FRIEND '97 Regional Hydrology: Concepts and Models for Sustainable Water Resource Management

(Proceedings of the Postojna, Slovenia, Conference, September-October 1997). IAHS Publ. no. 246, 1997.

297

Prediction of design storms and floods

URSZULA SOCZYNSKA, BARBARA NOWICKA,
URSZULA SOMOROWSKA
The University of Warsaw, Faculty of Geography and Regional Studies,
Krakowskie Przedmiescie 30, 00-927 Warsaw, Poland

ELZBIETA KUPCZYK & ROMAN SULIGOWSKI

Pedagogical University in Kielce, Institute of Geography, Konopnickiej 15,
25-406 Kielce, Poland

Abstract In this work a method to estimate design floods based on rainfall

information has been proposed and verified. A design hyetograph has been
developed on the basis of 30-years of pluviographic records from 36
meteorological stations. The general formula describing rainfall intensityduration-frequency relationship has been derived from recorded series. The
spatial distribution of the derived parameters made it possible to estimate
rainfall intensity quantiles at each site within the territory of Poland. The
typical temporal pattern of storms have been obtained for the genetic type of
rainfall and for the selected pluviographic regions. The determined synthetic
storms were then transformed into outflows for the same return periods. The
method has been verified by fitting probability distributions to maximum
annual flows calculated from direct observations from the analysed catchments. This method, apart from giving maximum flows, gives more comprehensive information in the form of the hydrograph of stated probability
(design hydrograph), as the hydrological basis for hydrotechnical design.
ESTIMATION OF THE DESIGN STORM
Frequency analysis of extreme rainfall has been based on an annual exceedance
series. Frechet's model (Sevruk & Geiger, 1981) selected from the family of
generalized extreme value distributions the one that proved to fit best to the empirical
series and made possible the estimation of quantile values beyond the range of the
observed data sets. The probability density function of Frechet's distribution is
expressed as follows:
f(x)= aV a /

x>z

(1)

x<z

The Maximum Likelihood Method was used to estimate the two parameters (a, (3).
The third (s)displacement parameterwas derived from the rainfall data
properties. The graphic form of rainfall intensity-duration-frequency relationship is
illustrated in Fig. 1. The general formula developed to describe the intensityduration-probability relationships is of the following form:
IP = ^-t?

(2)

where Ip is rainfall intensity (at given probability level), tr is the rainfall duration, A
and B are parameters for each range of return intervals.

Urszula Soczynska et al.

298

tZ_i

i_i

'

"

18

i_

12

tr[h]
Fig. 1 Rainfall intensity-duration-probability of exceedance relationship at Gdansk.

Figures 2 and 3 illustrate spatial variations of parameters A and B respectively,

for Polish areas for the return period of 100 years. The design hyetograph was
developed on the basis of the detailed data covering the years 1961-1990. The
maximum rainfall storms grouped into 13 classes according to their duration were
extracted from independent rainfall events ranging from 10 minutes to 24 h. Rainfall
series were then subdivided according to the type of events on the basis of statistical
analysis of rainfall characteristics and also the main features of genetic type. The key
rainfall characteristic was found to be the function of the total depth of a rainfall
event to its duration. Rainfall episodes were subdivided into three sections
corresponding to time of duration intervals. The variation of depth-duration
relationship expressed the modification of storm characteristic and also proved the
occurrence of changes within the precipitation producing mechanism. The distinction
between groups of storms is due to the physical background and may be related to
the original type of precipitation (Sumner, 1988). The episodes belonging to
section 1 are linked with convection type processes; those in section 2 result from
frontal activity, while those in section 3 result from rainfall generated by a moving
depression or convergence zone on a synoptic scale. The length and slope of the
selected sections obtained from the depth-duration curve at each site proved to be
useful and provided objective criteria for regional analysis of rainfall. The stations
were classified into the following four regions by applying the hierarchical cluster
analysis method (Johnson & Wichern, 1982):
(a) sites with continental features and pre-mountainous sites,
(b) maritime sites,
(c) mountainous area sites,
(d) the group of sites strongly influenced by local circulation.
For evaluation of homogeneity of the regions, discordancy measures based on the
L-moments method (Hosking & Wallis, 1993) have been estimated.
The typical temporal pattern of storms has been obtained for each type of rainfall
and for selected pluviographic regions. These storm profiles can be accepted as the
pattern of the design hyetograph.

