Comparative Evaluation of Three Schaake Shuffle Schemes in Postprocessing GEFS Precipitation Ensemble Forecasts

Natural weather systems possess certain spatiotemporal variability and correlations. Preserving these spatiotemporal properties is a significant challenge in postprocessing ensemble weather forecasts. To address this challenge, several rank-based methods, the Schaake Shuffle and its variants, have been developed in recent years. This paper presents an extensive assessment of the Schaake Shuffle and its two variants. These schemes differ in how the reference multivariate rank structure is established. The first scheme (SS-CLM), an implementation of the original Schaake Shuffle method, relies on climatological observations to construct rank structures. The second scheme (SS-ANA) utilizes precipitation event analogs obtained from a historical archive of observations. The third scheme (SS-ENS) employs ensemble members from the Global Ensemble Forecast System (GEFS). Each of the three schemes is applied to postprocess precipitation ensemble forecasts from the GEFS for its first three forecast days over the mid-Atlantic region of the United States. In general, the effectiveness of these schemes depends on several factors, including the season (or precipitation pattern) and the level of gridcell aggregation. It is found that 1) the SS-CLM and SS-ANA behave similarly in spatial and temporal correlations; 2) by a measure for capturing spatial variability, the SS-ENS outperforms the SS-ANA, which in turn outperforms the SS-CLM; and 3), overall, the SS-ANA performs better than the SS-CLM. The study also reveals that it is important to choose a proper size for the postprocessed ensembles in order to capture extreme precipitation events.