Merging Regional Climate Models and Remote Sensing Observations to Better Estimate Mountain Snow Water Equivalent

Abstract

Large‐scale, high‐resolution estimation of snow water equivalent (SWE) in mountainous regions remains challenging. Continental‐scale approaches, including SWE reconstruction and regional climate model (RCM) simulations, are computationally expensive at high resolutions. We present Blender, a variational data assimilation method that improves modeled SWE and snowfall time series by modifying snowpack mass and energy balance while minimizing changes to input precipitation and energy fluxes, subject to remotely sensed snow cover fraction constraints. Blender combines the advantages of RCM simulation and SWE reconstruction and can be applied as a post‐processing step to many gridded snow data sets. We tested Blender at 500 m resolution in the Tuolumne watershed, Sierra Nevada, USA, during water years 2013, 2016, and 2017 (low, average, and high snow accumulation). We tested Blender with prior inputs from an RCM and ERA5‐Land (ERA5L), and evaluated posterior (output) performance using 18 Airborne Snow Observatory (ASO) flights, and compared its performance with the Western United States Snow Reanalysis (WUS‐SR). Blender improved spatial SWE patterns relative to ASO in all years for the RCM prior (RMSEP: 17.2% → 11.5%, R : 0.39 → 0.72) and ERA5L prior (RMSEP: 13.9% → 12.9%, R : 0.003 → 0.59), approaching WUS‐SR performance (RMSEP = 9.5%, R  = 0.82). Blender improved snow water storage (SWS) estimates for the RCM prior but not for ERA5L prior, although ERA5L posterior SWS remained reasonably accurate (median absolute error of 27%). Blender provides a flexible and efficient pathway toward higher‐resolution, large‐scale SWE estimation in mountainous regions.