The MARS-Crop Yield Forecasting System (M-CYFS) is used since 1993 to forecast the yields of all major crops in the European Union (EU) based on gridded runs of the WOFOST crop model. Using 28 years of observation, from 1988 to 2015, we quantified the variability in crop yield reported by all 28 EU Member States (MS) that can be explained by each individual WOFOST crop model based predictors and a few simple meteorological variables. A linear regression is used as a screening tool to quantify the relationship between each predictor and the yield residuals from the trend throughout the crop cycle for 168 country/crop combinations, assuming the yield residuals from the trend depend on the inter-annual climate variability. The results are plotted and analyzed at different level: every 10 days for each country crop/combination and each predictor; synthetized every 10 days for each country/crop combination keeping the predictor showing the best relationship with the yield residuals; finally, the best predictor found for each MS during the entire growing season is used to evaluate the ability of the model to estimate yield variability of each crop at European scale.
While 61% of the grain maize (Zea mays L.) yield variability can be anticipated 80 days before harvest with the simulated water limited biomass for countries where rainfed maize prevails, 41% of the soft wheat (Triticum aestivum L.) yield variability can be reproduced a month before harvest, the best estimates being obtained where wheat is predominantly exposed to water stress. For the other crops analyzed, the results are also found to be reliable for crops predominantly exposed to water stress and becoming unreliable in agricultural systems exposed to an oceanic climate with a high level of inputs. The agro-meteorological processes related to an excess of water (nitrogen losses, diseases, anoxia, harvest conditions) would need to be disentangled and better integrated into the crop modeling system to improve the predictors.
The monthly cumulated meteorological predictors are performing only slightly worse than the crop model predictors and help to characterize the main processes responsible for the yield variability. Nevertheless, the predictive capacity of the meteorological predictors is spatially and temporally incoherent and differs according to the crop phenology. In comparison, the M-CYFS crop model predictors are more consistent since the predictors summarize the succession of agro-meteorological conditions determining the yield throughout the entire growing season.