You are on page 1of 9

WATERRESOURCES

RESEARCH,
VOL.30,NO. 4, PAGES1009-1017,
APRIL 1994

Variation of Great Lakes water levels derived


from Geosat altimetry
CharlesS. Morris
JetPropulsion
Laboratory,CaliforniaInstituteof Technology,
Pasadena

StephenK. Gill
National
Ocean
Service,
National
Oceanic
andAtmospheric
Administration,
SilverSpring,
Maryland

Abstract.A technique
for usingsatellite
radaraltimetrydatato estimate
thetemporal
variationof the water level in moderateto largelakesand enclosedseasis described.
GreatLakesdatafromthefirst2 yearsof the U.S. Navy'sGeosatExactRepeat
Mission
(November1986to November1988),for whichthereis an improvedorbit, are
usedto demonstrate the technique.The Geosatresh!tsare comparedto the lake level
datacollectedby the Great Lakes Section,NationalOceanService,NationalOceanic
andAtmospheric Administration, andarefoundto reproduce thetemporalvariationsof
thefivemajorlakeswith root-mean-square error(rms)rangingfrom 9.4 to 13.8cm and
a combinedaverageof 11.1 cm. Geosatdata are alsoanalyzedfor Lake St. Clair,
representinga moderate-sized lake, with a resultingrms of 17.0cm. Duringthis study
period,the water level in the Great Lakes variedin a typicalannualcycle of about0.2
m (0.5m for Lake Ontario)superimposed on a generaldeclineof approximately 0.5 m.
The altimeterdata reproducedthe generaldeclinereasonablywell for all the lakes, but
theannualcycle was obscuredin somelakes due to systematicerrorsin the altimeter
data.Currentand future altimetrymissionswill havemarkedlyimprovedaccuracy
whichwill permit many moderate(25 km diameter)or larger lakes or enclosedseasto
be routinely monitored.

Introduction Research, 1994; C. S. Morris and S. K. Gill, Evaluation of


the TOPEX/Poseidon altimeter system over the Great
Monitoringthe hydrologiccycle is of paramountimpor- Lakes, submitted to Journal of Geophysical Research,
tanceto determiningmagnitudeof global changeand under- 1994).
standingthe consequencesof our changingenvironment Many inland bodies of water throughout the world are
[Dozier, 1992]. It has direct implicationson the Earth's monitored with in situ measurement systems. However,
energybudget,as well as agricultureand other aspectsof satelliteradar altimetershave the potential of providingan
mankind's qualityof life. One aspectof the hydrologiccycle additional source of water level data [Mason et al., 1984,
is the variationof inlandbodiesof (normallyfresh)water. 1985;Guzkowskaet aI., 1986a, b; Rapley et al., 1987;Harris
Typically,these include lakes, rivers, marshes,and small et al., 1992]. For well-monitored bodies of water, altimeter
seas. Variations in the level and area of inland waters reflect
data can supplementin situ data that may not be readily
the amountof rain/snowmelt [e.g., Harris et al., 1992], available to the global change researcher or for which the
evaporation,and even humanintervention(suchas pump- measurement technology and data quality may not be
ing). known. Altimeter observations can also benefit studies of
Overthepast20 years,a numberof radaraltimetershave those bodies of water for which the in situ data are not
beenplacedin orbit.Designed
primarilyfor oceanographiccollected routinely or in dense enough networks to ade-
research,theseinstrumentsmeasurethe rangebetweenthe quately provide average levels. When combined with other
satelliteand the ocean surface. Because of limitations on the
satellite data sets depicting spatial extent, water volume of
knowledge of the satellite'sorbit,altimetricdatacouldonly these water bodiescan be routinely monitored. In addition to
beusedin a relativesenseto studytransient oceanographicwater levels, altimeter data also provide statistics on wave
features.
Theabsolute measurement of sealevel(relativeto
height and have the potential to establishfreeze/thaw dates.
anellipsoid)
waslimitedby radialorbiterror,whichcouldbe
Unlike many other remote sensing systems, altimeter mea-
a meter or more. However, within the last few years,
surementsare not dependenton cloud cover. Distribution of
knowledge oftheEarth'sgravitational fieldhassignificantly
altimeter data to researchers may occur as soon as a few
improved,resultinginmoreaccurate orbits.Fortherecently
launchedTOPEX/Poseidon satellite,the orbiterroris less weeks after the data are taken, dependingon the policiesof
than5 cm(B. D. Tapleyet al., Precision orbitdeterminationthe specificmission.
forTOPEX/Poseidon, submitted to Journalof Geophysical In this paper we have chosen data from the most recently
completed mission, the U.S. Navy's Geodetic Satellite
Copyright
1994bytheAmerican Geophysical Union. (Geosat),to examinethe potential for usingaltimetric data to
Papernumber94WR00064. monitor the water level variation of the Great Lakes and
0043-1397/94/94
WR-00064505.00 Lake St. Clair. These lakes were chosen because of their

