Ensemble member forecast spread and its implication for road weather forecasting

The Federal Highway Administration (FHWA) began an aggressive research and development program in 2000 to develop a prototype winter Maintenance Decision Support system (MDSS) designed to support state departments of transportation snow and ice control operations. The MDSS utilizes state-of-the-art weather forecasting and data fusion techniques and merges them with computerized winter road maintenance rules of practice. The result is a set of guidance aimed at maintenance managers that provides a forecast of surface conditions and treatment recommendations customized for specific routes.

Several weather models are used by the Road Weather Forecast System (RWFS), which generates a consensus forecast used within other MDSS components. The overall forecast spread predicted by the ensemble model members of the RWFS, for a moderate snow event on 27-29 November 2004 is presented to illustrate the large differences often found between model members. The winter 2004-2005 field demonstration was conducted over Colorado with an emphasis on three regions including the E-470 roadway which bypasses eastern Denver, a section of I-70 near Genesee in the foothills west of Denver, and I-70 in the mountains near Vail Pass.

Several meteorological parameters (including air and dewpoint temperature, cloud cover, and quantitative precipitation forecast (QPF)) are examined from several numerical weather prediction models (Eta, GFS, RUC, MM5, WRF and MAV MOS). For a single weather event it is shown that the predictions from each model for air temperature have a spread as large as 6C at times and more than 8C for dewpoint temperature. The cloud conditions also vary greatly with a forecast range from scattered to overcast skies. The total QPF amounts (liquid equivalent) for one run time show a difference between all the models of 0.5 inches, while the later run times for this same case are within 0.2 inches. The models also vary greatly in precipitation rates from no precipitation falling to 0.05-0.1 inches per hour.

It is important to be aware of the forecast differences for each weather parameter. The large spread that is evident between all of the models lends credence to the ensemble forecasting approach utilized in the RWFS which results in a final consensus forecast that is often times better than many of the individual models.