Title: Climate change mitigation management using an optimal stochastic control framework
Authors: Daisuke Murakami - The Institute of Statistical Mathematics (Japan) [presenting]
Matthew Ames - Institute of Statistical Mathematics (Japan)
Pavel Shevchenko - Maquarie University (Australia)
Tor Andre Myrvoll - SINTEF (Norway)
Tomoko Matsui - The Institute of Statistical Mathematics (Japan)
Yoshiki Yamagata - ()
Abstract: In climate change mitigation management, it is crucial to understand the uncertainty in carbon dioxide emissions and atmospheric concentration, temperature, and damage caused by climate change. We investigate a stochastic control framework in order to find an optimal strategy for mitigating climate change using the Bellman equation. In the proposed method, the state variables consist of carbon dioxide emissions, carbon dioxide concentration and temperature, whereas the control variable is the mitigation cost. The optimal mitigation strategy that minimizes the discounted sum of damage and mitigation cost functions is estimated. While the damage cost function links temperature with economic influence, the mitigation cost function links carbon dioxide emissions and mitigation cost with economic influence. Since there may exist some unknown factors and uncertainty in both functions and it is difficult to specify them beforehand, we stochastically model them so that we do not need to define them explicitly. Furthermore, we compare methods of solving the Bellman equation via reinforcement learning and deep reinforcement learning. In the simulation experiments for future several dozens of years with different scenarios, we study the characteristics of the strategy which is the most suitable for mitigating climate change.