668 Global Change: Facing Risks and Threats to Water Resources (Proc.

of the Sixth World

FRIEND Conferen ce. Fez, Morocco, October 2010). IAHS Publ. 340 , 2010.

A comparison of two conceptual models for the simulation of

hydro-climatic variability over 50 years in a large Sudano-
Sahelian catchment

DENIS RUELLAND J , VIVIANE LARRAT 2 & VI NCE NT GUI NOT 3

I CNRS-UMR HydroSciences Montp ellier, Place E. Bataillon, F-34395 Montpe llier Cedex 5, France
ruelland@msem.univ-montp2.fr
2 JRD-UMR HydroSciences Montpellier, Place E. Bataillon, F-34395 Montp ellier Cedex 5, France
3 UM2-UMR Hy droSciences Montpellier, Place E. Bataillon, F-34395 Montp ellier Cedex 5, France

Abstr act Environmental and climatic changes have occurred in western African regions over many decades.
These changes, which are variable in space and time, can have lasting effects on water resources.
Hydrological modelling can help to assess the impact of these changes by representing the processes
governing the relationship between climatic data and river flow regimes. This paper compares two reservoir-
based hydrological models (HydroStrahler and GR4J) operating on a daily time step. The models are tested
over an approx. 50-year period in a large, poor ly gauged Sudano-Sahelian catchment that has undergone
significant hydro-climatic variabi lity. A calibration/va lidation exercise is performed using lumped and semi-
distributed approaches. The simulations are compared via a multicriteria analysis based on a variety of
goodness-of-fit indices. Both models simulate the rainfall-runoff relationship over the catchment area with a
fair degree of realism. Given the calibration strategy, the flood dynamics are better reproduced using the
HydroStrahler model, while the GR4J model gives a more accurate estimate of cumulated discharge. The
semi-distributed approach allows for a better representation of the hydrological processes within the
watershed, which does not necessarily lead to improved outlet simulations compared to the lumped
approach. A sensitivity analysis of the parameters also shows that equifinality problems are reduced when
the calibration is considered within a multiobjective framework. Furthermore, the robustness of the
simulations indicates that the models can be applied to various climatic conditions. The performance indexes
prove satisfactory in validation periods containing either wet or dry spells. As a result, these models can be
used to forecast future water availability using mid-term climatic scenarios in the basin.
Key words rainfall-runoff modelling; hydro-climatic variability; hydrostrahler; GR4J ; spatialization; sensitivity;
River Bani

I NTRODUCTION
T he severe drou gh t that West Africa has endured over the last 40 years has had significant
hyd rol ogical im pacts. It is we ll established tha t the decrease in mean annual disc ha rge of the
region ' s largest rivers has sometimes doub led the decrease in rainfall fo r the 1970- 2000 period
(see e.g. Le be l et al., 2003 ; An dersen et al., 2005). For exa m ple, since the 1970s, the Ban i Ri ver
(the main tri butary of the Niger River) has und ergone drastic decreases in runoff (aroun d 68 %)
compared with the 1950- 1960s (R ue lla nd et al., 2009) . Amo ng the main western African river
systems, a deficit of this magnitude has only ever been observed in the Senega l catc hmen t (Servat
et al., 1997). Th e reduction of water resourc es co nstitutes one of the mai n factors lim iting
development in those areas . T he flood of the inne r Niger de lta, fed by the Niger R iver and the Bani
River, has decreased dramaticall y (Mari ko, 2003) and led to deterio ratio n in living conditions for
fis hermen, farmers and herdsmen who depend on ann ua l floods. Over one mill ion people live off
the delt a ' s resources, an d th e social balance is threatened by the decli ning flood s (Mane et aI.,
2007). T his is extreme ly worry ing for the local populations, even mo re so d ue to dem ographic
growth that has been accele rating in the last decades.
Faci ng the g loba l changes that pa rtic ularly a ffec t the reg ion requires furt her insight in to the
flow ph eno men a in upstream ca tc hments of the de lta and improved capacity of environme nta l risk
forecasting . Hydrological mo de lling ca n give us a greater understandi ng into the im pac t of the
changes on water resources, by representi ng the processes governing the relationship bet ween
cl imatic data and river flow regimes . B efore a model ca n be used to make pred iction s it mu st be
adapted to va rious climatic co nditio ns . For this purpose, a wide spa tiote mpo ral scale is
recomme nded. In West Africa, the pau city of descrip tive data on suc h a scale mil itates in favour of

Copy right © 2010 IAHS Press

 Cerecterisetion de la secheresse hydropluvlometrique du Bani 663

Legende

Alluvions recentes

Alluvions anciennes

Continental Terminal

Gres

Grauwackes et conglornerats

Roches volcaniques et dolerites

Schistes, micaschistes, quartzites

1+'+'+"1 Granites, migmatites
L.:.±.:.±.:.±J

i\
~ .3°W/9°N o
~
50
km
100

Fig. 2 Carte gcologique du bassin du Bani a Douna (d 'apres DNGM et al., 1980; Brunet-Moret et al.,
1986) .

caracterc alcatoire ou non de la serie (Kendall & Stuart, 1943) , le test de Pettitt (Pettitt, 1979) teste
la presence ou l'absence de rupture (changement de moyenne au sein de la serie) et foumit une
date de rupture, la procedure de segmentation (Hubert et al., 1989) est une methode appropriee a la
recherche de multiples changements de moyenne.
Pour ce qui est de l'analyse pluviornetriquc, les tests ont concernc des series de cumuls
annuels et mensuels. Des calculs d'indices pluviometriques, de variation relative de cumuls de
pluie, de profils pluviornetriques joumaliers et de caracteristiques des pluies journalieres (date et
jour pluvieux) cornplcteront cette analyse.
Pour ce qui est de I'analyse des ccoulcments, les tests ont concerne des modules annuels et
mensuels et des debits caracteristiques des differentes phases de I'hydrogramme de crne : debits
maximum et minimum (Qmax et Qmin) , debits egales ou depasse 10 jours et 355 jours par an (DCC
et DCE) , debits egales ou depasses I, 3, 6 et 9 mois par an (DC I, OC3, OC6 et OC9) , debits moyens
maxima annuels de 5 et 30 jours consecutifs (VCX5 et VCX30), debits moyens minimum annuel de
60 jours consccutifs (VCN60). Des calculs de variabilite relative cornpleteront cette analyse.

