FLOOD ROUTING WITH REAL-TIME METHOD FOR FLASH FLOOD FORECASTING IN THE PLAIN BOU SALEM

International Journal of Technical Research and Applications e-ISSN: 2320-8163,
www.ijtra.com Volume 3, Issue 1 (Jan-Feb 2015), PP. 75-79
75 | P a g e
FLOOD ROUTING WITH REAL-TIME METHOD
FOR FLASH FLOOD FORECASTING IN THE
PLAIN BOU SALEM
Medjerda River in Tunisia
Abidi Sahar and Hajji Olfa
PhD student, Rural Engineering; Water and Forest
National Agronomy Institute-Tunis INAT
Tunis, Tunisia
sahar.abidi@yahoo.fr – olfa.hajji@yahoo.fr
Habaieb Hamadi
General Director
National research Institute of rural engineering, Water and Forests (INRGREF) Ariana, Tunisia
habaieb.hamadi@yahoo.fr
Abstract— Flooding problem raised seriously in the watershed
of Medjerda indeed flood risk factors still exists for some cities.
Studies forecasting and flood management may be important to
address these problems. The plain of Bou Salem had long known
catastrophic floods. Sudden rain, releases of dams and tributaries
flows caused historic flooding at the level of this plain. We
recovered thirty floods in the station of Bou Salem during the
period 1973-2013. Among the thirty floods, we distinguish mainly
three Flash floods. In fact, Flash flood is a short and sudden local
flood with great volume, it has a limited duration which follows
within few (usually less than six) hours of heavy or excessive
rainfall, rapid rain, or after a sudden release of water from a
dam. This communication is designed to analyze the results of the
flash floods forecasting by simple propagation models namely
Muskingum and Regression. The method of forecasting depends
on the upstream station flow and models coefficients of
antecedent floods. Forecast periods range from 2 to 8 hours, with
a pitch of 2 hours. We used numerical criteria, such as Nash
coefficient, peak relative error and time separating observed and
calculated pic, to evaluate the results. We noted that the
satisfaction of all criteria together is not touched. The results
were satisfactory with Nash coefficient ranging from 71% to
99.8%.
Index Terms— Flash flood, plain of Bou Salem, forecasting,
Muskingum, Regression.
I. INTRODUCTION
When it rains in a catchment area, all the difficulty lies in
the definition of rainwater division between its various possible
destinations (evaporation, infiltration or streaming…) and in
the definition of the concerned physical processes to realize
these tasks.
The Flash flood are definite by [1] as being suddenly
appear, often not easily foreseeable, fast climb time and had
relatively important specific discharge. These floods are
usually associated with intense rainfall events and occur in
basins of moderate size. The tributary flows and dams’ releases
aggravate the situation. This is the case of the plain of Bou
Salem which became a flood stage. It is crossed by the river
Medjerda receiving water from Jendouba station, tributaries
Mellegue, Tessa and Bouheurtma. Plain Bou Salem has
experienced devastating floods that have caused serious human
and material damage.
II. STUDY AREA
Medjerda is the most important river of Tunisia by the size
of its watershed (23,500 km2
) and the importance of annual
contributions which represent on average a third of the surface
water resources of the country (915 million m3
in average).
Plain Bou Salem is located in northern Tunisia and high
valley of Medjerda (Fig 1). It is limited by the hydrological
station of Jendouba (upstream) and Bou Salem (downstream).
The plain is crossed from west to east by the Medjerda River
over a length of 10 km and a catchment area of 6808 km2
.
Fig. 1 Localization of Bou Salem plain in the high valley of
Medjerda
The main tributaries which flow into Medjerda River at the
plain of Bou Salem are (Fig. 2):
At the left side:
- Oued Bouhertma: it converges near Bou Salem after a course
of 64 km, it drains a catchment area of 390 km².
At the right bank
- Oued Mellègue: it covers a distance of 317 km before
converging just after Jendouba, it drains a catchment area of
4,497 km².
- Oued Tessa: It flows into Medjerda River just after
Mellegue. It covers a distance of 143 km, it drains a watershed
of 2,410 km².

