Vous êtes sur la page 1sur 18

Hydrological Sciences Journal

ISSN: 0262-6667 (Print) 2150-3435 (Online) Journal homepage: http://www.tandfonline.com/loi/thsj20

Producing time series of river water height by


means of satellite radar altimetry—a comparative
study

Emmanuel Roux , Joecila Santos da Silva , Augusto Cesar Vieira Getirana ,


Marie-Paule Bonnet , Stéphane Calmant , Jean-Michel Martinez & Frédérique
Seyler

To cite this article: Emmanuel Roux , Joecila Santos da Silva , Augusto Cesar Vieira Getirana ,
Marie-Paule Bonnet , Stéphane Calmant , Jean-Michel Martinez & Frédérique Seyler (2010)
Producing time series of river water height by means of satellite radar altimetry—a comparative
study, Hydrological Sciences Journal, 55:1, 104-120

To link to this article: http://dx.doi.org/10.1080/02626660903529023

Published online: 11 Mar 2010.

Submit your article to this journal

Article views: 548

View related articles

Citing articles: 15 View citing articles

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=thsj20

Download by: [177.15.120.2] Date: 14 August 2017, At: 09:59


104 Hydrological Sciences Journal – Journal des Sciences Hydrologiques, 55(1) 2010

THSJ

Producing time series of river water height by means of satellite


radar altimetry—a comparative study

Emmanuel Roux1, Joecila Santos da Silva3,4, Augusto Cesar Vieira Getirana2,3, Marie-Paule Bonnet2,
Producing time series of river water height by means of satellite radar altimetry—a comparative study

Stéphane Calmant4, Jean-Michel Martinez2 & Frédérique Seyler1


1
IRD, US140 ESPACE, Centre IRD de Guyane, BP165, Route de Montabo, 97323 Cayenne Cedex, French Guiana
emmanuel.roux@ird.fr
2
LMTG, IRD-CNRS-Université de Toulouse III, Toulouse, France
3
Programa de Engenharia Civil, COPPE/Universidade Federal do Rio de Janeiro, PO Box 68506, CEP 21945-910,
Rio de Janeiro, Brazil
4
LEGOS, IRD-CNRS-CNES-Université de Toulouse III, Toulouse, France
Received 12 February 2008; accepted 5 August 2009; open for discussion until 1 August 2010
Downloaded by [177.15.120.2] at 09:59 14 August 2017

Citation Roux, E., Santos da Silva, J., Vieira Getirana, A. C., Bonnet, M.-P., Calmant, S., Martinez, J.-M. & Seyler, F. (2010)
Producing time series of river water height by means of satellite radar altimetry—comparative study. Hydrol. Sci. J. 55(1), 104–120.

Abstract Satellite radar altimetry is complementary to in situ limnimetric surveys as a means of estimating the
water height of large rivers, lakes and flood plains. Production of water height time series by satellite radar
altimetry technology requires first the selection of radar ground target locations corresponding to water body
surfaces under study, i.e. the definition of “virtual limnimetric stations”. We propose to investigate qualitative
and quantitative differences between three representative virtual station creation methodologies: (a) a fully
manual method, (b) a semi-automatic method based on a land cover characterization that allows the water body
surface under study to be located; and (c) an original fully automatic procedure that exploits a digital elevation
model and an estimation of the river width. The results yielded by these three methods are comparable: maximum
absolute magnitudes of water height differences being 0.46, 0.26 and 0.15 m for, respectively, 95, 90 and 80% of
the water height values obtained. Moreover, more than 67% and 92% of time series jointly produced by the
methods present root mean square differences lower than 20 and 50 cm, respectively. The results show that the
fully automatic method developed herein provides as reliable results as the fully manual one. This opens the way
to use of satellite radar altimetry for the generation of water height time series on a large scale, and considerably
extends the applicability of satellite radar altimetry in hydrology.
Key words satellite radar altimetry; Amazon basin; water height time series; image classification; digital elevation model

Produire des séries temporelles de hauteur d’eau des cours d’eau grâce à l’altimétrie radar
satellitaire—une étude comparative
Résumé L’altimétrie radar satellitaire est complémentaire des relevés limnimétriques in situ pour l’estimation de
la hauteur d’eau des grands fleuves, lacs et plaines d’inondation. L’obtention de séries temporelles de hauteur
d’eau par altimétrie radar satellitaire requiert la sélection de cibles radar au sol correspondant aux surfaces en eau
d’intérêt, i.e. la définition de “stations limnimétriques virtuelles”. Nous proposons de comparer qualitativement
et quantitativement trois méthodes de création de stations virtuelles: (a) une méthode entièrement manuelle; (b)
une méthode semi-automatique basée sur une caractérisation de l’occupation du sol qui permet de localiser les
cours d’eau; et (c) une procédure originale entièrement automatique qui exploite un modèle numérique de terrain
et une estimation de la largeur de la rivière. Les résultats fournis par ces trois méthodes sont comparables: la
différence absolue maximale de hauteur d’eau est de 0.46, 0.26 et 0.15 m pour, respectivement, 95, 90 et 80% des
valeurs de hauteur obtenues. De plus, plus de 67% et de 92% des séries temporelles produites conjointement par
les méthodes présentent des erreurs quadratiques moyennes inférieures à 20 et à 50 cm, respectivement. Ces
résultats montrent que la méthode entièrement automatique proposée ici fournit des résultats aussi fiables qu’une
méthode entièrement manuelle. Il est par conséquent envisageable d’exploiter l’altimétrie radar satellitaire afin
de générer des séries temporelles de hauteur d’eau à grande échelle, ce qui étend considérablement l’applicabilité
de l’altimétrie radar satellitaire en hydrologie.
Mots clefs altimétrie radar satellitaire; basin Amazonien; séries temporelles de hauteur d’eau; classification d’images; modèle
numérique de terrain

ISSN 0262-6667 print/ISSN 2150-3435 online


© 2010 IAHS Press
doi:10.1080/02626660903529023
http://www.informaworld.com
Producing time series of river water height by means of satellite radar altimetry—a comparative study 105

INTRODUCTION echo, the so called tracker (onboard) or retracker


(ground reprocessing) algorithm. Indeed, contrary to
Water resources management in the great tropical
the oceanic domain that the system was tuned for, the
basins of the world is a complex issue. For example,
shape of the energy distribution in the reception win-
in the Amazon basin, the largest river basin of the
dow is highly variable over continental waters. Frap-
world, the hydrological regimes of the main tributar-
part et al. (2006) have shown that, among the four
ies, their typology and water availability, are affected
retrackers used for the ground processing of the
by human economic activities, such as deforestation,
ENVISAT data, ICE-1 (Wingham et al., 1986) was
mining, and cereal monoculture. Water use must be
the one performing best over rivers. We therefore
regulated between agriculture and rural development,
used the ICE-1 solution for this work. Travel time
energy and transport, water supply and sanitation,
picked up by the retracking algorithm is then con-
environment and tourism. Many tropical basins are
verted into satellite–ground distance by taking into
trans-boundary and conflicting situations arise that
account corrections associated with electromagnetic
may increase in time. In the humid tropics, they
pulse propagation through the atmosphere and iono-
present a dense river network, very variable in shape,
sphere, and solid Earth tides. Orbitography provides
length, width and volume. The water rise during
the satellite height relative to a reference ellipsoid.
flooding varies spatially from 2 to 20 m, but is regu-
The difference between this height and the distance
Downloaded by [177.15.120.2] at 09:59 14 August 2017

