An Evaluation of Snow Initializations in NCEP Global and Regional Forecasting Models

Snow plays a major role in land-atmosphere interactions, but strong spatial heterogeneity in snow depth (SD) and snow water equivalent (SWE) makes it challenging to evaluate gridded snow quantities using in situ measurements. First, a new method is developed to upscale point measurements into gridded datasets that is superior to other tested methods. It is then utilized to generate daily SD and SWE datasets for water years 2012-14 using measurements from two networks (COOP and SNOTEL) in the United States. These datasets are used to evaluate daily SD and SWE initializations in NCEP global forecasting models (GFS and CFSv2, both on 0.5 degrees x 0.5 degrees grids) and regional models (NAM on 12 km x 12 km grids and RAP on 13 km x 13 km grids) across eight 2 degrees x 2 degrees boxes. Initialized SD from three models (GFS, CFSv2, and NAM) that utilize Air Force Weather Agency (AFWA) SD data for initialization is 77% below the area-averaged values, on average. RAP initializations, which cycle snow instead of using the AFWA SD, underestimate SD to a lesser degree. Compared with SD errors, SWE errors from GFS, CFSv2, and NAM are larger because of the application of unrealistically low and globally constant snow densities. Furthermore, the widely used daily gridded SD data produced by the Canadian Meteorological Centre (CMC) are also found to underestimate SD (similar to GFS, CFSv2, and NAM), but are worse than RAP. These results suggest an urgent need to improve SD and SWE initializations in these operational models.