Introduction

This chapter begins by motivating the need for more precise and resource-efficient irrigation scheduling in the face of growing freshwater scarcity. It highlights the inefficiencies associated with conventional open-loop irrigation practices and advocates for closed-loop approaches. It argues that a tighter integration of optimal control and machine learning can improve scheduling precision: optimal control enables irrigation decisions that balance crop water requirements with efficient resource use, while machine learning leverages field data to support adaptive decision-making. The chapter reviews existing optimal control applications (with emphasis on model predictive control [MPC]) and machine learning applications in irrigation scheduling and identifies key gaps that, when addressed, can further improve their impact on irrigation practice. It then presents the main objectives of the book: (i) estimation of soil moisture and soil hydraulic parameters using remotely sensed measurements with identifiability analysis and parameter selection; (ii) performance-triggered model reduction to enable computationally efficient soil moisture estimation in large-scale fields; (iii) mixed-integer MPC formulations with zone control for homogeneous and spatially heterogeneous fields; (iv) a unified scheduling framework that couples mixed-integer MPC with management zones delineated using k-means clustering and estimated hydraulic parameters, long short-term memory-based soil-moisture surrogate modeling, and decentralized reinforcement learning agents to improve computational efficiency; (v) a semi-centralized multi-agent reinforcement learning (SCMARL) framework for irrigation scheduling in large-scale fields with spatial variability, where state augmentation is used to address non-stationarity; and (vi) a hierarchical scheduling–control framework in which SCMARL, trained under a partially observable Markov decision process setting, leverages daily weather information to provide daily irrigation schedules, while an MPC control layer employs hourly weather information to track these schedules.