lar and persistent in the downstream part of the main


derived from the radar measurement, including the
tributaries. The water height has well-marked annual
aforementioned corrections, gives the height of the
variations which tend to increase in amplitude with
water surface with respect to the reference ellipsoid.
climate change. The Amazon basin includes 50 000 km
Altimetry satellites are kept on a fixed ground track
of waterway, and, despite a low population density,
in order to overfly the same locations at regular time
water quality degradation occurs in several parts of
intervals, called repeat cycles. The repeat cycle
the basin. In such a context, as stated by van Griens-
depends on the orbit parameters selected for the mis-
ven et al. (2006): “Two instruments play an import-
sion. For radar altimetry missions, this typically
ant role in integrated impact assessment: monitoring
ranges from 10 to 35 days. The ability of radar altim-
and modelling”. The Amazon basin covers about 6
eters to monitor stage elevation of continental water
000 000 km2. Sixty-three percent of this area is situ-
surfaces has been demonstrated (Birkett, 1998; De
ated in Brazil and is monitored by 435 limnimetric
Oliveira et al., 2001; Calmant & Seyler, 2006;
stations (ANA, 2008). Installing and maintaining the
Frappart et al., 2006). In the Amazon basin, satellite
operational networks of hydrological in situ stations
data very significantly increase the number of water
in remote basins is complicated and very costly.
stage measurement points. The intersections between
Human-based measurements are subjective, and the
the river system and the satellite tracks are about ten
completeness of data records and their quality
times denser than the actual in situ gauging network.
depend on operator availability and assiduity.
Moreover, if radar altimetry sampling periods, which
In this context, radar altimetry is a promising
correspond to satellite cycle durations, range from 10
complementary tool for water-level monitoring in
(TOPEX/POSEIDON) to 35 days (ENVISAT), it has
large river basins, such as the Amazon basin. Basi-
been shown that daily water stage estimation can be
cally, radar altimetry measures the two-way travel
provided by adequate interpolation techniques asso-
time of a radar pulse between the satellite antenna
ciated with in situ data (Roux et al., 2008). Addition-
and the Earth surface at the nadir of the altimeter.
ally, satellite altimetry already allows river discharge
This measurement is performed through recording
to be estimated by establishing rating curves using
the energy bounced back by the reflecting facets
altimetric heights and either remotely measured dis-
within the altimeter footprint (typically a few km2).
charge (Coe & Birkett, 2004; Kouraev et al., 2004),
Note that the radar wavelength used, namely K-band,
or local model-based discharge estimates (Leon
was selected because it does not penetrate water,
et al., 2006). Also, lake heights can be predicted
making energy reflected by water surfaces predomi-
(Coe & Birkett, 2004) and flood-plain water storage
nate in the total energy received onboard the satellite.
estimated (Frappart et al., 2005).
This energy is processed through a ∼0.1-μs reception
The exploitation of satellite altimeter data is per-
window. An important consideration is the algorithm
formed by defining “virtual limnimetric stations”.
used to derive the two-way travel time from the radar
Basically, a “virtual station” can be defined as any
106 E. Roux et al.
Downloaded by [177.15.120.2] at 09:59 14 August 2017

Fig. 1 Three-dimensional manual selection: (a) JERS-1 background image, one satellite ground track and the rough geo-
graphical selection of the intersection; (b) data selection into the plane perpendicular to the river direction; (c) locations of
selected data; (d) water height time series corresponding to the selected data. The lines link the medians of overpass values.

intersection between a water body surface and a sat- selection method based on a land-cover characteriza-
ellite ground track (Fig. 1). Ground track locations tion obtained with the JERS-1 SAR image classifica-
show significant variations that can reach 1 km either tion methodology proposed by Martinez & Le Toan
side of theoretical ground tracks. Thus, in the follow- (2007). The third method is original and fully auto-
ing, a “track” includes all satellite radar ground tar- matic. It only exploits a digital elevation model
gets and has approximately 2 km width. Several (DEM) of the studied area and basic knowledge on
techniques have been considered in the literature to the river width along the drainage network. The
define these virtual stations (see the section below on results of the investigations are then presented and
literature review of “virtual station” definition). discussed. Finally a conclusion is drawn.
However, these procedures are not extensively dis-
cussed and, to our knowledge, have not been quanti-
LITERATURE REVIEW OF “VIRTUAL
tatively compared in terms of water height time
STATION” DEFINITION
series produced. In this context, it is difficult to com-
pare and/or jointly analyse water height time series Birkett (1998), using AVISO Topex/Poseidon GDR
provided by different selection methods. This paper data, proposed a multi-criteria selection methodol-
first reviews altimetric data selection procedures ogy. The author: (a) supposes that river surfaces are
found in the literature. Then, three procedures are associated with a radar backscatter coefficient greater
chosen for investigation and the methodology for than 20 dB; (b) uses 1:1 000 000 Operational Navi-
comparison is detailed. The first method is the fully gation Charts (ONC) which roughly indicate the
manual three-dimensional (3-D) selection method bank-to-bank geographical crossing points between
used in Santos da Silva et al. (2007a,b). The second satellite tracks and rivers; (c) identifies the water sur-
is a semi-automatic and precise geographical face by considering the uniformity of the elevation
Producing time series of river water height by means of satellite radar altimetry—a comparative study 107

measurements, i.e. the along-track standard deviation method (Martinez & Le Toan, 2007). Land-cover
of water height variations; (d) identifies the water sur- characterization involves nine classes: (1) Permanent
face by utilizing an elevation filter (manual selection water (permanent lakes), (2) Occasionally flooded
based on ONC map heights and observation of the forest, (3) Never flooded forest, (4) Never flooded
altimetric elevations along-track); and (e) separates savannah, (5) Occasionally flooded non-vegetated/
river channel from nearby flood plain by examining herbaceous areas, (6) Always flooded forest, (7)
the seasonal variation of radar backscatter and eleva- Occasionally flooded savannah, (8) Permanent water
tion (river channels showing higher-amplitude eleva- (main rivers), and (9) Totally submersed vegetation.
tion variations and lower-amplitude backscatter Intersections between classes of interest and satellite
variations, and flood plains having the opposite trend). ground tracks are identified, and rectangular selec-
De Oliveira et al. (2001) and Mercier (2001) tion windows are defined manually.
tested three criteria in order to automatically select In Leon (2006), virtual stations are defined by
altimetric data related to water surfaces: (a) the “lati- manually digitizing rectangular windows that sur-
tudinal measurement density”, i.e. the percentage of round the intersections between the satellite tracks and
valid measurements inside a 1-km-wide window the rivers located by means of JERS-1-SAR images.
(along the latitudinal dimension) moved along the In Kouraev et al. (2004), the authors performed
satellite ground track (selection threshold is set to a more rigorous geographical selection of the alti-
Downloaded by [177.15.120.2] at 09:59 14 August 2017

