Exploring Use of Machine Learning Regressors for Daily Rainfall Prediction in the Sahel Region: A Case Study of Matam, Senegal

Abstract

Rainfall is the major source of water for rain-fed agricultural production in Sub-Saharan Africa. Overdependency on rain-fed agriculture renders Sub-Saharan Africa more prone to adverse climate change effects. Consequently, timely and correct long-term daily rainfall forecasting is fundamental for planning and management of rainwater to ensure maximum production. In this study, we explored use of regressors: Gradient Boosting, CatBoost, Random Forest and Ridge Regression to forecast daily rainfall for Matam in the northern geographical part of Senegal. Gradient Boosting model is therefore considered a better model with smaller values of Mean Absolute Error, Mean Squared Error and Root Mean Squared Error of 0.1873, 0.1369 and 0.3671 respectively. Further, Gradient Boosting model produced a higher score of 0.69 for Coefficient of Determination. Relative Humidity is perceived to highly influence rainfall prediction.