Scenario Planning for Sea Level Rise via Reinforcement Learning

Climate change and sea level rise impacts will affect coastal communities with multiple threats, including increased frequency of compound events, such as storm surge combined with heavy precipitation. Accurately modeling how the stakeholders, such as governments and residents, may respond to sea level rise scenarios (i.e., scenario planning) can assist in the creation of policies tailored to local impacts and resilience strategies. In this paper, our contributions are twofold. Firstly, considering a single-agent model for government, we numerically show that the government's policy on infrastructure improvement should be based on the observed sea levels rather than the observed cost from nature. The latter refers to the straightforward policy that any responsive (but not proactive) government would follow. Through a reinforcement learning algorithm based on a Markov decision process model we show that the precautionary measures, (i.e., infrastructure improvements triggered by the sea levels) are more effective in decreasing the expected cost than the aftermath measures triggered by the cost from nature. Secondly, to generate different scenarios we consider several sea level rise projections by NOAA, and model different government and resident prototypes using cooperation indices in terms of being responsive to the sea level rise problem. We present a reinforcement learning algorithm to generate simulations for a set of scenarios defined by the NOAA projections and cooperation indices.