Active-passive correlation spectroscopy: a new technique for identifying ocean color algorithm spectral regions

Active-passive correlation spectroscopy: a new techniquefor identifying ocean color algorithmspectral regions

Frank E. Hoge and Robert N. Swift

A new active-passive airborne data correlation technique has been developed which allows the validation ofexisting in-water ocean color algorithms and the rapid search, identification, and evaluation of new sensorband locations and algorithm wavelength intervals. Thus far, applied only in conjunction with the spectralcurvature algorithm (SCA), the active-passive correlation spectroscopy (APCS) technique shows that (a) theusual 490-nm (center-band) chlorophyll SCA could satisfactorily be placed anywhere within the nominal 460-510-nm interval, and (b) two other spectral regions, 645-660 and 680-695 nm, show considerable promise forchlorophyll pigment measurement. Additionally, the APCS method reveals potentially useful wavelengthregions (at 600 and -670 nm) of very low chlorophyll-in-water spectral curvature into which accessorypigment algorithms for phycoerythrin might be carefully positioned. In combination, the APCS and SCAmethods strongly suggest that significant information content resides within the seemingly featureless oceancolor spectrum.

1. Introduction

Primary production in the oceans is presently esti-mated to be about one-third of the total global plantfixation of carbon dioxide.' The actual mean andvariance of carbon fixation in the sea, however, is high-ly uncertain. To assist the scientific community inimproving the estimates of global productivity, NASAdeveloped the Coastal Zone Color Scanner (CZCS) andlaunched it aboard the Nimbus-G spacecraft in Oct.1978. This color sensor was primarily designed tomeasure or infer the photosynthetic pigments con-tained in phytoplankton residing within the upper-most water column or near-surface layer of the ocean.Estimates of the attenuation coefficient of downwell-ing irradiance as well as seston apparently can also beobtained using two-band ratio algorithms.2 Regard-ing planktonic pigments in general, only the principalphotosynthetic pigment, chlorophyll a (plus otherchlorophyll-like pigments such as phaeopigment a),has been inferred using algorithms applied to oceancolor upwelled spectral radiance.

Chlorophyll algorithms for application to remotelysensed upwelled spectral radiance have to date beendeveloped using shipboard truth data. These on-sta-tion point measurements are sometimes supplementedby underway fluorometric chlorophyll determinations.The amount and diversity of such truth data are then

Frank Hoge is with NASA Goddard Space Flight Center, WallopsFlight Facility, Wallops Island, Virginia 23337; R. N. Swift is withEG&G Washington Analytical Services Center, Pocomoke City,Maryland 21851.

Received 28 January 1986.

essentially limited by the speed of the surface vessel.Recently, airborne laser systems have demonstratedthat chlorophyll a and phycoerythrin fluorescencealong with laser-induced water Raman backscattercould be obtained both rapidly and at a high samplingdensity.3-6 Accordingly, at laser pulse rates readilyavailable today large numbers of independent mea-surements may be obtained over wide oceanic regions.Recently, oceanic upwelled spectral radiance measure-ment capabilities have also been added to an airbornelaser spectrofluorometer. 78 Voluminous laser-in-duced and solar-induced spectra synoptically gatheredby such an airborne active-passive instrument allowrather high precision studies of the ocean color radi-ance spectrum via computational algorithms. A typi-cal airborne flight mission produces 50,000 pair ofactive-passive observations.

The airborne laser-induced chlorophyll fluorescencefrom the water column has been shown to yield highcorrelation to the actual pigment concentration as de-rived by shipboard extractions and underway fluores-cence.3 9"10 These findings are in spite of known pho-toplankton fluorescence variability caused by ambientlight and nutrient changes."1 Thus, under the singularassumption that the airborne laser-induced chloro-phyll a fluorescence is a reliable indicator of the pig-ment concentration, it will be used as the sole source oftruth data for identification and evaluation of the pas-sive ocean color chlorophyll algorithm spectral regions.

In the initial portion of this paper the new active-passive correlation spectroscopy (APCS) technique isintroduced and described. The method is then ap-plied to airborne active-passive field data to illustrateits potential for the rapid search, identification, andevaluation of spectral bands or wavelength positionssuitable for use in ocean color algorithms for measur-

1 August 1986 / Vol. 25, No. 15 / APPLIED OPTICS 2571

Table I. Comparison of Active-Passive and Conventional Correlation Spectroascopy

A - P CORRELATION SPECTROSCOPY

TARGET ! Ocean Atmosphere

SPECIES Water Column Consttuentn . Trace Gases

INITIAL Constitutent-Driven SpectralDESIRED Variability Locations

QUANTITYFINAL Major and Minor Constituent Trace Gas Concentrations

Concentrations

FILTER Actively Measured Mask Gas, Interferogra.ELEMENT I (Simultaneously)

