Available online at www.sciencedirect.com

www.elsevier.com/locate/asr

Advances in Space Research 51 (2013) 1542–1550

From TOPEX/Poseidon to Jason-2/OSTM in the Amazon basin

Frederique Seyler a,⇑, Stephane Calmant b, Joecila Santos da Silva c,Daniel Medeiros Moreira d, Franck Mercier e, C.K. Shum f,g

a IRD/ESPACE-DEV, 500 Rue Jean Franc�ois Breton, 34093 Montpellier, Franceb IRD/LEGOS, 14 Av. Edouard Belin, 31400 Toulouse, France

c UEA/CESTU, Av. Djalma Batista 3578, 69058-807 Manaus, Brazild UFRJ/CPRM, Av. Pasteur 404, 22290-040 Rio de Janeiro, Brazil

e CLS, Collecte Localisation Satellites, 8–10, rue Hermes, Parc Technologique du Canal, 31520 Ramonville Saint-Agne, Francef Division of Geodetic Sciences, School of Earth Sciences, Ohio State University, 125 South Oval Mall, 43210 Columbus, OH 43210, United States

g Institute of Geodesy and Geophysics, Chinese Academy of Sciences, Wuhan, China

Available online 12 November 2012

Abstract

A major interest of radar altimetry over rivers is to monitor water resources and associated risk in basins where there is little or noconventional in situ data. The objective of the present study is to calibrate altimetry data in a place where conventional data are avail-able, and use the results to estimate the potential error committed in the estimation of water levels in an ungauged or poorly gaugedbasin. The virtual stations extracted with Jason-2 in this study concern a very broad sample of river channel width and complexity. Min-imum channel width has been estimated at 400 m. Unlike TOPEX/Poseidon (T/P), Jason-2 seems to have the capability to distinguish theriver bed from its floodplain. The quality of the results obtained with Jason-2 is incomparably better than that obtained with T/P.Despite the fact that no absolute calibration has been assessed for river in this study, the bias calculated converge around 0, 35 m, whichcould be then the error estimated on the water stage derived from Jason-2 ranges, when no other validation is available. ICE3 algorithmseems to be performing as well as ICE1, and further research is needed to design retracking algorithm specifically for continental water.� 2012 COSPAR. Published by Elsevier Ltd. All rights reserved.

Keywords: Altimeter calibration; Jason-2; Amazon basin; Hydrology

1. Introduction

The family of TOPEX/Poseidon (T/P) satellites (Fuet al., 1991) extends over 20 years of altimetry history, sinceT/P was launched August 10, 1992 from Kourou in FrenchGuiana. It is only in 1996 that the retracking of T/P archiveby the Science Working Team on the T/P project, achieves2–3 cm error in estimating the ocean surface. This accuracyhas been reached because of reduced radial orbit errors(Bertiger et al., 1994), reaching the sub-centimetric accu-

0273-1177/$36.00 � 2012 COSPAR. Published by Elsevier Ltd. All rights rese

http://dx.doi.org/10.1016/j.asr.2012.11.002

⇑ Corresponding author.E-mail addresses: frederique.seyler@ird.fr (F. Seyler), stephane.cal-

mant@ird.fr (S. Calmant), jsdsilva@uea.edu.br (Joecila Santos da Silva),daniel.moreira@cprm.gov.br (D.M. Moreira), franck.mercier@cls.fr(F. Mercier), ckshum@osu.edu (C.K. Shum).

racy for the Jason-2 mission (Bertiger et al., 2010). It is alsoabout that time that are emerging the first applications ofaltimetry for inland waters (Morris and Gill, 1994; Birkett,1995a,b, 1998; Ponchaut and Cazenave, 1998). It is only 10years later that are published early works on multi altime-ter mission for inland waters (Berry et al., 2005; Frappartet al., 2006). Jason-1 that was highly anticipated as a fol-lowing of T/P, was for inland waters a “gap” in data sincevery few data of Jason-1 are useful for monitoring inlandwaters. It was not until 2008 and the launch of Jason-2to continue the Poseidon family on inland water bodies.

In the studies of the ocean, it is possible to combine datafrom different missions as long as the relative biases areestimated. This is due to the spatial and temporal continu-ity of the ocean environment. The case of rivers and lakes isdifferent. Most lakes of small area will no longer be

rved.

1 In French – means Earth and space network.

F. Seyler et al. / Advances in Space Research 51 (2013) 1542–1550 1543

monitored if the trace of the orbit changes during a missionor between two different missions. In the case of rivers,there exists a spatial continuity but limited to the hydro-graphic network. The water level in the river system resultsfrom the combination of hydrodynamic factors (slope, flowvelocity, roughness of the bed), morphological factors(shape of the section that is variable throughout the hydro-logical cycle), the contribution of the tributaries and thevarious exchanges with the watershed and the floodplain(direct rainfall, evaporation, diffuse runoff, groundwatercontribution and subtraction). It is easily understandablethat this combination of factors is highly and non linearlyvariable in space and time. In addition, the use of satellitealtimetry for monitoring river stage allowed defining thenotion of virtual station (Frappart et al., 2006; Leonet al., 2006; Roux et al., 2008). A virtual station is consti-tuted by the measurement points from the ground trackportion located at the intersection with the river. This con-cept was designed partly to be close to terrestrial waterlevel monitoring networks that consist of fixed stations.The interest and value of a water level measurement stationis linked to its lifespan in a specific place. This persistenceallows following changes in the hydrological regime inthe long term and predicting extreme events. Monitoringand forecasting is of course even more significant in thecontext of climate change. It is why the concept of lineageof the missions is an important concept for hydrology, as ata family of satellite corresponds a common orbit, whichensures continuity in the river or lake monitoring. It isthe first objective of this study to compare the resultsobtained by T/P and by Jason-2 in estimating the waterstage at virtual stations.