76 | P a g e
Fig. 2. Tributaries of Bou Salem plain
A. Climate context
At Medjerda, the average annual rainfall varies from 1,000
mm in the northern part and the southern part receives only 300
mm. Precipitation plain Bou Salem (Fig. 3) varies between 240
and 700 mm with an average of 500 mm. Rainfall is also very
irregular from one year to another. The average temperature in
the basin varies between 38° C in July and August and less than
6° C in January.
Fig. 3. Annual average precipitation of Bou Salem
B. Historic flooding in Bou Salem
We recovered thirty flood of the station Bou Salem during
the period 1973-2013. These floods occur mainly during the
spring and winter seasons. From these floods, we noted the
existence of five major floods where the runoff volume for a
few days can reach the average annual volume. The following
table summarizes some characteristics of main floods:
TABLE I. HISTORIC FLOODING
Flood Peak
flow
(m3
/s)
Precipitation
(mm)
Rain
duration
(day)
Specific
flow
(l/s/km²)
March
1973
3220 123 6 473
Mai 2000 977 67 4 143
January
2003
1020 187 3 150
November
2004
529 83 4 77
April 2009 550 55 2 81
February
2012
882 50 2 129
The flood of March 1973 is characterized by a single high
peak inflow and considerable rainfall in the entire basin during
6 days. The volume spilled from Mellegue dam, Jendouba
basin and Tessa tributary reached 155, 259 and 75 million m3
respectively.
The flood of Mai 2000 had a high speed entering with a
single point, and localized rainfall. Precipitation focused on the
sub-basins Tess and Mellegue. The volume spilled reached
from Mellegue dam 93 106
million m3
and Jendouba basin 56
million m3
.
The flood of January 2003 had unique inflow and sudden
rainfall peak. The dam releases Mellegue (82 million m3
) and
Bouheurtma (36 million m3
) and contribution of Jendouba
basin (108 million m3
) and Tessa tributary (18 million m3
)
participated in the aggravation of this flood.
The flood of November 2004 had a flat unique inflow. The
precipitation was located at Mellegue dam with moderate
intensities for 4 days. The dam releases Mellegue (3.6 million
m3
) and Bouheurtma (12 million m3
) and contribution of Tessa
tributary (5 million m3
) and Jendouba basin (42 million m3
)
caused this flood.
The flood of April 2009 caused by sudden and significant
rainfall, releases from the dams of Mellegue (14 million m3
)
and Bouheurtma (69 million m3
) and contribution of Tessa (15
million m3
and Jendouba basin (54 million m3
).
The flood of February 2012 provoked by sudden and
significant rainfall, releases from Bouheurtma dams (68 million
m3
) and contribution of Jendouba watershed (196 million m3
).
C. Choice of flash floods in Bou Salem
Several definitions of flash floods have been found in the
literature. Table 2 presented the selection of flash floods in Bou
Salem according to the characteristics defined by [2].
Based on this characteristic we identified three flash floods:
the one of January 2003, April 2009 and February 2012. In the
next chapter, we will precede to the prediction of these three
floods with the models of Muskingum and Regression.
TABLE II. CHOICE OF FLASH FLOODS IN BOU SALEM
Flash flood
characteristic
1973 2000 2003 2004 2009 2012
Sudden onset and
evolution
(rapid hydrological
response, rise time of
rapid flood, violence)
X X X
Torrential rains that
are origin
X X X X
Importance des débits
dans les rivières
X X X X X X
Local inundation
(geographically)
X X X
Difficulty of
predicting the flow
and possibly damage
they generate
X X X X X
III. APPLIED METHODS
A. Forecasting methods
The method used in this study for flow forecasting is based
on the coefficients resulting from the reconstruction of flood
hydrographs with propagation models. These coefficients will
be taken, for each method and for each period of prediction,
from a previous flood haven same degree of humidity and same
season like the flood to be predicted. Prediction delay varies
from 2 to 8 hours with a pitch of 2 hours.
B. Reconstitution methods
The reconstitution is an operation consisted to calculate the
simulation coefficients of each flood, by each method and for
each calculation period. In this study, we will use the results of
reconstitution of 26 flood of Bou Salem station in Medjerda
River [3].
Two models were used to predict the flash flood of Bou Salem:
- Muskingum model,
- Regression model.
1) Muskingum model
Since its development in 1939 [4], this model is widely
used in hydrological engineering. Cunge [5] showed that the
Muskingum model is numerically equivalent to the Saint-
Venant equations via the diffusion equation of a wave.
Muskingum model proposes a relationship between the inflow
Qa (t) and outflow Qs (t) of type [6]:

77 | P a g e
)()()()( 321
tQadtQatQadtQ saas
 (1)
Where ‘Qa’ and ‘Qs’ are the upstream and downstream
flow (expressed in cubic SI), ‘t’ is the calculation time, ‘d’
represent the calculation delay and ‘a1, a2, a3’ are the
coefficients of Muskingum model can be calculated by least
squares‘ method.
2) Regression methods
Through two downstream data and information well before
the upstream taken at time t-τp, we can write [6]:
)()()()( 321 dtQbtQbdtQbdtQ sspas   (2)
Here ‘τp’ is the propagation times between the two upstream
and downstream stations and ‘b1, b2, b3’ are the regression
model coefficients can be calculated by the least squares
method.
We note that the upstream flow has both Jendouba and Tessa
flows and outgoing Mellegue and Bouheurtma dams.
C. Performance Measures
Graphic criteria used to optimize the results are observed and
simulated hydrographs, Error and correlations between
observed and calculated rates.
Numeric criteria chosen to test the effectiveness of the models
are:
The peak relative error (PRE):
maxcal
maxobsmaxcal
Q
QQ
PRE

