Quantification and prediction of the impact of ENSO on rainfed rice yields in Thailand

Climate variability driven by the El Nino-Southern ˜ Oscillation (ENSO) significantly impacts rainfed rice yields in Thailand, a critical agricultural region heavily reliant on monsoon rainfall. This study quantifies and predicts the effects of ENSO-induced climate signals on rice yields using advanced artificial intelligence (AI) techniques. We employed a three-stage methodology, integrating Multiple Linear Regression (MLR) with Variance Inflation Factor (VIF) analysis to assess the relative contributions of ENSO indices and local climate variables, followed by the development of two AI models: ENSO–CropNet, a deep neural network (DNN), and an ensemble Random Forest-XGBoost (RF-XGBoost) model. The results revealed that ENSO indices, particularly NINO3 and NINO3.4, significantly reduced rice yields in several provinces, with temperature and rainfall variability playing critical roles. The ENSO–CropNet model demonstrated high predictive accuracy (R² = 0.89, MAE = 1.04, RMSE = 1.45), surpassing the RF-XGBoost model (R² = 0.82, MAE = 3.62, RMSE = 3.84). Feature importance analysis identified rainfall, minimum temperature, and ENSO indices as key predictors. The study found that ENSO-driven climate variability led to a 12 % decline in rice yields across northern provinces. The findings underscore the significant role of ENSO-induced climate variability in rainfed rice production, with AI models such as ENSO–CropNet offering highly accurate predictions. These results highlight the potential of AI techniques to enhance agricultural forecasting and resilience in climate-vulnerable regions like Thailand.