National wheat yield prediction of France as affected by the prediction level

The objectives of this study are to: (1) investigate the effects of the prediction level on the prediction accuracy of total national yield; (2) examine a prediction model, using a non-linear trend function, simulation results and a value for planted area; (3) compare two different methods to estimate global radiation and evaluate the prediction results established with these estimates. France is selected as study area. Total national yields are predicted at sub-regional, regional, and national level. Results at sub-regional and regional level are summed into predicted total national yields. The total yield prediction accuracy is examined. The prediction model to be tested assumes a dependency between magnitude of the trend function and variation in yield (a multiplicative model). To evaluate this model its prediction results are compared with the results of a model which assumes no relation between trend and variation in yield (an additive model). To justify the use of simulation results and to evaluate their effects on the prediction results, these results are also compared with those of three simple trend models. The two compared methods to estimate global radiation are: (a) the Ångström–Prescott equation, for missing data average values are used; (b) the same equation, however, for missing data values are estimated using a cloud cover and temperature range method (proposed method). This study suggests that better results may be obtained if predictions are executed at sub-regional or regional level and subsequently summed into a total national value. On these levels the multiplicative model performs better than the other models used. The proposed radiation routine improves the prediction accuracy slightly.