Estimation of global solar radiation data based on satellite-derived atmospheric parameters over the urban area of Mashhad, Iran.

Abstract

Global solar radiation is the total amount of solar energy received on a horizontal surface and defined as the sum of direct, diffused, and reflected solar radiation. Global solar radiation is an important variable in agricultural, meteorological, hydrological, and climatological studies. The purpose of this paper is to develop an effective method to estimate the daily global solar radiation using different atmospheric properties detected from satellite data, including cloud fraction, cloud optical depth, aerosol optical depth, aerosol exponent, aerosol index, and precipitable water vapor from Moderate Resolution Imaging Spectroradiometer (MODIS) and ozone monitoring instrument (OMI) daytime data in the urban area of Mashhad, Iran, during the years from 2000 to 2018. Based on seven combinations of the atmospheric properties, models were developed using a standard statistical method, namely, multiple linear regression method and a specific class of artificial neural networks, namely, feedforward multilayer perceptron. The efficiency of the models was compared for the assessment of the daily global solar radiation based on the combinations of the input data. For both methods, 80% percent of the data are used for model development and the remaining data for validation. Results of pairwise statistics indicate that, on average, the estimates were more accurate using the artificial neural networks than the regression method. Results show that in both methods, the accuracy of estimation improves when cloud fraction is used as a predictor. This implies the significant effect of cloud cover on solar radiation. However, using the cloud optical depth decreases the accuracy of the estimation of global solar radiation, i.e., the least accurate model is the one with cloud fraction and cloud optical depth for the neural network method and the model with CF and AE for the regression method. The estimation error comes from the inaccuracy in measuring cloud optical depth that depends on satellite sensor resolution and the inhomogeneity of types and microphysical properties of clouds over the study area. Due to the arid climate of the study area, the precipitable water vapor content does not considerably affect radiation attenuation. The best estimate is earned by cloud fraction and aerosol index as inputs indicating the simultaneous role of aerosol and cloud in global solar radiation. Aerosol index considers the effect of absorbing aerosols such as black carbon and dust and is a complementary information to the cloud cover. The results imply that both methods have the potential to achieve an operational stage, taking advantage of the better availability of satellite data. Even though the artificial neural network is found to be more accurate than multiple linear regression, using the regression method is recommended because it is more easy to use. Results show that the effective variables vary in different seasons. In both methods, estimation error is highest in the spring and lowest in the fall and winter. The high inaccuracy may be due to the high sensitivity of radiative transfer to atmospheric condition in spring. On the other hand, the high accuracy may be caused by the less solar radiation fluctuations during fall and winter because of the lower solar radiation flux.