362
G.A. Schultz
the movement of such cells towards the catchment under consideration may serve
as basis for computation of rainfall input for a rainfall-runoff model.
16.2.4Transformation of Area Precipitation into a Real-time Forecast of a
Runoff Hydrograpb
With the aid of a distributed rainfall-runoff model and known model input, (e.g.
observed and forecast rainfall) it is possible to produce a flood hydrograph forecast in real-time.
Model Structure. There are, of course, many potential model structures which are
more or less suitable for transforming precipitation measurements together with
QPF into a real-time flood hydro graph forecast. If no hydrological model of the
distributed system type can be applied (e.g. since no adequate data are available) it
is certainly possible to use a simple lumped system model like the Unit Hydrograph. It has to be borne in mind, however, that this way the important information on spatial precipitation variability is lost and thus the forecast will be of lower
accuracy and reliability. In order to take advantage of the spatial precipitation
variability it is necessary to produce the flood forecast on the basis of a distributed
system hydrological model. There are plenty models of this type "on the market"
and here only one such model developed at the authors institute will be presented
briefly as an example.
The structure of this physically based rainfall-runoff model is related to the
three hydrological processes: (1) runoff generation, (2) runoff formation on hillslopes and (3) runoff concentration within the river network. For each process the
catchment is sub-divided into spatial units considering the most relevant physical
characteristic of each process. The heterogeneity of other relevant characteristics
within these units is described by distribution functions or average parameters. In
order to couple the three model components the fluxes of water from each one has
to be distributed among the units of the next component. Figure 16.3 shows in
form of a flow chart the coupling of the three process models which form the
rainfall-runoff model. As can be seen in the top process component one of the
catchment characteristics used in the model is soil storage capacity. This is determined in the way discussed in Sect. 16.2.3. More details on the model structure,
estimation of initial conditions and model performance are given in the literature
(e.g. Schumann and Funke, 1996). As mentioned before any other type of distributed system model may also be suitable for rainfall input obtained from weather
radar or satellite imagery.
Real-time Model Parameter Updating. The quality of a flood forecast depends
not only on the model quality but also on the accuracy of the parameter estimation
in real-time. If, for a certain flood event, the ftrst forecast has to be made the actual values of the model parameters are not known. It is customary therefore to
start the ftrst forecast with the aid of average parameter values known from previous forecasts. Since the flood forecast has to be updated, e.g. every hour, while
new information (rainfall and runoff measurements) comes in the parameters have
to be re-estimated every time a forecast is made on the basis of new information
G.A. Schultz
the movement of such cells towards the catchment under consideration may serve
as basis for computation of rainfall input for a rainfall-runoff model.
16.2.4Transformation of Area Precipitation into a Real-time Forecast of a
Runoff Hydrograpb
With the aid of a distributed rainfall-runoff model and known model input, (e.g.
observed and forecast rainfall) it is possible to produce a flood hydrograph forecast in real-time.
Model Structure. There are, of course, many potential model structures which are
more or less suitable for transforming precipitation measurements together with
QPF into a real-time flood hydro graph forecast. If no hydrological model of the
distributed system type can be applied (e.g. since no adequate data are available) it
is certainly possible to use a simple lumped system model like the Unit Hydrograph. It has to be borne in mind, however, that this way the important information on spatial precipitation variability is lost and thus the forecast will be of lower
accuracy and reliability. In order to take advantage of the spatial precipitation
variability it is necessary to produce the flood forecast on the basis of a distributed
system hydrological model. There are plenty models of this type "on the market"
and here only one such model developed at the authors institute will be presented
briefly as an example.
The structure of this physically based rainfall-runoff model is related to the
three hydrological processes: (1) runoff generation, (2) runoff formation on hillslopes and (3) runoff concentration within the river network. For each process the
catchment is sub-divided into spatial units considering the most relevant physical
characteristic of each process. The heterogeneity of other relevant characteristics
within these units is described by distribution functions or average parameters. In
order to couple the three model components the fluxes of water from each one has
to be distributed among the units of the next component. Figure 16.3 shows in
form of a flow chart the coupling of the three process models which form the
rainfall-runoff model. As can be seen in the top process component one of the
catchment characteristics used in the model is soil storage capacity. This is determined in the way discussed in Sect. 16.2.3. More details on the model structure,
estimation of initial conditions and model performance are given in the literature
(e.g. Schumann and Funke, 1996). As mentioned before any other type of distributed system model may also be suitable for rainfall input obtained from weather
radar or satellite imagery.
Real-time Model Parameter Updating. The quality of a flood forecast depends
not only on the model quality but also on the accuracy of the parameter estimation
in real-time. If, for a certain flood event, the ftrst forecast has to be made the actual values of the model parameters are not known. It is customary therefore to
start the ftrst forecast with the aid of average parameter values known from previous forecasts. Since the flood forecast has to be updated, e.g. every hour, while
new information (rainfall and runoff measurements) comes in the parameters have
to be re-estimated every time a forecast is made on the basis of new information