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

net/publication/352820366

Application of HEC-HMS for rainfall-runoff modelling in Mozambique and

France

Conference Paper · June 2021

DOI: 10.3311/FloodRisk2020.18.9

CITATIONS READS

2 215

1 author:

Helder Guta
Université Grenoble Alpes
9 PUBLICATIONS   7 CITATIONS   

SEE PROFILE

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

ANR Project: Concentration and Grain Size Measurements through Acoustic Turbidity View project

ADVP and UB-Lab 2C/3C View project

All content following this page was uploaded by Helder Guta on 29 June 2021.

The user has requested enhancement of the downloaded file.

 Application of HEC-HMS for rainfall-runoff modelling in Mozambique
and France

Helder Guta1,a
1
Grenoble Alpes University

Abstract. In the present study, the suitability of HEC-HMS to simulate event-based rainfall-runoff in
the catchments of Revubue in Mozambique and Ardèche in France was evaluated. These are tributaries of two important
rivers in Africa (Zambezi) and Europe (Rhone). Despite the low mean discharges in both rivers, it is common to observe
floods in certain periods (October – January). The hydrologic models were calibrated and verified based on observed
(historical) precipitation and runoff data. As for many developing countries, availability of reliable hydrologic data
remains an important issue in Mozambique. Indeed, the study highlighted some limitations imposed by the scarcity and
lower resolution of hydrological data in Mozambique. Despite the difficulties related to the unsatisfactory quality of
the data in Revubue basin, it was found that the model performs relatively well in terms of time and magnitude of peak
flows of the studied flood events. Very good results were observed for Ardèche catchment. The good performance
suggest that HEC-HMS may be suitable for flood modelling and/or forecasting in the studied catchments.

1 Introduction to develop hydrologic models for flood forecasting in

many regions of the world. Oleyiblo and Li (2009)
The south of France is well-known to be exposed to validated models for two Chinese catchments. Chu and
flash floods. These events represent one of the most Steinman (2009) applied joint event-based and continuous
destructive natural hazards in the Mediterranean region analysis for a catchment in Michigan, USA. Halwatura and
(Adamovic et al. 2016). On the other hand, the relatively Najim (2013) have validated a rainfall-runoff model for a
long Mozambican coast exposes the country to tropical Sri-Lanka basin, whereas Unduche et. al (2018) concluded
cyclones in the Indian Ocean, often accompanied by heavy that the model was suitable to be applied for flood
rain, leading to floods. Furthermore, the country is at the forecasting in Canada. The models developed herein are
downstream of several large catchments. Therefore, floods event-based, implying that they are suitable for simulating
observed in Mozambique are sometimes caused by heavy rainfall-runoff processes of single events or floods. During
rainfall occurred in the upstream countries. Important such intense rainfall events, other processes such as
efforts to improve flood management and mitigate its evapotranspiration are negligible comparing with rainfall
negative impacts were observed in the past decades. For and surface runoff (USACE 2017). They must however be
example, non-structural measures such as flood accounted if the models are to simulate rainfall-runoff
forecasting systems and flood plain zoning (Hipolito and processes on a continuous base. Event-based models
Vaz 2013) are recognized as important tools in the river require less data, since it includes only the determinant
basin management. The increased use of computer hydrologic processes during floods.
programs allowed improving the modelling capabilities of In France, the rainfall-runoff warning is issued by
many hydrologic processes. In the present study, the Meteo-France, while Vigicrues is a service responsible for
suitability of the open-source program HEC-HMS to flood warning. The Regional Directions of Environment,
perform simulations of rainfall-runoff processes in Planning and Habitation (DREAL) include Services for
the catchments of Revubue (sub-basin of Zambezi river in Flood Forecasting (Service de Prevision de Crues - SPC)
Mozambique) and Ardèche (sub-basin of Rhone river in that operate flood forecasting models. The forecasting
France) rivers was tested. The Rainfall-runoff models can performed by these services is shared through Vigicrues in
be useful for a wide range of applications, such as real-time. The SPC-Grand Delta (SPC-GD), attached to
hydrologic design of hydraulic structures, flood DREAL Auvergne-Rhone-Alpes uses rainfall-runoff
forecasting, etc. models for flood forecasting in Ardèche catchment. The
HEC-HMS was developed by the Hydrologic level of accuracy of an existing model is used as reference
Engineering Center, from the US Army Corps of for the present model developed in HEC-HMS.
Engineers (USACE). Over the past years, it has been used
a
Corresponding author: [email protected]
DOI 10.3311/xxx
 Budapest University of Technology and Economics (BME)

