A Climate Change Prediction System Based on Machine Learning Algorithms

In order to improve the precision of traditional methods, we propose a climate change forecasting system based on machine learning. Based on the analysis of the historical return data from the BCC-CSM Seasonal Climate Forecast System and the National Meteorological Information Center of China, a variety of methods have been used to examine the various factors affecting the forecast results. The Long Short Term Memory (LSTM) network is used to predict Chinese summer rainfall in 2020 and 2021. The results show that the LSTM network performs better than the BP neural network and the model output method. In particular, the performance of LSTM networks has been greatly affected by parameter optimization, including the training rate, and the learning rate. The selection of appropriate starting month data in particular enhances seasonal forecasting accuracy, with April data proving particularly effective for predicting summer precipitation. Additionally, incorporating the sea ice component factor positively contributes to seasonal precipitation prediction. In 2021 and 2022 summer precipitation return experiments, the LSTM network showed remarkable prediction performance, achieving 74 and 71 Ps, and 55.52% for anomaly symbols, respectively. The results exceed, on average, the results of national consultations and business models over the same period. In conclusion, the LSTM network offers improved prediction accuracy compared to traditional methods, presenting a promising avenue for enhancing seasonal precipitation forecasting.