Extended Abstract (1.2M)JP2.1
Application of a QPF-POP relationship to ensembles to generate better probabilistic precipitation forecasts
Christopher J. Schaffer, Iowa State University, Ames, IA; and W. A. Gallus and M. Segal
Gallus and Segal (Weather and Forecasting, 2004) and Gallus et al. (Weather and Forecasting, 2007) used a precipitation-binning technique to show that at grid points where the quantity of precipitation was larger, the probability that those grid points would receive at least a small amount of precipitation was greater. It is believed that this relationship holds because models are better at forecasting where atmospheric conditions are favorable for precipitation than at forecasting actual amounts. It was also noted that probability of precipitation (PoP) values increased even further if two different models showed an intersection of grid points. These findings suggest that PoPs can be further improved, if these relationships were applied to ensemble forecasts.
This study post-processes WRF ensemble precipitation data, testing new methods of obtaining PoP forecasts to show that PoPs will increase with a combined increase in precipitation amount and number of ensemble members predicting precipitation. At each point on the grid, a characteristic precipitation amount must be determined based on the amounts forecasted by the ensemble members. For instance, the maximum amount forecasted by any member could be used, or the method could use an ensemble average value. By binning the characteristic amount and determining the number of members forecasting a specific amount of precipitation at each grid point, refined PoPs are calculated. By using these refined probabilities (obtained from 20 cases from a 4-km dataset run by the Center for the Analysis and Prediction of Storms during 2007, in support of the NOAA Hazardous Weather Testbed Spring Program), the reliability of the probabilistic precipitation forecasts is improved, when compared to the traditional method of equally-weighting ensemble members to determine PoPs. The Brier scores calculated from the refined PoPs are more favorable than those of the traditional method for determining PoPs, as are the ROC areas associated with the refined PoPs. The method was also tested on a 15-km Winter 2006 dataset, which yielded similar results. Finally, a Summer 2008 dataset was used as a training dataset in order to be tested against the Summer 2007 dataset.
Joint Poster Session 2, Joint Poster Part II
Tuesday, 2 June 2009, 3:00 PM-4:00 PM, Grand Ballroom Center
Next paper