Agriculture is the main sectoral user of water globally. Increasing pressures on freshwater resources, coupled with population growth and climate change, mean there is a need to improve agricultural water productivity to tackle water scarcity and food insecurity. Improving irrigation scheduling practices is a key solution to these challenges, in particular in high-productivity agricultural regions where potential gains from improving irrigation application technologies have largely been exhausted. This thesis consists of four journal articles (two published, one in review and one work in progress) that together form a body of research focused on identifying and evaluating solutions for improving agricultural water productivity through improved irrigation scheduling practices. Work presented in the thesis focuses primarily on case studies in the central United States, where agriculture is the main sectoral user of water, and where there are growing issues of water scarcity related to intensive abstractions for irrigation. Analyses with the thesis leverage novel crop-water modelling tools, including AquaCrop-OSPy that was developed directly as part of this thesis (Chapter 2) The first empirical paper of the thesis (Chapter 3) examines how uncertainty in the knowledge of soil texture and local weather conditions impacts the water productivity of irrigation decisions made by farmers. This analysis demonstrates that a farmerâs choice of irrigation management rules has a much greater impact on productivity and profitability of water use than uncertainty in soil-moisture monitoring. A key conclusion from this paper is that perfect soil-moisture information is not required to make near-optimal irrigation decisions, and that greater benefits can be obtained by improving baseline irrigation management rules and heuristics. Building off of this finding, the second empirical paper of the thesis (Chapter 4) then examines how scheduling rules can be best adapted to improve the productivity and profitability of irrigation decisions under weather uncertainty. By comparing optimised irrigation heuristics that are applied in every year, to heuristics that are re-optimised during the season, an assessment of the added value of the adaptive in-season irrigation scheduling approaches was conducted. Results demonstrate that robustly optimised fixed management rules can achieve the vast majority of the profits obtainable with perfect foresight, and that the value of within-season adaptation is generally marginal unless perfect weekly weather forecasts are available to support in-season scheduling adaptation. Finally, the last empirical paper of the thesis (Chapter 5) explores the potential for AI approaches such as Deep Reinforcement Learning to enhance value of adaptive in-season irrigation scheduling in complex decision-making environments. An assessment is made of the added value of using Deep Reinforcement Learning for irrigation scheduling compared to optimised soil-moisture thresholds. These results show that only under restrictive water caps and minimised weather variability does Deep Reinforcement Learning increase profits compared to optimised soil-moisture thresholds.
Assessing the value of improved information and management strategies for optimal irrigation scheduling
Kelly, T. (Author). 1 Aug 2023
Student thesis: Phd