Sei sulla pagina 1di 8

INTERCOMPARISON OF THE LUMPED VERSUS SEMI-DISTRIBUTED HEC -

HMS HYDROLOGICAL MODEL IN THE KALAMAS RIVER BASIN



L. I. NTOANIDIS


Environmental Engineer - Hydrologist, 2 Tinou Str, 15562, Cholargos, Greece,
E-mail: l.ntoanidis@gmail.com


ABSTRACT

This paper compares the lumped and semi-distributed structure of the HEC-HMS model,
using daily data of the 1449.3 km
2
Kalamas river basin in Greece. The hydrologic analysis
of the study area is attempted for four storm events in two ways: a) with division of the
area in two sub-basins (lumped model) and b) with division of the area in eight sub-basins
(semi-distributed model). The model calibration procedure was conducted on two out of
four flood events, and the model parameters were optimised using the trial and error
technique for some parameters and the Univariate Algorithm for the rest.
Results show that all applied individual mathematical models provide a reliable
representation of the components of hydrological cycle; however, they provide better
results when they are applied in small catchments. The results arising from both the
lumped and the semi-distributed application of HEC-HMS lead to the conclusion that the
semi-distributed form of the model has an advantage against the lumped form in the case
of simulating complex flood hydrographs and when spatially detailed rainfall is available.
With regard to the calibration of HEC-HMS, for both lumped and semi-distributed
application scenarios, the resulting conclusion is that, for a more objective calibration of
the model, which will support more reliable forecasts, a global multi-objective optimisation
algorithm should be applied.

Keywords: hydrological modelling; semi-distributed models; calibration; HEC-HMS.


1. INTRODUCTION

Hydrological models can be classified into two major types: lumped and distributed.
Lumped models were developed since the 1960s (e.g. the Stanford catchment model,
Crawford and Lindsey, 1966). They consider the catchment as an undivided entity and
use lumped values of input variables and parameters. For the most part (for a review, see
Singh, 1995), they have a conceptual structure based on the interaction between storage
elements representing the different processes with mathematical functions to describe the
fluxes between the storage (e.g. HSPF, Donigian et al., 1995; GR, Perrin et al., 2003).
In the last three decades, lumped models were challenged by distributed models whose
spatial structure allows taking into account the spatial variability of processes within
catchments and consequently the prediction of local hydrological responses for points
within the catchment. In distributed models, parameters need to be defined for every
spatial element and for each process representing equation. In principle, parameter
adjustment should not be necessary for this type of models because parameters should
be related to the physical characteristics of the surface, soil and land-use. However, in
practical applications, calibration procedures are requested for both lumped and
distributed models; consequently the models require effective or equivalent values for
some parameters.
A third smaller category is the so-called semi-distributed models. They represent a
catchment by dividing it into smaller sub-catchments, with uniform as possible
characteristics. They were initially developed to combine the advantages of both lumped
and distributed models. Semi-distributed models are commonly used in the operative
hydrologic forecast services because of their well balanced ratio between the model
spatial accuracy and duration of simulation and calibration effort.
The scope of this paper is to investigate primarily the hydrological response of the
Kalamas river basin in flood events and secondly to illustrate the comparative application
of the model HEC-HMS in lumped and semi-distributed form.


2. STUDY AREA

The study area is part of the Kalamas river basin situated in western Greece. The basin
covers a 1449.3 km
2
area. The length of river from its springs up to the Kioteki streamflow
gauge (outlet) is 108 km. The area is a combination of agricultural land and forests with
mainly calcareous rocks and with mild slopes. Due to the calcareous geological
background, groundwater flow is promoted and plays a decisive role in the maintenance
of baseflow in the river.
The mean annual precipitation volume amounts in 2770 x 10
6
m
3
, while the mean annual
runoff is estimated around 1800 x 10
6
m
3
of water. Climate ranks in the mild
Mediterranean climate, with dry periods to coincide with warm. The average annual
temperature varies between 13.5C and 14.5C for the period 1995-2004. Precipitation is
abundant in the region, with an average annual rainfall of 1550 mm.
The bedrock mainly consists of limestone (47%), loam (14%), flysch (24%) and silt (10%).
The limestone rocks are mainly found in the northern part of the region upstream of the
Soulopoulo while flysch and marl are found in the south-western part of the basin between
Kioteki and Soulopoulo. The steeper slopes occur in the south-western part of the study
area (maximum and mean are 71 and 16% respectively), while the northern part has more
gentle slopes (maximum and mean of 58% and 12% respectively). Elevation ranging from
sea level up to 2157m upstream of Soulopoulo, while the average altitude of the study
area is 544m.

