Time-varying impact of climate on maize and wheat yields in France since 1900

Climate services that can anticipate crop yields can potentially increase the resilience of food security in the face of climate change. These services are based on our understanding of how crop yield anomalies are related to climate anomalies, yet the share of global crop yield variability explained directly by climate factors is largely variable between regions. In Europe, France has been a major crop producer since the beginning of the 20th Century. Process based and statistical approaches to model crop yields driven by observed climate have proven highly challenging in France. This is especially due to a high regional diversity in climate but also due to environmental and agro-management factors. An additional level of uncertainty is introduced if these models are driven by seasonal-to-decadal surface climate predictions due to their low performances before the harvesting months of both wheat and maize in western Europe. On the other hand, large scale circulation patterns can possibly be better predicted than the regional surface climate, which offers the opportunity to rely on large scale circulation patterns as an alternative to local climate variables. This method assumes a certain degree of stationarity in the relationships between large scale patterns, surface climate variables, and crop yield anomalies. However, such an assumption was never tested, because of the lack of suitable long-term data. This study uses a unique dataset of subnational crop yields in France covering more than a century. By calibrating and comparing statistical models linking large scale circulation patterns and observed surface climate variables to crop yield anomalies, we can demonstrate that the relationship between large scale patterns and surface variables relevant for crop yields is not stationary. Therefore, large scale circulation pattern based crop yield forecasting methods can be adopted for seasonal predictions provided that regression parameters are constantly updated. However, the statistical crop yield models based on large-scale circulation patterns are not suitable for decadal predictions or climate change impact assessments at even longer time-scales.