Catchment classification: multivariate statistical analyses for physiographic similarity in the Upper Niger Basin

JamilatouChaibouBegouInt. Journal of Engineering Research and Applications www.ijera.com
ISSN: 2248-9622, Vol. 5, Issue 9, (Part - 1) September 2015, pp.60-68
www.ijera.com 60|P a g e
a
Catchment classification: multivariate statistical analyses for
physiographic similarity in the Upper Niger Basin
JamilatouChaibou Begou1,2
, Pibgnina Bazie2
and Abel Afouda1
1
Graduate Research Program (GRP) Climate Change and Water Resources, West African Science Service
Centre on Climate Change and Adapted Land Use (WASCAL), University of Abomey-Calavi, Cotonou, Benin.
2
Centre Regional AGRHYMET, PO Box 11011, Niamey, Niger.
Abstract
The objective of this study was to determine physiographic similarity, as indicator of hydrologic similarity
between catchments located in the Upper Niger Basin, and to derive the dominant factors controlling each group
singularity. We utilized a dataset of 9 catchments described by 16 physical and climatic properties distributed
across a wide region with strong environmental gradients. Catchments attributes were first standardized before
they underwent an integrated exploratory data analysis composed by Principal Component Analysis (PCA)
followed by Hierarchical Clustering. Results showed a clear distribution into 2 major clusters: a group of
easterly flat catchments and another of westerly hilly catchments. This nomenclature came from the
interpretation of the main factors, topography and longitude, that seem to control the most important variability
between both clusters. In addition, the hilly catchments were designated to be dominated by forest and
ACRISOL soil type, two additional drivers of similarity. The outcome of this study can help understanding
catchment functioning and provide a support for regionalization of hydrological information.
Keywords: catchments, Hierarchical Clustering, physiographic similarity, Principal Component Analysis,
Regionalization.
I. Introduction
A core issue in hydrology is to make prediction
of hydrological variable where it is not measured.
This situation is of particular importance especially in
developing countries where many river basins are
ungauged [1-5]. This lack of information constraints
water resources management and constitutes a
stumbling block to adaptation to climate change in
the sector of water hence increasing the vulnerability
of rural population, particularly.
With the aim of predicting hydrological variables
in ungauged basins, regionalization procedures are
usually used. Different types of regionalization exist,
and can be classified as [6] in: 1) regression methods,
and 2) methods based on distance measures between
gauged and ungauged sites. The former methods
consist in deriving statistical relationships between
catchment attributes and the optimized model
parameters. Notwithstanding being considered as the
most common regionalization approach for flow
prediction in ungauged catchment [7], statistical
methods are limited in use due to the presence of
equifinality in calibrated model parameters. In fact, it
becomes difficult to associate individual parameters
with the physical characteristics of the catchment
(each parameter can take several values) and thus,
instead, complete parameter sets should be
transferred to ungauged sites [8]. Another drawback
of these methods is that most statistical models
consider linearity between catchment attributes and
model parameters [9, 10]. Consequently, in order to
address the issue of model parameters non-
uniqueness and propagate prediction uncertainty from
gauged to ungauged catchment, similarity methods
should be suitable.
Hydrologic similarity is an essential concept in
regionalization [11-13]. Many similarity concepts
have been proposed in the literature that attempt to
represent various hydrologic processes occurring at
different locations. [14], for instance, proposed three
similarity concepts: spatial proximity, similar
catchment attributes and similarity indices. In the
first concept, catchments that are close to each other
are assumed to behave hydrologically similarly.
Geostatistical methods are based on this similarity
measure. Many authors have indicated, for instance,
the predominance of kriging methods on
deterministic models in regions where the gauging
network is sufficiently dense (e.g. [15, 16]).
Nonetheless, it was pointed out that spatial proximity
does not always involve functional similarity
between catchments [17, 18], and thus[19, 20]
suggest, instead, the application of hydrologically
more meaningful distance measures. In the second
concept, catchment attributes, such as catchment size,
mean annual rainfall, and soil characteristics are used
as indicators of physiographic similarity. Many
studies stressed the value of parameter
regionalization methods based on physiographic
similarity, as a proxy for functional similarity ([10,
RESEARCH ARTICLE OPEN ACCESS

