Abstract Rising production costs and increasing weather variability pose significant challenges to agricultural profitability, motivating computational approaches to optimize resource allocation. This paper presents a coupled ordinary differential equation (ODE) crop model that captures two key physiological phenomena often neglected in optimization studies: delayed nutrient absorption and cumulative stress tracking, which we address via finite impulse response (FIR) convolution and exponential moving average (EMA) filtering. We compare two optimization approaches for irrigation and fertilizer scheduling: genetic algorithm (GA) optimization for fixed seasonal strategies and model predictive control (MPC) for adaptive daily decision-making. Applied to corn production in Iowa across 21 stochastic weather scenarios spanning normal conditions to extreme drought and heat stress, our results yield a surprising conclusion. GA optimization achieves 35% higher mean revenue than farmer best practices across all 21 scenarios (842 vs. 626 per acre), and contrary to our initial hypothesis, the fixed GA strategies outperform adaptive MPC in all 21 scenarios. The GA trained under normal weather conditions achieves the highest mean revenue with the lowest risk (coefficient of variation 14. 7%), while MPC, despite its daily adaptation, cannot match the GA’s full-season optimization horizon. However, MPC offers a sustainability advantage: it achieves competitive returns while using dramatically less water and fertilizer than all other strategies. The practical conclusion is that for seasonal agricultural planning, pre-optimization for expected conditions outperforms adaptive control, and the choice of which GA strategy to deploy depends simply on whether extreme weather is anticipated.
Becker et al. (2026) studied this question.