A Tree-Based Policy Search Framework for Dynamic Adaptation to Climate Change in Water Resources Systems
Abstract
Long-term water resources planning is challenged by substantial uncertainty in projections of flood and drought risk under climate change. Recent studies focused on adaptation of infrastructure and operations have therefore developed a dynamic approach in which actions are taken in response to observations as they occur (i.e., a control problem). There remains an opportunity for computational methods to design and test dynamic adaptation policies by (1) identifying the relevant indicators to monitor, including climate, hydrologic, and human variables; (2) mapping key thresholds of these variables (signals) to candidate actions; and (3) validating against out-of-sample scenarios to ensure robustness to climate and other uncertainties. Here we present a generalized framework to address these challenges based on policy tree optimization, a heuristic policy search method in which adaptation policies are represented as binary trees. The policy search designs a mapping between plausible future conditions and one or more actions, assigning threshold values to trigger dynamic adaptations without predefining the structure or complexity of the policy. We demonstrate this method for an illustrative water supply planning problem in California focused on reservoir expansion and operation under climate change, where indicator variables include a set of long-term hydroclimatic statistics. The resulting policies are then validated against a held-out subset of downscaled GCM projections, and benchmarked against a baseline "no action" case as well as a perfect foresight case to establish the relative value of the adaptation policies. The validation procedure applied to this particular case study highlights challenges relevant to any dynamic adaptation approach, namely, the characterization of climate and other uncertainties (model-based versus synthetic) as well as the search for informative indicator variables. The framework is open-source and transferable across planning contexts focused on threshold-based dynamic adaptation under climate uncertainty.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.H21G..02C
- Keywords:
-
- 1807 Climate impacts;
- HYDROLOGY;
- 1880 Water management;
- HYDROLOGY;
- 1918 Decision analysis;
- INFORMATICS;
- 6309 Decision making under uncertainty;
- POLICY SCIENCES & PUBLIC ISSUES