Abstract
This paper addresses the use of reliability techniques
such as Rosenblueths Point-Estimate Method (PEM)
as a practical alternative to more precise Monte Carlo approaches
to get estimates of the mean and variance of uncertain
flood parameters water depth and velocity. These parameters
define the flood severity, which is a concept used
for decision-making in the context of flood risk assessment.
The method proposed is particularly useful when the degree
of complexity of the hydraulic models makes Monte Carlo
inapplicable in terms of computing time, but when a measure
of the variability of these parameters is still needed.
The capacity of PEM, which is a special case of numerical
quadrature based on orthogonal polynomials, to evaluate the
first two moments of performance functions such as the water
depth and velocity is demonstrated in the case of a single
river reach using a 1-D HEC-RAS model. It is shown that
in some cases, using a simple variable transformation, statistical
distributions of both water depth and velocity approximate
the lognormal. As this distribution is fully defined by
its mean and variance, PEM can be used to define the full
probability distribution function of these flood parameters
and so allowing for probability estimations of flood severity.
Then, an application of the method to the same river reach
using a 2-D Shallow Water Equations (SWE) model is performed.
Flood maps of mean and standard deviation of water
depth and velocity are obtained, and uncertainty in the extension
of flooded areas with different severity levels is assessed.
It is recognized, though, that whenever application of
Monte Carlo method is practically feasible, it is a preferred
approach.