See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/24254373

Nowcasting of rainfall and of combined sewage flow in urban drainage

systems

Article  in  Water Science & Technology · February 2009

DOI: 10.2166/wst.2009.098 · Source: PubMed

CITATIONS READS

35 145

4 authors:

Stefan Achleitner Stefan Fach

University of Innsbruck University of Innsbruck
117 PUBLICATIONS   779 CITATIONS    23 PUBLICATIONS   412 CITATIONS   

SEE PROFILE SEE PROFILE

Thomas Einfalt Wolfgang Rauch

hydro & meteo GmbH, Lübeck, Germany University of Innsbruck
129 PUBLICATIONS   1,063 CITATIONS    329 PUBLICATIONS   7,749 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

River Water Quality Model No. 1 View project

Hydrocyclone View project

All content following this page was uploaded by Wolfgang Rauch on 28 May 2014.

The user has requested enhancement of the downloaded file.

 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 2008

Nowcasting of rainfall and of combined sewage flow in urban

drainage systems
Stefan Achleitner 1,2*), Stefan Fach1), Thomas Einfalt3) and Wolfgang Rauch1)
1)
Unit of Environmental Engineering, Institute of Infrastructure
University of Innsbruck, Technikerstr. 13, A-6020 Innsbruck, Austria
2)
alpS – Centre for Natural Hazard Management, Grabenweg3, A-6020 Innsbruck, Austria
3)
hydro & meteo GmbH&Co.KG, Breite Str. 6-8, D-23552 Lübeck, Germany

*Corresponding author, e-mail: stefan.achleitner@uibk.ac.at

ABSTRACT
Nowcasting of rainfall may be used additionally to online rain measurements to optimize the
operation of urban drainage systems. Uncertainties quoted for the rain volume are in the range
of 5 % to 10 % mean square error (MSE), where for rain intensities 45 % to 75 % MSE are
noted. For larger forecast periods up to 3 hours, the uncertainties will increase up to some
hundred percents. Combined with the growing number of real time control concepts in sewer
systems, rainfall forecast is used more and more in urban drainage systems. Therefore it is of
interest how the uncertainties influence the final evaluation of a defined objective function.
Uncertainty levels associated with the forecast itself are not necessarily transferable to
resulting uncertainties in the catchment’s flow dynamics. The aim of this paper is to analyse
forecasts of rainfall and specific sewer output variables. For this study the combined sewer
system of the city of Linz in the northern part of Austria located on the Danube has been
selected. The city itself represents a total area of 96 km² with 39 municipalities connected.
It was found that the available weather radar data leads to large deviations in the forecast for
precipitation at forecast horizons larger than 90 minutes. The same is true for sewer variables
such a CSO overflow for small sub- catchments. Although the results improve for larger
spatial scales, acceptable levels at forecast horizons larger than 90 minutes are not reached.

KEYWORDS
Rainfall, runoff, CITY DRAIN, forecast, nowcasting, uncertainty

INTRODUCTION
With the growing number of real time control (RTC) concepts realized for operation of sewer
systems, the use of rainfall forecast increases as well (Berne et al.(2004); Einfalt et al. (2004);
Krämer et al. (2005)). However, with nowcasting of rainfall an additional source of
uncertainties is introduced that can affect the quality of the operation detrimentally.
Uncertainty levels associated with the forecast itself are not necessarily directly transferable
to resulting uncertainties in the catchment’s flow dynamics. Thus rainfall forecast can still be
useful as input data to a model-based predictive control concept.
The uncertainty associated with nowcasting of rainfall relies on various factors where the
quantification highly depends on the application itself. The instrumentation used, like rain
gauges or weather radar, and methods applied for forecasting affect the rainfall forecast
quality. Further the location, the topography, the size of the catchment as well as the time
horizon required for the forecast strongly influence the reliability of the rainfall forecast.

S. Achleitner, S. Fach, T. Einfalt and W. Rauch 1

 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 2008

Uncertainties associated with a forecast horizon of 30 minutes can be found in Pierce et al

(2004), where various systems and methods have been tested. Uncertainties quoted for the
rain volume are in the range of 5 % to 10 % mean square error (MSE), where for rain
intensities 45 % to 75 % MSE are noted. For larger forecast periods up to 3 hours, the
uncertainties will increase up to some hundreds of percents according to Golding (2000).
The aim of this paper is to analyse the uncertainty of rainfall forecasts and its effect to specific
sewer model output variables based on the case study of the sewer system of the city of Linz
(Austria). The sewer runoff was modelled with the conceptual software CITY DRAIN
(Achleitner et al., 2007). The intention was to provide a basis for decisions by operators
regarding the use of rainfall forecast as model input data, especially in the view of possible
future model-based predictive control concept to be implemented.