In hydrology, the need for temporal continuity of themissions is accompanied by a need for a spatial distributionas dense as possible. This is one of the great advantages ofaltimetry over conventional hydrological measurement net-works to be globally distributed in a dense network consti-tuted by the different satellite ground tracks. For example,the study of Frappart et al. (2005) calculated the storageof the inundation plain of the Rio Negro sub-basin. Thissub-basin of about 700 000 km2 counts 25 in situ stationsand T/P altimetric virtual stations added 88 monitoringpoints of the water stage. Using multi-mission sourcesallows densifying the monitoring network of virtual stations.In this respect the relatively loose mesh of the Jason-2 orbitis complementary to that of Envisat, much denser. Thereverse is true for the temporal resolution. The revisit periodgreater than one month prevents a number of hydrologicalapplications to be contemplated with Envisat data. In thisrespect, Jason-2 is better suited for some applications need-ing frequent observations. However for using multi-missiondata, it is necessary to determine possible bias between sen-sors. If there are a number of studies on absolute bias ofJason-1 and 2 in the oceanic domain (Dettmering and Bos-ch, 2010; Bonnefond et al., 2010; Haines et al., 2010; Merti-kas et al., 2010, 2011; Arnault et al., 2011; Watson et al.,2011; Washburn et al., 2011, among others), a few studies

of calibration over lakes (Birkett and Beckley, 2010; Chenget al., 2010; Cretaux et al., 2009, 2011), there is currently noestimate of absolute bias for Jason-2 on rivers. This is thesecond objective of this paper to determine if absolute cali-bration for Jason-2 is possible over river.

The advantage of using data processed by trackerdesigned specifically for inland waters has been repeatedlyshown (Frappart et al., 2006; Birkett and Beckley, 2010).Cited studies established that the algorithm ICE1 was morerobust than the other retracking algorithms (OCEAN,SEAICE and ICE2) for both Envisat and Jason-2 overinland waters. When estimating the water height, ICE1do not always give the less noisy result but it gives nearlythe better result in most cases. In addition comparing withother retracking algorithm, ICE1 has the lowest rate ofdata loss. As soon as 2004, considerable effort has beenmade by the CASH project (Contribution of SatelliteAltimetry to Hydrology, funded by the Reseau Terre &

Espace,1 on behalf of the French Ministry of Research)for processing the whole T/P archive for the nominal orbit,with the trackers used for the Envisat mission (namelyICE1, ICE2 and SeaIce). During the PISTACH project,Mercier et al. (2007) have developed for CNES a trackerspecific for inland waters called ICE3, that has beenapplied to the Jason-2 data. No studies comparing thetwo trackers ICE3 and ICE1 have been published and thisis the third objective of the present study.

2. Data and methods

Data used in this study were from two radar altimetersatellite mission, T/P and Jason-2/OSTM. Were also useddata from GPS acquisition campaigns and in situ limnimet-ric data.

T/P satellite was launched on August 10th, 1992 byNASA, U.S. space agency and CNES, the French spaceagency. It is given a sea level overall accuracy over onemonth better than 2 cm. In September 2002, T/P wasmoved to a new orbit, now used by Jason-1, liberatingthe orbit for Jason-2 after a tandem phase. The tandem for-mation was maintained (both ground track within 1 km ofeach other) from the launch of Jason-2 in June 2008 untilJanuary 26th, 2009, when Jason-1 was moved from thatorbit into the T/P one. This phase has been used for cali-brating the entire altimetric system (Quartly, 2010), mainlyover ocean. It could not have been used over land surfaceas very few continental Jason-1 data is available.

The T/P data used in this study encompass then the per-iod 1992–2002 (http://www.aviso.oceanobs.com/en/mis-sions/past-missions/topexposeidon.html). In the scope ofthe CASH project, the whole archive of T/P data on thenominal orbit has been retracked with the four algorithmsused for the Envisat mission, i.e. ocean, ICE1, ICE2, andSEAICE. In this study we used the T/P data of the CASH

1544 F. Seyler et al. / Advances in Space Research 51 (2013) 1542–1550

archive retracked by ICE1, in order to be coherent with theprocessing used for Jason-2.

Jason-2 mission was launched on June 20th, 2008. It hason-board the Poseidon 3 altimeter, coupled with the realtime tracking system DIODE of DORIS, which shouldallow better and more frequent measurements in coastalareas, inland waters and ice caps. Poseidon 3 altimeter isa two-frequency solid-state altimeter, measuring rangewith accurate ionospheric corrections, operating at13.575 GHz (Ku-band) and 5.3 GHz (C-band). Expectedprecision is about 2 cm over ocean. Raw data are processedby SSALTO (Segment Sol multimissions d’ALTimetrie,d’Orbitographie.2

In this study were used data from the geophysical datarecord (GDR-C), a fully validated product (Dumontet al., 2008), processed in delayed-mode processing (AVI-SO), distributed by CTOH (http://ctoh.legos.obs-mip.fr/).What is new with Jason-2 and allows retrieving a good pro-portion of data over land, unlike Jason-1, is an on-boardtracking technique called “open loop tracking”. The inter-nal memory of the altimeter includes a Digital TerrainModel (DTM) including elevations of areas flown alongtrack. These DTM, coupled with the DIODE system, areused to anticipate the time position of the reception win-dow of the radar in order to improve the data collectionover the continental domain (http://www.altimetry.info/html/alti/principle/waveform/onboard_tracking_en.html).

T/P and Jason-2 shared the same orbit at successivetimes with a gap between 2002 and 2008. The orbit is1336 km high, with a repeat time of 10 days, the distancebetween ground tracks being 315 km at the equator, andthe along track resolution is 700 m for land applications.

In order to compare the two retracking algorithm ICE1and ICE3, two different datasets were used in this study.First, the altimetric ranges were extracted from CTOHdatabase (http://ctoh.legos.obs-mip.fr/), using the geo-physical and propagation corrections recommended forcontinental water: Wet troposphere correction calculatedover the continents using NCEP (National Centers forEnvironmental Prediction) data (Mercier, 2003); iono-spheric, tropospheric, pole tide and solid earth tide correc-tions provided in the GDRs have been applied. The rangechosen was that retracked by the algorithm ICE1 (Bamber,1994). ICE1 has been determined as the more robust forhydrological studies comparing with the other three algo-rithms available in the GDRs (OCEAN, ICE2 and SEA-ICE) (Frappart et al., 2006; Silva et al., 2010). Second,we used altimetric ranges retracked with the ICE3 algo-rithm, made available in the scope of the PISTACH project(Mercier et al., 2007). ICE3 has been developed to processaltimetric products specifically for coastal areas andcontinental water bodies. Better geophysical correctionsare applied to IGDR with a 20 Hz sampling rate, and

2 In French when in italic: Orbitography and altimetry multimissionground segment.

pre-classification of the waveforms following GLOBCOV-ER land cover allows better discriminating the waterbodies.

For the three datasets (T/P retracked with ICE1, Jason-2 retracked by ICE1 and ICE3), we used the VALS soft-ware (http://www.hybam.org) to build the temporal series(series of water stage varying with time). VALS softwarewas designed by our team specifically for creating virtualstations. The method developed in VALS consists inextracting the data comprised within a polygon defined atthe crossing of the tracks and the water body. A secondstep consists in performing a projection of the portion ofthe track onto a plane perpendicular to the river flow.The display in this vertical plane across the river bed allowsthe extraction of the part of each cycle comprised withinthe two margins of the river at all period of the hydrolog-ical cycle. Generally the selected points for the cyclesacquired during the low flow period will be fewer thanthe points selected at high flow, as the river bed is nar-rower. Last, the median of the measurements for each cyclewithin the bed of the river is computed. The median is lesssensitive than the average to possible outliers. The methodused is best described in Silva et al. (2010).

Most of in situ gauging stations used in this study aremaintained by ANA (National Agency for Water of Brazil)jointly with CPRM (Geological Survey of Brazil). All dataare available on hidroweb website (http://hidro-web.ana.gov.br/). Some of them are maintained in thescope of calibration projects funded by CNES/TOSCAprogram (program of the French Space Agency), IRD(French Institute for Development), CPRM, CNPq (Bra-zilian National Council for Scientific and TechnologicalDevelopment) and FINEP (Brazilian Financier of Studiesand Projects). Readings of the water levels are made twicea day by an observer at the gauging station, at 7:00 and17:00, and then the daily average value is calculated anddated at 12:00 local time. This is the reference daily waterlevel for the given station. Details on the operation of theANA/CPRM management of the Brazilian limnimetricnetwork is given in hidroweb website.

In order to level the in situ gauging stations to a com-mon reference level, GPS observations have been con-ducted using dual-frequency receivers and micro-centeredantennas and collected during various field campaigns.Position and ellipsoid height of each gauging station werecomputed using the GINS-PC software developed byCNES (Marty, 2009; Loyer et al., 2012) from GPSrecorded observations lasting from a couple of hours toseveral days. All the GPS solutions were relative toITRF2005 (Altamimi et al., 2007). To convert these GPSellipsoidal heights of the gauge onto orthometric heights,we use the EGM2008 geoid model (Pavlis et al., 2012).Absolute daily river levels are then computed from theriver stage at a given date corrected from the orthometricheight of the zero of the gauge station. Detailed descriptionof the leveling of the in situ gauging stations is given(Calmant et al., 2013).

Table 1Direct comparison between a track and a gauge nearby (distance <10 km).

ICE 1 Jason-2absolute bias (m)

ICE 3 Jason-2absolute bias (m)

Dbias

Coari (241) 0.582 ± 0.047 0.578 ± 0.042 0.004Santa Luzia (063) 0.503 ± 0.026 0.459 ± 0.024 0.044Urucurituba (152) 0.506 ± 0.018 0.467 ± 0.018 0.039

F. Seyler et al. / Advances in Space Research 51 (2013) 1542–1550 1545

Biases for Jason-2 were computed for the two differentdatasets (retracked by ICE1 and ICE3). First biases werecomputed by difference between the altimetric time seriesand that of the two closest leveled gauging stations, oneupstream and the other downstream of the virtual station(altimetric station). A constant linear slope model has beenused for best-fitting the bias between the altimetric stationand the two gauging stations. Only the concordant dates

Fig. 1. Location map; Blue dots represent the Jason-2 virtual station in the Astations. Magenta large dot marks the location of case A Fig. 2, red large dot mcase C Fig. 2. The background is from SRTM (Shuttle Radar Terrain Model

have been retained from the daily records of the in situgauging stations.

Second, for three tracks located at the vicinity of anin situ gauging station, it was performed a direct compari-son between water stage estimated at the virtual stationand the water level recorded at the in situ station (the threetracks are listed Table 1).

3. Results and discussion

3.1. From T/P to Jason-2

A number of 72 virtual stations (blue dots on Fig. 1)have been extracted from the Jason-2 mission within theAmazon basin. They have been chosen to be well distrib-uted within the basin, from the largest stem of the Amazonnear the mouth until the farthest reaches into the Andeanpiedmont, minimum channel width being estimated at

mazon basin extracted for this study. Green symbols are the T/P virtualarks the location of case B Fig. 2, yellow large dot mars the location of the

).

Fig. 2. Three cases of comparison between T/P, Jason-2 and in situ gauge station; grey line/black triangles represent time series of ranges calculated for T/P mission from 1992 to 2002, grey lines black dots represent time series for Jason-2 from 2008 on. Red line represents the time series of leveled river stage atthe closest in situ gauge station. The graph above the time series represents the differences in meter between the altimetry time series and the conventionalone. (For interpretation of the references to colours in this figure legend, the reader is referred to the web version of this article.)

1546 F. Seyler et al. / Advances in Space Research 51 (2013) 1542–1550

F. Seyler et al. / Advances in Space Research 51 (2013) 1542–1550 1547

400 m. For each one of these virtual stations, T/P data (T/P) have been processed in the same way (described hereabove) in order to extract virtual stations. From these 72virtual stations, only 41 could be extracted for T/P. Fromthese 41, when the time series is compared (for the concor-dant dates) with the time series of the closest in situ gaugestation, only 18 have a RMS between the two stations lessthan ±1.50 m. These 18 cases are reported Fig. 1 (greensymbols) and Fig. 2 gathers some of these favorable cases.In the three parts of the figure, the beginning of the greytime series (From 1992 to 2002) is made of T/P data andthe end of Jason-2 data, from 2008. The red line representsthe daily time series of the nearest gauging station. Thegraph above the time series illustrates the differencesbetween virtual and in situ series. Within these 18 favorablecases, it is possible to distinguish three different situations.The first case is the most frequent and is illustrated by thePurus case (Fig. 2A, magenta large dot on Fig. 1). Only theupper part of the stage variations is captured by T/P. Itcould be due to the loss of data during low water whenthe river narrows. But the constant level of the low waterenables interpreting this lack of data as the measurementby T/P on the dried floodplain. Few cases (four cases) showa phase concordance with the in situ time series, but the dif-ferences between the two series is of ±2–3 m with a highnumber of outliers (Fig. 2B, red large dot on Fig. 1). In thissituation, the floodplain do not completely dries up, but themeasurements at low flow are highly contaminated by themargins of the river. In some very rare cases (two cases),the T/P time series matches quite well the in situ time series.This is the case of the track 63 near Manaus (Fig. 2C, yel-low large dot on Fig. 1); another one concerns the MadeiraRiver. These two cases concern very large river withoutdrying floodplain during the low flow season. The typicaldifference is more or less the same than that of the Jason-2 time series, only the number of outliers is higher for T/

Table 2Absolute bias calculated for Jason-2 mission with the two retracking algorithm

Jason-2track

Sub-basin Upstreamin situ gauge

Downstreamin situ gauge

Distanceupstream

139 Amazonas Itacoatiara Parintins 249152 Amazonas Jatuarana Iracema 72228 Amazonas Parintins Obidos 32063 Madeira Manicore Fazenda Vista

Alegre102

152 Madeira Borba Urucurituba 132254 Madeira Porto Velho Humaita 163076 Negro Barcelos Moura 28241 Negro Barcelos Moura 103254 Negro Curicuriari Tapuraquara 128063 Solimoes Codajas Anama 53076 Solimoes Codajas Anama 53089 Solimoes Sao Paulo de Olivenc�a Fonte Boa 28165 Solimoes Fonte Boa Tefe 95178 Solimoes Sao Paulo de Olivenc�a Fonte Boa 169254 Solimoes Fonte Boa Tefe 237

P than for Jason-2. T/P is lacking of resolution in compar-ison with Jason-2, and even using the same process for thetwo missions, T/P is not able to distinguish between theriver and the floodplain. It can perform good measure-ments at high water when the major bed of the river isflooded but cannot lock onto the minor bed of the riverat low water, and the measured range is either that of thedried floodplain or contaminated by the margins. It is note-worthy that these results are much less optimistic than pre-vious studies conducted in the Amazon basin: For exampleBirkett et al. (2002), who found for T/P mission a meanRMS of 1.1 m when comparing with about 50 gauge sta-tions in the Amazon basin. In the present study, in situgauges are referenced to sea level. In Birkett et al. (2002),the water stage were referenced to the gauge zero. There-fore the comparison between the altimetry series and thein situ series were performed by matching the mean of bothseries. This relative leveling considerably reduced the devi-ation. Particularly, the cases shown Fig. 2A (the most fre-quent one) should demonstrate lesser RMS if relativelyleveled. It is interesting also to recall that the errors werereported in various precursor studies to be higher for lowwater than for high water, which is not surprising whenin most cases, T/P is measuring the whole dried floodplaininstead of the river stage.