50%); (b) the “water level fluctuation”, by stating metric data by considering 28.5-m resolution Landsat
that water surfaces have, in the Amazon Basin, an images and selecting only parts of the satellite
annual amplitude of at least 1 m; (c) the “backscatter ground tracks that cover the main channel of the river
coefficient”, s°, with the condition s° > 20 dB. How- (Ob’ River).
ever, Mercier (2001) shows that selection criteria that Zakharova et al. (2006) used JERS-1-SAR
only take into account the intrinsic information of images and defined a virtual station by a rectangular
altimetric data do not permit: (i) selection of water window. However, several rectangular windows can
surfaces with a geographical resolution lower than be defined when islands are present.
1 km, or (ii) permanent water surface to be distin- The previous manual data selection procedures
guished from occasionally flooded areas. As a result, are two-dimensional. But a simple (even if precise)
Mercier (2001) performs precise geographical delim- geographical selection of altimetric measurements
itations of areas within which the satellite’s nadir that “fall” in the water surface, using one image, has
locations (satellite ground tracks) and water (lakes two main drawbacks: (a) the water surface is only
and rivers) intersect. For lakes, the author uses the known at the image capture time. So, if the chosen
lake contours provided by the CIA World Data Bank image corresponds to high water stage, data related
II, available in the Generic Mapping Tool (GMT) to non-flooded areas during low water stage are
open-source software, and automatically selects the selected and can be unreliable (see Fig. 1(b)). How-
altimetric measurements. For large rivers such as the ever, if the chosen image corresponds to low water
Amazon River, Mercier (2001) uses Synthetic Aper- stage, possibly valid data related to flooded areas
ture Radar (SAR) images provided by the JERS-1 during a high water stage period are ignored; (b) 2-D
satellite, without, however, presenting in detail what geographical selection does not permit identification
his selection criteria and tools are (form of the selec- of the off-ranging phenomenon described by Berry
tion, selected pixels’ features, etc.). (2000). Radar echo retracking is proposed in order to
Berry et al. (2005) also used the CIA World overcome this inland-specific artefact (Berry, 2000;
Data Bank II database to define masks and select Frappart, 2005). This phenomenon often occurs
inland altimetric data. These masks are set slightly when the satellite nadir is not yet (or no more) above
wider than maximum water extent and a detailed the river surface and the altimeter is already (or still)
Amazon River mask is also defined. Then retracking locked on the river surface, leading to an underesti-
of radar echoes is performed by means of an expert mation of the water height (see Fig. 1(b)). This is
system approach (Berry, 2000) and echoes present- referred to as a “locking” and a “hooking” phenome-
ing a “leading edge and a significant power within non in Santos da Silva et al. (2007a,b) and Frappart
the waveform” are selected (Berry, 2006). (2006), respectively.
Frappart et al. (2005) used a land-cover charac- In Santos da Silva et al. (2007a,b) the authors
terization based on a JERS-SAR image classification use a 3-D procedure for the altimetric data selection.
108 E. Roux et al.

First a 2-D geographical selection of the satellite The data set used in this study ranges from 1
track–river intersection is performed by using JERS- October 2002 to 12 November 2006, corresponding
SAR images (i.e. a selection in the longitude × lati- to a four-year-long period. The repeat cycle of
tude plane). Altimetric heights associated with the ENVISAT being of 35 days, data from 35 to 42 satel-
selected radar nadir locations are then projected onto lite cycles have been exploited.
the plane perpendicular to the main direction of the
river (this direction is characterised manually), i.e.
Methods for virtual station definition (altimetric
the river section plane. Finally, a second data selec-
data selection)
tion is performed manually in the river section plane.
This 3-D procedure has been developed to avoid the Three data selection methods have been chosen for
two previously mentioned drawbacks of the 2-D geo- investigation, according to: (a) their representative-
graphical selection. This is given in more detail in the ness of the methodologies already described in the
Method Section below, as it is one of the three meth- literature, and (b) their level of automation:
ods extensively studied herein.
All the previously mentioned selection proce- Three-dimensional manual selection (MANU)
dures are summarized in Table 1. In the following, First a map or an image is selected to identify water
we present in detail, and compare, three methods for surfaces. Here, the high-water stage mosaic of the
Downloaded by [177.15.120.2] at 09:59 14 August 2017

altimetric data selection. Amazon basin provided by the Global Rain Forest
Mapping (GRFM) Project (Freeman et al., 2002) is
used. SAR data characteristics used for this mosaic
METHOD
construction can be summarized as follows (Hess
Area of study et al., 2003): L-band (1275 MHz) frequency and HH
polarization, 34 to 43 degrees for incidence angle,
The three proposed methodologies have been applied
pixel size resampled at 100 m, acquisition period
on the entire Rio Negro basin (Fig. 2). With a mean
from May to August 1996 for the high-water stage
annual discharge of 29 000 m3 s-1, the Negro River is
mosaic and from October to November 1995 for the
the second largest discharge tributary of the Amazon
low-water stage mosaic, –15 to –20 dB for noise
after the Madeira River. The drainage area of the Rio
equivalent s°, ±0.2 dB calibration uncertainty.
Negro is nearly 700 000 km2 and extends through
The MANU selection method consists of a two-
parts of Colombia, Venezuela and Brazil. The Negro
step manual procedure:
River basin is consequently a typical great and
remote trans-boundary tropical basin. Most of the (1) A geographical delimitation of the intersection
published studies related to satellite radar altimetry between the water surface under study and the
consider the Rio Negro basin. ground track of the satellite is realized by digi-
tizing a shape that can be complex (multiple
polygon, exclusion of non-flooded areas,
Altimetric data
islands, etc.) (cf. Fig. 1(a)). Then, the user has
In the following, we consider the 18-Hz Ku-range to indicate the main flow direction of the river
ENVISAT data contained in the Geophysical Data by pointing two points along the river. Heights
Records (GDRs) delivered by the Archiving, Valida- associated with the radar nadir locations falling
tion and Interpretation of Satellite data in Oceanogra- in the previously defined zone are automatically
phy (AVISO) at the Centre National d’Etudes projected onto the plane perpendicular to the
Spatiales (CNES). In particular, the ICE-1 retracked main direction of the river, i.e. the river section
version of these data is considered. In fact, among the plane. This data representation permits correct
retrackers Ocean, ICE-1, ICE-2 and Sea Ice, the ICE- visualization of the altimetric heights associated
1 algorithm (Wingham et al., 1986) gave the most with all satellite passes (cf. Fig. 1(b)).
accurate results for continental hydrology studies in (2) A second selection is manually performed in
Frappart et al. (2006). The corrections applied to the river section plane (cf. Fig. 1(b)) (the same
these data are on-board instrumental drifts and biases, tool for shape digitizing is used). Data associ-
ionospheric, dry and wet troposphere, solid earth and ated with a low intra-pass variability and a high
poles tide corrections (Chelton et al., 2001). inter-pass variability are selected. These criteria
Downloaded by [177.15.120.2] at 09:59 14 August 2017

Table 1 Summary of altimetric data selection methods found in the literature.


