Seasonal Weather Pattern Prediction From Enso Indices Using Machine Learning

The accurate prediction of seasonal weather patterns holds significant importance in supporting agriculture, disaster management, and economic planning in Bangladesh. However, the non-linear characteristics of weather and climatic patterns makes it quite challenging. Recently, the significant impact of El Nino-Southern Oscillation (ENSO) indices on regional climate variability have increasingly been recognized. This study investigates the correlation between nine ENSO indices and both temperature and rainfall patterns across Bangladesh and also evaluates the effectiveness of machine learning (ML) models in predicting these weather variables. Historical monthly data from 29 meteorological stations, spanning 1977 to 2022, were analyzed. Six supervised ML models—Random Forest (RF), Decision Tree (DT), K-Fold Cross-Validation (KFCV), XGBoost (XGB), Linear Regression (LR), and K-Nearest Neighbors (KNN) were applied. Performance was evaluated using R² score, Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). Results revealed that ENSO indices have a notable impact on climate parameters in Bangladesh. Among these models, XGB achieved the highest R² scores for temperature prediction, with values of 0.8824 (T _max ), 0.9706 (T _min ), and 0.9559 (T _avg ). RF and KFCV also showed strong performance, with RF achieving R² values of 0.8770 (T _max ), 0.9699 (T _min ) and 0.9531 (T _avg ) and KFCV achieving R² scores of 0.8606 (T _max ), 0.9619 (T _min ), and 0.9438 (T _avg ). Rainfall prediction, however, yielded lower accuracy, with RF recording the highest R² of 0.6273. The study highlights the impact of ENSO indices and concludes that XGB, RF, and KFCV are highly effective in modeling seasonal climate patterns influenced by ENSO.