Machine Learning Prediction of the Rainy-Season Onset in Southeast Africa

Authors:
Preprint
Publisher website
Google Scholar
Abstract
Accurate prediction of the rainy-season onset is crucial for many in Southeast Africa. This paper explores the application of the Random Forest (RF) machine learning method to forecast the start of the October-November-December (OND) rainy season across the region. Our RF model was trained to predict onset dates using historical precipitation data, observations and forecasts of key climate drivers, such as the El Niño-Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD), and onset forecasts derived from dynamical seasonal forecast models. Comparing our RF model with the widely used ECMWF System 5.1 dynamical forecast showed improvements of 2-15 days in RMSE, particularly for forecast initialised in August-September, and CRPSS gains of up to +0.15 for individual countries for the same start dates. The results demonstrate that our RF model captures regional variability in onset timing and consistently outperforms climatology. Notably, predictions for the 2023 and 2024 rainy seasons reveal significant shifts driven by ENSO phase transitions. This study highlights the robustness of our RF approach in dealing with complex climate variables and underscores the potential for improving seasonal rainfall forecasts in the region, helping stakeholders make more informed decisions based on these forecasts.

This publication has 12 references indexed in Scilit: