Comparison of four heuristic regression techniques in solar radiation modeling: Kriging method vs RSM, MARS and M5 model tree

In this study, four different heuristic regression methods including Kriging, response surface method (RSM), multivariate adaptive regression (MARS) and M5 model tree (M5Tree) have been investigated for accurate estimating of solar radiation with different input data. Monthly solar radiation (SR) from Adana and Antakya stations, which are located in Eastern Mediterranean Region of Turkey is estimated based on the input data of maximum temperature (Tmax), minimum temperature (Tmin), sunshine hours (Hs), wind speed (Ws), and relative humidity (RH). In Adana station, the best MARS model provided slightly better accuracy than the Kriging, RSM and M5Tree while the Kriging was found to be the better than the MARS, RSM and M5Tree in Antakya station. The predictions of M5Tree model are shown inaccurate results for both maximum errors and minimum agreement compared to another models. The effect of periodicity input is examined to obtain the accurate predictions of solar radiation for these stations based on the four heuristic -based modeling Kriging, MARS, RSM, M5Tree approaches. Periodicity input data improved the root mean square errors of the best MARS, RSM, M5Tree and Kriging models as 34%, 37%, 46% and 39% for Adana station and by 51%, 47%, 38% and 49% for Antakya station, respectively. The periodic Kriging models performed superior to the periodic MARS, RSM and M5Tree models.