References Satellite(s) Region(s) Information used Selection criterion of free surface water: Map(s)/Image(s)

Derived from geographical Derived from altimetric data


information
Birkett (1998) TOPEX/POSEIDON Paraguay River Tracker waveforms Bank to bank crossing points Minimum number of continuous Operational
(T/P) Amazon River Backscatter between satellite track and waveforms = 2 navigation charts
coefficient river Radar backscatter coefficient >
Altimetric heights ONC elevation values 20 dB
Nadir coordinates Uniformity of the along-track
elevation measurements
Discrimination of river channel
and nearby flood plain by
means of seasonal variation of
radar backscatter coeff. and of
elevation
De Oliveira, T/P Amazon basin Altimetric heights not specified Measurement density not specified
Mercier et al. Backscatter coefficient Water level fluctuation
(2001) Nadir coordinates Backscatter coefficient
Mercier (2001) T/P Great Lakes of North Altimetric heights Precise automatic (lakes) and Measurement density CIA World Data
America (Superior, Backscatter manual (rivers) selection Water level fluctuation Bank II (WDB II)
Michigan, Huron, Erie, coefficient Backscatter coefficient JERS1-SAR
Ontario) Nadir coordinates
Issykkul Lake
(Kyrgyzstan)
11 European lakes
Aral Sea
Amazon basin
Frappart (2005) T/P Negro River basin Nadir coordinates Rectangular windows defined JERS1-SAR
according to a JERS1-SAR- images
based land cover classification
characterization
Leon (2006) ENVISAT Upper Negro River Nadir coordinates Rectangular windows defined JERS1-SAR
manually
Kouraev et al. T/P Ob’ River Nadir coordinates Manual selection of the part of 28.5-m resolution
(2004) the satellite ground tracks Landsat images
that cover the main channel
of the Ob’ River
Berry et al. (2005) ERS-1/2 World Tracker waveforms Masks for lakes and the Radar echoes retracking WDB II
Berry (2006) T/P Amazon River Nadir coordinates Amazon River and Presence of a leading edge
ENVISAT automatic selection Significant power within the
waveform
Zakharova et al. T/P Amazon River Nadir coordinates Manual definition of JERS-SAR
(2006) rectangular windows
(several windows if islands
Producing time series of river water height by means of satellite radar altimetry—a comparative study

make the intersection


between the satellite track
and the river complex)
Silva et al. ENVISAT Negro River basin Nadir coordinates Manual definition of possibly Manual selection of data JERS-SAR
109

(2007a,b) T/P Altimetric heights complex geographical presenting low (resp. high) intra-
selection (resp. inter-)pass variability
110 E. Roux et al.
Downloaded by [177.15.120.2] at 09:59 14 August 2017

Fig. 2 Negro River basin delimitation, main tributaries and virtual stations defined by the three selection methods (MANU,
AUTO-JERS and AUTO-SRTM). The river network is extracted from the land-cover characterization (Class 8) obtained
from JERS mosaics (Martinez & Le Toan, 2007). Grey oblique lines represent ENVISAT satellite ground tracks.

are also used by Birkett (1998), De Oliveira generally with any methods based only on geograph-
et al. (2001) and Mercier (2001) (see Table 1 ical selection (even if most of the published
and section on Literature Review). approaches only require rectangular selections).
In the present study, geographical selections (or
In this study, 43 virtual stations defined by Santos da
masks) are made one pixel larger than Class 8 areas
Silva et al. (2007a,b) are taken into account. For
that intersect with satellite ground tracks (note that, as
comparison purposes, it would have been “theoreti-
satellite ground track location shows significant vari-
cally” possible to manually define a virtual station
ations that can extend 1 km either side of the theoreti-
for each selection performed by the two other
cal track, a “track” is approx. 2-km wide in the
approaches. We used a significant number of previ-
following, as it includes all the satellite ground target
ously defined stations (Santos da Silva, 2007a,b) for
locations). Berry et al. (2005) also defined masks
two reasons: (a) the MANU selection technique is
slightly wider than maximum water extent. This
time consuming and performing a selection for every
allows virtual stations to be defined for narrow rivers
river–satellite track intersection would have been
and avoids the multiplication of virtual stations for
particularly tedious; and (b) virtual stations in Santos
parcelled out satellite track–narrow river intersec-
da Silva (2007a,b) were created independently of the
tions. In fact, such rivers can be characterized by dis-
objectives pursued in this comparative work and
continuous areas associated with the Class 8 label
allow the results to be as objective as possible.
(see, for example, Fig. 4). Additionally, intersections
whose surface is smaller than 150 000 m2 are filtered.
Precise geographical selection (AUTO-JERS) However, despite the one-pixel tolerance and the
This method is semi-automatic and is based on the area-based filter previously mentioned, a manual
land-cover characterization derived from a JERS-1 post-processing of the results has to be performed in
SAR image classification (Martinez & Le Toan, order to assemble several virtual stations related to
2007). Altimetric data that “fall” in pixels labelled the same river but separated by islands or river dis-
“Class 8: Permanent water (main rivers)” are continuities (see Fig. 5(b), (c) and (d) for examples).
selected (Fig. 3). This method can be compared with This post-processing is a time-consuming operation
the approach of Mercier (2001) for lakes and more that has been partially automated by means of a
Producing time series of river water height by means of satellite radar altimetry—a comparative study 111

Fig. 3 Two-dimensional precise geographical selection: (a) Eighth class of the JERS-1-based land-cover characterization
(permanent water: rivers), one satellite ground track and the geographical selection performed automatically; and (b) water
height time series corresponding to the selected data.
Downloaded by [177.15.120.2] at 09:59 14 August 2017

DEM-based selection (AUTO-SRTM) Hydro-


graphical networks can be obtained by means of a
DEM. We state that, for the river water height moni-
toring, it is possible to define a virtual station at any
intersection between the drainage network extracted
from a DEM and a satellite ground track. To test this
assumption, we used the Shuttle Radar Topography
Mission (SRTM) DEM at 90-m spatial resolution.
Flow accumulation for all the Negro River basin was
extracted with the r.terraflow.short function of the
GRASS GIS, a flow computation routine for massive
grids (integer version), with the Single Flow Direc-
tion (SFD) option. The r.terraflow function was
designed and optimized especially for massive data
sets, but uses well-known approaches in order to
Fig. 4 An example of a parcelled out intersection between compute flow direction and accumulation:
a satellite track and a narrow river. Performance of geo-
graphical selections one pixel larger than the strict inter- (1) r.terraflow first floods the terrain until all the
section provides a continuous selection area.
sinks (flat areas that do not spill out, including
one-cell pits) are filled;
BASH1 script calling GRASS (2006) geographical (2) flow directions are then computed using either
information system (GIS) functions for drawing, the Multiple Flow Direction model or the SFD
selection and concatenation of the virtual stations model. Both methods compute downslope flow
(defined by vectorial maps in the GIS). directions by inspecting the 3 × 3 window
It is worth noting that the choice of the JERS around each cell. The SFD method assigns a
mosaic for application of the MANU method is con- unique flow direction towards the steepest
sistent with the use of the land-cover characterization downslope neighbour;
exploited by the AUTO-JERS approach and makes (3) once flow directions are computed for every
the results of these two methods comparable. cell in the terrain, r.terraflow computes flow
accumulation by routing water using the flow
directions and keeping track of how much water
1
BASH is a free shell-compatible command language interpreter. flows through each cell. (For more detail on the
See http://tiswww.case.edu/php/chet/bash/bashtop.html. algorithms see Arge et al., 2003.)
112 E. Roux et al.
Downloaded by [177.15.120.2] at 09:59 14 August 2017

Fig. 5 Examples of virtual stations for the three altimetric data selection methods.

