WCD-30. Quantile Mapping Using GEFSv12 Reforecasts for improving Probabilistic Precipitation Forecasts

Abstract
Version 12 of the Global Ensemble Forecast System (GEFSv12) was implemented in September 2020. Among the improvements include an increase from 20 to 31 ensemble members, generated 4 times each day (i.e., 00, 06, 12 18 UTC) to + 16 days lead time. Once daily, for 00 UTC initial conditions, the 31 members are extended to +35 days. The GEFSv12 is also accompanied by a 20-year reanalysis and reforecast database spanning 2000-2019; it provides 5 members of the 00 UTC cycle daily, and once weekly 11 members to +35 days. With these improvements, the current quantile mapping procedure used by the National Weather Service National Blend of Models (NBM) precipitation forecasts can be improved to provide better estimates of 6-hourly probabilistic quantitative precipitation forecasts (PQPF). Quantile mapping is a post-processing technique that permits amount-dependent bias correction based on the differences between forecast and analyzed cumulative precipitation distributions. Cumulative distributions functions (CDFs) are estimated for the forecast grid and analyzed grid for each member for every grid point, month, and lead time. When the real-time forecast becomes available, the quantile relative to the climatological distribution of forecasts is determined, and then the analyzed precipitation amount related to that same quantile from the climatological forecast is estimated. The resulting estimate is then remapped to the same grid point for the real-time forecast for that same ensemble member, month, and lead time. The quantile-mapped QPF generated with this procedure, which is currently used by the NBM, has been shown to consistently outperform the raw model output of QPF. In the current quantile mapping procedure used by the NBM, supplemental locations are used to provide enough precipitation samples within the last 60 days of training data to generate realistic CDFs. However, recent work has shown that this does not work well in mountainous regions and transition seasons, particularly in the dry fall to the start of wetter winter. A newly developed quantile mapping procedure takes advantage of the multi-decadal reforecasts from the GEFSv12 for generating the forecast CDFs. Early results show that the reforecast-based quantile mapping adds terrain detail back into the resulting PQPF. This presentation will also show that these changes result in more skillful, reliable and geographically detailed precipitation guidance within the GEFSv12. Results from 2017-2019 will be presented demonstrating that the new quantile mapping procedure improves the QPF using brier skill scores and reliability diagrams. Several cases will be shown highlighting the regional and seasonal strengths and limitations of this technique, with focus on lead times 6 to 240 hours using the 6-hr QPF.