Abstract
Solar radiation (Rs), a critical variable in agricultural and eco‐environmental processes, is not measured at many meteorological stations. Prediction of Rs has drawn increasing attention in the recent years. However, due to the dynamic nature of atmosphere, accurate estimation of Rs from routinely measured meteorological variables is a challenging task. Studies have demonstrated that machine‐learning approaches outperformed traditional statistical methods. This paper presents an application of Support Vector Machines (SVMs) for monthly mean daily Rs estimates using measured maximum, minimum, and mean air temperatures (Tmax, Tmin, and Tmean, respectively). Twenty‐four stations covering different climatic regions—cold (C), severe cold (SC), mild (M), hot summer and cold winter within the Yangtze River Plain (HSCW‐A), and hot summer and cold winter within the Sichuan Basin (HSCW‐B)—across China were gathered and analysed. Five SVM models with different input attributes were created. These models were also compared with two empirical temperature‐based methods. Root mean squared error (RMSE), Nash‐Sutcliffe coefficient (NSE), and coefficient of residual mass (CRM) were employed to compare the performances of different methods. The newly developed model, SVM using Tmax‐Tmin, and Tmean, outperformed other models with an averaged RMSE of 1.637 MJ m^−2^ and NSE of 0.813. On a regional scale, when Rs was estimated using the parameters developed at other sites, estimation of Rs within HSCW‐B and SC regions were more reliable than in other zones. Especially in HSCW‐B region, estimates of Rs using the parameters developed at Zunyi gave better performance (RMSE = 1.117 MJ m^−2^, NSE = 0.922) than that using the parameters obtained from their own data (RMSE = 1.309 MJ m^−2^, NSE = 0.894). The results showed that the SVM methodology may be a promising alternative to the traditional approach for predicting solar radiation at any locations where the records of air temperatures are available. Copyright © 2010 Royal Meteorological Society