Authors: Xinxin Sui, Andrew W. Wood, Guoqiang Tang, Rhys Pulling, Cenlin He–Colorado School of Mines
Title: A long-term baseline testbed dataset for benchmarking snow model performance in the western United States
Presentation Type: Poster Presentation
Abstract: Snow is a critical water resource in the dry and mountainous regions of the western United States. To represent snow accumulation and melt processes, a wide range of modeling strategies has been introduced, ranging from relatively simple temperature index schemes to more complex, physically explicit energy balance models, and more recently, AI-based modeling approaches. The diversity of approaches highlights the need for standardized datasets and protocols for benchmarking the value of different approaches across a large variety of locations. To this end, we have developed a nearly 75-year-long daily snow water equivalent (SWE) dataset for over 792 SNOTEL station locations across the western United States, based primarily on temperature index (TI) modeling. The dataset leverages a SNOTEL meteorological record extension that provides serially complete (gap–filled) and quality–controlled precipitation and air temperature records from 1950-2023. Results show that a simple daily timestep two-parameter TI model can effectively reproduce observed SWE dynamics and snow–rainfall partitioning across many sites, achieving high skill scores. In certain conditions (high elevation, low humidity), however, calibrated temperature thresholds and degree-day factors are required to compensate for the absence of radiative and other physical processes. Overall, this benchmark dataset can serve as a resource for the quantitative assessment of modeling advances in the snow modeling research community.