Geoscience Reference
In-Depth Information
precipitation forecasts in driving hydrological forecasting models (e.g. Wu
et al.
2011). It is an on-going
research question as to how to best make use of such forecasts when they are known to have deficiencies,
especially in mountain areas and in the representation of local convective rainfalls.
At least for cases where the catchment response time is greater than the lead time for decision making,
this makes it important that any rainfall-runoff model used be capable of real-time adaptation to take
account of any errors in the forecasts resulting either from errors in the inputs, whether from radar or
rainfall, or from error in the model structure. This requires, however, that the flood warning centre also
receive information about river stages in real time, at one or more gauging stations in a catchment, so that
model forecasts can be compared with observed stages or discharges in real time and the model adapted
to produce more accurate forecasts (at least until the gauge is washed away or the telemetering system
fails). Some ways of doing this are discussed in Section 8.4.
There is another reason why observed stage information might be useful in flood warning, particularly
in larger catchments where the time delays in the channel system are sufficiently long compared with the
required lead time for a forecast. In general, two to six hours would be the minimum feasible lead time to
allow a warning to be transmitted to the public but longer lead times might be needed for decisions about
deploying demountable flood defences, for example. A measurement of the discharge or stage upstream
can be used as part of the system for forecasting the stage and discharge and timing of the flood peak
further downstream. Such observations can also help constrain the uncertainty in propagating forecasts
from the headwaters of a catchment in predicting the risk to flood-prone areas downstream.
In general, flood warnings are issued in relation to the forecast stage of the river at a critical gauging
point without modelling the detailed pattern of inundation upstream of that point. In many situations this
may be adequate, since if flooding is predicted to occur somewhere in the flood plain, then a general
warning can be issued. In large rivers, however, such as the Mississippi, the progress of the flood wave
downstream may be very much controlled by the pattern of inundation during the flood, including the
effects of dyke failures which are inherently difficult to predict ahead of time. Thus it may be necessary
to use a hydraulic routing model in forecasting the expected depths downstream, continually revising the
calculations as conditions change (although transfer function methods can also be used for this purpose
for specific sites; see, for example, the work of Beven
et al.
(2008b), which includes the use of data
assimilation where measured levels are available, and Section 8.5). The use of a hydraulic model adds
the requirement of knowing the channel and flood plain topography, together with parameters such as
effective resistance coefficients. Topography is normally provided as a series of cross-sectional profiles
surveyed across the flood plain and channel at different sites, but the increasing use of two-dimensional,
depth-integrated models will lead to the use of topographic data in the form of detailed digital elevation
maps of the flood plain. Channel form, can of course change during a flood event due to erosion and
deposition. Models of sediment transport in rivers have not advanced to the stage where they can be used
operationally and most current hydraulic flood routing models use a “fixed bed” assumption.
8.4 Rainfall-Runoff Modelling for Flood Forecasting
Any rainfall-runoff model that has been calibrated for a particular catchment can be used in the prediction
of flood discharges. The US National Weather Service River Forecast System (Burnash, 1995), for
example, is a development of the Sacramento model, a form of lumped explicit soil moisture accounting
model with many parameters to be calibrated (see, for example, Sorooshian
et al.
, 1992; and Gupta
et al.
,
1999). Using methods such as those discussed in Chapter 7, the predictions may also be associated with
an estimate of uncertainty in the predictions. Qualifying the estimates in this way may be important.
Experience suggests that uncertainty in both measurements and predictions of flood peaks increases with
peak magnitudes. In addition, even if a model has been calibrated for a certain range of discharges,
uncertainty is bound to increase as predictions are made outside this calibration range for extreme events.
