Crop model sensitivity to the estimated daily global solar radiation data

The results of the previous studies have suggested that the estimated RG values are loaded with an error, which might compromise the precision of the subsequent crop model applications. Therefore a detailed analysis of the error propagation was made using two crop models i.e. CERES-Barley and CERES-Wheat. Database of meteorological data originating from 8 stations in Austria and Czech Republic was used in order to carry out the analysis. It has been found that even application of the method based on sunshine duration that yield the lowest bias in RG estimates significantly influences number of key crop model outputs. It has been also noted that in 5–6 seasons out of 100 cases the deviation greater than ±10 % is to be expected whilst the occurrence of ±25% could not be also ruled out. The precision of the yield estimates and other crop model outputs is lower then expected but still acceptable for most application with mean bias error in range of 2.0–4.1% when estimates based on the diurnal temperature range and cloud cover are used. In this case yield deviations over ±10% occurred in about 20% cases (depending on the crop) whilst the probability of significant yield departure (±25%) doubled of that found for the previous method. The methods based on the diurnal temperature range and daily precipitation sum showed an increase of the systematic bias of yield of winter wheat and considerably higher number of seasons with yield departures over ±25%. Utilisation of the methods based on the diurnal temperature range only for the purposes of seasonal yield forecasting or climate change impact assessment is questionable as the probability of significant yield departure is very high (as well as the systematic error). These findings should act as an incentive to the further research aimed at development of more precise and widely applicable methods of estimating daily RG based more on the underlying physical principles and/or remote sensing approach. Overall decrease of the existing uncertainties in the RG estimates would clearly result into increase of the reliability of subsequent applications that use RG as input variable.