Future Directions

This chapter outlines future research directions for improving the methods developed in this book. It first discusses extensions to the soil-moisture and hydraulic-parameter estimation frameworks, including validation under extreme moisture conditions as well as comparisons with approaches that employ separate parameter sets to address soil hysteresis. It then highlights opportunities to enhance learning-based multi-agent model predictive control schedulers, including offset-free model predictive control to handle model mismatch and the use of transformer-based time-series models as alternatives to long short-term memory networks. The chapter also considers how semi-centralized multi-agent reinforcement learning approaches can be adapted to new agricultural settings using transfer learning, and how incorporating weather forecasts into the agents’ observation space may improve scheduling performance. Finally, it discusses extensions to the hierarchical scheduling–control framework, including the explicit incorporation of seasonal water allocations and higher-resolution variable-rate irrigation.