A New Probabilistic Crop Yield Emulator: Development and Applications

Climate change significantly threatens global food security, while advancements in negative emission technologies, such as Bioenergy with Carbon Capture and Storage (BECCS) from crop residues, offer potential for climate mitigation. Crop yields are influenced by climatic factors, including temperature, precipitation, and atmospheric CO₂, as well as human management practices such as irrigation and fertilization. Crop residues, as unavoidable byproducts of food production, provide a sustainable resource for bioenergy generation without requiring additional cropland. To synergistically achieve the Sustainable Development Goals (SDGs) of Zero Hunger and Climate Action, a comprehensive analysis of future food crop yields through numerical modelling and exploration of diverse climatic and socio-economic scenarios incorporating region-specific adaptation strategies is crucial.

A new crop emulator, blending information from state-of-the-art global gridded crop models (GGCMs) and observational data from field experiments, has been developed to facilitate probabilistic projections of crop yields under diverse climatic and socio-economic scenarios. It can be integrated into simple climate models, such as the compact Earth system model OSCAR, or used standalone. For policy relevance, it is constructed at a sub-national scale with the flexibility to be aggregated to broader regional levels while remaining computationally efficient for large scenario ensembles. Aligned with the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) framework, it simulates yields for four major food crops: maize, rice (two growing seasons), soybean, and wheat (spring and winter varieties) driven by atmospheric CO₂(C), growing season temperature (T), water availability (W), including precipitation and irrigation, and nitrogen fertilization (N). While crop yield responses to C, T, and W are calibrated using ISIMIP3b simulations conducted under fixed human forcing, responses to N are calibrated against long-term field experiments, addressing inter-model uncertainty and integrating diverse data sources. Applying observational constraints via Bayesian inference further improves the model’s accuracy.

This paper describes the calibration, integration, and validation of the crop emulator and illustrates its performance and potential through two example studies. The first examines historical crop yields under static human inputs, and the consistency of these results with ISIMIP3a outputs validates the emulator’s ability to emulate GGCMs. The second study uses dynamic human inputs and constraints derived from field experiments (e.g., open-top chamber and free-air CO₂ enrichment experiments), showing good agreement with FAO statistics and demonstrating the emulator’s capability to represent human management impacts. Beyond these examples, the crop emulator's potential extends to various future applications, such as coupling with integrated assessment models (IAMs), reanalysis of the Sixth Assessment Report (AR6) scenarios, and contributions to the upcoming AR7.