Crop yield forecast combination of process-based and statistical crop models at the local level

with Andrej Ceglar (Joint Research Centre-European Commission). Work in progress.

Process-based crop models provide a clear physiological mechanism for linking weather to crop yield outcome, with many of the essential parameters in these models having been established through laboratory experiments but are typically heavily parametrised objects, show coarse resolution and fail to capture farmer behaviour which may lead lead to unknown errors in new, untested environmental circumstances. On the other hand, statistical models are data-driven approaches than can capture farmer management behaviour. Another advantage is the tremendous volume and breadth of available data. However, the estimated effect of climate and environmental regressors on yield are usually highly sensitive to the chosen regression method and often regression models showing similar performance may lead in some cases to different conclusions with respect to effect of temperature and precipitation. Hence, both tools represent imperfect but complementary ways of describing a variable, crop yield, which is also often imperfectly measured. There is then scope for combination of the two. This approach has been indeed adopted by the Agricultural Model Intercomparison and Improvement Project (Rosenzweig et al, 2013) and some of its gains have already been explored in some case study regions (Schauberger et al, 2017) and a simple combination of the two (Roberts et al, 2017).