Analyse pluviometrlque
52 stations ont ete retenues pour une analyse de leurs series annuelles et mensuelles d'apres les
crite res suivants: stations situecs a l'interieur ou a moins de 150 kms du bass in sans lacunes et
dont les series debutent avant 1960 et se terminent apres 1980.
L'analyse des resultats des tests ou procedures statistiques revele que I'ensemble du bassin du
Bani a ete touche par une diminution de sa pluviornetrie annu elle et mensue lie mais a des degres
670 Denis Ruelland et al.

Located in a Sudano-Sahelian climatic regime , the Bani catchment is characterized by a

monsoon climate with a strong north-south rainfall gradient (Fig. I) and considerable rainfall
variability since the mid-20th century . As a result, the flow at the Douna gauging station fell by
68% between 1952-1970 and 1971-2000, with a decrease in the deep water recharge and baseflow
contribution to the annual flood (Ruelland et al. , 2009) . Some of the low-water periods were so
severe that the river flow stopped periodically at Douna during the 1980s. This decrease in runoff
time series is also found at other discharge stations in the study area (67% for Dioila , 61% for
Pankourou, 56% for Bougouni) .

MA TERIALS AND METHODS

Hydrological models
In order to represent the seasonal and inter-annual vanations in runoff from the catchment,
discharge was linked to rainfall. In light of the data scarcity in the catchment, a conceptual
approach was favoured. Two rainfall -runoff models were compared: the HydroStrahler model
(Billen et al., 1994; Ruelland et al. 2008, 2009) and the GR4J model (Perrin et al ., 2003). Using
daily rainfall/PE data, these models represent, in a simplified manner, the flow processes in a
catchment and make it possible to simulate runoff at its outlet with a daily time step.

(a) HydroStrahler (HS) (b) GR4J

(PE
Interception - - t -- ---t--

Shallow [-~ Production

reservoir sat \ reservoir
'--"-'H-~ Superficial runoff
ssrr ~
Sub-surface
Percolation

runoff ~---
10% 90%
Surface runoff
infr
Infiltration

Deep I Routing
reservoir reservoir

~w~~seflow - ! - -- -{ F(.'(2)
Calibration parameters Calibration parameters
'"1: soil saturation content (mm ) xt : maximum capacity of the PR (mm)
ssrr: sub-surface runoff rate (Id) x2 : groundwater exchange coef . (mm )
in!': deep infiltration rate (Id) x3 : 1day ahead max . capacity of the RR (mm)
gwrr: groundwater runoff rate (Id) x4 : time base of unit hydrograph UH (d)

Fig. 2 Schemes of (a) HydroStrahler (after Ruelland et al., 2009) and of (b) GR4J (adapted from Perrin
et al., 2003) (R: rainfall ; PE: potential evapotranspiration; Q: streamflow) .

HydroStrahler takes into account two reservoirs in the watershed (Fig. 2(a)): (i) a shallow
reservoir, with short residence time, supplied by rainfall and feeding evaporation,
surface/subsurface runoff and infiltration, and (ii) a deep reservoir , with longer residence time , fed
by infiltration and generating the baseflow. The model involves four parameters as shown in
Fig. 2(a). These parameters need to be calibrated for each catchment being modelled in order to
account for three sources of runoff: immediate, rapid and delayed runoff.
GR4J simulates runoff via two functions (Fig. 2(b)). First, a production function that accounts
for rainfall (net rainfall) and evapotranspiration, determines the rainfall fraction (effective rainfall)
 A comparison of two conceptual models for the simulation of hydro- climati c variability over 50 years 671

part icipating to flow and supplying the production reservoir (interception and percolation). Next, a
routi ng function ca lculates runoff at the catchment outlet. The wa ter quant ity feedi ng the routi ng
part of the model is then made up from the perco latio n added to the wa ter remai ning fraction. This
flow is then divided into two fract ions: (i) 90% co mpose a rapid runoff that is routed through a
unit hydrograph (UH I) and a routing reservoir; (ii) 10% are attributed to a more delayed runoff
that is routed through a unit hyd rograph (HU2). The objec tive of the unit hyd rograph s is to
acco unt for differences in runoff delays between the two conceptual reservoirs. GR4J also require s
the ca libration of four para meters as show n in Fig . 2(b) .

