Evaluating the Performance of the EcPoint Post-Processing Method for Ensemble Rainfall Forecasts over south China

Abstract

Developing reliable weather prediction systems is still a challenging task due to the complexity of the Earth System and the chaotic behavior of its components. Small errors introduced by observations, their assimilation and the forecast model configuration escalate chaotically, leading to a significant loss in forecast skill with time. Traditionally, rainfall forecasts have been generated at grid-based spatial resolutions, providing valuable information on regional precipitation patterns. However, Weather varies markedly within a grid box and forecasts for specific sites have occasionally failed inevitably. The grid-based forecasts may not always meet the needs of decision-makers at specific points of interest. The major challenge is dealing with variations in sub-grid variability, that is to say, the variation seen amongst rainfall point values within a given model grid box, more especially in convective situations. While ensemble forecasts have shown promise in capturing the uncertainty inherent in average rainfall predictions of much larger grid boxes, their utility at point locations has not been extensively explored. Most evaluation studies focus on grid-based verification metrics, which may not accurately reflect forecast performance at individual points of interest. EcPoint, a post-processing approach developed at the European Centre for Medium Range Forecasts (ECMWF), is tailored to forecast rainfall at point locations. In this study, we evaluate the performance of the EcPoint post-processing method over the south China region. The analysis focuses on the reliability, accuracy and discrimination skill of this post-processing method over the three provinces in south China (Anhui, Zhejiang and Jiangsu). We examine performance versus lead time, seasons, and altitude. Through verifications, the study highlighted the added value of the post-processing method over Raw ensemble forecasts. One year of verification demonstrates that, between the Raw ensemble and post-processed EcPoint forecasts, EcPoint is the more reliable and skillful system, adding significant value to most rainfall events occurring during the day and the seasonal associated events, as well as the topography-associated rainfall events. To complement the one-year verification analysis, a case study was conducted on an extremely heavy rainfall event observed on June 2, 2022 at 12 UTC. The analysis demonstrated EcPoint’s ability to provide more localized and refined forecasts whereas Raw didn’t provide any possibility of rainfall, particularly at short lead times. At longer lead times, EcPoint ensembles maintained relatively low probabilities, but offered improved performance in capturing rainfall variability, while Raw ensemble exhibited broader but less precise rainfall predictions with a tendency of over warning of some areas. Future work can extend the evaluation to more diverse climatic and topographic regions of China to enhance the general applicability of the method. Although based solely on the global ECMWF-IFS model, EcPoint performs well over the small domain of south China (three provinces). Besides verifying EcPoint, the study confirms that the post-processing method can significantly improve the forecast performance.