Advanced Machine Learning–Based Temperature Forecasting for the Jombang Region Using ERA5 Reanalysis Data (2020–2025)

Abstract

Accurate temperature forecasting plays a crucial role in supporting climate-sensitive sectors such as agriculture, environmental management, and public health, particularly in tropical regions with complex atmospheric dynamics. This study presents a machine learning-based framework for short-term air temperature forecasting in the Jombang region, Indonesia, utilizing ERA5 reanalysis data from 2020 to 2025. The dataset was preprocessed through temporal alignment, anomaly handling, and lag-based feature engineering to effectively capture diurnal temperature variations. Three machine learning models—Long Short-Term Memory (LSTM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—were developed and evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The experimental results demonstrate that XGBoost achieved the best performance, with an MAE of 0.5982 °C, RMSE of 0.8279 °C, and MAPE of 2.2320%, outperforming both LSTM and Random Forest.These findings suggest that boosting-based ensemble learning is more effective in modeling nonlinear temperature patterns compared to recurrent and bagging-based approaches. This study contributes to the advancement of localized temperature forecasting models in tropical regions and provides a practical reference for selecting suitable machine learning methods for regional climate applications.