The main goal of this study is to apply the program and source in Massif Central and flows into the Rhône river
methodology used to validate the Ardèche model to the near Pont-Saint-Esprit (Adamovic 2014). At Vogue, the
Revubue catchment in Mozambique, where no catchment area is 623 km2. The hydrological year consists
hydrological model has ever been developed. Only a part mainly of two periods. There is a rainy season (September-
of the Ardèche catchment will be studied (Ardèche at February) with maximum precipitation intensity in autumn
Vogue), where effects of dams is negligible for moderate and high discharge. During the dry season (March-
and large floods (SPC-GD 2013). There are no dams in August), the runoff is low (Adamovic et al. 2016).
Revubue River, indicating that in both cases the rainfall-
runoff process is potentially independent of external
factors (other than the main hydrologic processes such
infiltration, initial retention, etc.).
In section 2, a short description of the study areas and
the data used for the analysis will be described. The
methodology to perform flood modelling based on unit
hydrograph will be presented in section 3. Section 3 will
present the results and discussion, followed by conclusions
in section whereas section in 4.

2 Study areas and hydrological data .

2.1 Study areas

Revubue River is one of the main tributaries of the

Zambezi River in Mozambique and its basin is located
entirely in Mozambique. The two main gauges in the
Revubue river are E-367 Maue-Angonia (upstream) and E-
302 Tete (downstream). The distance between these
stations is 250 km, roughly the length of the river itself.
Figure 1 illustrates the Zambezi basin in Mozambique,
with focus on Revubue sub-basin. According to a report by Figure 2 Map of Ardèche catchment (adapted from
Coba et al. (2017), the average instantaneous discharge is Adavomic et al. 2016)
about 30 m3/s. During floods, the flow rises to over 1000
m3/s. Based on the same report, the Rio Revubue can be 2.2 Hydrological data
described as being composed of 3 sub-basins: Higher
Revubue, Lower Revubue Revubue and Condezi. The For Revubue basin, it is assumed that some of the
black line in Figure 1 illustrate the repartition of the most reliable stations are those currently used for the
catchment. monitoring inundation risks in Revubue basin. These are
P333 at Maue in the upstream region and P438 at Tete, at
the outlet. They are included in the monitoring reports
“boletim hidrologico”. However, more stations, at the
upper Revubue should be included in order to retrieve an
average rainfall by the Thiessen method. The included
stations in each sub-basin are illustrated in Figure
1Erreur ! Source du renvoi introuvable.. Unfortunately,
it was not possible to properly cover the entire catchment
due to lack of reliable data. The data is collected by the
Regional Water Administration of Zambeze (ARA-
Zambeze) on daily basis. The only station with reliable
flow data is the station downstream. Upstream stations
data consists only of level measurements and no rating
curves are available. This imposes some important
limitations. Ideally, the modelling should comprise a
rainfall-runoff analysis for each subbasin, followed by a
flood routing down the basin. With only one reliable flow
Figure 1. Map of Revubue catchment (adapted from Coba et gauge this can not be performed. The catchment will be
al. 2017); yellow points are correspond to locations of divided only in three subbasin, but the flow generated in
raingauges and flow gauges. each of them cannot be verified. Only the total runoff at
the outlet can be analyzed in terms of observed and model
The Ardèche river has a length of 125 km. It is one of results.
the major tributaries of the Rhône river that drains into the For Ardèche basin, three stations were used:
Mediterranean Sea near Arles, southern France. It has its Antraigues, Aubenas and Barnas. They were selected in
 FLOODrisk 2020 – 4th European Conference on Flood Risk Management