Once the flow accumulation is computed, the (river-related) altimetric data. For a given satellite
drainage network can be easily obtained by simply track–drainage network intersection, we compute the
considering the cells in which the flow accumulation is buffer size as a function of the flow accumulation. As
above a given threshold. The higher this threshold, the an intersection is defined by several cells (pixels) with
less dense the resulting drainage network, ignoring different flow accumulation values, we consider, for
small-sized rivers. In this study, the drainage network the ith intersection, the average of the flow accumula-
has been extracted by considering an accumulation tion values: Acc(i). The flow accumulation can be con-
threshold equal to one million cells, corresponding to sidered as analogous to discharge. In the literature, the
8100 km2 watersheds. This threshold was empirically river width W is currently linked to the river discharge
defined in order to make the SRTM-based drainage Q by the relationship W = aQb (see, for example, Pis-
network and Class 8 of the JERS-based land-cover tocchi & Pennington, 2006, for a recent application of
characterization visually coincide. At this step of the this relation and older references). Consequently, for
procedure, both the rivers extracted from the DEM and the ith intersection, the buffer radius r(i) corresponding
the satellite track–river intersections are one cell wide. to the half river width W(i) is defined by:
Consequently, we propose to define buffers around
satellite track–river intersections in order to define the β
geographical areas that possibly contain valid r ( i ) = 0.5W ( i ) = 0.5α Acc ( i ) (1)
Producing time series of river water height by means of satellite radar altimetry—a comparative study 113

where a and b are parameters to be estimated and The three methods described above were imple-
Acc(i) is the mean accumulation value for the ith sat- mented in BASH language scripts calling GRASS
ellite track–drainage network intersection. In order to GIS specific functions.
estimate a and b, values for W have been extracted
from transverse section measurements freely avail-
Comparison of methods
able from the Brazilian Water Agency (Agência
Nacional de Águas, ANA, 2007). Currently, 40 trans- The following aspects of the methods are
verse sections are available for the entire Negro River investigated:
basin. Then, to facilitate parameter estimation, we
(1) Number and locations of virtual stations.
considered the following linear form of equation (1):
(2) Station “morphologies”, i.e. shape and com-
plexity of geographical selections as a function
log (W ) = log (2r ) = log(a ) + b log (Acc ) (2) of the method. This point is addressed from a
qualitative point of view. [Indeed, even if we
Finally, log(a) and log(b) (i.e. a and b) were esti- focus on river monitoring, the aim of the paper
mated with the classical least-squares method in the is not to compare the three methodologies in
R (R Development Core Team, 2005) environment terms of river location identification but in
Downloaded by [177.15.120.2] at 09:59 14 August 2017

for statistical computing (R function lsfit). The terms of water height estimation. It is quite dif-
results are shown in Fig. 6. Estimation provided a ferent according to altimetric data characteris-
residual standard error (RSE) equal to 0.2997 and R2 tics (footprint size, off-ranging phenomenon,
= 0.9188. We found a = 0.0187 and b = 0.653, and etc.).] Then a quantitative comparison between
equation (1) can be written as: the three approaches is done in order to charac-
terize method differences in terms of water
W = 0.0187Q 0.653 (3) height and its variability. Methods are consid-
ered pair-wise, by performing paired sample
Equation (3) enables the river width to be estimated and one-sided Wilcoxon tests on overpass val-
for every satellite track–drainage network intersection ues. The Wilcoxon test is a non-parametric test
and, consequently, one can compute the buffer radius to compare two related samples and does not
(with equation (1)) that characterizes areas which require assumptions about the sample distribu-
possibly contain valid (river related) altimetric data. tions. The null hypothesis is that the distribution
Once parameters a and b are obtained, the creation of of the paired sample differences is symmetric
virtual stations for the entire basin is fully automatic. about the median (Hollander & Wolfe, 1973).

Fig. 6 (a) Logarithm of the river width as a function of the logarithm of the flow accumulation (points), and the model
estimate (line). R2 = 0.9188 and Residual Standard Error = 0.2997; and (b) river width as a function of the flow accumula-
tion (points), and the model estimate (line).
114 E. Roux et al.

Characteristic values computed and discussed


for the quantitative investigation are: max i | MAD(cS1i ) − MAD(cS2i ) | < d
(3) Number of selected points per satellite over-
pass. Water height derived from very few alti-
MAD(cSji) being the median absolute deviation
metric data can be erroneous. The extreme case
of values related to the ith overpass of Station
is when only one value is selected for a satellite
Sj.
overpass. In such a case, the height value can be
the “true” one, but it also can correspond to an
inconsistent value due to the off-ranging effect,
an altimeter failure, etc. RESULTS
(4) Water height. For each satellite overpass, when Number and locations of virtual stations
several altimetric measurements have been
selected, the water height is characterized by The 2-D precise geographical selection method
the median of the altimetric values, the median using the JERS land-cover characterization (AUTO-
being a more robust estimator, less sensitive to JERS) initially provided 219 geographical selec-
extreme values and a fortiori to outliers, than tions. After the post-processing of these 219 areas
the average (see time series in Figs 1 and 3). (see Method Section, paragraph on Precise geo-
Downloaded by [177.15.120.2] at 09:59 14 August 2017

Methods are compared at two levels: graphical selection), 128 virtual stations were
– at the satellite overpass level, by taking into defined. The SRTM-based method (AUTO-SRTM)
account: (i) all selected overpasses, and (ii) yielded 131 virtual stations. In order to get compara-
overpasses for which at least five measure- ble results between methods, seven of these stations
ments have been selected; and were assembled with the nearest ones only separated
– at the virtual station level, by computing by islands. Consequently, 124 virtual stations were
the RMS differences between water height finally obtained with the SRTM-based method.
time series associated with the virtual sta- Eventually, in the following, 43 virtual stations
tions jointly defined by two methods. Nor- defined by Santos da Silva, et al. (2007a,b) with the
malized cumulated histograms of the RMS 3-D manual method (MANU) are taken into
differences are computed. This allows account.
deduction of the percentage of virtual sta- The AUTO-JERS and AUTO-SRTM methods
tions that can be considered as identically have 102 virtual stations in common, i.e. stations cor-
defined by two methods, in terms of water responding to the same satellite ground track–river
height and as a function of a RMS differ- intersections. The AUTO-JERS and MANU methods
ence tolerance. have 40 stations in common, the AUTO-SRTM and
(5) Water height variability. Overpasses that com- MANU procedures 38, and the number of satellite
prise at least five selected values are consid- track–river intersections jointly considered by the
ered. Variability is characterized by the median three methods is 38.
absolute deviation, MAD: Figure 2 shows the geographical distribution
of the stations for each method. In Fig. 2, the
N Negro River network is extracted from the land-
MAD(x ) = 1 /( N − 1)∑ xi − xmed cover characterization (Class 8) obtained from
i =1 JERS mosaics (Martinez & Le Toan, 2007).
Consequently, all the stations defined by the
where N is the number of observations, xi is the AUTO-JERS method are represented on the river
ith observation and xmed is the median of the in Fig. 2. Stations manually defined (MANU
observations. The overpass-based and the time method) are not necessarily represented on the
series-based (or station-based) investigations of river. Eventually, stations obtained by the AUTO-
the methods are performed too. Two stations, SRTM methodology can be located on tributaries
S1 and S2, related to the same satellite track– that do not appear in Fig. 2, given the fact that, in
river intersection, provide identically variable this case, the drainage network used for station
results according to a MAD difference toler- definition comes from the SRTM data and are
ance, d, if: independent from JERS.
Producing time series of river water height by means of satellite radar altimetry—a comparative study 115

