Application of retracked satellite altimetry for inland hydrologic studies

Application of retracked satellite altimetry for inland hydrologic studies

MANMAN ZHANG†, HYONGKI LEE†, C. K. SHUM†, DOUG ALSDORF†,

FRANK SCHWARTZ†, KUO-HSIN TSENG†, YUCHAN YI†,

CHUNG-YEN KUO*‡, HONG-ZENG TSENG‡, ALEXANDER BRAUN§,

STEPHANE CALMANT¶, NAZIANO FILIZOLAjand FREDERIQUE SEYLER††

†Geodetic Science School of Earth Sciences, The Ohio State University, Columbus, Ohio, USA

‡Department of Geomatics, National Cheng Kung University, No.1, Ta-Hsueh Road, Tainan

701, Taiwan, ROC

§Department of Geosciences, University of Texas at Dallas, Richardson, Texas, USA

¶Universite de Toulouse, LEGOS UMR CNRS/UT3/CNES/IRD, Brasilia, Brazil

jUniversidade Estadual de Amazonas, Manaus, Brazil

††Universite de Toulouse, LMTG UMR CNRS/UT3/IRD, Brasilia, Brazil

We explored the application of satellite radar altimetry for the monitoring of small

inland bodies of water and hydrologic studies using a water-detection algorithm,

optimally retracked TOPEX/POSEIDON data at 10-Hz sampling, and investi-

gated the use of radar backscatter to improve land cover classification. The

procedure was demonstrated over Manitoba and south-western (SW) Ontario,

and the Amazon River Basin study regions. Compared with an L-band synthetic

aperture radar data generated water-land cover mask, the water-detection algo-

rithm detected more water points over the Amazon basin. High correlation of 0.98

between the retracked 10-Hz altimetry and the gauge measurements in Manitoba

confirmed that the retracked TOPEX data are more accurate than the non-

retracked data, and with higher along-track spatial resolution by virtue of its

higher sampling at 10 Hz.

1. Introduction

There are hundreds of thousands of reservoirs and millions of lakes in the world.

However, it is still unclear how the storage of these bodies of water changes seasonally

and interannually, and how they interact with global climate change. Although water

level measurements are critical to hydrological, hydraulic and climatologic studies,

their availability is hindered by the limitation of the traditional hydrological obser-

ving tools—in situ gauge measurements. The stage gauge measurements are onlyavailable at limited sites within a watershed and are especially restricted by the

inaccessibility of remote study regions. The availability of in situ data is limited by

logistical problems or political reasons resulting in the unavailability of hydrologic

data to users. Satellite measurements, in particular active remote-sensing measure-

ments such as satellite altimetry, have been demonstrated to be effective for hydro-

logic studies over a large basin or at the global scale with adequate temporal and

spatial resolutions.

*Corresponding author. Email: [email protected]

International Journal of Remote SensingISSN 0143-1161 print/ISSN 1366-5901 online # 2010 Taylor & Francis

http://www.tandf.co.uk/journalsDOI: 10.1080/01431161.2010.483495

International Journal of Remote Sensing

Vol. 31, No. 14, 20 July 2010, 3913–3929

Downloaded By: [Calmant, Stéphane] At: 15:16 9 August 2010

mailto:[email protected]

http://www.tandf.co.uk/journals

Satellite radar altimetry, such as TOPEX/POSEIDON, ENVIromental SATellite

(ENVISAT), JASON-1 and JASON-2 are originally designed for deep ocean circulation

studies, but capable of observing global surface water topography height with an

accuracy of several centimetres and a spatial resolution of ,100 km with approximately

weekly or monthly sampling. Birkett (1995, 1998, 2000) demonstrated the capability ofusing TOPEX satellite altimeter to monitor inland lakes with an area larger than 300 km2.

Hydrologic studies were performed using satellite altimetry over the Amazon and other

large river basins (Birkett et al. 2002, Coe and Birkett 2004, Berry et al. 2005, Calmant

et al. 2009). At present, research institutions providing surface water height data products

for hydrologic studies include the Laboratoire d’Etudes en Geophysique et

Oceanographie Spatiales (LEGOS) (http://www.legos.obs-mip.fr), and US Department

of Agriculture and NASA’s Goddard Space Flight Center (GSFC) (http://www.pecad.-

fas.usda.gov/cropexplorer/global_reservoir/). Both data centres primarily used 1-HzTOPEX/POSEIDON (and other altimetry) data or averaged data of several individual

10-Hz data without radar waveform retracking. Retracking is important for small water

bodies to enhance the altimeter data accuracy which could be low due to the land

contamination in the received signal. Retracked radar altimetry hydrologic data products

are available resulting from the ‘Contribution of Spaceborne Altimetry to Hydrology’

(CASH) Project (http://ocean.cls.fr/cash/), which was established by the Reseau Terre et

Espace, French Ministry of Research and Technology. The CASH project has provided

the retracked TOPEX/PODEISON data archive with the algorithms used in theENVISAT waveform retracking procedure, which are OCEAN, ICE1, ICE2 and

SEAICE retrackers (Benveniste 2002). These TOPEX/PODEISON retracked hydrologic

data products currently have about 80 altimetric time series over eight large fluvial basins

in the world (http://ocean.cls.fr/html/cash/welcome.html).

