Rainfall Variability Modeling Using Machine Learning

Abstract

Rainfall variability represents a major challenge for water resources management, agriculture, and climate risk assessment in semi-arid regions such as northeastern Algeria. This study investigates the spatio-temporal variability of annual rainfall and evaluates the performance of machine learning techniques for rainfall modeling and prediction. Annual precipitation data from twelve meteorological stations distributed across northeastern Algeria were analyzed for the period 2007–2022. To capture both temporal and spatial controls on rainfall variability, a set of explanatory variables was constructed, including year, geographic coordinates (latitude and longitude), elevation, and antecedent rainfall at one- and two-year lags. Prior to model development, the dataset underwent comprehensive preprocessing, including quality control, handling of missing values, standardization, and feature selection to reduce redundancy and multicollinearity. The dataset was split into training (70%) and testing (30%) subsets to ensure robust model evaluation. Several machine learning models were trained and compared in terms of their predictive performance using standard statistical metrics (e.g., RMSE, MAE, and coefficient of determination, R²). The results demonstrate that machine learning approaches are capable of effectively capturing the non-linear relationships governing rainfall variability in northeastern Algeria, outperforming conventional approaches in terms of accuracy and generalization ability. The proposed framework provides a valuable tool for improving rainfall prediction and supporting sustainable water resources management and climate adaptation strategies in data-scarce semi-arid environments.