Accurate rainfall prediction during the monsoon season is vital for disaster risk reduction, mitigating floods, landslides, and related hazards. While traditional rainfall forecasting primarily relies on local meteorological variables such as humidity, temperature and pressure, the potential of atmospheric radionuclides remains largely underexplored. This study investigates Beryllium-7 ( \(^7\text {Be}\) ), a cosmogenic radionuclide, as a novel predictor for monsoonal rainfall forecasting. Acting as a tracer of large-scale atmospheric circulation and vertical air mass exchange, \(^7\text {Be}\) captures synoptic-scale processes and monsoon dynamics that precede precipitation. Its ability to provide early atmospheric signals makes \(^7\text {Be}\) particularly suitable for lag-based rainfall prediction. \(^7\text {Be}\) data from five CTBTO radionuclide stations and Malaysian rainfall stations were analyzed using the CatBoost Regressor. Lag and sliding window techniques were applied to capture temporal dependencies, while Optuna was used for hyperparameter optimization. Results show that \(^7\text {Be}\) intensity, combined with temporal features, significantly improves rainfall prediction. The CatBoost model achieved R \(^2\) values of 0.9237 (validation) and 0.9020 (test) for Kota Bharu, 0.9507 and 0.9312 for Kuala Terengganu, and 0.9187 and 0.8686 for Kuantan. These consistent outcomes underscore \(^7\text {Be}\) ’s robustness as a novel and reliable predictor in monsoonal rainfall forecasting, offering valuable insights for climate research and disaster management.
