ForeSWE: Forecasting Snow-Water Equivalent with an Uncertainty-Aware Attention Model

On November 13, 2025, researchers unveiled ForeSWE, a novel probabilistic spatio-temporal forecasting model designed to improve the prediction of Snow-Water Equivalent (SWE), a vital metric for assessing snowpack water content in snow-dominant watersheds. Traditional forecasting methods have struggled to account for the complex influences on SWE, including environmental conditions and topography, often lacking in providing uncertainty estimates. ForeSWE addresses these gaps by combining deep learning techniques with classical probabilistic approaches, utilizing an attention mechanism to capture spatiotemporal features and interactions, alongside a Gaussian process module for quantifying prediction uncertainty. Evaluated using data from 512 Snow Telemetry (SNOTEL) stations across the Western US, the model demonstrated significant improvements in forecasting accuracy and prediction intervals compared to existing methods. This advancement is crucial for water management decisions, as accu…