DEVELOPMENT OF A METEOROLOGICAL, AGRICULTURAL, STREAM HEALTH, AND HYDROLOGICAL (MASH) COMPREHENSIVE DROUGHT INDEX   By Elaheh Esfahanian       A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Biosystems Engineering  Doctor of Philosophy 2016      ABSTRACT  DEVELOPMENT OF A METEOROLOGICAL, AGRICULTURAL, STREAM HEALTH, AND HYDROLOGICAL (MASH) COMPREHENSIVE DROUGHT INDEX By Elaheh Esfahanian Droughts are one of the costliest of natural disasters, posing a significant threat to both man-made and natural systems. Hundreds of drought indices are currently available for the monitoring of drought magnitude, severity, and extent; however, most of these indices were primarily designed for the analysis of drought impact on human concerns, such as crop production and freshwater supplies, and do not consider greater environmental aspects such as stream health. To the best of my knowledge, no universal drought index has been developed with the ability to comprehensively quantify different aspects of drought (e.g. meteorological, agricultural, hydrological, and stream health). In addition, there is no general agreement for drought definition even within each drought category. This means that different drought indices, even in the same category, can report contradictory results.    In order to address these issues, we designed a study based on the following research objectives: 1) development of an index capable of determining the impact of drought on aquatic ecosystems and stream health; 2) creation of a universal drought index for the measurement of multiple impacts of drought (e.g. meteorological, hydrological, agricultural, and stream health); and 3) determination of a predictive drought model that is able to capture both the categorical and overall impacts of drought. To address the first objective, we coupled a soil and water assessment tool (SWAT) with a regional-scale habitat suitability model to investigate drought conditions in the Saginaw River Watershed. Using the ReliefF algorithm as our variable selection method along with partial least squared regression, six predictive stream health drought   models were developed to monitor stream health drought conditions. Of these models, the version with five flow-related variables was determined to be the best tool for predicting both stream health and drought severity. For objective two, thirteen commonly used drought indices from the following categories were integrated to devise a definition of drought that is both categorical and universal: meteorological (4 indices), hydrological (4 indices), agricultural (4 indices), and stream health (1 index). The three closest indices to each other in each category were selected and then averaged to obtain the categorical drought scores; next, the simple average method was used to aggregate the categorical scores, which then provided the universal drought score.  For objective three, the ReliefF algorithm was used to select the best variable set for each of the categorical drought scores as well as for the universal drought score. The highest ranked variables were then used in the development of the various predictive drought models via the adaptive network-based fuzzy inference system. The adaptive network-based fuzzy inference system successfully produced four predictive drought models, including the three categorical models (meteorological, agricultural, and hydrological) and the universal drought model.   iv             Dedicated to my loving husband, Alireza, and my wonderful parents, Vahid and Maryam              v  ACKNOWLEDGEMENTS I would like to thank all the people who have helped and supported me throughout my journey. First, I would like to thank my major advisor Dr. Pouyan Nejadhashemi for always being ready to provide me with invaluable help, advice, and input. I am extremely grateful for all the encouragement, motivation, support, and direction throughout my time in the program. In addition, I would like to thank my co-advisor, Dr. Jade Mitchell, and my committee members, Dr. Timothy Harrigan and Dr. Nathan Moore, for all their guidance and support.  I would also like to thank Zhen Zhang (Statistics) and Ying Tang (Geography) for their contributions to different aspects of my research. And many thanks to Barb and Jaime Lynn, who made the paperwork and administrative side of the process so easy and convenient for me.  Thanks Fariborz Daneshvar, Melissa Rojas-Downing, Matthew Herman, Mohammad Abouali, Umesh Adhikari, Pouyan Hatami, Sean Woznicki, Georgina Sanchez, and Subhasis Giri for all the laughter and great memories. and encouragement during the pursuit of my graduate degree: they have always been there for me; and to my sister, Shiva, for always being ready to cheer me up. I would also like to thank Dr. Abdol Esfahanian and his lovely wife for their endless support, and for making my life so much easier while I was thousands of miles away from home. Finally, many thanks to my lovely husband, Alireza Ameli, for always being there for me no matter what. His endless love, support, and understanding have helped me to make it across the finish line! vi  TABLE OF CONTENTS LIST OF TABLES ........................................................................................................................................ x LIST OF FIGURES .................................................................................................................................... xii KEY TO ABBREVIATIONS .................................................................................................................... xiv 1. INTRODUCTION ................................................................................................................................ 1 2. LITERATURE REVIEW ..................................................................................................................... 4 2.1. Overview ................................................................................................................................... 4 2.2. Drought Definitions .................................................................................................................. 4 2.3. Drought Classification .............................................................................................................. 5 2.4. Modern Impact of Drought around the Globe ........................................................................... 7 2.5. Causes of Drought ..................................................................................................................... 9 2.6. Drought Indices ....................................................................................................................... 11 2.6.1. Palmer drought severity index ............................................................................................ 17 2.6.1.1. Applications ................................................................................................................ 18 2.6.1.2. Advantages .................................................................................................................. 18 2.6.1.3. Limitations .................................................................................................................. 18  Standardized precipitation index ......................................................................................... 19 2.6.2.1. Applications ................................................................................................................ 20 2.6.2.2. Advantages .................................................................................................................. 21 2.6.2.3. Limitations .................................................................................................................. 21 2.6.3. Crop moisture index ............................................................................................................ 22 2.6.3.1. Advantages .................................................................................................................. 22 2.6.3.2. Limitations .................................................................................................................. 22 2.6.4. Palmer hydrological drought index ..................................................................................... 23 2.6.5. Base-flow index .................................................................................................................. 23 2.6.5.1. Applications ................................................................................................................ 24 2.6.5.2. Advantages .................................................................................................................. 24 2.6.5.3. Limitations .................................................................................................................. 24 2.6.6. Surface water supply index ................................................................................................. 25 2.6.6.1. Advantages .................................................................................................................. 25 2.6.6.2. Limitations .................................................................................................................. 25 2.6.7. Normalized difference vegetation index ............................................................................. 26 vii  2.6.7.1. Applications ................................................................................................................ 26 2.6.7.2. Advantages .................................................................................................................. 27 2.6.7.3. Limitations .................................................................................................................. 27 2.6.8. Vegetation condition index ................................................................................................. 28 2.6.8.1. Applications ................................................................................................................ 28 2.6.8.2. Advantages .................................................................................................................. 29 2.6.8.3. Limitations .................................................................................................................. 29 2.6.9. Recent developments in drought indices ............................................................................. 29 2.6.9.1. Effective precipitation ........................................................................................................ 29 2.6.9.2. Reconnaissance drought index ........................................................................................... 30 2.6.9.3. Flow duration curve ........................................................................................................... 30 2.6.9.4. Standardized runoff index .................................................................................................. 30 2.6.9.5. Water balance derived drought index ................................................................................ 31 2.6.9.6. Reclamation drought index ................................................................................................ 31 2.6.9.7. Indices based on soil moisture ........................................................................................... 31 2.6.9.8. Indices based on remote sensing ........................................................................................ 32 2.6.9.9.   Drought monitor ............................................................................................................... 33 2.7. Climate Change ....................................................................................................................... 33 2.7.1. Drought and Climate Change .............................................................................................. 35 2.8. Bioassessment ......................................................................................................................... 36 2.8.1. Stream Health ...................................................................................................................... 43 2.8.1.1. Fish as Indicators ............................................................................................................... 43 2.8.1.1.1. Index of biotic integrity ............................................................................................... 44 2.8.1.2. Macroinvertebrates as indicators ....................................................................................... 45 2.8.1.2.1. Benthic index of biotic integrity .............................................................................. 46 2.8.1.2.2. Hilsenhoff biotic index ............................................................................................ 47 2.8.1.2.3. Ephemeroptera, Plecoptera, and Trichoptera Index ................................................. 47 2.8.2. Effects of climate change on bioassessment programs ....................................................... 48 2.9. Drought Risk Assessments ...................................................................................................... 48 2.10. Drought Modeling ................................................................................................................... 50 2.10.1. Drought forecasting............................................................................................................. 50 2.10.2. Probabilistic characterization of drought ............................................................................ 54 2.10.3. Spatio-temporal drought analysis ........................................................................................ 56 2.10.4. Drought modeling under climate change scenarios ............................................................ 57 2.10.5. Land data assimilation systems ........................................................................................... 59 2.10.6. Drought Management ......................................................................................................... 60 viii  2.11. Summary ................................................................................................................................. 62 3. INTRODUCTION TO METHODOLOGY AND RESULTS ............................................................ 64 4. DEFINING DROUGHT IN THE CONTEXT OF STREAM HEALTH ........................................... 66 4.1. Abstract ................................................................................................................................... 66 4.2. Introduction ............................................................................................................................. 66 4.3. Materials and Methodology .................................................................................................... 70 4.3.1. Study area ............................................................................................................................ 70 4.3.2. Modeling process ................................................................................................................ 71 4.3.3. Soil and Water Assessment Tool ........................................................................................ 72 4.3.4. SWAT model calibration and validation ............................................................................. 73 4.3.5. Regional-scale Habitat Suitability Model ........................................................................... 74 4.3.6. Drought Model Input Variables .......................................................................................... 77 4.3.7. Variable Selection: ReliefF algorithm ................................................................................ 78 4.3.8. Partial Least Square Regression .......................................................................................... 80 4.3.9. Climate Models ................................................................................................................... 82 4.4. Results & Discussions ............................................................................................................. 86 4.4.1 SWAT Model Calibration and Validation .......................................................................... 86 4.4.2 Variable Selection ............................................................................................................... 87 4.4.3.1. Current Drought Severity Model ................................................................................ 87 4.4.2.2. Future Drought Severity Model .................................................................................. 89 4.4.3 Drought Severity Model ...................................................................................................... 89 4.4.3.1. PLSR predictively for median flow ............................................................................ 89 4.4.3.2. Accuracy, precision, and sensitivity of drought models in predicting drought zones . 95 4.4.4 Drought model performance under future climate scenarios .............................................. 97 4.4.5 The impact of climate change on future drought ................................................................ 98 4.5. Conclusion ............................................................................................................................ 101 4.6. Acknowledgments ................................................................................................................. 102 5. DEVELOPMENT AND EVALUATION OF A COMPERHENSIVE DROUGHT INDEX .......... 104 5.1. Abstract ................................................................................................................................. 104 5.2. Introduction ........................................................................................................................... 105 5.3. Materials and Methodology .................................................................................................. 109 5.3.1. Study area .......................................................................................................................... 109 5.3.2. Modeling process .............................................................................................................. 111 5.3.3. Categorical drought index development ........................................................................... 112 ix  5.3.3.1 Meteorological Drought Indices ................................................................................... 113 5.3.3.2 Agricultural Drought Indices ........................................................................................ 114 5.3.3.3 Hydrological Drought Indices ....................................................................................... 115 5.3.3.4 Stream Health drought Index ........................................................................................ 116 5.3.4. Input parameters ................................................................................................................ 117 5.3.5. Transformation and Clustering ......................................................................................... 119 5.3.6. Aggregation ....................................................................................................................... 120 5.3.7. Drought indices comparison ............................................................................................. 121 5.3.8. Drought model development ............................................................................................. 122 5.3.8.1. Parameter selection ................................................................................................... 122 5.3.8.2. Development of predictive drought models .............................................................. 123 5.4. Results and Discussions ........................................................................................................ 125 5.4.1 Statistical Analysis of Drought Indices ............................................................................. 125 5.4.2 Categorical Drought Indices ............................................................................................. 127 5.4.3 Comparison of Categorical Drought Scores and MASH .................................................. 128 5.4.4 Variable Selection ............................................................................................................. 131 5.4.5 Categorical and MASH drought models ........................................................................... 133 5.4.6 Identifying the drought vulnerable areas ........................................................................... 136 5.5. Conclusion ............................................................................................................................ 138 5.6. Acknowledgements ............................................................................................................... 139 6. CONCLUSIONS ............................................................................................................................... 140 7. FUTURE RESEARCH RECOMMENDATIONS ............................................................................ 142 APPENDICES .......................................................................................................................................... 143 APPENDIX A: Study One .................................................................................................................... 144 APPENDIX B: Study Two ................................................................................................................... 161 REFERENCES ......................................................................................................................................... 179   x  LIST OF TABLES Table 1. Summary of popular drought indices ............................................................................................ 12  Table 2. Classification of SPI values (adapted from McKee et al., 1993; 1995) ........................................ 20  Table 3. Biological response to increasing levels of stress (adapted from USEPA, 2011b; Davies and Jackson, 2006)............................................................................................................................................. 41  Table 4. Stressor identification process (adapted from USEPA, 2000; USEPA, 2011b) ........................... 42  Table 5. Reference table of drought zones (adapted from Hamilton and Seelbach, 2011) ......................... 77  Table 6. CMIP5 models developer, name, resolution, and components (Petkova et al., 2013; IPCC, 2013) .................................................................................................................................................................... 84  Table 7. Statistical criteria for SWAT model calibration and validation for nine USGS gauging stations within the Saginaw Bay Watershed ............................................................................................................ 86  Table 8. Top 15 ranked variables ................................................................................................................ 89  Table 9. Current Drought Severity Model performances............................................................................ 90  Table 10. Future Drought Severity Model performances ........................................................................... 91  Table 11. Confusion matrix for drought zones: First model ....................................................................... 96  Table 12. Confusion matrix for drought zones: Fourth Model ................................................................... 96  Table 13. Overall first model performance against 47 future climate scenarios ......................................... 97  Table 14. ANFIS models frameworks and characteristics ........................................................................ 124  Table 15. p-values from pairwise comparison of drought indices. Red colored p-values indicate no significant mean differences at the 0.05 level. .......................................................................................... 129  Table 16. Frequency of drought indices combinations in each drought category over 30-year period .... 130  Table 17. Top five ranked variables that were used for development of the drought predictive models. 132  Table 18. Best ANFIS models for each drought category and MASH ..................................................... 134  Table S1. Selected variables for development of current and future drought severity models. ................ 151 xi  Table S2. Confusion matrix for drought zones: Second model ................................................................ 156  Table S3. Confusion matrix for drought zones: Third model ................................................................... 156  Table S4. Confusion matrix for drought zones: Fifth Model .................................................................... 157  Table S5. Confusion matrix for drought zones: Sixth Model ................................................................... 157  Table S6. The first drought model performance using RCP 8.5 (maximum and minimum values are presented in red) ........................................................................................................................................ 158  Table S7. The first drought model performance using RCP 6.0 (maximum and minimum values are presented in red) ........................................................................................................................................ 159  Table S8. The first drought model performance using RCP 4.5 (maximum and minimum values are presented in red) ........................................................................................................................................ 160  Table S9. Meteorological drought indices, reference, input parameters, procedure, classification, and index value ................................................................................................................................................ 164  Table S10. Agricultural drought indices, reference, input parameters, procedure, classification, and index value .......................................................................................................................................................... 165  Table S11. Hydrological drought indices, reference, input parameters, procedure, classification, and index value .......................................................................................................................................................... 167  Table S12. Stream health drought index, reference, input parameters, procedure, classification, and index value .......................................................................................................................................................... 168  Table S13. Input parameters ..................................................................................................................... 169  Table S14. Mean difference (numbers in black) and standard deviation (numbers in red) among drought indices ....................................................................................................................................................... 175  Table S15. Saginaw River watershed calibration and validation results .................................................. 176  Table S16. Transformed drought categories * .......................................................................................... 177  Table S17. Transformed non-drought categories * ................................................................................... 178    xii  LIST OF FIGURES Figure 1. Saginaw Bay Watershed .............................................................................................................. 71  Figure 2. Drought zones variable selection and modeling process ............................................................. 72  Figure 3. Fish response curve to flow reduction (adapted from Zorn et al., 2008) ..................................... 76  Figure 4. ReliefF ranking histogram map ................................................................................................... 88  Figure 5. The variance explained percentage for each PLSR for the Current Drought Severity Model: a) First model, b) Second model, c) Third model. .......................................................................................... 92  Figure 6. The comparison of measured vs. predicted median flow histogram for the Current Drought Severity Model, a) First model, b) Second model, c) Third model. ........................................................... 94  Figure 7. Probability of increasing drought conditions under projected climate change (2040-2060) compare to current condition (1990-2010). ................................................................................................ 99  Figure 8. Percent change in (a) temperature and (b) precipitation from current (1980-2000) to future climate change (2040-2060)...................................................................................................................... 100  Figure 9. Saginaw River Watershed ......................................................................................................... 110  Figure 10. Categorical drought scores development and modeling process ............................................. 112  Figure 11. Measured versus modeled histograms of categorical drought and MASH: (a) CMI, (b) CHI, (c) CAI, and (d) MASH. ................................................................................................................................. 135  Figure 12. Drought vulnerable areas based on MASH in the Saginaw River watershed.......................... 137  Figure S1.Locations of precipitation, temperature, and streamflow monitoring stations ......................... 144  Figure S2. Distribution of median flow values ......................................................................................... 145  Figure S3. Sample histogram of ranking for parameter #20 (average flow rate from 23 months prior to the month of interest) ...................................................................................................................................... 146  Figure S4. The relationship of MSE with the number of PLSR components for Current Drought Severity Models: a) First Model, b) Second Model, c) Third Model. ..................................................................... 147  Figure S5.The relationship of MSE with the number of PLSR components for Future Drought Severity Models: a) Fourth Model, b) Fifth Model, c) Sixth Model. ...................................................................... 148 xiii  Figure S 6. The variance explained percentage for each PLSR for the Future Drought Severity Model: a) Fourth Model, b) Fifth Model, c) Sixth Model. ........................................................................................ 149  Figure S7. The comparison of measured vs. predicted median flow histogram for the Future Drought Severity Model: a) Fourth Model, b) Fifth Model, c) Sixth Model. ......................................................... 150  Figure S8. Location of temperature, precipitation, and streamflow gauging stations .............................. 161  Figure S9. Measured versus modeled histograms of categorical drought and MASH: (a) CMI, (b) CHI, (c) CAI, and (d) MASH .................................................................................................................................. 162  Figure S10. Drought vulnerable areas based on categorical drought indices in the Saginaw River watershed: (a) meteorological, (b) hydrological, (c) agricultural, (d) stream health ................................ 163    xiv  KEY TO ABBREVIATIONS  A: Agricultural ADI: Aggregate Drought Index AMO: Atlantic Multidecadal Oscillation ANFIS: Adaptive Neuro-Fuzzy Interference System  ANN: Artificial Neural Network  ARI: Adverse Resource Impacts ARIMA: Autoregressive Integrated Moving Average  ARS: Agricultural Research Service AVHRR: Advanced Very High Resolution Radiometer BCG: Biological Condition Gradient BFI: Baseflow Index  B-IBI: Benthic Index of Biotic Integrity BMP: Best Management Practice CADDIS: Causal Analysis/Diagnosis Decision Information System CAI: Categorical Agricultural Index CART: Classification and Regression Tree CDF: Cumulative Distribution Function CDI: Combined Drought Index CDL: Cropland Data Layer CHI: Categorical Hydrological Index CMI: Categorical Meteorological Index xv  CMI: Crop Moisture Index  CMIP5: Coupled Model Intercomparison Project Phase 5 CPC: Climate Prediction Center CSHI: Categorical Stream Health Index CWA: Clean Water Act DEP: Deviation of EP from MEP DM: Drought Monitor DMAPS: Drought Monitoring and Prediction System DMSNN: Direct Multistep Neural Network  DSI: Drought Severity Index DSS: Decision Support System EDI: Effective Drought Index  ENSO: El Nino Southern Oscillation EP: Effective Precipitation EPA: Environmental Protection Agency EPT: Ephemeroptera, Plecoptera, and Trichoptera Index ETDI: Evapotranspiration Deficit Index FDC: Flow Duration Curve FL: Fuzzy Logic  GCMs: General Circulation Models GPCC-DI: Global Precipitation Climatology Centre Drought Index H: Hydrological HBI: Hilsenhoff Biotic Index xvi  HDI: Hybrid Drought Index HRUs: Hydrologic Response Units HUC: Hydrologic Unit Code IBI: Index of Biotic Integrity IPCC: Intergovernmental Panel on Climate Change LPD: Liter per Day M: Meteorological MASH: Meteorological, Agricultural, Stream health, and Hydrological MCDA: Multi-Criteria Decision Analysis MEP: Mean of Effective Precipitation MFs: Membership Functions MMIs: Multimetric Indices MSE: Mean Square Error NAO: North Atlantic Oscillation NCDC: National Climatic Data Center NDMC: National Drought Mitigation Center NDVI: Normalized Difference Vegetation Index  NDWI: Normalized Difference Water Index NED: National Elevation Dataset NIR: Near Infrared NLDAS: North American Land Data Assimilation System NOAA: National Oceanic and Atmospheric Administration NPDES: National Pollutant Discharge Elimination System xvii  NPS: Nonpoint Source  NRCS: USDA Natural Resources Conservation Service NSE: Nash-Sutcliffe Efficiency Coefficient NSF-DOE-NCAR: National Science Foundation, Department of Energy, National Center for Atmospheric Research  NWS: National Weather Service PBIAS: Percent Bias PDI: Precipitation Drought Index  PDO: Pacific Decadal Oscillation PDSI: Palmer Drought Severity Index PHDI: Palmer Hydrological Drought Index  PLSR: Partial Least Square Regression PMDI: Palmer Modified Drought Index PMF: Probability Mass Function  RCMs: Regional Climate Models RCPs: Representative Concentration Pathways RD: Rainfall Deciles RDAI: Regional Drought Area Index  RDI: Reclamation Drought Index RDI: Reconnaissance Drought Index RMSE: Root Mean Square Error RMSNN: Recursive Multistep Neural Network  RSR: Root-Mean-Squared Error-Observations Standard Deviation Ratio xviii  S: Stream Health SAF: Severity-Area-Frequency SARIMA: Seasonal Autoregressive Integrated Moving Average  SEP: Standardized Value of DEP SHI: Stream Health Index  SI: Stressor Identification SMDI: Soil Moisture Deficit Index SMI: Soil Moisture Index SPEL: Standardized Precipitation Evapotranspiration Index SPI: Standardized Precipitation Index  SRI: Standardized Runoff Index SSURGO: Soil Survey Geographic SWAT: Soil and Water Assessment Tool SWDI: Soil Water Deficit Index SWIR: Short-Wave Infrared SWSI: Surface Water Supply Index  TDI: Temperature Drought Index TMDLs: Total Maximum Daily Loads UK: United Kingdom UN-ISDR: United Nations International Strategy for Disaster Reduction USGS: US Geological Survey VCI: Vegetation Condition Index  VCI: Vegetation Condition Index xix  VDI: Vegetation Drought Index VegDRI: Vegetation Drought Response Index VegOut: Vegetation Outlook VIC: Variable Infiltration Capacity VIS: Visible WBI: Water balance Derived Drought Index WDCC: Western Drought Coordination Council WET: Whole effluent toxicity WQS: Water Quality Standards Z-index: Palmer Moisture Anomaly Index      1  1. INTRODUCTION Drought is a natural event that occurs in most climate zones as an effect of the long-term reduction of precipitation within a region. Of all existing natural hazards, drought is the most detrimental in terms of human impact (Wilhite, 2000b; Mishra and Singh, 2010). Globally, drought causes approximately $8 billion in damage annually, making it t type of natural disaster (Wilhite, 2000b; Keyantash and Dracup, 2002). Although a natural phenomenon, various human activities can directly trigger droughts by impeding the ability of the land to capture and hold water, including: intensive farming, excessive irrigation, deforestation, the over-exploitation of available water, and erosion (Wilhite, 2000a; Mishra and Singh, 2010). Droughts are generally classified as being meteorological, agricultural, or hydrological (Wilhite and Glantz, 1985; American Meteorological Society, 1997; McMahon and Finlayson, 2003; Dai, 2011): meteorological droughts are a result of a prolonged period of below-average precipitation caused by anomalies in atmospheric circulation patterns (Dai, 2011). Agricultural drought is caused by a period of soil moisture loss triggered by a shortage of precipitation (Mishra and Singh, 2010; Dai, 2011). Hydrological drought is caused by a period of reduction in streamflow, runoff, and inflow to reservoirs as a result of precipitation deficiency (Whitmore, 2000). It is difficult to determine the exact start and end dates of a drought, as the various impacts of a given drought increase slowly, accumulate over time, and can even remain after the end of the drought (Mishra and Singh, 2010). These characteristics have led to drought being known as a creeping phenomenon (Whitmore, 2000; Mishra and Singh, 2010). Several indices have already been developed to monitor and quantify different types of drought; these indices are the primary tools for the assessment of drought severity, duration, and 2  intensity (Heim, 2002; Mishra and Singh, 2010).  Each drought index requires specific input parameters to measure drought; however, precipitation is typically used, either alone or in combination with other parameters (Heim, 2002; Mishra and Singh, 2010; Sheffield and Wood, 2011). In the case of meteorological drought, precipitation is traditionally the primarily parameter used; soil moisture content is commonly used for agricultural drought (along with the secondary parameters of precipitation and evapotranspiration); and hydrological drought parameters typically include streamflow and precipitation (Dai, 2011). Most drought indices quantify drought impact based on effects on human activities such as agricultural production while neglecting environmental sustainability effects like stream health; to rectify this oversight, the goal of the first study was to investigate the impacts of drought on stream health. The specific objectives were:  Identification of a method of variable selection for determination of the most influential parameters in the development of stream health drought models.   Development of a predictive model for quantifying the aggregate risk of drought on stream health.   Evaluation of the impact of climate change on stream health drought models.  The effects of drought are non-structural and spatially extensive (Wilhite et al., 2000c); resulting in widespread impacts on different sectors, including hydrology, meteorology, agriculture, natural ecosystems, and human wellbeing. There is currently no universal definition used for drought, since each sector measures it differently (Whitmore, 2000; Heim, 2002; Svoboda et al., 2002); this absence of a universal definition is itself one of the main obstacles to the effective study of drought (Mishra and Singh, 2010).  3  Despite recent advances in the scientific study of drought, monitoring methods are still in need of significant improvement; these changes would streamline drought preparation and management practices, as well as reduce vulnerability to drought in several different sectors (Svoboda et al., 2002). One method of improving drought monitoring is the combination of existing indices to better evaluate the overall impacts of a drought (Zargar et al., 2011). Meanwhile, hundreds of indices have been developed for each drought category due to the fact that no general agreement exists on how to formulate categorical drought indices (e.g. meteorological, agricultural, or hydrological). This means that different drought indices can report contradictory results.    The goal of the second study was the creation of a universal definition by introducing an overall drought index that considers multiple aspects of drought, including the meteorological, agricultural, hydrological, and stream health. This universal definition would substantially improve the current system of drought monitoring, thereby enabling decision-makers to more effectively allocate resources for the reduction of drought impacts across different sectors.    The objectives of the second study are:  Definition of the four categorical drought indices (meteorological, agricultural, hydrological, and stream health) based on commonly used drought indices.  Creation of a universal definition of drought via the combination of the categorical scores.   Selection of the best variable sets for construction of predictive drought models.  Development of predictive drought models for each drought category as well as the universal drought index.    4  2. LITERATURE REVIEW 2.1. Overview This literature review provides an overview of drought concepts and characterizations, risk assessment, and modeling. Section 2.2 provides drought definitions, which explain direct drought causation. Section 2.3 contains drought classifications that identify the impacts of drought on different sectors. Section 2.4 describes the global impacts of drought. Section 2.5 discusses the indirect causes of drought considering atmospheric and hydrological interactions. Section 2.6 describes the various drought indices that are used to measure drought characterizations such as severity, duration, intensity, frequency, and spatial extent. Following these sections, a discussion of climate change and its impact on drought, bioassessment and stream health (including the benefits of bioassessment and newly established tools, stream health indicators, and the effects of climate change on current bioassessment programs). Drought risk assessment is later addressed, including results from past studies that developed the current drought risk analysis guidelines. Lastly, drought modeling and its various components are discussed, including drought forecasting, probabilistic characterization of drought, spatiotemporal drought analysis, drought modeling using climate change scenarios, land data assimilation systems, and drought management.    