Fig.1: The Kalamas river basin with the flow and rain stations.



3. METHODOLOGY AND DATA AQUISITION

3.1 Methodology

As mentioned in the introduction, the aim of this paper is the analysis of the hydrological
response of the Kalamas basin in flood events using the hydrological model HEC-HMS, in
conjunction with Geographical Information System ArcView 3.2. The catchment analysis
was undertaken for four flood events, in two ways: a) by dividing the study area into two
sub-basins, and b) by dividing the study area into eight sub-basins.
The connection between hydrology and spatial information is accomplished with the HEC-
GeoHMS module of ArcView. The interaction between the HEC-GeoHMS and HEC-HMS,
is shown in Figure 2.
The final result of these processes is the production of two input files for HEC-HMS: the
map file (background map file) and the basin file (basin model file) which as mentioned
above can be lumped or distributed. The map file is a schematic representation of the
sub-basins and watercourses of the basin and the relationships between them. The basin
file contains the entire basins hydrological and geomorphological data as produced
through the HEC-GeoHMS. The remaining data required for the operation of HEC-HMS is
the creation of a meteorological file (meteorological component) containing the rainfall
events and methods for spatial and temporal distribution and the determination of some
specific parameters in mathematical models (hydrograph routing, rainfall losses, etc).


Fig.2: Flood event simulation process flowchart using HEC-HMS and HEC-GeoHMS
(reproduced from M.R. Knebl et al, 2005)

3.2 Data aquisition

HEC-GeoHMS uses as inputs the digital elevation model (DEM), digital maps of soil types
and land uses and some auxiliary files for spatial data in vector format such as the file with
the meteorological and river-stage stations in the basin and the river network for a more
complete and accurate representation of the basin.

Elevation data
The DEM of Kalamas basin produced by digitizing maps of Hellenic Military Geographical
Service (HMGS) in the laboratory of Hydrology and Water Resources, National Technical
University of Athens and has a resolution of 50m, which is good enough for the scale of
the basin.

Land use (LU) and geology data
The digital land use map produced by the Organization of Land and Mapping of Greece
(OLMG) and contains the polygons of land use in the basin of Kalamas, according to the
European coding (program Corrine Land Cover). From the LU map can be estimated the
Manning friction coefficient for the river channel and flood plains, which is necessary in
some routing models. The soil type map of the region contains respectively the polygons
with the different rock types present in the study area. In combination with the land use
map and with the help of tables from the literature, the runoff curve numbers can be
identified for each sub-basin, needed to calculate the runoff loss volume with the SCS
method.

Rainfall data
In this study the rainfall data for the study basin come from four stations under the
authority of Public Power Corporation of Greece (PPC) with temporal step of 1hour. The
raw data of the four rainfall stations were given as cumulative rainfall recorded in tapes of
the rainfall recorders. From the reading of the tapes derived the hourly rainfall in digital
form for input to the model. For the control of rainfall measurements of each station,
comparison of the daily cumulative rainfall was conducted with rainfall measurements of
an adjacent station.
The rainfall stations, while providing a good temporal resolution of rainfall, it is though
punctual. In this study, the method used for the derivation of areal rainfall is the method of
Thiessen polygons. Weighting coefficients for each sub-basin were calculated taking into
account all four stations in the region and also for the three out of four because for two
flood events station Polylofo was not functioning.

River discharge data
The river discharge data are necessary for two main reasons: a) the calibration of the
model and b) the determination of certain parameters required by the model. In this study,
these data are available in river station Kioteki, which coincides with the outlet point of the
whole basin. There is another hydrometric station at Soulopoulo, but the data was not
available for the simulations periods. In these positions are established river-stage
recorders functioning under the PPC authority that record the stage of the river with a time
step of one hour. The locations of the stations are shown in Figure 1.
To calculate the river discharge from the river-stage records requires the stage-discharge
curve at specific locations. For the period 1995 - 2004 which involves the simulations
periods, the stage-discharge curve at station Kioteki has a polynomial form as follow:

Q = 2.7688h
2
+ 11.448h + 2.3206, R
2
=0.963

where Q is the discharge in m
3
/s and h the rivers stage in m.


4. MODEL SELECTION AND APPLICATION

The HEC-HMS model provides the option of a variety of mathematical models to
represent different components of the hydrological cycle, depending on the needs of the
simulation and the quantity and quality of input data. In the present study concerning the
time scale was chosen the event simulation and for the spatial scale was chosen the
lumped simulation, since the study area was divided in two sub-basins in the first scenario
and in eight sub-basins in the second. Figure 3 present the HEC-HMS map file of the
study area for the two scenarios. For both scenarios of spatial distribution, the same
mathematical models and the same calibration procedure was applied.

