Precipitation Prediction in Ecuador Using Machine Learning Models

Abstract

Precipitation prediction plays a critical role in water resource management, agriculture, and climate risk mitigation, particularly in regions characterized by strong climatic variability, such as Ecuador. This study investigates the application of machine learning techniques to precipitation prediction using a long-term climatic dataset spanning 1950 to 2023. Three regression models were evaluated: Decision Tree Regressor, Random Forest Regressor, and Gradient Boosting Regressor. The dataset was preprocessed through temporal decomposition, logarithmic transformation of precipitation, and cyclical encoding of seasonal effects to capture long-term trends and annual variability. Model performance was assessed using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R2). Results show the Gradient Boosting Regressor achieving the best performance (MAE = 27.63 mm, RMSE = 37.29 mm, R2 = 0.84). Scenario-based predictions and sensitivity analysis further demonstrate the model’s physical consistency and practical applicability. The findings confirm the potential of machine learning models, particularly gradient boosting, as reliable tools for precipitation prediction and exploratory climate analysis in Ecuador.