2.2. Drought Definitions Simply defined, a drought is an extended deficit in the amount of water compared to normal conditions governed by the hydrological cycle. The hydrological cycle is the movement of water through land, ocean, and atmosphere; its main components are precipitation, evaporation, run-off, snow-melt, and soil and groundwater storage (Sheffield and Wood, 2011). Due to the large number of diverse definitions for droughts, determination of a universal and 5  precise definition of drought has proven unfeasible (Yevjevich, 1967; Mishra and Singh, 2010). Drought definitions are categorized as either conceptual or operational: conceptual definitions utilize relative concepts to describe drought in simple terms, while operational definitions are much more in-depth. Operational definitions are used to identify the frequency, severity, duration, and termination of drought, and are used in preparation for future droughts (SOEST, 2003; Mishra and Singh, 2010). Some of the most commonly used definitions are provided below:  The smallest daily streamflow value of the year (Gumbel, 1963);  Extended periods during which lack of moisture results in crop failure (Unger, 1984);  A sustained decrease in the amount of precipitation normally received in a specific area (WMO, 1986);  A naturally occurring phenomenon that results when precipitation levels fall significantly below normally recorded levels, causing severe hydrological imbalances that negatively impact land resource production systems (UN Secretary-General, 1994);  A sustained period (e.g. a season, a year, or several years) of deficient rainfall anomalous to the statistical multi-year mean of a given region (Schneider, 1996). 2.3. Drought Classification Droughts are typically categorized as one of four major types: meteorological, hydrological, agricultural, and socio-economic (Wilhite and Glantz, 1985; American Meteorological Society, 1997); however, Mishra and Singh (2010) introduced groundwater drought as a new type of drought. Similarly, Sheffield and Wood (2011), instead of socio-economic category, introduced ecological and regional categories as new types of drought in 6  order to focus on the environmental impacts of drought (ecological drought) rather than the socio-economic impacts.  A meteorological drought occurs when there is a significant deviation from the mean precipitation in a region over an extended period of time; precipitation data are used to identify and analyze this type of drought (Mishra and Singh, 2010; Sheffield and Wood, 2011).   Hydrological drought refers to a period of deficiency in the supply of water (both surface and subsurface) of a given water resource management system (Panu and Sharma, 2002; Mishra and Singh, 2010; Sheffield and Wood, 2011). The following datasets are used to analyze hydrological droughts: streamflow, lake and reservoir levels, and groundwater levels (Mishra and Singh, 2010; Sheffield and Wood, 2011).   Agricultural drought is defined as a period of soil moisture deficiency leading to a reduction in the moisture supply available for crops and other types of vegetation (Panu and Sharma, 2002; Sheffield and Wood, 2011); this type of drought is driven by meteorological and hydrological droughts (Sheffield and Wood, 2011). Several drought indices have been used to study agricultural drought, featuring a combination of hydrometeorological variables such as precipitation, soil moisture, and temperature (Mishra and Singh, 2010).    Socio-economic drought refers to a combination of meteorological, hydrological, and agricultural droughts which result in adverse social and economic impacts on humans. This type of drought differs from those in the other three categories due to its direct link to the relationship between supply and demand for a given economic good (i.e., water): when the demand for water exceeds the supply, the result is a socio-economic drought (American Meteorological Society, 2004; Mishra and Singh, 2010; Sheffield and Wood, 2011).  7   Groundwater drought is defined as a lack of groundwater recharge over a prolonged period of time as a result of low precipitation and high evapotranspiration. This type of drought is mainly associated with low groundwater heads, small groundwater gradients, low groundwater storage, and low well yields (shallow wells may even dry up) (van Lanen and Peters, 2000; Mishra and Singh, 2010). Groundwater levels and gradients are used to quantify the effects of this kind of drought (van Lanen and Peters, 2000).   Ecological drought measures the impacts of drought on ecosystems and it is caused by a reduction in soil moisture due to low precipitation (causing a reduction in evapotranspiration), which adversely affects local vegetation (Sheffield and Wood, 2011).  Regional drought is defined as a period during which more than 70% of a given area (within a larger region) is affected by drought (Fleig et al., 2011).   2.4. Modern Impact of Drought around the Globe Droughts affect many sectors of society, including the economy, agriculture, industry, infrastructure, and tourism. Drought may have led to the declines of Sumer in pre-Roman times, and to the Mayan civilization in the past millennium. In the 20th century, droughts have caused the most detrimental economic and social impacts of all natural disasters (Mishra and Singh, 2010; Sheffield and Wood, 2011). In recent decades, multiple continents have been severely affected by drought (Mishra and Singh, 2010). Drought has also proven the costliest natural disaster in the United States, with average annual damages estimated at approximately $6-$8 billon (Mishra and Singh, 2010; Sheffield and Wood, 2011). Droughts accounts for 41% of the total estimated cost of all weather-related disasters in the U.S. (Cook et al., 2007; Mishra and Singh, 2010). Regional droughts in 1988 (central US) and 1996 (state of Texas) resulted in estimated losses of $46 billion (Sheffield and Wood, 2011).  8  In the past two centuries, various regions of Canada (particularly the Canadian prairies) have experienced severe droughts. The prairies are one of the most drought-prone regions due to their high precipitation variability; in 2001-2002, one of the most severe prairie droughts on record caused significant damage to water-related resources. In the 1890s, 1930s, and 1980s southern regions experienced multi-year droughts. During the 20th century, western Canada experienced at least 40 long-term droughts, and eastern Canada also suffered from major drought events (Environment Canada, 2004; Mishra and Singh, 2010).  Over the past 30 years, Europe has experienced several major droughts, resulting in economic losses of Wood, 2011; European Communities, 2012). In 2003, a prolonged drought associated with a heat wave that affected large parts of Europe cost 0; European Communities, 2012). Lehner et al. (2006) conducted a study on the possible impact of global climate change on drought frequency in Europe and concluded that, based on their proposed climate change scenarios, southern and southeastern Europe are more likely to experience significant increases in drought frequency than northern and northeastern Europe.  In Asia, agricultural production has declined in recent decades due to increasing water stress, which is a result of rising temperatures, a reduction in the number of rainy days, and the increasing frequency of El Nino events (Bates et al., 2008; Mishra and Singh, 2010). From 1998-2001, central and southwestern Asia experienced a severe drought and consequent famine which affected over 60 million people, particularly in Iran, Afghanistan, Pakistan, Tajikistan, Uzbekistan, and Turkmenistan (Barlow et al., 2002; Mishra and Singh, 2010). Since the late 1990s, most of northern China has experienced prolonged, severe droughts resulting in substantial economic and social losses (Zou et al., 2005; Mishra and Singh, 2010). India is one of 9  the most drought-prone countries, having experienced at least one drought per three-year period over the last five decades (Mishra and Singh, 2010).  Drought is also a recurring event in Australia, especially in the southern and eastern parts of the country in part because its rainfall is more strongly governed by El Nino. The millennium drought-2010) was the worst recorded drought since European settlement began (Bond et al., 2008; Mishra and Singh, 2010).  West Africa experienced a drought of unprecedented severity in the Sahel from the late 1960s to the mid-1980s which led to widespread famine and hundreds of thousands of deaths (Mishra and Singh, 2010; Sheffield and Wood, 2011). A slightly less severe drought occurred in East Africa during the mid-1980s which also caused famine and many deaths. In South Africa, multiple drought events occurred between the 1980s and early 1990s that were related to the El Nino Southern Oscillation (ENSO) (Sheffield and Wood, 2011).   2.5. Causes of Drought  The causes of drought are complex, as they are the outcome of the interaction of atmospheric and hydrological processes; drought is an extreme state of the hydrological cycle in which precipitation is below normal levels. Once established, dry hydrological conditions within a region cause the depletion of moisture from the upper layers of soil, subsequently causing a reduction in evapotranspiration rates and the sequential lowering of atmospheric relative humidity. Such decreases in relative humidity reduce the probability of rainfall (Bravar and Kavvas, 1991; Mishra and Singh, 2010); precipitation can also be reduced by both an increase in albedo and the accumulation of increase of fine particles in the air (Panu and Sharma, 2002; Nagarajan, 2009). Increases in albedo lower surface temperatures, resulting in local heat loss. 10  Lower surface temperatures cause a reduction in lifting air masses, which leads to a reduction in precipitation. Local heat loss causes a temperature gradient that induces a circulation capable of maintaining equilibrium with warmer surroundings, thereby depressing precipitation. Additionally, increases in the number of fine particles in the air can overseed clouds, which also can reduce precipitation (Panu and Sharma, 2002).  Another causation factor of drought are oceanic circulations that affect weather and climate; these circulations, with average patterns of current and heat storage, cause climate variations. Significant climatic variations occur when warm water from the western Pacific Ocean flows into the eastern-central equatorial Pacific Ocean (e.g. off the coast of Peru) (Panu and Sharma, 2002; Nagarajan, 2009). These anomalies in sea surface temperature create the El Nino effect, which has been associated with the onset of many recent droughts (Panu and Sharma, 2002; Nagarajan, 2009). The opposite occurs in the La Nina phenomena, which refers to the periodic cooling of sea surface temperatures in the eastern-central tropical Pacific Ocean (NOAA, 2012). These anomalies in sea surface temperature are due to large-scale atmospheric circulations which follow quasi-periodic cycles or oscillation (Panu and Sharma, 2002; Sheffield and Wood, 2011). Among these, ENSO has proven the most significant driver of global climate change, and oscillates approximately every two-to-seven years in the tropical Pacific Ocean (Sheffield and Wood, 2011). El Nino and La Nina are extreme phases of the ENSO, and represent warm and cold phases, respectively (Panu and Sharma, 2002; NOAA, 2012); the ENSO also affects hydrological features such as precipitation and streamflow over catchments (Panu and Sharma, 2002). There are other climate oscillations serving as primary drivers of regional climate variation which can act in other timescales, allowing them to interact with the ENSO.  These climate oscillations include the North Atlantic Oscillation (NAO), the Pacific 11  Decadal Oscillation (PDO), and the Atlantic Multidecadal Oscillation (AMO). The NAO affects the climate in eastern North America, Europe, and North Africa. The PDO manifests in the northern Pacific Ocean with a timescale of 20-to-30 years, and can interact with ENSO; it can also modify climate on a global scale. The AMO affects climate in the North Atlantic, especially in North America and Europe (Sheffield and Wood, 2011). However, like the weather, atmospheric drought is essentially unpredictable for timescales more than a month in advance despite significant efforts to improve our understanding. 2.6. Drought Indices Several drought indices have been developed to monitor drought conditions. Drought indices are prime tools for assessing drought effects and parameters; the parameters defined by these indices are duration, intensity, severity, and spatial extent. Each drought index requires different input parameters and uses a unique method to measure drought. The precipitation parameter is used in all indices, either alone or in combination with other meteorological parameters such as soil moisture and temperature (Heim, 2002; Mishra and Singh, 2010; Sheffield and Wood, 2011). Table 1 summarizes the most commonly used drought indices, including their respective strengths and limitations. In the following sections, some of the meteorological, agricultural, hydrological, and ecological drought indices were further explained.   12  Table 1. Summary of popular drought indices Index (References) Description and Use Strengths Weaknesses Meteorological Drought  Palmer Drought Severity Index (PDSI)  (Palmer, 1965; Alley 1984; Dai et al., 2004; Hayes, 2006; Mishra and Singh, 2010; Sheffield and Wood, 2011)  Utilizes a water balance model to depict departure of soil moisture from a given region (compared to normal conditions)   Uses precipitation and temperature as input parameters  Widely used by US governmental agencies  Good measure of intensity and duration of long-term drought   Facilitates direct comparisons between different regions and timeframes  Considers basic effects of surface warming  Values vary widely for extreme and severe drought classifications and frequencies in different locations  May lag in detecting emerging droughts by several months  All precipitation assumed to be rain  Rainfall Deciles (RD)  (Gibbs and Mahar, 1967; Hayes, 2006; Sheffield and Wood, 2011; Zargar et al., 2011)  Divides monthly precipitation events into deciles (10% each)  Can be computed for any chosen period  Used primarily in Australia   Relatively simple to calculate  Provides a precise statistical measurement of precipitation  Precipitation records covering extended periods needed to accurately calculate deciles  Standardized Precipitation Index (SPI)  (McKee et al., 1993; Edwards and McKee, 1997; Heim, 2002; Mishra and Singh, 2010; Sheffield and Wood, 2011)  Based on probability of precipitation  Calculated for any location with long-term monthly precipitation record   Quantifies precipitation deficit for multiple timescales  Solely based on precipitation  Temporal flexibility and versatility  Consistent classifications of severe and extreme drought frequencies in any location and timescale  For different lengths of precipitation records, SPI value discrepancies can be obtained as a result of different distributions  Dependent on nature of probability distribution  13  Table 1.     Percent of Normal  (Hayes, 2006; Sheffield and Wood, 2011; Zargar et al., 2011)  Calculated by dividing actual precipitation by normal precipitation   Normal precipitation typically considered to be a 30-year mean  Timescales can vary between one month and one year  Simple and transparent  Effective for comparing a single region and a specific period (within a given year)  Without normal distribution, mean and median values differ, causing inaccuracy   Unable to compare drought across multiple seasons or regions  Agricultural Drought    Palmer Moisture Anomaly Index (Z-index)  (Palmer, 1965; Dai et al., 2004; Sheffield and Wood, 2011; Zargar et al., 2011)  Calculates monthly standardized anomaly of available moisture  Used for monitoring short-term droughts  Input parameters: precipitation, streamflow, and temperature   Rapid response to changing conditions  Not used for monitoring long-term droughts  Antecedent conditions not considered   Crop Moisture Index (CMI)  (Palmer, 1968; Hayes, 2006; Mishra and Singh, 2010;  Sheffield and Wood, 2011)  Monitors short-term moisture supply (week-to-week) across crop regions  Derived from Palmer Index  Requires weekly temperature and precipitation values   Quick response to changing conditions  Can be used to compare moisture conditions at different locations   Easily computed from precipitation and temperature data  Not applicable to monitoring of long-term droughts   Rapid response to short-term changing conditions provides misleading information for monitoring of long-term conditions  14      Hydrological Drought    Palmer Hydrological Drought Index (PHDI)  (Palmer, 1965; Heim, 2000; Keyantash and Dracup, 2002; Mishra and Singh, 2010; Zargar et al., 2011)  Analyzes precipitation and temperature in the PDSI water balance model  Used for water supply monitoring and qualification of hydrological impacts of long-term drought conditions  Input parameters: precipitation, temperature, and streamflow/runoff    Used to monitor long-term droughts  Same as PDSI Baseflow Index (BFI)  (Institute of Hydrology, 1980; Gustard et al., 1992; Zaidman et al., 2001; Tallaksen and van Lanen, 2004; Sheffield and Wood, 2011)   Ratio of baseflow to total flow  Used for low-flow estimation and groundwater recharge assessment  Estimates low-flow indices at the ungauged site  Stored water in the basin used to quantify flow   Sensitive to missing data  Requires long-term records to separate baseflow from total flow Surface Water Supply Index (SWSI)  (Shafer and Dezman,1982; Heim, 2002; Hayes, 2006; Mishra and Singh, 2010;   Calculated based on monthly weighted sum of non-exceedance probabilities of snowpack, streamflow, precipitation, and   Simple to calculate and represent water supply conditions  Allows comparison of water supply availability among   The weight of each hydrological component in SWSI equation varies with spatial scale  Index measurement is  15      Sheffield and Wood, 2011)  reservoir storage components  Monitors abnormalities in surface water supplies  Developed in response to   Used for river basins in western US   regions with different variability  unique for each basin, making comparison between different basins difficult Ecological Drought     Normalized Difference Vegetation Index (NDVI)  (Rouse et al.,1974; Singh et al., 2003; Kogan, 2005; Sheffield and Wood, 2011; Brian et al., 2012)  Difference between near infrared and visible reflectance divided by sum of two wavebands   Advanced, very high-resolution radiometer (AVHRR)-based index used to monitor vegetation conditions and distributions  Detecting drought onset and measuring its intensity and duration   Measures general vegetative conditions in large area of coverage  Provides high spatial resolution of near real-time data for entire globe  Successfully used to identify stressed and damaged crops and pastures  Difficult to separate influences such as weather on vegetative health  Atmospheric conditions, especially cloud cover, considerably reduce index values and cause noise  Vegetation Condition Index (VCI)  (Unganai and Kogan, 1998; Heim 2002; Quiring and Ganesh, 2010; Mishra and Singh, 2010; Wardlow et   A pixel-wise normalization of NDVI to control local differences in ecosystem productivity  Suitable for monitoring of  agricultural droughts  A potentially global   Provides real-time data with high spatial resolution for monitoring drought  Captures rainfall dynamics more accurately than NDVI,   Limited utility during the cold season 16      al., 2012 )  standard of measuring times of drought onset, intensity, duration, and impact on vegetation  particularly in heterogeneous areas   Enables comparisons of impact of weather on areas with different environmental resources   Regional Drought    Regional Drought Area Index (RDAI)  (Bhalme and Mooley, 1980; Fleig et al., 2010, 2011; Sheffield and Wood, 2011)  Divides area affected by drought by the total area of region  Based on daily streamflow   Quantifies spatial extent of droughts  Requires spatially continuous or regional data  Drought Severity Index  (Dai et al., 2010; Sheffield and Wood, 2011)  Area-weighted intensity over the drought area  Quantification of average  severity of drought over a region  Same as above  17  2.6.1. Palmer drought severity index  The Palmer drought severity index (PDSI) was developed as a climatological tool to measure drought intensity, onset, and end date (Palmer, 1965; Alley, 1984). PDSI has been widely utilized in the U.S. by agencies such as the U.S. National Weather Service (NWS), the Climate Prediction Center (CPC), and the U.S. National Drought Monitor (Sheffield and Wood, 2011). This regional drought index uses precipitation and temperature for estimating moisture supply and demand within a two-layer, bucket-type soil model via the water balance equation (Alley, 1984; Dai et al., 2004; Mishra and Singh, 2010). PDSI represents the soil moisture departure within a specific region, as compared to the normal conditions, by using a water balance model (Sheffield and Wood, 2011). Dry and wet conditions are classified into 11 -0.49),  incipient drought (--0.99), mild drought (--1.99), moderate drought (--2.99), severe drought (---4.00) (Heddinghaus and Sabol, 1991). Several modified versions of PDSI have been developed, such as the Palmer Moisture Anomaly Index (Z-index), Palmer hydrological drought index (PHDI) (Palmer, 1965), and the Palmer modified drought index (PMDI) (Heddinghaus and Sabol, 1991). The Z-index is an intermediate term within PDSI calculating the monthly-standardized anomaly of available moisture (Palmer 1965, Zargar et al., 2011). This index is used to quantify agricultural drought impacts for short-term drought conditions (Zargar et al., 2011). The PHDI is used for water supply monitoring and for the qualification of the hydrological impacts of long-term drought conditions (Karl, 1986; Mishra and Singh, 2010; NCDC, 2013). And the PMDI was defined as a 18  real-time version of the PDSI for operational purposes (Heddinghaus and Sabol, 1991; Mishra and Singh, 2010).  2.6.1.1. Applications PDSI has proven valuable for use in many types of studies, including drought forecasting (Kim and Valdes, 2003; Ozger et al., 2009); exploration of the periodic behavior of droughts (Rao and Padmanabham, 1984); drought assessment over large geographic areas (Johnson and Kohne, 1993); the study of hydrologic trends and assessment of potential fire severity (Heddinghaus and Sahol, 1991); investigation of spatial and temporal drought characteristics (Lawson et al., 1971; Klugman, 1978; Karl and Koscielny, 1982; Diaz, 1983; Soule, 1992; Jones et al., 1996); and illustration of the areal extent and severity of various drought episodes (Palmer, 1967; Karl and Quayle, 1981).  2.6.1.2. Advantages  PDSI has proven to be is a reliable measure of the intensity and duration of long-term droughts, and has been utilized for many years (Mishra and Singh, 2010; NCDC, 2013). It uses precipitation and surface air temperature for its inputs, then outputs evaporation and run-off,  taking into account the basic effects of surface warming occurring in the 21st century (Dai et al., 2004; Sheffield and Wood, 2011). PDSI can also be used to evaluate wet situations (Alley, 1984), and is a standard measure of surface moisture conditions that facilitate direct comparisons of PDSI between different regions and timeframes (Alley, 1984; Dai et al., 2004).   2.6.1.3. Limitations  PDSI has several limitations, which have been detailed in multiple studies (Alley, 1984; Karl and Knight, 1985; Heddinghaus and Sahol, 1991; McKee et al., 1995). These limitations include the arbitrary selection of values for quantifying the intensity of drought and monitoring 19  the onset and end of a given drought or wet spell (Alley, 1984; Heddinghaus and Sahol, 1991); and that PDSI is better suited for evaluation of the agricultural impacts of drought than for determining the impacts of hydrologic droughts (Hayes et al., 1999). Additionally, the lag time between precipitation fall and runoff generated is not considered, which can result in values that are several months behind the actual values of emerging droughts (Hayes et al. 1999; Sheffield and Wood, 2011). PDSI also assumes no runoff occurrence until all soil layers have become saturated, which can lead to the underestimation of the runoff (Hayes et al., 1999; Mishra and Singh, 2010); and all precipitation is assumed to be rain, therefore snowfall, snow cover, and frozen ground are not considered, resulting in the potential inaccuracy of PDSI values determined for winter months and areas at high elevations (Hayes et al., 1999; Mishra and Singh, 2010; Sheffield and Wood, 2011). PDSI values also vary widely for extreme and severe drought classifications as well as frequencies in different locations (Hayes et al., 1999); and PDSI responds slowly to the conditions of a developing drought and also retains values reflecting a drought well after it has ended (Hayes et al., 1999; Mishra and Singh, 2010). Furthermore, PDSI is sensitive to both temperature and precipitation, often leading to a few mresponse to temperature and precipitation anomalies (Karl, 1986; Mishra and Singh, 2010). These rather significant limitations were the primary reason for the development of SPI, which was designed to resolve some of the most problematic issues inherent in PDSI.   2.6.2. Standardized precipitation index  McKee et al. (1993) developed the standardized precipitation index (SPI) as a probability tool to estimate the intensity and duration of drought events. SPI can be calculated for any location with a long-term monthly precipitation record of the desired time period. Computation of the SPI requires the fitting of a probability distribution to the historical precipitation records 20  for the timescale(s) of interest in order to define the relationship of the probability to the precipitation. The fitted probability distribution is then normalized to a standard normal distribution using the inverse normal (Gaussian) function. In a standard normal distribution, the mean and variance SPI for the location and desired time period are 0 and 1, respectively. Therefore, for any observed precipitation data, the SPI value is the deviation from the entire standard normal distribution (McKee et al., 1993; Edwards and McKee, 1997; Heim, 2002; Mishra and Singh, 2010).  Table 2 represents the classification scale for the SPI values. The index is negative for drought situations (less than median precipitation) and positive for wet conditions (greater than median precipitation). Negative values of SPI represent a higher probability of drought occurrence and more severe droughts (McKee et al., 1993; Hayes et al., 1999; NCDC, 2013). Table 2. Classification of SPI values (adapted from McKee et al., 1993; 1995) Class Index Value Extremely wet .0 Very wet  Moderately wet  Near normal - Moderate drought --1.0 Severe drought --1.5 Extreme drought SPI < -2.0  2.6.2.1. Applications SPI has proven valuable for widespread applications within drought studies, including forecasting (Mishra and Desai, 2005a; Cancelliere et al., 2007; Mishra et al., 2007), spatio-temporal analysis (Mishra and Desai, 2005b; Mishra and Singh, 2009), and climate impact studies (Mishra and Singh, 2009).  21  2.6.2.2. Advantages  as it is solely based on precipitation; making drought assessment possible without the use of additional hydrometeorological its temporal flexibility and versatility; it can be applied to a variety of timescales, from small timescale monitoring of water supplies (including soil moisture, which is important for agricultural production), to large timescale monitoring of water resources such as groundwater supplies, river flow, and lake water levels (Hayes et al., 1999; Livada and classification of severe and extreme drought frequencies for any given location and timescale, as a result of its normal distribution (Hayes et al., 1999).  2.6.2.3. Limitations  The length of the precipitation record plays an important role in calculating SPI values; discrepancies in SPI values can be obtained as a result of having different distributions due to varying lengths of precipitation records. Users should be aware of this inconsistency when interpreting and making decisions based on SPI values. Wu et al. (2005) conducted a study to evaluate the effect of precipitation record length on SPI calculation; they concluded that the different records of varying lengths with similar gamma distributions over different time periods result in consistent SPI values. However, discrepancies were observed for precipitation records of varying lengths with different gamma distributions (Wu et al., 2005; Mishra and Singh, 2010).  , as different SPI values are obtained when multiple types of probability distribution are used. This dependency causes bias in the SPI values when long timescales (longer than 24 months) are 22  involved; additionally, in the case of short timescales, the calculated SPI is not normally distributed for dry climates (Mishra and Singh, 2010). Some of the probability distributions used to simulate precipitation distribution when computing SPI values are gamma distributions (McKee et al., 1993; Edwards and McKee, 1997; Mishra and Singh, 2009); Pearson type III distributions (Guttman, 1998); and lognormal, extreme value, and exponential distributions (Todorovic and Woolhiser, 1976; Madsen et al., 1998; Lloyd-Hughes and Saunders, 2002; Wu et al., 2007). This also makes it so that SPI cannot really be compared spatially unless two pixels have the same maximum and same shape parameters.  2.6.3. Crop moisture index  The crop moisture index (CMI) was developed by Palmer (1968) and uses a meteorological approach for the monitoring of the short-term moisture supply (week-to-week) of various crop regions (Hayes, 2006; Mishra and Singh, 2010; Sheffield and Wood, 2011). CMI was derived from the Palmer drought severity index and requires weekly temperature and precipitation values in order to be computed (Hayes, 2006; Mishra and Singh, 2010).  2.6.3.1. Advantages CMI responds rapidly to changing conditions and is easily computed using precipitation and temperature data (Hayes, 2006; Mishra and Singh, 2010; Sheffield and Wood, 2011). And, because it is weighted by both location and time, CMI can also be used to compare moisture conditions at different locations (Hayes, 2006). 2.6.3.2. Limitations CMI is not effective for monitoring of long-term drought (Hayes, 2006; Mishra and Singh, 2010; Sheffield and Wood, 2011) since its rapid response to short-term changing conditions is misleading when monitoring long-term changing conditions (Hayes, 2006; Mishra 23  and Singh, 2010). In addition, Juhasz and Kornfield (1978) conducted a sensitivity analysis of CMI and reported that this index might erroneously indicate wetter conditions as temperature increases, as a result of the anomaly term formulation of evapotranspiration.  2.6.4. Palmer hydrological drought index  The Palmer hydrological drought index (PHDI) was introduced by Palmer (1965) for the purpose of assessing hydrological droughts. This index is a derivative of PDSI and uses the same water balance assessment on a two-layer soil model (Karl and Knight, 1985; Keyantash and Dracup, 2002). The PHDI can quantify the severity of either a drought or a wet spell, and is based on daily inflow (precipitation) and soil moisture storage (Karl and Knight, 1985; Sheffield and Wood, 2011). The principle that differentiates PHDI from PDSI is that the PHDI responds more slowly to the changes in weather leading to the termination of either a drought or a wet spell. In other words, PHDI rebounds gradually towards the normal state and only indicates the termination of a drought when the moisture deficit actually vanishes (Johnson and Kohne, 1993; Heim, 2000; Keyantash and Dracup, 2002). This characteristic of PHDI makes it suitable for the assessment of hydrological drought, which is a slower developing phenomenon than meteorological drought (Keyantash and Dracup, 2002).  The PHDI is used for water supply monitoring and qualifying the hydrological impacts of long-term drought conditions (Karl, 1986; Mishra and Singh, 2010; NCDC, 2013). Its limitations are the same as those of the PDSI; however, this index can also monitor long-term drought conditions (Sheffield and Wood, 2011).  2.6.5. Base-flow index The Institute of Hydrology (now the Centre for Ecology & Hydrology) introduced the base-flow index (BFI) as an indicator of catchment permeability (Zaidman et al., 2001; Tallaksen 24  and van Lanen, 2004). The BFI was first developed in the United Kingdom (UK) for use in determining the low flow characteristics of rivers (Tallaksen and van Lanen, 2004). The BFI is the ratio of the base flow to the total flow, which is calculated by applying smoothing and separation rules on a daily mean flow hydrograph (Gustard et al., 1992; Tallaksen and van Lanen, 2004; Sheffield and Wood, 2011). BFI values range from 0.1 for a flashy river with an impermeable catchment to nearly 1.0 for a very stable river with a permeable catchment (Gustard et al., 1992; Tallaksen and van Lanen, 2004).  2.6.5.1. Applications The BFI has many potential areas of application, including low-flow estimation and groundwater recharge assessment. This index is widely used in countries such as the UK, Canada (Pilon and Condie, 1986), Fiji (Green, 1986), Zimbabwe (Meigh, 1987), New Zealand (National Water and Soil Conservation Authority, 1984), Norway (Tallaksen, 1986), and Australia (Nathan and McMahon, 1990a).  2.6.5.2. Advantages The BFI is frequently used to estimate low-flow indices at ungauged sites and is closely related to other low-flow indices (Tallaksen and van Lanen, 2004). It can also quantify the flow from stored water within a basin (Gustard et al., 1992; Sheffield and Wood, 2011).  2.6.5.3. Limitations The BFI limitation is that long records are required in order to separate base flow from total flow (Sheffield and Wood, 2011). It is also sensitive to missing data; one day of missing data can cause data from the base-flow separation (Tallaksen and van Lanen, 2004).  25  2.6.6. Surface water supply index  The surface water supply index (SWSI) was developed by Shafer and Dezman (1982) as a hydrological drought index for monitoring abnormalities in surface water supplies (Hayes, 2006; Mishra and Singh, 2010). SWSI is calculated based on the monthly weighted sum of the non-exceedance probabilities of snowpack, streamflow, precipitation, and reservoir storage components. For winter data, only snowpack, precipitation, and reservoir storage is used to compute SWSI; for summer months, only streamflow, precipitation, and reservoir storage components are used in the calculation of SWSI (Garen, 1993; Mishra and Singh, 2010).  2.6.6.1. Advantages SWSI is used as a complement to the PDSI since it accounts for snow accumulation, subsequent runoff, and large topographic variation across a region (Hayes, 2006). SWSI is simple to calculate and represent water supply conditions (Hayes, 2006; Sheffield and Wood, 2011). Since the non-exceedance probabilities are used as the normalizing technique in calculating SWSI, it allows comparison of water supply availability among regions with different variability (Garen, 1993).  2.6.6.2. Limitations The weight of each hydrological component in the SWSI equation varies with the spatial scale (one basin to another) and the temporal scale (season or month). This variation is due to the differences in hydroclimatic variability which results in SWSIs with different statistical properties (Garen, 1993; Heim, 2002; Mishra and Singh, 2010); SWSI measurements are unique for each basin, making it difficult to compare SWSI values between multiple basins (Hayes, 2006; Sheffield and Wood, 2011).  26  2.6.7. Normalized difference vegetation index  The normalized difference vegetation index (NDVI) is an advanced, very high-resolution radiometer (AVHRR)-based index proposed by Rouse et al. (1974) to monitor vegetation conditions and distributions but cannot directly quantify drought. AVHRR is used by the National Oceanic and Atmospheric Administration (NOAA) series of Polar-orbiting Operational Environmental Satellites (Kogan, 2005; Brian et al., 2012). It is a five-channel passive scanning radiometer and its radiance is used to monitor drought conditions caused by sensitivity to cover visible, near-infrared, mid-infrared, and thermal infrared regions of the electromagnetic spectrum. NDVI is derived from channels 1 and 2 (visible and near infrared) based on the known radiometric properties of plants (Kogan, 2005; Quiring and Ganesh, 2010). NDVI values range between -1 to 1, where negative values indicate the presence of features such as clouds, water, and snow; near zero values indicate no vegetation; and values near 1 indicate the highest possible density of vegetation. This index is defined as NDVI= (NIR  VIS)/(NIR + VIS), where NIR stands for near-infrared and VIS stands for visible. Healthy vegetation generally reflects more near-infrared wavelengths than visible light wavelengths, while unhealthy vegetation shows little difference between visible and near-infrared reflected radiation (Singh et al., 2003; Boken et al., 2005; Quiring and Ganesh, 2010).  2.6.7.1. Applications NDVI has been proven to be a useful tool for: 1) mapping changes in vegetation cover and measuring drought impact in regions around the world (Anyamba et al., 2001; Gutman 1990; Ji and Peters, 2004; Singh et al., 2003); 2) providing accurate descriptions of continental land cover, vegetation classification and vegetation phenology (Tucker et al., 1987; Trapley et al., 27  1984; Justice et al., 1985); 3) monitoring rainfall and drought, crop growth conditions and crop yields (Kogan 1987, Dabrowska-Zielinska et al., 2002); and 4) detecting drought onset and measuring drought intensity and duration (Kogan 1995; Seiler et al., 2000; Quiring and Ganesh, 2010). 2.6.7.2. Advantages  NDVI measures general vegetative conditions over large areal regions, and has been successfully used to identify stressed and damaged crops and pastures. Also, because NDVI is an AVHRR-based index, it can provide high spatial resolution of near real-time data for the entire globe (Sheffield and Wood, 2011).  2.6.7.3. Limitations  Although NDVI has been used in a wide range of applications for drought monitoring, it does have limitations. First, it is difficult to separate out other influences, such as weather components, on vegetation health (Singh et al., 2003). Second, the presences of noise in AVHRR data restricts remote sensing of vegetation. Atmosphere components, especially clouds, considerably reduce NDVI values and cause noise (Guttman 1991; Singh et al., 2003). Third, in some semiarid environments, both soil characteristics and reflectance of lower plant communities such as mosses, lichens, algae, and cyanobacteria can lead to misinterpretation of drought conditions (Wardlow et al., 2012). Fourth, in tropical forests, the vegetation greenness within a pixel may saturate and make NDVI insensitive to increasing amounts of vegetation (Ripple, 1985; Ingram et al., 2005). Fifth, NDVI uses a limited amount of the total spectral information available within an image, which results in less information on vegetation coverage (Foody et al., 2001; Ingram et al., 2005). Sixth, for nonhomogeneous land cover, NDVI values are normally higher for more favorable environmental conditions such as forests, and lower for 28  less favorable environmental conditions such as dry steppes (Unganai and Kogan, 1998).  There is also a time lag between NDVI green-up and rainfall, so any detection of drought has to account for time differentials. 2.6.8. Vegetation condition index  The vegetation condition index (VCI) is also an AVHRR-based index developed by Kogan (1990). This index is a pixel-wise normalization of NDVI to control local differences in ecosystem productivity but cannot directly quantify drought. Pixel-based normalization minimizes the effect of short-term signals and also amplifies long-term ecological signals (Quiring and Ganesh, 2010; Wardlow et al., 2012). VCI, like NDVI, is computed from satellite AVHRR radiance (visible and near-infrared) (Mishra and Singh, 2010). It is defined as VCI = 100 (NDVI -NDVImin) / (NDVImax - NDVImin) and ranges from 0 to 100 for minimum and maximum NDVI, respectively. High values of VCI indicate healthy vegetation conditions (optimal) and low values of VCI indicate poor vegetation condition (unhealthy and unfavorable) (Singh et al., 2002; Quiring and Ganesh, 2010).  2.6.8.1. Applications VCI has been used for detecting and tracking drought in several regions around the world, including the U.S. (Kogan, 1995), China (Kogan and Sullivan, 1993), parts of the former Soviet Union (Kogan and Sullivan, 1993), Argentina (Seiler et al., 2000), Africa (Unganai and Kogan, 1998), and Kazakhstan (Gitelson et al., 1998). Besides successfully detecting and tracking drought, VCI can also be used to detect vegetation stress due to excessive wetness for both localized/short-term and widespread/long-term droughts (Kogan, 1995; Heim, 2002; Singh et al., 2003). Additionally, VCI is a potential global standard for measuring the time of drought onset, intensity, duration, and impact on vegetation. Studies such as Gitelson et al. (1998), Kogan 29  (1997), and Unganai and Kogan (1998) have indicated that VCI is suitable for monitoring agricultural droughts. However, according to Bayarjargal et al. (2006), Bhuiyan et al. (2006), Singh et al. (2003), and Vicente-Serrano (2006), VCI is not appropriate for monitoring meteorological droughts in some regions. 2.6.8.2. Advantages VCI, like NDVI, provides real-time data with high spatial resolution for monitoring drought (Heim, 2002; Quiring and Ganesh, 2010). Rainfall dynamics are best captured by VCI, particularly in heterogeneous areas, and VCI also quantifies the impact of weather on vegetation. In addition, it is possible to use VCI for comparing the impact of weather in areas with different environmental resources (i.e., ecological and climatic) (Unganai and Kogan, 1998).  2.6.8.3. Limitations VCI has limited utility during the cold season, since it is based on vegetation analysis; therefore, it is primarily useful for the summer growing season (Heim 2002; Mishra and Singh, 2010).  2.6.9. Recent developments in drought indices 2.6.9.1. Effective precipitation  Effective precipitation (EP) can more precisely monitor an ongoing drought and its duration, since it is a summed value of daily precipitation with a time-dependent reduction function (Byun and Wilhite, 1999; Mishra and Singh, 2010). The mean of EP (MEP), the deviation of EP from MEP (DEP), and the standardized value of DEP (SEP) are the three additional indices which complement EP (Mishra and Singh, 2010).   30  2.6.9.2. Reconnaissance drought index The reconnaissance drought index (RDI) was developed by Tsakiris and Vangelis (2005) to monitor the severity of meteorological droughts. In this index, the ratio between the precipitation and potential evapotranspiration records is first fitted into a lognormal probability distribution, and is then transformed into a standard normal distribution. Zarch et al. (2011) modified the RDI by using gamma distribution instead of lognormal distribution; it was concluded that, in most cases (i.e., different locations and timescales), RDI performs better with gamma distribution. 2.6.9.3. Flow duration curve The flow duration curve (FDC) was developed by Tallaksen and van Lanen (2004) for hydrological drought assessment using streamflow data. The streamflow data for a specific time period are ranked, their exceedance probabilities are calculated, and the calculated probabilities are then divided into five groups. These groups are comprised of: 0-10% (high flow), 10-40% (moist conditions), 40-60% (mid-range flow), 60-90% (dry conditions), and 90-100% (low flow) (Tallaksen and van Lanen, 2004; USEPA, 2011a).  