Climate change impacts on crop yield : development and evaluation of fundamental models as a basis for economic assessment

Physical climate changes due to greenhouse gas emissions are well understood. However, quantifying the economic consequences remains a major challenge. Nevertheless, such quantification is crucial for the development of effective climate protection and adaptation strategies. Especially at local and regional levels, there is insufficient knowledge about the multiple impacts of climate change on economic sectors and regions. This is particularly true for the agricultural sector, which is considered to be vulnerable to the effects of global climate change. Since climate change not only changes temperature but also precipitation patterns in space and time, a higher variability of individual weather and the resulting extreme events (e.g. storms, flooding or droughts) is expected. Accurate models that depict the weather and crop yields are important not only for projecting the effects of agriculture, but also for projecting the impact of climate change on the associated economic and ecological consequences and thus for mitigation and adaptation policies. There are various methodological approaches to modelling climate impacts on agriculture. On the one hand, there are holistic approaches such as integrated assessment models. On the other hand, there are process-based or mechanistic models that capture the relevant biophysical relationships. Finally, there are empirical or statistical models that explain the relationship between meteorological variables and agricultural yields. These modelling approaches are rooted in very different disciplines and involve different emphases and assumptions, often resulting in a lack of consistency. Based on this scientific discussion, the thesis aims at the design of statistical approaches in order to allow a convergence of the results of the different methods. The aim is to identify missing aspects in current statistical approaches, such as the absence of important variables (e.g. soil moisture) and addressing the timing of the occurrence of extreme events that affect plant growth. In addition, new statistical approaches from the field of machine learning will be introduced to complement the existing methods, which are mainly based on econometrics. Furthermore, the approach presented here enables a Germany-wide impact assessment for the main crops. Finally, the development of such statistical damage functions promotes the management of the effects of extreme events on the agricultural sector on several time scales and can be used for climate change impact assessment. The work is cumulative and consists of three scientific articles.