Fig.3 Schematic of the study area in the lumped and semi-distributed form of HEC-HMS

Rainfall loss model
In this study the curve number CN loss model of the Soil Conservation Service (SCS,
1972) was applied. It is easy to apply and consists of only one parameter (CN), which is a
function of soil type, land use and soil moisture status.
For this study CN values were taken from tables in greek bibliography (Koutsogiannis,
1993). The calculation of the composite curve number for each sub-basin using the
above equation was done with the help of HEC-GeoHMS and digital maps of land use
and soil types of the study area.

Baseflow model
For the representation of the baseflow in this study was chosen to implement the
exponential recession model described by the equation:
t
t
k Q Q =
0

where Q
0
is the initial value of the base flow at time 0, Q
t
is the value of the base flow at
any time t and k is a constant of exponential decline.

Direct runoff model
In this study was chosen to implement the Snyders synthetic unit hydrograph. The
equations that determine the parameters of the UH are:
24
3 3
75 . 2
) ( 75 . 0
3 . 0
p
p
p
p
C m t p
t
T
t
A C
Q
L L C t
+ =

=
=

where C
t
= basin coefficient, L
m
= length of the main stream from the outlet to the divide,
L
c
= length along the main stream from the outlet to a point nearest the watershed
centroid, Q
p
= peak of UH, A = watershed drainage area, C
p
= UH peaking coefficient,
T = duration of UH and t
p
= basin lag.
Channel routing model
The flood routing model applied in this study is the Muskingum model. This method was
chosen because it does not require knowledge of the river sections and the Manning
roughness coefficient and because the two parameters (K, X) it contains can be
determined either by measured flood hydrographs or even through calibration.
The parameter X is called storage coefficient and is a dimensionless parameter that
expresses the attenuation of the flood wave. The parameter K represents the travel time
of the flood wave through the reach.

Model calibration
In this study we applied both calibration techniques: trial and error and automated by
applying an optimization algorithm offered in HEC-HMS. In both cases, the calibration of
the model was based on the measured flows at Kioteki station, using two out of the four
floods events that simulated, with peaks on 12/2/99 and 27/12/96.
The trial and error technique was applied to the parameters C
p
and C
t
of Snyders unit
hydrograph, the parameters K and X of Muskingum routing model and the SCS CN
losses model. The choice of this technique was to obtain the best parameter values for
the model, which would be constant for each sub-basin and storm event. For the rest of
the models parameters, such as the flow at the inflection point of the hydrograph and the
exponential recession constant in the baseflow model and initial runoff losses in the SCS
CN model, applied the automatic calibration for each hydrograph separately, using the
Univariate Gradient optimization algorithm and as an objective function the Sum of
Squared Differences.
The equations of performance indices are shown in Table 1, where q
s
is the estimated
flow, q
o
is the observed flow for the same time, q
m
is the mean observed flow, V
S
is the
total estimated volume and V
O
is the total observed volume.
For indices PE, b and RMSE the perfect calibration occurs when they are equal to zero,
while for the index R
2
occurs when it is equal to unity. The RMSE and R
2
indices are a
measure of error of the volume and peak flow between the observed and estimated
hydrographs while the PE index is a measure of error only to peak flows and the index b
only to the total volume.

Table 1. Models calibration performance indices
Performance index Equation
Root Mean Square Error
( )

=
2 1
o S
q q
N
RMSE

Nash Sutcliffe coefficient
( )
( )

=
2
2
2
1
m o
o S
q q
q q
R

Total volume absolute error
100

=
O
O S
V
V V
b

Peak flow absolute error
100
) (
) ( ) (

=
peak q
peak q peak q
PE
o
o S


5. SIMULATION RESULTS

Figures 4a to 4d present the estimated flood hydrographs at Kioteki station for the four
simulated storm events, after the calibration of the model for both scenarios (lumped and
semi-distributed). Table 2 presents the results of the statistical performance indices as
derived from the application of HEC-HMS under lumped and semi-distributed form for the
four floods.
Table 2. Values of the performance indices after calibration of HEC-HMS under lumped
and semi-distributed form
PERFORM.
INDICES
FLOOD WITH PEAK ON
12/2/99
FLOOD WITH PEAK ON
27/12/96
Semi-distributed Lumped Semi-distributed Lumped
R
2
0.907 0.855 0.909 0.905
RMSE 0.515 0.644 0.392 0.401
PE 12.585 7.294 8.376 7.777
b 1.208 2.817 0.586 1.029
PERFORM.
INDICES
FLOOD WITH PEAK ON
22/12/99
FLOOD WITH PEAK ON
30/3/95
Semi-distributed Lumped Semi-distributed Lumped
R
2
0.788 0.618 0.893 0.893
RMSE 0.918 1.234 0.456 0.456
PE 9.468 6.816 3.664 2.039
b 0.195 0.991 0.179 0.304


a) b)

c) d)
Fig. 4 Results after calibration of HEC-HMS under lumped and semi-distributed form for flood
events with peak flows on a) 30/3/95 b) 22/12/99 c) 27/12/96 and d) 12/2/99