2.6.9.4. Standardized runoff index  Shukla and Wood (2008) developed standardized runoff index (SRI) based on the SPI concept to characterize hydrologic drought. The SRI integrates hydrologic processes that define seasonal loss in streamflow caused by climate impact. They concluded that SRI is a suitable complement to SPI for illustrating hydrological drought on monthly-to-seasonal scales (Shukla and Wood, 2008; Mishra and Singh, 2010).  31  2.6.9.5. Water balance derived drought index Vasiliades et al. (2011) created the water balance derived drought index (WBI) to assess hydrological drought characteristics. In order to calculate the WBI, a water balance model is used to simulate streamflow data for a specific time period. The simulated streamflow data is then normalized using Box-Cox transformation, and the transformed streamflow data is converted into a standard normal distribution (Zargar et al., 2011; Vasiliades et al., 2011). The results show that the WBI is a good indicator of both hydrological drought severity and duration (Vasiliades et al., 2011).  2.6.9.6. Reclamation drought index  The reclamation drought index (RDI) was introduced by Weghorst (1996) to identify the onset and end of a drought period (Weghorst, 1996; Niemeyer, 2008). This index can also be used as a tool for defining drought severity and duration (Hayes, 2006; Niemeyer, 2008), and is similar to the SWSI in that it is calculated at the river basin level. The RDI can also incorporate air temperature-variable demand and duration (Hayes, 2006; Zargar et al., 2011), in addition to the SWSI input parameters (precipitation, streamflow, snowpack, and reservoir storage), thus making it capable of analyzing both climate and water supply factors. RDI values range from 4 to -4, with 4 representing extremely wet conditions, and -4 representing extreme drought. Since the RDI is unique for each river basin, it cannot be used for inter-basin comparisons (Hayes, 2006).    2.6.9.7. Indices based on soil moisture Narasimhan and Srinivasan (2005) developed the soil moisture deficit index (SMDI) and the evapotranspiration deficit index (ETDI) to improve the spatial and temporal resolutions used in agricultural drought monitoring. Weekly soil moisture and evapotranspiration were simulated 32  using the calibrated hydrologic model soil and water assessment tool (SWAT). The SMDI and ETDI both feature higher spatial (16km2) and temporal (weekly) resolutions than the SPI or PDSI indices, with resolutions of 7,000-100,000 km2, monthly (Narasimhan and Srinivasan, 2005; Mishra and Singh, 2010). Hunt et al. (2009) developed the soil moisture index (SMI), which is capable of determining drought onset as well as identifying soil discharge. The SMI is a continuous function that provides the relative position of the actual water content between the wilting point and the field capacity (Hunt et al., 2009; Mishra and Singh, 2010; Combe et al., 2014). Recently, Martínez-Fernández et al., 2015 developed the soil water deficit index, which is effective for the monitoring of agricultural drought. This index uses soil water observations such as moisture content, available water content, and field capacity to calculate soil water deficits.  2.6.9.8. Indices based on remote sensing  The normalized difference water index (NDWI) was introduced by Gao (1996) in order to determine the water content of vegetation and spongy mesophyll in vegetation canopies. This satellite-derived index is a combination of NIR and short-wave infrared (SWIR) bands (Delbart et al., 2005; Mishra and Singh, 2010); in order to obtain information on the water content of vegetation, indices based on the NIR and SWIR bands have proven more reliable than those using the NIR and VIS bands (e.g. NDVI). As a result of the VIS, the NDVI represents the chlorophyll rather than the water content (which is located in the strong chlorophyll absorption region) (Chen et al., 2005). NDWI may also be a more sensitive index for drought monitoring than the NDVI as a result of its dependence on both the desiccation and wilting of a given vegetation canopy (Mishra and Singh, 2010). 33   2.6.9.9.   Drought monitor  A weekly drought monitor (DM) tool was introduced by the NOAA, USDA, and NDMC in order to consolidate and centralize drought assessment. The DM is a synthesis of different drought indices and incorporates climatic data and professional input from all levels. This tool categorizes drought on a scale from zero-to-four (D0 to D4), where D0 represent abnormally dry areas and D4 represent exceptional drought events (i.e., a drought of record). Three labels are used to indicate the sectors affected by drought: A represents agricultural impacts, W hydrological impacts, and F a high risk of wildfire. The drought monitor map identifies the general drought area, labels individual droughts by their intensity, as well as drought impact on different sectors (Svoboda, 2000; Heim, 2002; Mishra and Singh, 2010). Although DM map simplicity makes it attractive for both public and different applied areas. It has the limitation of presenting several drought temporal scales on one map, which can be misleading (Heim 2002; Mishra and Singh, 2010).  2.7. Climate Change  According to the U.S. Environmental Protection Agency (EPA), climate change refers to any significant, lasting change in temperature, precipitation, or wind patterns over several decades (USEPA, 2014). Climate change is occurring and our planet is warming (USEPA, 2014; approximately 0.8°C (1.4°F). This increase in temperature has been mainly occurring since the mid-1970s, with the period between 1983 and 2012 being the warmest three decades on record in over 800 years (Royal Society, 2014projected to rise by another 2°F to 11.5°F (USEPA, 2014). A wide range of observations have been recorded confirming this warming trend, including an increase in ocean temperatures, a rise 34  in sea level, the decline of snow and ice cover in the Northern Hemisphere, and an increase in the concentrations of greenhouse gasses (Royal Society, 2014).  This rapid increase in global temperatures is anomalous when comparing observations with models, the long-term records, and fingerprint studies, suggesting that these recent changes are not solely due to natural causes  the sun, volcanic eruptions, or internal fluctuations in the climate system (Royal Society, 2014). Besides these types of natural variability, human activities are causing climate change by increasing greenhouse gas concentrations in the atmosphere (Sheffield and Wood, 2011). By 2012, human activities had increased atmospheric CO2 by approximately 40% (compared to levels in the 19th century), with most of the increase having occurred since 1970. Concentrations of other greenhouse gases such as methane and nitrous oxide are also increasing as a result of human activities (Royal Society, 2014), which have significantly disturbed the balance of the natural carbon cycle via the burning of fossil fuels, deforestation, and other drastic changes in land use.  In nature, CO2 is continuously exchanged between the atmosphere, plants, animals, and the oceans; natural processes such as photosynthesis, respiration, the decomposition of plants and animals, and the continuous exchange of gas between the atmosphere and oceans keeps the carbon cycle in balance. However, since the rate at which human activities release CO2 into the atmosphere is now substantially faster than the rate at which natural processes can restore the balance, large amounts of CO2 will remain in the atmosphere for thousands of years (Royal Society, 2014). The Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment reported with 95% certainty that human activity is the dominant cause of observed warming since the mid-20th century (IPCC, 2013). Many different climate models have been used to simulate the multitude of factors affecting climate change; however, it should be noted that the 35  recent warming trends were only replicated in cases where climate models included human influences on the atmosphere. Climate change also affects aquatic ecosystems, biological, and ecological processes. The recent increase in air temperature has caused a significant rise in water temperature and hydrological variability (Bates et al., 2008; Britton et al., 2010). As a result, a notable change has been observed in species composition and range, organism abundance, phenology, and biodiversity (Walther et al., 2002; Root et al., 2003; Bates et al., 2008; Hellmann et al., 2008).  2.7.1. Drought and Climate Change  Climate change can lead to significant changes in the frequency of extreme climate events such as droughts, floods, heat waves, and extreme rainfall (IPCC, 2013). Drought frequency, intensity, and duration have increased in many regions since the 1970s, especially in the tropics and subtropics, the Mediterranean, and West Africa; while the frequency and intensity of droughts in central North America and northwestern Australia have decreased since 1950 (IPCC, 2013). In the early 21th century, the likelihood of changes in the intensity of future droughts are considered to be very low; however, in the late 21th century, drought risk will likely increase in presently dry regions (IPCC, 2013). Moreover, while concluded that it is more likely that human influence has contributed to observed trends of drought (Bates et al., 2008), the update to the IPCC assessment does not support this theory. The any significant effect on drought changes; this determination was primarily based on modeling uncertainties and low agreement between scientific studies (IPCC, 2013). However, the Working Group I AR5 Summary for Policymakers concluded that it is likely that the number of areas affected by drought will increase over time.  36  Evapotranspiration and soil moisture also play important roles in drought development; and more regions are experiencing drought as a result of recent increases in temperature and decrease in land precipitation, which results in both evapotranspiration and the reduction of soil moisture (IPCC, 2013; Dai et al., 2004). 2.8. Bioassessment The 1972 amendments made to the Federal Water Pollution Control Act of 1948, collectively called the Clean Water Act (CWA), established regulations for the discharge of pollutants into bodies of water, with the objective of restoring and maintaining the ecological USEPA, 2014). Bintegrated, adaptive community of organisms with a species composition, diversity, and functional organization comparable to that of its natural habitat (Frey, 1977; Karr and Dudley, 1981; USEPA, 2011b). With the passage of the CWA, significant effort was put into the improvement of water quality resource systems by developing thresholds and criteria for discharging specific contaminants into bodies of water (Karr and Dudley, 1981; USEPA, 2011b). Although there have been major  improvements in the quality of water resources and significant reduction in point-source pollutant discharge over the past four decades,  valuable aquatic resources are still being lost (Jelks et al., 2008; USEPA, 2011b).  These losses are further proof that there is a pressing need for analytical tools such as biological assessment that can operate at the ecosystem scale and directly measure the effects of different stressors on the biological integrity of aquatic ecosystems. Biological assessment is an important tool of water quality management as it helps address challenges such as habitat loss, hydrological alteration, invasive species, climate change, storm water, and nutrient loads 37  (USEPA, 2011b). These assessments are then used to measure the overall ecological integrity of a given aquatic ecosystem via surveys and other direct measurements of the waterbobiota (USEPA 2011c; USEPA, 2011b). Resident biota are species that spend all or part of their lives in aquatic environments, and are effective monitors of stream conditions. Biological assessments depict the relationship between stressors and their impact on an aquatic ecosystem (i.e., biological responses), which can be used to predict the environmental outcomes of potential water quality management actions (USEPA, 2011b).  The primary aim of the CWA is the restoration of the ecological integrity of water resource systems, and it employs many different regulatory and non-regulatory approaches to achieve this goal. Biological assessment plays an important role in supporting these approaches; its many benefits are detailed below:  1) Water Quality Standards (WQS): This is the regulation that designates waterbody uses, sets criteria to protect these uses, and establishes anti-degradation policies to protect waterbodies determine if a given body of water sustains healthy aquatic life. They can also provide information on the species composition of a particular site, which can be used to adjust the chemical water quality to match the chemical sensitivity of the resident species (USEPA, 2000; USEPA, 2011b).  2) Development of Total Maximum Daily Loads (TMDLs): Under section 303(d) of the CWA, states are required to develop a list of all impaired and threatened waters requiring TMDL development. Biological assessment provides valuable ecological evaluations of the status of individual bodies of water, based on the severity of the incurred biological damage, waterbodies are prioritized for TMDLs (USEPA, 2000; USEPA, 2011b).  38  3) National Pollutant Discharge Elimination System (NPDES) Permits: Under section 402 of the CWA, discharging point-source pollutants into U.S. waters require an NPDES permit, which can be issued by the individual states or the EPA. NPDES permits ensure that all affected waterbodies achieve their WQS. Biological assessments can directly measure the combined impacts of stressors on resident biota, allowing states and tribes to use bioassessment results independently or in combination with chemical or whole effluent toxicity (WET) data to determine the effectiveness of permit controls (USEPA, 2000; USEPA, 2011b).  4) Nonpoint Source (NPS) Pollution: Unlike point-source pollution, NPS pollution is difficult to control because it comes from diverse sources such as land runoff, precipitation, atmospheric deposition, and hydromodification. Many states have reported that NPS pollution is the leading cause of water quality impairment. Biological assessments are the most effective and sensitive indicators of the cumulative effects of multiple chemical and non-chemical unpredictable stressors caused by NPS pollution. Biological impairment caused by NPS pollution can also be determined using biological assessment, and restoration efforts such as voluntary best management practices (BMPs) can then be used to improve degraded water (USEPA, 2000; USEPA, 2011b).   There are various well-developed biological assessment programs whose benefits are based on their ability to characterize the biological conditions of a waterbody relative to U.S. WQS; integrate the cumulative effects of multiple stressors from a variety of sources; detect aquatic life deterioration caused by unmeasured stressors and unknown sources of impairment; provide field data on biotic response variables to support the development of empirical stressor 39  response models; and inform water quality and natural resource managers, stakeholders, and the public of the projected environmental outcomes of any potential future actions (USEPA, 2011b).  New tools have recently been established to improve the use of biological assessments in water quality management and further help states to develop more robust biological assessment programs. Three of these new tools are listed below and explained in greater depth in the following sections: 1) The Biological Assessment Program Review: This tool evaluates the biological assessment program rigor that indicates how well the information obtained from the assessment program can support management decision making. A template is provided that helps states evaluate and upgrade the technical capabilities of their biological assessment programs by determining the best places to invest resources (USEPA, 2011b). 2) The Biological Condition Gradient (BCG): A conceptual model that describes the change of 10 ecological attributes in response to a gradient of increasing anthropogenic stress. The gradient is divided into six levels, with level 1 representing no/low level of stress and level 6 representing a high level of stress. The 10 ecological attributes are taxonomic composition and tolerance (attributes I-V), nonnative taxa (attribute VI), organism condition (attribute VII), ecosystem function (attribute VIII), the spatial and temporal extent of detrimental effects (attribute IX), and ecosystem connectivity (attribute X) (Davies and Jackson, 2006). This model provides a way of mapping different biological indicators on a common scale of biological conditions to facilitate comparisons between programs and the development of a universal set of biological criteria. The ability to calibrate this model to the unique characteristics of individual geographical regions helps states and tribes to more precisely study their biological community in terms of aquatic life uses, CWA objectives, and potential 40  management actions. The BCG provides a framework (Table 3) that synthesizes existing field observations with expert knowledge into testable hypotheses (Davies and Jackson, 2006; USEPA, 2011b). This framework helps water quality managers define their desired environmental conditions, monitor and assess existing environmental conditions, select management measures to reach their desired conditions, measure the effectiveness of their restoration projects, and better communicate with stakeholders (USEPA, 2011b).  41  Table 3. Biological response to increasing levels of stress (adapted from USEPA, 2011b; Davies and Jackson, 2006)    Level of Exposure to Stressors Biological Condition Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 Natural structural, functional, and taxonomic integrity is preserved Structure & function similar to natural community with some additional taxa & biomass; ecosystem level functions fully maintained Evident changes in structure due to loss of some rare native taxa; shifts in relative abundance; ecosystem level functions fully maintained Moderate changes in structure due to replacement of some sensitive ubiquitous taxa by more tolerant taxa; ecosystem functions largely maintained Sensitive taxa markedly diminished; conspicuously unbalanced distribution of major taxonomic groups; ecosystem function shows reduced complexity & redundancy  Extreme changes in structure and ecosystem function; wholesale changes in taxonomic composition; extreme alterations from normal densities       Watershed,  habitat,          flow regime and water            chemistry as naturally occurs                                                                                                                                                                                                                                                                                                                                                                       Chemistry,   habitat,                                                       flow regime severely altered                                                                                                                                                                                                                                                                                             from natural  conditions   3) Stressor Identification (SI) and Causal Analysis/Diagnosis Decision Information System (CADDIS): (EPA) Offices of Water and Research and Development to identify unknown stressors within impaired waters. The SI is an iterative process that can be applied at any level of biological organization for any type of water body. This process is prompted by the information provided by a given biological assessments indicating biological impairment, and helps identify the stressors causing the reported impairment. The core components of the SI process are: listing of the potential causes of impairment, analyzing evidence using new and existing data, and 42  characterizing causes in order to accurately determine the most likely stressor(s) causing impairment (USEPA, 2000; USEPA, 2011b). Table 4 represents an overview of the stressor identification process within the context of water quality management. Decision-maker and stakeholder involvement plays an important role in defining the scope of each investigation as well as in the listing of potential causes; additional data can be added to the stressor identification step at any point in the process (USEPA, 2011b). CADDIS is the online application of SI, and provides scientists and engineers with useful guidelines for use in the evaluation of potential causes of aquatic system impairment (USEPA, 2011b; USEPA, 2016).   Table 4. Stressor identification process (adapted from USEPA, 2000; USEPA, 2011b) 1. Detect Biological Impairment 2. Stressor identification a. List candidate causes b. Analyze evidence c. Characterize causes 3. Identify and apportion sources 4. Management action: Eliminate or control causes; monitor results 5. Biological condition restored or protected    Decision-maker and stakeholder involvement Acquire data and iterate process as necessary 43  2.8.1. Stream Health  In order to conduct bioassessment analysis, first stream health should be defined. In general, a healthy stream is a flourishing, sustainable ecosystem that is resilient to stress and maintains its societal values over time (Meyer, 1997). Biological monitoring is an essential tool for the assessment of the health of biological communities living within a given stream system (Loeb and Spacie, 1994; USEPA, 2012a); many biological monitoring methods exist to measure the ecological conditions of stream systems. Of these, biological indicators have proven their value as a tool capable of detecting low levels of nonpoint-source pollutants, changes in physical habitats, and the effects of long-term disturbance events on aquatic ecosystems (Barbour et al., 1999; Nerbonne and Vondracek, 2001; Flinders et al., 2008). Fish and macroinvertebrate communities are commonly used as indicators in water-quality assessments (Barbour et al., 1999; Flinders et al., 2008; Carlisle et al., 2013); the best application is a combined measurement of both fish and macroinvertebrate communities, as the two groups offer complementary information regarding water quality, resulting in a complete assessment of overall stream health (Flinders et al., 2008; Carlisle et al., 2013).   2.8.1.1. Fish as Indicators Fish play many important roles within aquatic ecosystems, and thus are invaluable indicators of stream health (Karr, 1981; Carlisle et al., 2013). Their primary roles are as a source of food for other aquatic and terrestrial species, as well as the main consumers of macroinvertebrates and algae (Carlisle et al., 2013). Several advantages of using fish as indicators are: 1) the relative ease of collecting specimens and identifying them to species level; 2) they are reflective of integrated environmental health (described below); 3) they are good long-term indicators of water quality across river networks due to their mobility and long 44  lifespan; 4) they are located at the top of the aquatic food chain; and 5) they have known life history, distribution, and environmental requirements (Karr, 1981; Barbour et al., 1999; USEPA, 2012a ; Carlisle et al., 2013). In addition, fish assemblages cover a variety of trophic levels, including omnivores, herbivores, insectivores, planktivores, and piscivores, which provides an integrative view of stream environmental health (Karr, 1981; EPA, 2012a).  Lastly, the effect of toxicity and stress can be evaluated through missing taxa, as well as growth and reproductive depression (Karr, 1981). 2.8.1.1.1. Index of biotic integrity  The index of biotic integrity (IBI) was introduced by James Karr (1981) as a bioassessment tool for the evaluation of the biological integrity in streams (Karr, 1981; Karr, 1991). The IBI covers a range of ecological levels, from individuals within a population to entire ecosystems, in order to evaluate human effects on stream health (Karr, 1991). There are twelve IBI metrics for the Midwestern U.S., which have been divided into three groups: species richness and composition, trophic composition, and fish abundance and condition. The species richness and composition group, comprising six different metrics, evaluates the total number of native fish species; as well as the number of benthic, water-column, long-lived, intolerant, and tolerant species. The trophic composition group, using three metrics, assesses the percentage of individuals categorized as omnivores, insectivores, and piscivores in order to evaluate the trophic composition of the entire fish community. The fish abundance and condition group, also with three metrics, evaluates population density and fish condition by measuring the number of individuals in each sample, the percentage of hybrids, and the percentage of fish with disease or anomalies (Karr, 1991). Each metric is rated from 1-to-5, and the sum of these ratings provides their IBI value.  A score of 5 indicates the study site has a slight deviation from the undisturbed 45  condition, 3 indicates a moderate deviation, and 1 indicates a strong deviation from the undisturbed condition.  The IBI values obtained by adding together the 12 metric scores are categorized into six groups, ranging from excellent to no fish; these groups represent the integrity class of the site (Karr, 1981; Karr, 1991). Since IBI is a region-specific index, modification of the selected metrics is required before it can be used on any geographical region outside of the Midwestern U.S.  The substantial differences in biological communities and fish distribution between regions necessitate furtheMany stream health studies have modified these metrics in order to study an individual site within a specific region (Mebane et al., 2003; Meador et al., 2008; Zhu & Chang, 2008; Angradi et al., 2009; Navarro-Llacer, Baeza, & de las Heras, 2010; Launois et al., 2011; Pelletier et al., 2012; Musil et al., 2012). Musil et al. (2012) successfully modified ten of the IBI metrics to assess stream health in Europe; the resulting index was named the European fish index. Wan et al. (2010) modified nine IBI metrics to evaluate conditions within Minnesota streams; they named their new index the Minnesota fish index of biotic integrity.  2.8.1.2. Macroinvertebrates as indicators Macroinvertebrates are organisms without backbones that are large enough to be seen with the naked eye.  They inhabit all types of water and include insects in their larval form, crayfish, clams, snails, and worms (Carlisle et al., 2013; USEPA, 2012a). Macroinvertebrate assemblages are commonly studied as a means of assessing water quality, and are good indicators of localized conditions due to their limited migration and immobile lifestyle. Since they have sensitive life stages, complex lifecycles, and dissimilarity in their pollution tolerance, they respond quickly to stressors and are affected by even short-term environmental variations 46  (Barbour et al., 1999; Carlisle et al., 2013; USEPA, 2012a). Their wide range of pollution tolerance and trophic levels allows for the determination of the cumulative impacts of pollution. In addition, macroinvertebrates can inhabit freshwater systems for a year or more and therefore can monitor environmental conditions over relatively extended periods of time (Barbour et al., 1999; Carlisle et al., 2013; USEPA, 2012a). Further, the sampling of macroinvertebrates is relatively simple, similar to the process used with fish, and it is easy to identify family levels (Barbour et al., 1999; USEPA, 2012a). Many macroinvertebrate indices and metrics have been developed to evaluate stream health and integrity including the Benthic index of biotic integrity, Hilsenhoff biotic index, and Ephemeroptera, Plecoptera, and Trichoptera Index which are the most commonly used in the assessment of the biological condition of streams. 2.8.1.2.1. Benthic index of biotic integrity The Benthic index of biotic integrity (B-IBI) is a multi-metric index originally developed by Karans and Karr (1994) for the study of streams in the Tennessee Valley. The B-IBI was modeled after the fish IBI and focuses on taxa richness, composition, and biological processes.  It uses thirteen original metrics which are both relatively uncorrelated and reactive to human disturbances of the environment (Karans and Karr, 1994; Fore et al., 1996). These metrics include total taxa richness; taxa richness of intolerant snails, mussels, mayflies, caddisflies, and stoneflies; relative abundance of corbicula, oligochaetes, omnivores, filterers, grazers, and predators; dominance; and total abundance (Karans and Karr, 1994). Each metric is compared to the undisturbed site and then given a score of 1 (severe impact), 3 (moderate impact), or 5 (little to no human impact). The combined metric scores determine the final B-IBI value (Karans and Karr, 1994). Some modifications have been made to the B-IBI metrics in order to evaluate stream health in different regions using various sampling methods (Fore and Karr, 1996; 47  Lammert and Allen, 1999): Lammert and Allen (1999) modified B-IBI to account for taxonomic differences in southeastern Michigan by only using nine of the 13 metrics and adjusting the scoring criteria to work for the Raisin River in southeast Michigan.  2.8.1.2.2. Hilsenhoff biotic index The Hilsenhoff biotic index (HBI), also called the biotic index, was introduced by Hilsenhoff (1987) to evaluate organic and nutrient pollution within streams. This index is based on the tolerance values of the organic pollutants assigned to each taxon (Hilsenhoff, 1987; Barbour et al., 1999). This pollution-tolerance index is used to summarize tolerance information from macroinvertebrate communities and can target multiple types of stressors (Lenat, 1993).  The HBI is calculated by multiplying the tolerance value of each taxon by the abundance of that taxon; it is summed across the taxa and then divided by the number of individuals in the sample (Lenat, 1993; Fore et al., 1996). Both the tolerance values and HBI range from 0 to 10. For the tolerance values, a value of 10 is assigned to taxa known to occur in severely polluted streams, and a value of 0 is assigned to the taxa collected in undisturbed streams with very high water quality. An HBI rating of 0 indicates excellent water quality with no pollution, while a rating of 10 indicates very poor water quality and severe pollution (Hardy et al., 2004).  Different modifications have been made to the HBI in order to take into account the effects of different types of pollutants, ecoregions, stream sizes, and seasons (Lenat, 1993).  2.8.1.2.3. Ephemeroptera, Plecoptera, and Trichoptera Index The Ephemeroptera, Plecoptera, and Trichoptera index (EPT) was introduced by Lenat (1988) as a method of relating taxa richness to its appropriate water quality classification. The EPT index is named after the three most common, most intolerant insect groups in the benthic macroinvertebrate community. These organisms are very sensitive to environmental 48  perturbations and pollutants, making them valuable indicators of water quality; both EPT taxa richness and percent abundance metrics are used in classifying water quality (Lenat, 1988). EPT taxa richness is the total number of Ephemerortera (mayflies), Plecoptera (stoneflies), and Trichoptera (caddisflies) present (Lenat, 1988; Goetz and Fiske, 2013); while the EPT percent abundance (%) is calculated by dividing the taxa richness by the total number of taxa (Cuceiro et al., 2012). The five water quality classifications are excellent, good, good-fair, fair, and poor (Lenat, 1988). The value of the EPT index decreases with decreasing water quality (Barbour et al., 1996; Compin and Cereghino, 2003).   2.8.2. Effects of climate change on bioassessment programs Many of the activities used in bioassessment programs are climatically sensitive, including assessment design, implementation, and environmental management. Bioassessment program designs rely on multimetric indices (MMIs) and predictive models created using ecological traits to detect impairment. Since ecological traits are sensitive to temperature changes, the MMIs and predictive models should also be affected by climate change. The selection of reference sites and determination of reference conditions in individual bioassessment programs are also influenced by climate change, as it impacts reference conditions and can even ult of these factors, the successful implementation of bioassessment programs requires long-term stations capable of detecting changes in biotic conditions due to climate-related trends (USEPA, 2012b).   2.9. Drought Risk Assessments  Among natural disasters, drought (emergence) is the most difficult to detect due to its unpredictable timing, variable duration, cumulative severity and extent, and non-structural impacts (Wu and Wilhite, 2004; Whitmore, 2000). Drought causes a complex set of direct and 49  indirect impacts; direct impacts of drought include reduction in the productivity of both cropland and forests, increasing danger of fire, diminishing water levels, and loss of livestock (SOEST, 2003; Paul Venton, 2012). Indirect impacts of drought are characterized as the consequences of these direct impacts (SOEST, 2003). In addition, drought effects can be further categorized into economic, environmental, and social impacts. Economic impacts include declines in crop yields, food insecurity, income lost by farmers, and the forced sale of household assets and land. Environmental impacts are comprised of reservoir depletion; livestock losses; soil erosion; loss of biodiversity; and the reduction of air, water, and landscape quality. Social impacts of drought include declines in public safety, health, dislocation, water use conflicts, and quality of life issues (SOEST, 2003; Paul Venton, 2012). Based on these drought characteristics and impacts, applying established risk assessment principles to drought is a logical step.  According to the UN- loss of lives, reduced health status, livelihoods, assets and ecosystem services in connection with drought, which could occur in a particular community or a society over a specific time period in the -ISDR, 2009). Drought risk is a function of the frequency of occurrence, severity of drought, and its vulnerability (Knutson, et al., 1998; SOEST, 2003). Drought risk analysis consists of drought risk assessment and drought risk management. Drought risk assessment identifies and quantifies drought risk and its vulnerabilities (Hayes et al., 2004; Paul Venton, 2012), and drought risk management identifies the best management strategies for minimizing the adverse effects of droughts (Hayes et al., 2004).   50  The National Drought Mitigation Center (NDMC) and the Western Drought Coordination Council (WDCC) developed user-friendly guidelines to help individual communities conduct their own drought risk analyses (Knutson, et al., 1998; Hayes et al., 2004). In addition to the aforementioned guidelines, Wilhite (1991) developed a 10-step drought planning methodology which includes risk assessment guidelines for drought planners. This methodology focuses on the key elements of the drought planning process, such as drought preparedness and management (Wilhite et al., 2000a). In the U.S., the states of New Mexico, Texas, Hawaii, Georgia, Nebraska, and Colorado; and the tribal governments of the Navajo, Zuni, and Hopi Nations have already performed their own drought risk analyses and developed individual drought mitigation plans (Hayes et al., 2004).  2.10. Drought Modeling  Understanding drought modeling and its components is crucial for water resource planning and management. There have been significant improvements made in drought modeling over the past three decades. Six different components are used in modeling drought: forecasting, probabilistic characterization, spatio-temporal analysis, impact of climate change, land data assimilation systems, and drought management (Mishra and Singh, 2011). Reviews of the different drought modeling components are provided in the following subsections. 2.10.1. Drought forecasting Drought forecasting is one of the main components of drought hydrology, and plays an important role in drought management and mitigation. A major challenge for researchers is the development of methods capable of accurately predicting drought onset and end points for periods months and years in advance (Mishra and Singh, 2011). Different methods of forecasting drought, including their applications, advantages, and limitations, are discussed below. 51  1) Regression models estimate the relationship between a dependent variable with one or more independent variables. The value of the dependent variable is predicted by using independent variables in the regression analysis. The value of the dependent variable is represented by a drought quantifying parameter such as a drought index. Independent variables with available data include explanatory variables such as precipitation, temperature, and soil moisture (Mishra and Singh, 2011). A regression model was developed by Kumar and Panu (1997) to predict agricultural drought using the grain yield of a main crop as the dependent variable; this model is capable of forecasting the severity of an agricultural drought several months in advance.  The onset of a drought in northeastern Brazil was predicted using multiple linear regressions (Liu and Negron-Juarez, 2001), using NDVI for the dependent variable and multiple ENSO indices for the independent variables (Liu and Negron-Juarez, 2001; Mishra and Singh, 2011). Despite their frequent use, regression models do have limitations, such as its assumption of a linear relationship between all dependent and independent variables. This assumption makes these types of models less viable for use in long-term forecasting. Another Singh, 2011).    2) Time series models effectively consider the sequential linear correlation between observations. A time series for a specific drought quantifying parameter is modeled based on previous observations for forecasting drought (Mishra and Singh, 2011). The autoregressive integrated moving average (ARIMA) and seasonal autoregressive integrated moving average (SARIMA) are the two most commonly used time series models for this type of application (Box et al., 1994; Mishra and Singh, 2011). The primary advantages of these two models are their forecasting capability; and their ability to perform systematic searches for the 52  identification, estimation, and performance of diagnostic checks during model development (Mishra and Desai, 2005a; Mishra and Singh, 2011). ARIMA and SARIMA were used to quantifying parameters. The predicted results using the best model had a strong agreement with the actual recorded data (Mishra and Desai, 2005a). Durdu (2010) also used the ARIMA modeling approach to forecast drought in the Büyük Menderes River Basin in western Turkey, also using an SPI series as the drought quantifying parameters. Although these models are both powerful and flexible, they do have limitations: ARIMA can only model always assumed to be constant during the series period, which is not always an accurate assumption (Yanovitzky and VanLear, 2007).  3) Probability models such as Markov chain are often used for drought forecasting and the quantification of the uncertainties associated with drought-causing hydro-meteorological variables. The Markov chain is a stochastic process in which each subsequent value of the process is solely dependent on the current value, and not on the preceding sequence of values (Mishra and Singh, 2011). Both homogenous and non-homogeneous Markov chain models were used to derive a conditional scheme for the prediction of short-term drought classes at several sites in Alentejo, Portugal (Paulo et al., 2005). Also, by using first-order Markov chains, Ochola and Kerkides (2003) were able to correctly predict the number and duration of dry spells in Kenya over a given period. 4) Artificial neural network models are flexible nonlinear models with high predictive accuracy that can estimate any complex nonlinear relationship with the appropriate number of nonlinear processing units. Artificial neural network (ANN) models generally consist of three 53  layers: the input layer, the hidden layer, and the output layer (used for forecasting purposes) (Mishra and Desai, 2006; Mishra and Singh, 2011). The quantifying drought parameters are introduced in the input layer; the hidden layer processes the input information using the appropriate nonlinear transfer functions, and the output layer forecasts the future values of different lead times (Mishra and Desai, 2006; Morid et al., 2007; Mishra and Singh, 2011). The advantages of using ANN techniques are: 1) they remain robust in the presence of noisy or missing inputs, even within small subsets of data; 2) their capacity to adapt quickly to changing environemnts; 3) their ability to determine relationships between different input samples (Dawson and Wilby, 1998); and 4) definition of the intermediate relationship between inputs and outputs is not required (Morid et al., 2007). Morid et al. (2007) also conducted a drought forecasting study in the Tehran Province of Iran using ANN models to predict qualitative values of drought indices. The effective drought index (EDI) and the SPI were the predictands, and various combinations of past rainfall records and climate indices (including the SOI and NAO) were the predictors. A comparison between linear stochastic models with recursive multistep neural networks (RMSNN) and direct multistep neural networks (DMSNN) was conducted by Mishra and Deasi (2006) for drought forecasting purposes. Their results indicated that RMSNN and DMSNN are useful for short-term and long-term drought forecasting, respectively. The limitations of ANN techniques include their black box nature, their empirical nature of model development, higher computational burden, and their proneness to over fitting (Mishra and Singh, 2011).  5) Hybrid models are useful models for predicting drought since they have the potential to extract the benefits of individual models, enabling them to forecast drought with better accuracy and higher lead times (Mishra and Singh, 2011). Kim and Valdes (2003) proposed a 54  . Their hybrid model was an integration of wavelet transforms and neural networks. The PDSI was the drought index used, and the results indicated that utilization of the hybrid model improved the neural applicability of the adaptive neuro-fuzzy interference system (ANFIS) for drought forecasting using SPI in Anatolia, Turkey. The ANFIS method is an integration of ANN and fuzzy logic (FL). Their results indicated that ANFIS provides high accuracy and reliability, and generally performs better than ANN (Bacanli, 2009).  2.10.2. Probabilistic characterization of drought Droughts have a probabilistic characterization (Sen, 1980a; Mishra et al., 2009a; Mishra and Singh, 2011), which plays an important role in the efficient planning and management of water resources, especially in arid and semi-arid regions (Serinaldi et al., 2009; Mishra and Singh, 2011). Severity, duration, intensity, frequency, and area are the essential parameters for characterizing drought; various probabilistic analyses can be used to characterize drought parameters. These probabilistic analyses include univariate drought analysis, bivariate drought analysis, and multivariate drought analysis using copula (Mishra and Singh, 2011).   