Data-Driven Optimal Irrigation Control Based on DSSAT and Optimization Methodologies

Abstract

The escalating challenges posed by global climate change, coupled with growing concerns about agricultural water scarcity, necessitate innovative strategies for efficient and sustainable water use. Agriculture, a primary consumer of freshwater, accounted for 42\% of total freshwater withdrawals in the U.S. in 2015, highlighting the importance of optimized irrigation practices. Traditional rule-based irrigation management methods, which rely on empirical schedules, are increasingly inadequate in addressing the complexities of modern agriculture. This dissertation investigates data-driven optimal irrigation approaches by integrating a well-known dynamic crop growth simulator for over 40 crops, decision support system for agrotechnology transfer (DSSAT) with modern optimization technologies to foster agricultural sustainability.

The study explores two key optimization and control methodologies: simulation optimization (SO) and model predictive control (MPC). Decision support system for agrotechnology transfer(DSSAT), a cropping system software for over 42 crops, offers detailed simulations of crop growth, yield, and soil moisture dynamics under diverse environmental scenarios. As such, this work proposes to use DSSAT as a high fidelity digital twin (DT) for testing various irrigation optimization and control strategies to achieve enhanced irrigation efficiency with the goal of reducing irrigation water usage while maintaining or even increasing crop productivity.

In this dissertation, the SO approach plays a critical role in advancing precision irrigation through the use of heuristic and meta-heuristic techniques. Pattern search (PS) and the multi-objective genetic algorithm (MOGA) are two key methods employed to optimize irrigation schedules with a focus on maximizing crop yield and improving water use efficiency. PS, a direct search algorithm, is particularly suited for scenarios where gradient information is unavailable, allowing it to effectively handle the non-linear, non-differentiable nature of agricultural systems. MOGA, on the other hand, is a more complex evolutionary algorithm designed to balance competing objectives, such as yield and water use efficiency. By searching for solutions along the Pareto front, MOGA provides irrigation schedules that offer various trade-offs, giving farmers flexibility to choose a plan based on their priorities, whether that is minimizing water use or achieving maximum yield.

Each technique operates on a daily basis, where irrigation amount and scheduling decisions are evaluated and adjusted daily to respond to crop and its environmental changes dynamically. Specifically, the algorithms consider historical and forecast weather conditions, soil moisture, and crop growth to make decisions about irrigation timing and quantity. PS iteratively tests different irrigation schedules within a specified search space, adjusting decisions based on simulated crop responses. MOGA generates a population of potential irrigation plans and evolves them across generations using selection, crossover, and mutation, converging towards solutions that optimize both yield and irrigation efficiency. This iterative approach ensures that irrigation practices are responsive to the day-to-day variability in weather and soil conditions, aligning each irrigation event closely with crop requirements and its environment.

These techniques have demonstrated considerable effectiveness in achieving optimal irrigation schedules under varying environmental conditions. The long-term simulations enabled the examination of diverse weather patterns, soil properties, and seasonal variations, which are critical for assessing the robustness of PS and MOGA under realistic agricultural scenarios. Both techniques have shown their potential to improve crop yield by maintaining ideal soil moisture levels and preventing over-irrigation, thereby enhancing Irrigation Use Efficiency (IUE). These findings contribute valuable insights to precision irrigation management, suggesting that the combination can serve as a powerful tool for sustainable agriculture, helping to mitigate the impacts of climate variability on water resources.

MPC as an advanced and dynamic strategy for real-time irrigation management, enabling precise control over soil moisture levels throughout the crop growth cycle. Central to the MPC approach is the development of an estimated state-space model, grounded in the principles of the soil-water balance equation. This model captures the dynamic interactions between soil moisture, crop water demands, and environmental variables, allowing MPC to make accurate, data-driven decisions regarding irrigation scheduling. By accounting for these complex relationships, MPC facilitates an real-time data-driven irrigation strategy that aligns closely with crop requirements, optimizing water use and supporting sustainable agriculture.

The MPC framework operates on a receding horizon principle, where soil moisture predictions are continuously updated within a specified forecast period. By integrating the state-space model with DSSAT crop simulations, MPC anticipates crop water needs based on current soil conditions, recent irrigation activities, and future weather patterns. This proactive approach allows MPC to address potential water deficits or prevent over-irrigation before they impact crop health, thereby enhancing both crop yield and water use efficiency. The DSSAT simulations validate these predictions, enabling MPC to optimize soil moisture levels throughout the growing season in response to real-time environmental data.

To enhance predictive accuracy, the state-space model is refined using system identification techniques, particularly through the autoregressive with exogenous inputs (ARX) model. This approach incorporates essential inputs such as irrigation, precipitation, and evapotranspiration, to predict soil moisture. With these refinements, MPC can effectively adjust to short-term environmental changes as well as long-term seasonal trends, making it a robust tool for precision irrigation amidst climate variability. By integrating real-time data and predictive modeling, MPC promotes efficient water management, ensuring crops maintain ideal moisture levels—a critical step toward sustainable agriculture, especially in regions facing water scarcity.

In summary, this dissertation highlights the potential to address the pressing challenges of climate change and water scarcity. By harnessing predictive modeling and optimization, this research contributes substantially to the development of sustainable, efficient, and climate-resilient agricultural systems.