LOCATION Digitally Recorded During Stored in Instrument

i 1-2 > Experiment|

DYNAMICl Dynamic Constant

CONSTANT

PHYSICAL Spectral Reflectance Absorption and Emission(Includes Absorption

MEASUREMENT d Scatter and Emission;

ing chlorophyll. At this time only the three-band orSCA will be evaluated for photoplankton pigmentmeasurement. The APCS technique will also be simi-larly applied to determine spectral band combinationsfor the detection and measurement of the accessorypigment, phycoerythrin, since its laser-induced fluo-rescence is also measurable concurrently with the pas-sive ocean color spectrum. While not specifically ad-dressed herein, it is expected that the APCS techniquewill in future efforts be extended to standard two-bandratio algorithms, case II waters, other accessory pig-ments, Gelbstoff, water attenuation, etc. Researchersin other disciplines may also find the technique usefulfor land and atmospheric spectral pattern recognition.

II. Active-Passive Correlation Spectroscopy

A. Background

Conventional correlation spectroscopy has beenused in various configurations for '60 years.12 Indispersive, gas filter, and interferometer instruments aknown spectrum, gas, or interferogram is stored withinthe instrument for correlation with the atmospherictrace gas being observed in absorption or emission.Dedicated correlation instruments can possess one ormore possible advantages over standard spectrometersincluding high throughput, spectral simultaneity, datasimplicity, and high sensitivity or SNR.

Active-passive correlation spectroscopy describedherein differs from these previous applications in sev-eral distinct ways. First, it is applied to oceanic watersinstead of the atmosphere. Second, the fundamentalspectral reflectance, absorption, and emission charac-teristics of the ocean are not as narrow or well definedas those of the atmosphere. Third, since the spectralcharacter of the ocean is not highly defined, active orlaser-induced fluorescence emission spectroscopy is

used as known correlation information. A techniquewill be described for determining the spectral loca-tion(s) of the subtle variations in the passive oceancolor reflectance caused by water column constituents(phytoplankton pigments, dissolved organic material).In the interest of brevity active-passive correlationand conventional correlation spectroscopy are com-pared in Table I.

Relative to the APCS technique discussed here, thenearest equivalent remote sensing analog is the oil spilldetection method of O'Neil et al.1

3 They correlatedairborne laser-induced fluorescence spectra againstsimilar laboratory fluorescence emission spectra of thetarget oils overflown. Accordingly, their methodcould be labeled active airborne-laboratory correla-tion spectroscopy. Their active-only technique clear-ly allowed differentiation among a dye, two crude oils,and the general fluorescence background of ocean wa-ter.

Another airborne active-passive multispectral scan-ner (APMSS) instrument (besides the AOL) has beenbuilt and successfully flown.14 Active and passive ter-restrial images were obtained. This same APMSSinstrument was later used to make active-passive mea-surements of water depth.15 No computations of spec-tral correlation between the two data types were re-ported for either experiment.

B. Passive Ocean Color Spectrum

In Fig. 1 an idealized solar-stimulated ocean colorreflectance spectrum is presented for the case of highand low chlorophyll concentration. It is not distin-guished by numerous or prominent spectral features.The most obvious chlorophyll-induced characteristicsare (a) the slope change in the nominal 450-570-nmregion, (b) the spectral flattening in the 450-510-nmsection, and (c) the fluorescence emission in the 675-

2572 APPLIED OPTICS / Vol. 25, No. 15 / 1 August 1986

CONVENTIONAL CORRELATION SPECTROSCOPY

CHLOROPHYLL

IhJ

-1 \ \Fluorescence

Slope Changes

Curvature Changes \

400 460 560 550 600 650 70 750WAVELENGTH (nm)

Fig. 1. Idealized low-altitude oceanic spectral reflectance for low-and high-chlorophyll bearing waters. The principal differentialchlorophyll features are presently known to be the slope and curva-ture in the blue/blue-green region and the fluorescence emission in

the red.

695-nm segment. For remote sensing of chlorophyll(and phaeopigments) band ratios,2 curvature,7'8 16-8

and amplitudes of solar-induced spectral fluorescenceemission,1920'2 ' have all been used with varying degreesof success. The band ratio changes have been usedmost often and form the basis of the satellite CZCSchlorophyll biooptical algorithms.22 Increasingamounts of chlorophyll pigment also lead to spectralflattening. These curvature variations have recentlybeen used to infer quite accurately chlorophyll fromairborne platforms.7 '8 16-8 Only the fundamental in-tensity (and not slope or curvature) of the fluorescenceemission line has to date been used to infer chlorophyllconcentration. 8 "9 '2 0'2 1

C. Active or Laser-induced Ocean Color FluorescenceSpectrum

Figure 2(a) represents a typical airborne fluores-cence spectrum3 4 6 resulting from laser stimulation at532 nm. In contrast to passive ocean color spectra, it isdistinguished by rather narrow identifiable spectrallines. The laser, phycoerythrin, water Raman, andchlorophyll lines are identified by L, P, R, and C,respectively. This is the type of spectrum againstwhich all correlations herein are performed. The re-sults reported here will be confined only to correlationsusing the individual spectral lines of the chlorophylland phycoerythrin pigments. Future plans call for theaddition of a shorter wavelength laser transmitter toproduce fluorescence from other phytoplankton pho-topigments.

