Rainfall Runoff Modelling
Spending on large infrastructure projects in Australia,
such as new bridges and roads, is worth at least A$1BN each year. These
projects therefore represent a major national investment. The consequences
of structural failure are potentially severe, not only in economic terms
but also in social disruption and potential for loss of life.
An important consideration in the design of such
structures is their structural integrity in relation to flooding. To
examine this question flood hydrologists examine flood behaviour under
various design scenarios and storm patterns. Runoff routing models are
often used to estimate design flood hydrographs. These models route
rainfall excess through a conceptual representation of catchment storages
to produce a surface runoff hydrograph at one or more points in the
catchment. Distributed, temporary storage on the catchment surface is
represented by a number of conceptual storages linked according to the
geometry of the channel network. This situation is depicted below:
Calibrating Rainfall-Runoff Models
Using a rainfall-runoff model floods in the stream
network can in principle be predicted at any point in the catchment,
provided suitable parameter values can be found for the runoff routing
model. There are a number of approaches to this problem provided suitable
stream flow data are available. In the hydrological literature it is
common practice to fit separate models for each storm event and then to
combine or "pool" them in some way to derive common estimates.
The Environmetrics Group
has been collaborating with CSIRO Land and Water’s Hydroclimatic
Processes and Impacts Group (http://www.clw.csiro.au/research/catchment/hydroclimatic/hydroclimatic.htm)
to develop a new approach to this important problem. The approach we have
developed includes a diagnostic for testing whether it is reasonable to
pool parameter estimates, and is described in Campbell
et al. (1999). We adopt a Bayesian approach to model
calibration that allows us to integrate available data with the
considerable expert opinion that is available. Our results suggest that
naïve pooling of parameter estimates is not appropriate. Instead there is
some evidence in our work that parameter estimates tend to cluster into
groups, perhaps determined by storm track and intensity. It may therefore
be appropriate to use different rainfall-runoff model parameters for
different storm patterns.
The Regionalisation Problem
It is often the case in practice that stream flow data
are not available to calibrate a rainfall-runoff model. The general
approach taken in the hydrological literature to these problems is to link
the rainfall-runoff parameters to catchment characteristics using
regression techniques. That is, rainfall-runoff parameters are related to
physical features of the landscape- the so-called "regionalisation"
problem. By measuring these physical features in the location of interest
we can then derive parameters for the rainfall-runoff model. In Campbell
and Bates (2001) we developed a new framework for regionalisation within a
Bayesian statistical paradigm. Using this approach we are able to
integrate available stream flow and catchment characteristic data with
expert opinion to define regionalisation relationships. Our results
suggest that the approach has substantial predictive capability for our
case study.
Related Work
We have recently completed work on extending our model
calibration approach to continuous simulation models, described in Bates
and Campbell (2001). Continuous simulation models use a continuous record
of stream flow, not just discrete storm events, to model the
rainfall-runoff process. In this work we have made a number of innovations
in the design of our model calibration algorithm to respect a number of
physical constraints.
For more information
Eddy
Campbell : Ph: +61-(0)8-9333-6203 Fax:
+61-(0)8-9333-6121
References Cited:
Bates, B. C. and Campbell, E. P. (2001). A Markov chain
Monte Carlo scheme for parameter estimation and inference in conceptual
rainfall-runoff modeling. Accepted by Water Resour. Res.
Campbell, E. P. and Bates, B. C. (2001).
Regionalization of rainfall-runoff model parameters using Markov chain
Monte Carlo samples. Accepted by Water Resour. Res.
Contact: Allan Adolphson
Ph: +61-(0)2-9325-3261 Fax: +61-(0)2-9325-3200
|
To
top