1009
19447973, 1994, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/94WR00064 by Central South University, Wiley Online Library on [13/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1010 MORRIS AND GILL: VARIATION OF GREAT LAKES WATER LEVELS

number, size, and the fact that they are thoroughlymoni- Whentransitioning
fromlandto water,thealtimeter requires
tored. The large sizeof the Great Lakesprovidesbetween5 a certainamountof time to "lockup." Unfortunately,
Geo.
and 10 satellite tracks over each lake. In addition, there are sat,as comparedwith othersatellitealtimeters,waspartic.
two tracks over Lake St. Clair, a moderate-sizedlake. Theutady poor at obtaininglock when coming off of land.
sizeof the lakespermitsthe altimeterdatato beusedwithout The time requiredto lockup over inland bodiesof water
specialreprocessingto eliminateinterference fromland.The dependson the type of terrain being traversedprior to
six well-monitoredlakes provide a reasonablesampleto encountering waterand,in the caseof Geosat,theattitudeof
evaluatethe performanceof the altimeter. the spacecraft.In general,the moreruggedthe terrain,the
longerthe time neededto find the properrangeto the water.
The situationis further complicatedfor Geosatbecauseof
Geosat
the changingattitude of the spacecraft.The ability of the
Mission Overview Geosat altimeter to lockup as it transitioned from landto
Geosatradar altimeter satellitewas launchedby the U.S. water was dependenton the off-nadir pointing of the altim-
Navy in April of 1985to map the Earth's geoidin what is eter[Cheneyet al., 1991]. The largerthe off-nadirpointing,
referred to as the Geodetic Mission (GM). Most of the data the greater the data loss during the land-water transition.
from that mission are still classified. Once the GM was The attitude variation was primarily due to variationsof
completed,the satellitewas placed into a 17.05-dayexact solar radiation pressure on the satellite which produced
repeatorbit, the sameorbit as NASA's 1978Seasatsatellite. short-term (period of about 5 hours) and long-term(11-
The primary purposeof the Exact Repeat Mission(ERM) month period) variations in the satellite's attitude. The
was to provide data for oceanographicanalysis.The com- practical result is that for a repeat cycle, a given passover
plete ERM data set from November 11, 1986, to December the Great Lakes could include an extensive track of data.
31, 1989, is unclassified. The Geophysical Data Record while for the next cycle there may be little or no data for the
(GDR) ocean data set, which includes the Great Lakes data same pass.
used in this study, is available on CD-ROM from the There are severalcorrectionswhich must be appliedto the
National Oceanic and AtmosphericAdministration(NOAA) radar altimeter range estimate; these are documentedby
and is summarized by Cheney et al. [1991]. To date, the Cheney et al. [1991]. Most of these corrections result from
limiting factor on accuracy of satellite altimetry has always the radar pulsetraveling throughthe Earth's atmosphereand
beenthe knowledgeof the radial componentof the satellite's tendto delay the return signal. Specifically, theseincludethe
orbit. Recently, the University of Texas has recomputedthe dry tropospheric, wet tropospheric (for water vapor), and
orbit for the first 44 repeat cycles of the Geosat ERM the ionosphericcorrections.The first two are estimatedfrom
(November 11, 1986, to November 24, 1988), using an meteorological models, and the latter is estimated from an
improved TEG 2B gravity field (C. K. Shum et al., unpub- ionosphere model. Of these corrections, the ionosphere
lished manuscript, 1994). These improved orbits were pro- correction has the potential for the greatest error [see
vided by C. K. Shum (personal communication, 1993). The Cheney et al., 1991]. For the latitude of the Great Lakes, the
resultingradial orbit error has been reduced by more than a ionospherecorrection is typically much smaller than at the
factor of two. The remaining Geosat data are significantly equator, where the greatest errors occur. Nonetheless, there
degraded because increased solar activity, which increases is the potential of several centimeters error in the altimeter
drag, increased the orbit error. The data used in this study rangeestimatebecauseof uncertaintiesin thesecorrections.
are limited to the first 44 cycles of the ERM for which the In addition to the atmosphericcorrections, a correction
also has to be made for the solid Earth tide. Other tides are
improved orbit is available.
smalland thus, their effectsare ignoredfor this application.
Discussionof Altimetry Observations For instance, tidal loading is only a fraction of a centimeter
The basic quantity reported in the Geosat ERM GDR [Parke, 1982]anddirecttidal forcingchangesthe waterlevel
ocean data set is the altimeter's 1-s average height estimate by -+2 cm or less [Ferrel, 1874; Sanchez et al., 1985].
of the ocean (lake) surface above an ellipsoid for a given A detailed discussion of the radar altimeter measurement
location at a given time. This height estimaterepresentsan techniqueis given by Chelton et al. [1989].
averagewater level over an area, typically with a radius of
severalkilometers.This area, the altimeter'sfootprint, var- Lake Level Data for the Great Lakes
ies in size as a functionof wave height;the largerthe waves, Lake level datafor theseanalyseswere obtainedfromthe
the larger the footprint. Great Lakes Section, Ocean and Lake Levels Division,
The altimeter's water level height estimateis derived by NationalOceanService(NOS). The Sectionis responsible
accuratelyknowing the satellite's orbital position and thefor management of a permanentnetwork(in 1988)of 51lake
range from the satellite to the ocean (lake) surface, which is
level stationslocatedthroughoutthe Great Lakes Basin,
measured by the radar altimeter. The altimeter sends a includingthe connectingwaterwaysand Lake St. Clair.
13.65-GHz chirp to the Earth's surface,and the range is Severalstations havebeenin operationsincethe mid-1800s.
estimatedby timingthe return signal.The altimeterincludes The data are usedto supportregulation,navigation and
an onboardtracker to estimatewhen the return pulse will charting,river andharborimprovement,powergeneration,
arrive. The trackerassumesa slowlyvaryingsurface.When and scientific studies. All data are referenced to a common
the tracker encountersa rapidly changingslope,suchas is datum:International
GreatLakesDatum(IGLD), whichis
often producedby land, it can becomeconfused.When this also usedby a comparableCanadiannetworkof stations.
occurs, the altimeter tracker is said to be "out of lock" and Figure1 showsthelocationof theNOAA stationslocatedon
the range estimates are invalid or lost completely. The the Great Lakes and Lake St. Clair. The selected stations
trackerwill then attemptto reestablishthe properrange. used in this study are indicated.
19447973, 1994, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/94WR00064 by Central South University, Wiley Online Library on [13/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MORRIS AND GILL: VARIATION OF GREAT LAKES WATER LEVELS 1011