order to cover the study region in roughly equal loss method parameter, since it allows to estimate the
proportion. The used data was obtained from Meteo- hyetograph that corresponds approximately to the
France, with a time step of 3h. observed direct runoff. Two of the most widely used
It is crucial to confirm that the rainfall data is consistent methods were tested, namely, SCS-Curve Number (SCS-
with the occurrence of the flood events. This simple CN) and the Initial and Constant Rate.
verification may illustrate potential inconsistencies in the The strength of the model is because it includes
data used to build the model. It suffices to check, for the cumulative rainfall, soil type, land use and antecedent soil
chosen events, that there is indeed an increase in rainfall in moisture (Maidment 1993). The effective rainfall is given
the selected stations, before and/or during the flood events. by:
This verification was used to exclude some several events
from the analysis. This was mostly the case for Revubue, (𝑃−𝐼𝑎 )2
𝑃𝑒 = (1)
but few events in Ardèche also displayed inconsistencies 𝑃−𝐼𝑎 +𝑆
between the rainfall and runoff data. This suggests that
even in catchments well equipped and monitored, it is where Pe is the precipitation excess at time t, P is
difficult to avoid data errors completely. In the program, accumulated rainfall depth at time t, Ia is the initial
the rainfall data will be used to define a meteorological abstraction (initial loss), often described as 𝐼𝑎 = 0.2𝑆 and
model. In the present case, the meteorological model is 𝑆 = (25400 − 254𝐶𝑁)/𝐶𝑁 is the potential maximum
heavily simplified, since only precipitation data is retention, which describes the ability of the watershed to
required. Weighting factors should be assigned to each abstract and retain storm precipitation. The curve number
rain gauge in order to determine the a uniform precipitation CN varies between 30 (soils with high infiltration rates) to
depth over the catchment. Herein, they were estimated 100 (water bodies/impermeable surfaces). Some
from Thiessen polygons. Additionally, in HEC-HMS the assumptions have been adapted to local characteristics. For
user can specify the temporal distribution of that uniform example, Subramanya (2008) refers that the Indian
precipitation. experience indicates that 0.1𝑆 < 𝐼𝑎 ≈ 0.3𝑆. Thus, this is a
parameter subject to calibration if data is available.
The initial and constant rate method is a simpler
Sub- Area method, but has shown to be applicable in wide range of
Rain gauges
basins (km2) conditions (USACE 2017). The concept of the method is
Alto P-333 P-434 P- 43 that the maximum potential rate of precipitation loss (fc)
10241
Revubue Angonia Mitengo Zobue remains constant during a flood event USACE (2000).
Baixo Additionally, until the accumulated precipitation exceeds
2495 P-438 TETE
Revubue the initial loss volume, no runoff occurs. The effective
Condezi 3559 P-902 Tsangano
rainfall is determined by:
Total 16.295
Table 1. Rain gauges in Revubue basin
0 𝑖𝑓 𝑃𝑖 < 𝐼𝑎
𝑝𝑒 = { 0 𝑖𝑓 𝑃𝑖 > 𝐼𝑎 𝑎𝑛𝑑 𝑝𝑡 < 𝑓𝑐
3 Flood modelling based on Unit 𝑝𝑡 − 𝑓𝑐 𝑖𝑓 𝑃𝑖 > 𝐼𝑎 𝑎𝑛𝑑 𝑝𝑡 < 𝑓𝑐
Hydrograph theory (2)

3.1 Loss function – SCS CN/Initial and loss where pe is the excess rainfall during a time interval
method (time step), pt is the total precipitation during the time
interval and Pi is the accumulated rainfall at a given time.
In order to estimate the effective rainfall, the base flow Note that the rainfall intensity is directly taken into account
should be separated from direct runoff. The effective in this method, contrarily to SCS method. As pointed by
rainfall volume is defined as approximately equal to the USACE (2000), the method predicts same loss for 25 mm
volume of direct runoff. Once the effective rainfall volume rainfall in 1 hour or 1 day.
is determined, it should be distributed in time, using the Given the advantages and limitations of each method,
selected loss method. the two methods were tested for both catchments. Overall,
First, it is necessary to choose a method for base flow the SCS-CN seems to provide better results for Revubue,
separation. Here, the fixed base method according to Chow whilst the initial-constant method seems to perform
et al. (1988) is considered. In this method, the based flow slightly better for Ardèche
in the beginning of the event is extended up to the time of
peak. From then, it is connected to the inflection point, 3.2 Transform function – SCS Unit
where recession begins. This point can be identified by Hydrograph
plotting the logarithmic values of the discharge. In the
recession stage, a linear trend is observed Shaw (1994). The unit hydrograph (UH) concept, introduced in the
The runoff hydrograph can be integrated to retrieve the United States of America by Sherman in 1932, was a major
volume of effective rainfall. The next step is to distribute step forward in hydrological analysis (Shaw 1994). He
the estimated effective runoff volume at each time step. defined it as the hydrograph of direct runoff resulting from
This requires selecting a loss method. Note that this unit depth (often 1 cm) of effective rainfall falling in a
approach should serve as a prelaminar determination of the given duration, such as 1 hour or 1 day, generated
 Budapest University of Technology and Economics (BME)