21, 22]. The third similarity concept is based on
hydrologic function defined by similarity indices
such as the aridity index of Budyko (e.g. [23, 24]),
which has proved to be a valuable measure of
catchment behavior.
Similarity of hydrological function between
catchments could be derived by a classification
scheme. As discussed by [13], the ultimate goal of
classification is to understand the interaction between
catchment structure, climate and catchment function.
Additionally, [25]proposed four objectives of
catchment classification which are: 1) nomenclature
of catchments, 2) regionalization of information, 3)
development of new theory, and 4) hydrologic
implications of climate, land use and land cover
change. Many authors attempted to classify
catchments around the world into similar groups. For
instance, [26]used 8 physiographic and
meteorological variables to organize 21 catchments
located within the Nile basin, into 2 homogeneous
regions by applying a multivariate statistical analysis.
In a different approach, [27]used self-organizing
maps to classify around 300 Italian catchments
according to several descriptors of the streamflow
regime and geomorphoclimatic characteristics. As for
[28], they distinguished only six dominant classes for
331 catchments across the continental United States
using four similarity metrics. It is worth noting the
work by[29]involving 24 worldwide large drainage
basins, among which, the Niger basin. In fact,
[29]considered sixteen geomorphologic and climatic
variables into multivariate statistical analyses and
obtained 6 clusters along with the description of the
major controlling factors driving the
hydrosedimentary response of each group. However,
large river basins, as it is the case in [29], usually
encompass several climatic regions and exhibit
strong environmental gradients. Consequently, a
global classification at such spatial scale can still hide
significant internal heterogeneities among
subcatchments, hence limiting our understanding of
the hydrological functioning occurring at smaller
catchments. Therefore, it is essential to break down
the scale and provide more detailed classification
scheme, and this is essential especially when
prediction in small ungauged catchments is foreseen.
However, only one a priori classification of the Niger
basin exists and have been proposed by the Niger
Basin Authority (e.g. [30]) which subdivided the
whole basin into 4 physio-climatic regions: the Upper
Niger, the Niger Inner Delta, the Middle Niger,
andthe Lower Niger. Nevertheless, this classification
falls short of providing a quantitative assessment of
the degree of (dis)similarity within and between the
so-called homogenous regions.
In the light of these examples, the main objective
of this study was to classify subcatchments of the
Upper Niger into similar groups according to their
physio-climatic parameters. The specific objectives
were to: 1) reduce the dimension of the input dataset
containing catchment attributes by a Principal
Component Analysis, and 2) perform a hierarchical
clustering of subcatchments based on the reduced
dataset. This study provides the first ever
quantification of similarity among catchments with
respect to physiographic characteristics on a large
tropical river basin at finer spatial scale. Nor
descriptors, neither statistics themselves are actually
novel in the broad literature, but their combined use
in that particular area to evaluate the gain of
homogeneity with increasing number of clusters,
is.The questions that will be addressed in this study
were: (i) can the Upper Niger further be separated
into similar groups of catchments based on their
physical characteristics, and if so, (ii) what are the
dominant controls of similarity between catchments.
II. Material and methods
2.1. The study area
The present study was conducted within the
Upper Niger (Fig. 1). This basin is composed by to
mutually independent subbasins: the Upper Niger
subbasin controlled by the Koulikoro gauging station
and the Banisubbasin at the Douna outlet, each
covering an area of 120,000 km2 and 101,000 km2,
respectively. The study area is shared by four West
African countries: Guinea, Cote d’Ivoire, Mali and
Burkina Faso, in a lesser extent. To avoid confusion,
the parent basin is called the Upper Niger and its
subbasin upstream Koulikoro is called the Upper
Niger subbasin.
Altitudes are unevenly distributed across the
Upper Niger. The extreme west and south of the
basin are hilly zones. The Tinkissosubcatchment, for
instance, is situated in the Fouta-Djalon Mountain,
which culminates at more than 1000 m in the region
of Dabola[31]. Similarly, the south of the basin is
shaped into plateau and mountains, the most
important of which is situated between Milo and
Dion rivers and reaches its highest point at 1500 m.
In contrast, the Bani watershed’s topography is
gently sloping, with altitudes ranging from 249 m to
826 m. Average annual precipitation (period 1981-
2000) varies from 1500 mm y-1 in the humid
Guinean zone in the south-west (region of
Kissidougou) to 620 mm y-1 in the Sahelian zone in
the North-east (region of Segou). The vegetation is
dominated by the presence of closed evergreen forest
in the highlands of the Fouta Djalon Mountain,
whereas the Baniis mainly the domain of savannah
with small spots of deciduous forest. ACRISOL is the
most important soil grouping on the majority of
subcatchments, except at the subcatchments
controlled by Bougouni and Kouro1 gauging stations.