Hydro-climatic data
Daily rainfall series were derived from 72 raingauges cov ering the area (F ig. I) . Duri ng the 1950-
2000 period an aver age of 65 gauges per day (with a minimum of 39) was used to interpolate daily
rainfa ll maps by the inverse distance we ighted method that pro ved to be optimally acc urate among
the classically available methods for data reconstru ction in the given context (Ruelland et al.,
2008) . Potenti al eva potra nspiration (PE) forcing consisted of monthl y maps (Fig. I ) produced by
the Climatic Research Unit (Univers ity of East Anglia, UK) from about 100 stations spre ad ove r
West Afr ica, using Penman 's method and spline interpol ation (M itchell & Jones, 2005). Since PE
presents low var iatio ns during the year, monthly data we re disaggregated evenly to the daily time
step within eac h month. Last ly, four discharge ga uging statio ns (Do una, Dioi la, Bou gouni and
Pankourou) correspo nding to the sub-catchments of the Ban i River were selecte d on the basis of
the numb er and quality of the time series ava ilable (Fig. I).

Model calibration and validation

A procedure was developed to automatica lly calibrate the models and to calculate run off. For eac h
catchment, the spatial average of daily rainfall and PE is ca lculated over the entire mul tiyear
period und er examination. The models were then run with these data as inputs, whi le their
respective four parameters were changed within a defin ed range. These systematic test runs aim ed
to optimize a given statistical criterion betw een the ca lculated and observed va lues of spec ific
flows over the calibration period. In this study, the fo llowing criteria were considered: (i) a goo d
agreem ent between the ave rage simulated and observed catchment runoff volume ; (ii) a goo d
overall agree ment of the shape of the hydro graph ; (iii) a goo d agree ment of the peak flows. In
order to obtain a successful calibration by using automa tic optimiza tion routines, it was necessary
to formul ate numerical perform ance measur es that reflect the ca libration objectives. Thi s was done
by consider ing the ca librat ion probl em in a multi -objective framework (see Madsen, 2000 ;
Ruelland et al., 2009) . The following num erical performance statistics measure the different
calibration objectives stated above:
Nash, Nas h-Sutcliffe coe fficient (Nas h & Sutcliffe, 1970):

Nash = I - [I N (

",' 1
QOb'.' - Q'im.,
)'
/ I Qob'., - QOb' ]
N (

1=1
_ )'
( I)
YE, volume error:
p p p

YE = (V"b' - V,om)/ VObs =(L Vobs.!' - L V'im.y) / L V obs . y (2)

.1'=1 .1'=1 .1'=1

YE. vg , annual average relative volum e error:

I P I
YE,", = P~ (IVob,." - ¥;im .y / Vob' . , ) (3)

PE. vg , annual average peak erro r:

PE avg -lp f
- _~
(IQp"k
ohs ,}'
_QP"k
Slm ,y
1/ QP"k)
obs. y
(4)
672 Denis Ruelland et al.

where Na nd P are the numb er of time steps and the number of years in the period, respectively, Q ohs.1
and Q sim,1 are the observed and simulated discharge at time t, Q obs is the average observed discharge
over the period , Q~~:,~ and Q~~~~ are the observed and simul ated peak discharge for the year y , Vohs
and are the total volumes of the observed and simulated hydrogr aphs over the period, and Vobs,y
v"i m
and the volumes of the observed and simulated hydrographs for the year y.
v" i'RY are
This multi- objective calibration probl em was transformed into a single-obj ective opt imization
problem by defining a scalar objective function Fagg that aggregates the various objective functions:
Fagg = (i - Nas h ) + IVE I+ VE avg + PE avg (5)
Mod el calibr ation was then performed in a 4-0 paramet er space by searching for the min imum
value of this aggregated function . Two main strategies were considered for calibration/validation
of the models:
10-day lumped modelling: spatially-distributed forcing s are aggregated over the entire basin
for use in the lump ed version of the model s. The optimal paramet er set is then estimated by
calibrating the lump ed model to simulate streamflow at the basin outlet.
10-day semi-distributed modelling: the values of Q used in equations ( 1)-( 4) are the specific
discharges for each of the sub-catchments. This strategy does not account for any transfer time
between the upstream and downstream stations. However, an analysis of the observed hydro-
graphs indicates that such a transfer delay time is smaller than the 10-day simulation time step.
The 1950-2000 simulation period was divided into three parts. Calibration was perform ed for a
20-year period (1967-1985) and validation was carried out over two 15-year periods. This choice
was made for several reason s. First, the period 1967-1 985 exhibited no data gaps for any of the four
reference gauging stations; consequently the amount of information used for the semi-distributed
calibration of the model was maxim ized. Secondly, the two validation periods are distingui shed by
contrasted climatic behaviours, the period 1952-1 966 being wet and the period 1986-2000 being
rather dry, which allowed the models to be tested for their suitability in differing climatic situations.
Validation consisted of running the models with the parameters optimized during the calibration
phase . The first two years of simulations (both in calibration/va lidation) were used as a model warm-
up to eliminate the influence of initial conditions in the models' reservoirs.

RESULTS
Efficiency of the models
A comparison of the efficienc y of the mod els is summarised in Table 1 and Fig. 3. Goodness-of-fit
scores are rather good for both the calib ration and validat ion phases, even if they tend to
deteriorate slightly during the validation periods for both models (Table I) . Over the whol e 1952-
2000 simulation period, HydroStrahler gives better result s in lump ed mode (values of Nash index
of 88% vs 81% in semi-distributed mod e) whil e better perform ances are observed for the semi-
distributed application of GR4J (83% vs 79% in lump ed mode). Neve rtheless, simulations based
on HydroStrahler seem to more accurately represent the hydrograph shape regardless of the chosen
mode used for spatial discr etization of the catchment. In particul ar, simulations based on GR4J
tend to shorten the flood event by making the flood ascent phase begin too late and the depletion
phase sta rt too early (Fig . 3).
Figure 4 illustrates these observation s in greater detail. Indeed, an analys is of annual scores of
the aggregated objective function (extracted from simulations with both model s) confirms that in
lumped mode, HydroStrahler is mor e accurate than GR4J (the obta ined values being mainl y
situated below the first bisector) . How ever, for the most recent years ( 1986-2000), which are also
among the driest, GR4J tend s to give results that correspond with those of HydroStrahler.
However, in semi-distributed mod e, this graph confirms that GR4J is more accurate than
HydroStrahler for both validation periods .
 A comparison of two conceptual models for the simulation of hydro- climatic variability over 50 yea rs 673

