High-resolution snow water equivalent estimation: A data-driven method for localized downscaling of climate data

Abstract

Estimating high-resolution daily snow water equivalent (SWE) in mountainous regions is challenging due to geographical complexity and the irregular availability of high-resolution meteorological data. This study introduces a method for downscaling SWE based on low-resolution climate models. It is based on the dependence between climate estimators and SWE. Although SWE changes rapidly, patterns often repeat under similar meteorological conditions. We implement this principle to downscale SWE to a 500 m resolution using a k-nearest neighbor algorithm with a customized distance metric. To evaluate the performance of our approach, we conducted tests for California's Sierra Nevada and Colorado's Upper Colorado River Basin in the western United States using different low-resolution climate models (ec-earth3-veg, mpi-esm1-2, and cnrm-esm2-1) at both 100 and 9 km scales. We performed a cross-validation analysis and compared our results with commonly used gridded SWE datasets, as well as with point-scale time series. The results demonstrate that our approach enables the generation of downscaled SWE, which closely matches reanalysis data in terms of statistical properties. The outputs demonstrate that, for each region, performance depends on the choice and accuracy of the climate model inputs, such as precipitation and temperature data. Overall, the cnrm-esm2-1 model demonstrates superior accuracy in Colorado, outperforming other models at both the 100 and 9 km resolutions. Conversely, the ec-earth3-veg model achieves the best performance in California with 9 km resolution climate data. Across models, a 9 km resolution typically provides slightly better accuracy compared to a 100 km resolution. This opens up possibilities for applications in regions with limited in situ or meteorological measurements. The approach also has the potential to recreate unmeasured historical SWE values.