(735ac) Real-Time Irrigation Scheduling in Precision Agriculture: Comparison of Simulation-Based Optimization Approaches

This work investigates the application of simulation-based optimization methods within the Decision Support System for Agrotechnology Transfer (DSSAT) framework, aiming to enhance sustainability of agriculture against the global climate change and water scarcity. DSSAT is a software application program that comprises crop simulation models for over 42 crops, which has gained wide acceptance and has been used to simulate various applications at different spatial and temporal scales (Jones et al. 2003). This work explores a computationally efficient direct search method in the form of pattern search and a computationally more expensive population-based method of multi-objective genetic algorithm (MOGA). The objective is to optimize real-time irrigation scheduling (i.e., irrigation time/date and amount) to maximize crop yield as well as water use efficiency (WUE) under variable weather conditions.

Pattern search is a direct search method used for optimization problems where the gradient is not available (Torczon.1997). In the context of DSSAT, pattern search is a good choice since the derivative information is hard to obtain. In addition, pattern search is simple and flexible for efficient exploration of high-dimensional spaces and handling non-smooth or even non-continuous DSSAT objectives. However, pattern search employs a local search strategy, which may have difficulty avoiding local optima. On the other hand, MOGA utilizes a population-based search strategy by maintaining a population of diverse candidate solutions and iteratively applies genetic operators (selection, crossover, mutation) to generate new solutions. The objective function is evaluated for each individual in the population, guiding the selection of solutions for the next generation. This population-based approach allows MOGA to explore a wider search space and potentially avoid local optima. Another advantage of MOGA is its explicit handling of multiple (conflicting) objectives (Deb et al., 2002).

In this study, we simulate the growth and water usage of maize (corn) in the Piracicaba region of Brazil, referencing soil properties, and management practices within the DSSAT system. We utilize 36 years of weather data, spanning from 1985 to 2020, sourced from NASA’s POWER (Prediction Of Worldwide Energy Resources) project. We compare the performances of pattern search and MOGA to the performance of the DSSAT built-in irrigation management strategy in terms of crop yield and WUE. We also investigate their flexibility in shifting weight between yield and WUE to address different conditions (e.g., water sufficient vs. water scarce regions). Finally, we compare the performances of the above-mentioned methods under different levels of weather forecast uncertainties.

Reference

• Jones, James W., et al. "The DSSAT cropping system model." European journal of agronomy3-4 (2003): 235-265.
• Torczon, Virginia. "On the convergence of pattern search algorithms." SIAM Journal on optimization1 (1997): 1-25.
• Deb, Kalyanmoy, et al. "A fast and elitist multiobjective genetic algorithm: NSGA-II." IEEE transactions on evolutionary computation2 (2002): 182-197.