D. Active-Passive Correlation SpectroscopyMethodology

Figures 2(a) and (b) schematically illustrate pairedactive and passive measurements as obtained with theAirborne Oceanographic Lidar. For all practical pur-poses the spectral pair is taken at the same time tk andessentially within the same footprint. The active dataare processed only by normalization with the water

PASSIVEa)

U-

e)

a.-

Fig. 2. Schematic illustrations of (a) typical 532-nm laser-inducedfluorescence spectrum showing the laser backscatter L, phycoery-thrin fluorescence P, water Raman backscatter R, and chlorophyllfluorescence C; (b) idealized solar-stimulated ocean color reflec-tance spectrum S(X1); (c) water-Raman-normalized laser-inducedfluorescence spectrum F'(Xi); (d) so-called spectral curvature ob-tained by applying second difference operator to logSi(X); (e) repre-sentative chlorophyll spectral correlation function pc(Xj) obtainedby correlating normalized chlorophyll fluorescence F'(X\) with A2

logS(Xj) for all Xj.

Raman backscatter.4-69 23 24 Thus from the originallaser-induced fluorescence spectrum, F(Xi), the nor-malized spectrum F'(Xi) = F(Xi)/F(XR) is produced asillustrated in Fig. 2(c). The along-track or temporalchlorophyll profile F'(XC,tk) is illustrated within Fig.2(c), and actual profiles of field data will be presentedlater.

The passive ocean color spectra S(Xi) represented inFig. 2(b) can be used in an unprocessed form, butunfortunately the high variability of the entire spectralwaveform completely masks the desired subtle colorvariations impressed on the clear ocean water spec-trum by the water column constituents. The slope orfirst derivative spectrum dS(Xi)/dX could likewise beused in the correlation process, but it too containedvariability which we were not prepared to correct. (Asshall be pointed out later, the field data sets usedherein were not specifically gathered for this correla-tion study but were simply the best available at thistime.) The curvature spectrum [related to the secondderivative d2S(Xi)/dX 2] does, however, yield resultswhich are reasonably free of unwanted environmentalfactors.16-'8 Furthermore, it can be applied to uncali-


AC TIVE

brated and uncorrected passive ocean color data takenfrom relatively low aircraft altitudes. All data usedhere were obtained at 150-m flight altitude. Withthese considerations in mind we chose to process thepassive color data with a curvature algorithm. Thealgorithm chosen is essentially that of Campbell andEsaias,'8 Al, logS(Xi), where AJ( ) is a difference opera-tor defined by

Af(X) = + -- 2 (1)

If w = 30 nm and the operator is applied twice asrequired, Campbell and Esaias18 showed that

A2 logS(X,) = -logG(X), (2)

where

S2(X )

=S(X 1 + 30) S(X - 30)

Thus processing the S(Xi) spectra is performed bystarting with data from the bluest spectral band of theinstrument, S(Xi - 30), then selecting data from thespectral position 30 nm redder S(Xi) and another 30 nmto the red S(Xi + 30). These three spectral radiancesare used to compute G,,(Xi) via Eq. (3). Illustrated inFig. 2(d) is the -logGJXj) or A2 logS(Xj). The indexlabel is changed to j to identify the fact that the origi-nal 32 band spectra were first linearly interpolated to"create" additional bands having a width of 1 nminstead of 11.25 nm as actually provided in the passiveocean color subsystem (POCS) of the AOL. The Xj aresuccessively incremented by 1 nm, and the calculationis repeated until a complete curvature spectrum, suchas schematically illustrated in Fig. 2(d), is obtained.For the actual data set to be discussed herein, theprocedure is started at 443 nm and conducted 288times until the final curvature value is obtained fromthe raw radiances at 701,731, and 761 nm. The curva-ture spectrum G,,(Xj) is thus shorter than the spectralradiance S(Xi) by 60 nm (30 nm each on the blue andred ends). A curvature spectrum is produced fromeach raw spectrum S(Xi).

