Fixing Inconsistencies between GFS Snow Accumulations and Precipitation Type

Some operational models in the National Centers for Environmental Prediction (NCEP) numerical modeling suite can output multiple snow accumulation products. The North American Mesoscale model (NAM) and Global Forecast System (GFS) each output two snow accumulation products: 1) Water Equivalent of Accumulated Snow Depth (WEASD) and 2) Snow Depth (SNOD). These two snow accumulation products are computed differently and can depict inconsistent snowfall totals during mixed precipitation events. WEASD includes both snow and sleet in its water-equivalent total and requires the user to select their own snow-to-liquid ratio (often 10:1). In contrast, SNOD utilizes the densities of different precipitation types within the model’s own Land Surface Model (LSM) to produce an actual “snow” accumulation. Due to its utilization of the densities of different precipitation types, SNOD is typically the more accurate snow accumulation product during mixed precipitation events.

The winter storm of 1–4 February 2022 was an exceptional mixed precipitation event, with noteworthy sleet and freezing rain accumulations forecasted and observed across a large portion of the central and eastern United States. Prior to the event, many forecasters noticed unusually high snowfall totals in GFS SNOD forecasts. While NAM SNOD correctly reduced snowfall totals in areas of sleet and freezing rain, GFS SNOD did not. An examination of the GFS by the NCEP/Environmental Modeling Center (EMC) Model Evaluation Group (MEG) and Land-Surface Team of the Model Physics Group revealed that SNOD was artificially inflated by sleet and freezing rain in the LSM, with multiple precipitation types contributing to SNOD as if they were snow. Snow in the GFS was also found in environments where it was not expected to exist (i.e., beneath low-level warm layers where sleet was expected). Updates to address the issues in the SNOD computation were included in the GFS upgrade in November 2022 (GFSv16.3).