Estimation of daily global solar radiation using empirical and machine-learning methods: A case study of five Moroccan locations

Abstract Reliable solar radiation data are essential to study the feasibility and to determine the optimal size of solar plants in a particular location. However, the network of solar radiation measurement stations is very weak around the world. This has led to the development of a wide variety of models and techniques for estimating solar radiation from commonly measured meteorological variables (Ambient air temperature, relative humidity, wind velocity …). In this work, 22 empirical models, Artificial Neural Networks (ANN) techniques, and tree-based ensemble methods were tested in estimating the daily global solar radiation (GSR) in five Moroccan locations. Each database was partitioned into two sets: the training set and the validation set. The training set is used to calibrate the models, while the validation set is used to assess their credibility. The best-performing model at each station was selected based on three statistical indictors, the coefficient of correlation (R), the normalized mean absolute error (nMAE) and the normalized root mean square error (nRMSE). The results on the validation dataset revealed that the Random Forest method (R: 87.53–96.20%; nMAE: 5.84–11.81%; nRMSE: 7.85–15.33%) outperformed the all tested models in terms of accuracy. The other machine learning methods achieved generally good performance (R: 81.73–95.14%; nMAE: 5.88–13.86%; nRMSE: 8.22–18%). Among the empirical models (R: 56.58–93.46%; nMAE: 6.96–21.83%; nRMSE: 9.89–26.96%), the Temperature - Geographic factors model (TG1) (R: 72.38–93.46%; nMAE: 6.96–17.94%; nRMSE: 9.89–22.39%) can be recommended for estimating the GSR for the all considered stations.