Figure 1: localization of the Upper Niger basin and the study catchments
2.2 Catchments and catchments’ attributes
A total of 9 candidate catchments were selected
and range in size from 6379 km2
to 101,456 km2
and
were hereinafter given the name of their
corresponding outlet. For example, Bougouni
referred to the subcatchment controlled by the
Bougouni outlet. Three of them are included in the
Bani, while five are located on the Upper Niger
subbasin, (Fig. 1), and are referred to as Group I and
Group II, respectively. The Dounasubbasin, which is
actually the Bani, is the biggest catchment and was
added on purpose to test similarity across spatial
scale. In addition to belonging to two hydrologically
non-connected subbasins, Group I and Group II
individuals were chosen to be non-nested sites in
order to provide a better structure of independence
between subcatchments. Furthermore, these
subcatchments have not been affected by
anthropogenic activities able to significantly modify
their flow regime and have been chosen to be located
in the headwaters of both subbasins.
This study make the implicit assumption that the
physical similarity based on the selected catchment
attributes, is a proxy of hydrological functioning of a
catchment. Therefore the choice of catchment
attributes (CAs) is of great importance. Selected CAs
are related to the shape (e.g. area, length) and the
topography (e.g. slope, elevation) of each subbasin
and its main tributary reach and were derived by
application of the SWAT model (at watershed
delineator and HRU analysis processing steps
required for SWAT model setup). The same input
spatial data (Table 1) were used to characterize
Group I and Group II subcatchments. The selection

of the appropriate CAs can also depend on the
physical meaning of the model parameters (Mps) that
will subsequently be involved in information
regionalization. For instance, in the SWAT model,
the curve number parameter (CN2) which is
considered among the most sensitive Mps, depends
on the soil and land use characteristics of the
catchment [32]. Therefore, two other characteristics
related to land use and soil were considered as
descriptors: Forest and ACRISOL. Forest represents
the proportion of area covered by forest, and
ACRISOL gives information about the soil texture
based on the relative proportion of sand, silt and clay.
As ACRISOL remains the dominant soil in the
majority of the study catchments, its proportion is
used to indicate the presence of more than 35% of
clay in each catchment. Forest and ACRISOL were
calculated using the following equations:
Forest 100
Af
A
 
  
 
(1)
ACRISOL ,
Aacs
A
 
  
 
(2)
WhereAf is the area covered by forest within a
watershed, Aacs is the area covered by ACRISOL,
and A is the total area of the watershed.
Last, it is very common to use climatic
characteristics such as long-term annual precipitation
as indicator of similarity. Thus, average annual
precipitation was computed for each subcatchment on
the period 1981-2000. A detailed description of the
16 CAs is given in Table 2.
Table 1:Input data for SWAT model to derive catchments attributes on the Upper Niger basin.
Data type Description Resolution/period Source Processing
Topography
Conditioned
DEM
90 m USGS hydroshedsa
SWAT Watershed
Delineator
River River network 500 m USGS Hydroshedsa
SWAT Watershed
Delineator
Land use/cover GLCC version 2 1 km Waterbaseb SWAT HRU Analysis
Soil FAO Soil Map Scale 1:5000000 FAO c SWAT HRU Analysis
Precipitation
data
Rainfall Daily/1981-2000 AGRHYMET
Arithmetic mean
a
http://hydrosheds.cr.usgs.gov
b
http://www.waterbase.org
d
http://www.fao.org/geonetwork
Table 2:Summary of catchment attributes derived by the SWAT model as input for multivariate statistical
analyses on the Upper Niger basin.
Attribute Description Units
Slo1 Subbasin slope %
Len1 Longest path within the subbasin m
Sll Field slope length m
Csl Subbasin tributary reach slope m
Wid1 Subbasin tributary reach width m
Dep1 Subbasin tributary reach depth m
Lat Latitude of the subbasin centroid -
Long Longitude of the subbasin centroid -
Elev Mean elevation of the subbasin m
ElevMin Minimum elevation of the subbasin m
ElevMax Maximum elevation of the subbasin m
Shape_Leng Subbasin perimeter m
Shape_Area Subbasin area m2
a
P Average annual precipitation on the subbasin (mm) mm
Forest Proportion of forest on the subbasin %
ACRISOL Proportion of ACRISOL on the subbasin (%) %