1) Univariate drought analysis is a traditional approach for characterizing droughts which uses probability distribution functions to fit the sample frequency distribution (Cancelliere and Salas, 2004; Serinaldi et al., 2009; Mishra and Singh, 2011). In this approach, each of the drought parameters are considered to be independent and are investigated separately, therefore, the correlation between the different variables is not defined (Tallaksen et al., 1997; Fernandez and Salas, 1999a,b; and Cancelliere and Salas, 2004; Chen et al., 2011). In recent years it has been recognized that drought parameters are generally both dependent and 55  stochastically associated (Serinaldi et al., 2009; Michele et al., 2013); hence, the better approach is to derive the joint distribution of drought parameters (Mishra and Singh, 2011). Cancelliere and Salas (2004) derived a probability mass function (pmf) of drought length and its first-order moments. A periodic Markov chain was assumed to estimate the drought occurrence probability within a given length of time. Cebrian and Abaurrea (2006) developed a stochastic model consisting of a Poisson cluster process and a marked process for describing drought severity. The Poisson cluster represented drought occurrence and the marked process represented the series of duration, deficit, and maximum intensity (Cebrian and Abaurrea, 2006). 2) Bivariate drought analysis deals with two drought variables, most commonly, the duration and severity of drought. This approach characterizes drought by deriving a joint distribution of the drought variables (Chen et al., 2011; Mishra and Singh, 2011). Although bivariate distributions are commonly applied in drought analysis, they also have various drawbacks. First, bivariate distributions involve complex mathematical deviations or parameters fitted from either generated or observed data (Shiau, 2006; Chen et al., 2011). Second, bivariate models cannot be applied to marginal distributions within different families. For instance, these models cannot be used to correlate hydrological variables with marginal gamma and Gumbel distributions (Frees and Valdez, 1998; Shiau, 2006, Mishra and Singh, 2011). Several studies have been conducted to evaluate drought bivariate characteristics (Shiau and Shen, 2001; Gonzalez and Valdes, 2003; Kim et al., 2003b; Salas et al., 2005; Kim et al., 2006; Mishra et al., 2009; Cancelliere and Salas, 2010). Shiau and Shen (2001) formulated a joint distribution of drought duration and severity to investigate drought characteristics and the frequency and risk of occurrence for hydrologic droughts. Gonzalez and Valdes (2003) 56  investigated the frequency and risk of the occurrence of droughts in terms of their duration and severity using PDSI (Gonzalez and Valdes, 2003).  3) Multivariate drought analysis using copula. Droughts are multivariate events with correlated random variables. It is difficult to develop joint multivariate drought models due to the significant mathematical treatments, data requirements, and limited availability of models (Shiau and Modarres, 2009; Mishra and Singh, 2011); multivariate drought analysis using copula overcomes these limitations and also provides uncertainty reduction in its estimates of the frequency distribution parameters (Shiau, 2006; Song and Singh, 2010; Chowdhary and Singh, 2010; Mishra and Singh, 2011). Copulas are functions that create multivariate distribution functions by joining together univariate distribution functions. Multivariate distribution construction using copulas models the dependent structure of random variables independently from their marginal distributions (Shiau, 2006; Mishra and Singh, 2011). Several recent studies have used copulas to analyze the multivariate probability of drought. Shiau (2006) derived joint drought duration (exponential distribution) and severity (gamma distribution) by utilizing two-dimensional copulas and defining the drought characteristics via SPI. Song and Singh (2010) used a trivariate Plackett copula to construct the joint distribution function of drought duration, severity, and inter-arrival time based on streamflow data from the Wei River Basin.  2.10.3. Spatio-temporal drought analysis  Spatio-temporal drought analysis (i.e., regional drought analysis) also plays an important role in the short- and long-term management of water resources. The spatial coverage of drought behavior (Panu and Sharma, 2002; Mishra and Singh, 2011). In regional drought analysis, the 57  spatial distributions of climatic and hydrologic variables are investigated at different thresholds in order to classify the severity of a drought in a given region (Shin and Salas, 2000; Mishra and Singh, 2011). Climatic and hydrologic variables such as precipitation, soil moisture, streamflow, and the moisture content of the air have been used in several regional drought analysis studies (Shin and Salas, 2000). Alegria and Watkin (2007) performed a regional drought analysis by conducting a meteorological drought intensity-duration-frequency analysis using both the annual and warm season precipitation records of Sonora, Mexico. Vicente-Serrano (2006) analyzed differences in drought spatial patterns on the Iberian Peninsula using a wide range of precipitation characteristics; this analysis was conducted over different timescales using SPI.  2.10.4. Drought modeling under climate change scenarios d-1970s (Royal Society, 2014); and the duration, intensity, and areal extent of these changes in climate differ at the regional and even the local level (Mishra and Singh, 2011). General circulation models (GCMs) are powerful tools for assessing climate change impacts on drought. These global models simulate atmospheric processes and their interactions with the land and oceans over different timescales; and also account for the greenhouse gas concentrations in the atmosphere (Gosh and Mujumdar, 2007; Sheffield and Wood, 2011). The GCMs are linked to multiple projections of CO2 emission rates, thus producing different climate change scenarios. These scenarios indicate future greenhouse gas emission rates and also analyze the impact and vulnerability of potential climate changes (Gosh and Mujumdar, 2007). GCMs can successfully model large-scale processes and smoothly varying fields such as surface pressure; however, their coarse spatial resolution prevents them from modeling fine-scale (150-200km) processes and non-smooth fields such as precipitation (Gosh and Mujumdar, 2007; Sheffield and Wood, 2011). 58  Therefore, in order to model hydrologic variables, it is necessary to downscale the GCM outputs. After downscaling, large-scale GCM outputs can be used to model hydrological and drought variables at smaller scales (e.g. local scale) (Gosh and Mujumdar, 2007; Mishra and Singh, 2011).  Of the various downscaling techniques, dynamic downscaling and statistical downscaling and Singh, 2009b). In dynamic downscaling, regional climate models (RCMs) are derived from GCM outputs using one-way nesting approaches at fine-grid scales (Jones et al., 1995; Gosh and Mujumdar, 2007). RCMs (i.e., limited-area models) have higher spatial resolutions (50km) and are applied at the regional scale (Gosh and Mujumdar, 2007; Sheffield and Wood, 2011). On the other hand, statistical downscaling produces future scenarios by statistically relating meso-scale climate features (most commonly atmospheric circulation) to regional scale hydrological variables (Wilby et al., 1998; Wetterhall et al., 2003; Gosh and Mujumdar, 2007). Statistical (1) it adjusts quickly to new areas, requires few parameters; (2) has fewer computational demands; and (3) has lower implementation costs (Wilby et al., 2000; Wetterhall et al., 2005; Gosh and Mujumdar, 2007; Khan and Coulibaly, 2010). Blenkinsop and Fowler (2007) used six RCMs to develop a drought index based on monthly precipitation anomalies; the precipitation anomalies derived from the GCMs were dynamically downscaled and used to investigate future drought scenarios for six catchments across Europe. Mishra and Singh (2009b) constructed severity-area-frequency (SAF) curves using six GCMs in two different scenarios to investigate the impact of climate change on method was applied. The results indicated that, compared to the historical records, there will 59  likely be more severe droughts of greater spatial extent between 2001 and 2050 (Mishra and Singh, 2009b).   2.10.5. Land data assimilation systems Obtaining hydrological data remains a challenge due to the scarcity and inadequacy of hydrometric stations around the world (Sheffield and Wood, 2011; Mishra and Singh, 2011). Mishra and Coulibaly (2009) reported that, in recent decades, there has even been a reduction in the number of hydrometric stations as a result of war, natural disasters, lack of funding, and insufficient institutional agendas. This continual decrease in the number of hydrometric stations will continue to affect drought monitoring; however, remote sensing is an important tool for combating these issues, as it provides high spatial and temporal resolution data on both the regional and global scale (Sheffield and Wood, 2011; Mishra and Singh, 2011). Remote sensing is vital to the decision-making process as it provides invaluable data for use in monitoring drought conditions via satellite-borne sensors (Mishra and Singh, 2011). Satellite observations have significantly improved the accuracy and spatial extent of modern hydrological models as a result of their ability to measure hydrological variables such as precipitation, evaporation, soil moisture, total water storage, and lake and river levels (Sheffield and Wood, 2011; Mishra and Singh, 2011). However, there are several challenges associated with remote sensing, including the limited penetration depth of measured light (just a centimeters into soil or a single meter into snowpack); interference caused by vegetation, clouds, and radio frequencies; temporal and spatial coverage disconnections; and limited atmospheric windows due to strong atmospheric absorption and radiation scattering (National Research Council, 2007; Sheffield and Wood, 2011; Mishra and Singh, 2011; Rodell, 2012). In order to mitigate these issues, the best approach is the combination of satellite observations with different hydrological models. Data assimilation 60  is an effective means of merging observations with models (Sheffield and Wood, 2011; Mishra and Singh, 2011); in fact, data from multiple sources with different resolutions and accuracies can be integrated via data assimilation (Mishra and Singh, 2011).  In recent decades, several studies have been conducted to estimate soil moisture using the North American land data assimilation system (NLDAS). The NLDAS is a multi-institutional project that creates accurate land surface model (LSM) datasets from observed and modeled atmospheric data (Robock et al., 2003; Luo and Wood, 2008; Mishra and Singh, 2011). Robock et al. (2003) evaluated real-time NLDAS land surface models for use in the calculation of land hydrology on the southern Great Plains during the warm season. Luo and wood (2007) later developed a drought monitoring and prediction system (DMAPS) by using real-time NLDAS forcing in a variable infiltration capacity (VIC) land surface model. Their results indicated that DMAPS can provide near real-time qualitative assessment of drought, and can even predict the onset of a drought several months in advance (Luo and wood, 2007). In addition, Luo and Wood (2008) used real-time NLDAS forcing to drive the VIC land surface model to estimate soil moisture values.   2.10.6. Drought Management Droughts are the costliest of all natural disasters, and have become much more frequent in their occurrence in recent years. The potential impacts of drought include a rise in water demand, hydro-meteorological variability, and societal vulnerability; because of this, it is vital to effectively manage water resources during drought periods (Merabtene et al., 2002; Mishra and order to satisfy the continuous water demand (Merabtene et al., 2002). The most important 61  aspects of drought management are the decision support system (DSS) approaches and multi-criteria decision analysis (MCDA), which are discussed below (Mishra and Singh, 2011):  1) DSS are user-friendly graphical model interfaces for water resource systems. DSS is based on the integration of different models and its outputs are provided to policy makers so they can issue warning or suggest preparedness action plans (Mishra and Singh, 2011). Merabtene et al. (2002) developed a DSS to evaluate drought vulnerability of the water supply system as well as optimal water supply strategies for Fukuoka City, Japan. Their proposed DSS was introduced to minimize the risks of drought damage and improve utilization of water resource systems during times of drought (Merabtene et al. 2002). Pallotino et al. (2005) proposed a DSS based on the scenario analysis approach. They examined a set of statistically independent hydrological scenarios in order to obtain a robust decision policy. Their results indicated that this DSS could be easily adopted by practitioners and end-users of the water resource systems in Sardinia, Italy (Pallottino et al., 2005).  2) MCDA is an integration tool for use in assessing alternative options. The aim of MCDA is to identify alternative actions based on appropriate quantitative and qualitative assessment criteria, enabling decision-makers to make informed decisions regarding potential set of alternatives. Effective drought management involves the combination of different perspectives, including meteorological, hydrological, ecological, environmental, and socio-economic. Using MCDA in drought management is advisable for making decisions regarding preliminary planning for drought impact and the implementation of strong drought management plans (Traore and Fontane, 2007; Mishra and Singh, 2011). However, MCDA method have some limitations that should be considered, including difficulty in understanding the risks associated with prolonged or persistent drought conditions (if risks 62  are unquantified) and the challenge of translating into publically understandable terms (Mishra and Singh, 2011). Rossi et al. (2005) used simulation models with MCDA in order to assess drought mitigation strategies on a water supply system in Sicily, Italy. The effect of several drought mitigation alternatives were assessed using the simulation model, and the MCDA was applied to economically, environmentally, and socially rank alternatives. Their results indicated that the proposed methodology could be adopted in the decision-making process for comparing drought mitigation strategies (Rossi et al., 2005). Traore and Fontane (2007) developed a method to manage drought impacts based on MCDA using strategic, tactical, and emergency measures for the Niger River in Mali, Africa. Their results revealed the importance of considering tactical and emergency management as well as strategic objectives in managing drought impacts (Traore and Fontane, 2007).   2.11. Summary  Drought is a natural phenomenon that it is expected to affect more areas in the future due to climate change. Overall, the majority of existing drought indices were developed to study and evaluate drought impacts on human needs such as crop production and freshwater supplies; however, drought impacts are not limited to human concerns. Other components, such as natural habitat and stream health are also affected by drought; thus, in order to attain sustainable water management, all relevant areas of drought impact must be considered. To the best of my knowledge, no study has been conducted to evaluate the impact of drought on stream health, which is a major indicator of environmental sustainability. The main purposes of this study are: 1) to evaluate the impacts of drought on stream health by the development of a new index based on the bioassessment approach; 2) to develop an overall/comprehensive drought index that 63  considers different drought impacts, including meteorological, agricultural, hydrological, and stream health.  64  3. INTRODUCTION TO METHODOLOGY AND RESULTS This dissertation consists of two research papers that have been submitted to scientific journals. The first study focuses on development of a new index to capture drought impact on aquatic ecosystems. The second study builds upon the first by introducing a comprehensive drought index that incorporates different aspects of droughts, including meteorological, agricultural, hydrological, and stream health.   a new concept of stream health drought. To accomplish this, a hydrological model was calibrated and validated using observed streamflow data obtained from nine monitoring stations within the Saginaw Bay Watershed. The hydrological model outputs (daily streamflow data for all stream segments) were used as input data for a regional-scale habitat suitability model capable of quantifying the impact of flow reduction on fish assemblages. In order to develop the drought predictive models, 66 physiographical and climatological variables were examined; due to the large number of variables, the ReliefF algorithm was used to rank the most influential variables. The top-ranked variables were then used to develop six predictive drought models using the partial least square regression technique. The final model was selected based on goodness-of-fit (R2) and accuracy measures.  Finally, the performance of the best predictive drought model was examined based on 47 different climate scenarios.  Development and Evaluation of a Comprehensive Drought meteorological, agricultural, and hydrological indices, in order to develop an overall/comprehensive drought index. To accomplish this, 13 commonly used drought indices were selected and normalized; four for each meteorological, agricultural, and hydrological 65  drought category and one for the stream health category. The three closet drought indices to each other in each category were identified and averaged; then the scores for each drought category were averaged to obtain the overall score. Predictive categorical and overall drought models were developed based on the categorical and overall drought scores. In order to obtain the drought predictive models, 90 variables were used (the same variables that were used to calculate the 13 original drought indices); due to the large number of variables, the ReliefF algorithm was used to rank and select the best set of variables. The selected variables were then used in an adaptive neuro-fuzzy inference system to develop four predictive drought models: one predictive model was developed for each drought category (i.e., meteorological, agricultural, and hydrological), and the last was developed as the overall drought model.    66  4. DEFINING DROUGHT IN THE CONTEXT OF STREAM HEALTH 4.1. Abstract  Droughts affect many sectors, such as agriculture, economic, social, human health, and ecosystems. Many drought indices have been developed; yet, none of them quantifies the impacts of drought on stream health. The purpose of this study is to define a new drought index capable of assessing fish vulnerability. To accomplish this, a hydrological model, called the Soil and Water Assessment Tool, and the Regional-scale Habitat Suitability model were integrated in order to understand the state of drought within 13,831 stream segments within the Saginaw Bay Watershed. The ReliefF algorithm was used as the variable selection method, and partial least squared regression was used to develop two sets of predictor models capable of determining current and future drought severities. Forty-seven different climate scenarios were used to investigate drought model predictability of future climate scenarios. The results indicated that the best drought model has a high capability for predicting future drought conditions with R2 values ranging from 0.86 to 0.89. In general, the majority of reaches (94%) will experience higher drought probability under future climate scenarios compare to current conditions. The procedure introduced in this study is easily transferable to other watershed to measure the impacts of drought on stream health.  Key words: Great Lakes; Stream Health; Climate Change; Risk   4.2. Introduction   Droughts are temporary events that can occur almost in all climatic zones and are related to the reduction in received precipitation during a period of time (Wilhite et al., 2014; Mishra and Singh, 2010). Drought ultimately impacts both surface and groundwater resources (Mishra and Singh, 2010). Droughts rank first, among all the natural hazards that affect the human well-67  being (Wilhite, 2000b; Mishra and Singh, 2010); and they are the most costly natural disasters of the world (Wilhite, 2000b; Keyantash and Dracup, 2002). Globally, droughts cause an average of $6 to $8 billion in damages annually (Wilhite, 2000b; Keyantash and Dracup, 2002). Therefore, it is important to predict the timing and extent of droughts to help with development of mitigation strategies.    Drought is typically classified as either meteorological, hydrological, agricultural, or ecological drought (Wilhite and Glantz, 1985; American Meteorological Society, 1997; McMahon and Finlayson, 2003; Sheffield and Wood, 2011). Moreover, for each type of drought several drought indices have been developed. Meteorological droughts occur when there is a significant deviation from the mean precipitation in a region (Mishra and Singh, 2010; Sheffield and Wood, 2011). The Standardized Precipitation Index (McKee et al., 1993, 1995; Mishra and Desai, 2005a, 2005b; Cancelliere et al., 2007; Mishra et al., 2007; Mishra and Singh, 2009a) and Percent of Normal (Hayes, 2006; Sheffield and Wood, 2011; Zargar et al., 2011) are examples of commonly used meteorological drought indices. Hydrological droughts refer to a period of deficiency in the supply of water (both surface and subsurface water) (Panu and Sharma, 2002; Mishra and Singh, 2010; Sheffield and Wood, 2011). Streamflow, lake/reservoir levels, and groundwater levels are the parameters that are used to define hydrological drought (Mishra and Singh, 2010; Sheffield and Wood, 2011). Common hydrological drought indices are the Palmer Hydrological Drought Index (Palmer, 1965; Heim, 2000; Keyantash and Dracup, 2002; Mishra and Singh, 2010; Zargar et al., 2011), the Baseflow Index (Institute of Hydrology, 1980; Gustard et al., 1992; Zaidman et al., 2001; Tallaksen and van Lanen, 2004; Sheffield and Wood, 2011), and the Surface Water Supply Index (Shafer and Dezman,1982; Heim, 2002; Hayes, 2006; Mishra and Singh, 2010; Sheffield and Wood, 2011). Agricultural droughts are defined as a 68  period of soil moisture deficiency, which reduces moisture supply for vegetation and crop yield (Panu and Sharma, 2002; Sheffield and Wood, 2011). This type of drought is driven by meteorological and hydrological droughts (Sheffield and Wood, 2011). Several drought indices have been used to study agricultural drought including the Palmer Drought Severity Index (Alley 1984; Rao and Padmanabham, 1984; Johnson and Kohne, 1993; Kim and Valdes, 2003; Dai et al., 2004; Özger et al., 2009) and the Crop Moisture Index (Palmer, 1968; Hayes, 2006; Mishra and Singh, 2010; Sheffield and Wood, 2011). These indices use a combination of hydrometeorological variables such as precipitation, soil moisture, and temperature to analyze agricultural drought (Mishra and Singh, 2010). Ecological drought indices measure the impacts of drought on ecosystems (Sheffield and Wood, 2011); yet, few indices have been developed to quantify these impacts. Examples include the Normalized Difference Vegetation Index that is generally used to monitor the health of a canopy (Rouse et al., 1974; Singh et al., 2003; Kogan, 2005) and Vegetation Condition Index (Kogan, 1995; Singh et al., 2003; Quiring and Ganesh, 2010; Wardlow et al., 2012).  In general, a concept of drought that has been received the least attention is ecohydrological aspects of drought that can be summarized as stream health. A healthy stream is an ecosystem that is flourishing, sustainable, resilient to stress, and maintains its societal values over time (Meyer, 1997). Many biological monitoring methods exist to measure the ecological conditions of stream systems. Among these methods, biological indicators are widely used for detecting the presence of point and non-point source pollutants, changes in physical habitat, and the effects of long-term disturbance events on ecosystems (Barbour et al., 1999; Nerbonne and Vondracek, 2001; Flinders et al., 2008). Fish are the most commonly used biological communities for water-quality assessments (Barbour et al., 1999; Flinders et al., 2008; Carlisle et 69  al., 2013). Fish are sources of food for aquatic and terrestrial species, while being primary consumers of macroinvertebrates and algae (Carlisle et al., 2013). This links fish communities to other biotic characteristics of the ecosystem, which allows fish to be representative of the larger picture within the stream system. Furthermore, fish are relatively easy to collect and identify, provide long-term and regional impacts due to their mobility and lifespan, and their environmental requirements are well-known (Karr, 1981; Barbour et al., 1999; Carlisle et al., 2013). Additionally, fish assemblages cover a variety of trophic levels such as omnivores, herbivores, insectivores, planktivores, and piscivores, which provides an integrative view of stream environmental health (Karr, 1981; Barbour et al., 1999).   Flow is a key driver of stream ecological processes that affect aquatic organism performance, distribution, and abundance (Hart and Finelli, 1999; Bunn and Arithington, 2002). Alteration of flow regimes especially during dry seasons can significantly affect the ecosystem health (Stewart-Koster et al., 2010; Hamaamin et al., 2013). Drought perturbs stream ecological conditions by altering native biological communities such as fish assemblages (Lake, 2003). Drought can cause reductions and alterations in fish populations and their structure by reducing spawning and recruitment (Lake, 2003). Therefore, it is important to quantify the impacts of drought on stream biota.  In this study, we are defining a new drought index in the context of stream health. In general, the majority of drought indices are sensitive to the impacts of drought to human usages including drinking or crop production neglecting other aspects of environmental sustainability such as stream health. Therefore, this study is unique because it uses fish integrity as an indicator to define drought. By coupling the hydrologic model with a regional-scale habitat suitability model, the drought model will be developed capable of identifying drought zones for all streams 70  within the study area. This allows targeting the streams that are more prone to degradation due to extreme climatological conditions allowing mitigation practices to be more effectively deployed. 4.3. Materials and Methodology   4.3.1. Study area The study area for this study is the Saginaw Bay Watershed located in the east central ea of 16,122 km2, its final outlet drains into Lake Huron, Figure 1. Most of this area is agricultural and forest lands (37% and 37%, respectively), with the agricultural lands dominated by corn and soybean crops. The remaining lands are pasture (9.5%), urban (7.5%), wetlands (8%), and water (1%). The Saginaw Bay -digit hydrologic unit code (HUC 040802) and consists of six 8-digit HUC watersheds, the Tittabawassee (HUC 04080201), Pine (HUC 04080202), Shiawassee (HUC 04080203), Flint (HUC 04080204), Cass (HUC 04080205), and Saginaw (HUC 04080206).There are 13,831 stream segments within the Saginaw Bay Watershed with different sizes and temperatures; with the majority of streams being warm water streams (Einheuser et al., 2013). The Saginaw Bay Watershed has been designated as area of concern by the US Environmental Protection Agency due to fish consumption advisories caused by excessive agrochemical utilization and contaminated sediments (USEPA, 2013).  71   Figure 1. Saginaw Bay Watershed 4.3.2. Modeling process The goal of the modeling process is to predict drought zones based on stream health. In order to accomplish this goal, a multi-step modeling process was developed (Figure 2). First, the Soil and Water Assessment Tool, a hydrological model, was used to obtain daily streamflow data (1972-2012) for all stream segments in the Saginaw Bay Watershed. The daily streamflow data was used as an input into a regional-scale habitat suitability model in order to assess the impacts of flow fluctuation on fish assemblages. Next, the changes in fish assemblages were translated into drought zones. Knowing drought zones for each stream segment, it was hypostasized that a 72  drought predictive model could be developed using physiographical and climatological variables. Selected variables were then used to accomplish two general goals: 1) develop a drought model capable of determining current drought severity (using ReliefF algorithm) and 2) develop a drought forecast model capable of predicting future drought severity (using time series variables). Finally, the partial least square regression was used to create drought predictive models using the previously selected variables.   Figure 2. Drought zones variable selection and modeling process  4.3.3. Soil and Water Assessment Tool  In this study, the Soil and Water Assessment Tool (SWAT) was used to simulate daily streamflow data for 13,831 stream segments of the Saginaw Bay watershed. SWAT is a 73  physically based, continuous time model developed by the US Department of Agriculture - Agricultural Research Service (Gassman et al., 2007). In this spatially explicit model, a watershed is delineated into multiple subwatersheds, which are further segmented into hydrologic response units (HRUs) with homogenous land cover, soil, slope, and management practices. This model uses physiographical and climatological characteristics of a region to simulate streamflow, runoff, soil erosion, as well as nutrient, sediment, and pesticide loadings (Gassman et al., 2007; Neitsch et al., 2011).   Different sources were used to obtain the physiographical and climatological data needed to run SWAT model. The National Elevation Dataset (NED) of the US Geological Survey (USGS) with a spatial resolution of 10 m was used to represent the topography data of the region (NED, 2014). The Natural Resources Conservation Service (NRCS) Soil Survey Geographic (SSURGO) database was used to identify soil characteristics in the area of interest (NRCS, 2014a). The 2012 Cropland Data Layer (CDL) of the United States Department of Agriculture-National Agricultural Statistics Service (USDA-NASS) with a spatial resolution of 30 m was used to represent land use/land cover data (NASS, 2012). Climatological data were obtained from 16 precipitations and 13 temperature National Climatic Data Center (NCDC) stations. Daily precipitation and temperature data were obtained at these stations for the period of 1972 to 2012.  4.3.4. SWAT model calibration and validation The SWAT model was calibrated and validated against the observed daily streamflow data of nine USGS gauging stations (presented in the Supplementary Material, Figure S1) from 2001 to 2010. The first half of this period (2001 to 2005) was used for calibration and the second half (2006 to 2010) was used for validation. Three statistical variables were used to examine the 74  quality of calibration and validation: Nash-Sutcliffe model efficiency coefficient (NSE), root-mean-squared error-observations standard deviation ratio (RSR), and percent bias (PBIAS). Passing criteria for these three variables are NSE > 0.5, RSR < 0.7, and PBIAS < ±25 on a monthly basis (Moriasi et al., 2007).  4.3.5. Regional-scale Habitat Suitability Model  The calibrated SWAT model was run from 1972 to 2012 in order to obtain the streamflow data needed for a regional-scale habitat suitability model. This model was created with the goal of introducing regional environmental flow standards for Michigan Rivers (Zorn et al., 2008). Environmental flow is defined as the quantity and quality of the water flow required to sustain freshwater ecosystems (Poff et al., 2010). Therefore, developing standards for environmental flow can protect aquatic ecosystem from adverse impacts. The regional-scale habitat suitability model predicts the effect of flow reduction on fish assemblages during summer months (July, August, and September) (Zorn et al., 2008; Hamilton and Seelbach, 2011). In Michigan, summer months are the period with the lowest flow for most streams and one of the most biologically stressful periods. In order to characterize this period, an index flow was developed. The index flow is the median of the daily flow values of the lowest summer month of the flow regime (Hamilton and Seelbach, 2011). Critical flow reduction will be calculated based on the percentage of index flow. In this model, about 40 fish species were used as stream health indicators. The fish data were obtained from fish surveys at 1,720 sites from 1980 to 2006. Three habitat variables were used to define the optimal fish species habitat conditions. These variables were catchment area, July mean water temperature, and base flow yield. The fish species were divided into characteristic and thriving species. Characteristic species were defined as those species that have all three of their habitat variable scores within 1.5 standard deviations of the 75  optimal values. And, thriving species were defined as those species that have all three of their habitat variable scores within 1 standard deviation of the optimal values (Zorn et al., 2008). Zorn et al. (2008) classified Michigan rivers into 11 groups based on their catchment size (stream, small river, and large river) and water temperature (cold, cold-transitional, cool, and warm). The fish assemblage response curves for all 11 river classes were created to determine the effect of flow reduction on characteristic and thriving species. Each response graph (e.g. Figure 3) has two curves, one for thriving species and one for characteristic species response. Based on the Biological Condition Gradient concept the Groundwater Conservation Advisory Council recommended to divide the response curve into four risk zones, i.e. A, B, C, and D. The potential risk of flow reduction increases from zone A to zone D, where zone A represents no risk to the fish population, zone B shows alert and attention to the fish population, zone C represents concern and prevention of flow reduction, and zone D shows Adverse Resource Impacts (ARI) to the fish population. The threshold between zone A and B represents a 10% reduction in the thriving fishes population. The threshold between zone B and C shows a 20% reduction in the thriving fishes population. The line between zone C and D is called the ARI line, which shows the threshold of flow reduction for causing ARI to characteristic fish species. This corresponds to 10% reduction in the characteristic fishes population (Zorn et al., 2008).  76   Figure 3. Fish response curve to flow reduction (adapted from Zorn et al., 2008) In order to define drought zones, first, the index flow for each stream in Saginaw Bay Watershed is determined. The SWAT model was run for 41 years, i.e. from 1972 to 2012, to simulate the daily streamflow data. Using this data, the lowest daily median flow values of the lowest summer month were calculated for each stream. Based on the Zorn et al. (2008) model criteria, four drought zones (A, B, C, and D) were defined: Zone A, representing the no drought condition, Zone B, showing the moderate drought condition, Zone C,  indicating the severe drought conditions, and Zone D, referring to the extreme drought conditions. This information was used to create the reference table of drought zones (Table 5). Using the reference table, drought condition can be identified for each stream segment by multiplying the index flow (obtained from the SWAT model for the period of study 1972 to 2012) to correspondent value of Table 5.    77  Table 5. Reference table of drought zones (adapted from Hamilton and Seelbach, 2011) Ecological Stream Types Zone A Zone B Zone C Zone D Cold Streams >86% None >80%-86%  Cold Small Rivers >89.5% None >79%-89.5%  Cold Transitional Streams None >96% None  Cold Transitional Small Rivers None >98% None  Cold Transitional Large Rivers None >97% None  Cool Streams >94% >85%-94% >75%-85%  Cool Small Rivers >85% >81%-85% >75%-81%  Cool Large Rivers >86% >81%-86% >75%-81%  Warm Streams >90% >82%-90% >76%-82%  Warm Small Rivers >92% >87%-92% >83%-87%  Warm Large Rivers >90% >84%-90% >78%-84%   4.3.6. Drought Model Input Variables  In this study, a total of 66 variables were initially considered for development of the drought model as independent variables. These variables were categorized as follows: precipitation (25 variables), streamflow (24 variables), land use (8 variables), soil (8 variables), and drainage area (1 variable).  The precipitation variables included the total precipitation for the month of interest, the total precipitation of each of the previous 12 months, and the average precipitation in past  months was included, i.e. , where  is the precipitation of the i-th month before the month of interest. To further elaborate on the last group, the independent variable for n = 1 correspond to the average precipitation of the month of interest and one month before that; when setting n = 2 that corresponds to the average precipitation of the month of interest and that of the previous two months; and so on. The monthly average flow rates of the prior 1 to 24 months were also considered as independent variables. The land use was categorized as agricultural, forested, urban, and water areas. The actual area (4 variables) and 78  percentage of these land uses (4 variables) for the area above each stream segment were calculated summing up to eight land use variables. Soil data was divided into four hydrologic soil groups of A, B, C, and D (NRCS, 2007). The soil groups were categorized based on their infiltration and water transmission rates. Group A soils, with gravel or sand texture, consist of gravel or sand (>90%) and clay (<10%). These soils have high infiltration and water transmission rates. Group B soils have a loamy sand or sandy loam texture and consist of 10-20% clay and 50-90% sand. These soils have moderate infiltration and water transmission rates. Group C soils, with loam or silt loam texture, have 20-40% clay and less than 50% sand. These soils have low infiltration and water transmission rates. Group D soils, with clayey texture, have more than 40% clay and less than 50% sand. The infiltration and water transmission rates of these soils are very slow (Cronshey et al., 1986; NRCS, 2007). Like the land use variables, the actual area and percentage values of each soil groups were calculated for each subbasin adding up to eight soil variables. The last variable was drainage area, which was calculated as the total area above the outlet of each stream segment.   4.3.7. Variable Selection: ReliefF algorithm  The ReliefF algorithm is a commonly used method for feature selection (Robnik-Sikonja and Kononenko, 2003). This method ranks the independent variables according to their relevance or importance in classifying or predicting the dependent variable (Kononenko, 1994; Robnik-Sikonja and Kononenko, 2003). The ReliefF algorithm is the improved version of the Relief algorithm, which was originally developed for binary classification (Kira and Rendell, 1992b; Robnik-Sikonja and Kononenko, 2003). ReliefF is capable of handling data with strong dependencies or outliers (Kononenko, 1994; Robnik-Sikonja and Kononenko, 2003; Mahlein et al., 2013).  79  For any given sample, ReliefF searches for the nearest neighborhoods of the same-class, also known as hits, and of the different-class, also known as misses. With k being the number of each neighborhood samples, there will be k nearest hits and k nearest misses for each sample. The nearest hits and misses are usually defined by the Euclidean distance (L2 norm) or Manhattan distance (L1 norm). The relevance of variables are determined by the sum of the Euclidean distance, or the Manhattan distance, between the nearest hits and nearest misses of all samples (Robnik-Sikonja and Kononenko, 2003; Mahlein et al., 2013), as follows:  Where, n is equal to 1 for the Manhattan distance and is equal to 2 for the Euclidean distance, si is the i-th sample, NHi is a set of k nearest-hit to the si, and NMi is a set of k nearest-miss to the si, and Wi is the weight of the feature. This equation is set up in such a way that each feature is penalized, if it differs greatly from that of the nearest-hits, and rewarded otherwise, in case of nearest-misses. ReliefF runtime scales linearly with the number of independent variables, i.e. if the number of independent variables are doubled, the algorithm would take twice as long. However, the computational requirement of the algorithm increases non-linearly with increasing sample size. This means that if the number of observations are doubled, the time needed to perform the required computation will increase more than twice. This is mostly due to the sorting and the distance calculation of each sample to all the other samples, which results to  distinct distance values that are required to determine the nearest-hits and nearest-misses. In this study, there were more than 6.6 million observations (13831 streams × 40 years ×12 months), where each observation has 66 independent variables. Having these many observations made it impossible to use a single ReliefF run. Therefore, a subset of 10,000 samples was randomly 80  selected from the original data set, and ranked. This procedure was repeated 2500 times; resulting in 2500 different rankings for each independent variable. A histogram of the ranking for each independent variable was constructed and averaged to determine the final score. Selected variables were then used to accomplish two general goals: 1) Develop the most accurate drought model capable of determining current drought severity for all stream segments within the study area. This model is called the Current Drought Severity Model and was tested by using three sets of variables (the top 5, 10, and 15 ranked variables obtained from ReliefF) (Table S1). 2) Develop the most accurate drought forecast model capable of predicting future drought severity. This model is called the Future Drought Severity Model and was tested against three sets of variables that include all precipitation and streamflow variables from 6, 12, and 18 months prior to the month of interest.  4.3.8. Partial Least Square Regression  The partial least square regression (PLSR) is a statistical approach used for modeling linear relations between multivariate measurements (de Jong, 1993; Wold et al., 2001). However, the main advantage of PLSR over general linear regressions is its ability to deal with strongly collinear, noisy, incomplete, and large arrays of independent variables (Wold et al., 2001; Carrascal et al., 2009).  In order to train, test, and select the best PLSR model, 10-fold cross validation was used. In 10-fold cross-validation, the dataset is randomly divided into 10 equally sized exclusive subsets or folds. Nine folds of the data are used for training (90%) and the remaining (1-fold) is used for testing (10%) (Hamaamin et al., 2013). This process was repeated 10 times with a new non-overlapping testing fold until all folds were used for testing the PLSR model. 