The regions studied in previous research are usually over large river basins, and

almost all of them (except the CASH project) did not use retracked measurements at 1

Hz, or at 10 Hz or higher sampled data. The returned waveforms over the large bodies

of water are well behaved according to the altimeter instrument design criteria overocean, primarily because these inland bodies of water are large (.300 km2) and

directly underneath the satellite groundtrack. As a result, the surface water heights

have been accurately computed assuming that the return radar waveforms conform to

the Brown model (Brown 1977) which is designed for deep-ocean radar altimeter

scatter modelling. When a body of water is smaller, the shape of the return waveform

changes and the non-retracked data cannot be directly used without retracking of the

radar waveform. This paper extends the technique developed by Lee et al. (2008,

2009), which optimally retracks individual 10-Hz TOPEX/POSEIDON waveformsfor geodynamics and wetland applications, to apply to several study regions contain-

ing small bodies of inland water for hydrologic studies. Another problem in the study

of small bodies of water is the unavailability, in general, of high-resolution masks with

classification of water and land covers at the sub-satellite points of the satellite

measurements. One such high-resolution mask was developed by Hess et al. (2003)

over the Amazon basin. They used JERS-1 L-band synthetic aperture radar (SAR) to

map Amazon wetland extents, land inundation and vegetation extents with 100-m

resolution and for high-water and low-water seasons. These masks have a potentialdisadvantage because even a two season per year classification may not be adequate to

completely classify the inundation extents of the study region. Other generated masks

used data from space-borne optical/infrared sensors, such as LANDSAT Thematic

Mapper (TM) and Moderate Resolution Imaging Spectroradiometer (MODIS)

3914 M. Zhang et al.


http://www.legos.obs-mip.fr

http://www.pecad.-fas.usda.gov/cropexplorer/global_reservoir

http://www.pecad.-fas.usda.gov/cropexplorer/global_reservoir

http://ocean.cls.fr/cash

http://ocean.cls.fr/html/cash/welcome.html

https://www.researchgate.net/publication/3231593_The_average_impulse_response_of_a_rough_surface_and_its_applications_IEEE_Trans_Antennas_Propag?el=1_x_8&enrichId=rgreq-72e9ed91-417e-4d4a-bb2d-40e1f363fd6e&enrichSource=Y292ZXJQYWdlOzIzMzIyNzkxNDtBUzo5OTI0Njg4NzkzMTkyNUAxNDAwNjczNzExMDc5

images (Enjolras and Rodriguez 2009), which often have problems with cloud and

smoke cover contaminations, or problems from the canopy above the water area.

Generation of masks is usually tedious and thus global high-resolution (,100 m)

masks are in general unavailable. A potential new classification method using infor-

mation from one satellite system, i.e. satellite radar altimetry, to classify the waterfrom other land covers using radar backscatter has been demonstrated by Papa et al.

(2003) using TOPEX/POSEIDON radar altimetry.

In this paper we propose an algorithm for semi-automated water-detection and

retrieval of accurate surface water height measurement by retracking of TOPEX radar

altimeter measurements, land cover classification using radar backscatter coefficients,

and based on the statistics of the generated water height inferred by radar altimetry.

The algorithm is tested in two study regions as a demonstration: one region covering

Manitoba and SW Ontario, and another covering the Amazon basin. Each of theseregions is distinct in its climatology, land cover and the size of water inundation areas.

The objective is to test the developed algorithm to detect and obtain accurate surface

water height measurements over the relatively small bodies of water. The combination

of the semi-automated classification method and the optimally 10-Hz retracked

measurement explores the potential of the application of satellite altimetry over

small terrestrial bodies of water, which can maximize the usage of satellite radar

altimetry to study surface water hydrology.

2. Method

One of the challenges in this study is to observe or to predict land surface character-

istics using satellite data. In particular, there is a need for robust classification between

water and non-water regions using satellite radar altimetry data. Prior to this work,

the procedure to accomplish this task included investigation of land cover character-

istics using satellite image or existing land cover maps, retrieving the satellite data at

the candidate points, and investigating the validity of the measurement at the candi-date points manually. This procedure is time-consuming, making it practically impos-

sible for large-scale studies. Moreover, the dependence on a priori information of land

properties from maps/masks could result in the loss of valuable measurement of the

bodies of water due to the limitation of the resolution or the inappropriate collection

time of the map. For example, in the Amazon region, there is a body of water with

seasonally varying extents and not detectable by satellite optical images as ‘water’, but

is verified by the detailed in situ hydrological map, as will be shown in this paper. Even

using a high-resolution Amazon mask—we used the high-water mask from the dual-season masks—developed using L-band SAR (Hess et al. 2003), as shown in the

paper, could miss water classifications presumably because of the more frequent

fluctuations of inundation extents, which may happen more than twice a year. In

this paper, the algorithm/procedure is developed and tested for semi-automatically

detecting water and extracting surface water height measurements over relatively

small inland bodies of water in two example study regions.