METHODOLOGY

Conceptual model of the sewer system

The urban drainage system of Linz in Austria was chosen as case study and implemented in
the open source software CITY DRAIN (Achleitner et al., 2007). The combined sewer system
covers a paved area of 2500 ha including the city and its surrounding suburban areas. Totally
300 000 people equivalents are connected. The conceptual urban drainage model consists of
30 sub catchments, as this made it possible to investigate the impact of the spatial rainfall
distribution. In addition to the catchments, combined sewer overflow structures, transport
sewers and pumping stations are represented in the urban drainage model. In Figure 1 the
conceptual model is illustrated. The conceptual model and its calibration is described in detail
in Kleidorfer et al. (2007) and Möderl et.al. (2007) (see also Figure 1). The grey shaded area
in the conceptual model (right side of Figure 1) complies with the urban area which is in dark
grey (left side of Figure 1).

Figure 1. Real world situation (left side) and conceptual model of the sewer system (right side)

The urban drainage model was calibrated using historic rainfall data as model input. As the
model was originally designed to determine the efficiency of the entire sewer system the
historic rainfall data used for calibration was not spatially distributed. For the use of spatially

2 S. Achleitner, S. Fach, T. Einfalt and W. Rauch

 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 2008

distributed rainfall data in this study, the model had to be adapted. Differences in simulation
results associated with this change from point rainfall to spatially distributed rainfall as model
input were found to be insignificant (see Kleidorfer et al. (2006))

The aim of this work was to illustrate the uncertainties of nowcasting of rainfall and its effect
to uncertainty in combined sewage flow predictions. Simulation results based on spatially
distributed measured rainfall as inputs were taken as the baseline scenario for comparison.
Thus, the uncertainty introduced by forecasting alone was investigated, without having
uncertainties due to e.g. calibration, measurement failures associated.

Availability of weather radar and raingauge data

Rainfall data were available from daily raingauges (20 stations) of which three stations also
measure in high resolution (5 minute data), and the raingauge Linz ZAMG with a 10 minute
time step. Radar rainfall was provided by the University of Graz as composite over all
Austria.

A number of extreme rainfall events were chosen based on the rain intensity, the total depth of
rainfall and the duration of the rain event. The number of events finally used for forecasting
was reduced to five events between June 2002 and July 2003. The decision was also based on
quality criteria of the weather radar data delivered. The selected events cover a period of 34
days. The radar composites had a time step of 10 minutes and a nominal resolution of 1018.6
m. The rainfall events selected were as followed:

- Event number 1: 06.06.2002 – 08.06.2002

- Event number 2: 03.07.2002 – 05.07.2002
- Event number 3: 06.08.2002 – 08.08.2002
- Event number 4: 10.08.2002 – 13.08.2002
- Event number 5: 22.07.2003 – 25.07.2003

Data quality
All data was extensively quality controlled by visual inspection of experienced staff. For the
raingauge data, there was one station where the data was wrong but could be corrected to give
reliable data. No other raingauge measurements were suspicious.

The radar data showed a number of particularities. The data being a composite concealed their
origin from the different radar stations. This made it slightly more difficult to attribute quality
effects of e.g. attenuation to a direction of origin. For the city of Linz, both radars covering
the area are situated relatively far away, with approximately 100 km (Salzburg radar) and 160
km (Vienna radar) (see Figure 2).

S. Achleitner, S. Fach, T. Einfalt and W. Rauch 3

 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 2008

Figure 2: Overview on the urban drainage area and radar coverage

The large distance affects the potentially attainable accuracy because the radar beam is
already at 2 – 4 km above Linz, and the size of an original measurement point is considerably
larger than 1 km².

The provided composite product is a projection of the maximal value inside the measurements
from different elevations over a point. Thus, the projected value is not necessarily
representing the lowest non-shielded radar beam which is the most representative for rainfall
at the ground.

Radar data processing

The radar data and the raingauge data for identical locations were compared on the basis of
daily volumes. An adjustment of the radar data to raingauges is usually performed by a
comparison of the rainfall measurements of the two instruments at the same location, the radar
data primarily having been transformed by a standard Z-R relationship to mm/h or mm.

The relation of raingauge rainfall to radar rainfall is giving a so-called correction factor which
is equivalent to an adjustment of the previously defined Z-R relationship to a more exact
value. Hypothesis for a successful adjustment is that this relationship between raingauge and
radar measured rainfall is homogeneous in space and time, which is a simplifying assumption.
However, many successful results (e.g. Einfalt et al. (2008)) have shown that this approach is
working well on average, also for rare events.

