PNNL

Abstract

Concerns about food security under climate change have motivated eorts to better understand the future changes in yields by using detailed process-based models in agronomic sciences. Results of these models are often used to inform subsequent model-based analyses in agricultural economics and integrated assessment. Fundamental feedbacks between economic decision making and crop performance (e.g. management for intensification) is often ignored, as better integration of the dierent model types is hampered by the computational requirements of detailed deterministic process-based crop models and optimization-based economic modeling frameworks. Conducting a large simulation set with perturbations in atmospheric CO2 concentrations, temperature, precipitation, and nitrogen inputs, Phase II of the Global Gridded Crop Model Intercomparison (GGCMI), an activity of the Agricultural Model Intercomparison and Improvement Project (AgMIP), constitutes an unprecedentedly data-rich basis of projected yield changes across 12 models and five crops (maize, soy, rice, spring wheat, and winter wheat) using global gridded simulations. Results from Phase II of the GGCMI eort, a targeted experiment aimed at understanding the interaction between multiple climate variables (as well as management) are presented first. Then the construction of an emulator or statistical representation of the simulated 30-year mean climatological output in each location is outlined. The emulated response surfaces capture the details of the process-based models in a lightweight, computationally tractable form that facilitates model comparison as well as application in subsequent modeling eorts such as integrated assessment. Results show considerable spatial heterogeneity in sensitivity to input variables, confirming the need for a global study, as well as dierences across models. It is clear however that models robustly produce yield responses that are nonlinear and sensitive to interactions across the input variables. Results illustrate the utility of model emulation for climate impacts studies.