Columbia River Treaty 2014/2024: Monte Carlo Simulation in HEC-WAT

Abstract

The Flood Risk Analysis (FRA) compute option is an enhancement to the Hydrologic Engineering Center's (HEC) Watershed Analysis Tool (HEC-WAT) that allows the user to perform risk analysis using a Monte Carlo simulation approach. Hydrologic inputs as well as parameters of the models in the HEC-WAT program sequence can be defined as random variables to represent uncertainty. In order to separately consider epistemic and aleatoric uncertainty (knowledge uncertainty and natural variability), the HEC-WAT sequence of programs is executed for a large random sample of flood events, which are clustered into realizations. Random variables with epistemic uncertainty vary from one realization to another, while variables driven by aleatoric uncertainty are re-sampled for every flood event in a realization. The Monte Carlo nature makes FRA simulations (with thousands of flood events) much more computationally intensive than deterministic HEC-WAT simulations, and so not all results from all models need to be saved nor even the programs/models executed for every randomly sampled flood event. Computation of an event can be aborted at any step of the program sequence according to user-defined criteria. Conversely, interesting events can be tagged so that detailed results are saved. Interesting results can be replicated, modified or analyzed in more detail. In order to achieve repeatability, random number generators throughout the FRA compute are seeded in a deterministic, pseudo-random hierarchical fashion. Monte Carlo iteration continues until the mean or a quantile of one or more userdefined variables chosen from the sequence of models converge within certain tolerance. At the back end of a simulation, statistical summaries are assembled by the Performance Metrics plug-in. Data is collected throughout a simulation to estimate variables of interest such as average annual damage, or annual exceedance probability.