Dynamic downscaling ensemble forecast of an extreme rainstorm event in South China by COSMO EPS

Abstract

In this study, the high-resolution ensemble prediction system COSMO (Consortium for Small Scale) EPS is used to predict the extreme rainstorm that occurred from 27 to 31 August 2018 in Guangdong Province, China, which leads to intensities exceeding historical extreme values. COSMO EPS is run with a 2.8-km grid spacing, allowing for an explicit treatment of deep convection, and 24 members of the EPS are initialized and laterally driven by the ICON (ICOsahedral Nonhydrostatic) global model. We compare the predictions of COSMO EPS against observations derived from the global precipitation measurement (GPM) and with ensemble forecasts of both mesoscale EPS and global EPS provided by GRAPES (Global and Regional Assimilation and PrEdiction System), and with the deterministic forecasts of global models ICON and ECMWF (European Centre for Medium-Range Weather Forecasts). Model performances are evaluated both by gridpoint-based scores, such as the equitable threat score (ETS), and by the Method for Object-based Diagnostic Evaluation (MODE) for spatial verification. According to our results, COSMO EPS could perform better forecasts for the rainstorms taken place in eastern Guangdong than other models. However, the location and coverage area of its predicted rainstorm is eastward and smaller in contrast with the observations. Therefore, COSMO EPS exhibits relative high performance by object-based spatial evaluations, while it could not display evident superiority in terms of the gridpoint-based scores. The cause analysis of this extreme rainstorm shows that Guangdong Province of China is mainly affected by monsoon depression. Southwesterly and southerly winds continuously transport water vapor from the South China Sea to Guangdong Province. The southwest monsoon low-level jet advances northward over time, which promotes the occurrence and development of continuous heavy precipitation in the coastal areas of Guangdong. In an additional experiment, we investigate the benefit of assimilation of radar data, by applying the latent heat nudging (LHN) approach based on surface-based radar observations to the COSMO EPS. Subsequently, the prediction by assimilation of radar data more reasonably reproduces the spatial distribution of precipitation observations, while the coverage and intensity of the rainstorm in eastern Guangdong are still not reflected satisfactorily.