a
Calculated on the period 1981-2000
2.3 Multivariate statistical analyses
Multivariate statistics used in this study are
Principal Components Analysis (PCA) and Cluster
Analysis (CA), and were performed under R package
FactoMineR[33, 34], version 1.28.
PCA and CA are frequently used in hydrological
studies [25, 26,35], and commonly applied in a pre-
processing of a set of variables prior to the
classification, to provide a convenient lower-
dimensional summary of thedataset, or as a
classification tool itself. PCA reduces a dataset
containing a large number of variables to a dataset
containing fewer new variables that are linear
combinations of the original ones. These linear
combinations are chosen to represent the maximum
possible fraction of the variability contained in the
original data and are called Principal Components
(PCs). CA attempts to separate observations into
groups of similar characteristics called clusters.
The methodology utilized in this study was
based on the Hierarchical Clustering on Principal
Components (HCPC) function proposed by [36]. This
method combines three exploratory data analysis
methods, Principal Component methods, Hierarchical
Clustering and partitioning, to improve data
analysis.The chosen Principal Components method is
the PCA, because retained CAs are quantitative
variables. PCA was used herein as a pre-process for
clustering, i.e., the hierarchical clustering is solely
built on the determined PCs. In that case, the
clustering is more stable than the one obtained from
original variables [36]. Input variables, i.e., CAs,
were standardized because they are not measured on
comparable scales. The appropriate number of PCs
was chosen based on the scree plot technique [37].
Then, a hierarchical agglomerative clustering was
performed on the PCs previously determined. The
measure of distance between data points was based
on the Euclidean distance (the same was used in
PCA) and the agglomerative method for merging two
clusters used the Ward's criterion. According to this
criterion, the total inertia (variability) is decomposed
in within-group and between-group inertia, and the
pair of groups to be merged is chosen that minimizes
the growth of within-group inertia. Equation (3) gives
the formula for calculating the total inertia of a
dataset:
     
2
2 2
1 1 1 1 1 1 1 1
,
q qI IQ Q QK K K
qkiqk k q k iqk qk
k q i k q k q i
x x I x x x x
       
       (3)
Total inertia = Between-group inertia + Within-group inertia
Where xiqk is the value of the variable k for the
individual i of the cluster q, qkx is the mean of the
variable k for cluster q, kx is the overall mean of
variable k and Iq is the number of individuals in
cluster q.
The last step consists in choosing the appropriate
number of clusters when it is not preassigned, that is,
the stopping point of clustering that maximizes
similarity within clusters and maximizes dissimilarity
between clusters. HCPC function suggests an
―optimal‖ number Q of clusters when the decrease in
within-group inertia between Q - 1 and Q is from far
greater than the one between Q and Q + 1 (see
[36]for a thorough description of the HCPC
function). Results of HCPC function can be presented
in different ways: (1) a factor map, which displays
results of the hierarchical clustering on the map
induced by the first PCs, (2) a 2-dimensional
dendrogram or hierarchical tree, and (3) a 3-
dimensional dendrogram in which the hierarchical
tree is incorporated into the factor map. The latter
representation can solely be used to get an integrated
visualization of the dataset. However, dispersion of
data points is somehow masked in that way.
Therefore, the factor map was presented in the results
section for a better visualization of individuals’
dispersion on the plan formed by PCs, while the
hierarchical treeoffers a good insight of the
variability increase between clusters.
III. Results
3.1 Catchments clustering
It is interesting to briefly describe the
intermediate result of PCA. It permitted to determine
2 PCs that explain 81.33% (63.84% for Dim1 and
17.49% for Dim2) of the total variance of the original
data set. The subsequent clustering was then
performed on these PCs. Results are presented on
Fig. 2 and Fig. 3.

