marin, F. R. (2019). Irrigation requirements are lower than those usually prescribed for a maize crop in southern Brazil. Experimental Agriculture , 55 (4), 662 - 671. 158 Sridhar, V., & Anderson, K. A. (2017). Human - induced modifications to land surface fluxes and their implications on water management under past and future climate change conditions. Agricultural and Forest Meteorology , 234, 66 - 79. Stewart, D. W., Dwyer, L. M., & Carrigan, L. L. (1998). Phenolog ical temperature response of maize. Agronomy Journal , 90(1), 73 - 79. Su, Z. (2002). The Surface Energy Balance System (SEBS) for estimation of turbulent heat fluxes. Hydrology and earth system sciences , 6 (1), 85 - 100. Sun, M., Zhang, X., Huo, Z., Feng, S., H uang, G., & Mao, X. (2016). Uncertainty and sensitivity assessments of an agricultural hydrological model (RZWQM2) using the GLUE method. Journal of Hydrology , 534, 19 - 30. Suriharn, B., Patanothai, A., Pannangpetch, K., Jogloy, S., & Hoogenboom, G. (2007). Determination of cultivar coefficients of peanut lines for breeding applications of the CSM - CROPGRO - Peanut model. Crop Science , 47(2), 607 - 619. Sutton, J. C. (1982). Epidemiology of wheat head blight and maize ear rot caused by Fusarium graminearum. Canad ian journal of plant pathology , 4 (2), 195 - 209. Tang, R., Li, Z. L., Chen, K. S., Jia, Y., Li, C., & Sun, X. (2013). Spatial - scale effect on the SEBAL model for evapotranspiration estimation using remote sensing data. Agricultural and Forest Meteorology , 17 4 , 28 - 42. Tasumi, M., Allen, R. G., Trezza, R., & Wright, J. L. (2005). Satellite - based energy balance to assess within - population variance of crop coefficient curves. Journal of Irrigation and Drainage Engineering , 131 (1), 94 - 109. Thiery, W., Davin, E. L., Lawrence, D. M., Hirsch, A. L., Hauser, M., & Seneviratne, S. I. Journal of Geophysical Research: Atmospheres , 122 (3), 1403 - 1422. Tolk, J. A., Evett, S. R., Xu, W., & Schwartz, R. C. (2016). Const raints on water use efficiency of drought tolerant maize grown in a semi - arid environment. Field Crops Research, 186, 66 - 77. Tomlinson, S. A. (1996). Comparison of Bowen - ratio, eddy - correlation, and weighing - lysimeter evapotranspiration for two sparse - cano py sites in eastern Washington. Water - Resour. Invest. Rep , 96 - 4081. Tsuji, G. Y. (1998). Network management and information dissemination for agro - technology transfer. In Understanding Options for Agricultural Production (pp. 367 - 381). Springer Netherlands . Twine, T. E., & Kucharik, C. J. (2009). Climate impacts on net primary productivity trends in natural and managed ecosystems of the central and eastern United States. Agricultural and Forest Meteorology , 149(12), 2143 - 2161. USDA - National Agricultural Sta tistics Service (NASS) (2017). Crops U.S. state and county databases. Washington DC. Available at URL: http://www.nass.usda.gov/index.asp . 159 Vakhtang, S., Hansen, J., Sharda, V., Porter, C., Agga rwal, P., Wilkerson, C. J., & Hoogenboom, G. (2019). A multi - scale and multi - model gridded framework for forecasting crop production, risk analysis, and climate change impact studies.forecasts. Agricultural and Forest Meteorology , 250 , 243 - 261. Vamerali, T ., Saccomani, M., Bona, S., Mosca, G., Guarise, M., & Ganis, A. (2003). A comparison of root characteristics in relation to nutrient and water stress in two maize hybrids. In Roots: The Dynamic Interface Between Plants and the Earth (pp. 157 - 167). Springer , Dordrecht. van Dam, J. C. (2000). Field - scale water flow and solute transport: SWAP model concepts, parameter estimation and case studies=[Waterstroming en transport van opgeloste stoffen op veldschaal]. [sn]. Van Eeuwijk, F. A., Bustos - Korts, D. V., & Malosetti, M. (2016). What should students in plant breeding know about the statistical