Oct 15, 1998 - issue to be addressed is the precision of the ac-9 measurements. We present various models for the spectral relationships of each of the ...
JOURNAL
OF GEOPHYSICAL
RESEARCH,
VOL. 103, NO. Cll, PAGES 24,955-24,968, OCTOBER
15, 1998
Global relationships of the inherent optical properties of the oceans
A. H. Barnard, W. S. Pegau, and J. R. V. Zaneveld Collegeof Oceanicand AtmosphericSciences,Oregon State University,Corvallis
Abstract. We have collecteddata duringeight separateresearchcruisesfrom open ocean to estuarineoceanicenvironments.Inherent opticalpropertydata collectedduringthese cruiseswere incorporatedinto a large databasetotaling 1914vertical samples.The range of eachinherentopticalpropertywithin this databasespansover 2 ordersof magnitude. Using this database,we examinethe spectralrelationshipsof each of the inherentoptical propertiesbasedon the measurementsmade at 488 nm. The resultsof this studyshow that there are dependencies in the individualinherentopticalproperties(IOP) spectra that are linearlyrelated. The informationfrom the regressionmodelsis usedto explain the lineardependencies observedin the globaldata set.A separatedata set collectedfrom a recentcruiseis usedto compareregionalrelationshipswith the global.The implication of this researchis that over a diversityof oceanicregimes,there are fundamental,firstorder relationshipsin the individualIOP spectra.These relationshipscan provide an estimateof the individualIOP spectralrelationshipswhen no information about the IOP is available,as is often the casein oceancolor remote sensing.More detailed models, however,may be necessaryin order to more accuratelypredict the IOP spectral relationshipson regionalscaleswhere the expectedrange of variabilityis small. 1.
Introduction
The inherentopticalproperties(IOP), definedas thoseoptical propertiesthat are independentof the ambientlight field [Preisendorfer, 1976], determine the magnitudeand spectral signatureof the light propagatingthroughthe water. The IOP are the absorption,scattering,and beam attenuationproperties of the water and its dissolvedand suspendedmaterials. The apparentopticalproperties(AOP), suchas radianceand irradiancelevels(percentof surfacevalues)and the diffuse attenuationcoefficientare connectedto the IOP via the equation of radiative transfer. Together with the backscattering coefficient,the absorptioncoefficientdeterminesthe diffuse reflectanceof the ocean,which is used for remote sensing purposes.Phytoplankton,detritus, and dissolvedmatter all have different absorptionand scatteringcharacteristics that potentially allow one to infer their concentrationsfrom the remotelysensedspectralradiance. Various models have been proposedto estimatethe IOP basedon the AOP [Prieurand Sathyendranath, 1981;ZaneveM, 1989; Gordon, 1991; Kirk, 1994; Lee et al., 1996]. However, AOP measurementscan only be made during the daylight hoursand canbe difficultto interpretowingto varyingsurface conditions,winds, ship reflections, and sky conditions.The AOP are very difficultto measurenear the surface,where the influenceon the remotelysensedradianceis the largest.Furthermore,discriminationof the particulateand dissolvedfractionsof the IOP cannotbe inferred directlyfrom the measure-
and dissolvedmaterialis estimatedby analyzingthe fractionof the water sampleretainedon a filter pad (usually0.2/xm) and the fractionthat passesthroughthe filter [Trt;iper and Yentsch, 1967; Mitchell and Kiefer, 1988a]. Pigment extraction techniques using organic solventsof the filter-pad sample have been usedto separatethe phytoplanktonand detrital portions of the absorptioncoefficient[Kishinoet al., 1985].While these analysesprovide high spectralresolution,they typicallyonly providedata from discreteportionsof the water column.Furthermore, the determinationof absolutevalues using these methodsis still subjectto debateowingto uncertaintiescaused by the filter and extraction technique [Mitchell and Kiefer, 1988b;Bricaudand Stramski,1990;Mitchell, 1990]. The beam attenuation
coefficient
can be used to determine
suspendedparticulateload [Pak et al., 1988].Most commonly, in situ measurements
of the beam attenuation
coefficient
are
made using broadband, single-wavelengthtransmissometers [PetzoldandAustin, 1968;Voss,1992].While theseinstruments do provide high vertical resolutionestimatesof beam attenuation, they are limited in the amountof spectraldata they can provideduringa singleprofile. Recent advancesin opticalinstrumentationhaveallowedfor noninvasivemeasurementof the spectralabsorptionand beam attenuationcoefficients(WET Labs Inc., ac-9 meter) of the oceanson vertical scalessimilar to traditional conductivitytemperature-depth(CTD) measurements [Mooreet al., 1992; ZaneveM et al., 1992; Roeslerand Zaneveld, 1994; Bricaud et al.,
1995a]. Over the past 3 years, we have made spectral IOP measurementsin six different areas during eight separatereThe mostcommontechniqueusedto determinethe absorpsearchcruisesusingthis instrumentation.Our approachis to tion coefficientis spectrophotometricanalysisof filtrat½sof incorporate the datacolIected fromthesecruises into a large water samples.The absorptioncoefficientdue to particulate databaseto examinethe spectralvariabilityof eachof the IOPs Copyright1998by the American GeophysicalUnion. over large time and spacescales.This data setis uniquein that it containsspectraldata on each of the IOP parametersand Paper number 98JC01851. 0148-0227/98/98JC-01851509.00 encompasses a wide variety of oceanicregimes. ments of AOP.
