Improving Extraseasonal Summer Rainfall Prediction by Merging Information from GCMs and Observations

Abstract

A new prediction approach for summer (June?August) rainfall in China was designed by considering both preceding observations and numerically predicted summer rainfall through a multivariate linear regression analysis. First, correlation analyses revealed close relationships between summer rainfall in parts of China with the Antarctic Oscillation (AAO), the Arctic Oscillation (AO), and sea surface temperatures (SSTs) in the preceding winter (December?February). The Huang-Huai Valley, two subregions of the Jiang-Huai Valley, the southern Yangtze River, south China, and southeastern Xinjiang were then chosen as targets for their regional climate characteristics. Following this, an extraseasonal (one season in advance) regression prediction model for regionally averaged summer rainfall was constructed by using these three climate factors and a 3-month lead-time forecast of summer rainfall, undertaken by an atmospheric general circulation model (GCM) forced by observed SSTs, as predictors region by region. To improve the accuracy of prediction, the systematic error between the original regression model?s results and its observational counterparts, averaged for the last 10 yr, was corrected. Using this new approach, real-time prediction experiments and cross-validation analyses were performed for the periods 2002?07 and 1982?2007, respectively. It was found that the new prediction approach was more skillful than the original or corrected GCM prediction alone in terms of sign, magnitude, and interannual variability of regionally averaged summer rainfall anomalies in all regions. The preceding observations were the major source of the prediction skill of summer rainfall in each region, and the GCM predictions added additional prediction skill in the western Jiang-Huai Valley and southeastern Xinjiang, in both of which the GCM prediction was used as a predictor.