In comparison with these poor results of T/P, we canenhance the improvements made with the Jason-2 mission.It has been possible to extract time series for the 72 virtualstations (blue dots Fig. 1) with Jason-2, including upstreamsituations on the Amazonas of Peru, or narrow Napo inEcuador, or upper Rio Negro in Brazil, also for minor trib-utaries of the Amazon like Rio Ic�a. Mean RMS is 0.31 m.If a large part of that improvement can be expressed interms of a better resolution of the altimeter, these are verypromising results for the next generation of nadir altimeterlike SARAL/AltiKa.

s ICE1 and ICE3.

(km)Distancedownstream(km)

Totaldistance(km)

Ice1bias (m)

StDev(m)

Ice3bias (m)

StDev(m)

�18 267 0.369 0.036 0.176 0.071�30 102 0.669 0.020 0.716 0.076�143 175 0.236 0.023 0.203 0.021�98 200 0.475 0.036 0.330 0.044

�6 138 0.615 0.099 0.539 0.088�80 243 �0.122 0.030 �0.148 0.042�134 162 0.430 0.045 0.406 0.047�61 164 0.646 0.038 0.649 0.036�100 228 �0.081 0.019 �0.133 0.029�47 100 0.463 0.043 0.306 0.042�47 100 0.240 0.042 0.268 0.041�423 451 �0.512 0.056 �0.748 0.068�163 258 0.410 0.034 0.398 0.041�288 457 �0.456 0.057 �0.404 0.044�17 254 0.278 0.047 0.309 0.044

Table 3Relative bias (Db in m) between Ice1 and Ice3 water level and samplingrate (Tx) by direct comparison of retracking algorithm for each track.

Db (ice1–ice3) m tx (ice1) tx (ice3)

1–228 Amazonas 0.116 98 962–139 Amazonas 0.190 96 923–050 Amazonas 0.257 98 974–152 Amazonas 0.122 99 965–063 Amazonas 0.008 96 886–076 Solimoes 0.065 100 967–241 Solimoes �0.050 97 918–254 Solimoes �0.037 97 879–165 Solimoes 0.058 97 8110- Solimoes 0.116 96 8611- Solimoes 0.036 97 8912- Amazonas 0.083 95 8914–191 Maranon 0.131 94 9315–204 Maranon �0.145 94 9316–102 Javari 0.350 97 9418–026 Javari �0.007 96 9219–178 Jutai 0.049 100 9820–178 Jutai �0.005 100 9821–089 Jutai 0.061 100 9422–102 Jutai 0.096 100 9623–178 Jurua �0.292 100 9824–102 Jurua �0.059 99 9725–089 Jurua �0.254 100 9626–026 Jurua 0.134 100 9827–076 Purus 0.122 99 9729–165 Purus �0.011 100 9430–102 Purus 0.004 95 9231–026 Purus 0.154 96 9434–152 Madeira 0.106 97 9435–063 Madeira 0.153 98 9837–063 Madeira 0.075 100 9538–076 Madeira 0.106 100 9839–254 Madeira �0.108 98 10040–241 Madeira �0.036 98 9641–241 Madeira 0.002 95 9542–241 Madeira �0.009 91 9243–050 Xingu 0.226 93 9444–050 Xingu 0.208 94 9845–050 Xingu 0.169 79 7746–050 Iriri �0.078 93 9247–089 Negro 0.067 98 9248–089 Negro 0.042 100 9450–254 Negro 0.055 92 9251–165 Negro 0.110 94 9352–013 Negro 0.047 93 9353–178 Negro 0.016 100 9754–063 Negro 0.118 94 8855–076 Negro 0.079 97 9656–241 Negro 0.047 96 9457 178 Uaupes 0.055 99 9858 102 Ica �0.46 94 8959 013 Ica _ 0.061 99 9560 191 Ica 0.029 91 8861 026 Ica 0.041 92 8262 115 Ica 0.056 91 8263 089 Ica 0.090 95 9164 026 Napo 0.047 95 9365 165 Japura 0.065 100 9766 089 Japura 0.072 100 9667 178 Japura 0.089 96 8869 102 Japura 0.046 98 9770 191 Japura 0.124 79 7871 026 Japura 0.097 96 9472 115 Japura 0.082 92 94

1548 F. Seyler et al. / Advances in Space Research 51 (2013) 1542–1550

Nevertheless, the first objective of the study, which wasto determine if a continuity of monitoring could beobtained by the T/P to Jason-2 family, is not reached, orat least only partially. Only 18 virtual stations can be rea-sonably used for a long term monitoring, with suspiciousresults at low water for T/P.