24,955
24,956
BARNARD ET AL.: GLOBAL INHERENT
OPTICAL PROPERTY RELATIONSHIPS
It is the purposeof thisstudyto examinethe generalspectral dard once each samplingday. A BarnsteadNanopure© water relationshipsof the IOP on a globalbasis.From a descriptive systemis usedto producethe cleanwater calibrationstandard. standpoint,this researchrepresentsone of the first studiesto When properlycalibrated,the ac-9hasbeenshownto havean of -0.005 m- • (Twardowski et al.,submitted manuexaminethe large-scaledistributionsof the IOPs, measured accuracy of the ac-9dataincludesapplying simultaneouslyat various wavelengths.From an analytical script,1997).Postprocessing standpoint,this researchcan provideinformationon the gen- temperatureand salinity correctionsto the absorptionand eral spectraltendenciesin eachof the IOPs that spana diver- beam attenuation measurementsfollowing the methods desityof oceanicregimes.It is the goalof thisresearchto exam- scribedby Pegauet al. [1997]. The absorptionmeasurements ine the causesof thesespectraltendenciesand how theycanbe are further correctedfor scatteringerrors to accountfor the usedto obtainmore informationon the IOP variability.As the portion of scatteredlight not measuredby the detector[ZanIOP data were collected during various cruises,a secondary eveldet al., 1994;Moore et al., 1996].As the scatteringcorrecissueto be addressed is the precisionof the ac-9measurements. tion of the absorptionmeasurementutilizesthe absorptionand We presentvariousmodelsfor the spectralrelationships of beam attenuation measurements at 715 nm, the IOP measureeach of the individual IOPs. As we will show, these relationmentsat this wavelengthare not includedin this analysis. Fractionationof the total absorptioncoefficientinto disshipsare quite strongwhenconsidering wide rangesof oceanic environments. However, the reader is cautioned that the re- solvedand particulateportionsis accomplished by utilizingtwo sultspresentedin this work may not be applicableto all oce- ac-9meters.A 0.2/•m pore sizefilter is placedon the intakeof anic scales and environments. In fact, as we will show, these the absorptiontube of one of the ac-9 meters to provide a relationshipscan be quite different on local scales.More de- direct measurementof the absorptioncoefficient of chrotailed modelsmay be necessaryin order to more accurately mophoric dissolved materialaa(h)' In someinstances, occaadjustmentof predictthe IOP spectralrelationshipson regionalscales.Fur- sionalproblemscreatedby the filter necessitated thermore,the relationshipsmay changetemporallydepending theaa spectrum byreferencing thedissolved absorption proabsorption on the magnitude of the values owing to differencesin the filesto the valuemeasuredat 676 nm.The dissolved compositionof very clearwaterscomparedwith turbidwaters. measurements at 676 nm are therefore not included in this study.The secondac-9 is operatedunfiltered,providinga direct measurementof the absorptionand beam attenuation 2.
