Functional diversity among crop models reveals contrasting resilience of soybean and maize to heatwaves and CO₂ enrichment

Background: Crop models are widely used to assess climate-change impacts on agricultural production, yet they often produce divergent projections under warming, elevated CO₂, and heatwave conditions. These differences are thought to arise from how models represent key physiological processes, but this link is rarely examined explicitly. Objectives: We aimed to evaluate whether divergence among crop models can be attributed to identifiable structural representations of physiological processes, and to assess how these differences influence simulated resilience of soybean and maize under warming, elevated CO₂, and short-term heatwaves. Methods: We compared three process-based crop models—WOFOST, BioCro, and DSSAT (CROPGRO-Soybean and CERES-Maize)—for irrigated systems in Nebraska, USA. Models were calibrated and evaluated using AmeriFlux and USDA data, and then subjected to controlled perturbations including uniform warming (+1.5 to +4.0 °C), atmospheric CO₂ increases (400–800 ppm), and phenology-specific heatwaves imposed during vegetative, reproductive, and maturity stages. Results: All models reproduced mean historical yields but showed strongly divergent responses to climate stressors. Under +4.0 °C warming, maize yields declined by 25–46%, compared with 8–25% for soybean. CO₂ fertilization partially offset soybean losses, yielding increases of +18% to +44%, with no effect on maize. WOFOST showed the strongest CO₂ response, DSSAT the highest sensitivity to heat due to explicit reproductive stress functions, and BioCro intermediate responses driven by biochemical canopy photosynthesis. Short heatwaves reduced yields by 1–5%, with the strongest effects during reproductive stages. Conclusion: Model divergence followed consistent and interpretable patterns linked to structural representations of photosynthesis, phenology, and heat-stress responses. These differences reflect underlying physiological formulations rather than random uncertainty. Implications: Understanding the structural drivers of model divergence improves mechanistic transparency and helps interpret uncertainty in crop–climate projections, supporting more reliable assessments of climate impacts and adaptation strategies.