LAUSR

Abstract The accurate knowledge of diffuse solar radiation is of vital importance for climatology, sustainable energy, agriculture and biological activities. However, the spatial coverage of diffuse solar radiation measurements is limited in many regions, due to the lack of measuring devices, high operation and maintenance costs. Therefore, numerous empirical models have been proposed in different regions and climates for predicting diffuse solar radiation. The aim of this study was to establish, test and compare various models for predicting diffuse solar radiation in China. The performances of newly proposed models were compared with empirical models in this study. Using daily observations at 17 stations during 1993-2015, 97 models with 11 independent variables were established at each station. Meanwhile, the performances of newly-established models were compared with empirical models. The results showed: (1) larger model errors were found at Ejinaqi, Wulumuqi and Kashi stations, due to the dusty air conditions. Relatively poor model performances were also observed at Sanya station, owing to the rainy weather characteristics. (2) The comparisons for the five categories of models showed that the fourth category models with four input parameters generally had higher accuracies, except the case at Wulumuqi. (3) Comparisons of Kd-based with KD-based models showed that kd-based models generally had higher accuracies， the mean MBE, MAE, MARE, RMSE, MPE, t-stat, RRMSE, R and centered RMS for Kd-based models at all 17 stations were −0.43 MJ m−2 day−1, 1.5453 MJ m−2 day−1, 0.2583 MJ m−2 day−1, 2.1422 MJ m−2 day−1, 1.7611%, 15.2127, 0.3134, 0.8111 and 1.9969 MJ m−2 day−1, respectively. (4) By comparing with the models in literature, the newly-established models were better than the models in terms of model performances. The models proposed in this study were valuable for some areas without diffuse radiation record, which also supported the development and utilization of solar energy in China and other regions around the world.