Station “morphologies” respectively, only 1318 (30.0%) and 1657 (36.3%)


overpasses (see Table 2). Eventually, 1680 over-
Figure 5 shows different typical virtual station “mor-
passes are associated with the 43 stations provided
phologies”, for some stations that significantly differ
by the MANU approach.
from one method to another. For manually defined
The results concerning the number of selected
selections (MANU method), we have represented the
measurements for satellite overpasses jointly consid-
selected satellite nadir locations, the geographical
ered by two methods are presented in Table 3. They
selection alone, performed during the first step of the
show that the MANU method allows a significantly
MANU procedure, not being representative of the
greater number of measurements to be selected than
finally selected altimetric data.
the two other methods. The AUTO-JERS approach
Figure 5(b) shows very similar selections
selects the least number of altimetric measurements.
between AUTO-JERS and AUTO-SRTM methods.
In particular, with such a method, a significant
However, significant differences in data selection
number of overpasses (1125, i.e. 25.6%) is character-
can be observed between these two methods (Fig.
ized by only one altimetric measurement, making
5(c) and (d)). The AUTO-SRTM method selects
water heights unreliable.
areas that (a priori) do not correspond to water,
according to JERS-based land-cover characterization
Downloaded by [177.15.120.2] at 09:59 14 August 2017

(Fig. 5(a) and (c)). In Figs 5(b), and 6(c) and (d), it Water height
may be noticed that the MANU method allows selec-
Table 4 shows the results of paired samples and one-
tion of data a long distance away from the river.
sided Wilcoxon tests for water heights (medians) and
at the overpass level.
Median heights obtained with AUTO-SRTM
Number of selected points per satellite overpass
appear to be significantly lower than those resulting
The AUTO-JERS method allowed 128 virtual sta- from the other two methods. The results related to
tions to be defined, associated with 4399 satellite overpasses that contain at least five selected measure-
overpasses. In total 124 virtual stations, correspond- ments are presented in the second part of Table 4.
ing to 4558 overpasses, were obtained with the Water heights obtained with the MANU method are
AUTO-SRTM method. However, AUTO-JERS and significantly higher than those obtained by the two
AUTO-SRTM selected at least five height values for, other methods. Water heights provided by the AUTO-

Table 2 Number of virtual stations and of altimetric values provided by the three selection methods.
MANU AUTO-JERS AUTO-SRTM

Stations 43 128 124


Satellite overpasses 1680 4399 4558
Overpasses with only one value 5 (0.3%) 1125 (25.6%) 129 (3.0%)
Overpasses with at least five values 1263 (75.2%) 1318 (30.0%) 1657 (36.3%)
Stations with at least one pass of at 35 (81.4%) 71 (55.5%) 69 (55.6%)
least five values

Table 3 Comparison of the three altimetric data selection methods in terms of number of selected measurements per
cycle (Nb).
Method X vs Method Y

MANU vs MANU vs AUTO-JERS vs


AUTO-JERS AUTO-SRTM AUTO-SRTM

Number of common satellite overpasses 1501 1451 3681


Paired, one-sided Wilcoxon test > > <
Nb(X) vs Nb(Y) (p << 0.01) (p << 0.01) (p = 0.0080)
Distribution of Nb(X)–Nb(Y) 95% confidence interval [–19, 51] [–20, 63] [–5, 17]
[min, max] [–31, 64] [–30, 73] [–21, 54]
116 E. Roux et al.

Table 4 Comparison of the three altimetric data selection methods in terms of median water height (h).
Method X vs Method Y

MANU vs MANU vs AUTO-JERS vs


AUTO-JERS AUTO-SRTM AUTO-SRTM

All satellite Paired, one-sided Wilcoxon test NS > >


overpasses h(X) vs h(Y) (p << 0.01) (p = 0.00040)
Distribution of 95% confidence [–0.36, 0.46] [–0.62, 0.84]
h(X)–h(Y) interval
[min, max] [–3.03, 2.60] [–68.86, 36.13]
At least five Paired, one-sided Wilcoxon test > > >
selected values h(X) vs h(Y) (p = 0.015) (p << 0.01) (p = 0.0008)
per overpass Distribution of 95% confidence [–0.33, 0.39] [–0.26, 0.42] [–0.28, 0.33]
h(X)–h(Y) interval
[min, max] [–0.92, 1.86] [–0.77, 2.60] [–19.65, 1.68]
Downloaded by [177.15.120.2] at 09:59 14 August 2017

Fig. 7 RMS differences between water height time series of coinciding stations for each pair of methods.

SRTM method are still the lowest. However, the rela- This maximum bound is 0.26 and 0.15 m for the 90
tive differences are small, the maximum (absolute and 80% confidence intervals, respectively.
magnitude) bound of the 95% confidence interval The normalized cumulated histograms in Fig. 7
being 0.46 m (between MANU and AUTO-SRTM). can be interpreted as the percentage of virtual
Producing time series of river water height by means of satellite radar altimetry—a comparative study 117

stations that can be considered as identically defined the median, the 80th and the 95th percentiles,
by two methods as a function of the RMS difference. respectively.
Of the stations jointly defined by MANU and In the following, results are presented as in the
AUTO-JERS methods, 67.5 and 95% present RMS previous paragraph: Table 6 gives the paired samples
differences of less than 0.2 and 0.5 m, respectively. and one-sided Wilcoxon test results, the extremes
These proportions are 73.7 and 92.1% for MANU and the 95% confidence interval of differences
and AUTO-SRTM methods, respectively. In the case between selection methods considered pair-wise. The
of AUTO-JERS and AUTO-SRTM, these propor- test results show significant differences in variability
tions are 60.8 and 83.3%. For the latter two methods, in all cases. However, we show that the methodolo-
stations for which the RMS differences are high are gies slightly differ in terms of absolute magnitudes of
those that comprise a low number of selected altimet- the differences: the maximum absolute bound of the
ric measurements per overpass. In fact, Fig. 7(d) 95% confidence interval, 0.31 m, is obtained
reports the RMS differences between time series between MANU and AUTO-SRTM (this value is
jointly defined by AUTO-JERS and AUTO-SRTM only 0.15 m for the 80% confidence intervals).
methods, as a function of the minimum number of Figure 8 presents the cumulative percentage of
measurements per overpass. It appears that by taking stations considered to have identical variability as a
into account only overpasses characterized by at least function of a MAD difference value. Of the 102 sta-
Downloaded by [177.15.120.2] at 09:59 14 August 2017

five measurements, 85.7 and 97.9% of stations tions jointly defined by the AUTO-JERS and AUTO-
jointly defined by the two methods present RMS dif- SRTM methods, 53.1 and 79.6% present MAD dif-
ferences equal to 0.2 and 0.5 m, respectively. ferences below 0.20 and 0.50 m, respectively. The
MANU and AUTO-SRTM methods present the
greatest divergence in terms of results variability,
Water height variability
with “only” 20.0 and 70.0% of stations “identically”
Table 5 presents the median, the 80th and the 95th variable to within 0.2 and 0.5 m, respectively. These
percentiles of the MAD distribution over all over- proportions are 34.4 and 81.2% for the MANU and
passes with at least five measurements and for AUTO-JERS methods, respectively.
each method. The three methods provide very sim-
ilar variabilities, with about 0.1, 0.2 and 0.5 m for
DISCUSSION
The MANU and AUTO-SRTM methods permit the
definition of virtual stations at the very upstream part
of the hydrological network: for AUTO-SRTM, this
Table 5 Intra-pass variability: median, 80th and 95th per-
centiles of the median absolute deviation (MAD) values can be realized by considering a lower accumulation
computed for all selected satellite overpasses and as a threshold (1 million cells is used in this study).
function of the altimetric data selection method. Another advantage of the MANU method is that, by
MANU AUTO-JERS AUTO-SRTM
proposing an altimetric measurement selection in the
river section, it does not take into account only the
Median (m) 0.12 0.10 0.10 information derived from images or DEM. The
80th percentile (m) 0.24 0.21 0.22
95th percentile (m) 0.46 0.52 0.56
AUTO-JERS and AUTO-SRTM selection proce-
dures, in turn, exploit images or DEM established at