81  For the development of both current and future drought severity models, PLSR was used to predict the relationship between the median flow of each stream segment for the month of interest (dependent variable) and the independent variables (Table S1). The initial test showed that the observed median flow is highly skewed (Figure S2); therefore, the dependent and independent variables were transformed using  before the model development.  In order to evaluate the performance of these models, the accuracy, precision, and sensitivity of each model, in predicting drought zones, were determined. Accuracy refers to the overall correctness of the model (Eq. 1). Precision is an estimate of how correct the model outputs are for each class (Eq 2). Sensitivity (Eq. 3) refers to the model ability to correctly pick instances of a certain class (Aruna et al., 2011).  Accuracy, precision and sensitivity are calculated as follows:        (1)          (2)        (3) Where, P stands for the number of positive cases and N stands for the number of negative cases. The definition of positive and negative changes depending on what is being evaluated. For example, if the model is evaluating the presence or absence of the drought, then P indicates the number of observations point that drought is present and N represents those observations that show no sign of drought. While considering the definition of the positive and negative, TP stands for True Positive, meaning observations that are positive and they are correctly classified as positive. TN stands for True Negative, meaning observations that are negative and they are correctly classified as negative. FP, stands for False Positive, meaning the observations that are 82  negative but mistakenly classified as Positive. Finally, FN stands for False Negative, meaning the observations that are positive but mistakenly classified as negative. 4.3.9. Climate Models In this study, 16 different general circulation models (GCMs) from the Coupled Model Intercomparison Project Phase 5 (CMIP5) were used. CMIP5 aims to provide a multi-model dataset including long term and near term experiments that offer better understanding of climate change and climate variability (Taylor et al., 2012). The long-term experiments were used in this study, which range from mid-1900 to 2100 and beyond (Taylor et al., 2012). The names, modeling centers, atmospheric resolutions, and their components are listed in Table 6. For each GCM, data driven by three Representative Concentration Pathways (RCPs) scenarios (RCP4.5, RCP6.0, and RCP8.5 (Moss et al., 2010) was extracted for the future period of 2040-2060 and the historical period of 1980-2000. The latter is referred to as the control period data. Additionally, the weather station data for daily observation for the same time period as control data were obtained from the NCDC and used as the baseline in the analysis. The locations of these weather stations are presented in Figure S1. The climate is projected to change significantly in the future and such changes are associated with large uncertainty, which cannot be neglected for impact assessments, especially when extreme situations are of the primary concern (IPCC, 2013). In impact assessment studies, one way to take uncertainty into consideration is by adopting an ensemble approach where the ensemble could refer to differences in the models, emission scenarios, initial conditions, etc. (Parker, 2013).  In the current study, the ensemble was constructed based on different GCMs and emission scenarios. Although many other downscaling methods like quantile mapping, multiple regression, and artificial neural networks exist in the literatures (Maraun et al., 2010; Winkler et 83  al., 2011a and 2011b; Hessami et al., 2008; Sailor et al., 2000; Hewitson and Cran et al. 1996; Schoof and Pryor et al., 2001), the delta method is chosen in the current study due to its ability to produce a large ensemble of projections (Winkler et al., 2011b). The delta method is also widely used in hydrological studies (Morrison et al., 2002; Merritt et al., 2006; Adam et al., 2009; Elsner et al., 2010; Dessu and Melesse, 2013; Woznicki and Nejadhashemi, 2014).   The assumptions of the delta methods are as following: a) relative change is better simulated by GCMs compared to absolute values (Fowler et al, 2007; Loukas et al., 2007), b) the number and temporal sequence of wet days remains unchanged (Fowler et al, 2007; Dessu et al., 2013; Htut et al., 2014), c) the GCMs biases for both mean and variability will be similar for the control and future periods, ignoring GCMs biases in the distribution of simulated variables (Boyer et al., 2010; Winkler et al., 2011b; Htut et al., 2014), and d) the spatial pattern and temporal variability of the present climate is maintained in the future (Diaz-Nieto and Wilby, 2005; Boyer et al., 2010). The delta method was used to downscale the climate data to allow local scale analysis and for the impact model, which requires data inputs at daily time steps. The variables used in the delta method included the daily maximum temperature, daily minimum temperature, and daily total precipitation. In the delta method, the difference/ratio between the future and control period were calculated for the monthly averaged daily temperature/precipitation, and then applied to the observed daily time series. The temperature change factors are additive and can be negative values. On the other hand, the precipitation change factors are calculated as ratios with precipitation being zero bounded (Woznicki and Nejadhashemi, 2012).  84  Table 6. CMIP5 models developer, name, resolution, and components (Petkova et al., 2013; IPCC, 2013) Modeling Center/ID Model Atmospheric Resolution (latitude × longitude) Components   The First Institute of Oceanography/FIO FIO-ESM 2.8° × 2.8° Atm1,Aero2,LS4,O5,OB6,SI7 Institut Pierre-Simon Laplace/IPSL IPSL-CM5A-LR 3.75° × 1.9° Atm1,Aero2,LS4,O5,OB6,SI7 IPSL-CM5A-MR 2.5° × 1.25° Atm1,Aero2,LS4,O5,OB6,SI7 Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine-Earth Science and Technology /MIROC MIROC5 1.41° × 1.39° Atm1,Aero2,LS4,O5,SI7 Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), and National Institute for Environmental Studies/MIROC MIROC-ESM 2.81° × 1.77° Atm1,Aero2,LS4,O5,OB6,SI7 MIROC-ESM- CHEM 2.81° × 1.77° Atm1,Aero2,AtmCH3, LS4,O5,OB6,SI7 Met Office Hadley Centre /MOHC HadGEM2-ES 1.875° × 1.25° Atm1,Aero2,AtmCH3, LS4,O5,OB6,SI7 Meteorological Research Institute /MRI MRI-CGCM3 1.125° × 1.125° Atm1,Aero2,LS4,O5,SI7 NASA Goddard Institute for Space Studies /NASA GISS GISS-E2-R 2.5° × 2.0° Atm1,Aero2,AtmCH3, LS4,O5, SI7 GISS-E2-H 2.5° × 2.0° Atm1,Aero2,AtmCH3, LS4,O5, SI7 85      National Center for Atmospheric Research/NCAR CCSM4 1.25° v 0.9° Atm1,Aero2, LS4,O5, SI7 National Institute of Meteorological Research/Korea Meteorological Administration /NIMR/KMA HadGEM2-AO 1.875° × 1.25° Atm1,Aero2, LS4,O5, SI7 NOAA Geophysical Fluid Dynamics Laboratory/NOAA GFDL GFDL-CM3 2.5° × 2.0° Atm1,Aero2,AtmCH3, LS4,O5, SI7 GFDL-ESM2G 2.5° × 2.0° Atm1,Aero2, LS4,O5,OB6,SI7 GFDL-ESM2M 2.5° × 2.0° Atm1,Aero2, LS4,O5,OB6,SI7 National Science Foundation, Department of Energy, National Center for Atmospheric Research (NSF-DOE-NCAR) CESM1(CAM5) 1.25° × 0.9° Atm1,Aero2, LS4,O5, SI7                  1 Atmosphere; 2 Aerosol; 3 Atmospheric Chemistry; 4 Land Surface; 5 Ocean; 6 Ocean Biogeochemistry; 7 Sea Ice 86  4.4. Results & Discussions 4.4.1 SWAT Model Calibration and Validation The results of SWAT model calibration and validation, using statistical criteria, such as NSE, RSR, and PBIAS, are presented in Table 7. The reported NSE, RSR, and PBIAS are the overall values for both the calibration and validation for the period of 2001 to 2010. The SWAT model met the statistical criteria for all nine USGS stations according to evaluation criteria, defined by Moriasi et al. (2007), the performance rating for all of the stations are in the range of very good, good, and satisfactory. Therefore, the model can be used to simulate streamflow data for the region satisfactorily.   Table 7. Statistical criteria for SWAT model calibration and validation for nine USGS gauging stations within the Saginaw Bay Watershed USGS Station  NSE* (performance rating) RSR** (performance rating) PBIAS*** (performance rating) 04144500 0.64 (Satisfactory) 0.60 (Satisfactory) 14.27 (Good)  04148140 0.54 (Satisfactory) 0.68 (Satisfactory) -9.68 (Very Good)  04148500 0.71 (Good) 0.54 (Good) 16.34 (Satisfactory)  04147500 0.63 (Satisfactory) 0.61 (Satisfactory) -1.53 (Very Good)  04151500 0.64 (Satisfactory) 0.60 (Satisfactory) 13.77 (Good)  04154000 0.54 (Satisfactory) 0.68 (Satisfactory) 9.65 (Very Good)  04155500 0.61 (Satisfactory) 0.63 (Satisfactory) 9.84 (Very Good)  04156000 0.73 (Good) 0.52 (Good) 6.44 (Very Good) 87   04157000 0.80 (Very Good) 0.45 (Very Good) 10.90 (Good) * Nash-Sutcliffe model efficiency coefficient,  ** Root-mean-squared error-observations standard deviation ratio,  *** Percent bias. 4.4.2 Variable Selection 4.4.3.1.Current Drought Severity Model   The ReliefF algorithm was used for the development of the most accurate drought model capable of determining current drought severity for all streams. The ranking of all 66 variables is presented in the histogram map (Figure 4). The y-axis represents the 66 independent variables, and the x-axis represents their ranking. The color in Figure 4 shows how often a variable has obtained a certain rank during the 2500 different random sampling. The final number was normalized to scale between 0 and 1 (abundance). The dark blue indicates that the variable never obtained that rank. As the color spectrum moves from dark blue to dark red, it indicates that the variable obtained that rank more often than other independent variables. As an example, Figure S3 shows the histogram of the ranking of variable #20 (average flow rate from 23 months prior to the month of interest). This variable mostly obtained rank 7 during 2500 iterations. Therefore, in Figure 4, on line 20, rank 7 is being shown the most red. The final ranking for each variable was determined based on the average of these 2500 different ranking.   In general, the average flow rate variables were ranked much higher than any other type of variables (Figure 4). This is mainly due to the high correlation between median flow and average flow rate. After the average flow rate variables, the precipitation variables were ranked the highest. This indicates that the stream system in general is less flashy during the dry season and more influenced by the groundwater system. Therefore, the median flow is not as dependent 88  on precipitation as it was to the average flow rate. Finally, the physiological variables were shown to be the least related to the changes in the median flow indicating that the changes to the median flow are insensitive to the total drainage area, land use, and soil type.     Figure 4. ReliefF ranking histogram map Out of the 66 original variables, the top 5, 10, and 15 ranked variables were used to develop three sets of drought models. The list of the top 15 ranked variables are presented in Table 8. All of the top ranked variables are related to flow rate. The only difference between the variables is the month from which the flow rate was calculated.    Physiographical Variables Average Flow rate Variables Precipitation Variables Abundance 89  Table 8. Top 15 ranked variables Ranking Variables 1 Average flow rate from 1 month prior to the month of interest 2 Average flow rate from 2 months prior to the month of interest 3 Average flow rate from 24 months prior to the month of interest 4 Average flow rate from 12 months prior to the month of interest 5 Average flow rate from 13 months prior to the month of interest 6 Average flow rate from 11 months prior to the month of interest 7 Average flow rate from 3 months prior to the month of interest 8 Average flow rate from 23 months prior to the month of interest 9 Average flow rate from 14 months prior to the month of interest 10 Average flow rate from 10 months prior to the month of interest 11 Average flow rate from 4 months prior to the month of interest 12 Average flow rate from 22 months prior to the month of interest 13 Average flow rate from 15 months prior to the month of interest 14 Average flow rate from 9 months prior to the month of interest 15 Average flow rate from 5 months prior to the month of interest  4.4.2.2.Future Drought Severity Model   The future drought severity model should be capable of predicting drought conditions for all streams within an area of interest. The results from the ReliefF analysis showed that the physiological variables were the least related to the median flow. Therefore, they were not considered for the model development (Table S1). Overall, three sets of variables, all from 6, 12, and 18 months prior to the month of interest, were used in order to predict the future drought severity 6, 12, and 18 months in advanced, respectively. These variables included all average flow rate and precipitation variables for their respected period (Table S1). 4.4.3 Drought Severity Model 4.4.3.1.PLSR predictively for median flow The statistical analysis of the Current Drought Severity Model performances is presented in Table 9. Three models were developed using the top 5, 10, and 15 ranked variables. 10-fold cross validation was used to insure that the models were not over-trained or over-fitted by increasing the number of PLSR components. Mean Square Error (MSE) obtained from the 10-90  fold cross validation decreased asymptotically by increasing the number of PLSR components (Figure S4). This shows that the model is not over-fitted because the MSE values do not increased when incorporating more PLSR components. The performance of each model was studied on all streams and on stream orders (1 to 7). In the case of all streams, there is little to no change in R2 values (0.86) between the three models (Table 9performances. When comparing the stream order 6 and 7 (<2% of all streams) are exceptions in which R2 values slightly improve as the number of independent variables increase (Table 9). However, improvement in model predictability is minimal (R2 for stream order 6 will be changed from 0.64 in model 1 to 0.66 for the model 3) while the number of independent variables tripled from 5 to 15, respectively.  As presented in Figure 5, the first, second, and third PLSR models were able to explain 86.5%, 87% and 87.1% of the variance of the output (median flow), respectively. Since the difference between these three models are not considerable; the first model, which requires the least number of independent variables, was selected as the best predictive Current Drought Severity Model for this study.    Table 9. Current Drought Severity Model performances Model Number of Variables  R2 Stream order (All)   (1) (2) (3) (4) (5) (6) (7) First 5 0.86 0.74 0.71 0.72 0.76 0.81 0.64 0.74 Second 10 0.86 0.74 0.72 0.72 0.76 0.81 0.66 0.76 Third 15 0.86 0.74 0.72 0.72 0.76 0.81 0.66 0.76  The statistical analyses for the Future Drought Severity Model performances are shown in Table 10. Similar to the Current Drought Severity Model, three models were evaluated using 91  three sets of variables (flow rate and precipitation variables from 6, 12, and 18 months prior to the month of interest). The same procedure (10-fold cross validation) was used for the fourth, fifth, and sixth models to make sure the models are not over-fitted by increasing the number of PLSR components. The MSE of the Future Drought Severity Models was similar to the Current Drought Severity Models, where the MSE values decreased asymptotically by increasing the number of PLSR components, Figure S5.   Among these models, the fourth and fifth models have higher R2 values (0.85) and thus perform better than the sixth model (R2 = 0.76). Similarly, with respect to the stream order, both the fourth and fifth model performed better than the sixth model. The lower R2 value for the sixth model is due to its PLSR model deficiency in explaining the output variance (Figure S6). The fourth and the fifth model can explain more about 84% of the variance; however, the sixth model can explain 75% of the variance at most, Figure S6. This could be due to the fact that the variables used to predict drought 18 months in advance are not sensitive enough to detect future drought conditions.  Table 10. Future Drought Severity Model performances Model Number of Variables R2 Stream order (All)   (1) (2) (3) (4) (5) (6) (7) Fourth 34 0.85 0.71 0.67 0.67 0.71 0.76 0.58 0.69 Fifth 16 0.85 0.69 0.66 0.66 0.69 0.74 0.56 0.67 Sixth 7 0.76 0.53 0.46 0.46 0.50 0.61 0.32 0.49     92    Figure 5. The variance explained percentage for each PLSR for the Current Drought Severity Model: a) First model, b) Second model, c) Third model.                                     (a)                                       (b)                                 (c) 93  The histogram of measured and predicted median flow obtained from the Current and Future Drought Severity Models are presented in Figures 6 and S7, respectively. The horizontal axis represents the  of the median flow values and the vertical axis represents the number of counts. Overall, the histogram of the predicted median flows is very similar to the histogram of the measured flows for all models. However, the peak in the predicted histogram is higher than the peak in the measured histogram. This can be explained due to the fact that the predictive models overestimate the median flows for all stream orders (Table 10) and overestimate the median flows for stream orders 6 and 7 (Table 9). This difference was most pronounced for the sixth model (Figure S7). Therefore, it is expected that the error level is higher at the higher and lower median flow rates due to the shift from the two tails of the distribution for the sixth model to the peak of the median flow rates. This also explains the lower R2 values for the sixth model (R2 = 0.76).     94      (a)   (b) (c) Figure 6. The comparison of measured vs. predicted median flow histogram for the Current Drought Severity Model, a) First model, b) Second model, c) Third model.         95  4.4.3.2.Accuracy, precision, and sensitivity of drought models in predicting drought zones For each model, drought zones were identified by comparing the predicted median flow values against the reference table for the drought zones (Table ). The results from the six drought severity models are summarized in six confusion matrices (Tables - and S2-S5). The diagonal of the matrix shows the number of zones that have been correctly classified by the model. The overall accuracy of the model is determined by dividing the sum of the diagonal values by the sum of the all values of the matrix. As seen in Tables - and S2-S5, the first model through fifth model have an accuracy higher than 70%. However, the sixth model has accuracy of only 59%. This reflects the conclusions drawn in the previous section that the sixth model has a lower predictability than the other developed models. In general, all models have lower sensitivities and precisions for zone B and C compared to zone A and D. Especially for zone C, all cases had a sensitivity and precision percentage below 8%. This low predication is due to the small range that zone B and C cover. As mentioned earlier, there are eleven different classes of streams based on their sizes and temperature (Zorn et al., 2008), and each of these classes have specific index flow zoning ranges. For zone B and C and all stream classes, the zoning ranges cover only 10% or less of the index flow, while the remaining index flow reduction is covered by zone A and D. Therefore, accurately predicting these small ranges can be difficult, which ultimately reduce the sensitivity and precision of the models for zones B and C.  When comparing the performance of the Current Drought Severity Models, the first model performed the best. The first model had a higher accuracy, precision, and sensitivity compared to the other models. Among the Current Drought Severity Models (Tables , S2, and 96  S3), the first model has higher precision in classifying zone A (83%), and lower precision in classifying zone D (67%). In addition, it has a higher sensitivity in classifying zone D (66%), and slightly lower sensitivity in classifying zone A (84%). Therefore, it was concluded that the first model is the best among the Current Drought Severity Models. Among the Future Drought Severity Models, the fourth model performed slightly better than the fifth model. The fourth model has a slightly better accuracy (71%), a better precision in classifying zone D (62%), and a better sensitivity in classifying zone A (84%) compared to the fifth model. The sixth model, as expected, has a low accuracy (59%), precision, and sensitivity among all models. Therefore, it was concluded that the fourth model is the best among the Future Drought Severity Models. Table 11. Confusion matrix for drought zones: First model            Drought           Zone Predicted  A B C D Sensitivity Actual A 3,410,732 99,090 82,247 484,990 84% B 89,968 85,271 10,230 87,386 31% C 66,450 8,292 9,082 66,863 6% D 530,891 78,431 57,450 1,272,877 66% Precision 83% 31% 6% 67% Accuracy = 74%  Table 12. Confusion matrix for drought zones: Fourth Model             Drought           Zone Predicted  A B C D Sensitivity Actual A 3,400,622 81,426 66,065 503,081 84% B 107,300 82,647 6,523 75,077 30% C 83,139 6,350 6,304 54,046 4% D 756,301 85,320 52,635 1,026,748 53%  Precision 78% 32% 5% 62% Accuracy = 71%  97  4.4.4 Drought model performance under future climate scenarios In order to evaluate the capabilities of the drought models in predicting future drought caused by climate change, the best-selected drought model (the first model) was used. The predicted values resulting from the first model were compared with the calibrated SWAT model. In this section, zone A is referred to as the no-drought condition, and zones B, C and D are combined and denoted as drought conditions.    The results for the first drought model performance against each climate change scenario are presented in Tables S6 through S8. The overall performance of the first model for predicting future median flow and drought conditions are presented in Table . Based on R2 values, HadGEM2-ES performed better than the other models under RCP 8.5 and RCP 4.5. Under RCP 6.0, MIROC5 performed better than rest of the models. However, as shown in Table , the R2 values among the models did not vary substantially and ranged from 0.87 to 0.89. In addition, the model has reasonable R2 values ( > 0.866) for predicting future median flow rates with low standard deviation. Further, it has a high accuracy in predicting no-drought/drought condition (on average above 80%). The low standard deviation (0.0187) shows the consistency of the model performance under different climatic conditions. Table 13. Overall first model performance against 47 future climate scenarios The Best Drought Model Performance Log 10 (median flow in LPD*) No-drought/Drought RMSE R2 Accuracy Minimum 0.4940 0.8662 0.7689 Maximum 0.5407 0.8861 0.8470 Mean 0.5153 0.8750 0.8053 Standard deviations 0.0097 0.0040 0.0187 * LPD: Liter per day 98  4.4.5 The impact of climate change on future drought   The first model, which is the best model, was run for the 47 future climate scenarios for the period of 2040 to 2060 to understand the impacts of climate change on drought in the context of stream health. In addition, the status of drought conditions for the current period, 1990 to 2010, was evaluated to provide a reference condition in which future drought will be evaluated against. The future and current drought conditions were compared with each other using a cumulative distribution function (CDF) for all reaches within the Saginaw Bay Watershed, Figure 7. The probability of increasing drought conditions is categorized into three equal intervals. Reaches drawn as green show lower climb in drought probability while reaches drawn in red show higher probability of drought occurrences under future climate conditions. In general, the majority of the reaches (94%) will experience higher drought probability under future climate scenarios compare to current conditions. Specifically the mid-section of the watershed will experience the highest probabilities (66.68% to 100%) for future drought. In order to understand the possible causation for the increased drought in this region, the percent change maps for the two main drivers of drought (precipitation and temperature) were created (Figure 8). The percent change map of average temperature, Figure 8a, shows warmer temperatures are expected for all streams in the region due to an increase in average air temperature (2.3 °C). However, the region with the lowest increase in of precipitation (3.22% to 4.31%) was almost identical to the region of the worst future drought conditions (66.68% to 100%). This indicates that the stream system in the area of interest (high drought probability region) is predominately fed by surface runoff (sensitive to precipitation patterns), while the rest of the watershed is predominately groundwater fed.  99    Figure 7. Probability of increasing drought conditions under projected climate change (2040-2060) compare to current condition (1990-2010).   100  Figure 8. Percent change in (a) temperature and (b) precipitation from current (1980-2000) to future climate change (2040-2060).  (a)    (b)  101  4.5.  Conclusion Dracup, 2002), affecting both human and natural systems. Several drought indices have been developed to study the impacts of drought on the human dimension. However, this study is unique since the objective was to define a new drought index and predictive model as a proxy for stream health. The concept of the index flow was adapted from the Regional-scale Habitat Suitability Model to define the drought zones (Zorn et al., 2008). Initially, 66 variables were considered for development of the drought model. Variables were selected to develop two sets of models: 1) the most accurate drought model capable of determining current drought severity and 2) the most accurate drought forecast model capable of predating future drought severity. For the Current Drought Severity Models, the ReliefF algorithm was used to identify the top 15 ranked variables and for the Future Drought Severity Models, three models using some variables from 6, 12, and 18 months prior to the month of interest, respectively, were used. In general average flow rate and precipitation were among the highest ranked variables that were used for development of drought models.   Six drought models were developed using PLSR. The drought models were evaluated for median flow, classifying drought zones, and no-drought/drought conditions. Among the Current Drought Severity Models, the first model with five variables, having the highest R2 (0.86) and accuracy (0.74), was selected as the best; while the fourth model with 34 variables, having the highest R2 value (0.85) and accuracy (0.71), was selected as the best model among the Future Drought Severity Models. Overall, the first model was selected as the best model to predict drought severity.    102  The first model predictability was also tested using 47 future climate models, in which R2 values varied from 0.87 to 0.89. In addition, the average accuracy of the no-drought/drought ratio is 0.81. The future and current drought conditions were compared with each other using CDF curves and maps for all reaches within the Saginaw Bay Watershed. Approximately 7.2% of streams are expected to experience high increase (66.68% to 100%) in drought frequency under future climate scenarios. This region is highly correlated to the region that will experience the lowest increase in precipitation (3.22% to 4.31%), while the average temperature rises for the entire watershed is about 2.3 °C. This study introduced a new concept of evaluating the impact of drought as a proxy for the stream health. This is important since the introduced concept and modeling techniques can provide a road map for better allocation of resources to mitigate the impacts of the climate change on aquatic systems. The technique presented here is robust and transferable to other watersheds around the world. Due to limitations of the delta method, future studies should explore additional resources of climate data and methods for downscaling and bias correction in the development of the climate change ensemble. 4.6. Acknowledgments  This work is supported by the USDA National Institute of Food and Agriculture, Hatch project MICL02212.  Also, we would like to thank the climate modeling groups (listed in Table  of this paper) for producing and making their model outputs available. Additional thanks go to the World Climate Research Programme's Working Group on Coupled Modelling, which maintains the CMIP database and promotes the sharing of climate model outputs. Furthermore, we would like to thank the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison, which provides coordinating support and development of software 103  infrastructure in partnership with the Global Organization for Earth System Science Portals. The climate projections were developed with funding from the National Science Foundation under Grant BCS-0909378. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.    104  5. DEVELOPMENT AND EVALUATION OF A COMPERHENSIVE DROUGHT INDEX 5.1. Abstract Drousectors. Despite their wide range of impacts, no universal drought definition has been defined. The goal of this study is to define a universal drought index that considers drought impacts on meteorological, agricultural, hydrological, and stream heath categories. Additionally, predictive drought models are developed to capture both categorical (meteorological, hydrological, and agricultural) and overall impacts of drought. In order to achieve these goals, thirteen commonly used drought indices were aggregated to develop a universal drought index named MASH. The thirteen drought indices consist of four drought indices from each meteorological, hydrological, and agricultural categories, and one from the stream health category. Cluster analysis was performed to find the three closet indices in each category. Then the closet drought indices were averaged in each category to create the categorical drought score. Finally, the categorical drought scores were simply averaged to develop the MASH drought index. In order to develop predictive drought models for each category and MASH, the ReliefF algorithm was used to rank 90 variables and select the best variable set. Using the best variable set, the adaptive neuro-fuzzy inference system (ANFIS) was used to develop drought predictive models and their accuracy was examined using the 10-from R2 = 0.75 for MASH to R2 = 0.98 for the hydrological drought model. The results of this study can help managers to better position resources to cope with drought by reducing drought impacts on different sectors. 105  Keywords: Meteorological drought; Hydrological drought; Agricultural drought; Stream health drought; Drought monitoring; Drought predictive model 5.2.   Introduction Droughts are common and recurring phenomena affecting many sectors such as these sectors make it difficult to develop a universal/all-embracing definition of drought, since each sector measures drought differently (Whitmore, 2000; Heim, 2002). Drought definitions are generally categorized into meteorological, agricultural, hydrological, socioeconomic, and stream health (American Meteorological Society, 1997; Heim, 2002; Esfahanian et al., 2016). Meteorological drought is generally defined as a period of precipitation deficiency (several months or years) compared to a long term average (Whitmore, 2000; Heim, 2002; Mishra and Singh, 2010; Sheffield and Wood, 2011). The impacts of meteorological drought are a reduction in infiltration, runoff, deep percolation, and ground water recharge (NDMC, 2016). Agricultural drought is defined as a period of soil moisture deficiency resulting from precipitation shortage for a short period of time (few weeks duration) (Heim 2002; Sheffield and Wood, 2011). The impacts of agricultural drought are a reduction in crop biomass and yield, and plant growth (Heim, 2002; NDMC, 2016). Hydrological drought is defined as a period of deficiency in water supply due to prolong precipitation shortage (Heim, 2002). The impacts of hydrological drought are a significant reduction in streamflow, groundwater, reservoir, and lake levels (Whitmore, 2000; Heim, 2002; NDMC, 2016). The concept of socioeconomic drought, which is not the subject of this study, is based on the impacts of meteorological, agricultural, and hydrological droughts on the supply and demand of some economic goods (Heim, 2002; NDMC, 2016). 106  Finally, stream health drought is defined as a period of deficiency in streamflow causing irreversible impacts on aquatic ecosystems (Esfahanian et al., 2016).   Several drought indices have been developed to monitor and quantify drought. Drought indices are primarily tools to investigate drought duration, intensity, severity, and spatial extent (Mishra and Singh, 2010). Each drought index requires specific input parameters in order to measure drought. Precipitation is usually used alone or in combination with other parameters for this matter (Heim, 2002; Mishra and Singh, 2010; Sheffield and Wood, 2011). Usually for meteorological drought, precipitation is the primarily parameter (Dai, 2010). For agricultural drought, soil moisture content is commonly used with the secondary parameters of precipitation and/or evapotranspiration (Dai, 2010). For hydrological drought, streamflow is often used beside precipitation (Dai, 2010). Finally for stream health drought, index flow, stream size, and stream temperature are used to capture fish vulnerability to drought. The index flow is defined as the median of the summer month with the lowest daily flowrate for the given period (Hamilton and Seelbach, 2011; Esfahanian et al., 2016).  Despite the current progress in understanding the science behind droughts, there is still a need to improve drought monitoring methods, which will ultimately improve drought preparation and management practices, and reduce drought vulnerability on different sectors (Svoboda et al., 2002). One way to improve drought-monitoring techniques is to combine the existing indices to better capture the overall impacts of drought (Zargar et al., 2011) because results from categorical droughts can be significantly different, which can be both confusing and misleading. In general, the methods used for combining drought indices can be classified as: 1) decision matrix analysis (Svoboda et al., 2002; Balint et al., 2011; Zieses et al., 2014); 2) classification 107  and regression tree (CART) analysis (Tadesse and Wardlow, 2007; Brown et al., 2008); and 3) regression technique (Keyantash and Dracup, 2004; Karamouz et al., 2009).  In the decision matrix analysis, multiple criteria are first identified to guide the final outcome. This technique was used by Svoboda et al. (2002) to create the Drought Monitor, which is a composite of meteorological drought indices (such as Palmer Drought Severity Index and Standardized Precipitation Index), and hydrologic and remote sensing information. The relationship between the Drought Monitor components and drought severity were defined using the decision matrix analysis (Scoboda et al., 2002). Additionally, the Combined Drought Index (CDI) was introduced by Balint et al. (2011), which is the combination of the Precipitation Drought Index (PDI), Temperature Drought Index (TDI), and Vegetation Drought Index (VDI). The weighted average of the PDI, TDI, and VDI indices were used to compute the CDI. The assigned weight for the PDI was 50% and 25% weight was assigned for each TDI and VDI indices (Balint et al., 2011). Zieses et al. (2014) developed the Global Precipitation Climatology Centre Drought Index (GPCC-DI) with 1° grid spatial resolution, which is a combination of the Modified Standardized Precipitation Index (SPI-DWD) and Standardized Precipitation Evapotranspiration Index (SPEI). The GPCC-DI is calculated by taking the average of SPI-DWD and SPEI indices for each grid cell (Zieses et al., 2014).  The CART analysis is a tree-building technique, which constructs a set of decision rules to build predictive models. This technique was used by Tadesse and Wardlow (2007) to develop the Vegetation Outlook (VegOut) to predict future vegetation conditions. In this tool meteorological drought indices (Standardized Precipitation Index and Palmer Drought Severity Index), oceanic indices (such as Southern Oscillation Index, and Multivariate El Niño and Southern Oscillation Index), and satellite and biophysical data were combined using a rule-based 108  regression tree method. A year later, Brown et al. (2008) introduced a new index named Vegetation Drought Response Index (VegDRI) based on the CART concept. In this index, meteorological drought indices (Standardized Precipitation Index and Palmer Drought Severity Index), satellite-based vegetation measures, and biophysical information (such as land cover and soil available water capacity) were combined using CART analysis in order to develop the VegDRI empirical models for different seasons.  The regression technique estimates the linear and nonlinear behavior between the dependent and independent variables. This technique was used by Keyantash and Dracup (2004) to develop an Aggregate Drought Index (ADI) that considers meteorological, hydrological, and agricultural categories of drought. In this index, six hydrologic variables including precipitation, streamflow, reservoir storage, evapotranspiration, soil moisture, and snow water content were aggregated using principle component analysis (Keyantash and Dracup, 2004). In addition, the Hybrid Drought Index (HDI) was developed by Karamouz et al. (2009) using this technique. This index is a combination of the Standardized Precipitation Index, the Palmer Drought Severity Index, and the Surface Water Supply Index (Karamouz et al., 2009). An artificial neural network technique was used to predict the HDI based on the three drought indices (Karamouz et al., 2009).   Given the lack of a universal drought definition in monitoring drought, the goal of this study is to introduce a universal drought definition that considers several aspects of drought including meteorological, agricultural, hydrological, and stream health. This universal definition can improve drought monitoring, which can help decision makers to better allocate the resources to reduce drought impacts on different sectors. The objectives of this study are to: (1) define categorical drought indices (meteorological, agricultural, and hydrological) based on commonly 109  used drought indices; (2) define a universal definition of drought by combining the categorical scores; (3) select the best variable sets to construct predictive drought models; (4) develop predictive drought models for each drought category and the universal drought index.  5.3.  Materials and Methodology 5.3.1. Study area The Saginaw River Watershed is the largest watershed in Michigan, and is located in the eastern part of central Michigan (Figure 1). The watershed has a total area of 16,122 km2 and drains into Lake Huron. This area has a population of about 1.4 million in 22 counties (PSC, 2012). The dominant landuse is agricultural and forest lands (37% each), and the remaining landuses are pasture (9.5%), urban (7.5%), wetlands (8%), and water (1%). There are 145 subbasins in the Saginaw River Watershed, with the majority of them being warmwater streams. The elevation in the watershed ranges from 328 m to 176 m above mean sea level. The region consists of four different hydrologic soil groups A (24%), B (58%), C (16%), and D (1%). This watershed has been one of the most studied regions in the Great Lakes having high diversity in flora and fauna, agriculture, and recreational opportunities (Selzer et al., 2014). From the meteorological standpoint, Saginaw River Watershed has an average annual precipitation of 816 mm, and an average annual temperature of 9 °C, which is very similar to the State average values (U.S. climate data, 2016). From the agricultural standpoint, this region is one of the most productive agricultural regions in Michigan in which corn, soybean, and sugar beets are the main crops (U.S. Department of Agriculture, National Agricultural Statistical Service, 2011; Selzer et al., 2014). From the hydrological standpoint, this region is a rich resource for recreation activities such as walleye fisheries (Selzer et al., 2014). The value of recreational activities in this region is around $15.9 million annually (Selzer et al., 2014). From the stream health standpoint, the 110  Saginaw River Watershed provides habitats for more than 90 fish spices and it has the largest contiguous freshwater costal wetland in the United States (PSC, 2012; Selzer et al., 2014). However, the rapid industrial and population growth in the 20th century have caused significant ecosystem degradation in this region (Selzer et al., 2014). Some segments and the outlet of the watershed have been designated as a Great Lakes area of concern by the US Environmental Protection Agency due to degraded fisheries, fish consumption advisories, loss of recreational values, and sediment contamination (USEPA, 2015).    Figure 9. Saginaw River Watershed 111  5.3.2. Modeling process The goal of this study is to define a new combined drought index which considers different aspects of drought including meteorological (M), agricultural (A), stream health (S), and hydrological (H). This new drought index is named MASH.  The process started by calculating 13 drought indices for all subbasins within the Saginaw River watershed. For each drought category, except stream health for which only one drought index exists, four commonly used drought indices were selected. The modeling process consists of two phases: The Categorical Drought Index Development phase in which the overall drought index is defined for each drought category and MASH, and the Drought Model Development phase in which predictive models are developed to estimate the categorical drought indices and MASH (Figure 2). In the Categorical Drought Index Development phase, all 13 drought indices are calculated on a monthly basis for 145 subbasins within the study area over a 34-year-period (1979-2012). In order to make the indices comparable within each category, the value of each index was classified and then normalized using a linear scaling technique. Next within each category (meteorological, agricultural, and hydrological) cluster analysis was performed to calculate the categorical drought index based on the average value of the closest three out of four indices. The MASH index then was calculated by averaging the categorical drought indices.   In the Drought Model Development phase, the ReliefF algorithm was used for ranking the input variables for the drought predictive models (three categorical and one MASH).  For each drought predictive model, all combinations of the top two and three variables (out of the top five ranked variables) were used for the model development using the adaptive neuro-fuzzy 112  inference system (ANFIS). The accuracy of the developed models were then validated using the 10-fold cross validation technique.      Figure 10. Categorical drought scores development and modeling process 5.3.3. Categorical drought index development  The goal of this phase is to define the overall drought scores for each drought category (meteorological, agricultural, hydrological, and stream health). Therefore, in the first step, we    113  will define the common drought indices within each drought category, which will be further used for development of the categorical and universal drought indices.  