Quantifying the Mismatch between Snow and Climate in Global Reanalyses and Land Models

Accurate estimation of the amount of snow on the ground is critically important to water resources, as annual runoff production in cold regions is heavily influenced by annual snowmelt. In addition, snowcover has a large effect on land-atmosphere interactions through surface exchanges of radiation and energy. Yet, despite snow's importance, its quantification in large-scale gridded products is difficult because of uncertainties about how much snow is on the ground over large areas. Here, we present a new methodology to estimate large scale maximum snow water equivalent (SWE) from in situ point measurements by linking them to snowfall estimates. Using this linkage can result in a better estimation of SWE at large scales with fewer stations than using four other methods that relate SWE to elevation. We then use thousands point observations from the SNOTEL and COOP networks across the continental US to evaluate the snow components of land models in the Global Land Data Assimilation System (GLDAS), as well as in the CFS-R, MERRA, and ERA-Interim reanalyses. We find that, relative to model snowfall, maximum SWE depicted in the gridded products is substantially lower than that obtained from the station data across the continental United States. Furthermore, in a majority of the products that we tested, snowfall also makes up a lower fraction of total snow season precipitation than is suggested by the COOP and SNOTEL data. Although the low bias of snowfall divided by precipitation can partly be explained by biases in temperature, the substantial low bias of maximum SWE is likely to be indicative of land model deficiencies, such as ablation rates that are too high.