For the events for this study, little resemblance between radar and raingauge amounts was
visible, e.g. on the event of 12 August 2002 (Figure 3). Common observation for all events
was that the potential correction factor was high for strong events and low for less intense
events, indicating a systematic underestimation of high intensities. The spatially distributed
field of correction factors for the events could reduce this lack of high intensities, but a high
uncertainty remained for the interpretation of radar derived results.

4 S. Achleitner, S. Fach, T. Einfalt and W. Rauch

 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 2008

50

45

40

35

Radar [mm/d]
30

25

20

15

10

0
0 20 40 60 80 100 120
Raingauge [mm/d]

Figure 3: Observed relationship between radar and raingauge daily sums at the same locations

Forecast approach
For the radar forecasting, a tracking approach (Einfalt et al., 1990) was used to compute a
forecast for the following three hours at the arrival of every new radar image. The forecast
was calculated with a 10 minute time step for 180 minutes and based on the previously
adjusted radar images. The produced time series were catchment specific time series for the
subcatchments of the city on Linz and its connected sub urban areas, created by intersection
of the catchment topology with the radar pixels.

The CITY DRAIN model has been adapted such that 30 sets of consecutive forecast time
series as shown in Figure 4 were accepted as input. The time series contained a measured part,
followed by a forecast time series. In order to have a homogeneous type of model input
throughout a simulation run, measured parts from forecast time series were used instead of
rainfall derived directly from station data. Thus, for a simulation run a time series combined
of measured and forecasted rainfall data was used.

Figure 4. Generation of time series with measured and forecasted part

The state of the system calculated on basis of the measured rainfall data was made available
for the consecutive simulations with the rainfall data forecasted. Finally, forecast results were
compared to simulation results based on measured rainfall as input. The intention in doing
this was to assess the uncertainties due to forecast alone excluding uncertainties due to e.g.

S. Achleitner, S. Fach, T. Einfalt and W. Rauch 5

 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 2008

calibration, measurement failures. In the assessment it was of special interest to include the
impact due to an increasing forecast horizon. Further different locations were addressed
having differing drainage areas associated.

RESULTS
The conceptual model described earlier was used to point out the constraints due to
nowcasting of rainfall and of combined sewage flow. The evaluation was made for forecast
horizons +45, +90, +120 and +180 minutes. The simulation runs were analyzed concerning
the precipitation of single sub-catchments and the entire urban area. Additionally the
discharge, the rates of spill flow and the filling volume of CSO basins at selected points of the
urban drainage system which are of interest for the operator of the sewer system were
investigated. Exemplarily the precipitation of a single sub-catchment and the entire urban
area, the discharge into the main CSO tank and the resulting overflow are shown.

Figure 5. Graph of Precipitation of sub-catchment Gramastetten for different forecast horizons

Figure 5 illustrates the graph of precipitation of the single sub-catchment Gramastetten,

located in Northern part of the urban area. The rain intensity over a time period of five
minutes is plotted on the ordinate. The black line characterizes the measured precipitation and
the grey dots show the forecasted precipitation for forecast horizons of +45, +90 and +180
minutes respectively. It is obvious that the scattering of the forecasted precipitation increases
with the time period forecasted. For forecast horizons of more than 90 minutes the
uncertainties seems to be unacceptable with regard to managing any kind of RTC in the sewer
system.

Figure 6. Graph of Precipitation of four gathered sub-catchments for different forecast horizons

Due to significant divergence of precipitation measured and forecasted especially for time
periods of more than 90 minutes it was analyzed if the divergence will decrease with

6 S. Achleitner, S. Fach, T. Einfalt and W. Rauch

 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 2008

increasing catchment area. The idea was - since deviations vary spatially - that the
uncertainties are compensated on a larger spatial scale. Therefore the twelve sub-catchments
located within the urban area were gathered to four groups. The groups were built according
to their geographical position and location in the sewer system. The group specified by
“LINZ Stadt – B” consists of four sub-catchments. The associated precipitation for this part of
the urban area is shown in Figure 6. The other three aggregated sub-catchments indicate
nearly the same behaviour. Again the divergence from the measured values and with it the
uncertainties are only tolerable for a forecast horizon of less than 90 minutes.

