Assessing the impact of meteorological forcing and its uncertainty on snow modeling and reanalysis
Abstract. Large uncertainties in global model-based snow datasets, particularly in snow water equivalent (SWE), limit our understanding of snow storage and its response to climate change. These uncertainties are sensitive to meteorological inputs used to force offline snow models. In this study, we assessed the impact of three meteorological forcing datasets (i.e., ERA5, MERRA-2, and NLDAS-2) on ensemble SWE estimates within a probabilistic snow modeling and reanalysis framework across three snow-dominated mountainous watersheds in the western US. Prior (open-loop) SWE estimates show significant inter-dataset variability, primarily driven by differences in cumulative snowfall. SWE errors are dominated by bias, and no single-forcing dataset consistently outperforms the others across all domains or elevations. To assess the value of using multiple products, we construct a multi-forcing ensemble using least-square-based weighting informed by prior performance. The multi-forcing ensemble reduces errors compared to individual forcings and improves prior SWE accuracy across all regions. Assimilation of near-peak lidar-derived snow depth substantially corrects prior SWE errors, reducing the influence of forcing-driven biases accumulated during the snowfall season. As a result, random error is the dominant source of posterior error. Although assimilation narrows performance differences, the multi-forcing ensemble still yields slightly better overall accuracy and improved uncertainty characterization. This work demonstrates that integrating diverse meteorological forcings within a data assimilation framework improves SWE estimates (both model-based and reanalysis-based), especially where the optimal forcing dataset is uncertain.