Here, we use TOPEX/POSEIDON data primarily for demonstration purposes. The

methodology is developed to use satellite radar altimeter data over small bodies ofwater. To take advantage of the repeat track property of satellite tracks, the data are

reorganized in a high-level product, called the stackfile (Kruizinga 1997, Yi 2000), to

allow efficient access and accumulation of the data or the water level time series. The

data processing and water-classification procedure is described here. First, the

Satellite-based observations of hydrological processes 3915


skeleton of a 1-Hz regional stackfile database is built. Then, the 1-Hz regional

stackfile database is densified to build the 10-Hz (or higher sampled) and

retracked satellite altimetry stackfile (Lee et al. 2008, 2009). The retrieved surface

water height is referenced to the so-called TOPEX reference ellipsoid with a mean

equatorial radius of 6378.1363 km and an inverse flattening of 1/298.257. Whenaltimetry measured water height is compared with stage measurement, we used

water level anomaly, which is the height with respect to the averaged water height.

In this study, the along-track and cross-track gradient corrections (Yi 2000, Lee

et al. 2008) are not applied in the stackfile generation to retrieve the water level

time series. In the study regions, gradient corrections are found to be small and

thus are not applied.

To find water measurements, all of the available bins in the stackfile are first

automatically selected to find the location where satellite altimeter data is availablewithin the given area. Each bin represents a specific location for particular 1-Hz data.

Then, within each selected bin, the data are further organized into groups by their

individual 10-Hz number from 1 to 10, denoting the first to the 10th 10-Hz data point

within the bin. Thus, each 1-Hz bin is divided into 10 locations corresponding to the

10-Hz data. For each of the 10-Hz locations, data from all cycles are analysed. The

interval between each TOPEX repeat cycle is 10 days, and thus for 10 years there are

about 360 cycles to analyse a particular 10-Hz location. Three iterations are carried

out consecutively to select the location as described below. During the first iteration,the measurements with missing or extremely large height value, or with the back-

scatter coefficient s0 ,5 dB (Birkett 2000) are edited. Following that, the mean value

is computed for the standard deviation computation in the next step. During the

second iteration, for each fixed location, the difference between the mean and the data

of each cycle is computed, then the proportion of the data with difference smaller than

M1 is investigated for the same location. We select the location if the proportion is

larger than N1 and the standard deviation of data from all cycles is smaller than S1.

During the third iteration, the proportion of the data with difference smaller than M2is investigated for each fixed location, and we select the location if the proportion is

larger than N2. Finally, the locations whose time series show good seasonality are

selected. The details of the procedure are shown in figure 1. The estimated parameter

values corresponding to M1, N1, S1, M2 and N2 are found to be 38%, 68%, 20 m, 2 m

and 4 m, respectively. These values have been chosen to optimally select over-water

altimetry data for the study regions: Manitoba and SW Ontario, and the Amazon

river basin.

3. Results

We apply the algorithm/procedure to two different study regions: Manitoba and SW

Ontario, and the Amazon River Basin. The retrieved TOPEX water level time series are

compared with the available in situ stage gauge measurements from U. S. Geological

Survey (USGS), Agencia Nacional de Aguas (ANA) Brazil, Environment Canada as

well as radar altimeter derived water height data products from LEGOS, U.S.Department of Agriculture (USDA) and CASH. The performance of the algorithm

for water-point detection is then compared with the results using the high-water mask

obtained by L-band JERS-1 SAR data over the Amazon River Basin (Hess et al. 2003)

along TOPEX tracks.



3.1 Results for Manitoba and SW Ontario

There are a huge number of lakes smaller than 400 km2 in Canada, in particular 3883small lakes in Ontario (Environment Canada, http://scitech.pyr.ec.gc.ca/waterweb).

The observation of these small lakes is important for the usage of these water

resources. Here, we present results of the small inland bodies of water in Manitoba

and SW Ontario detected by satellite altimetry using the algorithm. To show that our

individual 10-Hz retracked measurement has great potential for accurate stage esti-

mation, we demonstrate that our data can improve the spatial resolution and accu-

racy of the traditional 1-Hz or the average of several 10-Hz altimetry measurements.

