New Publication: Bayesian-Belief DPS for Adaptive Water Supply Planning
We are excited to share that Fletcher Lab PhD candidate, Mofan Zhang’s paper, “Bayesian-Belief Direct Policy Search for Adaptive Water Supply Planning with Endogenous Learning,” is now published.
Long-term water infrastructure planning is increasingly challenged by deep uncertainty as climate conditions evolve over time. Traditional adaptive planning approaches can help systems respond to changing conditions, but they often rely on static representations of future climate uncertainty. This paper introduces Bayesian-Belief Direct Policy Search, a new framework that integrates Bayesian learning into a reinforcement-learning-based framework for adaptive infrastructure planning.
By endogenously modeling climate beliefs, Bayesian-Belief DPS complements flexibility with learning, enabling adaptive policies to respond to evolving uncertainty and improve robustness under deep uncertainty. The study shows that incorporating learning into adaptive planning can improve system performance, particularly under nonlinear climate change and long-lived infrastructure decisions.
More broadly, this work highlights the value of learning-based adaptive planning for supporting more reliable and robust water supply decisions in a changing climate.