A New Landsat-era Snow Reanalysis Dataset over the Western United States

Abstract:

Despite the importance of seasonal snowpack, even in well instrumented areas of the globe like the Western United States, there is a need for observationally-constrained gridded snow datasets at space-time resolutions and extents capable of resolving the key modes of variability seen in mountainous terrain. Herein we present a newly developed Landsat-era snow reanalysis dataset over the full Western United States (WUS). A Bayesian framework with a priori snow estimates updated through the assimilation of Landsat 5, 7, and 8 (Water Years (WYs) 1985-2019) fractional snow-covered area observations is used to generate continuous posterior estimates of snow water equivalent (SWE) at a spatial resolution of ~0.004° (~500 m) at the daily time scale. The posterior reanalysis estimates are evaluated through comparison with all recent (lidar-derived) Airborne Snow Observatory (ASO) snow products (in California, Washington, and Colorado in WYs 2013-2019) and other in situ snow data across the WUS. The evaluation shows significant improvements in posterior over prior estimates. Comparing prior and posterior SWE with ASO SWE, the mean of basin-wide correlation increases from 0.56 (prior) to 0.80 (posterior); root mean square error and mean difference reduce by averages of 55% and 29%, respectively. Preliminary results quantify the storage of snow water across WUS and how it varies spatially (by watershed and elevation) and temporally (seasonally and interannually over the 35-year record) and compare these large-scale estimates to other available products. Additionally, spatial and temporal variability, and how they are connected to underlying physiographic and meteorological drivers are illustrated.