Field
Methods
and Data
We have assembledthe slow descentrate optical platform (SlowDROP)to providesimultaneous hydrographical andop-
coefficients duetoparticulate plusdissolved materials, apa(•), andCpa (•), respectively. Astheac-9meters provide measure-
mentsof absorptionand attenuationcoefficientsin reference to pure water, we have chosento denotethe absorptionand CTD measurements.The platform is free falling and slightly attenuationin these terms. They are equivalentto the total negativelybuoyant,which eliminatesthe effectsof ship mo- absorptionor beam attenuationcoefficientminusthe contrition. A typical instrumentconfigurationon the platform in- bution by water. cludestwo spectralabsorptiona and beam attenuationc coThe particulate absorption fractionap(,•), whichcontains efficient(ac-9) metersand a CTD. Data from the instruments contributions by detrital,phytoplankton, and sedimentary maare integratedinto a singlestreamusinga WET Labs, Inc., terial, is estimatedby subtractionof the filtered absorption data acquisitionsystem(MODAPS). Platform descentrates measurements from the unfiltered measurements
tical measurements
on vertical
scales similar
to traditional
areapproximately 10-30 cms-•, providing betterthan10 cm vertical resolutionof the water column. Typically,the entire water column(near surfaceto bottom) is profiled, exceptin caseswhere the bottom depth is greater than 150 m due to restrictionsimposedby the length of the data cable. Spectralabsorptionand beam attenuationcoefficientmeasurementsare made usingWET Labs,Inc., ac-9 metersat nine wavelengthsh: 412, 440, 488, 510, 532, 555, 650, 676, and 715 nm. For the purposeof thispaper,onlya brief overviewof the ac-9 calibration,deployment,and processingproceduresimplementedduringeachfield experimentis given.For a detailed descriptionof theseprocedures,the readeris referredto WET Labs., Inc. (www.wetlabs.com) and M. L. Twardowskiet al. (Quantificationof the micro-to finescalein situchromophoric DOM absorptionand total absorptionin coastalwaterswith an ac-9, submittedto Journalof Atmosphericand Oceanic Technology,1997). The ac-9 metersare mountedverticallyon the opticsplatform,with tubingextensions placedon the flow tube intake nozzles.Submersiblepumpsare attachedto the outflow
ap(X)= apg(h)- ag(X)
The scattering coefficient for particlesbp(,•) is derivedby subtractionof the absorptioncoefficientfrom the beam attenuation coefficient,assumingRayleighscatteringof dissolved materialsis negligiblecomparedwith that of particulatematter
b(x) =
-
Over the past 3 years,we havecollectedIOP profilesusing the SlowDROPsystemduringeightresearchcruiseslocatedin six separategeographicalregions.A descriptionof the locations and datesof these cruisesis given in Table 1. The IOP data from these cruisesencompassa wide variety of oceanic regimes ranging from the equatorial Pacific to eutrophic coastal and estuarine
waters. While
all of the IOP
data were
collectedin the northern hemisphere,we considerit to be globalin the senseof the dynamicrangeobservedin eachlOP. In these terms,we considerthe data set to be global in magnituderange,not necessarily in a spatialsense.The readerwill nozzle of each of the flow tubes, so that in situ water is drawn alsonote that the global data set doesnot containwaters that along the optical path. To insure that the in situ water mea- wouldbe consideredto be purely"blue"(i.e., where the IOPs dependent on the characteristics of thewateritself). suredby eachac-9 is sampledfrom the samedepth,the intake are strongly Over 1,000 profilesof the IOP have been collectedduring tubesare positionedat the sameverticallevel on the platform these cruises,at an averagevertical resolutionof approxiand in very closehorizontalproximity. To avoid the possibleeffectsof instrumentationdrift, each mately 0.2 m. Thus the total number of depth samplescolac-9 is normallyfield calibratedrelative to a cleanwater stan- lected for each of the IOPs is well over 200,000 at each wave-
BARNARD
ET AL.: GLOBAL INHERENT
OPTICAL
PROPERTY
RELATIONSHIPS
24,957
Table 1. Global Data Set Description Approximate Number
Location
Data Set
Dates
Latitude, øN
Longitude,øW
of
Profiles
Number
Near coastal California Gulf of California
OCE95 GOC95
Oct. 1995 Dec. 1995
33.22 28.00
117.44 112.50
8 11
200 436
Equatorial Pacific East Sound,Washington North Atlantic Shelf ChesapeakeBay
ZF96 ES96 CMO96 COPE96
April-May 1996 May-June 1996 Aug.-Sept. 1996 Sept. 1996
01.00 48.65 40.30 36.80
165.00 122.89 70.30 75.83
9 12 10 17
235 340 306 203
GOC96
Nov. 1996
27.80
111.50
12
249
CMO97
April-May 1997
40.30
10
632
Gulf
of California
North Atlantic Shelf
70.30
of
Samples
Experimentabbreviations are asfollows:OCE95, Oceanside,California,1995;GOC95, Gulf of California,1995;ZF96, Zonal Flux (equatorial Pacific),1996;ES96,East Sound(PugetSound),Washington,1996;CMO96, CoastalMixing and Optics(North Atlantic Shelf), 1996;COPE96, ChesapeakeBay OutflowExperiment(Maryland), 1996;GOC96, Gulf of California,1996;CMO97, CoastalMixing and Optics(North Atlantic Shelf), 1997.Total numbersof profilesand samplesfor the 1995-1996data set are 77 and 1914,respectively.
