Improving Snow-to-Liquid Ratio Forecasts in the Western United States

Accurate and spatially detailed estimation of snow-to-liquid ratio (SLR) is a critical component of snowfall forecasting as it is used to convert the quantitative precipitation forecast (QPF) to snowfall amount. In the western contiguous United States (CONUS), snow climates vary significantly from the coast to the interior, necessitating the development of regionally specific SLR algorithms capable of accounting for the large variability possible during individual storms (i.e., from 4:1 to 40:1 or more in rare cases).

In this study, we investigate relationships between SLR and atmospheric conditions using high-quality manual snowfall observations collected by snow-safety personnel and other groups at several high-elevation sites in the continental western United States. Several machine learning techniques are applied to a combination of single-station, regional, and west-wide datasets and evaluated using observations from recent cool seasons. Prior work has shown a neural network can improve the forecast SLR. However, existing applications are trained only on lowland datasets and may not sufficiently represent the variability observed in complex terrain.

These experimental SLR forecasts are slated to be available in a quasi-operational setting prior to the 2020-2021 cool season. We anticipate these statistical techniques for determining SLR will be a valuable addition to an operational forecaster’s toolbox and enhance the quality of snowfall forecasts, especially in regions of complex terrain in the western United States.