Multifactor Descending Dimension Method of Objective Forecast for Summer Precipitation in Southwest China

In this paper, the existing criteria of EOF (Empirical Orthogonal Function) modes temporal stability are improved, and a scheme for electing stable high correlation prediction signals is proposed. Moreover, the temporal stability of EOF modes, the temporal and spatial characteristics, and the key signals to predictable modes of summer precipitation anomaly percentage from 1981 to 2016 in Southwest China are analyzed. Hence, a multi-factor descending dimension prediction model is established. The results show that the first nine modes are stable within 3 years lead and in recent 10 years for climate prediction. They account for nearly 70% of the variance contribution rate and primary to summer precipitation anomaly percentage in Southwest China. The optimal prediction signals and the corresponding equations for primary PCs (Principal Component) are selected and built with the stable high correlation concept and the optimal subset regression method. These equations are recognized for having good capabilities for PC fitting. Their complex correlation coefficients range from 0.62 to 0.84, all passing the significance test of 99.99%, and their sign coincidence rates are greater than 69%. The prediction model built on that basis has a good hindcast skill for the spatial distribution, variation trend, and abnormal level of summer precipitation in Southwest China. Its mean ACC (Anomaly Correlation Coefficient) score is 0.58. The TCCs (Temporal Correlation Coefficient) pass the 90% significance test in the whole region except for sporadic stations and pass the 99.9% significance test in most areas. While the mean PS (Prediction Score) score is 84, the mean PS score is 87.1 for precipitation in the most abnormal 10 years. According to forecast test in 13 years (1971–1980 and 2017–2019), this model’s mean PS score is 72, and the mean PS score is 77, which is higher than that of published forecasts in 2017–2019.