Figure 7. Graph of rate of spill flow of CSO Gramastetten for different forecast horizons

Additionally to the precipitation the rate of spill flow has been analyzed in the same way as
described before. The results are shown for the sub-catchment Gramastetten as this subbasin
is not influenced by any other catchment located upstream. As a consequence the
uncertainties in the rate of spill flow can be explained only by the uncertainties of the
precipitation – with the assumption that the conceptual model has been calibrated sufficiently.
If the rate of spill flow is compared to the precipitation for the same forecasted time period it
is obvious that the divergence of the rate of spill flow is significantly less than that of the
precipitation. Due to this positive result from the point of view of an operator also the rate of
spill flow of the main CSO basin Weikerlsee has been analysed. This main CSO basin collects
the combined sewage of the entire urban area. Because of the long flow transport time it was
expected that the spill flow rate can be predicted accurately. The results in Figure 8 confirm
this assumption. If the results of Figure 8 are compared with those of Figure 7 the difference
is clear.

Figure 8. Graph of rate of spill flow of the CSO basin Weikerlsee for different forecast horizons

CONCLUSIONS AND OUTLOOK

The research project presented in this paper dealt with the uncertainties in the results of a
conceptual model due to uncertainties of forecasted rainfall. The approach to forecast

S. Achleitner, S. Fach, T. Einfalt and W. Rauch 7

 11th International Conference on Urban Drainage, Edinburgh, Scotland, UK, 2008

precipitation and with it the runoff in a sewer system is seen as a vital information for the
operator preliminary to the installation of a warning system, real time control (RTC) or model
based predictive control (MBPC) technology. The reliability in using quantitative rainfall
forecast has been evaluated for different locations, system variables and forecast horizons.

From this case study it can be concluded that 1) the uncertainty in the forecasted precipitation
exceeds by far the associated uncertainty in flow predictions and 2) that a forecast horizon of
more than 90 minutes introduces uncertainties in flow predictions that are not tolerable. The
uncertainties of the forecasted precipitation can be explained by the quality of the radar data
and by the constraints limiting the calibration of the radar data. Although the forecasted
precipitation varies a lot from the measured values the forecasted filling process and rate of
spill flow of the main CSO basin showed acceptable results which the operator can use to take
action.

REFERENCES
Achleitner S., Möderl M. and Rauch W. (2007). CITY DRAIN - An open source approach for
simulation of integrated urban drainage systems. Environmental Modelling &
Software, 22 (8),1184-1195.
Berne A., Delrieu G., Creutin J.-D. and Obled C. (2004). Temporal and spatial resolution of
rainfall measurements required for urban hydrology. Journal of Hydrology, 299 (3-
4),166-179.
Einfalt T., Denoeux T. and Jacquet G. (1990). A Radar Rainfall Forecasting Method Designed
for Hydrological Purposes Journal of Hydrology, 114,229-244.
Einfalt T., Scheibel M. and Mittelstädt R. (2008). Analysis of a damage producing flash flood
in the Wupper area. 11th ICUD Edinburgh (Abstract accepted).
Einfalt T., Arnbjerg-Nielsen K., Golz C., Jensen N.-E., Quirmbach M., Vaes G. and Vieux B.
(2004). Towards a roadmap for use of radar rainfall data in urban drainage. Journal of
Hydrology, 299 (3-4),186-202.
Golding B. W. (2000). Quantitative precipitation forecasting in the UK. Journal of
Hydrology, 239,286-305.
Kleidorfer M., Meyer S. S. and Rauch W. (2006). Konzepterstellung für die Bewirtschaftung
und Sanierung der Kanalisation Linz - Randbedingungen der Bewirtschaftung
Hydrologische Grundlagen. IUT - Arbeitsbereich Umwelttechnik, Innsbruck.
Kleidorfer M., Fach S., Möderl M. and Rauch W. (2007). Umsetzung von
Kanalnetzsteuerungen in hydrologischen Modellen am Beispiel von Linz.
Österreichische Wasser- und Abfallwirtschaft, 59 (9 -10),131-137.
Krämer S., Grum M., Verworn H.-R. and Redder A. (2005). Runoff modelling using radar
data and flow measurements in a stochastic state space approach. Water Science and
Technology, 52 (5),1-8.
Möderl M., Kleidorfer M., Fach S. and Rauch W. (2007). Reduktion von
Mischwasseremissionen durch Optimierung eines Entwässerungssystems am Beispiel
der Stadt Linz. Wiener Mitteilungen - Kanalmanagement 2007, 203,K1-K22.
Pierce C. E., Seed A., Ebert E. E., Fox N., Sleigh M., Collier C. G., Donaldson N., Wilson J.,
Roberts R. and Mueller C. (2004). The Nowcasting of precipitation during Sydney
2000: an appraisal of the QPF algorithms. Weather and Forecasting, 19 (1),7-21.

8 S. Achleitner, S. Fach, T. Einfalt and W. Rauch

View publication stats