Research On Optimizing The Configuration Of NCEP Global Ensemble Forecast System

Ensemble forecasting has been one of the most important tasks for global numerical operational and research centers. The configuration of ensemble forecast system plays a critical role for developing practical ensemble forecast system, which need consider the balance of computing resource, ensemble size and model resolution. This study, using the current NCEP GEFS (Global Ensemble Forecast System), compared the impact of different ensemble sizes and different model horizontal resolutions respectively, together with the relative performance of ensemble sizes versus model horizontal resolution. Based on these results, the configuration of NCEP operational GEFS was optimized. The results of experiments were studied using different verification methods (RMSE, P AC, CRPS and BSS) from December 1st,2009 to January 31st,2010 for six variables (500hPa and 1000hPa geopotential heights; 850hPa and 2m temperature; and 10m u- and v-components of wind) over the Northern Hemisphere (NH) and Southern Hemisphere (SH) extra-tropics. The conclusions are written as follows,1. When the model resolution was constant, increasing ensemble size was beneficial to improve skill of ensemble. The conclusion can be reflected from two points in detail as follows. (1) The experiment was started from ensembles of the Lorenz-96 model. Results indicated that, with the increasing of ensemble size, the degree of improvement decreased. The impact of increasing ensemble size on probabilistic forecast was more helpful for improving forecast skill than on ensemble mean forecast. (2) The conclusions obtained from the Lorenz model were corroborated with NCEP GFS, a realistic model of the atmosphere. Results indicated that, with doubly increasing of ensemble sizes, over NH, the forecast ranges of significant improvements in the performance shortened when measured by deterministic metrics; nevertheless, they were significant for all six variables over SH. Increasing in the ensemble size led to significant improvements in the performance for all forecast ranges when measured by probabilistic metrics. The improvement in probabilistic forecast skill for the short lead time came from the resolution gain, which slowly decreased with increasing lead time. For the long lead time, it came from the reliability gain.2. When the ensemble size was constant, for most variables, increasing model horizontal resolution was beneficial to improve forecast skill for short forecast ranges. There are two points in detail as follows. (1) Results indicated that the improvements with increasing horizontal resolution were only significant during short forecast range for 500hPa and 1000hPa geopotential heights, and 10m u- and v-components of wind. (2) The performances of ensemble forecast for 850hPa and 2m temperature were more sensitive to horizontal resolution than other variables.3. The relative benefits of the T126L28 model with 70 members (70T126) and the T190L28 model with 20 members (20T190), which have equivalent computing costs, were compared. From the statistical verification, it was found that for most variables, increasing the model resolution was more (less) beneficial than increasing the ensemble size for short (long) lead times. Forecasts from the configurations of 70T126 and 20T190 for 2009 North American Christmas winter storm were compared.70T126 (20T190) performed better than 20T190 (70T126) for 14-day (3-day) forecast, which was corresponding to the results from the statistical verification.4. Based on the results above, a variable horizontal resolution and ensemble size coupled system which costs less computational resources was designed used to enhance week-1 and week-2 forecast of the NCEP GEFS, respectively. The results of the statistical verification and the case study showed that this configuration can not only save computational resource, but also have good performance.This study, based on the current NCEP GEFS, will be helpful in planning the future development of NCEP GEFS. Meanwhile, the conclusions obtained from this study are general, so other Numerical Weather Prediction centers around the world could use these experiences for reference, especially for the development of Chinese ensemble forecast system.