The solar resource has by far the highest energy potential on Earth leading to an increased attention to climate data sets of surface solar radiation (SSR) as an important source of information for solar energy assessment. Such an assessment is essential for solar energy projects policy and planning. The present work aims at evaluating the skill of satellite-based estimates of solar global horizontal irradiance (GHI) from the new “Surface Solar Radiation Data Set–Heliosat edition 3” SARAH-3 at 0.05° spatial resolution and 30min temporal resolution over the South West Indian Ocean, using 1-min pyranometer GHI measurements (in-situ) recorded at different stations of the Indian Ocean Station network (IOS-net) during the year 2020. Both data have been arithmetically averaged to different time scales from the annual means to hourly means for comparison. The data have also been classified into 4 clusters by the use of an unsupervised method of machine learning “K-means clustering” to depict the 4 types of solar regimes. Validation results show that SARAH-3 reproduces fairly the shapes of seasonal cycles and diurnal cycles with different amplitude compared to the in-situ measurements. Stations with higher GHI estimates in SARAH-3 also depict more clear days and fewer overcast days than the in-situ measurements.