 (3)
Where ‘Qcalmax’ is the maximum rate calculated and ‘Qobsmax’
represent the maximum observed flow.
Nash coefficient (Nash):






 n
1n
2
n
1i
2
)QmQoi(
)QciQoi(
1CNash (4)
Where Qoi is the observed flow, Qci is the calculated flow and
Qm is the average observed flow.
The peak time error: difference between the time of calculated
‘tQcal’ and observed ‘tQobs’ peak:
QobsQcal ttPTE  (5)
IV. RESULTS & DISCUSSIONS
We present in this part the results of each flood forecasting.
A. Flash flood of 11 January 2003
For predicting the flood, we used the coefficients of the
flood 13/12/1990.
We summarize in Table 3 the values of various evaluation
criteria.
TABLE III. FLOOD FORECASTING OF 11 JANUARY 2003
Muskingum model Regression model
Prediction
delay (h)
PRE
(%)
Nash
(%)
PTE
(h)
PRE
(%)
Nash
(%)
PTE
(h)
2 1.1 99.6 0 2 99.8 0
4 6 98.6 4 6 98 -3
6 12 97 4 16 94 -2
8 14 95 4 26 87 0
After this application, we note that:
- The peak relative error (PRE) is lower for the period of
two hours for both models and it increases with time forecast. It
is lower for Muskingum model.
- The Nash decreases with increasing delay of prediction
for both models.
- The peak time error (PTE) is the same for a period of two
hours for both models and it is lower for other periods for the
regression model.
We can conclude that the satisfaction of the three numerical
criteria at once is not possible to predict the flood January
2003. Nash is more important with the regression model for the
period of two hours with the model Muskingum for other
periods.
We present the results of forecasting the flood in January
2003 with the regression model for a period of two hours.
Fig. 4. Flow hydrograph forecasting of 11/01/2003 flood by
Regression model in 2 hours
Fig. 5. Prediction error of 11/01/2003 flood by Regression
model in 2 hours
Fig. 6. Flow correlation of 11/01/2003 flood by Regression
model in 2 hours
The prediction results of flood 11/01/2003 in 2 hours are
satisfactory. In fact, the shape of the hydrograph is reproduced
and the observed one is superposed with the calculated. The
maximum flow is reproduced by its shape, its value with a

78 | P a g e
small error (2%) and its time. The flow error varied from -42
to 55 m3
/s. The Nash coefficient is high, equal to 99.8%.
The prediction results of flood 11/01/2003 in 2 hours are
satisfactory. In fact, the shape of the hydrograph is reproduced
and the observed one is superposed with the calculated. The
maximum flow is reproduced by its shape, its value with a
small error (2%) and its time. The flow error varied from -42 to
55 m3
/s. The Nash coefficient is high, equal to 99.8%.
B. Flash flood of 16 April 2009
For predicting this flood, we used the coefficients of the
flood 25/05/2000. The results are summarized in the following
table:
TABLE IV. FLOOD FORECASTING OF 16 APRIL 2009
Predictio
n delay
(h)
PR
E
(%)
Nash
(%)
PTE
(h)
PRE
(%)
Nash
(%)
PTE
(h)
2 -0.1 97.5 2 2.7 99.6 -2
4 -0.2 91 4 4.3 97 0
6 -0.5 81 6 5 89 2
8 -1.2 71 8 4.2 80 4
The values of the performance criteria are variable from
one model to another. We do not have complete satisfaction for
all criteria. For the peak relative error (PRE), the Muskingum
model gives the best results. For the Nash coefficient and the
peak time error (PTE), the regression model prevails.
For both models and for all delay, Nash varies between 71
and 99.6%, which proves the good prediction of the flood of
April 2009 at Bou Salem.
The following figures show the results of forecasting flood
of April 2009 with the Muskingum model with a delay of 4
hours.
Muskingum model in 4 hours
Fig. 8. Prediction error of 22/04/2009 flood by Muskingum
model in 4 hours
Fig. 9. Flow correlation of 22/04/2009 flood by Muskingum
model in 4 hours
The model reproduces the hydrograph shape: the rising
limb is underestimated and the receding limb is overstated. The
peak is reproduced by its form and value with a very low error
of -0.1% and it appeared ahead two hours. The variation of the
flow error varied between -175 m3/ s and 50 m3
/ s.
C. Flash flood of 22 February 2012
By coefficients calculated during the reconstruction of the
flood 01/02/2003, we made the prediction of the flood of
February 2012 using both models Muskingum and Regression.
Performance criteria are summarized in Table 5.
TABLE V. FLOOD FORECASTING OF 22 FEBRUARY 2012
Predicti
on
delay
(h)
PRE
(%)
Nash
(%)
PTE
(h)
PR
E
(%)
Nash
(%)
PTE
(h)
2 -0.9 98 2 0.7 99 2
4 -1.9 96 4 1.3 97 1
6 -3.1 92 6 1.8 93 0
8 -4.5 88 8 9.6 82 2
This table shows also the relative error of the peak (PRE) is
smaller in absolute value for the regression model. For both
models, this error increases with delay prediction. For both
models, the value of Nash is inversely proportional to the time
of forecast. The peak time error (PTE) is lower for the
regression model for all forecasting delay.
To analyze the results of forecasting flood 22/02/2012, we
have chosen the results given by Regression model for delay of
6 hours:

79 | P a g e
Regression model in 6 hours
Fig. 11. Prediction error of 22/02/2012 flood by Regression
model in 6 hours
Fig. 12. Flow correlation of 22/02/2012 flood by Regression
model in 6 hours
Predicting the flood hydrograph of 22/02/2012 to six hours
with the regression model gave birth to a second peak in
advance of the observed peak. Raising limb is underestimated
and the receding limb is gently overestimated. The peak is
reproduced by its form and the value is overestimated by
1.8%. The error on the flow varies entre 2.5 m3/s and 141
m3/s. the cloud points given by the flow correlation are close
to the first bisector.
V. CONCLUSION
The characteristic location of Bou Salem, which lies in a low
plain where the confluence of Medjerda tributaries (Mellègue,
Tessa and Bouheurtma ) provoked a braking fairly important
to the flow of Medjerda toward the plain.
From historical floods Bou Salem (period 1973-2013), we
identified three flash floods: January 2003, April 2009 and
February 2012.
Two propagation models are challenged to predict flows at
Bou salem: Muskingum model and Regression model. For
these two models, we considered upstream flows: Jendouba (at
the main course Medjerda) and flow rates of 3 tributaries
(Mellègue, Bouheurtma and Tessa) which flow into the section
Jendouba-Bou Salem. The forecasting delay is varied from 2
to 8 hours with a pitch of 2 hours. To judge the validity of
predictive models, we took three numerical criteria: relative
error peak, Nash coefficient and peak time error.
Floods forecasting by both models and for all forecasting
delay showed that the satisfaction of three numerical criteria at
once is not possible. But generally forecast results were
satisfactory. The models reproduced well the shape of the
hydrograph, the peak and the flood. Taking into account three
criteria, we argued that it is the Regression model prevails.
The flash flood forecasting is a step in risk management,
which required the implementation of flood warning systems.
This system provides information when water levels rise
rapidly to cause flooding within hours.
ACKNOWLEDGMENT
I would like to express my gratitude to all those who gave me
the possibility to complete this article. I address my deep
recognition to my colleague HAJJI Olfa, Professor HABAIEB
Hamadi.
REFERENCES
[1] E. Gaume, Elements of analysis on flash floods, Paris. Brief
submitted for the grant of the title of Doctor of the National
School of the Rural Engineering of waters and forests and
Doctor of the National Institute of Scientific Research, 2002, p.
19.
[2] V. E. Borrell, J. Chorda, D. Dartus, Prévision des crues éclairs,
Geosciences 337(13), 2005; pp. 1109-1119.
[3] S. ABIDI, O. HAJJI, T. HEMASSI, H. HABAIEB, Influence of
considering tributaries to reconstruct flood hydrograph of an
extreme event on the upstream portion of Medjerda River,
International Research Journal of Public and Environmental
Health Vol 1(2), 2014, pp. 54-62.
[4] G. T. McCarthy, The unit hydrogroph and flood routing. U.S.
Corps Eng., Providence, R.I.J, 1939.
[5] A. Cunge, On the subject of a flood propagation computation
method (Muskingum Method), J. Hydraulic Res, 1969.
[6] H. Habaieb, Numerical Comparison of models of flood forecast,
application to watershed Belgian, French and Tunisian. Doctoral
thesis at the INP Toulouse – France, 1992, p.

FLOOD ROUTING WITH REAL-TIME METHOD FOR FLASH FLOOD FORECASTING IN THE PLAIN BOU SALEM

More Related Content

Viewers also liked (20)

Similar to FLOOD ROUTING WITH REAL-TIME METHOD FOR FLASH FLOOD FORECASTING IN THE PLAIN BOU SALEM (20)

More from International Journal of Technical Research & Application (20)

Recently uploaded (20)

FLOOD ROUTING WITH REAL-TIME METHOD FOR FLASH FLOOD FORECASTING IN THE PLAIN BOU SALEM