Table 1 Good ness -of-fit scor es for the various simulations: (LP) lumped and (SD) semi-distributed models.
1952-1966 (validation) 1967-1 985 (calibration) 1986- 2000 (validation)
Goodness-of- Nash YE YEavg PEovg Fagg Nash YE VEovg PEavg F,gg Nash VE VEovg PE, vg Fagg
fit criterions
DOlma 87% +0.27 0.26 0.25 0.9 88% +0 .00 0.23 0.22 0.6 8 1% - 0.24 0.45 0.32 1.2
(LP)
VJ Douna 77% +0.29 0.29 0.3 1 1.1 84% - 0 .03 0.27 0.22 0.7 84% - 0.25 0.46 0.22 1.1
:r: (SD)
DOlm a 76% +0. 17 0.18 0.25 0.8 82% +0.00 0.22 0.26 0.7 70% - 0.26 0.67 0.52 1.7
(L P)
;;t
x Douna 8 1% +0.23 0.23 0.14 0.8 82% +0.00 0.24 0.16 0.6 85% - 0.18 0.39 0.25 1.0
o (SD)

2000 -.--- - - - -- -- - - - ---, 2000 -,---- - - - -- - ------,

Lumped modelling Semi-distributed modelling

1600 Nash 1600 Nash VE

HydroS. 88% HydroS 81% +0.11
GR4J 79% GR4J 83% -+-0.10
~ 1200 UJ 1200

1o 800
io 800

400 40 0

--.
J F M A M J J A SON 0 J F M A M J J A SON 0
----- Qobs - QHvdroStrahler

Fig . 3 Comparison of the mean observe d and simulated hydrographs at the Douna station for bot h
models ove r 1952- 2000 (goo dness -of-fit scores corr espond to the whole 1952-2000 period ).

3.5 -,---- -- -- - - - -----", 3.5-,--- - -- - - - - - ---"

Lumped modelling Sem i-distributed modelling

~ 2.5 ~ 2.5
Cl:: Cl::
Cl Cl
.' '"
u::'" 1.5 '"
. ..
en
.: 1.5 I

.
..; . ..- .. . . ......
. ~~~ ~
0.5 "", •• : . • 0.5
...
•

0.5 1.5 2.5 3.5 0.5 1.5 2.5 3.5

Fagg HydroStrahler Fagg HydroStrahler

• Validation 11952-1966) • Calibrat ion 11967- 1985) • Validation 11986-200 0)

Fig. 4 Annual results of the scalar obje ctive function F agg at the Douna station for both models.

Calibrated parameters
The Bani River catchment is a heterogeneous basin in terms of rainfall , vegetation, soil, and
geology distributi on. Consequently, different parameter sets may be assumed a priori for each
sub-catchment. The analysis of the optimize d calibration parameters (Table 2) confirms that the
catchment portions have different hydrological properties, and shows that the lumped calibration
parameters (for both models) reflect an intermediate level of these various properties.
674 Denis Ruelland et al.

In the case of HydroStrahler, it is observed that the soi l saturation content (sat) increases
between downstream and upstream po rtions, wh ich may be partly link ed to an (expected) increase
in so il water-holding capaci ty due inter alia to the increased density of vege tation from north to
south. Th e subsurfac e runo ff rate (ssrr) also increases from north to south in line with the clim atic
gradient: as the rain intensifies, more water is avai lable to flow through the shallow reservoir. In
fact, rainfall is more significant in the upstream sub-bas ins (Bougo uni, Pankourou), which
explains their major co ntribution to the annual flood. In contrast, the infiltration rate Cinfr) and
most importantl y the groundwater runoff rate (gwrr) drastically decrease between down stream and
upstream porti ons: water flows rapidl y through the surface horizon s in the upstream sub-bas ins in
cont rast with the dow nstrea m Doun a portion where base runoff has a much more important
contribution to the spec ific discharge. This can be explained by the differences in hydrodynam ic
properties of the sub-basi n aquifers as a result of the geo logica l charac teris tics in the catchment
(Ruelland et al., 2009). In the same way , wit h GR4J, the capac ity of the production reservoir (x l)
increases between downstream and upstream porti ons while the l-day capacity of the rout ing
reservoir (x3) decreases. Once again, the behaviour of these parameters can be explained by the
limit ed contribution of baseflow to runoff in the up stream porti ons .
Howe ver, these assumptions need to be qualifi ed. The calibration parameters are essentially
theo retical; their values have a limited physical meanin g and can result from numeric solutions
that have few links w ith realit y. For instance, some G R4J parameters like the groundwater
exchange coefficient (x2) and the time base of unit hyd rograph (x4) were fairly unresponsive
dur ing ca libration, which shows no obvio us correlatio n with physical phenomenon. For insta nce,
variations of x4 below a value of less than 10 days did not lead to significa nt degradation s of the
aggregate d obje ctive func tion.

