An analysis of short range ensemble for cold season cyclones. Part I: An analysis of quantitative precipitation mean and probability forecasts

Ensemble forecasts are increasingly being used within the forecast environment to provide forecasters with information on the predictability of weather systems. Most ensemble information presented to operational forecasters is displayed as mean and probability fields with little or no information from individual members. It has been suggested that analysis of the mean and probability fields can be used to enhance the decision on snowfall amounts as well as the warning decision process. This assumes that the differences between members of the ensemble are the result of differences in initial conditions or model parameterizations. However, the differences between ensemble members can also be the result of systematic errors between different models. This is especially true in the first 24 to 72 hours when errors from initial conditions are relatively small. The short-range ensemble forecast (SREF) suite run by NCEP includes 4 models – the Eta, Regional Spectral Model (RSM), the Nonhydrostatic Mesoscale Model core of the Weather Research Forecast model (WRF-NMM) and the Advanced Research WRF (WRF-ARW). Two events will be analyzed to examine the SREF quantitative precipitation forecast (QPF) and compare those to the operational model QPF within 84 h of the forecast.

We found that the use of probability fields, by themselves, provided little additional information. In many cases analysis of the individual runs found that each model tended to cluster around the respective “parent” run. In one event, 4 out of 5 RSM members and 6 out of 6 WRF members produced precipitation in Sioux City, Iowa while only 1 of 10 Eta members produced precipitation, with a resultant 50 percent chance for precipitation. It appeared that the differences between members were the result of systematic model errors rather than inherent unpredictability. The inability to remove members that are determined to have a low probability of occurrence places equal weight on all 21 members. It is thus theorized that examination of SREF mean and probabilities, especially QPF, without knowledge of each individual member's performance, provides little additional information to aid in the forecast process, and in the case above, results in degradation of the forecast.

Ensemble forecasts can play an important role the first 24 to 72 h of the forecast. However, use of mean and probability fields by themselves can slow the recognition of significant events due to averaging of both the likely and the less likely members. After assessing the dynamics and thermodynamics from the deterministic models, we propose that forecasters can examine individual members and remove the members that appear least likely to occur. Mean and probability fields can then be calculated from this subset to help forecast snow and QPF amounts.