A review of the application of hybrid machine learning models to improve rainfall prediction

Rainfall is one of the most important meteorological phenomena that impacts many fields, including agriculture, energy, water resources management, and mining, among others. While machine learning (ML) models have shown great potential in rainfall forecasting as they perform well and sometimes better than some physical models, the complex physical processes involved in rainfall formation make single ML models insufficient for providing accurate rainfall estimates in most cases. Although there are comprehensive reviews of the performance evaluation of individual ML models in the literature, only a limited number of reviews exist that include hybrid models that specifically focus on rainfall forecasting. This paper presents an extensive review of the performance of hybrid ML models for rainfall forecasting. The vital information on the forecasting time scales, model inputs, and evaluation methods used for constructing these models has been analysed and discussed. The findings revealed that hybrid ML models composed by integrating data pre-processing techniques and optimisation algorithms may be a successful and efficient solution to enhance rainfall predictions at various timescales. Hybrid ML models used for rainfall predictions are capable of producing comparatively more accurate forecasts and reducing uncertainty for both short and longer lead times. Recent advances in physical-ML hybrid models for weather forecasting have also been highlighted. Overall, this review article provides useful information to researchers interested in developing early warning systems for precise and timely rainfall forecasting.