Abstract
Accurate rainfall forecasting is vital for managing water resources, preventing floods, supporting agricultural activities, and enhancing disaster preparedness. Traditional forecasting methods, such as linear regression, autoregressive models, and time-series analysis, are limited in their ability to capture the intricate and dynamic nature of rainfall patterns. To address these shortcomings, this study utilizes a fusion of observed rainfall data and climate change projections to improve the precision of rainfall predictions over daily, 3-day, and weekly intervals. The performance of several advanced machine learning models was assessed, with the Efficient Linear Support Vector Machine (ELSVM) showing the highest accuracy in daily rainfall forecasting, yielding an R² value of 0.3868, indicating its ability to effectively capture the variability in rainfall. For the 3-day forecasting interval, Exponential Gaussian Process Regression (Exponential GPR) marginally outperformed Long Short-Term Memory (LSTM), with Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) values of 15.84, 547.04, and 23.39, respectively. On the other hand, LSTM demonstrated higher error rates, with MAE, MSE, and RMSE values of 14.07, 363.03, and 19.05, respectively, and an R² value of 0.1662 for weekly forecasts. These findings underscore the significant potential of combining advanced machine learning models with data fusion techniques to enhance the accuracy and reliability of rainfall predictions, offering meaningful contributions to water resource management, climate adaptation, and the development of more robust forecasting systems.