The NOAA stationsare each generallyconfiguredwith a Lake Ontario Water Level


primary
digital
float-driven
electromechanical
gauge provid- 76.0•a ....................... •.
ingpunched-paper-tape output.Thesegaugescollectdataat
15-rain
intervalswith a 0.3-cmresolution
andwitha backup
analogfloat-drivenmechanicalgaugecollectingdata on a • 74.5• •
stripchartwith a 0.3-cmresolution.A stationobserver
makesdaily checks on the systemsfor correct time and to 600 800 1000 1200 1400 1600
makeindependentwater level measurements usingan elec- Days From 1/1/1985
trictapegauge(ETG). The ETG observations are usedto Lake Erie Water Level
completetheeditingandprocessingof thedataandto ensure 175.5•'
b
thedataare continuouslyreferencedto datum.The gauges
arelocatedin heatedwalk-in enclosuressittingon top of 174.5

wellsor sumpslocated several feet from the shoreline.The
wellsare connectedto the water with undergroundhorizon- 1735[-
173:0•,, I 1987 I,
tal intakepipes located at sufficientdepth to be below the 600 800 1000 1200 1400 1600
expected icethicknesses. Theseconfigurations act asstilling Days From 1/1/1985
wellsfor the high-frequencywind waveswhile allowingfull
transmission of the lake variation frequencies. Figure 2. Examples of lake level measurement consis-
Usinghourlydata, the daily, monthly,andannualaverage tency: (a) four stations on Lake Ontario and (b) four stations
water levels are computed for each station as standard on Lake Erie. Values are referenced to the International
outputproducts. Highest and lowest daily averagewater Great Lakes Datum (IGLD). Observed departures in the
levels for each month and frequency distributions of the Lake Erie data result from wind-driven events.
dailyaveragewater levels are also compiled.For purposes
of this paper, average daily lake levels for each lake were
estimatedby averaging the daily lake levels from several smaller magnitudethan those displayed by the Lake Erie
strategicallyspacedstationsfrom each lake. Only one sta- data.
tion was used (and in operation) for the Lake St. Clair
analysis.
Figure2 illustratesthe consistencyamongstationson two Data Analysis Methodology
of the lakes. In both examples, Lakes Ontario and Erie, a In principle, monitoring temporal lake level variations,
time seriesof daily lake levels are plotted for four stations.
using Geosat or other spaceborne altimeter, is straightfor-
Althoughtheselakes are adjacent, the Lake Ontario stations
ward. For a given lake, the altimeter-derived lake level
displaysignificantlybetter consistency.The brief, but sig-
heightvariationsare evaluated as a function of time by using
nificant,departures from the mean trend by some or all ofcollinearanalysis[e.g., Cheney et al., 1983]. Each overflight
the Lake Erie stationsare due to wind-driven events (e.g.,
of the lake provides an estimate of the departure from the
seiche).Lake Erie is extremely susceptibleto seicheactionmean lake level. However, this estimate also includes altim-
in whichthe basin is set into periodic "slosh" motion in eter error. For convenience in these analyses, we have
responseto meteorological forcing [Forrester, 1983]. Al- chosento look only at the relative lake level variation. It is
thoughthe remaininglack displaysomelack of consistency also possible, but not trivial, to reference the results to a
amongstations,theseperiodsare lessextensiveand of much specificpoint on the lake and present them in terms of a lake
level datum.
In our analyses,we define one half of an orbital revolution
to be a pass. A given pass begins at either the northern most
or southernmost point in the satellite's orbit and continues
for 180 degreesof the orbit. Following the convention used
by the TOPEX/Poseidon mission, ascendingpass numbers
are odd and the descendingpassesare even. The passesare
numberedconsecutivelyfrom the beginningof the cycle. In
its 17-dayrepeat cycle, Geosat has 244 orbits, or 488 passes.
Of these, only 21 passes actually cross the Great Lakes;
these are shown in Figure 3. The number of passescrossing
individual lakes range from 10 for Lake Superior to five for
Lake Ontario. In addition, Lake St. Clair, a moderate-sized
lake between Lakes Huron and Erie, also has two passes.
Geosat pass 465 crosses every lake except Lake Ontario.
Because the level of each lake may vary differently, each
lake is analyzed separately.
One of the major components of the altimeter error is
unmodeled variations in the orbit. This orbit error shows up
Figure 1. Distribution of NOAA/NOS lake level stations most prominently at a frequency of once per revolution.
(boxes)alongthe Great Lakes and Lake St. Clair. The Over the ocean, it is possible to empirically estimate this
subsetof stationsusedin the analysis by a solid once per revolutionvariation and remove it. This procedure
is denoted
box.Stationsalong theadjoiningwaterways arenotshown. has been routinely used by satellite oceanographers.Re-
19447973, 1994, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/94WR00064 by Central South University, Wiley Online Library on [13/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1012 MORRIS AND GILL: VARIATION OF GREAT LAKES WATER LEVELS