length. In order to reduce the number of data points to a manageablelevel, representativeprofilesof the IOP were se-
brational and/or instrumentionalerrors (i.e., biases). We
lected
lentto the accuracy of the ac-9measurements (0.005m-•).
from
each of the individual
cruise data sets and were
verticallybinnedby selectingthe median value in each depth interval. Data from each of the cruiseswere pooled into a globaldatasetresultingin 1914depth-binnedsamplesfor each of the IOP parametersdescribedaboveat eachof the respective wavelengths.As we are interestedin examiningthe spectral relationshipsof each of the IOPs on a global scale,no effort hasbeenmadeto excludeanywater typeor composition.
therefore assumethat the uncertaintyin the model is equiva-
Orthogonalleast squaresregressions are usedto derive the linear modelsof individual IOP parametersin order to minimize the errors in both the 488 nm estimate
and the estimate
at other wavelengths.The choiceof the 488 nm wavelengthas the independentx variablewas made for a numberof reasons. First, both the particulate and dissolvedmaterials affect the optical propertiesat 488 nm, which allowsthis wavelengthto Furthermore, no effort has been made to examine the rela- be includedin inversionsof all parameters.Second,for remote tionshipsbasedon verticalstructure.All portionsof the water sensingpurposes,488 nm is a wavelengthat whichreflectance column wereincorporated, ihcluding the bottomnepheloidis commonlymeasured.Third, the 488 nm band is near the center of the wavelength range of the ac-9 measurements, layer, if presentin shallowwater casts. which minimizeserrorsthat couldbe causedby usingtoo large a wavelengthinterval. The standarderror of the slope and y 3. Results and Discussion intercept as computedfor each of the regressions in the folThe spectralrelationshipsof each IOP are examinedin the lowing sections.As thesevalueswere alwayscomputedto be followingsections.As the resultswill show,the spectraldepen- lessthantheuncertainty in theac-9meter(-0.005 m-•), these dencyof each IOP can be describedbasedon simple linear values are not shown. regressions.Although not included in this presentation,the The magnitudeof eachof the opticalparametersat 488 nm spectraldependence was alsoexaminedusinglog-transformed included in the global data set varies over approximately2 data. While the log transformationmodels provide for more ordersof magnitude(Table 2). Note that individualcruisedata equal weightingof all data points, they are, however,more setsdo not exhibitthe samerangeof variabilityas observedin difficultto interpretand requirea zero intercept.In mostcases the globaldata set.For the data obtained,only 14% of the total examined,only a slightimprovementin the spectralIOP rela- numberof observations are in the upper order of magnitudeof tionshipmodelsusinga log transformationwasfound.Because each IOP parameterdata set. In order to examinethe influlinear modelsdo not requirea zero intercept,they canprovide ence of magnitudeon the spectralIOP relationships,we proan estimateof the possiblebiasesin the IOP measurements. vide results from the entire global data set and from two Sinceone of the goalsof this researchis to examinethe pre- reduced data