Table 2 Lumped (LP) and semi-distributed (SO) calibrated parameters over 1967-1 985 for both models.
HydroStrahler GR4J
sat ssrr infr gwrr xl x2 xl x4
Catchment (mm) (dol) (dol) (dol ) (mm) (mm) (mm) (d)
Doun a (LP) 470 0.00029 0.00003 0.01290 580 0.036 180 3.5
Douna portion (SO) 450 0.000 08 0.0002 1 0.0 18 40 590 0.076 220 10.5
Oiolla portion (SO) 460 0.000 14 0.000 07 0.01560 800 00023 125 1.5
Pankourou portion (SO) 500 0.00026 0.00003 0.000 01 940 0.260 100 4.0
Bougouni portion (SO) 500 0.0003 4 0.00007 0.0000 1 1420 0.001 75 7.5

Sensitivity analysis of the parameters

Sensitivity analys is of the calibrated parameters clea rly shows that some parameters of the mod els
are subjec t to greater variability than others durin g the calibration phase. The more sensitive
parameters were show n to be: (i) the subsurface run off rate (ssrr) and the infilt ration rate (infr) , for
HydroStrahler; (ii) the production reservoir capaci ty (xl) and the I-day capaci ty of the routin g
reservoir (x3), for GR4J. For these parameters, it is seen fro m Fig. 5 that onl y one num erical area
allow s for optimal calibration to be attained. Well-known equ ifinality problems are then reduced.
In con trast, the least sens itive parameters (sat and x4) present threshold s below which the
goodn ess-of-fit scores do not significantly deteriorate, which reveals probl ems of equifina lity
(notably in the case of GR4J). Th e threshold effect of sat can be explained by the fact that the
opt imal calibration never leads to generate immediate run off through the sha llow reservoir: as far
as sat is suffic iently high, the soi l saturatio n content is never affected, and the simulated sur face
runo ff is on ly composed of rapid (or subsurfa ce) runo ff. Thi s is probabl y due to the necessary
tim e-transfer of both surface and subsurface water fluxes at the study sca le. The threshold effect of
x4 can be explained by the fact that the x3 parameter contro lling the routing reservoir would be the
main factor in determini ng the wate r flux delay between the two conce ptua l reservoirs. For the
 A compariso n of two conceptual models for the simulation of hydro-climatic variability over 50 years 675

purposes of this study it was then pointl ess to place too much importance on these parameters in
the calibration phase: once the threshold values were found , it was just a matter of fixing them and
vary ing the other param eters. In light of this, it might be interesting to make changes to the
equa tions controlled by these parameters in order to strengthen their roles, or, in cont rast, to use
models with fewer parameters (e .g. GR3J, see Edijatno et al., 1999).
Figure 5 also shows that each objective function can constitute a source of equifi na lity.
Aggr egating the various functions into a single-o bjective ca libratio n could then emphasise the
problem rather than redu cing it. Therefore, it is esse ntial to make sure that the chose n agg rega ting
function Fagg (see equation (5)) does not interfere with the search for the ca libration optimum. It is
interesting to note that in both models the equifi nality area produced by Fagg mirrors the on e
produ ced by the Nas h criterion (F ig. 5). This latter criterion wo uld have a particular discriminant
role in the constitution of Fagg . No netheless, for both models, and for every pair of para meters
studied, Fagg is able to red uce the equifinality area in comp arison to the use of a single criterio n
(Nas h or VE for exa mple, Fig. 5). Co nsequently, this function does not prevent us finding the
optimal param eters, but makes it possible to min im ize uncertaint ies result ing from calibration. It
represents a suitable solution for calibrating models by offering an objective function that is rich,
easy -to-use , and that allows us to efficiently target the most efficient parameters.

Nash VEg Fagg

-
~

ssrr ssrr ssrr

50 0 500

..,
~
4 00 40 0

0::
C) ~ 30 0
""l
~
300 ~
:c
- 200 200

100 100
4 00 600 600 1000 120 0 1400 40 0

rl rl rl
Fig. 5 Sensitivity of the "key-parameters" depending on the goodness-of-fit indices used (Nash, volume
error, aggregated function): (a) I-1ydroStrahler; (b) G R4J .

DISC USSION: ABILITY OF THE TWO MODELS IN SIMULATING THE LO NG-TERM

HYDRO-CLIMATIC VARIABILIT Y OV ER THE CATCHMENT
In western Afr ica, as in num erous regio ns where wa ter avai lability causes severe probl ems, it is
crucia l to und erstand how river systems function. In order to represe nt the rain fall-runoff
processes in a large, poorl y-gauged tropical catchmen t, two conceptual hyd rological models
(HydroStrahler and GR4J) were compa red using lump ed and semi-distributed app roache s. The aim
was to model the wa ter balance and its changes when subjec t to climatic stress at vario us sca les.
This study shows that in these regions it is difficult to accurately reproduce the rainfall- runoff
relationship in a wide spatiotemporal scale. The conceptual character of the models, in conjunction
with the lack of, and\or quality of the data, make s it a delicate task. Neve rthe less, both of the
676 Denis Ruelland et al.

