Research On Statistical Methods For Predicting July Precipitation In The Middle And Lower Reaches Of The Yangtze River Based On Winter Sea Temperatur

The middle-lower reaches of the Yangtze River are vulnerable to summer monsoon rainfall,so it is important to improve the seasonal prediction ability of abnormal precipitation in this region.The theoretical studies have demonstrated that the sea surface temperature in the preceding winter can provide physically meaningful indicative signals.Despite the continuous development of prediction techniques,the statistical method is still an important tool for establishing precipitation forecast models with a lead time of one season.This study developed a flexible statistical forecast model and a non-linear neural network prediction model for July precipitation over the middle-lower reaches of the Yangtze River based on the prophase winter sea surface temperature.Furthermore,considering the actual demand of the operational forecasting community,more attention was paid to the abnormal years.However,the previous prediction models using statistical methods were relatively fixed and not flexible enough.Unlike previous studies that focused on selecting prediction factors,this study focused more on how to fully and flexibly use statistical methods to improve prediction skills when constructing prediction models.Before constructing these two prediction models,we first tested the performance of statistical methods through a large number of ideal experiments,such as testing the degree of over fitting generated when using different linear statistical methods,and the reasons for the unstable results when using neural networks in small samples,providing ideas and paving the way for the use of prediction models.According to the characteristics of observed samples and related theoretical knowledge,some more flexible and better–targeted methods were introduced in the forecast model.These special treatments include flexibly defining the prediction area in the middle and lower reaches of the Yangtze River based on the similarity of prediction factors,more flexibly and fully using sea surface temperature information dynamically,expanding artificial synthetic samples combined with theoretical background knowledge,and amplifying the numerical value of abnormal precipitation to enhance the prediction of abnormal precipitation.Rolling forecast experiments show that the linear correlation between prediction and observation is around 0.5,more than half of the abnormal precipitation years can be successfully predicted,and there is no contradictory prediction of the abnormal years.These results indicate that the flexible statistical forecast model is valuable in real-life applications.Furthermore,sensitivity experiments show that forecast skills without these special treatments are obviously decreased.Considering the adverse condition of the small observed sample size,a very simple neural network model was selected to extract the non-linear relationship between input predictors(sea surface temperature)and output predictands(July precipitation in the middle and lower reaches of the Yangtze River)in the forecast system.Considering the instability of the results when using neural networks with small samples,some flexible methods are used and high-quality initial weights and biases are selected through a large number of set experiments to build a prediction model,enhance the stabilization of forecast skills,and provide a method for setting super parameters.Finally,analyze the interpretability of the forecast system.The forecast experiments show that the linear correlation coefficient between the predictands from the forecast system and their corresponding observations is above 0.5,more than half of the abnormal precipitation years can be successfully predicted,and there is no contradictory prediction of the abnormal years.Repeated experiments have shown that the prediction technique of this prediction model is relatively stable.Different extraction methods of sea surface temperature information were used for comparative experiments.The above evaluations suggest that the two forecast systems are valuable in a real application sense.