PNNL

Abstract

A large percentage of peak runoff events in the Western United States are caused by snowmelt, particularly during rain-on-snow events. Distributed hydrological models that track snow accumulation and ablation processes have been extensively used to predict snow and streamflow regimes, as well as associated hydrologic extremes under altered environmental conditions. A key source of uncertainty in hydrologic modeling in snow-dominated regions is the magnitude, duration, and distribution of snow water equivalent (SWE). However, the spatial variability of the modeling skill, predictive uncertainties, parameter sensitivity and transferability, and model structural adequacy are discussed less frequently on regional to continental scales. With large-ensemble runs of the Distributed Hydrology-Soil-Vegetation Model (DHSVM) at each Snowpack Telemetry (SNOTEL) site, we benchmarked the snow modeling skill, explored the parameter space that can effectively characterize snowpack dynamics, and assessed the predictive accuracy and uncertainties on the daily scale across hydroclimate regimes represented by 246 SNOTEL sites in the Western United States. We found that DHSVM could well capture snowpack dynamics and peak SWE on a fine daily scale at most SNOTEL sites (mean Nash-Sutcliffe Efficiency of 0.9), and its predictive skill was generally higher in the Pacific Northwest and much of the Northern and Middle Rockies. Snow albedo parameterization was critical for snow modeling at most SNOTEL sites, and snow threshold temperature is important for sites characterized by warm winters. The snow modeling uncertainties quantified here can be evaluated in conjunction with the uncertainty in climate ensembles in the estimation of hydrologic extremes and streamflow forecasting.