Figure 2: Hierarchical clustering representation on the map induced by the first 2 Principal Components on the
Upper Niger basin. Catchments are colored according to the cluster they belong to, the barycenter of each
cluster is represented by a square and Dim1 and Dim2 are the first two Principal Components on which the
hierarchical clustering is built.
Figure 3: hierarchical clustering of the Upper Niger catchments. On the hierarchical classification or tree, each
rectangle represents a cluster of similar catchments. The barplot(inertia gain) gives the decrease of within-group
variability with increasing number of clusters.

Table 3: Description of hierachical clusters. In bold, positive v.test value indicating that the variable has a value
greater than the overall mean, and in italic, negative v.test value indicating that the variable has a value smaller
than the overall mean . All v.test values are significant at the probability p = 0.05
Variable v.test Mean in the category Overall mean p-value
Cluster 2
Long 2.01 - 6.43 - 8.32 0.045
Elev - 1.97 376.51 454.59 0.048
Cluster 3
Elev 2.63 520.42 454.59 0.0084
ElevMin 2.58 344.6 314 0.0097
Slo1 2.48 5.6 4.21 0.0131
Forest 2.35 79.01 52.54 0.0186
ElenMax 2.34 1219.2 1029.78 0.0194
Csl 2.24 0.18 0.13 0.0252
Acrisol 2.01 62.06 46.36 0.0442
Long - 2.49 - 9.81 -8.32 0.0126
IV. Discussion and conclusions
Overall, results of this study showed that the
Upper Niger can be classified into 2 major clusters of
similar catchments based on physiographic
characteristics. In addition, topographic variability
and geographical position of the subcatchment were
demonstrated to exert a stronger control on separating
clusters, and permitted to propose a kind of
nomenclature of clusters: the group of easterly flat
catchments assigned to the Bani, and the one of
westerly hilly catchments, assigned to the Upper
Niger subbasin. The latter is further characterized by
the dominance of Forest and ACRISOL as the major
soil type. These results expectedly answer the
questions posed at the beginning of this work.
However, due to limited availability of literature on
this area, it is difficult to show how these results fit in
with existing knowledge on that topic. A broader
comparison can only be made about the dominant
controls on similarity in different contexts. For
instance, [26]demonstrated that topographic
parameters (e.g., mean stream slope, minimum
elevation, and maximum elevation) provide the major
categorization of catchments of the equatorial Nile,
and proposed the same nomenclature of flat and hilly
regions. Likewise, [29]showed that the whole Niger
basin is close to the group of basins characterized by
topographic parameters (hypsometry and mean
elevation), which can be considered as the major
driving forces of its hydrosedimentary response.
Nonetheless, it is important to note that no
cluster analysis can produce a definitive classification
because the results are depending on the dataset used
and other kind of subjective choices (choice of
classification algorithm and distance metric, [25]). It
is also acknowledged that the actual limitation that
arose within this study was the absence of
geologicaldescriptors, limiting thus our
understanding of subsurface controls. In spite of the
limitations discussed above, these are encouraging
results, showing on one hand the relevance of
physical characteristics to give information about the
spatial dissimilarity characterizing a large tropical
river basin, and on the other, the value of statistical
analyses (such as the HCPC function) as a pertinent
tool for exploring similarity among catchments.
Concerning the assumption of correspondence
between physical and functional similarity made in
this study, [18]pointed out that this assumption may
not always be verified. Further studies can try to find
out its validity in the present case study, by
evaluating, for instance, the performance of a
regionalization method to transfer information within
and between clusters. The use of other similarity
concepts (such as similarity indices) applied to the
same catchments could also give a good platform of
discussion.
V. Acknowledgements
This study was funded by the German Ministry
of Education and Research (BMBF) through the West
African Science Service Centre on Climate Change
and Adapted Land Use (WASCAL; www.wascal.org)
that supports the Graduate Research Program Climate
Change and Water Resources at the University of
Abomey-Calavi. We would like also to thank
AGRHYMET Regional Centre for providing
additional funds through the French Global
Environment Facility (FFEM/CC) project.
References
[1.] H. Bormann, B. Diekkrüger, Possibilities
and limitations of regional hydrological
models applied within an environmental

