Precipitation prediction in several Chinese regions using machine learning methods

The severity of global climate change is exemplified by the significant increase in extreme precipitation events, leading to an urgent need for accurate rainfall prediction models to mitigate flood disasters that adversely affect economic and social development. With the rapid progress of machine learning in the big data era, novel solutions to regression problems are being proposed. In this paper, we try to construct and evaluate different rainfall prediction models based on specific humidity, relative humidity, horizontal and vertical water vapor flux, and lifting index as variables, using four classic machine learning algorithms: linear regression, random forest regression, support vector regression, and Bayesian ridge regression. The grid search method is employed for hyperparameter tuning, significantly improving the models' prediction accuracy and generalization ability. Evaluation of the predictive performance of the models on nine typical regions in China, including Zhengzhou, Beijing, and Chengdu, demonstrates that the random forest regression model has the highest predictive accuracy, with an average fitting degree of 0.8 or above, followed by support vector regression and Bayesian ridge regression models. Conversely, the linear regression model may have the poorest predictive performance. Therefore, the random forest regression model is recommended for future precipitation prediction, providing a valuable solution to various regression problems. The appropriate selection of variables for prediction and grid search for hyperparameter tuning are possibly the highlights of this paper.