Files
Abstract
Importance: Climate change may put our ability to feed future populations at risk. The prospects of those risks are limited by our understanding of how environmental factors impact crop yields. Present-day correlations of yields with individual factors may involve confounding variables whose relationships shift in future climates. Numerical models of crop yields are built on untested assumptions, and their projected impacts under climate change are not clearly attributed to underlying mechanisms. This dissertation aims to improve our understanding of climate impacts on agriculture through a series of studies that address (1) how yield responses are driven by factors other than temperature, (2) whether assumptions about crop maturities in models are consistent with observations, (3) how crop maturity rates factor into future yield responses, and (4) how changes to soil moisture in future climates affect not only crop yields but the ability of farmers to plant at all.Approach: This work extracts insight into maize crop production using creative combinations of agricultural models and observational data. Process-based models simulate daily crop growth based on numerical representations of physiological mechanisms, offering a direct link between weather inputs and crop yields. This work both seeks to understand what factors drive yield losses in models and uses those models as sources of synthetic data in “perfect model” experiments to test statistical methods of crop yield prediction. This work makes novel use of observational datasets to investigate whether modeled maturity parameters and responses reflect real-world crop behaviors. Lastly, this work addresses agricultural impacts disregarded in crop yield models, using a machine-learning approach to understand how soil and weather conditions prevent farmers from planting their intended maize crops, and whether prevented planting outcomes might become worse under climate change. This work focuses on the U.S. Corn Belt, the most productive maize region in the world, but results may generalize to other mid-latitude maize regions such as China.
Key Findings: We find that (1) commonly used statistical methods using temperature variables overproject yield losses under climate change by a factor of two because they omit the changing relationship of humidity and temperature under climate change. Methods using moisture variables produce consistent responses over time, suggesting moisture stress is more relevant than temperature stress in driving yield responses. Process-based models accurately reflect both (2) observed historical maize maturities and (3) the acceleration of maturity rates in warmer temperatures, but are inconsistent in how accelerated maturity impacts yields. Lastly, we find that (4) prevented maize planting is strongly influenced by winter and springtime soil moisture and drainage conditions, and is projected to become less frequent but more severe on average under climate change. Findings from this dissertation highlight the importance of agricultural models accurately reflecting the real-world drivers of crop production and suggest that U.S. maize production may fare better under climate change than previously suspected.