By David Lafferty and Ryan Sriver, University of Illinois
Understanding and managing climate risk in the agricultural sector is an important global problem. Climate models such as those used in the Coupled Model Intercomparison Project (CMIP) are key tools in this regard as they provide critical information about how climatological conditions may change in the future. However, they often face difficulties resolving fine scale processes due to coarse resolutions and systematic historical biases. For these reasons, downscaling and bias-correction are widely used post-processing techniques that aim to make model outputs more actionable for decision-makers and stakeholders.
Downscaling aims to interpolate gridded climate data to a higher resolution, and bias-correction aims to adjust for systematic biases. However, they can both introduce new uncertainties that propagate into modeled socioeconomic outcomes. For example, these post-processed climate fields may be physically implausible if methods are applied without expert knowledge, and the resulting representation of atmospheric variability is also a major challenge (Maraun et al., 2017). Many studies analyze the effects of these and other uncertainties, but they typically focus on hydrologic or meteorological variables; few extend their analysis into sectoral outcomes.
In a recent paper published in Communications Earth & Environment, we look at how uncertainties associated with downscaling and bias-correction propagate into simulated agricultural yields, focusing on U.S. maize (Lafferty et al., 2021). Specifically, we use an ensemble of statistically bias-corrected and downscaled climate models (NEX-GDDP), as well as the corresponding parent models (CMIP5), to drive a statistical panel model of U.S. maize yields. We then compare how simulated yields differ between the two ensembles in hindcasts and projections.
Our results show that there are large differences in the yield outcomes simulated by each of the CMIP5 models, but that most overestimate the variability of historical yields. Conversely, although the bias-corrected and downscaled models more closely match observations, they overestimate average yields and tend to underestimate yield variability as well as the severity of the largest historically observed yield drops (Figure 1).
We are able to trace these differences to the underlying representation of extreme degree days, which is a variable in the yield model that negatively affects yields. Extreme degree days is an integrated measure of the amount of heat a crop is exposed to over the growing season. Here, they are defined for temperatures above 29°C and thus correspond to the upper tail of the temperature distribution. Our work shows that both sets of models struggle to adequately capture the behavior of these tails, which ultimately affects how they simulate crop yield outcomes in the past as well as in the future. We find large differences in projections of decision-relevant metrics such as the magnitude of weather-induced yield shocks at various return periods, as shown in Figure 2. The spread among modelsleaves end-users with choices that require navigating potential trade-offs in resolution, historical accuracy, and projection confidence.
This paper raises important questions about the use of bias-corrected and downscaled climate information, particularly when combined with sectoral models that are sensitive to climatological extremes. Although we focus only on U.S. maize, other major crops such as soy, cotton, and wheat are often modeled using very similar econometric approaches (Schlenker & Roberts, 2009). Without careful consideration of the representation of temperature extremes and how they propagate through subsequent models, the use of bias-corrected and downscaled climate projections may lead to underestimates of the severity of impacts and consequently, poor decisions.
Lafferty, D. C., Sriver, R. L., Haqiqi, I., Hertel, T. W., Keller, K., & Nicholas, R. E. (2021). Statistically bias-corrected and downscaled climate models underestimate the adverse effects of extreme heat on U.S. maize yields. Communications Earth & Environment, 2(1), 196. https://doi.org/10.1038/s43247-021-00266-9
Maraun, D., Shepherd, T. G., Widmann, M., Zappa, G., Walton, D., Gutiérrez, J. M., et al. (2017). Towards process-informed bias correction of climate change simulations. Nature Climate Change, 7(11), 764–773. https://doi.org/10.1038/nclimate3418
Schlenker, W., & Roberts, M. J. (2009). Nonlinear temperature effects indicate severe damages to U.S. crop yields under climate change. Proc Natl Acad Sci USA, 106(37), 15594. https://doi.org/10.1073/pnas.0906865106