wherer/(•Pbi
n, p) is the total numberof altimeterlakelevel
estimates
fallinginlatitudebinq:>bin,
for alloverflights
ofpass
p. Theequivalent expressionfor the average geoidheight
(gavg)
of thebinis
1 n(qO
bin,P
)
gavg
(q>bin'
P)= E g(q>,
se,P) (2)
n(q•
bin'P) 1
whereg(•, •, p) is also provided on the GDR.
In the presentstudy, we require that each latitudebinhave
observationsfrom a minimum of three overflightsfor thatbin
to be included in the analysis. This minimum is a small
number of overflights when compared to the maximum
possibleof 44 (cycles).However, this is misleadingbecause
no passover a givenlake had data for more than 33 cycles
and a typical pass only had data for 20 repeat cycles.A
Figure 3. The distribution of Geosat data used in the weightingprocedure is included in the analysis (see (4)
study.Passnumbersare indicated.Significantbreaksin the below) so that latitude bins with more data, and hence better
data typically indicate the presenceof an island. determined averages are given a proportionally greater
weight in establishingthe average departure from the mean
lake level.
cently, orbit analystshave developedtechniquesto account After the average lake level and geoid values are deter-
for the once per revolution error. Thus new orbit solutions, mined for each bin, the residuals Ah are calculated for each
suchas the TEG 2B (C. K. Shum, personalcommunication, overflight. The overflight residuals are corrected to the
1993), for Geosat and those for TOPEX/Poseidon, already location of the bin average using the known geoid variation
have eliminated most (or all) of this error. Thus it is not as shown below:
necessaryto estimate or correct for this type of orbit error.
The variation of water level across a lake will emulate the Ah(q•bi
n, p, of)= h(qv,•, p, of) - havg(rPbin,
p)
local geoid. To accountfor this spatial variation, a seriesof
0.05-degreelatitudebins are establishedfor eachpassover a -- (#(q>, •:, P) --#(q>bin,P)) (3)
given lake. Each bin represents6.4 km along the altimeter's For a given overflight, the weighted average departure
groundtrack. As the along-track distance between 1-s aver-
fromthe meanlakelevelZkhavg
is determined
using
age Geosat height observations is 6.6 km, this results in one
observationper bin per repeat cycle. Ideally, each latitude Ahavg(p,
of)
bin will receive a height observation for every Geosat
no( p ,of )
overflight. In reality, the amount of usable data varies from
one overflight to the next. This produces variable length Ah(rPbin,P, of)n(rPbin,P)
no(p,of)
tracks over the lake. The analysistechniquedescribedbelow 1

is designedaccountfor this problem. n(rPbin,P)


1
Data Selection
(4)
The data extraction from the Geosat GDRs is performed
on the basis of latitude and longitude. For each 1-s height where no(p, of) is the number of valid 1-s altimeter
estimategiven on the GDRs, there is an associatedstandard observationsor footprints across the lake in a given over-
deviation that provides a measure of the reliability of the flight.For Geosat,no is equalto the numberof latitudebins.
height estimate. This standard deviation is the only other For the Great Lakes analysis, a minimum of three 1-s
criterionusedfor including/excluding datain the preliminary observationswererequiredfor a givenpassto be includedin
analysis.Height values with a standarddeviationof greater the analysis.The two Lake St. Clair passes,one overLake
than 10.0 cm are excluded from the analysis. Erie and another over Lake Huron failed to have the
There are a numberof data quality flags[seeCheneyet al., required number of in-lock footprints. These cases were
1991] supplied on the GDR. In this study, these were not examined separatelyalthough the analysis procedurewas
used for the preliminary data selection, but were used as the same.
reference in the final data editing, as discussedbelow. Foreachoverflight,
correctedresiduals
Ahre
sareobtained
Estimating Relative Lake Level Variation byremoving
theaverage departure
fromthemeanlakelevel.
Defining h(qv, •, p, of) as a !-s altimeter lake level Ahres(rPbi
n,p, of)= Ah(q•bin,
p, of) - Ahavg(
p, of) (5}
estimatemade at latitude(rp)and longitude(0 on pass(p)
during
overflight
number lakelevel(havg) TheAhres valuesarecombined
(of), theaverage for all overflights
forthat
for a given latitudebin (½bin)is given by passandexamined(as a functionof latitude)for "trends"
and"blunderpoints"or outliers.The blunderpointstypi-
1 n(•in,P) cally occurnearland and are the result of land contamina-
havg(qøbin'
P)= E h(qo,
•, p, of) (1) tion. For this study, 1-s observations
were consideredout-
t/(qO
bin,/19) I liersandeditedfromthedataif Ahre
s wasmorethan2:30
19447973, 1994, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/94WR00064 by Central South University, Wiley Online Library on [13/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MORRIS AND GILL: VARIATION OF GREAT LAKES WATER LEVELS 1013

