Evaluation of Various Methods of Estimating Solar Radiation for Crop Model Application In the Southeastern USA.

Global solar radiation (R_g) is an important input for crop models to simulate crop responses. Because the scarcity of long and continuous records of R_g is a serious limitation in many countries, including the USA, R_g is estimated using various models. For crop model application, empirical R_g models that use commonly measured meteorological variables, namely temperature and precipitation, are generally preferred. Although a large number of models of this kind exist, few have been evaluated for conditions in the USA. This study was conducted to evaluate the performances of 16 R_g models that are empirical and based on temperature and or precipitation for the southeastern USA, a major agricultural area in the country. Taking into account spatial distribution and data availability, thirty locations in the region were selected and their daily weather data spanning eight years obtained. Half of the data was used for calibrating the models and the other half for evaluation purpose. For each model, location-specific parameter values were estimated through regressions. Using the root mean squared error of prediction and the modeling efficiency as goodness-of-fit measures, the models were evaluated for each location based on the measured and estimated values of R_g. In general, the modeling efficiencies of both temperature and precipitation-based models were higher than those of only-temperature- or only-precipitation-based models. Among the models that use temperature or precipitation as the only input variable, the Mavromatis (2008) model had the best performance. The piecewise linear regression-based Wu et al. (2007) model (PLR-WU) performed best not only among the models that use both temperature and precipitation as inputs but also among all the sixteen models evaluated. The modeling efficiency of PLR-WU model was about 5 to more than 100% higher than those of the other models, depending on models and locations. The best performance of PLR-WU was mainly due to the use of both temperature and precipitation as inputs, temperature difference rather than temperature range, rain-day information instead of rainfall, and separate relationships for low and high radiation levels.