Multi-Objective Policy Trees for Dynamic Adaptation to Climate Change

Abstract:

Water resources systems face a wide range of uncertainty in future hydroclimatic and socio-economic conditions, justifying an adaptive planning approach. Recent advances in dynamic adaptation have designed policies linking infrastructure and management actions to indicator variables monitored over time via thresholds. Typically, one or more of these components are prespecified, constraining the flexibility of policy generation. The opportunity exists to develop methods that identify the most relevant indicators, actions, and thresholds to combine in developing policies for dynamic adaptation to climate change. Here we present a generalized framework to address these challenges based on multi-objective policy tree optimization, a heuristic policy search method in which adaptation policies are represented as binary trees. We demonstrate this framework using an illustrative water resources planning problem in California where infrastructure expansion, reservoir operations, conservation rules, and conjunctive use are adapted over time to balance flood risk, water supply, and environmental flow objectives. To capture the uncertainty in nonstationary forcing, indicator variables include long-term hydroclimatic statistics from downscaled GCM projections along with uncertain land use and economic conditions. Robustness of policy trees trained to this scenario ensemble is determined by validation against a set of synthetic scenarios representing a broader range of uncertainty in transient hydroclimatic trends, natural variability, and socio-economic conditions. ­­­We determine the relative value of these adaptation policies by comparing validation results with no-action and perfect foresight benchmarks. The framework developed in this study is widely transferrable across water resources systems challenged with adaptive planning under uncertainty.