The entireanalysisprocedureis then repeatedwith the new


hayõ values.
The finalstepto obtaina completetime seriesof Ahavg
-20• ' X -- Geoid
Vm-lation4 values for a given lake is to combine the results from the
differentpasses.This is done assumingthat the temporal
lake level variations are the same over the entire lake, an
assumptionwhich is consistent with the NOAAJNOS lake
level measurements
(e.g., Figure 2). For a given pass,the
sumof the time seriesof Ahav
g valuesequalszero. To
combine the different passeshaving different temporal dis-
tributionsof overflights,all passesare referencedto the pass
with the greatestnumberof overflights(the base pass). For
a given pass (p), this is accomplishedby interpolating the
45.0 45.2 45.4 45.6 45.8 46.0
basepassAhavgto the overflighttimesof pass(p) and
Latitude minimizingthe sum of the squaresbetween pass(p) and the
base pass.This procedureresults in an average offset value
Figure4. Comparison
of the spatialvariationof meanlake forpass(p), whichisthenappliedto all ofthepass(p)Ahavg
levelfor pass235 over Lake Huron derived from Geosatand values. Thus
the estimatedgeoid variation [Rapp and Pavlis, 1990]. The
correctedGeosat values demonstratethe improvementob- Ahfinal(P,
of) = AHavg(p,of) + offsetpass(P)(8)
tainedby correctingfor unequal data distributionacrossthe
lake (see text). Data Editing
Data editing, which occurs after (5), is the most time-
consumingportion of the analysis procedure because it is
performedmanually. Fortunately, the data set is relatively
cm.The vastmajorityof the observationshaveAhresvalues
of within + 10 cm. The limit of -+30 cm was chosen to
smallfor a given pass.Plotting Ahres valuesas a functionof
latitude provides a visual method for identifying question-
accountfor the possibility of cross-lake gradients resulting
abledata. In addition,the following guidelineswere followed
from seiche,which can range in extreme casesfrom 20 cm
during data editing. Suspect points were examined for con-
for Lake Ontario to 2 m for Lake Erie [Forrester, 1983].
sistencywith adjacent observations and the expected geoid
After editing, the analysis procedure (equations(1)-(5)) is
gradient, and for increases (factor of 2 or greater) in the
redone with the edited data set.
altimeter-derived significantwave height (average height of
Trends, affecting one or more latitude bins in the cross-
the highestone third of the waves). Geosatdata quality flags
lake Ahres values, can occur due to missingdata in the were also reviewed.
latitudebins. The trends, when observed, are usually near
In addition to specificpoints, entire overflights with large
land where fewer observations are available and reflect
off-nadir pointing or very small waves were examined. Of
errorsin the computed mean lake level for the latitude bins
particular concern are calm, nonoceanlike conditions, which
in question.The weightingprocedureutilized minimizesthe
can produce spuriousheight estimates. Although numerous
importanceof these bins. However, it is possiblethe im- passes were examined, no obvious problem cases were
provethe lake level meansby usingthe informationprovided discovered.
by Ahres.This can be donebecausethe relativevariationof
lakelevelacrossthe lake is typicallystatic,a conditionnot
Evaluating the Altimeter Measurement System
trueof the ocean.Thus the averagevalue of Ahres for each
bin shouldbe zero, if the mean lake level for that bin, For the Great Lakes and Lake St. Clair, lake level
relativeto the other meanlake level valuesalongthe pass,is measurementsare available for comparison. This provides
correctlydetermined.A nonzeromeanfor the Ahres implies an opportunity to evaluate the performance of the Geosat
thatthe averagelake level is in error by approximatelythat altimeter system. The following steps were used to derive
amount.By correctingto the averagelake level for the bin in the altimeter residuals (altimeter - lake level measurement).
question,an improved spacialvariationof lake level along The daily average NOAA/NOS lake level measurements
thesatellitegroundtrack
is obtained.Figure4 illustrates
the (lakemeas)for the lake in questionare interpolatedto the time
typicalimprovementthat is obtainedusingthis procedure;it of each Ahfinalvalue. An average offset between the lake
showsthatalthoughthe correctionfor mostlatitudebinsis level measurementsand the Ahfinal values (OffSetlake)is
lessthan5 cm, one bin requiresa 9-cm correction.The determined by minimizing the sum of the squares. The
proceduregives a smoother variation of mean lake level, altimeter residual (ALTres), the estimate of the error in the
whichis consistent
withthe estimated
geoidvariation[Rapp altimeter system measurement for a given lake (L), pass
andPavlis,1990].Thisprocedureis summarized by (6) and (p), and overflight(of) is given by
(7):
ALTres(L, p, of)= Ah•inal(p,of)
1 n(•o
bi•,p)
havg
cor(rp
bin,
P)= E Ahres(q0bin,
p, Of') - (lakemeas(L,of) - Offsetlake(L)) (9)
tl(tfi
bin'P) 1
(6) By directly combiningALTres valuesfrom differentover-
flights, passes, and lakes, a time series of the estimated
havg(qO
bin,P) = havg(qø
bin,P)--havg
cor(qo
bin,P) (7) altimeter error is obtained. By smoothing these estimates
19447973, 1994, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/94WR00064 by Central South University, Wiley Online Library on [13/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1014 MORRIS AND GILL: VARIATION OF GREAT LAKES WATER LEVELS

Geosat (GEM T2 Orbit) - Lake Superior cycleof LakeSuperior


is visible,despiteits smallrange
! loo
a .......................... TM (about20 cm). However, close examinationrevealsthat
Geosatunderestimates, particularlyin 1987, and overesti.
matesin 1988.This trendis apparentin the resultsfor allthe
lakes.These systematicaltimetererrors will be discussed

j -•0• + ++ •+1•
++++++ later.
-1•

, , • ....
• 1987 •
1•
•m
•,
12•
1/1/1985
1• 16•
Figure 6 presentsthe results for the other lakes. The
number
of passes
(fiveto seven)for eachof theremaining
•osat (T•2B Orbit) - •e Su•or ,, Great Lakes is displayedin Figure 2. It shouldbe notedthat
pass465 over Lake Huron and pass 149over Lake Erie were

1• • .+ +
++
.... too shortto consistentlyprovide the required minimumthree

-50•
-1• - - 1 ....
1987
!...... + 1988
J . a
Geosat (TEG-2B Orbit) - Lake Michigan
........... m•s,
=ix.6•
• •m 1/1/1•

Figure 5. Comparisonof the derived temporallake level


variationfor Lake Superior from (a) the GEM T2 orbit and
(b) the more recent TEG 2B orbit. Each point is derivedby
averagingthree or more Geosat 1-sobservations(footprints) 600 800 1987 •
1000
•om
12•
1/1/1985
1988 lOOO

across the lake. The solid line is the NOAA/NOS-measured


lake level variation.
! • b+..............
Geosat (TEG-2B
Orbit)
-Lake
Huron
•S = ix.s •

