Crop switching reduces agricultural losses from climate change in the United States by half under RCP 8.5

Extreme temperatures under climate change are projected to reduce average yields of major U.S. crops, though impacts vary spatially with some areas potentially benefiting from warmer conditions and irrigation. As climate shifts, relative productivity changes will drive farmers to alter crop choices and locations. Quantifying how regional productivity changes influence future cropping decisions and the extent to which crop shifting can mitigate damages is a central question for climate impacts on agriculture. Traditional econometric yield models face limitations for this task: they remove yield levels with region fixed effects (hindering prediction in new areas) and suffer a resolution-variance trade-off when allowing regional interactions. This study develops a hierarchical Bayesian yield model that estimates region-specific climate sensitivities while borrowing strength across space to predict yields in unobserved regions. Using this model for six U.S. crops, the authors evaluate potential economic outcomes of profit-maximizing crop reallocation under RCP 8.5 by 2050 and 2070, assessing how much crop switching can offset climate-induced losses.

The study builds on a large literature linking weather and climate to crop yields, including nonlinear temperature effects and meta-analyses of climate impacts and adaptation. Prior econometric approaches commonly use panel models with fixed effects to capture unobserved heterogeneity, but these remove yield level information critical for predicting suitability in new regions. Interacted coefficient models increase flexibility but risk overfitting and imprecise estimates. The paper also draws on research in spatial land-use optimization and comparative advantage in agriculture, as well as hierarchical modeling literature that supports partial pooling to capture spatially varying relationships. The authors position their approach as an empirical framework that combines spatially varying yield response estimation with optimization to assess adaptation via crop relocation.

Data: County-level yields (1949–2009) for six crops (barley, corn, cotton, rice, soybeans, wheat) across the contiguous U.S., matched with weather variables (growing degree-days, extreme degree-days, water deficit indices) and county-level covariates (annual mean temperature, isothermality, temperature seasonality, annual precipitation, precipitation seasonality, irrigation fraction by crop). Climate projections for 2050 and 2070 are from downscaled, bias-corrected CMIP5 models under RCP 8.5. Bayesian hierarchical yield model: Log yields are modeled as a linear function of crop water deficit, growing degree-days, extreme degree-days, and a linear time trend. Coefficients and intercepts vary by county. Their expected values are modeled as linear functions of the six county-level covariates, enabling partial pooling across space and prediction in counties without historical cultivation. The model jointly estimates county-specific weather response parameters and the meta-model linking those parameters to covariates. Two Bayesian variants are considered: one with county-specific error variances and one with uniform error variance across counties. Model validation and comparison: The Bayesian model is compared against a suite of OLS panel models varying in intercept and coefficient structures (uniform, county fixed effects, and interacted with covariates). Fit is evaluated using R^2 on all years and via temporal cross-validation: estimation on 1949–1994, evaluation on 1995–2009. Land-use optimization: A linear programming model maximizes total profits (local price times predicted yield minus management costs) by choosing crop areas per county and period. Constraints: (1) total crop area per county cannot exceed the county’s 2010 area under any of the six crops; (2) total national area allocated to each crop cannot exceed its 2010 national total (limiting large price feedbacks). Yields for future allocations are adjusted for the irrigation capacity of destination counties. Optimization is run for each posterior draw of yields to propagate uncertainty. Reported 2010 costs and prices are used; hidden costs are inferred by adjusting costs in counties where observed crops would otherwise not be optimal. Future climates are evaluated for 2050 and 2070 using the ensemble of downscaled CMIP5 models under RCP 8.5. Sensitivity analyses include imposing crop switching costs to assess how frictions affect reallocation and profits.

- Spatial variation in climate sensitivity: The effect of extreme degree-days (EDD) on yields varies substantially across counties and crops. Corn and cotton are less sensitive to extreme heat in the southern U.S., consistent with adaptation in varieties and practices. Wheat and barley show higher sensitivity in dry regions. The 95% range of estimated county EDD coefficients spans 2 (rice) to 12 (cotton) times the standard error of the average coefficient. Mean temperature explains a significant share of variation in EDD sensitivity across counties (8% for soybeans up to 63% for cotton). Spatial correlations in coefficients extend up to ~2000 km. - Model performance: On all years, Bayesian models perform similarly to flexible OLS models but avoid overfitting. Under cross-validation (train 1949–1994; test 1995–2009), flexible OLS models with interacted coefficients show severe overfitting (e.g., OLS with interacted intercepts/coefficients yields negative R^2 for corn −1.05, cotton −37.50, rice −1.59, soy −16.27). Bayesian models maintain better predictive power and outperform OLS for three crops under cross-validation (e.g., barley CV R^2: Bayes ~0.48 vs OLS interacted ~0.45; wheat CV R^2: Bayes ~0.50 vs OLS uniform ~0.49). - Current-period optimization: Even under current climate, profit-maximizing reallocation (subject to constraints) changes crops in 16% of counties (5% excluding corn–soy swaps) and increases total profits by 13% on average. - Future crop patterns (RCP 8.5): Corn retains large area but is less concentrated in the Midwest. Soybeans shift north, replacing spring wheat and barley. Great Plains wheat areas hollow out, while winter wheat expands north along the Mississippi. Cotton expands to higher latitudes, becoming dominant in southern California. Land in the southern U.S. unprofitable for any of the six crops grows to about 5% of included land by 2070. - Economic outcomes without reallocation: Total profits for the six crops fall from about $45.7B (current) to $35.8B in 2050 and $31.4B in 2070 (−31% by 2070). - Economic outcomes with reallocation: Current profits could rise to ~$51.8B under optimal allocation. By 2070, even with optimization, profits fall to ~$38.6B (≈16% below current observed profits), implying crop switching halves the climate-induced loss (from ~31% to ~16%). Relative to the current-period optimal allocation, losses by 2070 are ~26%. - Extent of switching: By 2070, 53% of counties experience crop switching under optimization (36% excluding corn–soy swaps). - Production changes by 2070 (relative to observed): Decreases in barley (−9%), corn (−37%), rice (−2%), soybeans (−6%); increases in cotton (+73%) and wheat (+2%). - Switching costs: Introducing switching costs (e.g., $180/acre) reduces reallocation by about half; as costs rise, outcomes converge to no-reallocation losses. Because optimal profits in 2050 are below current, losses persist under any switching cost level.

The study addresses how much crop switching can mitigate climate change impacts on U.S. agriculture by jointly estimating spatially varying yield responses and optimizing land use. Findings indicate substantial potential for adaptation through reallocation, halving aggregate profit losses by 2070 under RCP 8.5. However, significant residual losses remain, and the geographic pattern of adaptation implies disruptions: many counties switch crops, some regions experience persistent declines, and about 5% of currently cultivated land becomes uneconomical for any of the six crops by 2070. The hierarchical approach reveals existing adaptation (e.g., reduced heat sensitivity in warmer regions) and provides a framework to identify where crop transitions are most beneficial. Nonetheless, crop switching alone cannot fully offset damages; additional measures such as new seed varieties, irrigation and water management, and broader adaptation investments are needed. Policymakers should anticipate regional economic and social impacts, potential supply chain disruptions, and environmental consequences of changing crop distributions.

This paper introduces a hierarchical Bayesian yield modeling framework that captures spatially varying climate sensitivities and supports prediction in regions without historical cultivation, enabling robust evaluation of crop switching as an adaptation strategy. Coupled with a profit-maximizing land-use optimization constrained by current land and crop areas, the analysis shows that reallocation of six major U.S. crops can halve projected profit losses by 2070 under RCP 8.5, yet cannot eliminate them. The framework highlights where and how crop transitions could reduce damages, while revealing limits to adaptation via switching alone. Future research should incorporate dynamic technology change, CO2 fertilization effects, irrigation expansion feasibility, risk and multi-year shock dynamics, general equilibrium price feedbacks, farmer behavior and switching frictions, and broader land-use options beyond the six crops and current cultivated areas.

- Yield model captures adaptation reflected in historical within-year responses; it does not explicitly model future technological change, new varieties, or irrigation expansion. - Optimization assumes perfect knowledge of crop–weather responses and that realized weather aligns with expected climate. - General equilibrium price effects are not modeled; national crop area caps are used to limit price changes, but market feedbacks may alter profitability. - Risk aversion, unexpected shocks, and multi-year consequences of failures are not included. - Land-use constraints restrict cultivation to counties already growing at least one of the six crops in 2010 and limit each crop’s national area to 2010 totals; alternative land uses are not considered. - Field- and farmer-level constraints and switching frictions are simplified; hidden costs are inferred by adjusting costs to reconcile observed choices. - CO2 fertilization benefits and longer-term trends are not isolated in the main results; technology trends are not extrapolated to the future. - Some southern areas become unsuitable for any of the six crops, implying unmodeled transitions to other land uses or crops beyond the scope.