An assessment of the utility of a local model for operational mountain snowfall predictions

Forecasting mountain snowfall remains a challenging and often frustrating experience. Although the basic principles of mountain snowfall are known, subtle changes in the moisture and stability structure of the atmosphere can produce much different snow amounts than actually forecast. While the operational forecaster, combining experience with different observations, can more often than not make a reasonably good forecast, other tools that can aid the forecasting process are more than welcome.

At the Boulder (formerly Denver) National Weather Service (NWS) Weather Forecast Office (WFO) forecasters have long used a "model" for orographic snowfall prediction which evolved from an early statistical study by Owen Rhea (CSU Atmospheric Science Paper 197, 1973). Rhea used a statistical approach to compare precipitation measurements at selected Colorado mountain sites to a relatively simple calculated orographic precipitation rate based on a given 700-mb wind velocity and temperature. A later study by Rhea used a numerical model to refine the calculated orographic precipitation rates, and from this study maps of predicted snowfall were produced based on different prevailing conditions. In the 1980s the work was formulated into a simple computer program that allowed the forecaster to enter the moisture depth and 700-mb wind speed, direction, and temperature, and the program produced a map of expected snowfall over a specified duration. More recently Eric Thaler (the Boulder WFO Science Operations Officer) added an additional criteria of lapse rate and produced an expanded set of output that included ranges of snowfall for different liquid to snow ratios.

While the Rhea (and more recently Rhea/Thaler) model has provided very useful (and sometimes quite accurate) guidance to the forecaster, clearly the method considers only an estimate of the orographic snowfall potential based on relatively generalized conditions and without consideration of any additional dynamic forcing. At the Forecast Systems Laboratory (FSL) we have been testing several local models at 10-km resolution, recognizing that as computer power continues to increase and costs decline, a local forecast office might be able to run a local-scale model inhouse. With that goal in mind, in the past year one of the local models, a 10-km version of the RAMS model, locally called the Scalable Forecast Model (SFM), was set up to run at the Boulder WFO with its output displayed on their AWIPS (Advanced Weather Interactive Processing System). The SFM has been running more or less continuously four times a day, out to 18h, for over a year at the Boulder WFO. The model is initialized with a mesoscale analysis from the Local Analysis and Prediction System (LAPS) developed at FSL and the boundary conditions provided by NCEP's Eta model. With explicit physics and a 10-km scale, the model nicely captures orographic precipitation. However, since there has never been any systematic verification of its snowfall predictions in the mountains, it is regarded at this point with some caution by the WFO forecasters. In this paper we will assess the model's performance and potential usefulness over a range of conditions by comparing its forecasts not only to observations but also to the output from the Rhea/Thaler model as a benchmark.