6. CONCLUSIONS

The conclusions from the conducted simulations, referred to the application and
comparison of a lumped and semi-distributed form of the model, are presented below:
A model performance improvement based on the volume absolute error is observed
when using the semi-distributed version of the HEC-HMS in comparison to its lumped
counterpart. Moreover, the same conclusion applies prior to calibration of the model.
The best simulation of the total volume with the semi-distributed version of HEC-HMS,
also leads to the conclusion that the modules for calculating the volume of direct
runoff and base flow volume, are best applicable in small basins.
Referring to the floods on 27/12/96 and 30/3/95, it appears that differences in values
of performance indices between lumped and semi-distributed form of HEC-HMS is
very small. This is partially because the flood hydrographs on 27/12/96 and 30/3/95
show only one peak and partly because of the fact that rainfall data for that flood
events were from three out of four stations in the area. This observation leads to the
conclusion that the application of the semi-distributed form is preferred against the
lumped, only if it is to simulate complex hydrographs and detailed data of spatial
distribution of rainfall is available.
The results of the simulations indicate for both scenarios that while for the three out
of four indices, their value was improved after calibration, for the index absolute error
in peak PE, the value became worse or improved a little. These results may be due
to merits in the calibration procedure. The Univariate Gradient algorithm finds a local
minimum of the objective function, while it depends on the initial values of the
parameters. Moreover the objective function of sum of squared differences
emphasises on improving the indices R
2
and RMSE. We therefore suggest using a
multiobjective optimisation algorithm able to identify the global minimum for a better
calibration and performance of the model.


REFERENCES

1. Bardsley, E., Liu, S., 2003. An approach to creating lumped-parameter rainfallrunoff models
for drainage basins experiencing environmental change. Water Resources Systems
Hydrological Risk, Management and Development (Proceedings of symposium HS02b held
during IUGG2003 at Sapporo, July 2003). IAHS Publ. no. 281.
2. Beven, K.J., Binley, A.M., 1992. The future of distributed models: model calibration and
uncertainty prediction. Hydrological Processes 6, 279-298.
3. Chow, V.T., Maidment, D.R., Mays, L.W., 1988. Applied Hydrology. McGraw-Hill, New York.
4. Diskin, M.H., Simon, E., 1977. A procedure for the selection of objective functions for
hydrologic simulation models. Journal of Hydrology 34, 129-149.
5. HEC, 2000. Hydrologic Modeling System: Technical Reference Manual. US Army Corps of
Engineers Hydrologic Engineering Center, Davis, CA.
6. HEC-GeoHMS, 2003. Geospatial Hydrologic Modeling Extension: Users Manual. US Army
Corps of Engineers Hydrologic Engineering Center, Davis, CA.
7. Knebl, M.R., Yang, Z.L., Hutchison, K., Maidment, D.R., 2005. Regional scale flood modeling
using NEXRAD rainfall, GIS, and HEC-HMS/RAS: a case study for the San Antonio River
Basin, summer 2002 storm event. Journal of Environmental Management 75, 325-336.
8. Madsen, H., Wilson, G., Ammentorp, H.C., 2002. Comparison of different automated
strategies for calibration of rainfall-runoff models. Journal of Hydrology 261, 48-59.
9. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models. Part I a
discussion of principles. Journal of Hydrology 10, 282-290.
10. Nikolay, S., 2004. Distributed hydrological modeling: myth or reality? Postgraduate course,
Department of land and Water Resources Engineering, KTH.
11. Perrin, C., Michel, C., Andreassian, V., 2001. Does a large number of parameters enhance
model performance? Comparative assessment of common catchment model structures on
429 catchments. Journal of Hydrology 242, 275-301.
12. Yapo, P.O., Gupta, V.H., Sorooshian, S., 1996. Automatic calibration of conceptual rainfall-
runoff models: sensitivity to calibration data. Journal of Hydrology 181, 23-48.

Potrebbero piacerti anche