Firstly, to assess the improvement of the accuracy of our individual 10-Hz retrackedsatellite altimetry measurement, we retrieved the TOPEX/POSEIDON data at the

locations where the stage measurements are available for a comparison study. The

retrieved water level measurements, located in large Lakes in Manitoba and SW

Ontario, are compared with the gauge measurements available from Environment

Canada (http://scitech.pyr.ec.gc.ca/waterweb). Although these example TOPEX time

series are located in large lakes, we conduct the study primarily due to the availability

of gauge data in these lakes, which enables us to validate the technique by quantifying

the improvement of the accuracy and by showing higher spatial resolution. Wecompare both the retracked and non-retracked TOPEX altimetry water level mea-

surements in regions containing numerous small bodies of water in Manitoba and SW

Ontario. We then compare our retrieved data with data available from other investi-

gators, including LEGOS, USDA and NASA/GSFC and CASH. Finally, we study the

use of 10-Hz automatic gain control (AGC) and retrieved water height measurement

for the smaller body of water where the stage measurements are not available. The

locations of the identified water points are shown in figure 2 and table 1.

OSU stackfile

(Kruizinga 1997, Yi 2000)

Basic cell: Bin addressed by row# & column#,Row# denotes latitude, Column# denotes pass

Find automatically

all the bins within

the given region

Iterations for each bin within the region: Extract 10-Hz retracked data (Lee et al.2008) from TOPEX Geophysical DataRecords (GDR) and Sensor Data Records (SDR)

For every 10-Hz location in each bin:

Do 3 iterations for all cycles

Iteration 1: edit data (σ0 < 5 dB), compute mean.

Iteration 2: compute standard deviation (std) and #

of cycles in which the height deviates from mean

height < M1, select the location if the proportion

of # > N1 and std < S1.

Iteration 3: compute # of cycles in which the height

deviates from mean height <M2, select the location

if the proportion of # > N2.

Column#

Row#bin

Each Bin includes data from all cycles for fixed 1-Hz location ;

#;

Figure 1. Water detection algorithm flowchart.



http://scitech.pyr.ec.gc.ca/waterweb

http://scitech.pyr.ec.gc.ca/waterweb

Figure 3 shows that we are able to detect the water surface and construct a TOPEX

water level time series in the Reindeer Lake in Manitoba, Canada, while there is no

satellite data available from the published USDA and NASA/GSFC or from the

CASH data site. We compare the TOPEX water level time series with the stage data

from Environment Canada after the data been interpolated to the same observational

Figure 2. Location of the example water points in large lakes used in the comparison and thedetected bodies of water in Manitoba and Western Ontario, Canada. The large red circlesdenote the TOPEX altimeter data locations used in the comparison, the purple triangles are theavailable stage gauges used for the comparison, the red points are detected bodies of wateralong the satellite tracks, and the small purple box denotes the region used to study the 10-HzAGC and the retrieved height measurement over the small body of water.

Table 1. Locations of water points in lakes in Manitoba and SW Ontario, Canada.

TOPEX data Gauge data

Longitude (�) Latitude (�) Longitude (�) Latitude (�)

Reindeer Lake 258.3661 57.8550 258.3333 57.8806Cedar Lake 259.8181 53.2113 259.8539 53.9394Lake of Woods 264.9846 49.0674 265.1906 49.7102



Fig

ure

3.

Sta

ge

ga

ug

eti

me

seri

esco

mp

are

dw

ith

1-H

zn

on

-ret

rack

ed(G

DR

)a

nd

ind

ivid

ua

l1

0-H

zre

tra

cked

da

taa

tR

ein

dee

rL

ak

e,B

roch

et,M

an

ito

ba

,C

an

ad

a.



time. The comparison shows that the satellite observation has an excellent correlation

of 0.87 with the in situ observations. The individual 10-Hz retracked data has slightly

higher correlation with the stage gauge measurement than the 1-Hz non-retracked

data. The higher correlation indicates the accuracy is improved using the individual

10-Hz retracked data. In addition, it can be seen from figure 4 that the correlation ofindividual 10-Hz retracked data reaches 0.98, which is higher than either the LEGOS

data or the 1-Hz non-retracked data at Cedar Lake, where there are no available data

from the USDA and NASA/GSFC or from the CASH data sites. Figure 5 demon-

strates the improvement of the retracked altimeter observations, with correlation

value increasing from 0.34 to 0.87 at the Lake of Woods where the area has small

bodies of water (figure 2) and there is no CASH data. Finally, we compare the

retracked height using different retrackers such as the Offset Center Of Gravity

(OCOG) retracker (Bamber 1994), and 10%, 20% and 50% threshold retracker(Davis 1997). Table 2 shows that the highest correlations of 0.98 and 0.87 are achieved

using the 10% threshold retracker for data over Cedar Lake and the Lake of Woods,

respectively.

