Abstract: Diffuse solar radiation (R_dif) and direct solar radiation (R_dir) constitute the two fundamental components of the global solar radiation (R_g) received by horizontal surfaces, and they are crucial for photovoltaic research. Aerosols and clouds are key factors influencing their variations under all-sky conditions. However, the accuracy of existing radiation products is insufficient to support studies on aerosols radiative effects (ARE) and clouds radiative effects (CRE) on R_dif and R_dir. This study develops an interpretable ensemble model to derive high spatiotemporal resolution (10 km, 1 h) R_dif and R_dir in 2017 from European Centre for Medium-Range Weather Forecasts Reanalysis Version 5 (ERA5)-Land R_g and quantify the global ARE and CRE on their distribution. The developed Stacking model shows high accuracy and stability, with R_dif (R = 0.89, MBE = 3.17 W/m², RMSE = 50.98 W/m²) and R_dir (R = 0.91, MBE = −8.56 W/m², RMSE = 82.94 W/m²). SHapley Additive exPlanations (SHAP) values confirm the dominant role of R_g (SHAP > 0.2), followed by total cloud cover (TCC), while also revealing complex nonlinear interactions among features. Comparisons with other radiation products including ERA5-Single, Deep Space Climate Observatory/Earth Polychromatic Imaging Camera (DSCOVR/EPIC), and Clouds and the Earth’s Radiant Energy System (CERES) show that the developed dataset more accurately captures the spatial variability of R_dif and R_dir. Analysis of ARE and CRE indicates that their spatial distribution patterns closely align with aerosol optical thickness (AOT) and TCC distributions. The strongest aerosol-induced R_dif enhancement or R_dir attenuation (ARE = 62.92 W/m²) occurred in the Around Arabian during peak aerosol (June), while the most pronounced cloud-induced R_dif enhancement or R_dir attenuation (CRE = 63.94 W/m²) was observed in Central and Eastern South America during peak cloudiness (January). On an annual scale, CRE generally exceeded ARE in most regions. This study provides valuable reference for generating high-precision global R_dif and R_dir data and analyzing the ARE and CRE on R_dif and R_dir at a global scale.