Snow-to-Liquid Ratio Prediction over the Contiguous United States Using Machine Learning

Abstract Winter storms have widespread impacts across the contiguous United States (CONUS), resulting in travel and economic disruptions, loss of life, and damage to property and infrastructure. Snow-to-liquid ratio (SLR), the ratio of new (i.e., freshly fallen) snow to liquid precipitation equivalent, is used operationally to forecast snowfall amount during winter storms. However, no clear best prediction method for SLR has been established, and some operational SLR prediction methods were developed for specific regions, potentially limiting their utility for CONUS-wide applications. Using Community Collaborative Rain, Hail and Snow Network (CoCoRaHS) snowfall observations, station geographic information, and fifth generation European Centre for Medium-Range Weather Forecasts atmospheric reanalysis (ERA5) vertical profiles of atmospheric variables, we developed a random forest (RF)-based algorithm for predicting SLR across the CONUS. SLR forecasts and quantitative snowfall forecasts (QSFs) based on this algorithm were then derived from the NOAA High-Resolution Rapid Refresh (HRRR). The resulting RF SLR mean absolute error (MAE) was 2.92 across the CONUS compared to MAEs for current operational methods of 3.51–4.76. Improvements were also seen in QSF. Additionally, the RF outperformed operational SLR methods within individual regional snow climates. Despite these improvements, there were situations when the RF struggled, such as high (>20) and low (