Abstract
Accurate drought prediction is critical for mitigating its adverse effects and supporting effective water resource management. Hydrological deterministic watershed models provide a tool for predicting drought by simulating the complex interactions between precipitation, soil moisture, surface water, and groundwater. However, there is limited research evaluating the ability of hydrological models to predict droughts and resulting low streamflows, as well as how to parameterize these models to more effectively represent drought series and statistics. This thesis investigates the performance of the Weather Research and Forecasting Hydrological (WRFHydro) modeling system in simulating drought conditions in several watersheds located within the United States. Although WRF-Hydro has demonstrated some capability in predicting streamflow in the northeastern U.S. in this study, it has limited accuracy in predicting droughts. This study implements improved calibration strategies, including use of a variety of metrics for model calibration, aggregating daily streamflows to longer time-periods for calibration of a model with a shorter time step, and the use of multiple low-flow metrics to assess model performance. The results of this research demonstrate that calibration techniques, including the use of logspace metrics and the proper handling of zero flows, should improve predictions of low streamflow statistics and series. Use of a new censored maximum likelihood estimator for model calibration in log space was shown to improve the classification of streamflows recorded as zero. Incorporating aggregated streamflows during calibration also shows the potential for enhancing low streamflow predictions, although notable differences are observed across our study sites. The integration of the MOVE.2 method to extend retrospective National Water Model (NWM) output and create streamflow-derived drought categories was shown to better align with the U.S. Drought Monitor classifications. However, limitations in the NWM prediction of soil moisture and low streamflow events indicates research areas requiring further investigation. This work lays a foundation for refining hydrological models by emphasizing multivariate approaches, improving low streamflow calibration techniques, and addressing challenges posed by intermittent streamflow sites. These advancements contribute to the hydrologic modeling literature and should lead to improvements in drought characterization, drought prediction and water resource management.