Prediction of design storms and floods

299

Fig. 2 Spatial distribution of parameter A at / = eA-trB at 1% probability of

exceedance.

Fig. 3 Spatial distribution of parameter B at /

exceedance.

= eA-trB at 1% probability of

SELECTION AND TESTING OF MODELS

Six different conceptual models were applied for rainfall-runoff transformation. As
the main objective of the study was to work out the method to estimate design storms
for ungauged basins, the following criteria were applied when choosing models for
the analysis:
(a) model should have small number of parameters,
(b) parameters should be easily estimated from existing topographic maps. Applied

Urszula Soczynska et al.

300

models included:
(i) Wackermann model in original version I (Thiele & Euler, 1981) and
modified one II (Ignar, 1993),
(ii) three versions of the GIUH (Rodriges-Iturbe & Valdes, 1979; Soczynska &
Nowicka, 1989; Ostrowski, 1994),
(iii) Nash model with relationship proposed by Lutz (1984).
All the above-mentioned are the lumped type of models. The models were adopted for
11 basins. Most of them are located in mountainous regions. A total of 95 different
recorded flood events were used in the analysis. The SCS and runoff coefficient
methods were adopted for effective rainfall determination. Thirteen different parameters were evaluated for testing of chosen models. They comprised physiographic and
river bed parameters and effective rainfall characteristics. Physiographic parameters of
basins were calculated from the topographical maps using the ILWIS GIS package. The
type of parameters and characteristics used are summarized in Fig. 4.
A computer program in Turbo Pascal was used to simulate flood hydrographs
and testing of the models. Examples of flood hydrographs simulated for Sleza River
basin using different models are shown in Fig. 5.
DATABASE

Physiographic data

Hydrologie data

Flood events data

Parameters

Flood events data

(raster and vector

data - in ILWIS format)

(ASCII files)

(ASCII files)

river network maps;

river network maps
according to Strahler's
classification;
river network maps
according to modified
Strahler's classification;
landuse maps.

- basins area;
- maximum elevation;
- minimum elevation;
- avaraged river channel
slope;
- Horton's Law parameters
(R R ,R.,R A ' ^ Q ,);
- main river length

- river length from outlet

to basin's boundary;
- distance from the basin
outlet to projected
gravity center
- percentage of urbanised
areas.
Fig. 4 Structure of the computer database.

hydraulic parameters
of the riverbed:
V0 - initial stream
velocity
S - stream level slope
n - Manning's
coefficient
B - width of the river
bed
rainfall and flow data
(1-hour and daily data).

301

Prediction of design storms and floods

On the basis of the results obtained (analysis of the correlation coefficient R,

special correlation coefficient RS, total square error CBK Delleur et al, 1973) four
models have been selected for further research: Wackermann I and II, GIUH III and
the Lutz model.
DESIGN FLOOD ESTIMATION
Design storms were transformed into the design flood hydrographs assuming the
flood discharge has the same probability as the design storm. The following steps
were necessary to follow in order to verify the proposed method:
Wackermann SGGW Model

Wackermann Model

10

20

30

40

50

40

50

t[h]
GIUH UW1 Model

GIUH UW2 Model

10

20

30

t[h]
GIUH IMGW Model

10

20

Lutz Model

30

t[h]

a - observed hydrograph, b - calculated hydrograph

Fig. 5 Results
suits of
oi flood hydrograph simulation in Sleza River basin at Bialobrzezie

(VMarch 1980).

302

Urszula Soczynska et al.

600

90 MO MO 1 30

g|o J 40 4

500

10

2P

0,5

400
300

<P>

200

:;;|g:

100

fill

Exceedance probability (%)

O- Wackermann I

- X - Wackermann II

GIUH III

a Lutz
Confidence interval
Fig. 6 Comparison of the simulated probability curves with the theoretical one
calculated after Pearson III distribution model (Skawa River basin at Osielec).