5.3.3.1 Meteorological Drought Indices Palmer Drought Severity Index (PDSI), Rainfall Deciles (RD), Standardized Precipitation Index (SPI), and Reconnaissance Drought Index (RDI) were the four meteorological indices selected for this study. These indices are commonly used to monitor meteorological drought (Keyantash and Dracup, 2002; Hayes, 2006; Dai, 2010; Sheffield and Wood, 2011; Moorhead et al., 2015). The reference, input parameters, and description of each of these indices are presented in Table S9. All of these indices use precipitation as the input parameter to monitor drought (Hayes, 2006). The PDSI uses additional parameters such as temperature, available water content, and solar radiation while the SPI and RD only use precipitation and the RDI uses potential evapotranspiration in addition to precipitation.  Advantages/disadvantages: The RD, SPI, and RDI are computationally less complex compared to PDSI, since the latter needs a greater number of parameters for calculation (Hayes, 2006). The PDSI considers evaporation by comparing the actual soil moisture (precipitation plus available water content) to the soil moisture demand of a region (potential evapotranspiration) (Heim, 2002; Dai, 2010). On the other hand, the RD and SPI do not consider evaporation (Dai, 2010). The RDI is more comprehensive compared to the SPI, due to considering the balance between the input (precipitation) and the output (potential evapotranspiration) (Tsakiris and Vangelis, 2005; Zargar et al., 2011).  RD, SPI, and RDI require long-term precipitation data; however, the PDSI does not need long-term climatological data (Dai, 2010). The SPI and RDI can be measured for different time scales such as 1-month, 3-months, 6-months, up to 48 months, yet RD and PDSI cannot be used for different time scales (Keyantash and Dracup, 2002; 114  Tsakiris and Vangelis, 2005; Dai, 2010). Based on several studies conducted on drought indices, it was concluded that the SPI is more preferable compared to PDSI (Guttman, 1998; Hayes et al., nd clear assessment of drought intensity, duration, and spatial extent (Hayes et al., 2000; Zargar et al., 2011). However, the PDSI was indicated to be very complex and difficult to interpret (Zargar et al., 2011). This was supported by another study that evaluated the performance of meteorological drought indices based on six criteria of robustness, tractability, transparency, sophistication, expendability, and dimensionality (Keyantash and Dracup, 2002). The results of this study indicated that among all meteorological drought indices, the SPI and RD have the highest rank (the best among studied indices), and the PDSI has the lowest rank (Keyantash and Dracup, 2002; Zargar et al., 2011). The RDI was not included in this comparison.  5.3.3.2 Agricultural Drought Indices For this study, the Palmer Moisture Anomaly Index (Z-index), Soil Moisture Deficit Index (SMDI), Evapotranspiration Deficit Index (ETDI), and Soil Water Deficit Index (SWDI) were selected as the agricultural drought indices. The Z-index is one of the most widely used indices for capturing agricultural drought (Dai, 2011). The SMDI and ETDI have high spatial and temporal resolutions in monitoring agricultural drought (Narasimhan and Srinivasan, 2005). The SWDI was recently introduced, and uses soil water observations to analysis agricultural drought (Martinez-Fernandez et al., 2015). The reference, input parameters, and description for each of these indices are presented in Table S10. The SMDI only uses the soil moisture, and the ETDI only uses evapotranspiration as the input parameters to estimate drought condition (Narasimhan and Srinivasan, 2005). In addition, the SWDI only uses soil water storage 115  parameters such as soil moisture, available water content, field capacity, and wilting point for agricultural drought assessment (Martinez-Fernandez et al., 2015).   Advantages/Disadvantages: The Z-index is more computationally intensive compared to the other three indices, since it considers precipitation, temperature, available soil water content and other parameters (a total of eight variables) in its calculation (Palmer, 1965; Jacob et al., 2013; Ficklin et al., 2015). In contrast, the SMDI and ETDI only use soil moisture and evapotranspiration (potential and actual), respectively (Narasimhan and Srinivasan, 2005). However, the challenging part for calculating the SMDI will be acquiring soil moisture data (Moorhead et al., 2015). The SWDI is more comprehensive compared to SMDI in capturing soil moisture deficit, since it uses several soil parameters while the SMDI only uses one parameter (soil moisture). The spatial and temporal resolutions of SMDI and ETDI are higher compared to the Z-index. The Z-index has a coarse spatial resolution of 7,000 to 100,000 km2 and monthly temporal resolution (Narasimhan and Srinivasan, 2005). On the other hand, the SMDI and ETDI have high spatial resolution of 16 km2 and a weekly temporal resolution (Narasimhan and Srinivasan, 2005). The SMDI and ETDI with finer spatial and temporal resolutions can improve monitoring soil moisture and evapotranspiration deficits compare to the Z-index. 5.3.3.3 Hydrological Drought Indices In this study the Palmer Hydrological Drought Index (PHDI), Flow Duration Curve (FDC), Standardized Runoff Index (SRI), and Water Balance Derived Drought (WBI) were used as the hydrological drought indices. The PHDI is one of the most commonly used indices to monitor hydrological drought (Dai, 2011) while the FDC, SRI, and WBI were more recently developed (Tallaksen and van Lanen, 2004; Shukla and Wood, 2008; Vasiliades et al., 2011). The reference, input parameters, and description for each of these indices are presented in table 116  S11. These new indices only use runoff to monitor hydrological drought (Tallaksen and van Lanen, 2004; Shukla and Wood, 2008; Vasiliades et al., 2011). However, the PHDI is more computationally intense, since it uses eight different climatological data for calculating hydrological drought (Palmer, 1965; Jacob et al., 2013; Ficklin et al., 2015). Advantages/Disadvantages: The SRI and WBI need long-term historical data to be reliable for drought monitoring. However, the PHDI and FDC do not require long-term streamflow data. The normalization approach is used to calculate the SRI and WBI indices, which is similar to the SPI index (Shukla and Wood, 2008; Vasiliades et al., 2011). The SRI and WBI use different distributions in order to normalize the runoff data. The SRI uses a log normal distribution while the WBI uses the Box-Cox transformation (Shukla and Wood, 2008; Vasiliades et al., 2011). Then the transformed data are standardized into a normal distribution with a mean of zero and standard deviation of one (Shukla and Wood, 2008; Vasiliades et al., 2011). However, for the FDC calculation the threshold level approach is used (Tallaksen and van Lanen, 2004; USEPA, 2011a). In this approach, the threshold levels are defined based on a specific percentile flow for a certain period of time. The streamflow data for the selected time intervals are ranked, and their exceedance probabilities are calculated (Tallaksen and van Lanen, 2004; USEPA, 2011a). Then based on the defined thresholds, the wet and dry condition of the region can be determined. 5.3.3.4 Stream Health drought Index In contrary with other drought indices, only one index was defined for the stream health drought. The reference, input parameters, and description for the stream health drought index is presented in Table S12. This index uses long-term median and average flow rate to monitor stream health drought (Esfahanian et al., 2016). The index was developed based on the concept 117  of the regional-scale habitat sustainability model (Zorn et al., 2008; Hamilton and Seelbach, 2011), which predicts the effect of flow reduction on fish assemblages during the low flow period that is the most stressful for the fish assemblage. In the regional-scale habitat sustainability model, an index flow was defined as a threshold to evaluate the proportion of fish assemblage reduction (Zorn et al., 2008). The index flow is the median of the daily flow rate values for the summer month (July, August, and September) with the lowest average flow rate within a given period (Hamilton and Seelbach, 2011). The stream health drought conditions were defined considering different percent of index flow reduction, stream size, and stream temperature (Zorn et al., 2008; Esfahanian et al. (2016). The general associated ranges of drought classes are presented in Table S4. The index flow was calculated for each stream in the study area. Then it was multiplied by the general associated drought ranges to obtain the specific drought class for each stream on a monthly basis.  The stream health drought index model uses monthly median flows and average flow rates as the input data to predict the stream health drought condition for the future month based on the river continuum concept (Vannote et al., 1980; Esfahanian et al., 2016). The input data to the stream health drought index model can either be obtained directly by monitoring (e.g. United States Geological Survey-National Water Information System) or indirectly through hydrological modeling (e.g. using the Soil and Water Assessment Tool-SWAT). However, the initial calculation of the index flow required long-term flow data and understanding of the stream habitat that can limit the use of this technique in regions where rich datasets are not available. 5.3.4. Input parameters  The name, source, and description of all parameters used to estimate the 13 aforementioned drought indices are presented in Table S13. In general, seven different sources 118  were used to obtain the input data for drought indices calculation. The precipitation and temperature data were obtained from the National Climatic Data (NCDC) stations (16 precipitation and 13 temperature stations). The National Elevation Dataset of the US Geological Survey (USGS) with a spatial resolution of 30 m was used to obtain elevation data (NED, 2014). The soil characteristic data such as available water content were obtained from the Natural Resources Conservation Service (NRCS) Soil Survey Geographic (SSURGO) database (NRCS, 2014). The data warehouse provided by Abatzolgou (2013) was used to obtain solar radiation, wind speed, and relative humidity data. The average annual albedo data were obtained from Barkstrom (1984). The intermediate palmer parameters such as potential evapotranspiration, and index of drought severity were obtained from the MATLAB tool developed by Jacob et al. (2013) and modified by Ficklin et al. (2015). The remaining hydrological and climatological parameters such as actual and potential evapotranspiration, soil moisture, field capacity, available water content, and streamflow were obtained from a hydrological model (SWAT), which is developed by the USDA Agricultural Research Service (USDA-ARS). This physically based model was calibrated and validated using observed monthly streamflow data for nine USGS gauging stations. The calibration period was from 2001 to 2005, and the validation period was from 2006 to 2010. Three statistical methods were used to evaluate the model calibration and validation performances. These methods are Nash-Sutcliffe efficiency (NSE), root mean square error observations standard deviation ratio (RSR), and percent bias (PBIAS). The model (Moriasi et al., 2007). The calibration and validation information for each USGS streamflow gauging station is provided in Table S15. The locations of each USGS streamflow gauging stations are presented in Figure S8.  119  5.3.5. Transformation and Clustering As shown in Tables S9 to S12, each drought index has different associated ranges, and classifications of drought magnitude. Some of the indices are categorized into more detailed classes compared to others that have a boarder classification of wet and dry conditions. For instance, the PDSI is classified into 11 wet and dry categories. However, the SPI is classified into seven dry and wet categories. In order to make the indices comparable, a similar classification should be used. In this study, four drought categories including initial, moderate, severe, and extreme drought were identified and associated ranges were assigned to them (Table S16). Similarly, four non-drought categories including initial, moderate, severe, and extreme wet conditions were identified and associated ranges were assigned to them (Table S17). In order to obtain the overall drought score for each category and MASH, the drought indices were normalized to become comparable (Tables S16 and S17). The linear scaling technique was used to assign eight ranges (-100 to <-75,-75 to <-50,-50 to <-25,-25 to <0, 0 to <25, 25 to <50, 50 to <75, and 75 to 100) based on the defined classification (initial, moderate, severe, and extreme). Based on the normalized ranges, the calculated values for each drought index were transformed into a number between -100 and 100.  The following equations (1 to 8) were used to normalize each drought index: Initial Drought   (1)  Moderate Drought                                    (2) Severe Drought  (3) Extreme Drought  (4) Initial Wet    (5)  Moderate Wet                                     (6) 120  Severe Wet   (7) Extreme Wet   (8)  where,  is the normalized drought/wet index value,  is the initial drought/wet index value,  to  are the associated range of the initial drought/wet category,  to  is the associated range of the moderate drought/wet category,  to  are the associated range of the severe drought/wet category, and  to  are the associated range of the extreme drought/wet category. In order to obtain the categorical drought scores for each subbasin, cluster analysis was performed. The cluster analysis allows for identifying a more collective drought score since there is not a universal definition of drought within each category. The cluster analysis finds the closest three indices out of four within each category (except the stream health index) for each month and then finds the average of three closest indices. In the case that there is a tie between two sets of three indices, the average of four indices is calculated.  5.3.6. Aggregation  Since no preference was considered for each drought category, the simple averaging method was used to calculate the MASH score (equation (9)) for each month over the 34 year period:   where, MASH is the overall score of all four categorical drought scores, CAI  is the categorical agricultural score, CHI  is the categorical hydrological score, CSHI  is the categorical stream health score, CMI is the categorical meteorological score. The MASH score is a number between -100 and 100.  121  5.3.7. Drought indices comparison A paired T-test is frequently adopted for testing the difference between paired observations for two variables. However, the usual T-test assumes the samples are independent and normally distributed. The independence assumption can be violated due to the fact that observations within the same location (subbasin) can be correlated. Similarly, the observations during the same recorded time (year/month) can be correlated. Therefore, we consider a model-based approach to test the mean difference by adjusting for such correlations. We use a linear mixed-effects model (Pinheiro and Bates, 2006) with two nonnested random effects on location and recorded time respectively to model the difference:  where,  is the observed difference between two variables for subbasin  and  recorded time (month) from 1979 to 2012. The parameter  represents the grand mean of the difference .  is a random effect on the subbasin to account for the correlation between observations within the same subbasin, with its dispersion parameter measuring the variability due to subbasin.  is a random effect on the recorded time to account for the correlation between observations within the same recorded time, with its dispersion parameter measuring the variability due to recorded time.  is the residual with its dispersion parameter measuring the unexplained variability. By adjusting for the correlations, our goal is to test the null hypothesis  versus the alternative hypothesis . The associated p-value is calculated based on Satterthwaite's approximations for the degree of freedom and is implemented in the R package lmerTest (Kuznetsova et al., 2016). We applied the method to each pair of drought indices.  122  5.3.8. Drought model development In this phase, drought predictive models were developed for three of the drought categories (metrological, hydrological, and agriculture) and MASH. It is important to note that no predictive drought model is need for stream health since: 1) only one index was defined for this category; and 2) the stream health index can be directly reported from the observed streamflow data. In order to select the best variable set for the drought models, the ReliefF algorithm was used. Then the top five ranked variables were incorporated into the ANFIS. Finally, the 10-fold cross validation technique was used to determine the validity of the predictive models. 5.3.8.1. Parameter selection The ReliefF algorithm is a commonly used feature selection method, which is capable of handling data with strong dependencies (Kononenko, 1994; Robnik-Sikonja and Kononenko, 2003). This method is the improved version of the Relief algorithm enabling feature selection for numerical datasets (Kira and Rendell, 1992b; Kononenko, 1994; Robnik-Sikonja and Kononenko, 2003). In this method, the independent variables are ranked based on their relevance in predicting the dependent variable (Kononenko, 1994; Robnik-Sikonja and Kononenko, 2003). With k being the number of each neighborhood samples, this algorithm searches for k of the nearest neighbors of the same class, also known as nearest hits, and k of the nearest neighbors of the different class, also known as nearest misses, for each sample. Therefore, there will be k nearest hits and k nearest misses for each sample. The relevance of the variables for all samples is defined using the following equation:          (10) 123  where,  is the weight of the feature,  is the number of each neighborhood samples,  is the i-th sample,  is the k nearest hits to , and  is the k nearest misses to .  The input data, categorical meteorological, agricultural, hydrological, and the MASH scores were used in the ReliefF algorithm to rank the best variable set. The top five ranked variables were used to develop the predictive drought models.  5.3.8.2. Development of predictive drought models The categorical and MASH predictive drought models were created using the Sugeno-type fuzzy inference system (Takagi and Sugeno, 1985). The Sugeno-type fuzzy inference system has been widely used in modeling complex environmental and ecological systems, water resource problems, and drought forecasting (El-Sebakhy et al., 2007; Kisi et al., 2006; Bacanli et al., 2009; Einheuser et al., 2013, Hamaamin et al., 2013; Woznicki et al., 2015; Woznicki et al. 2016). In this technique, graphical membership functions (MFs) are used to represent the degree of membership of the input variables. Degree of memberships of zero and one represent no and full membership, respectively (Kaehler, 2006).  There are some challenges associated to modeling with fuzzy logic such as defining the membership function parameters and designing fuzzy rules (Bacanli et al., 2009). Due to these limitations, the ANFIS method was developed to improve the development of membership functions. ANFIS is a combination of fuzzy logic and artificial neural network (ANNs) methods which has the benefits of both methods in one framework (Bacanli et al., 2009). This multi-layer network uses ANNs to create MFs and minimize the output errors to be used in fuzzy logic (Jang, 1993).  124  The Fuzzy Logic Toolbox in MATLAB R2015b was used to develop ANFIS models (MathWorks, 2016). Five membership function shapes in combinations of 2, 3, or 4 were tested for each variable. The membership function shapes are triangular, trapezoidal, generalized bell, Gaussian, and Gaussian composite. The first two functions are linear and the remaining functions are nonlinear and fit better for ecological data (Marchini et al., 2011). Furthermore, there are two possible outputs for the membership functions, linear and constant. All possible combinations of two and three sets of variables out of top five ranked were used to create the predictive models. Information describing all of the possible combinations are presented in Table 14. As a result, a total of 3,600 models were created for the three drought categories and MASH adding up to 14,400 models.  The 10-fold cross validation technique was used to train, test, and select the best ANFIS model. The dataset is randomly and equally divided into 10 exclusive subsets (folds) in the 10-fold cross validation. Nine folds of the data are used for training (90%) and the remaining one fold is used for testing (10%). This process was repeated 10 times and each time the fold used for testing was substituted with one of the folds used in model training. Therefore, in this process the total of 144,000 models were trained and tested in order to select the best ANFIS models for the three drought categories and MASH. The final selection is based on the lowest Root Mean Square Error (RMSE) of the 10-fold cross validation. In case of a tie in RMSE,  was used as tiebreaker.  Table 14. ANFIS models frameworks and characteristics Number of input variables Possible combinations of input variables Number of membership functions  Membership function shapes Output membership function Sum of combinations 2 10 32 5 2 900 3 10 33 5 2 2,700 125  5.4.   Results and Discussions 5.4.1 Statistical Analysis of Drought Indices As discussed in the introduction section, there is no universal drought definition even within each drought category. This means that different drought indices, even in the same category (e.g. meteorological) can report diverse level of drought severity. In order to test this hypothesis, four commonly used drought indices in each category (methodological, hydrological, and agricultural) were tested using a linear mixed-effects model (Pinheiro and Bates, 2006). This model tests the mean difference between each pair of drought indices within each category. The results of this statistical analysis for each drought category are presented in Table 15. Each number in this table indicates the p-value between each pair of indices. p-values larger than 0.05 (in red) show no significant mean differences.  In the meteorological category, none of the indices had a significant mean difference. The similarity between the meteorological indices can be due to having similar approaches in monitoring meteorological drought. In most cases, a long-term historical precipitation record is used to calculate drought severity. In the hydrological category, only the SRI and WBI indices have a significant mean difference. This difference can be explained by examining the different normalization approaches used to calculate the hydrological drought for each index. SRI fits the historical runoff records into a log normal distribution, and then transforms it a standardized normalized distribution. However, the WBI uses the Box-Cox for transforming of the historical runoff records, and then transforms it into a standardized normalized distribution. In the agricultural category, the SMDI and ETDI are the only indices that have no significant mean difference. The similarity between the SMDI and ETDI can be due to using the same crop growth model in their calculations. Overall, in the meteorological and hydrological categories 126  most of the pairs showed similar behavior. However, in the agricultural category only one pair out of six pairs showed similar behavior.  For all drought indices, the mean difference and standard deviation values are presented in Table S14. For the meteorological indices (PDSI, RD, SPI, and RDI), the mean differences are small ranging from -1.32 to 1.29; however, the standard deviations are large and ranging from 10.72 to 58.97. Similarly, for the hydrological indices (PHDI, FDC, SRI, and WBI), the mean differences are small ranging from -0.06 to 1.97; and the standard deviations are large and ranging from 8.81 to 47.54. Therefore, despite the fact that for both meteorological and hydrological indices the long-term averages can be very similar, the results for individual events can be quite different. Finally, for the agricultural indices (Z-Index, SMDI, ETDI, and SWDI), the mean differences and standard deviations are both large ranging from -5.56 to 67.12, and 34.53 to 44.13, respectively. This indicates that there is a large contradiction between the agricultural indices for both long-term averages and individual events.   Regarding the categorical drought indices, CMI is not significantly different from other meteorological indices. While the mean varies from -0.15 to 1.14, the standard deviation is large ranging from 14.45 (RDI) to 50.34 (RD). The categorical hydrological index (CHI) does not have any significant mean difference with the PHDI, FDC, and SRI since the mean difference ranges of 0.12 to 1.56, and the standard deviation ranges from 9.35 to 36.34. However, similar to the CMI, the standard deviation is large, which can be misleading. In fact with each drought level having a range of 25 the CHI could be off by up to two drought levels. Comparing the CHI to with all other categorical and drought indices, no significant mean difference was observed with any of the meteorological indices and the Z-index. The meteorological indices have a mean difference range of 0.03 to 1.36, and a standard deviation range of 32.69 to 44.52 with the CHI. 127  In addition, the Z-index mean difference and standard deviation with the CHI are 0.73 and 32.12, respectively. Therefore, it can be concluded that CHI can be a good alternative to both meteorological and hydrological drought indices but caution should be exercised due to the possibility of large standard deviation. The categorical agricultural index (CAI) has no similarity with any of the agricultural indices or other drought indices. This implies that diversity in defining agricultural indices are much larger compare to hydrological and metrological drought categories. Finally, the categorical stream health index (CSHI) has no similarity with any of the drought indices, which implies that the CSHI should be separately calculated and combined with existing indices to capture the overall drought condition.  5.4.2 Categorical Drought Indices As described in the methodology section, cluster analysis was used in order to define a universal drought definition for each drought category. The three closest indices out of four in each drought category (methodological, hydrological, and agricultural) were identified and averaged for each month for over 30 years in the Saginaw River Watershed. In the case that there was two sets of three indices that had equal means, a set of four indices were selected and averaged. The results of this analysis are summarized in Table 16. What is unique about this analysis is that in contrary with similar studies (Scoboda et al., 2002; Karamouz et al., 2009) the indices were not combined to develop a new index for each category; rather the most common drought definition in each category was identified by averaging the closet drought scores. This helped define the near universal drought index known as the categorical drought index.  In the meteorological category, the combination of PDSI, SPI, and RDI was identified as the most selected combination in 62.7% of the time. Moreover, the combination of all four indices was the least selected combination (0.02%). In the hydrological category, the 128  combination of FDC, SRI, and WBI was identified as the most selected combination (49.68%). The PHDI, SRI, and WBI combination was selected as second with a small difference (7.96%) from the first ranked set. The least selected combination was the all four indices combination (0.01%). Finally, in the agricultural category, the Z-index, SMDI, and ETDI combination was identified as the most selected combination for 69.44% of the time. The Z-index, ETDI, and SWDI combination was selected as a distance second (12.21%) and the remaining combinations were selected about 9% of the time. The combination of all four indices was not selected at all. Therefore, it can be concluded that the RD, PHDI, and SWDI are the most different indices in meteorological, hydrological, and agricultural categories, respectively.  5.4.3 Comparison of Categorical Drought Scores and MASH The linear mixed-effects model was used to evaluate the mean difference between categorical (CMI, CHI, CAI, CSHI) and MASH scores. The results indicate that among the categorical drought scores, only CMI and CHI do not have a significant mean difference with each other (Table 15). However, the standard division is large (29.98), which reduces the reliability of using these indices interchangeably. Additionally, the MASH index did not show similar behavior to any of the categorical drought indices. This implies that the MASH is not biased toward any of the categorical drought indices while representing the overall drought conditions.    129  Table 15. p-values from pairwise comparison of drought indices. Red colored p-values indicate no significant mean differences at the 0.05 level.   Drought Index   Meteorological Hydrological Agricultural Stream Health Overall PDSI RD SPI RDI CMI PHDI FDC SRI WBI CHI Z-Index SMDI ETDI SWDI CAI CSHI MASH PDSI                                  RD  0.43                 SPI  0.98 0.58                RDI  0.96 0.59 0.94               CMI 0.85 0.56 0.78 0.81              PHDI  0.03 0.91 0.39 0.39 0.32             FDC  0.95 0.48 0.98 0.99 0.94 0.33            SRI  0.88 0.45 0.93 0.95 0.96 0.23 0.95           WBI  0.61 0.26 0.66 0.66 0.45 0.08 0.58 0.003          CHI  0.97 0.38 0.95 0.94 0.81 0.17 0.88 0.39 0.09         Z-Index  0.46 0.33 0.22 0.19 0.052 0.11 0.56 0.36 0.75 0.47        SMDI  6e-06 0.03 0.003 0.003 0.0003 0.008 0.0004 4e-06 0 1e-06 4.1e-05       ETDI  0.04 0.47 0.04 0.04 0.02 0.43 0.11 0.02 0.003 0.02 0.003 0.08      SWDI  0 0 0 0 0 0 0 0 0 0 0 0 0     CAI 3e-06 0.002 2e-05 1.5e-05 0 2e-06 0.0004 0 1e-06 0 2e-06 0 0 0    CSHI 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0   MASH  0 8.6e-05 0 0 0 0 0 0 0 0 0 0.0006 6e-06 0 0 0  130  Table 16. Frequency of drought indices combinations in each drought category over 30-year period           Combinations ranking       Meteorological Indices      Hydrological Indices          Agricultural Indices PDSI RD SPI RDI Frequency (%) PHDI FDC SRI WBI Frequency (%) Z-index SMDI ETDI SWDI Frequency (%) First x  x x 62.70  x x x 49.68 x x x  69.44 Second  x x x 26.16 x  x x 41.71 x  x x 12.21 Third x x  x 5.94 x x x  4.92 x x  x 9.86 Fourth x x x  5.18 x x  x 3.68  x x x 8.49 Fifth x x x x 0.02 x x x x 0.01 x x x x 0.00 131  5.4.4  Variable Selection The ReliefF algorithm was used to rank the best variable set for each predictive drought model. The top five selected variables for each drought category and MASH are presented in Table 17. The best two and three variable combinations of the top five ranked variables for the ANFIS predictive drought models were also identified in Table 17. As it was expected, the results indicate that variables related to precipitation data were ranked the highest for the meteorological category. The Gamma-precipitation, Gamma P-PET, and precipitation percentile are intermediate variables used to calculate the SPI, RDI, and RD, respectively. The Gamma-precipitation was obtained from fitting the monthly precipitation data into a gamma distribution. The Gamma P-PET was obtained from fitting the monthly precipitation divided by the monthly potential evapotranspiration data into a gamma distribution. And the precipitation percentile was obtained from ranking the cumulative precipitation of three months before the month of interest. For the hydrological category, the variables obtained from streamflow were mostly ranked as the highest variables. The streamflow exceedance probability, log-normal streamflow, and severity index for a wet/dry spell are used to calculate the FDC, SRI, and Palmer index, respectively. The streamflow exceedance probability was obtained from ranking monthly average streamflow data. The log-normal streamflow was obtained from fitting the monthly streamflow data into a log normal distribution. And the severity index for a wet/dry spell was obtained from the Ficklin et al. (2015) MATLAB code. For the agricultural category, the variables obtained from evapotranspiration deficit were ranked the highest most often. The monthly water stress anomaly, monthly soil moisture deficit, and Gamma P-PET are used to calculate the ETDI, SMDI, and the RDI, respectively. The monthly water stress anomaly was calculated from actual and potential monthly evapotranspiration. The monthly soil moisture deficit was obtained using 132  average, maximum and minimum monthly soil moisture. And the Gamma P-PET is the same variable that was selected for the meteorological category.  Finally for MASH, streamflow related variables were ranked the highest most often. The log-normal streamflow and streamflow exceedance probability were both selected as the top variables in the hydrological category as well. And the precipitation percentile variable was also selected as the top ranked variable in the meteorological category. Overall, it can be concluded that precipitation, streamflow, and evapotranspiration variables have a high influence on meteorological, hydrological, and agricultural drought, respectively. Meanwhile for MASH, the streamflow variables have the highest influence on determining the overall drought.  Table 17. Top five ranked variables that were used for development of the drought predictive models. Category Ranked Variables Meteorological 1 Gamma-precipitation 2 Gamma P-PET*,** 3 Precipitation percentile** 4 Dry/wet spell severity index*,** 5 Precipitation Hydrological 1 Streamflow exceedance probability** 2 Log-normal streamflow*,** 3 Severity index for an established wet/dry spell*,** 4 Dry/wet spell severity index 5 Precipitation percentile Agricultural 1 Monthly water stress anomaly*,** 2 Monthly soil moisture deficit*,** 3 Gamma P-PET** 4 Gamma-precipitation 5 Precipitation percentile MASH 1 Log-normal streamflow*,** 2 Streamflow exceedance probability 3 Precipitation percentile  4 Gamma-precipitation*,** 5 Monthly water stress anomaly** * The best two variables set used in developing the final ANFIS drought models  ** The best three variables set used in developing the final ANFIS drought models.  133  5.4.5 Categorical and MASH drought models The best ANFIS models for each drought category and MASH, including their statistical analysis and ANFIS configuration, are presented in Table 18. The ANFIS configuration consists of the input and output membership functions, and the number of membership functions. For all best selected models using two variables, the Gaussian membership function and linear output membership function were selected. For the best models selected using three variables, triangular and generalized bell were selected in addition to Gaussian membership functions. However, trapezoidal and Gaussian composite membership functions were never selected. The dominant combination for the number of membership functions were 4, 4 for two variables and 4, 4, 4 for three variables. The statistical analyses preformed on the models were found to be generally acceptable and consistent. The R2 of the drought models range from 0.64 to 0.97 for two variables and from 0.75 to 0.98 for three variables. The meteorological and hydrological drought models have the highest R2 of 0.91 and 0.97 using two variables and 0.95 and 0.98 using three variables, respectively. Furthermore, the RMSE values for both meteorological and hydrological drought models are low (below 10). Among the categorical drought models, the agricultural drought model had the lowest R2 value and the highest RMSE value. The MASH drought model had a R2 of 0.72 and RMSE of 18.93 using two variables, and a R2 of 0.75 and RMSE of 18.18 using three variables. It can be concluded that the predictive models developed from three variables are more reliable than those developed from two variables, due to higher R2 and lower RMSE values.   134  Table 18. Best ANFIS models for each drought category and MASH Drought Model Number of  Variables ANFIS Configuration RMSE R2 Shape MFs (MFs) Output MFs Meteorological 2 Gauss1 (4,4) Linear 8.87 0.91 3 Bell2 (4,4,4) Linear 6.18 0.95 Hydrological 2 Gauss (4,3) Linear 6.29 0.97 3 Gauss (4,4,4) Linear 4.82 0.98 Agricultural 2 Gauss (4,4) Linear 20.53 0.64 3 Triangle3 (4,4,4) Linear 16.32 0.77 MASH 2 Gauss (4,4) Linear 18.93 0.72 3 Gauss (4,4,4) Linear 18.18 0.75                     1 Gaussian; 2 Generalized bell; 3 Triangular The measured versus modeled histogram for the categorical drought models and MASH are presented in Figure 3 for three variables and Figure S9 for two variables. The x-axis represents the categorical drought and MASH scores and the y-axis represent the number of events. Overall, the histograms of measured scores are similar to the modeled ones. However, the predicted histogram of the agricultural model has a higher peak compared to the measured histogram for both two and three variables models. In the modeled histogram, the peak occurred almost 8000 times; however, in the measured histogram, the peak occurred about 6000 times. This can be due to the high RMSE and low R2 values of the agricultural model, which can results a shift in frequency classes. The histograms for the MASH model looked better than the agricultural model, but still have different peak values. The modeled histogram has a higher peak value (more than 8000) than the measured histogram (less than 8000) for two variables. However, using three variables improved the peaks in the predicted histogram.    135      Figure 11. Measured versus modeled histograms of categorical drought and MASH: (a) CMI, (b) CHI, (c) CAI, and (d) MASH.(a) (b) (c) (d) 136  5.4.6 Identifying the drought vulnerable areas  Identifying the drought prone areas is an important step toward developing mitigation strategies and actions to reduce drought impacts and vulnerability (Wilhite et al., 2014). In this section of the study, the goals are: 1) to demonstrate the application of MASH in identifying drought prone areas in the Saginaw River Watershed and 2) to compare the drought prone areas identified by MASH against drought prone areas identified by categorical drought indices. Drought prone areas were divided into three equal intervals of high, medium, and low priorities based on the number of drought events that occurred over the period of study (1979 to 2012). The map of drought prone areas based on MASH is presented in Figure 4 and the drought prone areas for the categorical drought indices are presented in Figure S10. The areas drawn as green show lower vulnerability while areas drawn in red show higher vulnerability to drought. Identifying drought prone areas, especially high propriety areas, can help policy makers and watershed managers to deploy mitigation strategies more effectively.   In the Saginaw River Watershed, 9% of the watershed was identified as high propriety areas based on the MASH index. In order to compare the high propriety areas of MASH with the The stream health category has the highest hit (69.85%) and the hydrological category had the lowest hit (9.23%). This indicates that 69.85% of high propriety areas for the categorical stream health drought match the high propriety areas of MASH; while only 9.23% of high propriety areas of hydrological category match with the high propriety areas of MASH. After the stream health category, the agricultural category has the highest hit with 56.56%, and the meteorological category has 10.29% hit, which is fairly close to the hydrological category. Overall, the stream health category had the highest overlap with the MASH index. 137  The percentages that each categorical drought score missed in identifying the high propriety areas compared to MASH were also calculated. The results indicate that the agricultural category had the highest miss (79.06%) and the stream health had the lowest miss (14.43%). This shows that the agricultural category failed to identify the overall high propriety areas 79.06% of the time, while the stream health category failed 14.43% of the time. Although the agricultural category was identified to have the second highest hit percentage of 56.56%, it also missed identifying high propriety areas 79.06% of the time. Therefore, it can be concluded that the agricultural category is less representative of the overall drought condition identified by MASH. After the agriculture category, the hydrological and meteorological categories were identified to have the highest miss of 64.24% and 61.07%, respectively.  Figure 12. Drought vulnerable areas based on MASH in the Saginaw River watershed 138  5.5.   Conclusion Despite the importance of understanding the overall impacts of droughts, no universal definition exists (Whitmore, 2000). Meanwhile, many drought indices have been developed to capture the impacts of drought. The goal of this study is to define an overall drought index. Thirteen commonly used drought indices were used that commonly represent meteorological, hydrological, agricultural, and stream health categories. Cluster analysis was used to find the closest three drought indices in each drought category and then averaged the scores to create categorical drought scores, which later were used to define the overall drought score named MASH.   The ReliefF algorithm was used to identify the top ranked variables that were later used to develop the predictive drought models. The results of the ReliefF algorithm indicated that precipitation, streamflow, and evapotranspiration were the most influential variables of meteorological, hydrological, and agricultural drought categories, respectively. In addition, the streamflow variables were selected as the top ranked variables for the overall drought. The predictive drought models were developed using ANFIS with two input variables and three input variables. The results of the predictive models using three input variables was better compared to using two input variables. The R2 values were 0.95 for meteorological, 0.98 for hydrological, 0.77 for agricultural drought categories, and 0.75 for MASH. The drought prone areas identified by MASH were compared with the drought prone areas identified by meteorological, hydrological, agricultural, and stream health categories. The results indicated that the stream health category had the highest hit (69.85%) and the hydrological category has the lowest hit (9.23%) compared to MASH. In addition, the 139  agricultural category had the highest miss (79.06%) and the stream health had the lowest miss (14.43%). Overall, the stream heath category had the closest identification of high propriety areas to MASH. This study introduced a comprehensive drought index (MASH) capable of quantifying drought with respect to metrological, agricultural, stream health, and hydrological aspects. Future studies should include other aspects of drought such as economic and social, which can further improve the general understanding of drought impacts on human and natural systems.  5.6.   Acknowledgements This work is supported by the USDA National Institute of Food and Agriculture, Hatch project MICL02212.   