(e.g., with a simple five-point running mean), significant •.+, •.. ++ •.,.1• +• • + +
features can be seen.
-50

Limitations -lOO • 1987 I 1988+


The methodology discussedabove works when the altim- Days From 1/1/198•

eter's groundtrack crossesa lake at a location which is wide


enoughto permit the altimeter to return valid height obser- Geosat (TEG-2B Orbit) - Lake Erie
vations. We estimate that the groundtrack over a given lake
must be at least 25 km long for this to occur. However, even
with a sufficientlylong groundtrack, topography, islandsor
other conditions may prevent valid height data from being
•!I
c.............m%'=
-50
5
•3.8•
+

+
collected. Under these circumstances or for smaller lakes, it
may still be possible to obtain usable height estimates.
However, this requires reprocessing the individual histo-
-100
600
, , •,
800 1987 1000 1200
Days From 1/1/1985

1988 1400 1600

gramsof returned altimeter power called waveforms,which


are archived on Sensor Data Records (SDRs). SDRs are d ma8'. •.4
voluminous in size compared to the GDRs and not widely
available. Despite this, reprocessingthese data can provide
usefulresults [Rapley et al., 1987; Koblinsl,? et al., 1993]. It
is recommendedthat our procedure be consideredfor the
larger lakes and enclosed seas, while the waveform repro- -50
r
-100 I 1987 [ 1988
cessingbe reserved for smaller bodies of water. Days From 1/1/1985

Lake Level Variation Geosat (TEG-2B Orbit) - Lake St. Clair

The Great Lakes e o


The largest of the Great Lakes also has the greatest
numberof Geosatpasses.As is indicatedin Figure3, Lake
Superioris covered by a total of 10 passesrangingfrom the
short passes 207 and 340 at the west end of the lake to -•0
• •87 1•88
+ •
severallong passesin the middle of the lake. Thesepasses Da• •m 1/1/1985
produced a total of 217 estimates of the relative lake level
variation. Figure6. The derivedtemporallake level variationfor
Figure 5 presents a comparisonof the estimatedrelative Lakes(a) Michigan,(b) Huron,(c) Erie, (d) Huron,and
St. Clair based on the TEG 2B orbit. For all the lakes, except
lake level variation derived from the GEM T2 orbit, which is
Lake St. Clair, eachpoint is derivedby averagingthreeor
distributed on the NOAA Geosat CD-ROMs, and the im-
moreGeosat1-sobservations
(footprints)acrossthe lake.
proved TEG-2B orbit. The improvement is dramatic. The The solidline is the NOAA/NOS-measuredlake level vari-
root-mean-square(rms) of the (Geosat-NOAA/NOS) lake ation.LakeSt. Clairtypicallyhadonlyoneor twoin-lock
level residuals decreases from 25.1 cm to 9.4 cm. The annual observations.
19447973, 1994, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/94WR00064 by Central South University, Wiley Online Library on [13/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MORRIS AND GILL: VARIATION OF GREAT LAKES WATER LEVELS 1015

footprints
acrossthe lake. These passesare considered
special
cases
andarediscussed
in thes•ctionof LakeSt.
Clair.
Thermsof the lake level variationfor the remainingfour
GreatLakes varies from 13.8 cm for Lake Erie to 9.4 cm for
LakeOntario. Each of theselakes displaysa generaldecline
in waterlevel which is detectedby the altimeter.However,
the annualcycles of Lakes Michigan, Huron, and Erie,
whichare quite small, are not convincinglyshown. The
failureof the altimeter to detect these annual cyclesis a
directresultof systematicerrors that have a magnitudeand
periodsimilarto the annualcycle of theselakes.It is
interestingto note that Lake Erie, with the worst rms value
of the five Great Lakes, is also the lake most affectedby
-•00 1987
600
I , ,

800 1000
, ,

1200
, ,• •

14OO 1600
,,

wind-drivenevents, resulting in inconsistencyin the lake Days From 1/1/1985


levelstationdata(Figure2). AlthoughLake Ontariohasonly
59estimatesof temporal water level variation, it alsohasthe 6O

greatest
annualcycle(orderof 50 cm).The altimeter-derived
estimatesfor Lake Ontario accurately reflect the observed
variation.

Lake St. Clair '•' 2o


The previousresults were based on a minimum of three
in-lock footprints across a lake for a given overflight.
Clearly,the more footprintsacrossa lake, the more accurate
• -20
the lake level estimate. The question is, How well can
moderate-sizedlakes be monitored using the available in-
-40-
lock data?With two separate passesto evaluate, each with
only one or two in-lock footprints over the lake, Lake St. 1987 1988