aspects of genotype× environment interactions? Crop Science , 56(5), 2119 - 2140. Van Ittersum, M. K., Cassman, K. G., Grassini, P., Wolf, J., Tittonell, P., & Hochman, Z. (2013). Yield gap analysis with local to global relevance a review. Field Crops Research , 143 , 4 - 17. van Ittersum, M. K., Leffelaar, P. A., Van Keulen, H., Kropff, M. J., Bastiaans, L., & Goudriaan, J. (2003). On approaches and applications of the Wageningen crop models. European Journal of Agronomy , 18(3 - 4), 201 - 234. Varella, H., Buis, S., Launay, M., & Guérif, M. (2012). Global sensitivity analysis for choosing the main soil parameters of a crop model to be dete rmined. Agricultural Sciences , 3(07), 949. Velázquez, J. A., Anctil, F., Ramos, M. H., & Perrin, C. (2011). Can a multi - model approach improve hydrological ensemble forecasting? A study on 29 French catchments using 16 hydrological model structures. Advanc es in Geosciences , 29 , 33 - 42. Vigier, B., Reid, L. M., Dwyer, L. M., Stewart, D. W., Sinha, R. C., Arnason, J. T., & Butler, G. (2001). Maize resistance to gibberella ear rot: symptoms, deoxynivalenol, and yield1. Canadian Journal of Plant Pathology , 23 (1) , 99 - 105. Vilayvong, S., Banterng, P., Patanothai, A., & Pannangpetch, K. (2015). CSM - CERES - Rice model to determine management strategies for lowland rice production. Scientia Agricola , 72(3), 229 - 236. Vrugt, J. A., & Robinson, B. A. (2007). Treatment of u ncertainty using ensemble methods: Comparison of sequential data assimilation and Bayesian model averaging. Water Resources Research , 43 (1). Wada, Y., & Bierkens, M. F. (2014). Sustainability of global water use: past reconstruction and future projections. Environmental Research Letters , 9 (10), 104003. Wagena, M. B., Bhatt, G., Buell, E., Sommerlot, A. R., Fuka, D. R., & Easton, Z. M. (2019). Quantifying model uncertainty using Bayesian multi - model ensembles. Environmental Modelling & Software , 117 , 89 - 99. 160 Waha, K., Huth, N., Carberry, P., & Wang, E. (2015). How model and input uncertainty impact maize yield simulations in West Africa. Environmental Research Letters , 10(2), 024017. Wallach, D., Goffinet, B., Bergez, J. E., Debaeke, P., Leenhardt, D., & Auber tot, J. N. (2001). Parameter estimation for crop models. Agronomy Journal , 93(4), 757 - 766. Wallach, D., Keussayan, N., Brun, F., Lacroix, B., & Bergez, J. E. (2012). Assessing the uncertainty when using a model to compare irrigation strategies. Agronomy Jo urnal , 104(5), 1274 - 1283. runoff models. Water Resources Research , 27(9), 2467 - 2471 Warfield, C. Y., & Davis, R. M. (1996). Importance of the husk covering on the susceptibility of corn hybrids to Fusarium ear rot. Plant disease (USA) . White, J. W., & Hoogenboom, G. (1996). Simulating effects of genes for physiological traits in a process - oriented crop model. Agronomy Journal , 88(3), 416 - 422. White, J. W., & Hoo genboom, G. (2010). Crop response to climate: ecophysiological models. In Climate change and food security (pp. 59 - 83). Springer, Dordrecht. Williams, A., Hunter, M.C., Kammerer, M., Kane, D.A., Jordan, N.R., Mortensen, D.A., Smith, R.G., Snapp, S. and Davis, A.S. (2016). Soil water holding capacity mitigates downside risk and volatility in US rainfed maize: time to invest in soil organic matter?. PloS one , 11 (8), p.e0160974. Wise, K., Mueller, D., Sisson, A., Smith, D., Bradley, C., & Robertson, A. (201 6). Thank you to the sponsors who helped make this book possible. Wu, J., Zheng, C., Chien, C. C., & Zheng, L. (2006). A comparative study of Monte Carlo simple genetic algorithm and noisy genetic algorithm for cost - effective sampling network design under uncertainty. Advances in Water Resources , 29(6), 899 - 911 Wutzler, T., Lucas - Reichstein, M. (2018). Basic and extensible post - processing of eddy covariance flux data