Machine learning-based predictive analysis of ENSO response to temperature and precipitation in southwestern mountains

Abstract

 Accurately predicting the response of ENSO events to temperature and precipitation is of great scientific significance and application value. This study takes Zhaotong City as an example, based on the month-by-month temperature and precipitation observation data of Zhaotong City as well as the Oceanic Nino Index (ONI) from 1961-2023, constructs lagged features, and applies the Random Forest (RF) and XGBoost machine learning models to analyze the correlation between ENSO and climate, and carry out prediction studies. The results show that the RF model performs better in temperature prediction, with a mean absolute error (MAE) of 1.67°C, compared with 1.70°C for the XGBoost model; the precipitation prediction error is significantly higher than that of temperature, and the model's prediction values of extreme high temperature and heavy precipitation events are generally low, and there is a phenomenon of lagging or advancing of the rainy season phase. The study reveals the impact of ENSO on the climate of southwest mountains, indicating the potential of machine learning methods in long-term climate prediction, which can provide important technical support for regional prediction and disaster prevention and mitigation, but the precipitation prediction needs to be further combined with local environmental factors to improve the accuracy.