Improving large-area mechanistic yield simulations through disentangling genotype × environment × management factors

• Applicant: Professor Claas Nendel, Ph.D.
• Subject Area: Plant Cultivation, Plant Nutrition, Agricultural Technology
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 512314528

A reliable prediction of agricultural yields is of great interest for farmers, authorities and decision-makers across a large body of thematic institutions. Mechanistic agroecosystem simulation models (AEM) have been developed to describe the complex system of crop–soil–atmosphere interaction, also including the formation of crop yields. In theory, such models should be able to simulate the yield of each and every crop in each and every corner of the world, since the underlying physics applies everywhere on this planet. However, model intercomparison studies have produced evidence that none of the many AEM is living up to this expectation. Two main reasons are held responsible for this observation: (i) the description of the processes responsible for the behaviour of the system in the models is incomplete or erroneous, and (ii) the information that the models use to arrive at a prediction is insufficient or insufficiently accurate. Applications of AEM in a gridded design for yield predictions over large continuous areas increase in number, but proper validation is currently missing. When calibrated for best performance in reproducing these observed yield patterns, it remains unclear if the underlying processes and their representative state variables are similarly well simulated. The fact that for the vast majority of agricultural fields across Germany no information on variety, fertiliser schemes and fine-scale soil information is available, limits the potential of using AEM for yield predictions. This project proposes using AEMs to inversely estimate some of the input parameters and driving variables by using observed crop yields at a fine scale and additional information that constrains the inverse parameter estimation. Utilising a multi-annual and multi-local data set of on-farm winter wheat yields for 100 tiles of 10 × 10 m² size, which is accompanied by multi-sensor satellite imagery at 10m scale, this project proposes to (i) explore current limitations of AEMs for accurate reproduction of observed winter wheat yields across Germany, to (ii) improve the internal validity of three AEMs in reproducing these yields and to (iii) identify priorities for further research towards the improvement of future yield predictions using AEM. The overarching hypothesis is that the accuracy of current mechanistic agro-ecosystem models in blindly simulating winter wheat yields across the soil-climate space of Germany is mainly limited by insufficient information on genotype × environment × management, not by the model structure.

• DFG Programme: Research Grants