Enhancing Flood Resilience through Multi-Agent Systems and Operational Research Models: Evidence from Kerala Flooding

Flooding remains one of the most disruptive natural hazards for communities, infrastructure, and public services, especially where emergency response must be organised under uncertainty and resource constraints. Using Kerala as a district-level case study, this paper examines how Multi-Agent Systems (MAS), Deep Q-Networks (DQN), Multi-Agent Reinforcement Learning (MARL), Linear Programming (LP), Network Flow optimisation, and linear regression can be used within a comparative decision-support framework for flood response planning. In this study, flood resilience is interpreted in an operational sense, namely the ability of a response system to prioritise needs and allocate limited resources in a timely and structured way during a flood event. The analysis uses historical flood and geospatial information associated with the 2018 Kerala flood and is implemented as a simulation-based retrospective study rather than a live operational deployment. The results show that reinforcement-learning-based models are useful for exploring adaptive prioritisation behaviour under changing conditions, while optimisation-based models provide transparent baseline allocations under explicit constraints. Linear regression is retained as a supporting analytical component for examining relationships between flood-related variables rather than as a central optimisation tool. The contribution of the paper lies in offering a structured comparative case study that clarifies the different roles, strengths, and limitations of learning-based and optimisation-based methods for district-level flood response support.