Evaluation of temperature-based global solar radiation models in China

Estimation of global solar radiation (R-s) from the daily range of air temperature (Delta T) offers an important alternative in the absence of measured R-s or sunshine duration because of the wide availability of air temperature data. In this paper, we assessed 16 R-s models including modified versions of the Bristow and Campbell (B-C) and the Hargreaves (Harg) models across a wide range of agro-ecological conditions in China. Using long-term data from 15 sites in Northeast, North China Plain and Northwest China, we explored the main factors affecting model parameters and the predictive accuracy and proposed empirical equations to estimate these parameters. Two schemes in calculating Delta T were employed: Delta T-1 (based on single day T-min) used in the Harg and Delta T-2 (based on 2-day average of T-min) used in the B-C model. Results showed that the original B-C model performed similarly to the best performing modified Harg model, but significantly outperformed the original Harg model with a 4-7% higher accuracy. The common practice of fixing some parameters in the B-C model caused the most significant effect and resulted in a 3-9% lower accuracy than that of the original model. In contrast, modifications had the smallest effect and yielded little improvement and are thus unnecessary. The Delta T scheme had a moderate effect, with Delta T-1 generally resulting in a higher accuracy especially in high altitude areas. This indicates that the effect of cold or warm air advection is negligible in the B-C model even in a temperate climate. The accuracy of the temperature-based models was affected mainly by the magnitude of Delta T, with larger Delta T resulting in higher accuracy. Parameters of the B-C model correlated significantly with many commonly used geographical and meteorological factors, meaning that they are more easily obtainable without the need for calibration and consequently R-s measurement. The parameter of the original Harg model correlated insignificantly with all the examined factors, meaning that it has to be calibrated. Main findings in this study are valuable in clarifying the relative impact of different approaches in applications on model accuracy, in demonstrating the advection effect, in identifying the dominant factors of model parameters and thus in increasing their availability. (C) 2009 Elsevier B.V. All rights reserved.