hydrological model s studied here allow the runoff in the catchment to be correctly simulated ove r a
nearly 50-year period characterized by signific ant hydro-climatic variability. The Nash coefficients
obtained at the Douna gauging station with HydroStrahler (in lumped mode) and with GR4J (in
semi-distributed mode) are always over 80%, in both calibration and validation (Table I).
Consequently, the simulation results show a good agreement with the flows observed at the outlet
during both wet and dry periods. Even if the semi-distributed approach generally yields poorer
results with HydroStrahler than with GR4J, the simulation qualit y remains acceptable. The spatial
heterogeneousness of the basin is found through the calibrated parameters. The spatial variability
between these parameters mainly reflects the different volumes of precipitated water between the
catchment's portions. They can also be explained by variable capacities of infiltration linked to the
geologic characteristics. Nevertheless, even though the semi-distributed modelling gives a better
account of the hydrological processes inside the watershed, this study shows that moving from a
lumped to a semi-distributed approach does not necessarily improve the simulated hydrograph at the
outlet of a watershed with heterogeneous physical charac teristics. These findings contradict the work
of some authors (see e.g. Baudez et al., 1999; Bourqui et al., 2006; Das et al. 2008) and reinforce the
idea that distributed or semi-distributed approaches do not clearly demonstrate any practical
superiority over a lumped approach in simulating streamflow (see also Loague & Freeze, 1985;
Michaud & Soroshian, 1994; Refsgaard & Knuds en, 1996; Loumagne et al., 1999).
Few hydrological studies have been carried out on this kind of spatiotemporal sca le in western
Africa. However, one can refer to Schuol et al. (2008) who tested the dail y physically-ba sed
SWAT model (So il Water Assess ment Tool) in severa l large West African catchm ent s and over a
dry period ( 1970- 1995). The simulations with this model (that runs according to a distributed
spatial approach) were confro nted with the obse rvations of several stations. The Nas h coe fficients
between 0 and 70% (depend ing on the validation/ca libration phase at the Douna station,
respectively) show a less accurate representation of the rainfall- runoff relationship than that
obtained with the daily models used in the present study. The difference betwe en the input data
obviously exp lains those poorer result s as we recentl y showed (Laurent & Ruelland , 20 I0) that it
was also possible to obtain satisfactory simulations over the Bani catchment with SWAT. The
conceptual GR2M model (modele du Genie Rural Mensuel a 2 pararn etres) has also been applied
in West Africa (Deze tter et al., 2008), notably ove r the Bani catchm ent. In that study, the
efficiency of the model is esti mated via the Nash coe fficient. The best coe fficient obtained by
calibration (at a monthly time step) at the Douna station over 1922-1 995 is 74% . For the same
catchment and over the 1952- 2000 period, we obtain a Nash criterion value of 88% with
HydroStrahler (lumped mode) and 83% with GR4J (semi-distribut ed mode) at a 10-day time step.
Eve n then, the models used in this study seem to be more accurate. The use of four param eters
(compared with two with GR2M) can partially explain those better result s. Furthermore, the use of
other obje cti ve functions criteria, such as the YE or PE, allows the simulations to be tested more
thoroughly. Still, it can be argued that the time period in the prese nt study is shorter (1952-2000 vs
1922-1 995) . Meanwhile, the simulation period used with GR2M includ es mainly wet years in
wh ich the rainfall-runoff relationship seems eas ier to reproduce. In additi on, the Nas h coefficient
tends to give more weight to high values and thus to wet yea rs through its very design (see e.g.
Perr in, 2000).
There are still some questions that need to be addr essed regard ing our simulations. The
analysis of the results does not allow us to make any co nclusions to catego rically determi ne the
superiority of one mod el ove r the other (HydroStrahler vs GR4J ) in the study context.
HydroStrahler simulations more accura tely rep roduce the shape of the hydrograph (Fig. 3) while
GR4J parameters tend to raise more equ ifinality probl ems. On the other hand, the simulated
cumulated discharge over a 50-yea r period presents a significant gap in co mparison to the
observe d cumulated discharge, in part icular with HydroStrahl er (Fig. 6), which may be
probl ematic in terms of a long-term prospective assess ment of water resources. Mor eover, many
anthro pogenic issues that might have disrupted the hydrol ogical functioning of the catchm ent have
not been considered. First, land cove r changes are not exp licitly tak en into accou nt in the
parameterization of the model s. However, recent remote sensing studies (Ruelland et al. , 20 1Oa,b)
 A comparison of two concept ual models for the simulation of hydro- climatic variability over 50 years 677

o .,~ - '--.---r-,---.----,--,-.-----,--,-.-----,------;,-.,----,------;r-.,--.-----;r--.-----r-----;--.-----r-----;-'
~~*~~~~~~~~~~~~v~*~~~**~~
00000000 0000000000000000~
Fig. 6 Comparison of the cumulated discharge obtained at the DOlma station from lumped and semi-
distributed simulations with HydroStrahler and GR4J.

ind icate that the dow ns tream Sahe lian par t of the ca tchmen t has und ergone drastic cro pland
expansion and defor estati on since the 1960s. Th ose cha nges in vege tatio n cover may have had an
imp act on soil wa ter-holding cap ac ity and run off becau se of alterations to surface features (see e.g.
D 'H erbes & Va lentin, 1997). Nonethe less, in the mos t produ ct ive sub-basins (in the upstr eam
pa rt), relative ly small land cover changes have been observe d becau se of wea ker demographic
pressur e and a grea ter capacity of the natura l env iro nme nt to regenerate (R ue lland et al., 20 lOa).
The n, if land cove r cha nges in the Bani ca tchme nt we re taken into account in the mode ls it
prob ably would not dram at ically improve the hydrol ogical simulations . In co ntras t, hydrological
functioning could be disrupted by the pro fus ion of dams that have been bui lt in we stern Africa
since the successive drou ght years of the 1970 -1 980s (see e.g . Cecc hi et al., 2009). The refo re,
improveme nts need to be made in order to take these da ms into acco unt whe n mak ing es tima tes
abo ut lon g-t erm wa ter balance. T hese improveme nts are esse ntia l so that the mod els ca n be used to
forecas t the imp acts of global changes in the reg io n and ove rco me major obstac les to ag ricu ltural
pro duction/ ma nage me nt based developm ent due to water scarcity.