Following the confirmation of the accuracy achieved using the individual 10-Hz

retracked satellite altimetry data, we applied the same methodology to detect smaller

bodies of water, and studied the usage of backscatter coefficient for land surface

classification in an inland region away from Cedar Lake (figure 2). The backscatter

coefficient (s0) is the ratio of the returned power and the transmitted power, and it hasthe potential to classify different land surface characteristics, as demonstrated by

Papa et al. (2003). The backscatter coefficients in TOPEX GDR are only available at

1-Hz frequency, and thus 10-Hz AGC is instead used. It has been known that there is

,20 dB difference between the AGC and s0 (Birkett 1998). We detected the body of

water while there are no available data from LEGOS, the USDA and NASA/GSFC

or the CASH project. To explore the difference between AGC value for different land

Figure 4. Stage gauge time series compared with LEGOS data, 1-Hz non-retracked (GDR)and individual 10-Hz retracked data at Cedar Lake, Manitoba, Canada.



properties and varying water level, we extract the AGC value during winter and

summer. The results are shown in figure 6. From the upper left plot, we can see the

10-Hz AGC value along TOPEX pass #178 in the chosen region is low during the

winter season when there is snow coverage. The corresponding water height is also very

low, as shown in the lower panel. These demonstrate that the low AGC value indicates

snow coverage or low water level in this region. Therefore, this indicates the advantageof our algorithm in high-latitude lakes where the AGC value cannot be used to classify

water signals during the winter season due to snow attenuation. The upper right plot

illustrates that AGC is high during the summer. As shown in the lower panel of figure 6,

at that time the water level is very high possibly leading to large inundated areas. Hence,

the high AGC value could be used to detect inland flooded regions.

3.2 Results for the Amazon River Basin

The Amazon River, with more than 1100 tributaries, is the world’s largest river in

terms of water volume, and seven of its tributaries are more than 1600 km (Alsdorf

Figure 5. Stage gauge time series compared with 1-Hz non-retracked (GDR) and retrackedTOPEX data at the Lake of Woods, SW Ontario, Canada.

Table 2. Comparison of the correlation coefficients between the gauge measurement anddifferent retracked water heights for the Lake of Woods, SW Ontario, Canada and Reindeer

Lake and Cedar Lake, Manitoba, Canada.

Retracker

Gauge OCOG 10% threshold 20% threshold 50% threshold

Lake of Woods 0.76 0.87 0.84 0.80Reindeer Lake 0.85 0.83 0.87 0.79Cedar Lake 0.88 0.98 0.97 0.91



et al. 2001). About 20% of world’s river water flowing into oceans is the discharge

from Amazon River into the Atlantic Ocean, which is about five times that of the

Congo River, and 10 times that of the Mississippi River. Due to the massiveness of the

water basin and the prohibitive costs of obtaining stage data, scientists have used

satellite radar altimetry to study Amazon hydrology (Birkett et al. 2002, Berry et al.

2005).We demonstrate here the potential of satellite altimetry to study smaller rivers with

higher spatial resolution using retracked individual 10-Hz TOPEX data, by applying

the algorithm to detect smaller bodies of water. The Amazon mask we used (Hess

et al. 2003) was made from L-band SAR imageries with a resolution of 100 m for dual-

season, including high-water and low-water conditions. After delineation of the wet-

land, land surface properties are classified into nine categories (Hess et al. 2003) using

pixel attributes. Among the nine levels of classifications, open water, aquatic

TOPEX 10-Hz AGC at 20:38:23~20:38:29 on 12 January 1996 TOPEX 10-Hz AGC at 4:8:56:~4:9:03 on 29 July 1996

Figure 6. The retrieved decadal water height for the small body of water denoted by the purplestar point and the 10-Hz AGC along TOPEX pass #178 when the water level is low (12 January1996) and when the water level is high (29 July 1996). The AGC measurement times are denotedby the red circles in the time series plot. This study region is away from Cedar Lake, Manitoba,as denoted by the purple box in figure 2.



macrophyte (floating grass), flooded shrub, flooded woodland and flooded forest are

considered as water. All other non-flooded and unclassified areas are excluded as non-

water surface in the comparison.

We compared the water data points obtained using our algorithm with the points

found using the high-water level mask. Figure 7 shows the mask region and thecomparison results. The red points indicate that the algorithm detected the bodies

of water missed by the mask. In the lower plot of figure 8, we compare the water height

measurement using individual 10-Hz retracked data with stage measurement at Santa

Maria de Boiacu Gauge in Branco river. The good agreement indicates the individual

10-Hz retracked data could also provide good data accuracy in the Amazon River

Basin.

We choose three regions (figure 7) including seven example water points (or loca-

tions) to validate the points selected only by the algorithm using the detailed hydro-logical map, and the results are shown in figures 8–10. The satellite altimeter

measurements at these seven example points in the three tested regions are not

available from LEGOS and USDA and NASA/GSFC. Figure 9 shows the validation

of the region 1 in Xingu River in figure 7, and part of figure 7 is shown in the upper

right panel of figure 9 for convenience. The lower plot verifies the existence of water

detected only by the algorithm at the purple points using the detailed hydrological

map in which the detected water locations are shown as water bodies. Similarly,

figures 8 and 10 verify the existence of water at the selected points in other two regionsusing the hydrological maps.

