The accurate estimation of solar irradiance remains a significant challenge in tropical environments, where high cloud variability and rapid weather fluctuations frequently disrupt solar radiation patterns. These conditions hinder the reliability of solar energy forecasting, which is critical for the integration of renewable energy and grid stability. In response to this challenge, this study introduces a novel machine learning framework that leverages complementary cloud information from satellite- and ground-based sources to estimate global horizontal irradiance in Jakarta, Indonesia. Three machine learning models were trained using a comprehensive dataset combining physical solar parameters with cloud index from GK2A satellite and cloud fraction from sky images. Among these, CatBoost demonstrates the most robust performance, achieving the lowest root mean square error (RMSE) of 54.48 W/m² and mean absolute percentage error (MAPE) of 20.71 % under all-sky conditions. Notably, under clear-sky conditions, CatBoost achieves exceptional accuracy with an RMSE of only 20.47 W/m² and MAPE of 1.01 %. These results show the effectiveness of integrating multi-source cloud information to improve solar irradiance estimation. The proposed framework not only enhances the methodological accuracy but also improves the reliability of solar power forecasting, supports efficient grid management, and promotes the integration of renewable energy in tropical urban environments. • Solar irradiance estimation in tropical environments. • Integration of satellite-derived CI and ground-based CF from sky images. • GHI estimation at 10-minute intervals. • CatBoost performed best under clear skies, with an RMSE of 20.47 W/m².
Aditya et al. (Thu,) studied this question.