Acknowledgements Th is wo rk was carried out as part of an intern al Hyd roSciences Lab proj ect
and o f the AN R RESSAC program (vulnerabi lity of surface water reso urces to anthro poge nic and
cli mate changes in Sahel).

REFERENCES
Andcrse n, I.. Dione. 0 .. Jaroscwich- Holder, M. & O livry, J-C. (2005) The Niger River Basin: a vision for sustainable
management. In: Direct ions in Development (cd. by K. G. Gol itzen), 145. The World Bank. Was hington DC, USA.
Baudcz, l -C , Loumagne, C, Michcl, c., Palagos. Boo Go rnendy, V. & Bart olli, F. ( 1999) Modelisation hyd rologiquc et
heterogeneite spatiale des bassins : Vers une comparaison de l'approche globale et de l'approche distribuee, Etude et
Gestion des Sols 6(4), 165-1 84.
Billen , G., Ga m ier, J. & Hanset, P. ( 1994) Mod elling phytop lankton development in whole drain age network : the RiverStrah ler
mod el app lied to the Se ine river system. Hydrobiologia 289, 119-1 37.
Bourqui , M ., Loumagne, C; Chahinian, N. & Plantier, M. (2006) Acc ounting for spatial varia bility: a way to imp rove lumped
modelling a pproac hes? An assessme nt on 3300 chi mera catchme nts. In: Large Samp le Basin Experime nts fo r
Hydrological Model Param eterization: Results of the Model Parema eter Experiment - MOP EX (ed . by V. A ndreass ian et
aI.), 300-3 10. IAH S Publ. 307 . IAHS Press, Wa llingfo rd, UK.
Brun et-Moret, Y., Chaperon, P., Lam agat, J.-P . & Mo linier, M. ( 1986) Monographi c hydrologique du fl eu ve Niger , Tom e I,
Niger Superieur . ORS TOM Ed., Co llection rnonographi es hyd ro logi ques 8, Paris, France.
678 Oenis Ruel/and et al.

Cecchi, P., Gourdin, F., Kone , S., Co rbin, D ., Etienne, 1. & Ca senave , A. (20 09) Les petits barr age s du nord de la Cote d'l voire:
inventaire et potenti alites hydro logique s. Secheresse 20, I 12-1 22.
Das, T., Bardossy, A ., Zehe , E. & He, Y. (200 8) Com parison of co nceptual mode l performance using d ifferent rep resentations
of spatial variab ility.J. Hydro/. 356 , 106- 118.
Dezetter, A., Girar d, S., Pature l, J.-E., Mahe, G ., Ado in-Bardin, S. & Servat, E. (2008) Simul ation of runoff in West A frica : is
there a single data -mode l combination that produ ces the best simulation results? J. Hydrol. 354 , 203-2 I2.
D 'H erbes, J.-M . & Valentin , C. ( 1997) Land surface conditions of the Niamey regi on : ec o logical and hydrological implication s.
J. Hydrol. 188, 18-42.
Ed ija tno, N. N., Ya ng, X., Makh lou f, Z. &Mic hel , C. ( 1999) GR3J : a daily waters hed mod el with three free parameters. Hy drol .
Sci. J. 44, 263- 278.
Laur ent, F. & Ruelland , D. (2010) Mo delisation iJ base physiqu e de la vari abili te hydroc limat ique iJ l' echelle d ' un grand bassin
versant tropi cal. In: Globa / Change : Facing Risks and Threats 10 WaleI' Reso urces (P roc. 6th FRIEND Int. Conf. Fez,
Morocco, 25- 29 Oc tober 20 I0) . IAHS Pub I. 340 . IAHS Press, Wallingford , UK.
Lebel, T., Diedh iou, A., & Laurent , H. (2003) Seaso nal cycle and interannual variability of the Sahe lian rain fall at hydrol og ical
scales. J. Geophy s. Res. 108, 83- 89.
Loague, K. M. & Freeze, R. A. ( 1985) A comparison of rain fall-runoff modellin g techni ques on small upland catchments.
WaleI' Resour. Res. 21(2), 229- 248.
Loumagne, c., Michel, C; Palagos, B., Baud ez, J.-c. & Bartoli , F. (19 99) . From a global to a semi-d istributed app roach in
rain fall- runo ff modellin g. La Houil le Blan che 6, 81- 88.
Mad sen , H. (2000) Automatic ca libratio n of co nceptua l rai nfall- runoff model using multiple objectives . J. Hydro/. 235,
276-288 .
Mar ie, J., Mor and , P. & N'Dj im, H. (200 7) Ave nir du fl euve Niger. IRD ed., Pari s, Fra nce.
Mariko, A. (2003) Caracteri sation et suivi de la dynamique de J'inondation et du couvert vegeta l dans le Delta interieur du
N iger (Ma li) par teledetection . PhD Thesis, Universite Montpellier 1/, Montpellier , Fran ce.
Michaud, J. D. & Sorooshia n, S. (1994) Com pariso n of simple versus compl ex di stributed runoff mod els on a mid sized
serniarid watershed . WaleI' Resour. Res. 30(3), 593-606.
Mit chell, T. D. & Jon es, P. D. (2005) An impro ved method of cons tructing a database of month ly cli mate observ ations and
asso ciated high-r esolutio n grids. Int. J. Clima tal. 25,693- 712.
Nash, J. E. & Sutcliffe , J. V. (1970) R iver flow forecasting throu gh conceptual models, a discussion of prin cipl es. J. Hydrol. 10,
282-290.
Perrin , C. (2000) Ve rs une amelioration d' un rnodele global plui e-debit au travers d 'une app roche comparati ve. PhD T hes is,
Institut National Polytechnique, Grenob le, Fra nce.
Perrin, C.; Michel, C. & Andreass ian, V. (2003) Improvement of a parsimoniou s model for stream flow simulation. J. Hydrol.
279, 275- 289.
Refsgaard, 1. C. & Knudsen, 1. (1996) Operational validat ion and intercomp ar ison of di fferent types of hydro logi ca l mod els.
Water Resour. Res. 32, 2 189-2202.
Ruell and, D., Ardoin -Ba rdin, S., Bill en, G . & Serv at, E. (2008) Sensiti vity of a lump ed and semi-distributed hydro logic al mo del
to several methods of rain interp olation on a large basin in West Africa. J. Hydrol. 36 1, 96-1 17.
Ruelland , D., Guinot, V., Levavasseur, F. & Ca ppe laere , B. (2009) Mode lling the long-term impact of clima te change on
rain fall-runoff processes over a large Sud ano-Sahelian catchme nt. In: New App roaches to Hy drological Prediction in
Data Spa rse Reg ions (Proc . Symp . HS.2 at the Joint IAH S & IAH Convention, Hyderabad , India, 6- 12 Se ptembe r 200 9),
59-68. IAH S Publ. 333. IAH S Press, Wallin gfor d, UK.
Ruelland , D., Levavasseur, F. & T ribotte, A. (201Oa) Pattern s and dynam ics of land- cover chan ges since the 1960s ove r th ree
experim enta l area s in Ma li. l nt. J. App l. Earth Obs. Geoinf. 12( 1), 11- 17.
Ruell and, D., Tribott e, A., Puech, C. & Dieulin , C. (20 10b) Compariso n of meth ods fo r LUCC moni toring o ver 50 yea rs from
ae ria l photographs and satellite images in a Sahelian catchm ent. Int. J. Remote Sens. (in press).
Schuol, 1., Abbas pour, K. C., Srinivasan, R. & Yang, H. (2008) Estimation of freshwater avail ability in the West African sub-
co ntinen t using the SWAT hydrol ogic model. J. Hydrol. 352, 30-49.
Servat, E., Paturel, J.-E., Lubes -N iel, H., Kouarne , B., Travaglio, M. & Marieu, B. (199 7) De la diminution des ecou lernents en
Afriqu e de J'Oue st et Centrale. CR Acad. Sci. 325 , 679-682.
~ Global Change:
~
Facing Risks and Threats
to Water Resources