The presence of water can also be indicated by the AGC value. Similar to what has

been done in the Manitoba and SW Ontario study region, we extracted the 10-Hz

AGC at the typical low water season and high water season for region 1, as shown in

the lower left and lower right plots, respectively, in figure 9, in which the AGC is

overlapped in the detailed local map. The detailed local map shows that there are

bodies of water in the locations where the AGC value is high. Furthermore, the AGC

value became higher in February, when the water level is higher, whereas it becamelower in February and September when the water level is lower. This implies that the

high AGC value indicates the existence of water. In addition, as shown in figure 9,

Figure 7. Comparison of the algorithm-detected points and the mask-detected points in theAmazon River Basin. Blue points denote the water points selected by both the water-detectionalgorithm and using the Hess mask, and the red points denote the water points detected only bywater-detection algorithm. Three green boxes denote the three tested regions for water pointsdetected only by the algorithm.



another body of water detected only by the algorithm, denoted by the uppermost

purple point along pass #37, has a relatively low AGC value. This may indicate thatthe low AGC value cannot be used to verify the non-existence of water because the

low AGC value could be obtained over the small body of water surrounded by

vegetation (Papa et al., 2003). Overall, when the AGC value is high, the existence of

Figure 8. Validation of region 3 in the Amazon basin. The mid-left plot is a detailed local mapused to verify the existence of water at the purple points selected only by the algorithm. The redpoints denote the stage gauge closest to the detected points. The stage gauge Katuema does nothave data during the TOPEX/POSEIDON operation time (1992–2002). The mid-right plot is toshow the location of the detailed local map in figure 7. The upper plots are example time seriesof altimeter measurement. The lower plot is the comparison of gauge measurement and TOPEXmeasurement at Santa Maria de Boiacu Gauge and shows excellent agreement.



TOP

EX

10-

Hz

AG

C a

t 8:1

4:42

~8:1

5:2

on 1

Sep

tem

ber

1996

TO

PE

X 1

0-H

z A

GC

at 2

1:49

:39~

21:4

9:59

on

16 F

ebru

ary

1997

Water level anomaly (m)

Fig

ure

9.

Va

lid

ati

on

of

reg

ion

1in

the

Am

azo

nb

asi

n.T

he

up

per

rig

ht

plo

tsh

ow

sth

elo

cati

on

of

the

det

ail

edlo

calm

ap

(reg

ion

1)

infi

gu

re7

.Th

elo

wer

left

an

dlo

wer

rig

ht

plo

tsp

rese

nt

the

10

-Hz

AG

Ca

lon

gT

OP

EX

pa

ss#

37

inth

ed

eta

iled

loca

lm

ap

at

dry

sea

son

(1S

epte

mb

er1

99

6)

an

dw

etse

aso

n(1

6F

ebru

ary

19

97

),re

spec

tiv

ely

,an

dth

eti

mes

are

als

oin

dic

ate

db

yth

ere

dci

rcle

sin

the

up

per

left

plo

t.T

he

up

per

left

plo

tis

the

retr

iev

edT

OP

EX

wa

ter

hei

gh

tm

easu

rem

ent

at

the

low

erp

urp

lep

oin

ta

lon

gp

ass

#3

7.

Th

est

ag

em

easu

rem

ent

isco

mp

are

dw

ith

the

TO

PE

Xa

ltim

eter

mea

sure

men

ta

nd

sho

ws

go

od

sea

son

ala

gre

emen

t.T

he

det

ail

edlo

calm

ap

inth

elo

wer

plo

tsv

erif

ies

the

exis

ten

ceo

fw

ate

ra

tth

ep

urp

lep

oin

tsd

eno

tin

gth

eb

od

yo

fw

ate

rse

lect

edo

nly

by

the

alg

ori

thm

,an

dth

ere

dp

oin

tsin

the

det

ail

edlo

calm

ap

den

ote

the

sta

ge

ga

ug

escl

ose

stto

the

det

ecte

dp

oin

ts.T

he

sta

ge

ga

ug

eF

aze

nd

aC

ipa

ub

ad

oes

no

th

av

ed

ata

du

rin

gth

eT

OP

EX

/PO

SE

IDO

No

per

ati

on

tim

e(1

99

2–

20

02

).T

he

up

per

mid

dle

plo

tis

an

oth

erex

am

ple

tim

ese

ries

of

alt

imet

erm

easu

rem

ent

at

the

up

per

mo

stp

urp

lep

oin

ta

lon

gT

OP

EX

pa

ss#

37

.



the body of water can be inferred, but when the AGC value is low, it could not be used

to confirm the non-existence of water.

