Comparison of deterministic and data-driven models for solar radiation estimation in China

Solar radiation is an indispensable input for many applications, contributing to different fields, including energy, meteorology, ecology, agriculture and industry. A lot of parameterization schemes have been developed for estimating solar radiation in sites around the world. This paper presented a comparative study on the performances of four shortwave solar radiation (SSR) models in different climates, including Yang’s hybrid model (YHM), an efficient physically based model (EPP), an hourly solar radiation model (HSRM) and a neural network model (ANNM). Daily meteorological variables observed at 837 stations in China were used as model inputs for YHM and ANNM. MODIS atmospheric and land products (MOD08_D3, MYD08_D3, MOD08_M3, MYD08M3, MOD09CMG, and MYD09CMG) were used to derive the required parameters for EPP and HSRM. Cloud fraction and solar zenith angle were found to be the major parameters influencing the model accuracies. The results indicated that YHM performed superior to EPP, ANNM and HSRM with daily mean RMSE of 2.414, 2.535, 2.855 and 3.645 MJm−2day−1, respectively. The monthly mean RMSE for all models were generally higher in July (3.37MJm−2day−1) and lower in January (1.997 MJm−2day−1). It was observed that the monthly mean RMSE was 2.95 MJm−2 day−1 in humid areas, while it is 2.773 MJm−2day−1 in semi-arid areas. Monthly and annual mean SSR (ASSR) during 2002–2015 were calculated to reveal the spatial and temporal variations of SSR across China using daily meteorological data, MOD08_M3 and MOD08_M3 products based on YHM and EPP models. The result showed that there was not obvious variation trends for ASSR in China, the largest value (14.521 MJm−2day−1) was observed in 2003, while the smallest ASSR (14.182 MJm−2day−1) was in 2014; the ASSR values were generally higher in Qinghai-Tibet and lower in northeastern China.