140  6. CONCLUSIONS This research introduced new concepts for use in quantifying both individual and overall impacts of drought via the introduction of new indices, providing valuable information for decision makers to better allocate limited resources for drought mitigations. In the first study, a new index was developed capable of predicting the drought severity in the context of stream health. Different physiographical and climatological variables were then employed to develop drought monitoring and forecasting models. In the second study, the newly developed stream health drought index was used in conjunction with 12 other drought indices representing the meteorological, hydrological, and agricultural components of drought; thereby creating a comprehensive new drought index. Finally, predictive models were developed to estimate the comprehensive drought index and the four categorical drought indices. The following conclusions were based on the results of the two studies:  The average flowrate parameters were the top-ranked variables used in the development of the stream health drought models.  The introduction of stream health drought models allowed us to measure the impacts of drought on aquatic ecosystems, enabling policymakers to improve water resource management methods via use of bioassessment.  The developed stream health drought model is highly reliable for the study of future climatological conditions, thereby helping to mitigate the impacts of climate change on aquatic ecosystems.  The application of stream health drought models under future climate scenarios revealed that the majority of streams (93.6%) within the study area are expected to experience higher probability of degradation by the mid-21st century. 141   The precipitation, streamflow, and evapotranspiration variables were the most influential parameters used in the development of the categorical (i.e., meteorological, hydrological, and agricultural) drought models.  The streamflow variables were the top-ranked variables for development of the overall MASH drought model.   Categorical drought indices are recommended for use in monitoring sectorial drought, rather than a single index.  The universal drought index introduced in this study provides a comprehensive view of drought impacts on climate, hydrology, agriculture, and stream health.    142  7. FUTURE RESEARCH RECOMMENDATIONS This research added a new dimension to drought study by introducing drought in the context of stream health. In addition, a universal drought index was presented that accounts for meteorological, agricultural, hydrological, and stream health aspects of droughts. However, additional research should be performed on the applicability of these new developed models beyond the study area. The following are potential areas to expand upon in future research:  Evaluation of the performance of developed drought models in different regions with different climate variabilities   Development of additional stream health drought indices to capture the impacts of drought on different components of aquatic species (e.g. macroinvertebrates, plants, snails, etc.).    Examination of the uncertainty associated with the input data and model components. Uncertainty can be simply defined as a lack of exact knowledge (Refsgaard et al., 2007), which is an important consideration in modeling. Accounting for uncertainty throughout the modeling process will ultimately improve decision analysis by providing accurate pictures of likely outcomes.     Evaluation of additional indices within each drought category (e.g. meteorological, agricultural, and hydrological) in order to better define drought conditions.   Incorporation of the economic and social aspects into drought assessment in order to better understand its overall impact on human and natural systems.  Development of an early warning system within the decision support platform that is capable of predicting the comprehensive drought index.   143          APPENDICES         144  APPENDIX A: Study One    Figure S1.Locations of precipitation, temperature, and streamflow monitoring stations      145      Figure S2. Distribution of median flow values   146      Figure S3. Sample histogram of ranking for parameter #20 (average flow rate from 23 months prior to the month of interest)       147        Figure S4. The relationship of MSE with the number of PLSR components for Current Drought Severity Models: a) First Model, b) Second Model, c) Third Model.  (b) (a) (c) 148        Figure S5.The relationship of MSE with the number of PLSR components for Future Drought Severity Models: a) Fourth Model, b) Fifth Model, c) Sixth Model.   (b) (c) (a) 149        Figure S 6. The variance explained percentage for each PLSR for the Future Drought Severity Model: a) Fourth Model, b) Fifth Model, c) Sixth Model. (a) (b) (c) 150                                                       (a)                       (b) (c) Figure S7. The comparison of measured vs. predicted median flow histogram for the Future Drought Severity Model: a) Fourth Model, b) Fifth Model, c) Sixth Model.   151  Table S1. Selected variables for development of current and future drought severity models. Category  Variable Current Drought Severity Model Future Drought Severity Model Ranking Top Ranked Variables Time scale based variables  5 10 15 6 months 12 months 18 months Precipitation 54 Precipitation for the month of interest       55 Precipitation from one month prior to the month of interest       62 Precipitation from two months prior to the month of interest       66 Precipitation from three months prior to the month of interest       65 Precipitation from four months prior to the month of interest       59 Precipitation from five months prior to the month of interest       60 Precipitation from six months prior to the month of interest    x   63 Precipitation from seven months prior to the month of interest    x   57 Precipitation from eight months prior to the month of interest    x   61 Precipitation from nine months prior to the month of interest    x   64 Precipitation from ten months prior to the month of interest    x   58 Precipitation from eleven months prior to the month of interest    x   56 Precipitation from twelve months prior to the month of interest    x x  152          32 Two months average precipitation ending with the month of interest       34 Three months average precipitation ending with the month of interest       30 Four months average precipitation ending with the month of interest       38 Five months average precipitation ending with the month of interest       37 Six months average precipitation ending with the month of interest    x   35 Seven months average precipitation ending with the month of interest    x   33 Eight months average precipitation ending with the month of interest    x   29 Nine months average precipitation ending with the month of interest    x   28 Ten months average precipitation ending with the month of interest    x   27 Eleven months average precipitation ending with the month of interest    x   25 Twelve months average precipitation ending with the month of interest    x x  26 Thirteen months average     x x  153            precipitation ending with the month of interest       Streamflow 1 Average flow rate from one month prior to the month of interest x x x    2 Average flow rate from two months prior to the month of interest x x x    7 Average flow rate from three months prior to the month of interest  x x    11 Average flow rate from four months prior to the month of interest   x    15 Average flow rate from five months prior to the month of interest   x    19 Average flow rate from six months prior to the month of interest    x   20 Average flow rate from seven months prior to the month of interest    x   16 Average flow rate from eight months prior to the month of interest    x   14 Average flow rate from nine months prior to the month of interest   x x   10 Average flow rate from ten months prior to the month of interest  x x x   154          6 Average flow rate from eleven months prior to the month of interest  x x x   4 Average flow rate from twelve months prior to the month of interest x x x x x  5 Average flow rate from thirteen months prior to the month of interest x x x x x  9 Average flow rate from fourteen months prior to the month of interest  x x x x  13 Average flow rate from fifteen months prior to the month of interest   x x x  18 Average flow rate from sixteen months prior to the month of interest    x x  22 Average flow rate from seventeen months prior to the month of interest    x x  23 Average flow rate from eighteen months prior to the month of interest    x x x 24 Average flow rate from nineteen months prior to the month of interest    x x x 21 Average flow rate from twenty months prior to the month of interest    x x x 17 Average flow rate from twenty one months prior to the month of     x x x 155           interest       12 Average flow rate from twenty two months prior to the month of interest   x x x x 8 Average flow rate from twenty three months prior to the month of interest  x x x x x 3 Average flow rate from twenty four months prior to the month of interest x x x x x x Land use 48 Agriculture       44 Percent Agriculture       42 Forest       39 Percent Forest       53 Urban       52 Percent Urban       40 Water       36 Percent Water       Soil 51 Group A       45 Percent Group A       49 Group B       43 Percent Group B       47 Group C       46 Percent Group C       50 Group D       41 Percent Group D       Total Drainage Area 31 Total Area       156  Table S2. Confusion matrix for drought zones: Second model                        Drought           Zone Predicted  A B C D Sensitivity Actual A 3,475,503 96,794 78,135 424,041 85% B 98,613 86,999 10,552 76,635 32% C 73,124 9,531 9,780 58,201 6% D 575,476 89,683 66,157 1,206,881 62% Precision 82% 31% 6% 68% Accuracy = 74%           Table S3. Confusion matrix for drought zones: Third model             Drought           Zone Predicted  A B C D Sensitivity Actual A 3,504,238 96,727 75,344 395,412 86% B 102,626 88,060 10,722 71,336 32% C 76,366 9,811 9,959 54,460 7% D 600,376 94,258 69,978 1,172,227 61% Precision 82% 30% 6% 69% Accuracy = 74%     157  Table S4. Confusion matrix for drought zones: Fifth Model                          Drought          Zone Predicted  A B C D Sensitivity Actual A 3,374,769 82,711 66,849 544,987 83% B 104,756 81,446 6,460 79,932 30% C 81,633 6,204 5,906 56,798 4% D 776,502 82,185 51,095 1,026,000 53% Precision %78 32% 5% 60% Accuracy = 70%       Table S5. Confusion matrix for drought zones: Sixth Model            Drought          Zone Predicted  A B C D Sensitivity Actual A 3,013,851 112,807 93,637 854,265 74% B 100,057 66,561 6,319 99,903 24% C 79,508 5,999 6,034 59,135 4% D 1,109,043 94,052 53,437 683,459 35% Precision 70% 24% 4% 40% Accuracy = 59%    158  Table S6. The first drought model performance using RCP 8.5                                       (maximum and minimum values are presented in red) Model Name R2 Log 10 (median flow in LPD*) RMSE HadGEM2-ES 0.8861 0.5036 FIO-ESM 0.8798 0.5097 MIROC-ESM-CHEM 0.8780 0.5070 MIROC5 0.8776 0.5060 GISS-E2-H 0.8775 0.5056 MIRCO-ESM 0.8767 0.5108 GISS-E2-R 0.8766 0.5039 HadGEM2-AO 0.8764 0.5104 CESM1-CAM5 0.8755 0.5134 IPSL-CM5A-MR 0.8742 0.5261 GFDL-CM3 0.8741 0.5041 CCSM4 0.8740 0.5192 IPSL-CM5A-LR 0.8734 0.5203 GFDL-ESM2M 0.8715 0.5312 MRI-CGCM3 0.8681 0.5206 GFDL-ESM2G 0.8662 0.5183                        * LPD: Liter per day    159  Table S7. The first drought model performance using RCP 6.0                                              (maximum and minimum values are presented in red) Model Name R2 Log 10 (median flow in LPD*) RMSE MIROC5 0.8796 0.5101 MIROC-ESM-CHEM 0.8773 0.5124 IPSL-CM5A-MR 0.8765 0.5251 CESM1-CAM5 0.8764 0.5132 HadGEM2-AO 0.8761 0.5246 MIRCO-ESM 0.8753 0.5143 HadGEM2-ES 0.8752 0.5094 FIO-ESM 0.8748 0.5161 GISS-E2-H 0.8739 0.5175 IPSL-CM5A-LR 0.8732 0.5200 GISS-E2-R 0.8726 0.5109 GFDL-CM3 0.8723 0.5066 GFDL-ESM2G 0.8706 0.5206 MRI-CGCM3 0.8690 0.5216 CCSM4 0.8688 0.5284 GFDL-ESM2M 0.8687 0.5313                           * LPD: Liter per day   160  Table S8. The first drought model performance using RCP 4.5                                           (maximum and minimum values are presented in red) Model Name R2 Log 10 (median flow in LPD*) RMSE HadGEM2-ES 0.8861 0.4940 MIROC5 0.8798 0.5007 GISS-E2-R 0.8780 0.4968 HadGEM2-AO 0.8776 0.5055 CCSM4 0.8775 0.5272 MIRCO-ESM 0.8767 0.5135 MIROC-ESM-CHEM 0.8766 0.5087 FIO-ESM 0.8764 0.5180 CESM1-CAM5 0.8755 0.5161 IPSL-CM5A-MR 0.8742 0.5260 IPSL-CM5A-LR 0.8741 0.5208 GFDL-CM3 0.8740 0.5115 GFDL-ESM2G 0.8734 0.5206 MRI-CGCM3 0.8715 0.5289 GFDL-ESM2M 0.8681 0.5407                      * LPD: Liter per day            161  APPENDIX B: Study Two    Figure S8. Location of temperature, precipitation, and streamflow gauging stations     162      Figure S9. Measured versus modeled histograms of categorical drought and MASH: (a) CMI, (b) CHI, (c) CAI, and (d) MASH (a) (b) (c) (d) 163   Figure S10. Drought vulnerable areas based on categorical drought indices in the Saginaw River watershed: (a) meteorological, (b) hydrological, (c) agricultural, (d) stream health 164  Table S9. Meteorological drought indices, reference, input parameters, procedure, classification, and index value Meteorological Indices (References) Input Parameter Procedure Classification Index value Palmer Drought Severity Index (PDSI) (Palmer, 1965; Jacob et al., 2013; Ficklin et al., 2015) Precipitation, Temperature, Solar radiation, Wind speed, Relative humidity, Available water content, Albedo, Elevation Measures moisture supply and demand within a two-layer bucket-type soil model using the water balance equation  Extreme drought Severe drought Moderately drought Mild drought Incipient drought Near normal Incipient wet spell Slightly wet Moderately wet Very wet Extremely wet -4.0 --4.0 --3.0 --2.0 --1.0 0.5 < PDSI < -0.5       Rainfall Deciles (RD) (Gibbs and Maher, 1967) Precipitation Dividing a long-term monthly precipitation distribution into deciles(10% parts) Much below normal Below normal Near normal Above normal Much above normal Deciles 1-2  Deciles 3-4 Deciles 5-6 Deciles 7-8 Deciles 9-10  Standardized Precipitation Index (SPI) (McKee et al., 1993)  Precipitation  Fitting a probability distribution to a historical precipitation records, and transforming it into a standardized normalized distribution   Extreme drought  Severe drought Moderately drought Near normal Moderately wet Very wet Extremely wet  -2.0 -2.0 < SPI < -1.5 --1.0 -     Reconnaissance Drought Index (RDI) (Tsakiris and Vangelis, 2005; Zarch et al., 2011)  Precipitation, Potential evapotranspiration (PET)  Fitting a probability distribution to a historical precipitation/PET  records, and transforming it into a standardized normalized distribution   Extreme drought  Severe drought Moderately drought Near normal Moderately wet Very wet Extremely wet  -2.0 -2.0 < RDI < -1.5 --1.0 -        165  Table S10. Agricultural drought indices, reference, input parameters, procedure, classification, and index value Agricultural Indices (References) Input Parameter Procedure Classification Index Value Palmer Moisture Anomaly Index (Z-index) (Palmer, 1965; Jacob et al., 2013; Ficklin et al., 2015) Precipitation, Temperature, Solar radiation, Wind speed, Relative humidity, Available water content, Albedo, Elevation Measures the soil moisture anomaly for the current month in the Palmer model Extreme drought  Severe drought Moderately drought Near normal Moderately moist Very moist Extremely moist Z--2.75 -2.75 < Z--2.0 -2.0 < Z-index < -1.25 -1.25< Z-index < 1.0 -index < 2.5 -index < 3.5 Z-  Soil Moisture Deficit Index (SMDI) (Narasimhan and Srinivasan, 2005) Soil Moisture Uses a high-resolution hydrologic model coupled with a crop growth model to calculate weekly soil moisture deficit                  SMDIj: SMDI during any week SDj: weekly soil water deficit (%) Extreme drought Severe drought Moderately drought Mild drought Incipient drought Near normal Incipient wet spell Slightly wet Moderately wet Very wet Extremely wet -4.0 --4.0 --3.0 --2.0 --1.0 0.5 <  SMDI < -0.5       Evapotranspiration Deficit Index (ETDI) (Narasimhan and Srinivasan, 2005) Potential evapotranspiration, actual evapotranspiration Uses a high-resolution hydrologic model coupled with a crop growth model to calculate weekly evapotranspiration deficit          ETDIj: ETDI during any week WSAj: weekly water stress anomaly Extreme drought Severe drought Moderately drought Mild drought Incipient drought Near normal Incipient wet spell Slightly wet Moderately wet Very wet Extremely wet ETDI -4.0 -ETDI < -4.0 -ETDI < -3.0 -ETDI < -2.0 -ETDI < -1.0 0.5 <  ETDI < -0.5 ETDI < 1.0 ETDI < 2.0 ETDI < 3.0 ETDI < 4.0 ETDI                                 166  Table S10.      Soil Water Deficit Index (SWDI) (Martinez-Fernandez et al., 2015) Soil moisture, Available water content, Field capacity, Wilting point Uses soil water observations to calculate soil water deficit                            FC: field capacity AWC: available water content which is FC WP No drought Mild Moderate Severe Extreme  -2 < SWDI < 0 --2 --5  -10   167  Table S11. Hydrological drought indices, reference, input parameters, procedure, classification, and index value Hydrological Indices (References) Input Parameter Procedure Classification Index value   Palmer Hydrological Drought Index (PHDI)                   (Palmer, 1965; Jacob et al., 2013; Ficklin et al., 2015) Precipitation, Temperature, Solar radiation, Wind speed, Relative humidity, Available water content, Albedo, Elevation Measures moisture supply and demand within a two-layer bucket-type soil model using the water balance equation Extreme drought Severe drought Moderately drought Near normal Moderately moist Very moist Extremely moist -4.0 --4.0 --3.0 -2.0 < PHDI < 2.0     Flow Duration Curve (FDC) (Tallaksen and van Lanen, 2004)  Streamflow  Measuring the cumulative probability of the streamflow for a specific time period  High flows Moist conditions Mid-range flows Dry condition Low flows  0-10% 10-40% 40-60% 60-90% 90-100%  Standardized Runoff Index (SRI)                                 (Shukla and Wood, 2008) Runoff Fitting a log normal distribution to a historical runoff records, and transforming it into a standardized normalized distribution  Extremely wet Severely wet Moderately wet Near normal Moderately drought Severe drought Extreme drought     - --1.0 - -2.0  Water Balance Derived Drought Index (WBI)                 (Vasiliades et al., 2011) Runoff Normalizing historical runoff records using Box-Cox transformation, and standardizing it into a standard normal distribution  Extremely wet Severely wet Moderately wet Near normal Moderately drought Severe drought Extreme drought    - --1.0 -2.0 < WBI  -2.0    168  Table S12. Stream health drought index, reference, input parameters, procedure, classification, and index value Stream Health Index (Reference) Input Parameter Procedure Classification Index value*   Stream Health Index (SHI)                  (Esfahanian et al., 2016) Streamflow Calculating monthly median flowrate and Index flow values for each stream Extreme drought  IF IF IF IF IF IF IF IF IF Warm Small IF IF    Severe drought  Cold Streams: 0.8IFIF Cold Small Rivers: 0.79IFIF Cold Transitional Streams: None Cold Transitional Small Rivers: None Cold Transitional Large Rivers: None Cool Transitional Streams: 0.75IFIF Cool Transitional Small Rivers: 0.75IFIF Cool Transitional Large Rivers: 0.75IFIF Warm Streams: 0.76IFIF  Warm Small Rivers: 0.83IFIF Warm Large Rivers: 0.78IFIF    Moderate drought  Cold Streams: None Cold Small Rivers: None Cold Transitional Streams: SHI>96%IF Cold Transitional Small Rivers: SHI>98% IF Cold Transitional Large Rivers: SHI>97% IF Cool Transitional Streams: 0.85IFIF Cool Transitional Small Rivers: 0.81IFIF Cool Transitional Large Rivers: 0.81IFIF Warm Streams: 0.82IFIF Warm Small Rivers: 0.87IFIF Warm Large Rivers: 0.84IFIF    Initial Drought  Cold Streams: SHI>0.86IF Cold Small Rivers: SHI>0.895IF Cold Transitional Streams: None Cold Transitional Small Rivers: None Cold Transitional Large Rivers: None Cool Transitional Streams: SHI>0.94IF Cool Transitional Small Rivers: SHI>0.85IF Cool Transitional Large Rivers: SHI>0.86IF Warm Streams: SHI>0.9IF Warm Small Rivers: SHI>0.92IF Warm Large Rivers: SHI>0.9IF 169  Table S13. Input parameters Input Parameters Drought Indices Source/description Meteorological Drought Agricultural Drought Hydrological Drought Stream Health Drought PDSI RD SPI RDI Z-index SMDI ETDI SWDI PHDI FDC SRI WBI SHI Precipitation (mm) x x x x x    x     Precipitation stations of National Climatic Data Center (NCDC) http://www.ncdc.noaa.gov/data-access/land-based-station-data Total 3-month precipitation (mm)  x            Total precipitation for the preceding three months Precipitation percentile   x            Historical ranking of total 3-month precipitation for each month Gamma-precipitation   x           Fitted gamma probability density function for the monthly precipitation Potential evapotranspiration (mm)    x   x       Hydrological model (soil and water tool assessment (SWAT)) Evapotranspiration (mm)       x       Hydrological model (soil and water tool assessment (SWAT)) P-PET ratio    x          Ratio of monthly precipitation to monthly potential evapotranspiration                                              170  Table S13.               Gamma P-PET    x          Fitted gamma probability density function for the monthly P-PET ratio Streamflow (cms)          x x x x Monthly median and average streamflow up to 24 months prior from the month of interest, Obtained from SWAT Streamflow exceedance probability          x    Exceedance Probability of monthly streamflow Log-normal streamflow           x   Fitted log-normal distribution function for the monthly streamflow Lambda-streamflow            x  Lambda coefficient for the box-cox transformation of monthly streamflow Transformed-streamflow            x  Box-cox transformed monthly streamflow values Mean T-streamflow            x  Mean of the box-cox transformed monthly streamflow time series Standard deviation            x  Standard deviation of the box-cox transformed monthly streamflow time series  (mm)      x  x      Soil moisture content, obtained from SWAT FC (mm)        x      Soil field capacity, obtained from SWAT 171  Table S13.                     x      Available water content, Obtained from SWAT Palmer potential evapotranspiration (mm) x    x    x     Obtained from Ficklin et al.(2015) Matlab code (modified version of Jacob et al.,2013) Palmer percentage probability x    x    x     Obtained from Ficklin et al.(2015) Matlab code (modified version of Jacob et al.,2013) Palmer_X1 x    x    x     Severity index for a wet spell that is being established Obtained from Ficklin et al.(2015) Matlab code (modified version of Jacob et al.,2013). Severity index for a wet spell that is becoming established. Palmer_X2 x    x    x     Severity index for a dry spell that is being established Obtained from Ficklin et al.(2015) Matlab code (modified version of Jacob et al.,2013). Severity index for a drought that is becoming established.                               172  Table S13.              Palmer_X3 x    x    x     Severity index for a wet/dry spell that has been established Obtained from Ficklin et al.(2015) Matlab code (modified version of Jacob et al.,2013). Severity index for any wet spell or any drought that has become established. Palmer_X x    x    x     Dry/wet spell severity index Obtained from Ficklin et al.(2015) Matlab code (modified version of Jacob et al.,2013) SMDI_1M       x        Monthly soil moisture deficit, Narasimhan and Srinivasan, 2005 meanSD      x        Average monthly soil moisture deficit MeanSW (mm)      x        Average monthly soil moisture MaxSW (mm)      x        Maximum monthly soil moisture MinSW (mm)      x        Minimum monthly soil moisture Monthly water stress ratio       x       Calculated from monthly potential and actual evapotranspiration obtained by SWAT Average water stress ratio       x       Average of monthly water stress ratios                173  Table S13.              Maximum water stress ratio       x       Maximum monthly water stress ratios Minimum water stress ratio       x       Minimum monthly water stress ratios Monthly water stress anomaly (%)       x       Monthly water stress anomaly ETDI_1M              Monthly evapotranspiration deficit Temperature (°C) x    x    x     Minimum and maximum temperature values, obtained from NCDC temperature stations http://www.ncdc.noaa.gov/data-access/land-based-station-data Solar Radiation (W/m2) x    x    x     Abatzolgou (2013), http://metdata.northwestknowledge.net/ Wind Speed (m/s) x    x    x     Abatzolgou (2013), http://metdata.northwestknowledge.net/ Relative Humidity (%) x    x    x     Abatzolgou (2013), http://metdata.northwestknowledge.net/ Available water content (mm) x    x   x x     Natural Resources Conservation Service (NRCS) Soil Survey Geographic (SSURGO) database https://www.arcgis.com/home/item.html?id=a23eb436f6ec4ad6982000dbaddea5ea                                              174  Table S13.              Albedo x    x    x     Average annual values from Barkstrom (1984) Elevation (m) x    x    x     National Elevation Dataset of the US Geological Survey (USGS) with a spatial resolution of 30 m http://nationalmap.gov/elevation.html    175  Table S14. Mean difference (numbers in black) and standard deviation (numbers in red) among drought indices Drought Index   Meteorological  Hydrological  Agricultural  Stream  Health  Overall PDSI  RD  SPI  RDI  CMI  PHDI  FDC  SRI  WBI  CHI  Z-Index  SMDI  ETDI  SWDI  CAI  CSHI      MASH  PDSI  0 0                                RD  -1.32 45.67 0 0                              SPI  -0.04 38.93 1.29 58.97 0 0                            RDI  -0.07 38.29 1.26 58.63 -0.03 10.72  0 0                          CMI  -0.18 28.81 1.14 50.34 -0.15 14.96 -0.12 14.45 0 0                        PHDI  -1.52 22.93 -0.2  46.75 -1.48 45.57 -1.45 44.86 -1.34         37.04  0 0                      FDC  -0.08 44.54 1.24 53.14 -0.05 49.86 -0.02 51.11 0.1       45.5 1.44 47.54 0 0                    SRI  -0.14 32.43 1.18 44.22 -0.10 32.68 -0.07 33.71 0.04           27.81 1.38 35.91 -0.06 28.69 0 0                  WBI  0.45 32.64 1.78 44.26 -0.49 32.92 0.52 33.80 0.64           28.07 1.97 35.9 0.54 30.11 0.59 8.81 0 0                CHI  0.03 32.69 1.36 44.52 0.071 35.16 0.1 36.17 0.22           29.98 1.56 36.34 0.12 24.26 0.17 9.35 -0.42 10.57 0 0              Z-Index  0.77 30.08 2.09  55 0.8 17.84 0.84 17.85 0.95           15.29 2.29 38.40 0.85 45.81 0.91 30.33 0.32 30.83 0.73      32.12 0 0            SMDI  -4.79 36.68 -3.47 46.42 -4.76 46.50 -4.73 45.65 -4.61       39.52 -3.27 39.65 -4.71    49 -4.65 37.94 -5.25 37.87 -4.83     38.49 -5.56   40.96 0 0          ETDI  -2.76 40.71 -1.44 52.78 -2.73 35.52 -2.7 35.33 -2.58         32.40 -1.24 44.85 -2.68 52.73 -2.62 35.88 -3.22 35.32 -2.8       37.86 -3.53   34.53 2.03    39.17 0 0        SWDI  62.33 41.23 63.65 51.60 62.36 35.94 62.39 35.62 62.51         33.14 63.85 43.54 62.41 56.18 62.47 34.68 61.87 33.87 62.29    38.04 61.56  36.98 67.12 44.13 65.09 34.88 0 0      CAI  4.51 31.27 5.83 48.82 4.54 30.66 4.57 30.23 4.69           25.27 6.03 37.05 4.59 45.49 4.64 30.51 4.05 30.52 4.47      31.83 3.74    24.99 9.3 27.38 7.27 25.12 -57.82 36.50 0 0    CSHI -36.86 74.54 -35.54 78.74 -36.83 80.86 -36.8 81.55 -36.68       77.27 -35.34 75.44 -36.78 66.81 -36.72 69.62 -37.32 69.62 -36.9     68.13 -37.63 77.64 -32.07 75.53 -34.1 79.55 -99.19 82.87 -41.37    74.98 0 0   MASH  -8.13 30.56 -6.8 45.84 -8.09 32.57 -8.06 33.11 -7.94         26.52 -6.61 35.57 -8.04 33.53 -7.99 22.04 -8.58 22.26 -8.16     21.55 -8.9     28.38 -3.33 34.65 -5.36 34.45 -70.45 38.89 -12.63   25.38 28.74 53.71 0 0 176  Table S15. Saginaw River watershed calibration and validation results USGS Gauging Stations NSE PBIAS RSR Calibration Validation Calibration Validation Calibration Validation 04151500 0.78 0.84 11.08 8.30 0.47 0.4 04154000 0.53 0.58 -2.39 -13.30 0.68 0.65 04148500 0.66 0.63 20.011 -13.92 0.58 0.61 04147500 0.58 0.58 11.77 -20.33 0.65 0.65 04155500 0.81 0.79 3.30 -12.23 0.43 0.46 04157000 0.79 0.85 12.34 0.67 0.46 0.39 04144500 0.78 0.69 14.69 -3.34 0.47 0.55 04156000 0.71 0.74 4.67 -1.57 0.54 0.52            177  Table S16. Transformed drought categories * Drought index General associated ranges (normalized associated ranges) Initial drought (0 to <25)  Moderate drought (25 to <50) Severe drought (50 to <75) Extreme drought (75 to 100) PDSI -1.0 < PDSI < 0 --1.0 --3.0 -4.0 RD 0.3 < Decile < 0.5    SPI -1.0 < SPI < 0 --1.0 --1.5 -2.0 RDI -1.0 < RDI < 0 --1.0 --1.5 -2.0 Z-index -1.0 < Z < 0 --1.0 --3.0 -4.0 SMDI -1.0< SMDI < 0 --1.0 --3.0 -4.0 ETDI -1.0< ETDI < 0 --1.0 --3.0 -4.0 SWDI -2.0< SWDI < 0 --2.0 --5.0 -10.0 PHDI -1.0< PHDI < 0 --1.0 -4-3.0 -4.0 FDC 0.5 < FDC < 0.6    SRI -1.0 < SRI < 0 --1.0 --1.5 -2.0 WBI -1.0 < WBI < 0 --1.0 --1.5 -2.0 CSHI (Cool-Stream) 0.94IF<MMIF 0.85IF 0.75IF 0IF CSHI Cool-Small Rivers 0.85IF 0.81IF 0.75IF 0IF CSHI Cool-Large Rivers 0.86IF 0.81IF 0.75IF 0IF CSHI Warm-Stream 0.9IF 0.81IF 0.76IF 0IF CSHI Warm-Small Rivers 0.92IF 0.87IF 0.83IF 0IF CSHI Warm-Large Rivers 0.9IF 0.84IF 0.78IF 0IF      * CSHI: Categorical stream health drought index; IF: Index flow; MMF: Monthly median flow    178  Table S17. Transformed non-drought categories * Drought index General associated ranges (normalized associated ranges) Initial wet (<0 to -25)  Moderate wet (<-25 to -50) Severe wet (<-50 to -75) Extreme wet (<-75 to -100) PDSI 0 < PDSI < 1.0     RD 0.5< Decile< 0.7   Deci SPI 0 < SPI <1.0    RDI 0 < RDI < 1.0    Z-index 0 < Z < 1.0    SMDI 0 < SMDI < 1.0    0 ETDI 0 < ETDI < 1.0     SWDI 0 < SWDI < 2.0    PHDI 0 < PHDI < 1.0     FDC 0.4 < FDC < 0.5  0.1 < F  SRI 0 < SRI < 1.0    WBI 0 <WBI < 1.0    CSHI (Cool-Stream) 1IF 1.06IF 1.15IF 1.25IF CSHI Cool-Small Rivers 1IF 1.15IF 1.19IF 1.25IF CSHI Cool-Large Rivers 1IF 1.14IF 1.19IF 1.25IF CSHI Warm-Stream 1IF 1.10IF 1.19IF 1.24IF CSHI Warm-Small Rivers 1IF 1.08IF 1.13IF 1.17IF CSHI Warm-Large Rivers 1IF 1.10IF 1.16IF 1.22IF     * CSHI: Categorical stream health drought index; IF: Index flow; MMF: Monthly median flow  179           REFERENCES         180  REFERENCES Abatzoglou, J.T., 2013. Development of gridded surface meteorological data for ecological applications and modelling. International Journal of Climatology 33, 121131.  Adam, J.C., Hamlet, A.F., Lettenmaier, D.P., 2009. Implications of global climate change for snowmelt hydrology in the twenty-first century. Hydrological Processes 23, 962972. Alegria, M.H., Watkin Jr., D.W., 2007. Annual and warm season drought intensity durationfrequency analysis for Sonora, Mexico. Journal of Climate 20, 18971909.  Alley, W.M., 1984. The Palmer drought severity index: limitations and assumptions. Journal of climate and applied meteorology 23, 11001109.  American Meteorological Society, 1997. Meteorological drought-Policy statement. Bulletin of the American Meteorological Society 78, 847-849.  American Meteorological Society, 2004. Statement on meteorological drought. Bulletin of the American Meteorological Society 85, 771773.  Angradi, T.R., Pearson, M.S., Jicha, T.M., Taylor, D.L., Bolgrien, D.W., Moffett, M.F., Blocksom, K.A., Hill, B.H., 2009. Using stressor gradients to determine reference expectations for great river fish assemblages. Ecological Indicators 9, 748764.  Anyamba, A. C., Tucker, C. J., and Eastman, J. R. 2001. NDVI anomaly patterns over Africa during the 1997/98 ENSO warm event. International Journal of Remote Sensing, 22 (10): 18471859.  Aruna, S., Rajagopalan, S.P., Nandakishore, L.V., 2011. Knowledge based analysis of various statistical tools in detecting breast cancer. Computer Science & Information Technology 2, 3745.  Bacanli, U.G., Firat, M., Dikbas, F., 2009. Adaptive neuro-fuzzy inference system for drought forecasting. Stochastic Environmental Research and Risk Assessment 23, 11431154.  Balint, Z., Mutua, F.M., Muchiri, P., 2011. Drought monitoring with the combined drought index. FAO-SWALIM: Nairobi, Kenya 325.  Barbour, M.T., Gerritsen, J., Griffith, G.E., Frydenborg, R., McCarron, E., White, J.S., Bastian, M.L., 1996. A framework for biological criteria for Florida streams using benthic macroinvertebrates. Journal of the North American Benthological Society 185211.  181  Barbour, M.T., Gerritsen, J., Snyder, B.D., Stribling, J.B., 1999. Rapid Bioassessment Protocols for use in Streams and Wadeable Rivers: Periphyton, Benthic Macroinvertebrates and Fish, second ed. (EPA 841-B-99-002. U.S.). Barkstrom, B.R., 1984. The earth radiation budget experiment (ERBE). Bulletin of the American Meteorological Society 65, 11701185.  Barlow, M., Cullen, H., Lyon, B., 2002. Drought in central and southwest Asia: La Niña, the warm pool, and Indian Ocean precipitation. Journal of Climate 15, 697700.  Bates, B.C., Kundzewicz, Z.W., Wu, S., Palutikof, J.P. (Eds.), 2008. Climate Change and Water. Technical Paper, International Panel on Climate Change (IPCC) Secretariat, Geneva. Bayarjargal, Y., Karnieli, A., Bayasgalan, M., Khudulmur, S., Gandush, C., Tucker, C.J., 2006. A comparative study of NOAAAVHRR derived drought indices using change vector analysis. Remote Sensing of Environment 105, 922. Bhalme, H.N., Mooley, D.A., 1980. Large-scale droughts/floods and monsoon circulation. Monthly Weather Review 108, 11971211. Blenkinsop, S., Fowler, H.J., 2007. Changes in European drought characteristics projected by the PRUDENCE regional climate models. International Journal of Climatology 27, 15951610. Boken, V.K., Cracknell, A.P., Heathcote, R.L., 2005. Monitoring and predicting agricultural drought: a global study. Oxford University Press.  Bond, N.R., Lake, P.S., Arthington, A.H., 2008. The impacts of drought on freshwater ecosystems: an Australian perspective. Hydrobiologia 600, 316.  Box, G.E.P., Jenkins, G.M., Reinsel, G.C., 1994. Time Series Analysis-C-Hall, Englewood Cliffs, NJ. Boyer, C., Chaumont, D., Chartier, I., Roy, A.G., 2010. Impact of climate change on the hydrology of St. Lawrence tributaries. Journal of hydrology 384, 6583.  Brian D.Wardlow, Martha C. Anderson, James P. Verdin, 2012. Remote Sensing of Drought.Taylor & Francis Group.  Britton, J.R., Cucherousset, J., Davies, G.D., Godard, M.J., Copp, G.H., 2010. Non-native fishies and climate change: predicting species responses to warming temperatures in a temperature region. Freshwater Biology 55, 1130-1141. Bravar, L., Kavvas, M.L., 1991. On the physics of droughts. I. A conceptual framework. Journal of Hydrology 129, 281297. 182  Brown, J.F., Wardlow, B.D., Tadesse, T., Hayes, M.J., Reed, B.C., 2008. The Vegetation Drought Response Index (VegDRI): A new integrated approach for monitoring drought stress in vegetation. GIScience & Remote Sensing 45, 1646.  Bunn, S.E., Arthington, A.H., 2002. Basic principles and ecological consequences of altered flow regimes for aquatic biodiversity. Environmental management 30, 492507.  Byun, H.-R., Wilhite, D.A., 1999. Objective quantification of drought severity and duration. Journal of Climate 12, 27472756.  Cancelliere, A., Salas, J.D., 2004. Drought length properties for periodic-stochastic hydrologic data. Water Resources Research 40.  Cancelliere, A., Di Mauro, G., Bonaccorso, B., Rossi, G., 2007. Drought forecasting using the standardized precipitation index. Water resources management 21, 801819.  Cancelliere, A., Salas, J.D., 2010. Drought probabilities and return period for annual streamflows series. Journal of Hydrology 391, 7789.  Carlisle, D.M., Meador, M.R., Short, T.M., Tate, C.M., Gurtz, M.E., Bryant, W.L., Falcone, J.A., rsstreams, 1993-2005. US Geological Survey. Carrascal, L.M., Galván, I., Gordo, O., 2009. Partial least squares regression as an alternative to current regression methods used in ecology. Oikos 118, 681690.  Cebrián, A.C., Abaurrea, J., 2006. Drought analysis based on a marked cluster Poisson model. Journal of Hydrometeorology 7, 713723.  Chen, D., Huang, J., Jackson, T.J., 2005. Vegetation water content estimation for corn and soybeans using spectral indices derived from MODIS near-and short-wave infrared bands. Remote Sensing of Environment 98, 225236.  Chen, Lu, Singh, Vijay P., and Guo, Shenglian, 2011. Drought Analysis Based on Copulas. Symposium on Data-Driven Approaches to Droughts.Paper 45. Chen, L., Singh, V.P., Shenglian, G., Hao, Z., Li, T., 2011. Flood coincidence risk analysis using multivariate copula functions. Journal of Hydrologic Engineering 17, 742755. Chowdhary, H., Singh, V.P., 2010. Reducing uncertainty in estimates of frequency distribution parameters using composite likelihood approach and copula-based bivariate distributions. Water Resources Research 46.  Combe, M., Vilà-Guerau de Arellano, J., Ouwersloot, H.G., Jacobs, C.M.J., Peters, W., 2015. Two perspectives on the coupled carbon, water and energy exchange in the planetary boundary layer. Biogeosciences 12, 103123. 183   Compin, A., Céréghino, R., 2003. Sensitivity of aquatic insect species richness to disturbance in the AdourGaronne stream system (France). Ecological Indicators 3, 135142.  Cook, E.R., Seager, R., Cane, M.A., Stahle, D.W., 2007. North American drought: reconstructions, causes, and consequences. Earth-Science Reviews 81, 93134.  Couceiro, S.R.M., Hamada, N., Forsberg, B.R., Pimentel, T.P., Luz, S.L.B., 2012. A macroinvertebrate multimetric index to evaluate the biological condition of streams in the Central Amazon region of Brazil. Ecological Indicators 18, 118125.  Cronshey, R., 1986. Urban hydrology for small watersheds. US Dept. of Agriculture, Soil Conservation Service, Engineering Division.  Dabrowska-Zielinska, K., Kogan, F., Ciolkosz, A., Gruszczynska, M., Kowalik, W., 2002. Modelling of crop growth conditions and crop yield in Poland using AVHRR-based indices. International Journal of Remote Sensing 23, 11091123.  Dai, A., Trenberth, K.E., Qian, T., 2004. A global dataset of Palmer Drought Severity Index for 1870-2002: Relationship with soil moisture and effects of surface warming. Journal of Hydrometeorology 5, 11171130.  Dai, A., 2011. Drought under global warming: a review. Wiley Interdisciplinary Reviews: Climate Change 2, 4565.  Davies, S.P., Jackson, S.K., 2006. The biological condition gradient: a descriptive model for interpreting change in aquatic ecosystems. Ecological Applications 16, 12511266.  Dawson, C.W., Wilby, R., 1998. An artificial neural network approach to rainfall-runoff modelling. Hydrological Sciences Journal 43, 4766.  De Jong, S., 1993. SIMPLS: an alternative approach to partial least squares regression. Chemometrics and intelligent laboratory systems 18, 251263.  Delbart, N., Kergoat, L., Le Toan, T., Lhermitte, J., Picard, G., 2005. Determination of phenological dates in boreal regions using normalized difference water index. Remote Sensing of Environment 97, 2638.  Dessu, S.B., Melesse, A.M., 2013. Impact and uncertainties of climate change on the hydrology of the Mara River basin, Kenya/Tanzania. Hydrological Processes 27, 29732986.  Diaz, H.F., 1983. Some aspects of major dry and wet periods in the contiguous United States, 18951981. Journal of Climate and Applied Meteorology. 22, 316.  184  Diaz-Nieto, J., Wilby, R.L., 2005. A comparison of statistical downscaling and climate change factor methods: impacts on low flows in the River Thames, United Kingdom. Climatic Change 69, 245268.  Durdu, Ö.F., 2010. Application of linear stochastic models for drought forecasting in the Büyük Menderes river basin, western Turkey. Stochastic Environmental Research and Risk Assessment 24, 11451162.  Edwards, D.C., McKee, T.B., 1997. Characteristics of 20th Century Drought in the United States at Multiple Scales. Atmospheric Science Paper No. 634, May 130. Einheuser, M.D., Nejadhashemi, A.P., Wang, L., Sowa, S.P., Woznicki, S.A., 2013. Linking biological integrity and watershed models to assess the impacts of historical land use and climate changes on stream health. Environmental management 51, 11471163.  El-Sebakhy, E.A., Raharja, I., Adem, S., Khaeruzzaman, Y., 2007. Neuro-fuzzy systems modeling tools for bacterial growth, in: IEEE/ACS International Conference on Computer Systems and Applications, pp. 374380.  Elsner, M.M., Cuo, L., Voisin, N., Deems, J.S., Hamlet, A.F., Vano, J.A., Mickelson, K.E., Lee, S.-Y., Lettenmaier, D.P., 2010. Implications of 21st century climate change for the hydrology of Washington State. Climatic Change 102, 225260.  Environment Canada, 2004. Threats to Water Availability in Canada. National Water Research Institute, Burlington, Ontario. NWRI Scientific Assessment Report Series No. 3 and ACSD Science Assessment Series No. 1. p. 128. Esfahanian, E., Nejadhashemi, A. P., Abouali, M., Daneshvar, F., Ameli, A., Herman, M., Tang, Y.,2016. Defining Drought in the context of Stream Health. Ecological Engineering 94, 668-681.  European Communities, 2012. Report on the review of the European water scarcity and drought policy. Brussels, 14.11.2012.   Fernández, B., Salas, J.D., 1999a. Return period and risk of hydrologic events. I: mathematical formulation. Journal of Hydrologic Engineering 4, 297307. Fernández, B., Salas, J.D., 1999b. Return period and risk of hydrologic events.  II:Applications. Journal of Hydrologic Engineering 4, 308316. Feyen, L., Dankers, R., 2009. Impact of global warming on streamflow drought in Europe. J. Geophys. Res. 114, D17116. Ficklin, D.L., Letsinger, S.L., Gholizadeh, H., Maxwell, J.T., 2015. Incorporation of the PenmanMonteith potential evapotranspiration method into a Palmer Drought Severity Index Tool. Computers & Geosciences 85, 136141.  185  Fleig, A.K., Tallaksen, L.M., Hisdal, H., Stahl, K., Hannah, D.M., 2010. Inter-comparison of weather and circulation type classifications for hydrological drought development. Physics and Chemistry of the Earth, Parts A/B/C 35, 507515.  Fleig, A.K., Tallaksen, L.M., Hisdal, H., Hannah, D.M., 2011. Regional hydrological drought in north-western Europe: linking a new Regional Drought Area Index with weather types. Hydrological Processes 25, 11631179.  Flinders, C.A., Horwitz, R.J., Belton, T., 2008. Relationship of fish and macroinvertebrate communities in the mid-Atlantic uplands: Implications for integrated assessments. Ecological Indicators 8, 588-598. Food and Agriculture Organization (FAO). (1983). Guidelines: Land evaluation for Rainfed Agriculture. Rome: FAO Soils Bulletin 52. Foody, G.M., Cutler, M.E., Mcmorrow, J., Pelz, D., Tangki, H., Boyd, D.S., Douglas, I., 2001. Mapping the biomass of Bornean tropical rain forest from remotely sensed data. Global Ecology and Biogeography 10, 379387. Fore, L.S., Karr, J.R., Wisseman, R.W., 1996. Assessing invertebrate responses to human activities: evaluating alternative approaches. Journal of the North American Benthological Society 212231. Fowler, H.J., Blenkinsop, S., Tebaldi, C., 2007. Linking climate change modelling to impacts studies: recent advances in downscaling techniques for hydrological modelling. International journal of climatology 27, 15471578.  Frees, E.W., Valdez, E.A., 1998. Understanding relationships using copulas. North American actuarial journal 2, 125.  Frey, D.G., 1977. Biological integrity of water: an historical approach. The integrity of water. US Environmental Protection Agency, Washington, DC 127140.  Gao, B.-C., 1996. NDWIA normalized difference water index for remote sensing of vegetation liquid water from space. Remote sensing of environment 58, 257266. Garen, D.C., 1993. Revised surface-water supply index for western United States. Journal of Water Resources Planning and Management 119, 437454.  