Evidence, limitations and implications for renewable energy planning of daily global solar irradiation in tropical climate zones of Cameroon, predicted by machine learning

Abstract

Accurate solar irradiance prediction is a crucial step for the efficient planning and deployment of solar energy systems, particularly in tropical regions like Cameroon where solar resources remain largely untapped. Despite growing global interest in solar energy prediction using machine learning techniques, this remains a significant challenge. Similar studies exist in Cameroon, but they remain few compared to those in Asia, Europe, or North Africa. In contrast, our study stands out due to its complementary geographic coverage, the inclusion of solar vortex mass (SVM), and the in-depth analysis of predictive meteorological variables (particularly cloud cover). This study aims to help fill this gap by predicting daily global solar irradiance on horizontal surfaces in five Cameroonian cities Bertoua, Douala, Garoua, Maroua, and Yaoundé characterized by relatively diverse climatic conditions. Four machine learning algorithms were used for this purpose: Linear Regression, Decision Tree, Random Forest, and Support Vector Machine. Meteorological parameters including mean air temperature, relative humidity, precipitation, wind speed, wind direction, day of the year, atmospheric pressure, and cloud cover were used as input variables for training the models. The dataset was obtained from NASA’s POWER database over 27 years and covers the period from 1999 to 2026. Model performance was evaluated using three statistical indicators: mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R 2 ). The results show that RMSE values range from 0.58 to 0.77 kWh/m 2 , MAE values from 0.40 to 0.59 kWh/m 2 , and R 2 values from 0.52 to 0.69, across all cities and models. Among the four algorithms evaluated, the Random Forest model consistently achieved the best predictive performance in all cities, followed by the Support Vector Machine, the Decision Tree model, and the linear model. These results suggest that the Random Forest algorithm is the most suitable model for predicting global daily solar irradiation in Cameroon and can serve as a preliminary selection tool for assessing solar energy resources in the region, while enriching the scientific literature on solar irradiation modeling in Central Africa.