Gain of one-month lead predicting spring rainfall over China: A comparison between FGOALS-f2 ensemble prediction system and its driving stretched-grid downscaling prediction system

Abstract

Accurately one-month lead precipitation prediction is crucial for the regional disaster mitigation but challenging. Based on 20 years (2001−2020) hindcasts, this study compared the one-month lead prediction skills for precipitation in April between the Chinese Academy of Sciences Flexible Global Ocean-Atmosphere-Land System, Finite-Volume version 2 (FGOALS-f2) and its driving stretched-grid downscaling prediction system (the Unified global-to-regional prediction system, UGPS), to explore the gain of spring rainfall one-month lead prediction in stretched-grid downscaling prediction system. As a result, UGPS can significantly improve the prediction skill of rainfall for the total amount and extreme frequency in regional absolute amplitude compared with FGOALS-f2. The biases can be evidently reduced by 40%–98% for total rainfall and 76%–98% for extreme rainfall days and non-rainfall days over most of China. And the prediction skills for the interannual anomalies are slightly increased in UGPS but not as significant as the amplitude. The most remarkable improvement in UGPS against FGOALS-f2 is shown in the topographic areas with high altitudes (>2000 m) with a >40% bias reduction of predicted 20-year averaged total rainfall amount and a significant upgradation of prediction skill score for anomalous rainfall amplitude even though the score is still negative. This study suggests that the stretched-grid downscaling prediction system can improve the one-month lead prediction of intensity and extreme, but improving the interannual variability is still dependent on a better prediction from the “master” global climate model like FGOALS-f2.