with REddyPr oc. Biogeosciences , 15(16), pp.5015 - 5030. Xystrakis, F., & Matzarakis, A. (2010). Evaluation of 13 empirical reference potential evapotranspiration equations on the island of Crete in southern Greece. Journal of Irrigation and Drainage Engineering , 137 (4), 211 - 222. Yan, W. (2016). Analysis and handling of G× E in a practical breeding program. Crop Science , 56(5), 2106 - 2118. Yan, Y., Liu, J., Zhang, J., Li, X., & Zhao, Y. (2017). Quantifying soil hydraulic properties and their uncertainties by modified GLUE method. International Agrophysics , 31(3), 433 - 445. Yang, H., Grassini, P., Cassman, K. G., Aiken, R. M., & Coyne, P. I. (2017). Improvements to the Hybrid - Maize model for simulating maize yields in harsh rainfed environments. Field Crops Research , 204, 180 - 190. 161 Yang, Y., Anderson, M. C., Gao, F., Wardlow, B., Hain, C. R., Otkin, J. A., Alfieri, J., Yang, Y., Sun, L. and Dulaney, W. (2018). Field - scale mapping of evaporative stress indicators of crop yield: An application over Mead, NE, USA. Remote Sensing o f Environment , 210 , 387 - 402. Yang, Z., Wilkerson, G. G., Buol, G. S., Bowman, D. T., & Heiniger, R. W. (2009). Estimating genetic coefficients for the CSM - CERES - Maize model in North Carolina environments. Agronomy Journal , 101(5), 1276 - 1285. Yang, Z., Zhang, Q., Yang, Y., Hao, X., & Z hang, H. (2016). Evaluation of evapotranspiration Hydrological Processes , 30 (23), 4292 - 4313. Yin, X., Struik, P. C., & Kropff, M. J. (2004). Role of crop physiology in predicting gene - to - phenotype relati onships. Trends in Plant Science , 9(9), 426 - 432. Zeng, S. (2018). GRIB2 Data Decoder Application in MyWIDA: A Developers Guide (No. ARL - TN - 0903). US Army Research Laboratory Adelphi United States. Zhang, B., Xia, Y., Long, B., Hobbins, M., Zhao, X., Hain, C., Li, Y. and Anderson, M.C., 2020. Evaluation and comparison of multiple evapotranspiration data models over the contiguous United States: Implications for the next phase of NLDAS (NLDAS - Testbed) development. Agricultural and Forest Meteorology , 280 , p.1 07810. Zhang, T., Mahmood, R., Lin, X., & Pielke Sr, R. A. (2019). Irrigation impacts on minimum and maximum surface moist enthalpy in the Central Great Plains of the USA. Weather and Climate Extremes , 23 , 100197. Zhang, Y., Zhao, Y., & Feng, L. (2019). Hi gher contributions of uncertainty from global climate change. Meteorological Applications , 26 (1), 74 - 82. Zhao, W., Li, J., Yang, R., & Li, Y. (2018). Determining placement criteria of moistur e sensors through temporal stability analysis of soil water contents for a variable rate irrigation system. Precision agriculture , 19 (4), 648 - 665. Zhou, B., Yue, Y., Sun, X., Ding, Z., Ma, W., & Zhao, M. (2017). Maize kernel weight responses to sowing date - associated variation in weather conditions. The Crop Journal , 5(1), 43 - 51. Zhou, Y., Li, X., Yang, K., & Zhou, J. (2018). Assessing the impacts of an ecological water diversion project on water consumption through high - resolution estimations of actual eva potranspiration in the downstream regions of the Heihe River Basin, China. Agricultural and Forest Meteorology , 249 , 210 - 227. model ensemble prediction of ter restrial evapotranspiration across north China using Bayesian model averaging. Hydrological Processes , 30 (16), 2861 - 2879. Zhu, G.F., Li, X., Su, Y.H., Zhang, K., Bai, Y., Ma, J.Z., Li, C.B., Hu, X.L. and He, J.H. (2014). Simultaneous parameterization of th e two - source evapotranspiration model by Bayesian 162 approach: application to spring maize in an arid region of northwest China. Geoscientific Model Development Discussions , 7 , pp.741 - 775. Zhu, X., & Troy, T. J. (2018). Agriculturally Relevant Climate Extreme s and Their Trends in the World's Major Growing Regions. Earth's Future , 6 (4), 656 - 672.