Attributing observed climate change impacts in agriculture using observationally-derived counterfactual climate data and statistical crop-yield modelling

Statistically rigorous methods to attribute large-scale, long-term changes in observed climate to anthropogenic forcing are well established and the attribution of individual extreme events has advanced rapidly too. Further attributing climate impacts in society and ecosystems can be important for understanding and assessing loss and damage, for informing adaptation policies, and for motivating both adaptation and mitigation efforts. For this purpose, a counterfactual climate dataset was recently published within the Inter-Sectoral Impact Model Intercomparison Project (Mengel et al., 2021). Constructed by removing the long-term shifts in daily reanalysis data that are correlated to global-mean temperature change, the dataset does not address anthropogenic climate change, but its large spatial and temporal coverage and the range of variables covered as well as minimum requirements on computational tools and data make it a desirable resource for this interdisciplinary problem.

Here, we trial the use of that counterfactual dataset for the quantification of climate impacts on agricultural crop yields, which are of paramount importance to many of the regions most exposed and vulnerable to climate change, not least for food security. We present results from case studies that examine the impacts of selected drought events and are chosen based on reports of substantial food security impacts, on the availability of crop yield data, and on the existence of published scientific literature of a corresponding climate attribution study. The latter allows the comparison with results using methods that isolate the anthropogenic (combined, or by individual forcing) climate change signal. Together with more systematic discussion, this gives an idea of the degree to which our results on the contribution of any climate variations to the observed impacts may be a proxy for the anthropogenic climate change contribution specifically.

Impacts are explicitly simulated using statistical crop models that are established in the agricultural and agronomic literature, built and validated based on the observed record. The use of the single-realisation, weather-preserving factual and counterfactual dataset gives a deterministic rather than probabilistic estimate, but parametric and structural crop-model uncertainty is characterised, and the robustness of the results to different observational crop-yield datasets assessed. We collaborate with local stakeholders to ensure appropriate consideration of non-climatic factors and to improve data availability and quality. Our work combines perspectives of climate attribution, disaster risk reduction, and agricultural science to enhance attributing loss and damage in agriculture to climate change.