A methodology for evaluation of flood forecast‐response systems: 2. Theory

Abstract

A methodology for evaluation of the performance of flood forecast‐response (FFR) systems is presented. It is built of (1) a model of the FFR process and (2) a set of performance measures. The FFR process is formulated as a finite, random duration, Markovian decision process which outputs the expected annual loss (cost of response plus flood damage sustained). The sequence of forecasts of the flood crest and the actual river stages are described by a two‐branch Markov chain of order one. The decision maker's response is represented by a strategy defined on a three‐dimensional state vector and time. The performance measures consist of a value, an efficiency, and an expected opportunity loss for each system component as well as the total system. They are defined on a vector of outputs, each output being computed for a distinct scenario of system operation. Two evaluation models are constructed: one for a single decision maker and another for an arbitrary collection of decision makers along a reach (e.g., a community).