Clair provides an interesting test. 60O 800 1000 1200 1400 1600
In all, there are 37 estimates of lake level for Lake St. Days From 1/1/1985
Clair.The resultingrms is 17.0 cm, significantlyhigherthan
60
that observed for the other !akes (see Figure 6e). The
altimeterdoes detect a downward trend in the lake level, but
the very small annual cycle (10 cm) is totally missed.Once
again, systematicerrors in the altimeter data obscurethe
subtlechanges.
Two other passes, pass 149 over Lake Erie and pass465
2O
over Lake Huron, failed to have three footprints over their
respectivelake and thus provide additional tests for smaller
bodiesof water. Pass 149 provided only five water level
estimates
for Lake Erie, with a surprisinglow rms of 10.8
cm. However, pass 465, which included 10 estimatesfor -40
Lake Huron, had arms of 15.3 cm.
1987 1988
Clearly,the expectedGeosatrmsfor moderate-sized lakes -60 , , l, I .... • I..... • .... • ,,I , ,
will be perhaps50% higherthan that found for the larger 600 =oo 4oo
Days From 1/1/1985
lakes.Thissuggeststhat only significant
waterlevelchanges
will be detected. Figure 7. (a) Five-pointrunningmeanof the Geosatresid-
uals from the Great Lakes with the lake level variation
removed based on the GEM T2 orbit. (b) Same as Figure 7a
Geosat Altimeter Error except basedon the improved TEG 2B orbit. (c) Same as
With a known lake level variation, which then can be Figure 7b with trend and discontinuity removed. Annual
removedfrom the analysis,the Great Lakes altimeter data cycle of the 8-cm amplitude (peak to trough) is superim-
canbe usedto evaluatethe Geosataltimetererror. Figure7b posed.
displays
therunningmeanof the combined
residuals,
after
removalof lake level variation, from the analysesof Lakes
Superior,
Michigan,Huron,Erie, andOntario(Figures5b the ERM. Althoughthe effect of other orbital maneuverscan
and 6a-6d). Ideally, these residualsshouldbe random. be seenin the plottedresiduals,these are short-livedeffects
However,it is obviousthatthereis a trend,a "discontinu- and not of significancefor deriving temporal lake level
ity"andan"annual"cyclein theresiduals. variation.
Thediscontinuity,
nearday 860, beginswith oneGeosat Of greater concern is the trend in the residuals, which
orbitmaneuverandendswith another.In all, therewere25 amountsto 8.5 cm/year. The trend is much larger than would
suchmaneuvers(aboutoncepermonth)duringthisphaseof be expectedfrom any known error source [Wagner and
19447973, 1994, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/94WR00064 by Central South University, Wiley Online Library on [13/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1016 MORRIS AND GILL: VARIATION OF GREAT LAKES WATER LEVELS

Cheney,1992; Tapleyet al., 1992],exceptpossiblydrift in which, without correction,limititsusefulnessformonitoring


the orbit computation.The derivedslopedisagreeswith the lakes andsmallseas.It is obvious thatthealtimetry errors,
slightnegativeslopefound by Tapleyet al. [1992]usingthe particularly orbiterrors,are beingreducedsufficiently so
TEG 2 orbit for the entire ocean, althoughthe methodology that centimeteraccuracy(with averaging)in lake level
used to derive the TEG 2B orbit is somewhat different. It is variabilitycan be achieved.The TOPEX/Poseidonmission
curiousthat usingthe same data over the sametime period alreadyproducesdata productswith errors much smaller
with the earlier GEM T2 orbit fails to show this trend, thanthe Geosatdatausedin the presentstudy(C. S. Morris
althougha much larger annualcycle is evident(Figure7a). It andS. K. Gill, Evaluationof the TOPEX/Poseidonaltimeter
is likely that the observedtrend is a regional effect in the system over the GreatLakes,submitted to Journalof
TEG 2B orbit. GeophysicalResearch, 1994). These data, when combined
When the trend in the residuals and the discontinuityare with other satellite data sets (to identify changesin lake
eliminated, the remaining residuals display an apparent area),will permit the water volume of lakes to be monitored
annualcycle with an amplitude(peak to trough)of about 8 from space.
cm and with maxima just after the beginning of the year One drawback to using spaceborne radar altimetersfor
(Figure 7c). The pattern of the residualsagreesextremely thispurposeis that they only look at nadir. It is not possible
well with that found by Tapley et al. [1992, Figure4b] in their to directthe satelliteto look at specificlakes.Sampling
of
analysis of the global mean sea surface variation (basedon lakesisby chance only.Someof thisproblemismitigated by
the TEG-2 orbit). They discuss in their paper possible the European's ERS 1 mission launched in 1991, whichis
sourcesof error that might contribute to a cyclic error. The spendingmuch of its time in a 35-day repeat orbit (as
annual cycle may represent an error in one or more correc- comparedwith the TOPEX/Poseidon10-dayrepeatcycle).
tions (dry and wet troposphere, ionosphere, etc.). Orbit This repeat cycle provides denser spacingof the ground
error,whichv•ouldbeexpected
to havea periodequalto or tracks and thus a better opportunity to sample additional
one half of the synodic period of the satellite, is another lakes.With severalfuture altimeterscurrently underconsid-
possibility. The synodic period is the period of rotation of eration, each with different repeat cycles and orbital inclina-
the satellite's orbital plane relative to the Sun. As the orbital tions, the prospectsfor lake coverage appears favorable.
plane rotates, the distribution around the orbit of solar
radiation pressureon the satellite changes.If not modeled
Acknowledgments.We thank C. K. Shum, University of Texas,for
correctly,an orbit error can result. With the synodicperiod providingthe TEG 2B orbits and for extensive discussionsrelatedto
being about 11 months, it is impossibleto distinguishbe- thispaperandR. Cheney,NOAA/NOS, for reviewingan earlydraft
tween a true annual cycle and an 11-monthcycle with only 2 of this paper. This work was performed,in part, at the Jet Propul-
yearsof data. The agreementbetween the presentstudyand sionLaboratory, CaliforniaInstitute of Technology,undercontract
that by Tapley et al. [ !992] suggeststhat whatever the cause with the National Aeronauticsand Space Administration.
of the periodic variation, it occurs over the Great Lakes as
well as the ocean.
The effect of these altimeter errors is to obscuretemporal References
lake level variations. For Geosat, it should be possibleto Chelton,D. B., E. Walsh, and J. MacArthur, Pulsecompression and
remove many of these systematicerrors for the study of sea level tracking in satellite altimetry, J. Atmos. Ocean. Tech-
nol., 6, 407-438, 1989.
otherlakesbasedon the error analysispresentedhere.When
Cheney,R. E., H. G. Marsh, and B. D. Beckley, Global mesoscale
the trend, discontinuity and annual cycle are removedfrom variability from collinear tracks of Seasat altimeter data, J.
the Great Lakes data, the rms error dropsfrom 11.1 to 8.3 Geophys. Res., 88, 4343-4354, 1983.
cm, and the new residuals show no obvious trends. Cheney, R. E., N. S. Doyle, B. C. Douglas, R. W. Agreen,L.
Miller, E. L. Timmerman, and D.C. McAdoo, The complete
Geosat altimeter GDR handbook, NOAA Manual NOS NGS 7,
Conclusions Natl. Ocean Serv., Rockville, Md., 1991.
Dozier, J., Opportunitiesto improve hydrologic data, Rev. Geo-
The Great Lakes, Lake St. Clair, and the connecting phys., 30, 315-331, 1992.
rivers are well-monitoredby NOAMNational OceanService Ferrel, W., Tidal researches,in United States Coast Survey Report
(NOS) and do not require additionalmonitoring.However, Appendix,268 pp., U.S. Gov. Print. Off., Washington,D.C.,
1874.
they do providea usefultest caseto evaluatethe potential Forrester,W. D., CanadianTidalManual, 134pp., Department
of
useof satelliteradar altimetersfor worldwidemonitoringof Fisheriesand Oceans, Ottawa, Ont., 1983.
the variation of the levels of lakes and enclosed seas. Guzkowska,M. A. J., C. G. Rapley, and I. M. Mason, Satellite
Althoughmany of the world's lakes are monitored,the use altimeter measurementsover land and inland water, Eur. Space
Agency Spec. Publ., ESA SP-254, pp. 1563-1568, 1986a.
of satellitealtimetrycanprovidea low-cost,readilyavailable Guzkowska, M. A. J., I. M. Mason, C. G. Rapley, and F. A.
set of data to supplementexistingdata or providedata for Street-Perrott,The prospectsfor hydrologicalmeasurements
us-
lakes not currentlymonitored.For Geosat,a preliminary ing ERS-1, Eur. SpaceAgency Spec. Publ., ESA SP•48, pp.
survey,basedonan examinationof the availableheightdata, 321-326, 1986b.
suggeststhat our method could be used to monitor about 20 Harris, A. R., I. M. Mason, C. M. Birkett, and J. A.D. Mansley•
lakes in addition to the Great Lakes and Lake St. Clair. Lake Remote Sensingfor global climate research,Eur. Space
Agency Spec. Publ., ESA-SP-341, pp. 173-178, 1992.
These include the major lakes in northern Canada and Koblinsky, C. J., R. T. Clarke, A. C. Brenner, and H. Frey,
Africa. For the recentlylaunchedTOPEX/Poseidonmission, Measurement of river level variations with satellite altimetry,
this number jumps to more than 40 lakes and enclosedseas. Water Resour. Res., 29, 1839-1848, 1993.
The results of the Geosat evaluation of the Great Lakes is Mason,I. M., C. G. Rapley,F. A. Street-Perrott,
and S. P.
Harrison, Satellite altimetric measurementsof lake levels, Eur.
both encouragingfor the use of altimetersin general and SpaceAgencySpec.Publ., ESA SP-221,pp. 165-169,1984.
discouragingin that the Geosat data have systematicerrors Mason,I. M., C. G. Rapley,F. A. Street-Perrott,
andM. A. J.
19447973, 1994, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/94WR00064 by Central South University, Wiley Online Library on [13/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MORRIS AND GILL: VARIATION OF GREAT LAKES WATER LEVELS 1017

