Inferring and Simulating Forecast Use in Reservoir Control Policies across the contiguous United States
Abstract
Reservoirs across the United States often release water based on guide curves to meet multi-sectoral objectives. These guide curves commonly connect water releases with factors such as the amount of water stored, time of year, and occasionally a forecast of future incoming flows. However, these guide curves are rarely publicly available. We argue that in order to devise improved operating policies for reservoirs under evolving objectives and forecast accuracy, it is imperative to first understand current operations to establish a baseline policy against which alternative policies can be compared. A major uncertainty in inferring policies is to understand which forecast for which season and events (flood, drought) are contributing to the release decision. This study explores whether the current operations of reservoirs can be inferred and simulated with policy functions identified through direct policy search (DPS). The analysis spans 314 reservoirs across the contiguous United States. At each reservoir, radial basis function (RBF) parameters are optimized using an evolutionary algorithm to build a policy function that reproduces historical release decisions. Three policies are tested that consider 1) storage and time of year only; as well as 2) an additional one-week forecast of water availability; or 3) an additional forecast of water availability with lead time that varies throughout the year. For the 3rd policy, the time-varying signature of forecast lead time is also derived based on observed releases. These three formulations are compared to determine if we can effectively simulate the range of reservoir releases under flood and drought risks across large systems. Initial results suggest that multi-lead time forecast use can be effectively modeled, especially for large systems dominated by spring snowmelt. These results indicate that realistic baseline policies can be inferred for many reservoir systems, which can be used to better represent flood and drought risk in large scale hydrology modeling research and infer the stresses of future climate conditions.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.H23G..05D
- Keywords:
-
- 1807 Climate impacts;
- HYDROLOGY;
- 1880 Water management;
- HYDROLOGY;
- 1918 Decision analysis;
- INFORMATICS;
- 6309 Decision making under uncertainty;
- POLICY SCIENCES & PUBLIC ISSUES