The ensemble output from a reservoir management system represents the best information we can obtain of the future predictions and builds on both our theoretical knowledge and all available data combined in a statistically optimal way. We will develop probabilistic decision models that will work well with an ensemble representation of the information. The focus is on finding efficient methods for computing decision variables (or controls) that maximize the expected value of an objective and perhaps simultaneously minimize some measure of risk. In the case of binary decisions, the methodology is trivial, but for continuous and integer control variables the optimization under uncertainty can be expensive, so it is crucial to make optimal use of ensemble information.
.png)