uniformly in time and space over the catchment area.

From the definition, the UH is always associated with a
given duration. This duration depends on the excess
rainfall duration used to derive it. From the assumptions of
the method, the duration should be long enough for the
entire catchment to respond, but it should not be too long,
as the assumptions of the method become unrealistic. In
the case of single storms, the determination of UH is
straightforward. For multi-period storms, a deconvolution
method should be applied to retrieve the UH. Because
Figure 3 Computed 1h SCS-UH in Ardèche
single-period storms are rare and deconvolution methods
are instable, synthetic hydrographs are widely used in
practical situations. The availability of data allows the
estimations of the parameters in a realistic manner. The
SCS UH is a (synthetic) dimensionless unit hydrograph,
defined by one parameter, the time of rise or time of peak
(Tp).
𝐶𝐴
𝑞𝑝 = (3)
𝑇𝑝
where A is the area (km2), C = 2.08, 𝑇𝑝 = 𝑡𝑟 /2 + 𝑡𝑝 ,
with 𝑡𝑝 = 0.6𝑇𝑐 . Tc is time of concentration and tr the
duration of the storm. In HEC-HMS, the UH is
systematically defined based on the duration equivalent to
the time step of the calculations. In the present case, as the
calculations will have 1h time step, the software computes Figure 4. Application of SCS-UH to Sep2000 flood event in
the 1h-UH, according to the specified lag time tp. Instead Ardèche (Observed- -; Estimated-)
of defining it based on time of concentration, we can
evaluate it from the rainfall and runoff data. Analysis of
several flood events have shown that it varies around tp = Other parameters to be defined in HEC-HMS model
4 - 6 h for Ardèche and 32 - 40h for Revubue. The adopted are the recession constant k=0.8 (0.9) and the ratio to peak
duration (after the calibration) for Ardèche (Revubue) was Rtp=0.5(0.4). Because these parameters do not change
tp=350 (2000) min, corresponding to 5 (33.3) h. This value dramatically, they can be considered as quasi-constant
seems to be adequate for a wide range of analyzed events. model parameters, contrarily to the parameters of the loss
Calculations of 1h SCS-UH can be performed in a simple method, which varies as function of the antecedent
excel sheet. This was tested for Ardèche catchment, where conditions (initial storage, saturation of soils, etc.).
higher data resolution is available. As this is only a
preliminary estimate, data from only one rain gauge can be 4 Results and discussion
used. Figure 3 shows the determined 1h SCS-UH using a
simple excel sheet, for the flood event of September 2000.
The loss method was the initial and constant. Figure 4 The results of calibration and verification of the model
shows the application of the determined SCS-UH for the are presented in this section. The SCS-CN method
same event. The good agreement of the model and preconizes a change in initial conditions based on the value
observed direct runoff indicate that the preliminary of CN. The initial abstraction and the continuous
estimation of the model parameters is relatively accurate. infiltration are all function of CN. By initial and constant
They can serve as initial parameters in the HEC-HMS rate method, the initial loss was taken as the main
model, and modified to improve the agreement with the calibration parameter, as function of antecedent soil
observed hydrographs. This corresponds to the calibration moisture. However, the loss rate seems to be heavily
process. However, this analysis was carried only for affected by antecedent conditions also. Thus, in the
Ardèche, where high temporal resolution data was calibration processes, the parameters to be determined are
available. For example, for this preliminary analysis of the mainly the CN value and both initial and constant rate
1h SCS-UH, 1h rainfall and runoff data was used. parameters. The results of the calibration process are
Therefore, this would be difficult if not impossible in displayed in the following Tables and Figures.
Revubue basin, where only 24h rainfall data is available.