change study in Benin (West Africa),
Physics and Chemistry of the Earth, Parts
A/B/C, 28(33–36), 2003, 1323-32.
[2.] E. Kapangaziwiri, D.A. Hughes, T.
Wagener, Incorporating uncertainty in
hydrological predictions for gauged and
ungauged basins in southern Africa,
Hydrological Sciences Journal, 57(5), 2012,
1000-19.
[3.] D. Mazvimavi, A.M.J. Meijerink, H.H.G.
Savenije, A. Stein, Prediction of flow
characteristics using multiple regression and
neural networks: A case study in Zimbabwe,
A/B/C, 30(11–16), 2005, 639-47.
[4.] M. Minihane. Estimating mean monthly
streamflow in the lugenda river, northern
mozambique. Putting prediction in
ungauged basins into practice: Canadian
Water Resources Association; 2013. p. 185-
96.
[5.] P. Ndomba, F. Mtalo, A. Killingtveit,
SWAT model application in a data scarce
tropical complex catchment in Tanzania,
A/B/C, 33(8–13), 2008, 626-32.
[6.] M. Hrachowitz, H.H.G. Savenije, G.
Blöschl, J.J. McDonnell, M. Sivapalan, J.W.
Pomeroy, B. Arheimer, T. Blume, M.P.
Clark, U. Ehret, F. Fenicia, J.E. Freer, A.
Gelfan, H.V. Gupta, D.A. Hughes, R.W.
Hut, A. Montanari, S. Pande, D. Tetzlaff,
P.A. Troch, S. Uhlenbrook, T. Wagener,
H.C. Winsemius, R.A. Woods, E. Zehe, and
C. Cudennec, C., A decade of Predictions in
Ungauged Basins (PUB)—a review.
Hydrological Sciences Journal, 58 (6), 2013,
1–58.
[7.] T. Wagener, H.S. Wheater, Parameter
estimation and regionalization for
continuous rainfall-runoff models including
uncertainty, Journal of Hydrology, 320(1–
2), 2006, 132-54.
[8.] A. B¨¢rdossy, Calibration of hydrological
model parameters for ungauged catchments,
Hydrology and Earth System Sciences
Discussions, 2006.
[9.] R. Merz, G. Blöschl, Regionalisation of
catchment model parameters, Journal of
Hydrology, 287(1–4), 2004, 95-123.
[10.] J. Parajka, R. Merz, G. Blöschl, A
comparison of regionalisation methods for
catchment model parameters, Hydrol Earth
SystSci, 9(3), 2005, 157-71.
[11.] G. Blöschl, Scaling in hydrology,
Hydrological Processes, 15(4), 2001, 709-
11.
[12.] C. Harman, M. Sivapalan, A similarity
framework to assess controls on shallow
subsurface flow dynamics in hillslopes,
Water Resources Research, 45(1), 2009, n/a-
n/a.
[13.] T. Wagener, M. Sivapalan, P. Troch, R.
Woods, Catchment Classification and
Hydrologic Similarity, Geography Compass,
1(4), 2007, 901-31.
[14.] G. Blöschl, Scaling and Regionalization in
Hydrology, Elsevier; 2011.
[15.] G. Laaha, J.O. Skøien, G. Blöschl, Spatial
prediction on river networks: comparison of
top-kriging with regional regression,
Hydrological Processes, 28(2), 2014, 315-
24.
[16.] J.O. Skøien, G. Blöschl, Spatiotemporal
topological kriging of runoff time series,
Water Resources Research, 43(9), 2007, n/a-
n/a.
[17.] G. Ali, D. Tetzlaff, C. Soulsby, J.J.
McDonnell, R. Capell, A comparison of
similarity indices for catchment
classification using a cross-regional dataset,
Advances in Water Resources, 40, 2012, 11-
22.
[18.] L. Oudin, A. Kay, V. Andréassian, C.
Perrin, Are seemingly physically similar
catchments truly hydrologically similar?,
Water Resources Research, 46(11), 2010,
n/a-n/a.
[19.] A. Bárdossy, G.G.S. Pegram, L. Samaniego,
Modeling data relationships with a local
variance reducing technique: Applications in
hydrology, Water Resources Research,
41(8), 2005, n/a-n/a.
[20.] Y. He, A. Bárdossy, E. Zehe, A review of
regionalisation for continuous streamflow
simulation, Hydrol Earth SystSci, 15(11),
2011, 3539-53.
[21.] P.F. Dornes, B.A. Tolson, B. Davison, A.
Pietroniro, J.W. Pomeroy, P. Marsh,
Regionalisation of land surface hydrological
model parameters in subarctic and arctic
environments, Physics and Chemistry of the
Earth, Parts A/B/C, 33(17–18), 2008, 1081-
9.
[22.] I. Masih, S. Uhlenbrook, S. Maskey, M.D.
Ahmad, Regionalization of a conceptual
rainfall–runoff model based on similarity of
the flow duration curve: A case study from
the semi-arid Karkheh basin, Iran, Journal of
Hydrology, 391(1–2), 2010, 188-201.
[23.] M. Sivapalan, M.A. Yaeger, C.J. Harman,
X. Xu, P.A. Troch, Functional model of
water balance variability at the catchment
scale: 1. Evidence of hydrologic similarity

