Abstract
Downward surface shortwave radiation (DSSR) plays an important role in the energy balance of the earth’s surface. Accurate estimate of DSSR is of great significance for the rational and effective use of solar energy. Some parameterization schemes were proposed to estimate DSSR using meteorological measurements given ground-based radiation observation sites are scare and uneven. With the development of remote sensing technique, remotely sensed data can be applied to obtain continuous DSSR in space. Commonly, the spatial resolution of most radiation products is relatively low and cannot meet the needs of certain fields. Moreover, some retrieval algorithms based on the radiation transfer models are complicated for non-professionals. In this study, a back-propagation (BP) neural network method with Levenberg−Marquardt (LM) algorithm (hereafter referred to as LM-BP) was applied to predict DSSR by building the relationship between measured DSSR and high-resolution remote sensing data from the Advanced Space-borne Thermal Emission Reflectance Radiometer (ASTER). The DSSR observations from the four-component radiation sensor installed at the land covered by vegetable, village, maize, orchard, Gobi, sandy desert, desert steppe, and wetland were used to validate the model estimates. The results showed that the estimates of DSSR from LM-BP agreed well with the site measurements, with the root mean square error (RMSE) and the mean bias error (MBE) values of 27.34 W/m2 and −1.59 W/m2, respectively. This indicates that by combining the LM-BP network model and ASTER images can obtain precise DSSR in heterogenous surface. The DSSR results of this study can provide accurate high-spatial resolution input data for hydrological, evapotranspiration, and crop models.