Exploring the Utility of Downscaled SREF Grids for Generating Probabilistic Snowfall Forecasts

Since the mid-1990s the National Weather Service forecast office in Salt Lake City has been providing forecasts to snow safety officials in the Wasatch Mountains of northern Utah to assist them in their avalanche control work. Forecasts are issued twice daily during the cool season and include expected snowfall, snow-water equivalent (SWE), and ridge top winds amongst other parameters. Two twelve-hour snowfall forecasts are provided in a basic two bin format with the ‘most likely' snowfall/SWE range being assigned a likelihood of 60-90 percent and the ‘next most likely' range being assigned a probability of 10-40 percent. Verification studies have shown that although the WFO generally does a good job forecasting storm total snowfall/SWE amounts, limited skill exists regarding each twelve-hour forecast bin. This is most evident in the ‘most likely' snowfall/SWE ranges as forecasts only successfully verify about 40% of the time. With this information in hand, the office is undergoing an effort to improve the science behind these forecasts to increase the accuracy of the probabilistic snowfall/SWE predictions.

In an effort to improve these probabilistic forecast ranges the Salt Lake City office is investigating an experimental technique utilizing Short-Range Ensemble Forecast (SREF) output from NCEP. Earlier results using SREF QPF directly showed no model skill in the highly complex terrain of the Wasatch Mountains in northern Utah. To incorporate the influence of terrain into the output, the QPF from each individual SREF member is being downscaled utilizing a PRISM-based climatology downscaling technique developed at the National Weather Service office in Boise ID. Data was collected from the winter of 2012-13 with over 60 events to assess the performance and potential utility of the downscaled SREF output. A variety of values were archived from the downscaled SREF output including the Max, Min, Median, and Mean of the individual members for each twelve-hour time step for comparison to observed values. In addition, a simple relative frequency approach was utilized to calculate a QPF range from the twenty-fifth to seventy-fifth percentile values from the individual downscaled SREF members. This range was verified against observed SWE totals and compared to the WFO's twelve-hour ‘most likely' forecast bins when measurable precipitation occurred. During the winter of 2012-13 the range produced from the twenty-fifth to seventy-fifth percentile values of the down-scaled SREF showed improvement over the WFO forecasts verifying nearly sixty percent of the time for events with SWE < .50” in 12 hours. The majority of the downscaled SREF outputs ‘misses' occurred in pattern types that can be readily identified by forecasters. Through intelligent use of this technique, it appears that forecasters may be able to provide better probabilistic information to the snow-safety community, aiding their operational decision making process. Results from the operational application of the technique from the early portion of the 2014-15 winter season will also be discussed.