Table 6 Comparison of the three altimetric data selection methods in terms of height median absolute deviation (MAD).
Method X vs Method Y

MANU vs MANU vs AUTO-JERS vs


AUTO-JERS AUTO-SRTM AUTO-SRTM

At least five Paired, one-sided Wilcoxon test > > >


selected values MAD(X) vs MAD(Y) (p << 0.01) (p << 0.01) (p = 0.0018)
per overpass Distribution of 95% confidence interval [–0.26, 0.23] [–0.31, 0.31] [–0.27, 0.31]
MAD(X)–MAD(Y) [min, max] [–2.06, 0.89] [–2.24, 1.26] [–54.60, 22.64]
118 E. Roux et al.
Downloaded by [177.15.120.2] at 09:59 14 August 2017

Fig. 8 Number of virtual stations that can be considered as identically variable in terms of MAD(cSji), i.e. the median abso-
lute deviation of values related to the ith cycle of station Sj.

a given time, whereas river width and even river MANU method is that, being entirely manual, it is
channel location change over time. time consuming and is likely to provide results that
If flood plains adjoin the river, data related to are poorly repeatable and reproducible.
these flood plains tend to be selected by the MANU The drainage network obtained with the SRTM
method. This explains why some selected altimetric does not reproduce the complexity of the river mor-
data are far away from the river channel (see Fig. 5). phology, the presence of islands for instance. This
This shows that, although the 43 virtual stations explains why selections resulting from AUTO-
defined with the MANU method were intended for SRTM can be very different from those provided by
river monitoring, selection criteria associated with the two other methods (Fig. 5(c) and (d)). Moreover,
the second selection step (low intra-pass height vari- this drainage network presents some aberrations, e.g.
ability and high inter-pass height variability) are sudden direction changes and rectilinear segments
favoured and the geographical selection associated that occur in flat areas (see Fig. 5(a) and (c)). An
with the first step of the procedure appears to have a additional limitation of the SRTM-based method is
negligible effect on the final selection in this case (of that the line obtained by means of the flow accumu-
course, this would not have been the case if stricter lation method does not necessarily coincide with the
geographical selections had been done according to line at equal distance from each true river bank (see
JERS data). One of the main drawbacks of the Fig. 5(a)). Eventually, the model (equation (1)) used
Producing time series of river water height by means of satellite radar altimetry—a comparative study 119

by the AUTO-SRTM method only represents an procedures in order to define these virtual stations,
averaged (or smoothed) river morphology and cannot making the results described in the literature difficult
take into account the river morphology variability to compare. The present study compares different
(see Fig. 5(a)). From a geographical point of view, all altimetric data selection procedures: (a) a fully man-
these drawbacks contribute to questionable river ual 3-D selection procedure, (b) a method exploiting
location identifications. Nevertheless, AUTO-SRTM a land-cover characterization that permits the loca-
is a fully automatic method that provides comparable tion of water body surfaces (rivers in this study), and
results, in terms of water height time series and (c) a fully automatic and original procedure that uses
results variability, to manual and consequently more a DEM and an estimation of the river width. We
subjective, less repeatable and reproducible, and show that the great majority of result differences, in
time-consuming methods. Moreover, most of drain- terms of water height and results variability, are com-
age network inconsistencies can be repaired by pre- parable with the absolute altimetric data errors
viously smoothing the DEM (Grohmann et al., reported in the literature. Consequently, from a glo-
2007), or by forcing flow directions to known bal point of view, we can consider the fully auto-
river locations as proposed by Hutchinson (1989), matic method to define all the virtual stations of a
Saunders (1999) and Getirana et al. (2007). These given basin. This opens the way to use of satellite
approaches would improve the drainage network radar altimetry for water height time series genera-
Downloaded by [177.15.120.2] at 09:59 14 August 2017

reliability and, consequently, the virtual station defi- tion on a large scale and, consequently, considerably
nition and have to be tested in further works. extends the applicability of satellite radar altimetry in
The previous remarks have to be moderated by hydrology.
the quantitative comparison of results. These results However, a minority of virtual stations jointly
show that the RMS differences between the water defined by the three methods are significantly differ-
height time series provided by the three methods are ent. This demonstrates that from the point of view of
of the same order of magnitude compared to the alti- a single station level, differences can be significant,
metric height RMS errors mentioned in the literature and, therefore, the reliability of a virtual station defi-
for the ENVISAT-RA2 altimeter and the ICE-1 nition method depends on the context. In the future,
retracker. These RMS errors range between 0.11 and we should focus on these differences in order to char-
0.53 m, depending on the water surface under study acterize them as a function of the context, i.e. the
(Frappart et al., 2006). river morphology, the relief, the tracking algorithm,
One may wonder why the water height obtained etc. This would allow one to automatically select the
with the MANU method is significantly higher than virtual station definition methodology that is better
that provided by the two other selection procedures. adapted to the observed context.
One explanation is that, being entirely manual, the
MANU method avoids the selection of altimetric
measurements associated with the “hooking” phe- REFERENCES
nomenon (see Fig. 1(b)). Hooking leads to underesti- ANA (Agência Nacional de Águas) (2007) Hydrological data base
mation of the water height (see Fig. 1(b)). http://hidroweb.ana.gov.br (accessed in 2007).
Consequently, the AUTO-SRTM method, which ANA (Agência Nacional de Águas) (2008) Evolução da Rede Hid-
rometeorológica Nacional—2007. Available at: http://
tends to select altimetric measurements out of water
www.ana.gov.br/GestaoRecHidricos/InfoHidrologicas/hid-
surfaces (see. Fig. 5(a) and (c)), is more likely to rometeorologia/rh_amazonica/
select radar nadir locations associated with the altim- Rede_Hidrometeorologica_Amazonia.asp.(accessed in 2008)
eter locking phenomenon and consequently to under- Arge, L., Chase, J. S., Halpin, P., Toma, L., Vitter, J. S., Urban, D. &
Wickremesinghe, R. (2003) Efficient flow computation on
estimate the water height. massive grid terrain datasets. GeoInformatica 7, 283–313.
Birkett, C. M. (1998) Contribution of the TOPEX NASA radar
altimeter to the global monitoring of large rivers and wetlands.
CONCLUSION Water Resour. Res. 34(5), 1223–1239.
Berry, P. A. M. (2000) Topography from land radar altimeter data:
Continental water height time series obtained with possibilities and restrictions. Phys. Chem. Earth 25, 81–88.
satellite altimetric data result from the definition of Berry, P. A. M. (2006) Two decades of inland water monitoring
using satellite radar altimetry In: 15 Years of Progress in
“virtual limnimetric stations”, that is the selection of
Radar Altimetry (Proc. Symp., Venice Lido, Italy, 13–18
radar nadir locations possibly related to the water sur- March 2006). European Space Agency Special Publ. ESA SP-
face of interest. In practice, researchers use different 614. ESA, Noordwijk, The Netherlands.
120 E. Roux et al.

