Abstract
The results of previous studies have suggested that estimated daily globalradiation (RG) values contain an error that could compromise the precision of subsequentcrop model applications. The following study presents a detailed site and spatial analysis ofthe RG error propagation in CERES and WOFOST crop growth models in Central Europeanclimate conditions. The research was conducted i) at the eight individual sites in Austria andthe Czech Republic where measured daily RG values were available as a reference, withseven methods for RG estimation being tested, and ii) for the agricultural areas of the CzechRepublic using daily data from 52 weather stations, with five RG estimation methods. In thelatter case the RG values estimated from the hours of sunshine using the ångström-Prescottformula were used as the standard method because of the lack of measured RG data. At thesite level we found that even the use of methods based on hours of sunshine, which showedthe lowest bias in RG estimates, led to a significant distortion of the key crop model outputs.When the ångström-Prescott method was used to estimate RG, for example, deviationsgreater than ±10 per cent in winter wheat and spring barley yields were noted in 5 to 6 percent of cases. The precision of the yield estimates and other crop model outputs was lowerwhen RG estimates based on the diurnal temperature range and cloud cover were used (mean bias error 2.0 to 4.1 per cent). The methods for estimating RG from the diurnal temperature range produced a wheat yield bias of more than 25 per cent in 12 to 16 per cent of the seasons. Such uncertainty in the crop model outputs makes the reliability of any seasonal yield forecasts or climate change impact assessments questionable if they are based on this type of data. The spatial assessment of the RG data uncertainty propagation over the winter wheat yields also revealed significant differences within the study area. We found that RG estimates based on diurnal temperature range or its combination with daily total precipitation produced a bias of to 30 per cent in the mean winter wheat grain yields in some regions compared with simulations in which RG values had been estimated using the ångström-Prescott formula. In contrast to the results at the individual sites, the methods based on the diurnal temperature range in combination with daily precipitation totals showed significantly poorer performance than the methods based on the diurnal temperature range only. This was due to the marked increase in the bias in RG estimates with altitude, longitude or latitude of given region. These findings in our view should act as an incentive for further research to develop more precise and generally applicable methods for estimating daily RG based more on the underlying physical principles and/or the remote sensing approach.
Highlights
Crop growth models, which have been developed since the 1960s, have been regarded as important tools of interdisciplinary research [1] and have since been used in a number of areas such as the assessment of agriculture potential of a given region [2], in the field of crop yield forecasting [3,4] or as a climate change impact assessment tool [5,6,7]
It is not surprising that the RG estimation methods based on sunshine duration (i.e. Eq (1)-(2) showed relatively good agreement when used for the crop-environment resource synthesis (CERES)-Barley crop model compared with the observed RG values (Table 4)
Whereas our study relied solely on the WOFOST model, Witt et al [42] used Crop Growth Monitoring Systems, where WOFOST outputs are coupled with the results of national statistics, which according to our opinion significantly reduces the magnitude of an error introduced by uncertainties in the crop model inputs
Summary
Crop growth models, which have been developed since the 1960s, have been regarded as important tools of interdisciplinary research [1] and have since been used in a number of areas such as the assessment of agriculture potential of a given region [2], in the field of crop yield forecasting [3,4] or as a climate change impact assessment tool [5,6,7]. Other types of mathematical models for crop growth and development (apart from the process-oriented methods) rely frequently on RG as one of the key independent variables [8,9] and their outputs might be clearly affected by any RG bias It has been noted many times [10,11,12] that continuous records of global solar radiation measurements are spatially scarce and that the ratio between the number of stations observing daily RG and those measuring temperature and precipitation is highly variable from less than 1:10 in Germany [13] or 1:20 in the Czech Republic [14] to 1:500 on the global level [15]. These methods include stochastic weather generators [2123], linear interpolation [24], use of higher order statistics [25], application of the neural network method [26,27], or the application of various empirical or semi-empirical relationships established between daily global radiation and other more frequently measured meteorological parameters [28,29,30]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