(a) estimation of the maximum probable rainfall as an input to hydrological models,

(b) estimation of the maximum probable flows by statistical analysis using data
series longer than 30 years,
(c) estimation of probable hydrographs using hydrological models,
(d) comparison of simulated maximum probable flows with the values of flood
probability curves described by Pearson III type distribution.
Rainfall-runoff transformation was conducted with the following assumptions:
(a) CN parameter of the SCS method was equal to 100 and respectively runoff
coefficient a = 1,
(b) considered design storms were of a duration longer than 8.5 h in all rainfallrunoff transformations.
Examples of results are shown for the Skawa basin at the Osielec cross-section.
Figure 6 shows a comparison of flood probabilities curves determined on the basis of
applied models and the statistical distribution. Figure 7 presents example of design
flood hydrographs estimated by the Lutz model for Skawa basin at the Osielec crosssection.
CONCLUSIONS
(a) Elaborated method of the design flood hydrograph estimation can be applied for
the ungauged natural basins.
(b) In mountainous basins with relatively large slopes and good conditions for direct
runoff creation, design storm can constitute an input to rainfall-runoff models
without reduction for losses.
(c) In most cases the simulated probable flows were less than observed ones,
especially for low probabilities (p < 20%).
(d) Results of simulations obtained from different models were close to each other
and therefore all of chosen models could be applied in practice.

Prediction of design storms and floods

10

20

30

40

50

60

70

80

303

90

t(h)
Fig. 7 Design flood hydrographs for the Skawa River basin at Osielec for
probabilitiesp = 1%, 50%, 90%.

(e) Application of the method for lowland basins requires additional studies
especially with regard to effective rainfall and concentration time determination.
Acknowledgements This project was sponsored mainly by the Committee of
Scientific Research in Poland. The authors would also like to thank the Faculty of
Geography and Regional Studies, Warsaw University, Institute of Geography,
Pedagogical University in Kielce, the Department of Land Reclamation and
Environmental Engineering, Warsaw Agricultural University and the Institute of
Meteorology and Water Management for giving possibility to perform the research.

REFERENCES
Delleur, J. W., Sarma, R. B. S. & Rao, A. R. (1973) Comparison of rainfall-runoff models for urban areas. /. Hydrol.
18(3-4).
Hosking, J. M. R. & Wallis, J. R. (1993) Some statistics useful in regional frequency analysis. Wat. Resour. Res. 29(2),
271-281.
Ignar, S. (1993) Metodyka obliczania przeplywow wezbraniowych w zlewniach nieobserwowanych (Methodology of
flood flows computations in ungauged basins). Wydawnictwo SGGW, Warszawa.
Johnson, A. R. & Wichern, D. W. (1982) Applied Multivariate Statistical Analysis. Prentice-Hall, New Jersey, USA.
Lutz, W. (1984) Berechnung von Hochwasser Abflussen unter Anwendung von Gebietskenngrossen. Mitteilungen IHW,
H. 24, Karlsruhe.
Ostrowski, I. (1994) Model regionalny malej zlewni "MOREMAZ-1" (A small watershed regional model). Mat.
Badawcze, seria: Hydrologia i Oceanologia 17. IMGW Warszawa.
Rodrigez-Iturbe, I. & Valdas, I. B. (1979) The gomorphologie structure of hydrologie response. Wat. Resour. Res.
15(6).
Sevruk, B. & Geiger, H. (1981) Selection of distribution types for extremes of precipitation. Operational Hydrology
Report no 15 (WMO no. 560). WMO Geneva.
Soczynska, U. & Nowicka, B. (1989) Application of GIUH and dimensionless hydrograph models in ungauged basins.
In: FRIENDS in Hydrology (ed. by L. Roald, K. Nordseth & K. A. Hassel) (Proc. Bolkesjoe, Norway, Conf.,
April 1989), 197-203. IAHS Publ. no. 187.
Sumner, G. (1988) Precipitation. Process and Analysis. John Wiley & Sons.
Thiele, F. & Euler, G. (1981) Vergleichende Untersuchung zur Ermittlung von Uebertragungsfunktionen aus
Einzugsgebietsgrossen nach verschiedenen Methoden. Untersuchung filer den DVWK-FA NiederschlagAbflussmodelle.