Revubue
CN Ia Initial Loss NSE Lag Recession Rtp
(mm) Const. Rate (min) Const.
13-Feb-14 90 0 0 0.2 0.78
12-Jan-15 90 15 27 0.2 0.66
17-Jan-15 82 15 22 0.6 0.65 20001 0.9 0.4
22-Feb-15 45 30 40 1.7 0.61 4002
2-Feb-17 65 15 15 0.65 0.22 3003
 FLOODrisk 2020 – 4th European Conference on Flood Risk Management

24-Mar-17 95 27 33 0.01 0.87

parameter may be the discharge at the beginning of the
event (Qinitial). For example, for a dry condition, it may be
Ardèche expected lower values of P7 and/or Qinitial. Although not
CN Ia Initial Loss NSE Lag Recession Rtp shown here, it was seen that small values of P 7 were
Time
(mm) Const. Rate (min) Const. correlated with lower values of CN or higher loss rate.
28-Sep-00 52 45 70 6.5 0.94 Overall, this trend was confirmed in both basins.
12-Nov-00 65 27 65 2 0.85
24-Dec-00 53 43 80 3 0.87
18-Oct-01 55 35 80 3 0.98 350 0.7 0.5
27-Oct-04 65 27 55 2.8 0.95
3-Nov-04 65 27 55 3.5 0.89
Table 2. Calibration events (1Upper Revubue; 2Condezi;
3Lower Revubue)

Revubue
Ia Initial Loss NSE Lag Recession
Time CN Rtp
(mm) Const. Rate (min) Const.
9-Mar-06 60 15 30 1.7 0.55
13-Feb-07 70 10 20 1.2 0.46 20001
4-Jan-08 70 10 28 1 0.70 4002 0.9 0.5
2-Apr-12 90 3 5 0.3 0.18 3003
27-Jan-12 85 3 5 0.6 0.71
Ardèche
Ia Initial Loss NSE Lag Recession
Time CN Rtp
(mm) Const. Rate (min) Const.
21-Oct-08 97 0 1 0.2 0.82
01-Nov-08 99 10 10 0.0 0.90
12-Oct-14 90 0 0 1.2 0.92
23-Nov-18 99 0 2 0 0.71 350 0.7 0.5
Figure 5. Calibration - Ardèche
22-Nov-19 99 4 6 0 0.94
19-Dec-19 99 4 7 0 0.93
Table 3. Verification events

A relatively good agreement between the simulated

and observed hydrographs is observed. This agreement is
quantified using the Nash-Sutcliffe efficiency (NSE)
criterion and the relative difference of the flood volume.
NSE values closer to 1 indicate better agreement of the
simulated hydrograph to the observed. They are also
displayed in the table of calibration and verification
results. In Ardèche catchment, the average value of the
Nash criterion taking 83% of the events with the best
agreement, corresponds to about 0.92, indicating a very
good performance. The model calibration from SPC-GD
displays an accuracy level of the same order. They
presented a NSE greater than 0.7 for 75% of the cases. One
general difficulty is to simulate floods with more than one
peak. This was observed also in the simulations from SPC-
GD. Results of two simulations from SPC-GD is shown
The performance is more modest for Revubue catchment,
where the average value of NSE of the 82% of the events
Figure 6. Calibration - Revubue
with the best agreement is about 0.66. However, given the
tremendous limitations linked to data resolution and
reliability, this result is considered satisfactory.
If the model is to be applied for forecasting floods, one
of the most important issues concerns the initial moisture
condition. The parameters determined by initial moisture
conditions can be analyzed in order to observe their
pattern. An important indicator of initial moisture
conditions is the accumulated rainfall over the past few
days Shaw (1994), for example previous 7 days (P 7). To
increase the reliability of this initial guess, one additional
 Budapest University of Technology and Economics (BME)

