Testing of Updated Methods for Probabilistic Quantitative Precipitation Forecasting in the Nws's National Blend of Models.

The National Blend of Models applies statistical postprocessing techniques to multi-model ensembles to improve their skill. These digital forecasts provide an objective set of guidance for the National Digital Forecast Database used to make NWS worded forecasts. This presentation will describe updates to the currently operational methodology for probabilistic quantitative precipitation forecasting (PQPF). The previous method used quantile mapping and dressing algorithms to create reliable probabilities. Testing of recent changes includes: (a) an improved quantile mapping algorithm, (b) the variable weighting of sorted ensemble members, with weights determined according to rank-histogram statistics, (c) the effects of the incorporation of ECMWF forecast data, and (d) comparison against an established parametric methodology, the Censored, Shifted Gamma Distribution of Scheuerer and Hamill (2015). The results show that the algorithmic updates and use of ECMWF data provide a substantial increase in forecast skill for lead times of 1-7 days.