An Intelligent Irrigation Decision Support System for Optimizing Cotton Water Use

Irrigation plays a crucial role in alleviating the negative effects of drought on crop production. However, increasing competition for water by other sectors, such as industry and domestic use, increases the pressure on available water supplies. Under these circumstances, agricultural producers must be able to manage their available supplies efficiently to optimize irrigation water use. The objective of this research is to develop a decision support system (DSS) for optimizing irrigation scheduling for cotton production using Deep Reinforcement learning (DRL). Our approach uses multiple DRL algorithms that enable an intelligent agent to learn cotton irrigation needs in an interactive environment by trial and error using feedback from its past actions and experiences. Aquacrop is used as an environment (cotton field) simulator and is coupled with a DRL model to simulate crop yield for different actions taken by the agent. Our proposed software estimates the daily irrigation needs of a 7-acre crop field irrigated by a center pivot system located at Clemson University's Edisto Research and Education Center (REC), near Blackville, South Carolina. This new system enables a closed-loop control scheme to adapt the DSS to local perturbations such as soil moisture and rainfall variabilities.