Vidyatullah, Icha Sulastri, Halmar Halide
Background. Extreme rainfall in tropical regions, particularly in Makassar and Banjarmasin, frequently triggers flooding that affects infrastructure, economic activities, and community livelihoods. Rainfall variability is influenced by local factors, geographic conditions, and large-scale atmospheric dynamics such as the Madden–Julian Oscillation (MJO), monsoon systems, and regional convective processes. Therefore, an accurate short-term prediction system is required to support mitigation efforts and early warning of hydrometeorological disasters. Objective. This study aims to identify extreme rainfall events based on hourly precipitation data and to compare the performance of the Autoregressive with Exogenous Input (ARX), Random Forest (RF), and Hybrid ARX–RF methods in predicting extreme rainfall up to 24 hours ahead. Methods. Hourly rainfall data were obtained from NASA POWER, and MJO index data were sourced from the Bureau of Meteorology for the period 2019–2025. Modeling was conducted using MATLAB by applying a time-series–based ARX method, an ensemble decision tree–based RF method, and a Hybrid approach based on performance weighting. Model evaluation was performed using the Pearson correlation coefficient and Root Mean Square Error (RMSE). Results. The results indicate that extreme rainfall events are associated with positive and highly variable MJO conditions with strong intensity. The RF model produced the highest correlation and the lowest RMSE for short- to medium-term horizons. The ARX model demonstrated greater stability for very short- and long-term predictions, while the Hybrid model provided more balanced predictions with lower uncertainty levels. Conclusion. The Hybrid ARX–RF model represents the most optimal approach for predicting extreme rainfall up to 24 hours ahead and is recommended as a foundation for developing operational forecasting systems and flood early warning in Makassar and Banjarmasin.
Keywords: extreme rainfall, ARX, Random Forest, Hybrid ARX–RF, Madden–Julian Oscillation, 24-hour prediction