The upper left plot of figure 9 compares the altimeter data with the stage data at

Altamira, which is one of the two stage data available during TOPEX/POSEIDON

operation time (1992–2002) over the study region. Unlike the good amplitude agree-

ment at Santa Maria de Boiacu Gauge, where the location of the altimeter and gaugedata is exactly the same, as shown in the lower part of figure 8, this 10-Hz retracked

data location, detected only by the algorithm, is far away from the Altamira gauge,

which is on the main stem, as shown in the lower left plot of figure 9. Consequently,

there is an amplitude difference between altimeter and gauge data. However, because

the two locations belong to the same tributary, as shown in the local detailed map, the

altimeter data agrees well with the stage gauge measurement in terms of the seasonal

variation, except around 1998. The satellite altimeter measurement was edited out

during 1998 when relatively lower water level can be observed from the stage data,which results in data loss due to the contamination of water return by the surrounding

topography.

As in the Manitoba and SW Ontario study region, we compare different retrackers

in the Amazon River Basin. The correlation coefficients between the stage and the

retracked water height measurements were 0.56, 0.31 and 0.45 using OCOG, 20% and

50% threshold retrackers, respectively. Compared with the results over the Manitoba

and SW Ontario region, the correlation coefficients are relatively low and there are

obvious amplitude differences between altimeter and gauge datasets. This could be

Figure 10. Validation of region 2 in the Amazon basin. The upper left plot is the detailed localmap used to verify the existence of water at the purple points selected only by the algorithm. Thered points denote the stage gauge closest to the detected points. The middle right plot is to showthe location of the detailed local map in figure 7. Other plots are example time series of altimetermeasurement.



due to the fact that the gauge station is located in the stem of the river while the

altimeter track point is passing across the river (figure 7), thus measuring the stage

variations in the inundation area connected to the river. Such inundation areas

usually present reduced water stage variations compared to the river, as found in

this study. Moreover, the OCOG retracker seems to be the optimal retracker over theAmazon, which agrees with Frappart et al. (2006), whereas the 10% threshold

retracker performed better in the Manitoba and SW Ontario study region. This

indicates that waveform retrackers could perform differently and care must be

taken to identify the best radar altimeter waveform retrackers for different regions.

4. Conclusions

To expand the application of satellite altimetry in the inland hydrological study, we

have developed a procedure to detect the inland small bodies of water using retrackedhigh-spatial sampling (10 Hz) TOPEX/POSEIDON satellite altimeter range and AGC

measurements. We applied the methodology over two climatically different study

regions: Manitoba and SW Ontario in Canada, and the Amazon River Basin. The

algorithm is based on the backscatter coefficients and the statistical analysis of altimetry

water level measurements. The TOPEX water level time series for small bodies of water

over the two study regions are retrieved using the individually retracked 10-Hz data

with spatial resolution of about 680 m along the satellite ground track.

In the Manitoba and SW Ontario study region, the retrieved 10-Hz water heightmeasurements are compared with available stage measurements as well as altimetry

products from LEGOS and USDA. This has shown that the 10-Hz waveform retrack-

ing obviously improved the correlation coefficient (for example, from 0.34 to 0.87 for

Lake of Woods). The individual 10-Hz retracked measurement also provides higher

correlation with stage measurement than using 1-Hz non-retracked data directly from

the geophysical data record (GDR). This demonstrates that the data accuracy could

be substantially improved by retracking technique. In the Amazon region, we

detected small bodies of water, validated using the detailed hydrological map; theseare not detected using the Hess et al. (2001) Amazon basin high-water mask generated

using L-band SAR. In addition to being capable of detecting more water bodies using

our technique, the measurement from TOPEX shows excellent agreement in season-

ality with stage data over small bodies of water. The usage of individual 10-Hz

retracked measurement selected by the algorithm improved the quantity and quality

of the data over the GDR or other publicly available data products. The combination

of the developed algorithm, optimal waveform retracking of the 10-Hz TOPEX data

and the radar backscatter data enabled detection of more small water bodies andachieved higher data accuracy, which could potentially greatly improve our ability to

quantify water level variation over small water bodies and thus to understand their

hydrology.

Acknowledgements

This research is funded by grants from NSF’s Hydrology program, NGA’s NURI

program, and the Ohio State University Climate, Water and Carbon Program. We

acknowledge NASA and CNES for providing the TOPEX/POSEIDON (GDR and



https://www.researchgate.net/publication/251432267_Water_level_changes_in_a_large_Amazon_lake_measured_with_spaceborne_radar_interferometry_and_altimetry?el=1_x_8&enrichId=rgreq-72e9ed91-417e-4d4a-bb2d-40e1f363fd6e&enrichSource=Y292ZXJQYWdlOzIzMzIyNzkxNDtBUzo5OTI0Njg4NzkzMTkyNUAxNDAwNjczNzExMDc5

SDR) data products; LEGOS, USDA and NASA/GSFC for providing high-level

radar altimetry data products; ANA Brazil and Environment Canada for providing

the river stage gauge data used for the study. We thank Mike Durand, the guest editor,

and an anonymous reviewer for the constructive comments which improved the

manuscript.

