Asymmetry of winter precipitation event predictions in South China

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Winter precipitation anomalies in South China (SC) frequently result in severe disasters. However, the evaluation of prediction performance and distinctions between positive precipitation anomaly events (PPA, wet condition) and negative precipitation anomaly events (NPA, dry condition) in current operational models remains incomplete. This study employed the Climate Forecast System version 2 (CFSv2) to assess winter precipitation prediction accuracy in SC from 1983 to 2021. Differences in predicting PPA and NPA events and the underlying physical mechanisms were explored. The results indicate that CFSv2 can effectively predict interannual variations in winter precipitation in SC, as there is a significant time correlation coefficient of 0.68 (0.62) between observations and predictions, with a lead time of 0 (3) months. The model revealed an intriguing asymmetry in prediction skills: PPA outperformed NPA in both deterministic and probabilistic prediction. The higher predictability of PPA, as indicated by the perfect model correlation and signal-to-noise ratio, contributed to its superior prediction performance when compared to NPA. Physically, tropical signals from the ENSO and extratropical signals from the Arctic sea ice anomaly, were found to play pivotal roles in this asymmetric feature. ENSO significantly impacts PPA events, whereas NPA events are influenced by a complex interplay of factors involving ENSO and Arctic sea ice, leading to low NPA predictability. The capability of the model to replicate Arctic sea ice signals is limited, but it successfully predicts ENSO signals and reproduces their related circulation responses. This study highlights the asymmetrical features of precipitation prediction, aiding in prediction models improvement.
Funding Information
  • National Natural Science Foundation of China (42005049, 42130607, 42288101)
  • Shandong Natural Science Foundation Project (ZR2020QD056, ZR2019ZD12)

This publication has 95 references indexed in Scilit: