Heterogenous Patterns of Crop Yield Growth Stagnation across U.S. Counties in the Next Decade

The full research paper is available here: Full Paper (PDF).

This paper introduces a hierarchical Bayesian framework that models the interaction between climate variables and crop yield growth at the U.S. county level, enabling probabilistic short-term yield projections.

The main methodological innovation is the development of a parametrized Beta-type likelihood model for crop yields, where both location and spread parameters depend on weather variables—specifically, precipitation, growing degree days (GDD), and extreme degree days (EDD). Key features include:

• Flexible link functions to ensure parameter domain validity and interpretability;
• Hamiltonian Monte Carlo (H-MCMC) for posterior sampling, allowing the model to handle complex dependencies and high-dimensional parameter spaces;
• A hierarchical prior structure that allows parameter variation across counties, incorporating both temporal and spatial heterogeneity;
• Estimation of joint weather effects using a multivariate prior on weather coefficients, with hyperpriors enabling correlation among climate impacts;
• Integration of IPCC-based climate forecasts (SSP2-4.5) into the simulation process, yielding interpretable and robust probabilistic forecasts.

Compared to previous methods (e.g., Ricardian or standard Beta models), this framework uniquely combines distributional flexibility, unit-level heterogeneity, and full Bayesian inference, making it particularly suited for short-term, spatially disaggregated climate impact assessment. The methodology is demonstrated through 2032 projections for corn and soybean yields under different crop growth regimes.

For a visual overview, see the slides ers-aaea2023 (PDF), presented at the AAEA 2023 meeting in Washington, DC.