Project Title: AI-Driven Optimization of Water Management in Paddy Field Irrigation   
Student: Atul Sunil
Course: MS in Computer Science (Artificial Intelligence)  

Abstract: 
Effective water management in rice paddy irrigation is crucial for agricultural sustainability, 
especially under the pressures of climate variability and growing water scarcity. Traditional 
irrigation methods (e.g. fixed schedules or continuous flooding) often use water inefficiently 
and lack adaptability to changing conditions. This project explores the use of artificial 
intelligence, specifically reinforcement learning (RL), integrated with agent-based modeling (ABM) 
to improve irrigation efficiency in paddy field systems. An agent-based simulation is developed 
to capture the dynamic interactions in a paddy irrigation network – including environmental 
factors (rainfall variability, river flow) and social factors (multiple farmers’ water usage 
decisions) – making the water management problem a complex adaptive system. A reinforcement 
learning agent is trained to make sequential irrigation decisions within this simulation, with 
the goal of optimizing water use (minimizing waste) while maintaining crop yields. The central 
research question is: Can an agent-based model augmented with reinforcement learning achieve more 
efficient paddy field irrigation under dynamic environmental and social conditions? To answer this, 
the RL-driven strategy is compared against conventional irrigation strategies in various scenarios. 
Preliminary results from simulation experiments indicate that the RL-based approach can significantly 
reduce water usage without compromising yield, by learning to anticipate rainfall and adjust allocation 
among fields. This aligns with recent studies where RL agents saved 10–30 of irrigation water in crop 
systems with no yield loss. Our findings demonstrate that a combined RL and ABM approach can 
adaptively manage paddy irrigation, improving water use efficiency under uncertainty. The dissertation 
discusses the design of the RL-ABM framework, experimental results, and the implications for real-world 
irrigation management. It concludes with reflections on limitations and future opportunities for deploying 
AI-driven irrigation in farming communities.