3.2. Bias calculation

Tables 1 and 2 report the bias calculated from the differ-ence between Jason-2 stations and in situ stations leveledby GPS. In Table 1, results are obtained by direct compar-ison, as the track was located in the immediate vicinity ofthe in situ gauge for three cases. In Table 2, the results wereobtained following the method described in chapter 2. Thismethod is based on a strong hypothesis: It supposes thatthe slope between the three stations (two in situ and onevirtual) may vary during the hydrological cycle but thatit remains the same all along the river reach considered.

Bias of Jason-2 when calculated by comparison with oneclose by in situ station is in average of 0.53 ± 0.03 m forICE1, and 0.498 m ± 0.028 cm for ICE3.

When compared with two in situ gauges leveled by GPS,mean values of bias are 0.244 ± 0.041 m for ICE1 and0.191 ± 0.048 m for ICE3. These values could seem closerto the absolute bias calculated by other authors in differentenvironments (lakes and ocean) than the values presentedabove. But these bias of Jason-2 are positive in 11 of the15 cases and negative in 4 cases. They are comprisedbetween �0.51 and +0.69 m. As negative bias occurs onlyin cases where there is a large distance between stations, itappears that the hypothesis of constant slope could not besatisfied when the distance is great. When removing thecases for a total distance greater than 200 km, mean biasfor ICE1 is 0.356 ± 0.047 cm and 0.352 m ± 0.044 cm forICE3, with no negative values.

The two sets of results appear to be far from the resultsfound by other authors for ocean and lakes: 0.162 m forlake Issyk Kul (Cretaux et al., 2011), 0.173–0.171 m foundby Mertikas et al. (2011) for descending and ascendingpasses over ocean, 0.148 m over the great lakes (Chenget al., 2010), and 0.15 m over Corsica (Bonnefond et al.,2010).

As there is no evident reason why the absolute biasshould be significantly different over rivers, it appears fromthese results that there does not exist any way of determin-ing the potential error of altimetric stage when comparedwith in situ gauged water level. Even when the gauging sta-tion is located nearby the altimetric track (less than 10 km),the hypothesis of constant slope seems to be irrelevant.

3.3. Comparison between ICE1 and ICE3

Table 3 reports the relative bias between the tworetracking algorithms, the sampling rate for each one (per-cent of calculated ranges in reference to the total number ofmeasurements), and the RMS for the two trackers, for the

F. Seyler et al. / Advances in Space Research 51 (2013) 1542–1550 1549

72 Jason-2 virtual stations. Relative bias is varying from�0.192 to +0.257 m, with a mean value of +0.051 m. Asthe mean RMS calculated from the comparison betweenthe two trackers is 0.29 m (min = 0.082, max = 1.408), wecannot assume a systematic bias between the two trackingalgorithm. By comparing the bias in relation to the in situstations (Table 2), there is no relevant difference betweenthe two trackers either. The bias calculated for each trackeris always of same sign, and very close from each other.Mean difference is + 0.054 m (minimum �0.052, maximumdifference 0.193). As for the sampling rate, it is better forICE3 than for ICE1 in three cases, equal in four cases,and better for ICE1 than for ICE3 in the 65 other cases.

4. Conclusion

Jason-2 altimeter gives far better results when estimatingthe river water level than T/P. A total of 72 virtual stationshave been extracted within the Amazon basin for Jason-2,with a mean RMS of 0.31 m, for rivers varying from sev-eral kilometers wide to 400 m. With the same methodologyof extraction, same geophysical corrections, and the sametracker (ICE1), only 18 virtual stations have a RMS lessthan ±1.50 m for T/P. Analyzing from various examplesthe reasons for such poor results for T/P, it seems thatthe results are comparable with those of Jason-2 in the caseof large rivers without floodplain. There is only a greaternumber of outliers with T/P than with Jason-2 in thesecases. For all the other situations, T/P captures only theranges for high flow. For low flow the values of the rangeare either contaminated by the surroundings of the river, orlost. To our knowledge, our study is the first one estimatingan error for T/P by comparing with water stage obtained atleveled in situ stations. This could explain that betterresults for T/P have been found previously when compar-ing the series only relatively. These results show that unfor-tunately, there is no continuity to be expected for T/P toJason-2 missions for continental rivers, with some rareexceptions. Nevertheless, the good results of Jason-2 thatcan be attributed to a better resolution of the altimeterseem promising for the next generation of nadir altimeterlike SARAL/AltiKa.

As far as we know, absolute bias has never been calcu-lated for Jason-2 altimetry missions applied to river level.In this study, comparing 15 virtual stations obtained withJason-2 with time series at the in situ gauges closestupstream and downstream from the virtual station, wefound a mean bias of 0.244 ± 0.041 m for ICE1 and0.191 ± 0.048 m for ICE3 algorithm. In both cases, the dis-tribution of biases is characterized by a large spreadaround the mean value. Such a result was also observedby Calmant et al. (2013) for the Envisat ranges alsoretracked with the ICE1 algorithm. Noteworthy, thesespreads around the mean values of biases are far abovethe potential error sources usually invoked in altimetry cal-ibration over oceans or even over lakes, such as orbit orpulse propagation errors. We propose that these great dis-

persions reflect the fact that the slope between the virtualand in situ stations is not only varying with time but alsoalong the course of the river. Therefore, it is not onlyimpossible to estimate an absolute bias for Jason-2 altime-ter in the case of rivers, but also impossible to estimate anerror of the altimeter range by comparing with one or sev-eral in situ gauging stations, even in the closest cases (lessthan 10 km apart).