Edited by:
ERIC SERVAT
UMR HydroSciences Montpe/li er (HSM ),
Universite Montpellier 2, France

SIEGFRIED DEMUTH
Hydrolo gical Processes and Climate Section, Division of Water Sciences,
Natural Sciences Sector, UNESCO , Paris, France

ALAIN DEZETTER
UMR HydroSciences Montpellier (HSM) ,
Universite Montpellier 2, France

TREVOR DANIELL
School of Civil and Environm ental Engineerin g,
University of Adelaide, Australia

Co-edited by: ENNIO FERRA RI, MUSTAPHA IJJAALI,

RAOUF JABRANE, HENNY VAN LANEN & YA N HUA NG

Proceedings of the Sixth World FRIEND Conference , Fez, Morocco ,

25-29 October 2010 .

IAHS Publication 340

in the IAHS Series of Proceedings and Reports
Published by the International Association of Hydrological Sciences 2010
IAHS Publication 340
ISBN 978 -1-907 16 1-13-1

Briti sh Library Cataloguing-in-Publicati on Data.

A ca ta logue recor d for this book is ava ilabl e from the British Library.

@AHS Press 2010

This publ ication may be reprodu ced as hard copy, in whole or in part, for educational
or nonprofit use, without special permissi on from the copyright holder, provided
acknowledgement of the source is made. No part of this publi cation may be
electronically reproduced, transmitted or stored in a retrieval system, and no use of
this publi cation may be made for electronic publi shing , resale or other commercial
purp oses without specific written permission from IAHS Press.

The pap ers included in this volum e have been reviewed and som e were exten sively
revised by the Editors, in collaboration with the authors, prior to pu blication .

IAHS is ind ebted to the employers of the Ed itors for the invaluable support and
se rvice s provided that en abled them to carry out their task effectively and effici ently.

The information, data and form ulae provided in this volume are reproduced by l ABS Press in
good fait h and as finally checked by the author(s); IA HS Press does not guarantee their
accuracy, completeness, or fitness for a gi ven purpose. The reader is responsible for taking
appropriate professional advi ce on any hydrological project and IA HS Press does not accept
responsibi lity for the reader' s use of the content of thi s volume. To the full est extent permitt ed
by the applicable law, l AB S Press shall not be liable for any damages arising out of the use
of, or i nabili ty to use, the content.
The designations employed and the presentatio n of material throughout the publication do not
imply the expression of any opin ion whatsoever on the part of IAHS concerning the legal
status of any country, territory, city or area or of its authori ties, or concerning the deli mi tation
of it s fro ntiers or boundaries.
The use of trade, firm , or corporate names in the publication is for the information and
convenience of the reader. Such use does not constitute an official endorsement or approval
by IAH S of any product or service to the exclusion of others that may be suitable.

Publications in the series of Proceedings and Reports are available from :

IAHS Press, Centre for Ecology and Hydrology, Wallingford, Oxfordshire OX10 8BB. UK
te/.: +44 1491 692442: fax: +44 1491 692448: e-mail: jilly@iahs.demon.co.uk
Printed by Information Press

Cover picture: Southern Morocco by Eric Servat