How Well Do Global Gridded Crop Models (GGCMs) Replicate the Responsiveness of Historical US Yields to Weather?
Understanding how yields of major calorie crops respond to weather shocks helps clarify the potential benefits and/or risks of future climate change. Empirical climate change economics studies derive yield responses to temperature and precipitation as simplified, reduced-form relationships using statistical models estimated on observed agricultural production, harvested area and weather. In crop simulations, yield changes are emergent behavior determined by complex interactions among multiple plant growth processes. Researchers at Boston University and Fondazione Eni Enrico Mattei (Italy), working within a multi-institutional Cooperative Agreement led by Stanford University, estimated econometric models of crop yields using observations and crop simulations for identical historical periods and US locations. The researchers compared the resulting reduced-form responses, thereby demonstrating that GGCMs’ yields tended to be more sensitive to adverse weather (extreme high temperature and/or low precipitation) exposures.
Different GGCMs simulate different crop yield changes when forced with the same pattern of historical change in climate. To have confidence in GGCMs’ projections of the future impacts of climate change, and to drive model development, it is important to assess their skill against observations, and understand why differences might arise. This paper develops a method for diagnosing differences in GGCMs’ responses, and for attributing these differences to model characteristics.
Global gridded crop models (GGCMs) are the workhorse of assessments of the agricultural impacts of climate change. Yet the changes in crop yields projected by different models in response to the same meteorological forcing can differ substantially. Through an inter-method comparison, the researchers provide a first glimpse into the origins and implications of this divergence—both among GGCMs and between GGCMs and historical observations. They examine yields of rainfed maize, wheat, and soybeans simulated by six GGCMs as part of the Inter-Sectoral Impact Model Intercomparison Project-Fast Track (ISIMIP-FT) exercise, comparing 1981–2004 hindcast yields over the coterminous United States (US) against US Department of Agriculture (USDA) time series for about 1000 counties. Leveraging the empirical climate change impacts literature, the research team estimate reduced-form econometric models of crop yield responses to temperature and precipitation exposures for both GGCMs and observations. Results demonstrate that up to 60% of the variance in both simulated and observed yields is attributable to weather variation. A majority of the GGCMs have difficulty reproducing the observed distribution of percentage yield anomalies, and exhibit aggregate responses that show yields to be more weather-sensitive than in the observational record over the predominant range of temperature and precipitation conditions. This disparity is largely attributable to heterogeneity in GGCMs' responses, as opposed to uncertainty in historical weather forcings, and is responsible for widely divergent impacts of climate on future crop yields.