Guzkowska,ERS-1 observationsof lakesfor climateresearch, altimeter, in Sea Level Changes: Determination and Effects,
Eur.SpaceAgencySpec.Publ.,ESA SP-233,pp. 235-241,1985. Geophys.Monogr., vol. 69, edited by P. L. Woodworth et al., pp.
Parke,
M. E., O1, P1, N2 modelsof the globaloceantideon an 167-180, AGU, Washington, D.C., 1992.
elasticEarth plus potentialsurfaceand sphericalharmonicde- Wagner, C. A., and R. E. Cheney, Global sea level changefrom
compositions for M2, S2, andK1, Mar. Geod.,6, 35-81,1982. satellitealtimetry,J. Geophys.Res., 97, 15607-15615,1992.
Rapley,
C. G., M. A. J. Guzkowska,
W. Cudlip,
andI. M. Mason,
An exploratorystudyof inlandwater and land altimetryusing S. K. Gill, National Ocean Service, National Oceanic and Atmo-
Seasatdata, Eur. SpaceAgencySpec. Publ., ESA CR-2433,377 spheric Administration, 1305 East-West Highway, Silver Spring,
pp.,1987. MD 20910.
Rapp,
R. H., andN. K. Pavlis,Thedevelopmentandanalysis
of C. S. Morris, Jet PropulsionLaboratory, California Instituteof
geopotential
coefficient
models to spherical
harmonic
degree
360, Technology,4800 Oak Grove Drive, Pasadena,CA 91109.
J. Geophys.Res., 95(B3),21889-21911,1990.
Sanchez,
B. V., D. B. Rao, and P. G. Wolfson,Objectiveanalysis
for tidesin a closed basin, Mar. Geod., 9, 71-91, 1985.
Tapley,
B. D., C. K. Shum,J. C. Russ,R. Suter,andB. E. Schutz, (Received July 12, 1993; revised December 16, 1993;
Monitoringof changesin global mean sea level using Geosat accepted December 31, 1993.)

You might also like