Testing Model Representations of Snowpack Liquid Water Percolation Across Multiple Climates

Abstract

AbstractSnowpack liquid water percolation is a sensitive model process that is crucial for snowpack runoff forecasts yet varies in sensitivity between climates and snow seasons. Therefore, models of varied complexity developed for different climates and purposes use different percolation representations. We investigated how liquid fluxes in a multilayer snow model vary as represented by discrete representations of gravity drainage and snow density thresholds. We evaluated performance and sensitivity to nonphysical parameters using point measurements of snow water equivalent (SWE) at a maritime site in Washington, USA, and measurements of both SWE and runoff in the French and Swiss Alps. At all three locations, the gravity drainage simulations reduced parameter sensitivity and increased model performance. Average Nash‐Sutcliffe efficiency improved from 0.06 to 0.61 between density threshold and gravity drainage simulations with default parameters. The disparity in model performance was particularly evident at the maritime site (Washington, USA), where the gravity drainage peak SWE was biased by 6% (0.10 m) but density threshold peak SWE was biased by 85% (1.51 m). Simulated runoff and SWE also decreased in performance and increased in parameter sensitivity when including a widely used two‐layer percolation routine. This demonstrates the importance of testing and evaluating models across a wide range of climates, with close attention paid to warmer regions, where percolation has high parameter sensitivity. This is particularly important for global snow modeling and climate change scenarios where multiple regions and snow seasons must be adequately represented with a single model implementation.