sets. The first reduced data set contains 14% of cision of the ac-9 measurements, the choice was therefore the total observationsand encompasses approximatelythe upmade to use the simpler linear models. per order of magnitudeof the IOP valuesin the globaldata set. The spectralrelationshipsare examinedusingthe estimates The secondreduceddata set,whichcontainsapproximatelythe of the individualIOP parameter at the 488 nm wavelengthas lower order of the IOP magnitude,includes86% of the obserthe independentvariablex and the measurementsat each of vationsin the globaldataset(that is,mostof the data are from the other wavelengthsas the dependentvariabley. relativelyless turbid water). Finally, comparisonsare made betweenthe relationships resolvedfrom the globaldatasetand y(X) = /3x(488 nm)+ m an IOP data set (CMO97) collectedduringa recentfield exThe/3 term is the slope,and the m term is they intercept.The perimentin order to indicatehowvariablethe relationshipsare uncertainties in the model are assumed to be random and at local or regional scales. equivalentto the accuracyof the ac-9 measurements (+0.005
Dissolved AbsorptionCoefficient ag(•.) m-•). Inaccurate calibrations or meterperformance, however, 3.1. Chromophoric may causesystematic biasesin the measurements. The y interTheaa at eachof theac-9measured wavelengths areplotted cepts returned from the model regressionsgreater than the versus aa (488nm) in Plate1. The resultsof the regressions, ac-9 measurementaccuracyare assumedto be causedby cali-
givenin Table 3, showthe spectralrelationshipsat eachof the
24,958
BARNARD
ET AL.'
GLOBAL
INHERENT
OPTICAL
PROPERTY
RELATIONSHIPS
0.6
ß 412
0.5 ß 440 ß 510
0.4 ß 532 ß 555 _
ß 650
... 0.2
_
0.1 -
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
ag(488nm)coefficient (m'•) Plate 1. Dissolved absorption coefficient aa measured at 488nmversus thedissolved absorption coefficient measuredat the six other wavelengthsfor the entire global data set. The linear regressionfits at each wavelength(seeTable 3) are shownas lines in the corresponding color.
wavelengths to be strongly linear,withr2 valuesgreaterthan ing that they are not significantlydifferent from zero and are 0.94 with the exceptionof the 650 nm band. The magnitudeof
thus unbiased.
aa (650nm)is typically lessthan0.01m-•, verynearthe
The spectrumof absorptionby dissolvedorganic material has often been assumedto decreaseexponentiallywith wavelength [Bricaudet al., 1981; Carderet al., 1989;Roesleret al., 1989]. Over the UV and visiblewavelengths,the spectrumof the absorptionby dissolvedorganicmaterialhasbeenmodeled with the generalexponentialfunction
accuracyof the ac-9 meter. Thus the low percentageof variability explainedby the model of the absorptionby dissolved materials at 650 nm is due to the decreasedsignal-to-noise ratio at thiswavelength.At eachwavelength,they interceptis lessthan the measurementaccuracyof the ac-9 meter indicat2.5
o 412 2
-
ß440
ß510
1.5
,,•',,,.,• .
555
ß650
ß676
1
•
•
-"'•
ß .:..•••.•'
•
-'-
.
0.5 ,.
,
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
a•,(488rim)coefficient (m'•) Plate 2. Sameas Plate 1, but for the particulateabsorption coefficient ap. SeeTable 4 for the linear regressionstatistics.