Nevertheless, it is worth noting that these results give afair estimation of the range of error that can be expectedwhen measuring river stage by altimetry. There is a remark-able convergence around the value of 0.35 m when estimat-ing bias for Jason-2. It seems then reasonable to apply abias of 0.15 m (calculated either for ocean or lakes) toJason-2 data in future studies, estimating a potential errorof ±0.35 m, when no other validation is available. And thiscould be always the case, even when in situ data is avail-able, unless permanent dispositive of validation can bedesigned, installed and maintained at the exact locationof the satellite tracks.

ICE3 algorithm seems to have no better results thanICE1 when estimating the range for Jason-2. Furtherresearch has to be devoted to find tracker better adaptedto continental water. Despite the complexity of continentalwaters, altimetry reveals being an invaluable tool for mon-itoring water resources in ungauged or poorly gaugedbasins. Great expectations is placed in SARAL/AltiKamission and in the continuation of Jason-2 mission, inorder to maintain the continuity of the great number of vir-tual station already functioning in the great tropical basinsand to ensure the needed complementarities between thedifferent missions.

Acknowledgments

CASAM and IHESA Projects supported this research.GPS field work was funded by CPRM and IRD (DinamicaFluvial do Sistema Amazonas-Solimoes) and by CNES(TOSCA project FOAM). The authors thank the studentsinvolved in the Research Initiation Program of RHASA,Research Team at Amazonas State University, Brazil,who produced the Jason 2 time series. The Ohio State Uni-versity (OSU) component of the research is partially sup-ported by NASA, and by the OSU Climate, Water, andCarbon (CWC) Program.

References

Altamimi, Z., Collilieux, X., Legrand, J., et al. ITRF2005: a new release ofthe international terrestrial reference frame based on time series of 454station position and Earth orientation parameters. J. Geophys. Res.112, http://dx.doi.org/10.1029/2007JB004949, 2007.

Arnault, S., Pujol, I., Melice, J.L. In situ validation of Jason-1 and Jason-2altimetry missions in the Tropical Atlantic Ocean. Marine Geodesy 34(3–4), 319–339, http://dx.doi.org/10.1080/01490419.2011.584833, 2011.

Bamber, J.L. Ice sheet altimeter processing scheme. Int. J. Remote Sens.15 (4), 925–938, http://dx.doi.org/10.1080/01431169408954125, 1994.

Berry, P., Galick, J.D., Freeman, J.A., Mathers, E.L. Global inland watermonitoring from multi-mission altimetry. Geophys. Res. Lett. 32,L16401, http://dx.doi.org/10.1029/2005GL022814, 2005.

1550 F. Seyler et al. / Advances in Space Research 51 (2013) 1542–1550

Bertiger, W.I., Bar-Sever, Y.E., Christensen, E.J., et al. GPS precisetracking of TOPEX/POSEIDON: results and implications. J. Geo-phys. Res. 99 (C12), 24449, http://dx.doi.org/10.1029/94JC01171,1994.

Bertiger, W.I., Desai, S.D., Haines, B., et al. Single receiver phaseambiguity resolution with GPS data. J. Geod. 84 (5), 327–337, http://dx.doi.org/10.1007/s00190-010-0371-9, 2010.

Birkett, C.M. The global remote sensing of lakes, wetlands and rivers forhydrological and climate research. In: Geoscience and Remote SensingSymposium, 1995 (IGARSS ‘95), Quantitative Remote Sensing forScience and Applications, vol. 3, pp. 1979–1981, 1995.

Birkett, C.M. The contribution of T/P to the global monitoring ofclimatically sensitive lakes. J. Geophys. Res. 100 (C12), 25179–25204,1995b.

Birkett, C.M. Contribution of the TOPEX NASA radar altimeter to theglobal monitoring of large rivers and wetlands. Water Resour. Res. 34,1223–1239, 1998.

Birkett, C.M., Mertes, L.A.K., Dunne, T., et al. Surface water dynamics inthe Amazon basin: application of satellite radar altimetry. J. Geophys.Res. 107 (D20), 8059, http://dx.doi.org/10.1029/2001JD000609, 2002.

Birkett, C.M., Beckley, B. Investigating the performance of the Jason-2/OSTM radar altimeter over lakes and reservoirs. Marine Geodesy 33(S1), 204–238, http://dx.doi.org/10.1080/01490419.2010.488983, 2010.

Bonnefond, P., Exertier, P., Laurain, O., Jan, G. Absolute calibration ofJason-1 and Jason-2 altimeters in corsica during the formation flightphase. Marine Geodesy 33, 80–90, http://dx.doi.org/10.1080/01490419.2010.487790, 2010.

Calmant, S., Santos da Silva, J., Moreira, D., et al. Detection of EnvisatRA2/ICE1 retracked altimetry bias over the Amazon basin rivers usingGPS, 51 (8), 1551–1564, 2013.

Cheng, K-H., Kuo, C.K., Tseng, H-Z., et al. Lake surface heightcalibration of Jason-1 and Jason-2 over the great lakes. MarineGeodesy 33, 186–203, http://dx.doi.org/10.1080/01490419.2010.487802, 2010.

