Assessing new intra-daily temperature-based machine learning models to outperform solar radiation predictions in different conditions

•Solar radiation values were predicted using only intra-daily temperature-based data.•Several machine learning models were assessed in different geo-climatic conditions.•The RMSE improvements ranged from 7.56% (arid site) to 45.65% (humid site).•Compared to empirical methods, the mean NSE was increased up to 60% (in summer).•The RMSE was reduced up to 32.27% when the models were used in a non-trained site.