Environmental Engineering Reference
In-Depth Information
Distributed Models and
Uncertainty in Flood RiskManagement
14
KEITH BEVEN
The Requirement for Distributed Models
in Flood Risk Management
sediment mobilization, transport and deposition;
velocity fields for pollutant dispersion; and habitat
evaluation) or fluxes that might feed back to affect
the transport of water (such as gulleying and
changes in channel cross-sections during floods).
Similar arguments are still beingmade (e.g. Loague
and VanderKwaak 2004). In fact, these advantages
have been difficult to demonstrate, often for good
reasons; see discussions by Refsgaard et al. (1996),
Beven (1996a, 1996b), and the long history of trying
to model the R5 catchment at Coshocton, Ohio,
recorded in the series of papers byKeithLoague and
colleagues (Loague 1990; Loague and Kyriakidis
1997; VanderKwaak and Loague 2001; Loague and
VanderKwaak 2002; Loague et al.2005). Indeed,
there is evidence that the application of distributed
models may often fail acceptance criteria (e.g.
Parkin et al. 1996; Choi and Beven 2007; but see
Ebel and Loague 2006, for a declaration of success).
However, certainly in terms of bothmaking use
of distributed inputs and making distributed pre-
dictions where they are required in the catchment
system, the use of distributed models provides the
potential to reflect the spatial nonlinearities in
the system more explicitly than any lumped ap-
proach. This is, however, at the cost of greater
computational requirements (in that calculations
must be made for every variable in every discrete
element used in representing the flow domain),
and the need to specify very large numbers of
parameter values (in that all model parameters
can potentially have different values in every
discrete element).
The increase in computer power since the first
distributed models were implemented on digital
Post-event analysis of any particular flood event
will reveal that both the rainfall or snowmelt
inputs that caused it and the effects in terms of
areas flooded and damages caused will be spatially
variable or distributed in nature. The hydrology
and hydraulics of the event will reflect the hetero-
geneities in the driving variables and catchment
and channel characteristics. The distributed
nature of the processes is important, and the log-
ical consequence is that in trying to predict flood
events for flood management purposes we should
use distributedmodels whenever local distributed
inputs interact with local nonlinear processes to
produce responses where the distributed impacts
might be significant.
Early discussions of distributedmodels focused,
generally rather optimistically, on suchadvantages
of distributed models (e.g. Freeze and Harlan 1969;
Freeze 1978; Beven and O'Connell 1982; Beven
1985). At the time these were potential future
advantages. They included the possibility of direct-
lymeasuring or estimating 'physically based' para-
meter values; the value of making use of
distributed input conditions; the possibility of pro-
ducing local predictions; and the need for distrib-
uted predictions of output fluxes for other purposes
(prediction of water quality variables; prediction of