Gassman, P.W., Reyes, M.R., Green, C.H., Arnold, J.G., 2007. The soil and water assessment tool: historical development, applications, and future research directions. Transactions of the ASABE 50, 12111250. Ghosh, S., Mujumdar, P.P., 2007. Nonparametric methods for modeling GCM and scenario uncertainty in drought assessment. Water Resources Research 43. 186  Gibbs, W., and Maher, J. 1967. Rainfall deciles as drought indicators. Bur. Meteorol. Bull. 48, 117. Gitelson, A.A., Buschmann, C., Lichtenthaler, H.K., 1998. Leaf chlorophyll fluorescence corrected for re-absorption by means of absorption and reflectance measurements. Journal of plant physiology 152, 283296. Goetz, S., Fiske, G.J., 2013. On the Relationship Between Stream Biotic Diversity and Exurbanization in the Northeastern USA, in: Geospatial Tools for Urban Water Resources. Springer, pp. 6178.  González, J., Valdés, J.B., 2003. Bivariate drought recurrence analysis using tree ring reconstructions. Journal of  Hydrologic  Engineering  8 (5), 247258. Green, I.R.A., Stephenson, D., 1986. Criteria for comparison of single event models. Hydrol Sci. J. 31, 395411. Gumbel, E.J., 1963a. Statistical forecast of droughts. Hydrological Sciences Journal 8, 523.  Gustard, A., Bullock, A., Dixon, J.M., 1992. Low flow estimation in the United Kingdom. Institute of Hydrology.  Gutman, G.G., 1990. Towards monitoring droughts from space. Journal of Climate 3, 282295.  Guttman, N.B., 1991. A sensitivity analysis of the Palmer Hydrologic Drought Index1. Wiley Online Library.  Guttman, N.B., 1998. Comparing the Palmer Drought Index and the Standardized Precipitation Index. JAWRA Journal of the American Water Resources Association 34, 113121.   Hallack-Alegria, M., Watkins Jr, D.W., 2007a. Annual and warm season drought intensity-duration-frequency analysis for Sonora, Mexico. Journal of Climate 20, 18971909.  Hamaamin, Y.A., Nejadhashemi, A.P., Einheuser, M.D., 2013. Application of fuzzy logic techniques in estimating the regional index flow for Michigan. Transactions of the ASABE 56, 103115.  internet screening tool. Fisheries Division Special Report 55.  Hardy, M.A., Maret, T.R., George, D.L., 2004. Reconnaissance of chemical and biological quality in the Owyhee River from the Oregon state line to the Owyhee Reservoir, Oregon, 200102. US Geological Survey.  Hart, D.D., Finelli, C.M., 1999. Physical-biological coupling in streams: the pervasive effects of flow on benthic organisms. Annual Review of Ecology and Systematics 363395. 187   Hayes, M.J., Svoboda, M.D., Wilhite, D.A., Vanyarkho, O.V., 1999. Monitoring the 1996 drought using the standardized precipitation index. Bulletin of the American Meteorological Society 80, 429438.  Hayes, M.J., Svoboda, M.D., Wilhite, D.A., 2000. Monitoring drought using the standardized precipitation index Drought, in: A Global Assessment, Vol. 1. Routledge, New York, pp. 168180.  Hayes, M.J., Wilhelmi, O.V., Knutson, C.L., 2004. Reducing drought risk: bridging theory and practice. Natural Hazards Review 5, 106113.  Hayes MJ. Drought indices. What Is Drought? Lincoln, Nebraska: National Drought Mitigation Center, 2006. Available at: http://www.civil.utah.edu/~cv5450/swsi/indices.htm. Heddinghaus, T.B., Sahol, P., 1991. A Review of the Palmer Drought Severity Index and Where Do We Go From Here? In: Proc. 7th Conf. on Applied Climatology, September 10-13, 1991, American Meteorological Society, Boston, Massachusetts, pp. 242246. Heim, R. R., Jr., 2000: Drought indices: A review. Drought: A Global Assessment, D. A. Wilhite, Ed., Routledge, 159167. Heim Jr, R.R., 2002. A review of twentieth-century drought indices used in the United States. Bulletin of the American Meteorological Society 83, 1149.  Hellmann, J.J., Byers, J.E., Bierwagen, B.G., Dukes, J.S., 2008. Five potential consequences of climate change for invasive species. Conservation Biology 22, 534543.  Hessami, M., Gachon, P., Ouarda, T.B., St-Hilaire, A., 2008. Automated regression-based statistical downscaling tool. Environmental Modelling & Software 23, 813834.  Hewitson, B.C., Crane, R.G., 1996. Climate downscaling: techniques and application. Climate Research 7, 8595.  Hilsenhoff, W.L., 1987. An improved biotic index of organic stream pollution. Great Lakes Entomologist 20, 3140.  Htut, A.Y., Shrestha, S., Nitivattananon, V., Kawasaki, A., 2014. Forecasting climate change scenarios in the Bago River Basin, Myanmar. Journal of Earth Science & Climatic Change 5,1.  Hunt, E.D., Hubbard, K.G., Wilhite, D.A., Arkebauer, T.J., Dutcher, A.L., 2009. The development and evaluation of a soil moisture index. International Journal of Climatology 29, 747759.  188  Ingram, J.C., Dawson, T.P., Whittaker, R.J., 2005. Mapping tropical forest structure in southeastern Madagascar using remote sensing and artificial neural networks. Remote Sensing of Environment 94, 491507.  Institute of Hydrology 1980.   Low Flow Studies report. Wallingford, UK.    IPCC, 2013: Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change [Stocker, T.F., D. Qin, G.-K. Plattner, M. Tignor, S.K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex and P.M. Midgley (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 1535 pp. Jacobi, J., Perrone, D., Duncan, L.L., Hornberger, G., 2013. A tool for calculating the Palmer drought indices. Water Resources Research 49, 60866089.  Jang, J.-S.R., 1993. ANFIS: adaptive-network-based fuzzy inference system. Systems, Man and   Cybernetics, IEEE Transactions on 23, 665685.  Jelks, H.L., S.J. Walsh, N.M. Burkhead, S. Contreras-Balderas, E. Diaz-Pardo, D.A. Hendrickson, J. Lyons, N.E. Mandrak, F. McCormick, J.S. Nelson, S.P. Plantania, B.A. Porter, C.B. Renaud, J.J. SchmitterSoto, E.b. Taylor, and M.L. Warren, Jr. 2008. Conservation status of imperiled North American freshwater and diadromous fishes. Fisheries 33 (8):372-407. Ji, L., Peters, A.J., 2004. A spatial regression procedure for evaluating the relationship between AVHRR-NDVI and climate in the northern Great Plains. International Journal of Remote Sensing 25, 297311. Johnson, W.K., Kohne, R.W., 1993. Susceptibility of reservoirs to drought using Palmer index. Journal of Water Resources Planning and Management 119, 367387.  Jones, R.G., Murphy, J.M., Noguer, M., 1995. Simulation of climate change over europe using a nested regional-climate model. I: Assessment of control climate, including sensitivity to location of lateral boundaries. Quarterly Journal of the Royal Meteorological Society 121, 14131449.  Jones, P.D., Hulme, M., Brifta, K.R., Jones, C.G., Mitchell, J.F.B., Murphy, J.B., 1996. Summer moisture accumulation over Europe in the Hadley center general circulation model based on the Palmer drought severity index. International Journal of Climatology. 16 (2), 155172.  Juhasz, T., Kornfield, J., 1978. The Crop Moisture Index: unnatural response to changes in temperature. Journal of Applied Meteorology 17, 18641866.  Justice, C.O., Townshend, J.R.G., Holben, B.N., Tucker, C.J., 1985. Analysis of the phenology of global vegetation using meteorological satellite data. International Journal of Remote Sensing 6, 12711318.  189  Kaehler, S.D., 2006. Fuzzy Logic Tutorial, the newsletter of Seattle Robotics Society (encoder). Available at: http://www.seattlerobotics.org/encoder/mar98/fuz/flindex.html  Karamouz, M., Rasouli, K., Nazif, S., 2009. Development of a hybrid index for drought prediction: case study. Journal of Hydrologic Engineering 14, 617627.  Karl, T.R., Quayle, R.G., 1981. The 1980 summer heat wave and drought in historical perspective. Monthly Weather Review 109, 20552073.  Karl, T.R., Koscielny, A.J., 1982. Drought in the United States: 18951981. Journal of Climatology 2, 313329.  Karl, T.R., Knight, R.W., 1985. Atlas of Monthly Palmer Hydrological Drought Indices (19311983) for the Contiguous United States. Historical Climatology Series 37, National Climatic Data Center, Asheville, North Carolina.  -index to their calibration coefficients including potential evapotranspiration. Journal of Climate and Applied Meteorology 25, 7786.  Karr, J.R., and D.R. Dudley. 1981. Ecological perspectives on water quality goals. Environmental Management 5:5568.   Karr, J.R., 1991. Biological integrity: a long-neglected aspect of water resource management. Ecological applications 1, 6684. Karr, J.R., 1996. Ecological integrity and ecological health are not the same. Engineering within ecological constraints 97, 109.  Kerans, B.L., Karr, J.R., 1994. A benthic index of biotic integrity (B-IBI) for rivers of the Tennessee Valley. Ecological applications 4, 768785.  Keyantash, J., Dracup, J.A., 2002. The quantification of drought: an evaluation of drought indices. The drought monitor. Bull. Am. Meteorol. Soc. 83 (8), 1167 1180. Keyantash, J.A., Dracup, J.A., 2004. An aggregate drought index: Assessing drought severity based on fluctuations in the hydrologic cycle and surface water storage. Water Resources Research 40.  Khan, M.S., Coulibaly, P., 2010. Assessing hydrologic impact of climate change with uncertainty estimates: Bayesian neural network approach. Journal of Hydrometeorology 11, 482495.  Kim, T.-W., Valdés, J.B., 2003. Nonlinear model for drought forecasting based on a conjunction of wavelet transforms and neural networks. Journal of Hydrologic Engineering 8, 319328.  190  Kim, T.-W., Valdés, J.B., Yoo, C., 2003. Nonparametric approach for estimating return periods of droughts in arid regions. Journal of Hydrologic Engineering 8, 237246.  Kim, T.-W., Valdés, J.B., Yoo, C., 2006. Nonparametric approach for bivariate drought characterization using Palmer drought index. Journal of Hydrologic Engineering 11, 134143.  Kira, K., Rendell, L.A., 1992. A practical approach to feature selection, in: Proceedings of the Ninth International Workshop on Machine Learning. pp. 249256.  approach. Hydrological Processes 20, 43514362. Klugman, M.R., 1978. Drought in the upper Midwest, 1931-1969. Journal of Applied Meteorology 17, 14251431. Knutson, C., Hayes, M., Phillips, T., 1998. How to Reduce Drought Risk. Western Drought Coordination Council: Preparedness and Mitigation Working Group.  Kogan, F.N., 1987. Vegetation index for a real analysis of crop conditions. In: Proceedings of the 18th Conference on Agricultural and Forest Meteorology, AMS, W. Lafayette, Indiana, 1518 September 1987, Indiana, USA, pp. 103106. Kogan, F.N., 1990. Remote sensing of weather impacts on vegetation in non-homogeneous areas. International Journal of Remote Sensing 11, 14051419.  Kogan, F., Sullivan, J., 1993. Development of global drought-watch system using NOAA/AVHRR data. Advances in Space Research 13, 219222. Kogan, F.N., 1995. Droughts of the late 1980s in the United States as derived from NOAA polar-orbiting satellite data. Bulletin of the American Meteorological Society 76, 655668.  Kogan, F.N., 1997. Global drought watch from space. Bulletin of the American Meteorological Society 78, 621636. Kogan, F.N., 2005. NOAA/AVHRR satellite data based indices for monitoring agricultural droughts. New York: Oxford University Press.  Kononenko, I., 1994. Estimating attributes: analysis and extensions of RELIEF, in: Machine Learning: ECML-94. Springer, pp. 171182.  Kumar, V., Panu, U., 1997. Predictive assessment of severity of agricultural droughts based on agro-climate factors1. JAWRA Journal of the American Water Resources Association 33, 12551264.  Kuznetsova, A., Brockhoff, P.B., Christensen, R.H.B., 2016. lmerTest: tests in linear mixed effects models. R package version 2.0-30.  191  Lake, P.S., 2003. Ecological effects of perturbation by drought in flowing waters. Freshwater biology 48, 11611172.  Lammert, M., Allan, J.D., 1999. Assessing biotic integrity of streams: effects of scale in measuring the influence of land use/cover and habitat structure on fish and macroinvertebrates. Environmental Management 23, 257270.  Launois, L., Veslot, J., Irz, P., Argillier, C., 2011. Development of a fish-based index (FBI) of biotic integrity for French lakes using the hindcasting approach. Ecological Indicators 11, 15721583.  Lawson, M.P., Reiss, A., Phillips, R., Livingston, K., 1971. Nebraska Droughts A Study of Their Past Chronological and Spatial Extent with Implications for the Future. Occasional Papers No. 1. Department of Geography, University of Nebraska, Lincoln, p. 147.  Lehner, B., Döll, P., Alcamo, J., Henrichs, T., Kaspar, F., 2006. Estimating the impact of global change on flood and drought risks in Europe: a continental, integrated analysis. Climatic Change 75, 273299.  Lenat, D.R., 1988. Water quality assessment of streams using a qualitative collection method for benthic macroinvertebrates. Journal of the North American Benthological Society 222233.  Lenat, D.R., 1993. A biotic index for the southeastern United States: derivation and list of tolerance values, with criteria for assigning water-quality ratings. Journal of the North American Benthological Society 279290.  Liu, W.T., Juárez, R.N., 2001. ENSO drought onset prediction in northeast Brazil using NDVI. International Journal of Remote Sensing 22, 34833501.  Livada, L., Assimakopoulos, V.D., 2007. Spatial and temporal analysis of drought in Greece using the standardized precipitation index (SPI). Theoretical and Applied Climatology 89, 143-153.   Lloyd-Hughes, B., Saunders, M.A., 2002. A drought climatology for Europe. International journal of climatology 22, 15711592.  Loeb, S. L., and A. Spacie, editors. 1994. Biological monitoring of aquatic systems. Lewis, Boca Raton, Florida, USA. Loukas, A., Vasiliades, L., Tzabiras, J., 2007. Evaluation of climate change on drought impulses in Thessaly, Greece. European Water 17, 1728.  Luo, L., Wood, E.F., 2007. Monitoring and predicting the 2007 US drought. Geophysical Research Letters 34, L22702.  192  Luo, L., Wood, E.F., 2008. Use of Bayesian mering techniques in a multimodel seasonal hydrologic ensemble prediction system for the Eastern United States. Journal of Hydrometeorology. 9, 866884.  Madsen, H., Mikkelsen, P.S., Rosbjerg, D., Harremoës, P., 1998. Estimation of regional intensity-duration-frequency curves for extreme precipitation. Water Science and Technology 37, 2936.  Mahlein, A.-K., Rumpf, T., Welke, P., Dehne, H.-W., Plümer, L., Steiner, U., Oerke, E.-C., 2013. Development of spectral indices for detecting and identifying plant diseases. Remote Sensing of Environment 128, 2130.  Maraun, D., Wetterhall, F., Ireson, A.M., Chandler, R.E., Kendon, E.J., Widmann, M., Brienen, S., Rust, H.W., Sauter, T., Theme\s sl, M., others, 2010. Precipitation downscaling under climate change: Recent developments to bridge the gap between dynamical models and the end user. Reviews of Geophysics 48.  Marchini, A., 2011. Modelling ecological processes with fuzzy logic approaches, in: Modelling Complex Ecological Dynamics. Springer, pp. 133145.  Martínez-Fernández, J., González-Zamora, A., Sánchez, N., Gumuzzio, A., 2015. A soil water based index as a suitable agricultural drought indicator. Journal of Hydrology 522, 265273.  Matallo, H., 2013, Drought vulnerability and risk assessment. United nations convention to combat desertification. MathWorks, 2016. Fuzzy Logic Toolbox. Available at: http://www.mathworks.com/products/fuzzy-logic/ (accessed 4.12.16).  McKee, T.B., Doesken, N.J., Kleist, J., others, 1993. The relationship of drought frequency and duration to time scales, in: Proceedings of the 8th Conference on Applied Climatology. American Meteorological Society Boston, MA, USA, pp. 179183.  McKee, T.B., Doesken, N.J., Kleist, J., 1995. Drought monitoring with multiple time scales, in: Proceedings of the 9th Conference on Applied Climatology. American Meteorological Society Dallas, Boston, MA, pp. 233236.  McMahon, T.A., Finlayson, B.L., 2003. Droughts and anti-droughts: the low flow hydrology of Australian rivers. Freshwater Biology 48, 11471160.  Meador, M.R., Carlisle, D.M., Coles, J.F., 2008. Use of tolerance values to diagnose water-quality stressors to aquatic biota in New England streams. Ecological Indicators 8, 718728.  Mebane, C.A., Maret, T.R., Hughes, R.M., 2003. An index of biological integrity (IBI) for Pacific Northwest rivers. Transactions of the American Fisheries Society 132, 239261.  193  Meigh, J.R., 1987. Low flow analysis of selected catchments in Malawi and Zimbabwe. Preliminary report on the Basement Aquifer Project, Institute of Hydrology, Wallingford, UK.  Merabtene, T., Kawamura, A., Jinno, K., Olsson, J., 2002. Risk assessment for optimal drought management of an integrated water resources system using a genetic algorithm. Hydrological Processes 16, 21892208.  Merritt, W.S., Alila, Y., Barton, M., Taylor, B., Cohen, S., Neilsen, D., 2006. Hydrologic response to scenarios of climate change in sub watersheds of the Okanagan basin, British Columbia. Journal of Hydrology 326, 79108.  Meyer, J.L., 1997. Stream health: incorporating the human dimension to advance stream ecology. Journal of the North American Benthological Society 439447.  Michele, C., Salvadori, G., Vezzoli, R., Pecora, S., 2013. Multivariate assessment of droughts: frequency analysis and dynamic return period. Water Resources Research 49, 69856994.  Miller, D.L., Hughes, R.M., Karr, J.R., Leonard, P.M., Moyle, P.B., Schrader, L.H., Thompson, B.A., Daniels, R.A., Fausch, K.D., Fitzhugh, G.A., others, 1988. Regional applications of an index of biotic integrity for use in water resource management. Fisheries 13, 1220.  Mishra, A.K., Desai, V.R., 2005a. Drought forecasting using stochastic models. Stochastic Environmental Research and Risk Assessment 19, 326339.  Mishra, A.K., Desai, V.R., 2005b. Spatial and temporal drought analysis in the Kansabati river basin, India. International Journal of River Basin Management 3, 3141.  Mishra, A.K., Desai, V.R., 2006. Drought forecasting using feed forward recursive neural network. Int. J. Ecol. Model. 198, 127138.  Mishra, A.K., Desai, V.R., Singh, V.P., 2007. Drought forecasting using a hybrid stochastic and neural network model. Journal of Hydrologic Engineering 12, 626638.  Mishra, A.K., Coulibaly, P., 2009. Developments in hydrometric network design: A review. Reviews of Geophysics 47.  Mishra, A.K., Singh, V.P., 2009a. Analysis of drought severityareafrequency curves using a general circulation model and scenario uncertainty. Journal of Geophysical Research. 114, D06120.   Mishra, A.K., Singh, V.P., Desai, V.R., 2009b. Drought characterization: a probabilistic approach. Stochastic Environmental Research and Risk Assessment. 23 (1), 4155. Mishra, A.K., Singh, V.P., 2010. A review of drought concepts. Journal of Hydrology 391, 202216.  194  Mishra, A.K., Singh, V.P., 2011. Drought modeling  A review. Journal of Hydrology 403, 157-175.  Moorhead, J.E., Gowda, P.H., Singh, V.P., Porter, D.O., Marek, T.H., Howell, T.A., Stewart, B.A., 2015. Identifying and Evaluating a Suitable Index for Agricultural Drought Monitoring in the Texas High Plains. JAWRA Journal of the American Water Resources Association 51, 807820.  Moriasi, D.N., Arnold, J.G., Van Liew, M.W., Bingner, R.L., Harmel, R.D., Veith, T.L., 2007. Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Transactions of the ASABE 50, 885900.  networks and time series of drought indices. International Journal of Climatology. 27 (15), 2103. Morrison, J., Quick, M.C., Foreman, M.G., 2002. Climate change in the Fraser River watershed: flow and temperature projections. Journal of Hydrology 263, 230244.  Moss, R.H., Edmonds, J.A., Hibbard, K.A., Manning, M.R., Rose, S.K., Van Vuuren, D.P., Carter, T.R., Emori, S., Kainuma, M., Kram, T., others, 2010. The next generation of scenarios for climate change research and assessment. Nature 463, 747756.  year fish to river obstacles: Functional and numerical linkages between dams, weirs, fish habitat guilds and biotic integrity across large spatial scale. Ecological indicators 23, 634640.  Nagarajan, R., 2009, Drought assessment. Center of studies in Resources Engineering Indian Institute of Technology Bombay, India. Springer.   Narasimhan, B., Srinivasan, R., 2005. Development and evaluation of Soil Moisture Deficit Index (SMDI) and Evapotranspiration Deficit Index (ETDI) for agricultural drought monitoring. Agricultural and Forest Meteorology 133, 6988.  Nathan, R.J., McMahon, T.A., 1990a. Evaluation of automated techniques for base flow and recession analyses. Water Resources Research. 26, 14651473.  National Oceanic and Atmospheric Administration (NOAA), 2012. Frequently asked questions about El Nino and La Nina. Available at: http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensofaq.shtml  National Research Council, 2007. Earth Science and Application from space: National Imperatives for the Next Decade and Beyond, ed. R.A. Andres and B. Moore III, 465 pp, National Academic Press, Washington, D.C. National Water and Soil Conservation Authority 1984.  An index for base flows. Streamland 24, Water and soil directorate, Ministry of Works and Dev., Wellington, N.Z.  195  Navarro-Llácer, C., Baeza, D., de las Heras, J., 2010. Assessment of regulated rivers with indices based on macroinvertebrates, fish and riparian forest in the southeast of Spain. Ecological Indicators 10, 935942.  NCDC, 2013. Climate of 2013-April, U.S. Palmer Drought Indices. Available at: https://www.ncdc.noaa.gov/oa/climate/research/prelim/drought/palmer.html   NDMC, 2016. Types of Drought. Available at: http://drought.unl.edu/DroughtBasics/TypesofDrought.aspx (accessed 4.12.16).  NED, 2014. National Elevation Dataset. Available at: http://nationalmap.gov/elevation.html (accessed 4.12.16).  Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2011. Soil and water assessment tool theoretical documentation version 2009. Texas Water Resources Institute.  Nerbonne, B. A. & B. Vondracek, 2001. Effects of local land use on physical habitat, benthic macroinvertebrates, and fish in the Whitewater River, Minnesota, USA. Environmental Management 28: 8799. Niemeyer, S., 2008. New drought indices. Water Management 80, 267274. NRCS, 2007. Chapter 7: Hydrologic Soil Groups. National Engineering Handbook: Section 4: Part 630, Hydrologic Engineering. Washington, D.C.: U.S. Dept. of Agriculture, Soil Conservation Service. NRCS, 2014.Web Soil Survey. Available at:   http://websoilsurvey.sc.egov.usda.gov/App/WebSoilSurvey.aspx (accessed 4.12.16).  Ochola, W.O., Kerkides, P., 2003. A Markov chain simulation model for predicting critical wet and dry spells in Kenya: analysing rainfall events in the Kano plains. Irrigation and Drainage 52 (4), 327342. Özger, M., Mishra, A.K., Singh, V.P., 2009. Low frequency drought variability associated with climate indices. Journal of Hydrology 364, 152162.  Pallottino, S., Secchi, G.M., Zuddas, P., 2005. A DSS for water resources management under uncertainty by scenario analysis. Environ. Model. Softw. 20 (8), 1031 1042.  Palmer, W.C., 1965. Meteorological drought. Weather Bureau Research Paper No. 45, US Department of Commerce, Washington, DC.58 pp. Palmer, W.C., 1967. The abnormally dry weather of 19611966 in the Northeastern United States. In: Jerome, S. (Ed.), Proceedings of the Conference on the Drought in the Northeastern United States, New York University Geophysical Research Laboratory Report TR-68-3, pp. 3256. 196  Palmer, W.C. 1968. Keeping track of crop moisture conditions, nationwide: The new crop moisture index. Weatherwise, 21 (4): 156161. Panu, U.S., Sharma, T.C., 2002. Challenges in drought research: some perspectives and future directions. Hydrological Sciences Journal 47, S19S30.  Parker, W.S., 2013. Ensemble modeling, uncertainty and robust predictions. Wiley Interdisciplinary Reviews: Climate Change 4, 213223.  United Nations Development Programme. Paulo, A.A., Ferreira, E., Coelho, C., Pereira, L.S., 2005. Drought class transition analysis through Markov and Loglinear models, an approach to early warning. Agric. Water Manage. 77, 5981. Pelletier, M.C., Gold, A.J., Gonzalez, L., Oviatt, C., 2012. Application of multiple index development approaches to benthic invertebrate data from the Virginian Biogeographic Province, USA. Ecological Indicators 23, 176188. Petkova, E.P., Horton, R.M., Bader, D.A., Kinney, P.L., 2013. Projected heat-related mortality in the US urban northeast. International journal of environmental research and public health 10, 67346747.  Pilon, P.J., Condie, R., 1986. Median drought flows at ungauged sites in southern Ontario, in: Proceedings of the 16th Canadian Hydrology Symposium: Droughtthe Impending Crisis. pp. 189196.  Pinheiro, J., Bates, D., 2006. Mixed-effects models in S and S-PLUS. Springer Science & Business Media.  Poff, N.L., Richter, B.D., Arthington, A.H., Bunn, S.E., Naiman, R.J., Kendy, E., Acreman, M., Apse, C., Bledsoe, B.P., Freeman, M.C., others, 2010. The ecological limits of hydrologic alteration (ELOHA): a new framework for developing regional environmental flow standards. Freshwater Biology 55, 147170.  PSC, 2012. Saginaw Bay Watershed and Area of Concern. Prepared by Public Sector Consultants, Inc. for the Planning & Zoning Center at MSU as part of a Great Lakes Restoration Initiative project. Lansing, MI. Available at: http://www.pscinc.com/LinkClick.aspx?fileticket=Hi_PFZU-xOI%3D&tabid=65 (accessed 3.01.16).  Qiao, F., Song, Z., Bao, Y., Song, Y., Shu, Q., Huang, C., Zhao, W., 2013. Development and evaluation of an Earth System Model with surface gravity waves. Journal of Geophysical Research: Oceans 118, 45144524.  197  Quiring, S.M., Ganesh, S., 2010. Evaluating the utility of the Vegetation Condition Index (VCI) for monitoring meteorological drought in Texas. Agricultural and Forest Meteorology 150, 330339.  Rao, A.R., PJournal of hydrology 68, 211229.  Refsgaard, J.C., van der Sluijs, J.P., Højberg, A.L., Vanrolleghem, P.A., 2007. Uncertainty in the environmental modelling processa framework and guidance. Environmental modelling & software 22, 15431556.  Ripple, W. J. ,1985. Asymptotic relationship characteristics of grass vegetation. Photogrammetric Engineering and Remote Sensing, 51, 1915  1921  Robnik-oretical and empirical analysis of ReliefF and RReliefF. Machine learning 53, 2369.  Robock, A., Luo, L., Wood, E.F., Wen, F., Mitchell, K.E., Houser, P.R., Schaake, J.C., Lohmann, D., Cosgrove, B., Sheffield, J., others, 2003. Evaluation of the North American Land Data Assimilation System over the southern Great Plains during the warm season. Journal of Geophysical Research: Atmospheres 108.  Rodell, M. (2012). Satellite Gravimetry Applied to Drought Monitoring. In M. C. Brian D. Wardlow, Remote Sensing of Drought. Taylor & Francis Group. National Research Council, Transforming Remote Sensing Data into Information and Applications. Washington, DC: The National Academies Press.  Root, T.L., Price, J.T., Hall, K.R., Schneider, S.H., Rosenzweig, C., Pounds, J.A., 2003. Fingerprints of global warming on wild animals and plants. Nature 421:57-60.  Rossi, G., Cancelliere, A., Giuliano, G., 2005. Case study: multicriteria assessment of drought mitigation measures. Journal of Water Resources Planning and Management. 131 (6), 449457. Rouse, J.W. Jr., R.H. Haas, D.W. Deering, J.A. Schell, and J.C. Harlan, 1974. Monitoring the vernal advancement and retrogradation (green wave effect) of natural vegetation. NASA/ GSFC type III final report. National Aeronautics and Space Administration, Greenbelt, MD. Royal Society, 2014. Climate change evidence and causes. An overview from the Royal Society and the US National Academy of Sciences. Sailor, D.J., Hu, T., Li, X., Rosen, J.N., 2000. A neural network approach to local downscaling of GCM output for assessing wind power implications of climate change. Renewable Energy 19, 359378.  198  Salas, J.D., Fu, C., Cancelliere, A., Dustin, D., Bode, D., Pineda, A., Vincent, E., 2005. Characterizing the severity and risk of drought in the Poudre River, Colorado. Journal of Water Resources Planning and Management 131, 383393.  Schoof, J.T., Pryor, S.C., 2001. Downscaling temperature and precipitation: A comparison of regression-based methods and artificial neural networks. International Journal of Climatology 21, 773790.  Seiler, R.A., Kogan, F., Wei, G., 2000. Monitoring weather impact and crop yield from NOAA AVHRR data in Argentina. Advances in Space Research 26, 11771185.  Selzer, M.D., Joldersma, B., Beard, J., 2014. A reflection on restoration progress in the Saginaw Bay watershed. Journal of Great Lakes Research 40, 192200.  Journal of hydrology 46, 265279.  Serinaldi, F., Bonaccorso, B., Cancelliere, A., Grimaldi, S., 2009. Probabilistic characterization of drought properties through copulas. Physics and Chemistry of the Earth 34 (1012), 596605. Schneider, S.H. (Ed.), 1996. Encyclopaedia of Climate and Weather. Oxford University Press, New York. Shafer, B.A., Dezman, L.E., 1982. Development of a Surface Water Supply Index (SWSI) to assess the severity of drought conditions in snowpack runoff areas, in: Proceedings of the Western Snow Conference. Colorado State University Fort Collins, CO, pp. 164175.  Sheffield, J., Wood, E.F., 2011. Drought: past problems and future scenarios. Earthscan Publication Ltd.  Shiau, J.-T., Shen, H.W., 2001. Recurrence analysis of hydrologic droughts of differing severity. Journal of Water Resources Planning and Management 127, 3040.  Shiau, J.T., 2006. Fitting drought duration and severity with two-dimensional copulas. Water Resources Management. 20, 795815.  Shiau, J.-T., Modarres, R., 2009. Copula-based drought severity-duration-frequency analysis in Iran. Meteorological Applications 16, 481489.  Shin, H.-S., Salas, J.D., 2000. Regional drought analysis based on neural networks. Journal of Hydrologic Engineering 5, 145155.  Shukla, S., Wood, A.W., 2008. Use of a standardized runoff index for characterizing hydrologic drought. Geophysical Research Letters 35.  199  Singh, V.P., et al., 2002. Vegetation and temperature condition indices from NOAA AVHRR data for drought monitoring over India. International Journal of Remote Sensing, volume 24, No. 22, 4393-4402.   Singh, R.P., Roy, S., Kogan, F., 2003. Vegetation and temperature condition indices from NOAA AVHRR data for drought monitoring over India. International Journal of Remote Sensing 24, 43934402.  Smakhtin, V.U., 2001. Low flow hydrology: a review. Journal of hydrology 240, 147186.  SOEST (University of Hawaii, school of ocean, earth science and technology), 2003. Drought risk and vulnerability assessment and GIS mapping project.  Commision on Water Resource Management, Department of Land and Natural Resources State of Hawaii.   Song, S.B., Singh, V.P., 2010. Frequency analysis of droughts using the Plackett copula and parameter estimation by genetic algorithm. Stochastic Environmental Research and Risk Assessment. 24, 783805.  Soulé, P.T., 1992. Spatial patterns of drought frequency and duration in the contiguous USA based on multiple drought event definitions. International Journal of Climatology 12, 1124.  Stanford, L.L., Spacie, A., 1994. Biological monitoring of aquatic systems. CRC Press.  Stewart-Koster, B., Bunn, S.E., Mackay, S.J., Poff, N.L., Naiman, R.J., Lake, P.S., 2010. The use of Bayesian networks to guide investments in flow and catchment restoration for impaired river ecosystems. Freshwater Biology 55, 243260.  Stocker, T.F., Qin, D., Plattner, G.K., Tignor, M., Allen, S.K., Boschung, J., Nauels, A., Xia, Y., Bex, B., Midgley, B.M., 2013. IPCC, 2013: climate change 2013: the physical science basis. Contribution of working group I to the fifth assessment report of the intergovernmental panel on climate change.  Svoboda, M., 2000. An introduction to the drought monitor. Drought Network News 12, 1520. Svoboda, M., LeComte, D., Hayes, M., Heim, R., others, 2002. The drought monitor. Bulletin of the American Meteorological Society 83, 1181-1190.  Tadesse, T., Wardlow, B., 2007. The vegetation outlook (VegOut): a new tool for providing outlooks of general vegetation conditions using data mining techniques. In seventh IEEE International Conference on Data Mining Workshops (ICDM Workshops), IEEE, pp. 667672.  Takagi, T., Sugeno, M., 1985. Fuzzy identification of systems and its applications to modeling and control. Systems, Man and Cybernetics, IEEE Transactions on 116132.  Tallaksen, L. 1986.  An evaluation of the Base Flow Index (BFI). Dept. of Geography, University of Oslo.  200  Tallaksen, L.M., Madsen, H., Clausen, B., 1997. On the definition and modeling of stream drought duration and deficit volume. Hydrological Sciences JournalJournal des Sciences Hydrologiques 42 (1), 1533.  Tallaksen, L.M., Van Lanen, H.A., 2004. Hydrological drought: processes and estimation methods for streamflow and groundwater. Developments in Water Sciences 48, Elsevier B.V., Amesterdam.  Tarpley, J.D., Schneider, S.R., Money, R.L., 1984. Global vegetation indices from the NOAA-7 meteorological satellite. Journal of Climate and Applied Meteorology 23, 491494.  Taylor, K.E., Stouffer, R.J., Meehl, G.A., 2012. An overview of CMIP5 and the experiment design. Bulletin of the American Meteorological Society 93, 485498.  Todorovic, P., Woolhiser, D., 1976. Stochastic structure of the local pattern of precipitation. In: Shen, H.W. (Ed.), Stochastic Approaches to Water Resources, Vol. 2. Colorado State University, Fort Collins, CO.  Traore, Z.N., Fontane, D.G., 2007. Managing drought impacts: Case study of Mali, Africa. Journal of Water Resources Planning and Management 133, 300308.  Tsakiris, G., Vangelis, H., 2005. Establishing a drought index incorporating evapotranspiration. European Water 9, 311.  Tsakiris, G., Loukas, A., Pangalou, D., Vangelis, H., Tigkas, D., Rossi, G., Cancelliere, A., others, 2007. Drought characterization. Drought management guidelines technical annex 85102.  Tucker, C.J., Choudhury, B.J., 1987. Satellite remote sensing of drought conditions. Remote sensing of Environment 23, 243251.  U.S. Department of Agriculture, National Agricultural Statistical Service, 2011. National Agriculture Statistic Services. Available at: http://www.nass.usda.gov/Statistics_by_State/Michigan/Publications/Annual_Statistical_Bulletin/stats11/cntyrnk.pdf (accessed 3.01.16).  Unganai, L.S., Kogan, F.N., 1998. Drought Monitoring and Corn Yield Estimation in Southern Africa from AVHRR Data. Remote sensing environment, 63:219-232.  Unger, P.W., 1984. Guidelines: land evaluation for rainfed agriculture. Food & Agriculture Org.  UN-ISDR (United Nations International Strategy for Disaster Reduction). 2009.  UNISDR terminology on disaster risk reduction. Geneva: UNISDR. 30 p. Available at:   http://www.preventionweb.net/files/7817_UNISDRTerminologyEnglish.pdf (accessed 8.8.14).   201  University of Hawaii, school of ocean, earth science and technology (SOEST), 2003. Drought risk and vulnerability assessment and GIS mapping project.  Commision on Water Resource Management.  UN Secretariat General, 1994. United Nations Convention to Combat Drought and  Desertification in Countries Experiencing Serious Droughts and/or Desertification, Particularly in Africa. Paris. U.S. climate data, 2016. Available at: http://www.usclimatedata.com/ (accessed 2.2.16). USEPA (U.S. Environmental Protection Agency), 2000a. Stressor Identification Guidance Document. EPA/82/B-00/025. Wahington, D.C. USEPA (U.S. Environmental Protection Agency), 2011a. An approach for estimating stream health using flow duration curves and indices of hydrologic alteration. AgriLIFE RESEARCH & EXTENSION, Texas A&M System, 21- 24.  USEPA (U.S. Environmental Protection Agency), 2011b, A primer on Using biological assessment to support water quality management, Office of Water, Washington, DC EPA 810-R-11-01. USEPA (U.S. Environmental Protection Agency), 2012a. Chapter 4 Macroinvertebrates and Habitat. Available at: http://water.epa.gov/type/rsl/monitoring/vms40.cfm USEPA (U.S. Environmental Protection Agency), 2012b. Implications of climate change for bioassessment programs and approaches to account for effects. Global Change Research Program, National Center for Environmental Assessment, Washington, DC; EPA/600/R-11/036A. Available from the National Technical Information Service, Springfield, VA, and online at http://www.epa.gov/ncea.  USEPA (U.S. Environmental Protection Agency), 2013. Great Lakes areas of concerns. Saginaw River and bay area of concern. Available at: http://www.epa.gov/glnpo/aoc/saginaw-river/index.html USEPA (U.S. Environmental Protection Agency), 2014, Summary of the clean water act. Available at: http://www2.epa.gov/laws-regulations/summary-clean-water-act USEPA (U.S. Environmental Protection Agency), 2015. Great Lakes areas of concerns. Saginaw River and bay area of concern. Available at:  http://www.epa.gov/glnpo/aoc/saginaw-river/index.html (accessed 1.1.16).  USEPA (U.S. Environmental Protection Agency), 2016. CADDIS Volume 1: Stressor identification. Available at: https://www3.epa.gov/caddis/si_home.html (accessed 3.1.16).  202  Van Lanen, H.A.J., Peters, E., 2000. Definition, effects and assessment of groundwater droughts. In: Vogt, J.V., Somma, F. (Eds.), Drought and Drought Mitigation in Europe. Kluwer Academic Publishers, Dordrecht, pp. 4961. Vannote, R.L., Minshall, G.W., Cummins, K.W., Sedell, J.R., Cushing, C.E., 1980. The river continuum concept. Canadian journal of fisheries and aquatic sciences 37, 130137.  Vasiliades, L., Loukas, A., Liberis, N., 2011. A water balance derived drought index for Pinios River Basin, Greece. Water Resources Management 25, 10871101.  Vicente-Serrano, S.M., 2006. Differences in spatial patterns of drought on different time scales: an analysis of the Iberian Peninsula. Water Resour. Manage. 20, 37 60.  Walther, G.-R., Post, E., Convey, P., Menzel, A., Parmesan, C., Beebee, T. J. C., Fromentin, J.- M., Hoegh-Gudberg, O. & Bairlein, F. ,2002.Nature 416, 389395. Wan, H., Chizinski, C.J., Dolph, C.L., Vondracek, B., Wilson, B.N., 2010. The impact of rare taxa on a fish index of biotic integrity. Ecological Indicators 10, 781788. Wardlow, B.D., Anderson, M.C., Verdin, J.P., 2012. Remote sensing of drought: Innovative monitoring approaches. CRC Press.  Chikira, M., Ogura, T., Sekiguchi, M., others, 2010. Improved climate simulation by MIROC5: mean states, variability, and climate sensitivity. Journal of Climate 23, 63126335. Available at: http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensofaq.shtml Weghorst, K. 1996. The reclamation drought index: guidelines and practical applications. cedb.asce.org, ASCE, Denver, Colo. Wetterhall, F., S. Halldin, and C. Xu (2005), Statistical precipitation downscaling in central Sweden with the analogue method, J. Hydrol., 306, 174-190.   Whitmore, J.S., 2000. Drought management on farmland. Springer Science & Business Media.  Wilby, R. L., H. Hassan, and K. Hanaki, 1998, Statistical downscaling of  hydrometeorological variables using general circulation model output, J. Hydrol., 205, 1  19.  Wilby, R.L., Hay, L.E., Gutowski, W.J., Arritt, R.W., Takle, E.S., Pan, Z., Leavesley, G.H., Clark, M.P., 2000. Hydrological responses to dynamically and statistically downscaled climate model output. Geophysical Research Letters 27, 11991202.  Wilhite, D.A., Glantz, M.H., 1985. Understanding: the drought phenomenon: the role of definitions. Water international 10, 111120.  203  Wilhite, DA, 1991. Drought planning: A process for state government. Water Resources Bulletin 27(1):2938. Wilhite, D.A., 2000a. Drought as a natural hazard: concepts and definitions in: A Global Assessment, Vol. 1. Routledge, New York, pp. 1-18.  Wilhite, D.A., 2000b. Drought: A Global Assessment, Vols. 1 and 2. Routledge, New York, 89-104, 1 and 2, Routledge, New York, pp. 129448.  Wilhite, D.A., Hayes, M.J., and Knutson, C.L., 2000c. Drought preparedness planning: building institutional capacity. National Drought Mitigation Center.  Wilhite, D.A., Sivakumar, M.V., Pulwarty, R., 2014. Managing drought risk in a changing climate: The role of national drought policy. Weather and Climate Extremes 3, 413.  Winkler, J. A., Guentchev, G., S., Perdinan, Tan P., -N., Zhong, S., Liszewska, M., Abraham, Z., Niedz´wiedz´, T., Ustrnul, Z., 2011a, Climate scenario development and applications for local/regional climate change impact assessments: An overview for the non-climate scientist. Part I: Scenario development using downscaling methods. Geography Compass 5(6) 275-300.  Winkler, J. A., Guentchev, G. S., Liszewska, M., Perdinan, Tan, P. N., 2011b. Climate scenario development and applications for local/regional climate change impact assessments: An overview for the non-climate scientist. Part II: Considerations when using climate change scenarios. Geography Compass 5(6), 301-328.  Wold, S., Sjöström, M., Eriksson, L., 2001. PLS-regression: a basic tool of chemometrics. Chemometrics and intelligent laboratory systems 58, 109130.  World Meteorological Organization (WMO), 1986. Report on Drought and Countries Affected by Drought During 19741985, WMO, Geneva, p. 118.  Woznicki, S.A., Nejadhashemi, A.P., 2012. Sensitivity Analysis of Best Management Practices Under Climate Change Scenarios1. Wiley Online Library.  Woznicki, S.A., Pouyan Nejadhashemi, A., 2014. Assessing uncertainty in best management practice effectiveness under future climate scenarios. Hydrological Processes 28, 25502566.  Woznicki, S.A., Nejadhashemi, A.P., Ross, D.M., Zhang, Z., Wang, L., Esfahanian, A.-H., 2015. Ecohydrological model parameter selection for stream health evaluation. Science of the Total Environment 511, 341353.  Woznicki, S.A., Nejadhashemi, A.P., Abouali, M., Herman, M.R., Esfahanian, E., Hamaamin, Y.A., Zhang, Z., 2016. Ecohydrological modeling for large-scale environmental impact assessment. Science of the Total Environment 543, 274286.  204  Wu, H., Wilhite, D.A., 2004. An operational agricultural drought risk assessment model for Nebraska, USA. Natural Hazards 33, 121.  Wu, H., Hayes, M.J., Wilhite, D.A., Svoboda, M.D., 2005. The effect of the length of record on the standardized precipitation index calculation. International Journal of Climatology 25, 505520. Wu, H., Svoboda, M.D., Hayes, M.J., Wilhite, D.A., Wen, F., 2007. Appropriate application of the standardized precipitation index in arid locations and dry seasons. International Journal of Climatology 27, 6579. Yanovitzky, I., VanLear, A., 2007. Time Series Analysis: Traditional and Contemporary Approaches, Chapter 4 in Advanced data analysis methods for communication research book. Yevjevich, V.M., 1967. An objective approach to definitions and investigations of continental hydrologic droughts. Hydrology Paper No. 23, Colorado State University. Zaidman, M. D., et al., 2001. Spatio-temporal development of streamflow droughts in north-west Europe. Hydrology and Earth System Sciences, 5(4), 733-751. Zarch, M.A.A., Malekinezhad, H., Mobin, M.H., Dastorani, M.T., Kousari, M.R., 2011. Drought monitoring by reconnaissance drought index (RDI) in Iran. Water resources management 25, 34853504. Zargar, A., Sadiq, R., Naser, B., Khan, F.I., 2011. A review of drought indices. Environmental Reviews 19, 333349.  Zargar, A., 2012. Handling uncertainty in hydrologic analysis and drought risk assessment using dempster-shafer theory. Doctor of Philosophy dissertation, University of British Columbia.  Zhu, D., Chang, J., 2008. Annual variations of biotic integrity in the upper Yangtze River using an adapted index of biotic integrity (IBI). Ecological Indicators 8, 564572. Ziese, M., Schneider, U., Meyer-Christoffer, A., Schamm, K., Vido, J., Finger, P., Bissolli, P., Pietzsch, S., Becker, A., 2014. The GPCC Drought Indexa new, combined and gridded global drought index. Earth System Science Data 6, 285295.  Zorn, T. G., Seelbach, P. W., Rutherford, E. S., Wills, T. C., Cheng, S. -T., Wiley, M. J., 2008. A regional-scale habitat suitability model to assess the effects of flow reduction on fish assemblages in Michigan streams. Michigan Department of Natural Resources, Fisheries Research Report 2089, Ann Arbor.  Zou, X., Zhai, P., Zhang, Q., 2005. Variations in droughts over China: 19512003. Geophysical Research Letters 32.