Evaluation of Noah-MP Snow Simulation across Site Conditions in the Western United States

Abstract

Abstract Quantifying spatiotemporal variability in snow water resources is a challenge especially relevant for regions that rely on snowmelt for water supply. Model accuracy is often limited by uncertainties in meteorological forcings and/or suboptimal physics representation. In this study, we evaluate the performance and sensitivity of Noah land surface model with multiparameterization options (Noah-MP) snow simulations from ten model configurations across 199 sites in the western United States. Nine experiments are constrained by observed meteorology to test snow-related physics options, and the 10th experiment tests an alternative source of meteorological forcings. We find that the base case, which aligns with the National Water Model configuration and uses observation-based forcings, overestimates observed accumulated snow water equivalent (SWE) at 90% of stations by a median of 9.6%. The model performs better in the accumulation season at colder, drier sites and in the melt season at wetter, warmer sites. Accumulation metrics are sensitive to model configuration in two experiments, and melt metrics, in six experiments. Alterations to model physics cause changes to median accumulation metrics from −13% to 2.3% with the greatest change due to precipitation partitioning and to melt metrics from −10% to 3% with the greatest change due to surface resistance configuration. The experiment with alternative forcings causes even greater and wider-ranging changes (medians ranging from −29% to 6%). Not all stations share the same best-performing model configuration. At most stations, the base case is outperformed by four alternative physics options which also significantly impact snow simulation. This research provides insights into the performance and sensitivity of snow predictions across site conditions and model configurations. Significance Statement The purpose of this work is to evaluate the performance and sensitivity of a land surface model’s simulation of snow across site conditions and in response to different model configurations. This is important because estimating snow distribution is a challenge especially relevant for regions that rely on snowmelt for water supply. While land surface models can provide useful large-scale estimates, they are often limited by uncertainties in forcings and/or suboptimal physics representation. The results, which show varying model behavior across geography, climate, vegetation types, and model configurations, highlight inadequacies in model physics representation, emphasize the need for accurate meteorological forcings, and suggest that customizing model configurations to the unique characteristics of the domain could yield more accurate and useful results.