Modeling and Simulating Adaptation Strategies Against Sea-Level Rise Using Multiagent Deep Reinforcement Learning

Abstract

Sea-level rise (SLR) problem, which is a major outcome of climate change, has been well documented and studied. Although it is globally observed due to climate change, local projections are needed to plan SLR adaptation strategies accurately. Since SLR is a community-wide multistakeholder problem at the local level, adaptation strategies can be more successful if the main stakeholders, e.g., government, residents, and businesses, collaborate in shaping them. Simulating the local socioeconomic system around SLR, including the interactions between essential stakeholders and nature, can be an effective way of evaluating different adaptation strategies and planning the best strategy for the local community. This work presents how such an SLR socioeconomic system can be modeled as a Markov decision process (MDP) and simulated using multiagent reinforcement learning (RL). The proposed multiagent RL framework serves two purposes. It provides a general scenario planning tool to investigate the cost–benefit analysis of natural events (e.g., flooding and hurricane) and agents’ investments (e.g., infrastructure improvement). It also shows how much the total cost due to SLR can be reduced over time by optimizing the adaptation strategies. We demonstrate the proposed scenario planning tool using available economic data and sea-level projections for Pinellas County, Florida, in a case study.