For Ardèche catchment, the modelled peak discharge

and its time of occurrence displayed a good agreement
with the observed data. The values of NSE is above 0.85
for most events evaluated. Overall, the model provides the
same level of accuracy as the model used by the authorities
responsible for flood monitoring in the region.
For Revubue catchment, the values of NSE remain
above 0.6 for most events. Additionally, the relative
difference in terms of flood volume remained below 15%.
Although the agreement is lower than in Ardèche, given
the data limitations, it is considered relatively good result.
The relatively large lag time may be a determining factor
allowing to obtain acceptable results with such low
resolution data.
The good performance of the developed models
suggest that HEC-HMS may be suitable for flood
modelling and/or forecasting in both catchments.
Particularly for case of Mozambique, the results suggest
that improvements in the data collection may significantly
enable the operation of flood forecasting systems.
Figure 7. Verification - Ardèche
6 References
1. Adamovic, M. (2014). Development of a data-driven
distributed hydrological model for regional scale
catchments prone to Mediterranean flash floods.
Application to the Ardèche catchment, France
(Doctoral dissertation, Grenoble).
2. Adamovic, M. Branger, F. Braud, I. Kralisch, S.
(2016). Development of a data-driven semi-
distributed hydrological model for regional scale
catchments prone to Mediterranean flash floods.
Journal of Hydrology, 541, pp. 173-189.
3. ARA-Zambeze. Avaliação das Cheias na Sub-bacia
do Revúbue 2018-2019.
4. Coba, Consultec, Salomon. (2017). Elaboracao do
Plano Estrategico para utilizacao e desenvolvimento
da bacia do rio Zambeze. Volume 1.
5. Chow, V., Maidment, D.R, Mays, L.W. 1988. Applied
Hydrology. McGraw-Hill, New York.
6. Chu, X. and Steinman, A., 2009. Event and
continuous hydrologic modeling with HEC-HMS.
Figure 8. Verification - Revubue Journal of Irrigation and Drainage Engineering,
135(1), pp.119-124.
5 Conclusions 7. Halwatura D. Najim M. (2013). Application of the
HEC-HMS model for runoff simulation in a tropical.
In the present study, event-based hydrological Environmental Modelling. & Software. Volume 46:
models of Revubue catchment in Mozambique and part of pp. 55-162.
Ardèche catchment in France was presented. The models 8. Hipólito ,J. R and Vaz, A.C. (2012) Hidrologia e
were built with the program HEC-HMS. The hydrological Recursos Hídricos, IST Press.
data was collected from ARA-Zambeze in Mozambique 9. Maidment, D.R. (1993). Handbook of hydrology.
(24h rainfall and 6h discharge), whereas for Ardèche McGraw-Hill.
catchment, higher resolution data was obtained from 10. Oleyiblo J., Li Z. (2009). Application of HEC-HMS
Meteo-France (3h rainfall) and HYDRO data base (1h for flood forecasting in Misai and Wan’an catchments
discharge). It was seen that most of the model parameters in China. Water Science and Engineering. 3(1):14-22.
were determined from the analysis of rainfall and 11. Subramanya, K. (2013) Engineering hydrology, 4e.
discharge data of the catchments. This preliminary Tata McGraw-Hill Education.
analysis was carried out using simple excel sheets. This 12. Shaw, E. M. (1994). Hydrology in practice, VNR Int.
allows evaluating in advance the acceptable range of 13. Service de Prévision de Crues. (2013). Règlement De
values for the main model parameters. Surveillance de Prévision et de Transmission De
L'information Sur Les Crues.
 FLOODrisk 2020 – 4th European Conference on Flood Risk Management

14. Unduche, F., Tolossa, H., Senbeta, D., & Zhu, E.

(2018). Evaluation of four hydrological models for
operational flood forecasting in a Canadian Prairie
watershed. Hydrological Sciences Journal, 63(8),
1133-1149.
15. US. Army Corps of Engineers (USACE). (2000).
Hydrologic Modeling System: Technical Reference
Manual. Davis, CA: U.S. Army Corps of Engineers,
Hydrologic Engineering Center.
16. U.S. Army Corps of Engineers (USACE). (2017).
Hydrologic Modeling System: Applications Guide.
Davis, CA: U.S. Army Corps of Engineers,
Hydrologic Engineering Center.

View publication stats