BARNARD
ET AL.: GLOBAL INHERENT
OPTICAL PROPERTY
RELATIONSHIPS
24,959
Table 2. Inherent Optical Property RangesWithin Each Individual Cruise Data Set at 488 nm
Data Set OCE95 GOC95 ZF96 ES96 CM 096 COPE96 GOC96 Global CM 097
apg 0.009-0.318 0.023-0.096 0.008-0.028 0.137-1.543 0.027 -0.209 0.092-0.523 0.032-0.257 0.008-1.543 0.051-0.107
ag 0.003-0.038 0.017-0.030 0.004-0.011 0.059-0.125 0.018-0.040 0.052-0.159 0.018-0.036 0.003-0.159 0.020-0.036
ap 0.006-0.297 0.001-0.070 0.000-0.019 0.070-1.476 0.003-0.171 0.019-0.367 0.001-0.226 0.000-1.476 0.018-0.080
Cpg 0.159-1.925 0.043-0.507 0.026-0.157 0.478-8.079 0.114-1.876 0.412-3.373 0.052-1.489 0.026-8.079 0.179-0.551
bp 0.146-1.696 0.016-0.411 0.016-0.131 0.283-6.595 0.064-1.705 0.318-3.116 0.018-1.241 0.016-6.595 0.125-0.475
Values are m- •
that the spectralshapeof the absorptionby dissolvedmaterials is independentof magnitudeof total absorption.As the global where/3•/i= exp [-S(,• i - •r)], aa(Xr) is the dissolved data set used in this study encompassesa variety of oceanic absorptionat a reference wavelengthand S is the spectral environments anda widerangeof aa values,we conclude that slope of the absorptionby dissolvedmaterials.The spectral simple linear models can be used to accuratelypredict the slopehasbeen estimatedfor variouswater typesand hasbeen spectrumof the absorptionby dissolvedorganic matter for found to be weaklyvariable in the global ocean environment most oceanic environments. [Bricaudet al., 1981; Carderet al., 1991;Roesleret al., 1989; Bloughet al., 1993;Hogeet al., 1993]. In typicaloceanicenvi- 3.2. ParticulateAbsorptionCoefficient a•,(X)
ag(X,): ag(h.r)•3 w
ronments, the spectral slopevariesfrom0.014to 0.018nm-1,
The ap valuesfromthe globaldatasetat eachwavelength
dependingon location[seeRoesleret al., 1989]. versusthe ap (488nm) are shownin Plate2. The absorption The averagespectralslopeS was determinedfor the global coefficientdue to particulatematter wasderivedfrom two ac-9 data set usingtwo methods.First, the slopesfrom the regres- measurements;the uncertaintyof each is assumedto be ran-
sionsof aa(X) versus aa (488nm),shownin Table3, werefit
dom and equalto 0.005m-1. Thus the uncertainty in the
to the exponentialmodel shownaboveusingthe absorptionby particulateabsorptionmodeldue to calibrationand instrument dissolvedmaterials at 488 nm as the reference wavelength. errorsisassumed to be0.007m-•. Thepercentage of variabil-
Thismethodresultedin a spectralslopeof S = 0.016 nm-1, withanr2 = 0.99. In thesecond method,the spectral slopeS
ity explainedis greaterthan 95% for eachof the wavelengths
(Table4). In general,thelowestr2 valuesandlargest y inter-
was determinedusingthe exponentialmodel at eachvertically ceptsare in the blue and red portionsof the spectrum.In the binnedsampleusingaa (488nm)asthereference wavelength. blue wavelengths,the shape of the absorptionspectrumis The ensemblespectralslopemeanwascomputedby takingthe stronglydependenton the concentrationof detritus and the average of sample spectralslopesover all of the depth bin chlorophyllpigmentstructureof phytoplankton[Morrowet al., samples.The mean slope over all profiles was S = 0.015 1989;Bricaudand Stramski,1990;Bricaudet al., 1995b]. It is nm-1, with a standarddeviationof +_0.003. The goodagree- thereforenot surprisingthat the 412 nm band showsthe lowest
mentbetweenthe twomethods (within0.001nm-•) suggestsr2 value,wherevariability in theproportion of detritalto pig-
that the spectralshapeof the absorptionby dissolvedorganic matter is accuratelymodeledusingthe simplelinear relationshipsresolvedfrom the global data set. Furthermore, these resultssuggestthat the absorptionby dissolvedmaterialscan be accuratelyestimated using the exponential form shown above,usingthe absorptionmeasuredat 488 nm as the refer-
mentedabsorbingmatter is expectedto be the greatest.In the red portion of the spectrum,differencesin the absorption spectraof detritusand phytoplankton,variabilityin the shape of the secondarymaximumof chlorophyllpigments,and vari-
Table3. Regression Results of aa(X) Versusaa (488nm)
Table4. Regression Resultsof ap(X) Versusap (488nm)
Using the Entire Global Data Set
Using the Entire Global Data Set
ationsin theblueto redabsorption ratioscausetheloweredr2
valuesand the increasedy intercepts. encewavelength anda spectral slopeS = 0.015 nm-1. We noted that 14% of the particulateabsorptionvaluesat Similarresultsin the spectralrelationshipswere found using 488 nm in the globaldata set are in the upper 90% of the data the lower and upper order of magnitudedata sets,suggesting range, 0.142-
