Speaker: Parth Modi, PhD student, Civil, Environmental, and Architectural Engineering, CU Boulder
Topic: Understanding the Performance and Value of Snow-Based Seasonal Streamflow Forecasts in the Western US

Abstract
Accurate water supply forecasting is critical for managing water resources, particularly in snow-dominated regions where drought can severely impact water availability. The overarching goal here is to enhance the accuracy, robustness, and practical utility of water supply forecasts (WSFs) by investigating how drought affects forecast performance, exploring the potential of advanced modeling techniques, and examining the relationship between forecast accuracy and decision-making value. The first chapter focuses on the relationship between snow water equivalent and streamflow volume in headwater catchments across the western US. We propose an adaptive sampling approach that shows training forecast models on drier years reduces errors by up to 20% during drought. The second chapter explores whether advanced machine learning models, specifically Long Short-Term Memory (LSTM) models, can enhance the predictive skill of WSFs. LSTMs were integrated into the Ensemble Streamflow Prediction (ESP) framework and tested by providing models with explicit information on snow. Forecasts incorporating a sophisticated representation of snowpack information performed similarly to operational forecasts. The final chapter assesses the relationship between forecast skill and forecast value – the financial benefit derived from using forecasts in decision-making. Despite the theoretical assumption that high forecast skill should lead to high forecast value, we find that this is not always the case, particularly during real-world drought conditions due to irregular error structures in the forecast. These findings emphasize the need for more sophisticated approaches to forecast evaluation, focusing on value across varying climate conditions rather than solely improving forecast skill metrics. While the analyses of this dissertation address key challenges related to different models and are context-dependent, they each offer new pathways for enhancing water resource management under increasing drought stress.

Bio
Parth Modi completed his MS in biological systems engineering from Virginia Tech in 2020 and is a doctoral research assistant in the Department of Civil, Environmental and Architectural Engineering at the University of Colorado Boulder. Currently, his research focuses on evaluating snow-based streamflow forecasts across the western US, particularly during drought conditions, and analyzing their implications for decision-making and economic outcomes. He has extensive experience in hydrologic modeling, working with both process-based models and machine-learning approaches, and has also contributed to projects examining the impacts of climate change on mesoscale hydrological processes.