Multivariate Variable-Weight Combination Prediction Model for Summer Precipitation in Guangxi Based on Statistical Methods and Intelligent Algorithms

Summer (June–August) precipitation data of 88 meteorological stations in Guangxi during 1961–2022, provided by the Guangxi Climate Center, were utilized along with 142 circulation characteristic indices and regionally Sea-Surface Temperature (SST) indices from the National Climate Center. Key predictors were selected via correlation analysis. Using these factors, individual prediction models for summer precipitation in Guangxi were constructed using multiple linear regression, random forest, and Back Propagation (BP) neural networks. Subsequently, these were integrated into a multivariate variable-weight combination model by using the variance reciprocal weighting method. The three individual models and the combined model were then used to fit summer precipitation data during 1961–2017 and to predict precipitation for the period 2018–2022. The results show that the multivariate variable-weight combination model consistently outperformed each model in fitting and predictive accuracy. To further validate the effectiveness of the combined approach, Empirical Orthogonal Function (EOF) analysis was applied to the summer precipitation anomaly series during 1961–2017 from the 88 stations. The multivariate variable-weight combination predictions were also evaluated on the time series corresponding to the first three EOF modes. The results corroborated the conclusions drawn from single-station predictions: The average comprehensive score for the trend anomaly Prediction skill (Ps) of the multivariate variable-weight combination predictions from the 88 stations during 2018–2022 reached 88.8, with annual Ps scores consistently exceeding 80, indicating significantly better and more stable performance compared with all three individual models. This study provides a new approach for summer precipitation prediction in Guangxi and the development of combined prediction models.