and space-time symmetry, Water Resources
Research, 47(2), 2011.
[24.] S. Tekleab, S. Uhlenbrook, Y. Mohamed,
H.H.G. Savenije, M. Temesgen, J.
Wenninger, Water balance modeling of
Upper Blue Nile catchments using a top-
down approach, Hydrol Earth SystSci,
15(7), 2011, 2179-93.
[25.] K. Sawicz, T. Wagener, M. Sivapalan, P.A.
Troch, G. Carrillo, Catchment classification:
empirical analysis of hydrologic similarity
based on catchment function in the eastern
USA, Hydrol Earth SystSci, 15(9), 2011,
2895-911.
[26.] J.M. KileshyeOnema, A.E. Taigbenu, J.
Ndiritu, Classification and flow prediction in
a data-scarce watershed of the equatorial
Nile region, Hydrol Earth SystSci, 16(5),
2012, 1435-43.
[27.] M. Di Prinzio, A. Castellarin, E. Toth, Data-
driven catchment classification: application
to the pub problem, Hydrol Earth SystSci,
15(6), 2011, 1921-35.
[28.] E. Coopersmith, M.A. Yaeger, S. Ye, L.
Cheng, M. Sivapalan, Exploring the physical
controls of regional patterns of flow duration
curves – Part 3: A catchment
classification system based on regime curve
indicators, Hydrol Earth SystSci, 16(11),
2012, 4467-82.
[29.] J. Raux, Y. Copard, B. Laignel, M. Fournier,
N. Masseï, Classification of worldwide
drainage basins through the multivariate
analysis of variables controlling their
hydrosedimentary response, Global and
Planetary Change, 76(3–4), 2011, 117-27.
[30.] ABN, Atlas Bassin du Niger (ABN:
Niamey, 2007).
[31.] Y. Brunet-Moret, P. Haoeron, J.P. Lamagat,
M. Molinier, Monographiehydrologique du
fleuveNiger.Tome 1-Niger superieur
(ORSTOM: Paris, 1986).
[32.] H. Sellami, Uncertaintyassessment of
hydrological impacts of climatic change in
smallmediterraneancatchments, doctoral
diss.,UniversitéCatholique de Louvain,
Belgique, 2014.
[33.] F. Husson, J. Josse, S. Lê, J. Mazet.
FactoMineR: Multivariate Exploratory Data
Analysis and Data Mining with R. R
package version 1.12, 2009. Available from:
http://factominer.free.fr.
[34.] S. Lê, J. Josse, F. Husson, FactoMineR: An
R Package for Multivariate Analysis,
Journal of Statistical Software, 25(1), 2008,
1-18.
[35.] A. Daigle, A. St-Hilaire, D. Beveridge, D.
Caissie, L. Benyahya, Multivariate analysis
of the low-flow regimes in eastern Canadian
rivers, HydrolSci J, 56(1), 2011, 51-67.
[36.] F. Husson, J. Josse, J. Pages, Principal
component methods - hierarchical clustering
- partitional clustering: why would we need
to choose for visualizing data?,technical
report (Agrocampus: France, 2010).
[37.] I.T. Jolliffe, Principal Component Analysis
(Springer, 2002).

Catchment classification: multivariate statistical analyses for physiographic similarity in the Upper Niger Basin

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Catchment classification: multivariate statistical analyses for physiographic similarity in the Upper Niger Basin (20)

Recently uploaded (20)

Catchment classification: multivariate statistical analyses for physiographic similarity in the Upper Niger Basin