Authors: Emily Golitzin – University of Utah
Title: Integrating Diverse Meteorological Inputs for Machine Learning-Based SWE Estimation in Complex Catchments
Presentation Type:
Abstract: Accurately estimating snow water equivalent (SWE) in mountainous catchments is a critical challenge for hydrologic forecasting and water resource management. Traditional modeling approaches struggle to capture the spatiotemporal heterogeneity of SWE distribution in complex terrain. In recent years, the proliferation of large-scale meteorological datasets along with advancements in computer hardware and machine learning (ML) techniques have spurred the development of data-driven hydrologic models, yet the application of ML to snow modeling remains relatively underexplored. We apply a flexible ML framework trained on the Airborne Snow Observatories, Inc. (ASO) 50-meter SWE product alongside a relatively simple feature space including static terrain attributes, in-situ SWE measurements, and gridded daily meteorological data, all processed at a user-defined spatial resolution, to create estimates of spatially distributed SWE across a variety of operationally important basins. Feature importance analysis reveals that precipitation inputs strongly influence model outputs. To investigate the sensitivity of model skill to the source of meteorological forcing data, we compare models trained on gridded precipitation from multiple forcing datasets with spatial resolutions ranging from 800m to 1/8 degree (approximately 12 km).