Comparison of estimates of daily solar radiation from air temperature range for application in crop simulations

Daily solar radiation is an input required by most crop growth, development and yield simulation models. It is, however, not observed at many locations, preventing the application of such models. The objective of this work was to (i) evaluate several existing models estimating solar radiation based on daily minimum and maximum air temperature and/or precipitation for seven sites in the world, and (ii) investigate the impact of the estimated solar radiation on grass reference evapotranspiration (ETo) and total plant dry biomass simulations for maize. Comparisons of the solar radiation models was made based on a single modular indicator, Irad, computed using a fuzzy expert system that aggregated several statistical indices, and distribution of mean daily errors over a year. The model estimations were also evaluated according to their ability to simulate ETo and total dry biomass that matches simulations from the observed solar radiation. According to the Irad indicator, there was no solar radiation model which consistently outperformed all the other models at all the sites tested, but Irad indicated models that relatively underperformed at all the sites. The graphical presentation of the mean fluctuation of errors over a year gave a good assessment of the solar radiation estimation models in investigating the temporal behaviour and magnitude of the residuals. Performance of the models according to Irad and simulations of grass ETo and total dry biomass agreed better for models that relatively underperformed. Ranking of the models according to the root mean square error (RMSE) in solar radiation estimation and the RMSE in the grass ETo simulations agreed very well. Comparison of the ranking of the models using the Irad (or the individual statistical indices thereof) and total biomass simulation was difficult because of the difference in the time scale used in calculation of the statistical indices. The difference in simulations of total dry biomass accumulated over the years, however, qualitatively agreed with the graphs of the mean fluctuation of errors over a year. In general, the Irad indicator demonstrated which solar radiation estimation models should be used for crop simulation modelling.