Long-term comparisons of net radiation calculation schemes

Six commonly used models for calculating daily net radiation were tested against measured net radiation. Meteorological data from 32 and 7 consecutive years obtained at two temperate sites were used. The extensive duration of the datasets ensured that all weather conditions and extreme events were captured. A set of statistical procedures was used to evaluate the performance of the models. The mean bias errors ranged from 0.0 W m(-2) stop to 24.8 W m(-2) stop and 0.1-24.7 W m(-2) stop and root mean square error from 11.0 W m(-2) stop to 28.1 W m(-2) stop and 10.0-27.9 W m(-2) stop at the two sites respectively, for days without snow cover on the ground. The best agreement was found when locally calibrated model coefficients were used. Only negligible differences in model performances were found between the two sites and the differences were lower than the inaccuracies of the net radiation instruments used. Including days with snow cover in the analysis lead to a slight increase in the bias and scatter of the predictions. Model performances were in general better during summertime than wintertime. Altered albedo values during winter caused by generally low sun angles were likely the cause of this. Analysis showed that at least 5 years of data were needed to obtain stable calibration coefficients for local calibration of the models. Based on the results from this study, and due to their physical background, two physical based models were recommended for calculating daily values of net radiation under temperate climate regimes. A simple adjustment of the calibration coefficients based on climate regime was suggested for these models.