Leveraging Precipitation Pattern Persistence for Snow Model Corrections in the Upper-Tuolumne Watershed

Snow model accuracy in mountainous areas is most constrained by the accuracy of meteorological forcing data. This is particularly true at fine-scale (< 100 m) spatial resolutions where forcing data must be downscaled or extrapolated to the model grid. As opposed to resolving fine-scale meteorology, previous studies have investigated the extent to which distributed snow depth could be reproduced using precipitation adjustments as a proxy for wind and terrain-enhanced snow accumulation. In this work, we test popular precipitation-adjustment schemes using snow patterns from 1) a satellite-trained snowpack reconstruction product, and 2) airborne lidar snow depth observations, across the California state Tuolumne river watershed. Results showed that snow accumulation patterns were driven from repeatable storm tracks and their interactions with static features like terrain and vegetation. As a result, snow accumulation patterns between two seasons with similar snow extents were well-correlated (r > 0.80). However, for seasons with abnormal climatic conditions, snow patterns from the snowpack reconstruction product, which leveraged satellite data from 1985 to 2016, identified seasons with similar conditions and patterns better than the airborne lidar collections (from 2013 to present-day). Composite snow patterns, averaged from patterns across multiple seasons, also outperformed any single pattern as interannual irregularities were smoothed.

Precipitation adjustments trained from historic snow depth patterns improved modeled snow depth heterogeneity and increased the snow depth coefficient of correlation from 0.31 to 0.78, on average. Additionally, snow depth observations across a small portion of the model domain (< 16%) were able to correct precipitation bias (within 5%) common among widely-used meteorological products. Results showed that the accuracy of simulations post-adjustment were more constrained by the accuracy of historic snow patterns as opposed to the precipitation adjustment scheme. Going forward, we suggest that long-term snowpack reconstructions, or hybrid datasets that include both snowpack reconstructions and airborne lidar observations, pose the best opportunity to 1) inform locations at which meteorological observations would be most valuable, 2) infer meteorological downscaling, and 3) resolve real-time snow depth evolution.