Berry, P. A. M., Garlick, J. D., Freeman, J. A. & Mathers, E. L. Kouraev, A. V., Zakharova, E. A., Samain, O., Mognard, N. M. &
(2005) Global inland water monitoring from multi-mission Cazenave, A. (2004) ‘Ob’ river discharge from TOPEX/Posei-
altimetry. Geophys. Res. Lett. 32(L16401), 1–4. don satellite altimetry (1992–2002)’. Remote Sens. Environ.
Calmant, S. & Seyler, F. (2006) Continental surface waters from sat- 93(1-2), 238–245.
ellite altimetry. C. R. Geosci. 338(14-15), 1113–1122. Leon, J., Calmant, S., Seyler, F., Bonnet, M., Cauhope, M., Frappart,
Chelton, D. B., Ries, J. C., Haines, B. J., Fu, L. & Callahan, P. S. F., Filizola, N. & Fraizy, P. (2006) Rating curves and estima-
(2001) Satellite altimetry. In: Satellite Altimetry and Earth Sci- tion of average water depth at the upper Negro River based on
ences: A Handbook of Techniques and Applications (ed. by J. Fu satellite altimeter data and modeled discharges. J. Hydrol.
& A. Cazenave). Academic Press, San Diego, California, USA. 328(3-4), 481–496.
Coe, M. T. & Birkett, C. M. (2004) Calculation of river discharge Martinez, J.-M. & Le Toan, T. (2007) Mapping of flood dynamics
and prediction of lake height from satellite radar altimetry: and spatial distribution of vegetation in the Amazon floodplain
example for the Lake Chad basin. Water Resour. Res. using multitemporal SAR data. Remote Sens. Environ. 108(3),
40(W10205), 1–11. 209–223.
De Oliveira Campos, I., Mercier, F., Maheu, C., Cochonneau, G., Mercier, F. (2001) Altimétrie spatiale sur les eaux continentales:
Kosuth, P., Blitzkow, D. & Cazenave, A. (2001) ’Temporal apport des missions TOPEX/POSEIDON et ERS-1&2 à
variations of river basin waters from Topex/Poseidon satellite l’étude des lacs, mers intérieures et bassins fluviaux. PhD The-
altimetry. Application to the Amazon basin. C. R. Acad. Sci. sis, Université Université Paul Sabatier, Toulouse III, France.
Series IIA Earth and Planetary Science 333(10), 633–643. Pistocchi, A. & Pennington, D. (2006) European hydraulic
Frappart, F. (2005) Hydrologie spatiale: développement d’applications geometries for continental SCALE environmental modelling.
pour l’utilisation de la télédétection sur les grands bassins fluvi- J. Hydrol. 329(3–4), 553–567.
aux. PhD Thesis, Université Paul Sabatier, Toulouse III, France. R Development Core Team (2005) R: A language and environment
Frappart, F., Seyler, F., Martinez, J., Leon, J. G. & Cazenave, A. for statistical computing. R Development Core Team, R Foun-
Downloaded by [177.15.120.2] at 09:59 14 August 2017

(2005) Floodplain water storage in the Negro River basin esti- dation for Statistical Computing, ISBN 3-900051-07-0, http://
mated from microwave remote sensing of inundation area and www.R-project.org.
water levels. Remote Sens. Environ. 99(4), 387–399. Roux, E., Cauhope, M., Bonnet, M.-P., Calmant, S., Vauchel, P. &
Frappart, F., Calmant, S., Cauhope, M., Seyler, F. & Cazenave, A. Seyler, F. (2008) Daily water stage estimated from satellite
(2006) Preliminary results of ENVISAT RA-2-derived water altimetric data for large river basin monitoring. Hydrol. Sci. J.
levels validation over the Amazon basin. Remote Sens. Envi- 53(1), 81–99.
ron. 100(2), 252–264. Santos da Silva, J., Roux, E., Calmant, S., Corrêa Rotunno Filho, O.,
Freeman, A., Chapman, B. & Siqueira, P. (2002) The JERS-1 Ama- Seyler, F., Vaz de Almeida, F. & Bonnet, M.-P. (2007a) Vari-
zon Multi-season Mapping Study (JAMMS): science objec- ações temporais de níveis d’água na bacia do Rio Negro esti-
tives and implications for future missions. Int. J. Remote Sens. madas pelo satelite altimétrico ENVISAT. In: Proc. of XVII
23, 1447–1460. Simposio Brasileiro de Recursos Hidricos, São Paulo.
Getirana, A. C. V., Bonnet, M.-P., Rotunno Filho, O. C. (2007) Santos da Silva, J., Roux, E., Calmant, S., Seyler, F., Corrêa
Obtaining drainage directions in flooded areas: a new “burn- Rotunno Filho, O., Joo Mansur, W., Guyot, J. & Bonnet, M.
ing” approach for large basins. Acapulco Mexico Jt Assem. (2007b) Spatial variability of floodplain hydrological cycle
Suppl., Abstract H22A-03, Eos Trans. AGU 88(23). Available with radar altimetry and satellite images. Rio Negro basin
at: http://www.agu.org/meetings/ja07/waissm07.html. (Brazilian Amazon). Eos Trans. Am. Geophys. Union 88(23) Jt
GRASS Development Team (2006) Geographic Resources Analysis Assem. Suppl., Abstract H31D.
Support System (GRASS) Software. ITC-irst, Trento, Italy. Saunders, W. (1999) Preparation of DEMs for use in environmental
http://grass.itc.it. modeling analysis. ESRI User Conference (24–30 July 1999,
Grohmann, C. H., Riccomini, C. & Alves, F. M (2007) SRTM-based San Diego, California). Available at: http://proceed-
morphotectonic analysis of the Pocos de Caldas Alkaline Mas- ings.esri.com/library/userconf/proc99/navigate/proceed.htm.
sif, southeastern Brazil. Comput. Geosci. 33, 10–19. Van Griensven, A., Fohrer, N., & McCulloch, C. (2007) Editorial
Hess, L. L., Melack, J. M., Novo, E. M. L. M., Barbosa, C. C. F. & notes. Water Resour. Manage. 21, 1991–1992.
Gastil, M. (2003) Dual-season mapping of wetland inundation Wingham, D. J., Rapley, C. G. & Griffiths, H. (1986) New tech-
and vegetation for the central Amazon basin. Remote Sens. niques in satellite altimeter tracking systems. In: Geoscience
Environ. 87, 404–428. and Remote Sensing (Proc. Int. Symp. Zürich, Switzerland, 8–
Hollander, M. & Wolfe, D. A. (1973) Nonparametric Statistical 11 September 1986). European Space Agency Special Publ.
Methods. John Wiley & Sons, Ltd., Chichester, UK. ESA SP-254, 1339–1344. ESA, Noordwijk, The Netherlands.
Hutchinson, M. F. (1989) A new procedure for gridding elevation Zakharova, E. A., Kouraev, A. V., Cazenave, A. & Seyler, F. (2006)
and stream line data with automatic removal of spurious pits. J. Amazon River discharge estimated from TOPEX/Poseidon
Hydrol. 106, 211–232. altimetry. C. R. Geosci. 338(3), 188–196.

Vous aimerez peut-être aussi