Identifying indicators of extreme wheat and maize yield loss in Europe

Regional and national yield predictions are generally based on a combination of expert knowledge, survey data, and model simulations. Predictions can influence crop prices and are used to estimate end-of-season stocks. In Europe, national yield predictions are made available to stakeholders several times during the growing season in the monthly MARS bulletin. Predictions are made with the MARS crop yield forecasting system, which relies on an in-depth analysis of past climate and crop statistics, short-term weather forecasts, and crop growth simulations. The skills and limitations of these products are important because they inform EU trade policies. The prediction of extreme yield losses is of particular interest because extreme events can significantly influence the quantity of tradable commodities. In this study, we evaluate the capacity of a large range of indicators to predict the occurrence of extreme yield loss events. Indicators of various complexity levels are considered: simple or combined climate variables, agro-climatic variables and outputs of complex dynamic crop models. Each indicator is used to predict the occurrence of extreme yield loss in France and Spain for both wheat and grain maize. The sensitivity and specificity of the indicators are evaluated by ROC (Receiver Operating Characteristic) analysis using de-trended regional yield time series. Indicators are ranked based on a score quantifying their ability to separate extreme from non-extreme yield loss events. No single indicator performs systematically well but several show acceptable scores. There is no obvious relationship between the level of complexity of a given indicator and its accuracy. Maximum temperatures and, later in the growing season, potential yield simulations rank highest for both crop species in France. Drought indices perform well in Spain for wheat and maize. The robustness of our ranking is evaluated by comparing the results obtained using several alternative methods. We argue that this transparent framework can be useful to evaluate and improve crop monitoring systems worldwide. In this perspective we provide a relation linking the most accurate indicator threshold values to the probability of extreme yield loss.JRC.D.5-Food Securit