References

ALSDORF, D., BIRKETT, C., DUNNE, T., MELACK, J. and HESS, L., 2001, Water level changes in a

large Amazon lake measured with spaceborne radar interferometry and altimetry.

Geophysical Research Letters, 28, pp. 2671–2674.

BAMBER, J.L., 1994. Ice sheet altimeter processing scheme. International Journal of Remote

Sensing, 15, 925–938.

BENVENISTE, J., 2002, ENVISAT RA-2/MWR Product Handbook, Issue 1.2, PO-TN-ESR-RA-

0050 (Frascati, Italy: European Space Agency).

BERRY, P.A.M., GARLICK, J.D., FREEMAN, J.A. and MATHERS, E.L., 2005, Global inland water

monitoring from multi-mission altimetry. Geophysical Research Letters, 32, L16401,

doi:10.1029/2005GL022814.

BROWN, G.S., 1977, The average impulse response of a rough surface and its applications. IEEE

Transactions on Antennas and Propagation, 25, pp. 67–74.

BIRKETT, C.M., 1995, The contribution of TOPEX/POSEIDON to the global monitoring of

climatically sensitive lakes. Journal of Geophysical Research, 100, pp. 25, 179–25, 204.

BIRKETT, C.M., 1998, Contribution of the TOPEX NASA radar altimeter to the global mon-

itoring of large rivers and lakes. Water Resources Research, 34, pp. 1223–1239.

BIRKETT, C.M., 2000, Synergistic remote sensing of lake Chas: Variability of basin inundation.

Remote Sensing of Environment, 72, pp. 218–236.

BIRKETT, C.M., MERTES, L.A.K., DUNNE, T., COSTA, M.H. and JASINSKI, M.J., 2002, Surface

water dynamics in the Amazon Basin: Application of satellite radar altimetry. Journal

of Geophysical Research, 107, LBA 26–1, doi:10.1029/2001JD000609.

CALMANT, S., SEYLER, F., and CRETAUX, J., 2009, Monitoring continental surface waters by

satellite altimetry. Surveys in Geophysics, 29, pp. 247–269.

COE, M.T. and BIRKETT, C.M., 2004, Calculation of river discharge and prediction of lake

height from satellite radar altimetry: Example for the Lake Chad Basin. Water

Resources Research Journal, 40, W10205.

DAVIS, C., 1997, A robust threshold retracking algorithm for measuring ice-sheet surface

elevation change from satellite Radar Altimeters. IEEE Transactions on Geoscience

and Remote Sensing, 35, pp. 974–979.

ENJOLRAS, V. and RODRIGUEZ, E., 2009, Using altimetry waveform data and ancillary informa-

tion from SRTM, Landsat and MODIS to retrieve river characteristics. IEEE

Transactions on Geoscience and Remote Sensing, 47, pp. 1869–1881.

FRAPPART, F., CALMANT, S., CAUHOPE, M., SEYLER, F. and CAZENAVE, A., 2006, Preliminary

results of ENVISAT RA-2-derived water levels validation over the Amazon basin.

Remote Sensing of Environment, 100, pp. 252–264.

HESS, L.L, MELACK, J.M., NOVO, E.M., BARBOSA, C.F. and GASTIL, M., 2003, Dual-season

mapping of wetland inundation and vegetation for the central Amazon basin. Remote

Sensing of Environment, 87, pp. 404–428.

KRUIZINGA, G.L., 1997, Validation and applications of satellite radar altimetry. Ph.D. Dissertation,

the University of Texas, Austin, Texas, USA.

LEE, H., SHUM, C., YI, Y., BRAUN, A. and KUO, C.-Y., 2008, Laurentia crustal uplift observed

using TOPEX/POSEIDON radar altimetry over land. Journal of Geodynamics, 46, pp.

182–193.



LEE, H., SHUM, C., YI, Y., IBARAKI, M., KIM, J.-W., BRAUN, A., KUO, C.-Y. and LU, Z., 2009,

Louisiana wetland water level monitoring using retracked TOPEX/POSEIDON alti-

metry. Marine Geodesy, 32, pp. 284–302.

PAPA, F., LEGRESY, B. and REMY, F., 2003, Use of the Topex-Poseidon dual-frequency radear

altimeter over land surfaces. Remote Sensing of Environment, 87, pp. 136–147.

YI, Y., 2000, The Ohio State Univeristy stackfiles for satellite radar altimeter data. Geodetic

Science Report No. 495, The Ohio State University, Columbus, Ohio, USA, reprinted

May 2010.



Application of retracked satellite altimetry for inland hydrologic studies

Documents

Transcript of Application of retracked satellite altimetry for inland hydrologic studies