Cretaux, J.-F., Calmant, S., Romanovski, V., et al. An absolutecalibration site for radar altimers in the continental domain: lakeIssyk-kul in Central Asia. J. Geod. 83, 723–735, 10.100/s00190.008.0289.7, 2009.

Cretaux, J.-F., Calmant, S., Romanovski, V., et al. Absolute calibration ofJason radar altimeters from GPS kinematic campaigns over lakeIssykkul. Marine Geodesy. 34 (3–4), 291–318, http://dx.doi.org/10.1080/01490419.2011.585110, 2011.

Dettmering, D., Bosch, W. Global calibration of Jason-2 by multi-missioncrossover analysis. Marine Geodesy 33, 150–161, http://dx.doi.org/10.1080/01490419.2010.490419, 2010.

Dumont, J.P., Rosmorduc, V., Picot, N., et al., OSTM/Jason-2 ProductsHandbook, 2008.

Frappart, F., Seyler, F., Martinez, J.-M., et al. Floodplain water storage inthe Negro river basin estimated from microwave remote sensing ofinundation area and water levels. Remote Sens. Environ. 99, 387–399,2005.

Frappart, F., Calmant, S., Cauhope, M., et al. Preliminary results ofEnvisat RA-2-derived water levels validation over the Amazon basin.Remote Sens. Environ. 100, 252–264, http://dx.doi.org/10.1016/j.rse.2005.10.027, 2006.

Fu, L., Christensen, E.J., Lefebvre, M. TOPEX /POSEIDON: the oceantopography experiment. Eos. Trans. Am. Geophys. Union. 72 (35),1991.

Haines, B., Desai, S.D., Born, G.H. The harvest experiment: calibration ofthe climate data record from TOPEX/Poseidon, Jason-1 and theOSTM. Marine Geodesy 33, 91–113, http://dx.doi.org/10.1080/01490419.2010.491028, 2010.

Leon, J.G., Calmant, S., Seyler, F., et al. Rating curves and average waterdepth estimation at the upper Rio Negro from altimetric spatial dataand calculated remote discharges. J. Hydrol. 328, 481–496, 2006.

Loyer, S., Perosanz, F., Mercier, F., et al. Zero-difference GPS ambiguityresolution at CNES–CLS IGS analysis center. J. Geod., http://dx.doi.org/10.1007/s00190-012-0559-2, 2012.

Marty, J.C. Documentation algorithmique du programme GINS, version5 juillet 2009, 563 available at <http://www.igsac.cnes.fr/documents/gins/GINS_Doc_Algo.html> (in French), 564, 2009.

Mercier, F. Satellite altimetry over non ocean areas: an improved wettropospheric correction from meteorological models. In: Joint EGS–AGU meeting, Nice, 2003.

Mercier, F., Picot, N., Lombard, A., et al. Improved Jason-2 altimetryproducts for coastal zones and continental waters (PISTACH Project).Eos Trans. AGU, 89 (53), Fall Meet. Suppl., Abstract G31B–0662, 2007.

Mertikas, S.P., Ioannides, R., Tziavos, I., et al. Statistical model and latestresults in the determination of the absolute bias for the radar altimetersof Jason satellites using the gavdos facility. Marine Geodesy 33, 114–119, http://dx.doi.org/10.1080/01490419.2010.488973, 2010.

Mertikas, S.P., Daskalakis, A., Tziavos, I.N., et al. Ascending anddescending passes for the determination of the altimeter bias of Jasonsatellites using the Gavdos facility. Marine Geodesy 34 (3–4), 261–276,2011.

Morris, C.S., Gill, S.K. Evaluation of the T/P altimeter system over thegreat lakes. J. Geophys. Res. 99 (C12), 24527–24539, 1994.

Pavlis, N.K., Holmes, S.A., Kenyon, S.C., Factor, J.K. The developmentand evaluation of the Earth gravitational model 2008 (EGM2008). J.Geophys. Res. 117, B04406, http://dx.doi.org/10.1029/2011JB008916,2012.

Ponchaut, F., Cazenave, A. Continental lake level variations from T/P(1993–1996). C. R. Acad. Sci. Paris, Ser. IIa 326, 13–20, 1998.

Quartly, G.D. Jason-1/Jason-2 metocean comparisons and monitoring.Marine Geodesy 33 (Special issue: OSTM/Jason-2 calibration/valida-tion), Supplement 1, 2010.

Roux, E., Cauhope, M., Bonnet, M-P., et al. Daily water stage estimatedfrom satellite altimetric data for large river basin monitoring. Hydrol.Sci. J. 53 (1), 2008.

Silva, J.S., Calmant, S., Rotuno Filho, O.C., et al. Water levels in theAmazon basin derived from the ERS-2 and Envisat radar altimetrymissions. Remote Sens. Environ. 114, 2160–2181, http://dx.doi.org/10.1016/j.rse.2010.04.020, 2010.

Washburn, S.A., Haines, B.J., Born, G.H., Fowler, C. The harvestexperiment LIDAR system: water level measurement device compar-ison for Jason-1 and Jason-2/OSTM calibration. Marine Geodesy 34(3–4), 277–290, 2011.

Watson, C., White, N., Church, J., et al. Absolute calibration in bassstrait, Australia: TOPEX, Jason-1 and OSTM/Jason-2. MarineGeodesy 34, 242–260, 2011.