Finally, we wish to produce a correlation spectrumfor some desired water column constituent. If theconstituent is chlorophyll, scatter plots of -logG,(Xj)vs F'(X,) are generated (internal to the computer) forall 1 < j S 288 and for all tk in a flight line. Thesescatter plot data then yield a chlorophyll spectral cor-relation function p(Xj). This spectral correlationfunction (SCF) for chlorophyll is illustrated in Fig.2(e). A spectral correlation function for phycoery-thrin pp(Xj) can be similarly calculated by using thenormalized laser-induced fluorescence spectrum(evaluated at Xp) F'(Xp). Typically from 5000 to10,000 pairs of active and passive waveforms are ob-tained during the traversal of a flight line. The spec-tral correlation functions, p,(Xj) and pp(Xj), are usuallyproduced from data acquired along the entire flightline. Both the identification and evaluation of spec-tral intervals are provided by this spectral correlationfunction. Values above ±0.8 are selected for further

study. The effectiveness of an algorithm is then quali-tatively judged by visually comparing along track pro-files of the active with the passively derived pigment.

Finer details of the actual handling of active andpassive ocean color spectra are provided in Refs. 7 and8. A brief summary is provided here for the conve-nience of the reader. Initially, all the passive datawere filtered to remove any observations contaminat-ed with sun glint. During the experiment sun glintwas minimized by avoiding sampling during the period1 h on either side of local noon and by pointing the finalsteering mirror of the AOL 15° off-nadir and awayfrom the sun. During processing, spectra containingsun glint were recognized by keying on the passivesignal in the longest wavelength channel, the center ofwhich was located at -761 nm. On some of the initialpasses (acquired within 1 h of local noon), this filteringprocedure amounted to removing as much as 20% ofthe passive observations, while on some of the laterpasses as few as 3% of the passive spectra were rejectedduring processing. The remaining active and passiveobservations were then subjected to a ten-point simpleaverage. This averaging was done to reduce the vol-ume of data and to minimize any effect on the resultsfrom shot-to-shot variability associated with the laser-induced fluorescence measurements. The spatial in-terval over which the data were averaged amounts to adistance of -125 m at the -100-m/s aircraft velocity.Along a typical flight line, some 5000 to 10,000 laser-induced fluorescence waveforms were recorded fromwhich fluorescence emission peaks corresponding tochlorophyll and phycoerythrin can be extracted. Anequal number of passive upwelled ocean color radiancespectra were obtained simultaneously. After filteringand averaging, some 500-1000 paired, active and pas-sive, observations remain for analysis.

E. Spectral Curvature Algorithm Forms

In the previous section all linear correlation scatterplots were performed as -logGJ(Xj) vs F'(Xi), where icorresponds to chlorophyll X, or phycoerythrin Xp.Since the laser-induced chlorophyll fluorescence hasbeen shown to be linearly related to the chlorophyllconcentration3 9 C, an algorithm of the form

C = A - B ogG,,(Xj) (4)

is thus inferred. This form has been found to producemore consistent results in our work than the form byCampbell and Esaias,18

logC = a - b ogG,,(X.) (5)

or the form by Grew and Mayo,16"17

logeC = a - flGJ(Xm), (6)

where Xm is 490 nm, and A, B, a, b, and a, are thelinear regression coefficients of all the respectiveforms. Our choice of algorithm in Eq. (4) does notsignificantly alter the results described since all threeforms of the algorithm produce quite good correlationwith actual chlorophyll measurements.7 8 16-8 Of


course, G,,(X) is common to all the algorithms in Eqs.(4), (5), and (6).

Campbell and Esaias18 found that the curvature al-gorithm effectively eliminates variations due tochanges in incident irradiance while at the same timeenhancing spectral features of the water medium.Utilizing a model based on earlier work by Smith andBaker,25 Campbell and Esaias18 were able to show thatthe irradiance reflectance of ocean water exhibits adistinctive curvature spectrum with a large negativecurvature at 490 nm. As chlorophyll-like pigments areadded to the water this negative curvature monotoni-cally approaches zero. This latter feature is the fun-damental physical basis for the high sensitivity of thealgorithm to chlorophyll in water.

As indicated previously, the curvature algorithm isessentially a difference operator applied twice to thelogarithm of the radiance in the middle spectral band.At least several important consequences result: (1)since this difference is extremely small, relative to thelarge radiance values involved, the leveraging imposedby the form of the algorithm necessitates high-sensorprecision and stability; (2) the calibration of the spec-tral radiance of the sensor is not a requirement al-though mission-to-mission stability is essential if theconstants in Eq. (4) are to become anchored for real-time use; and (3) the algorithm may be applied in realtime to ocean color spectra obtained at relatively lowaltitudes without serious consequences from atmo-spherically related effects. (Of course, at high-alti-tude, additive, independently varying, atmosphericpath radiance must be removed before further process-ing with the three-band algorithm.)

F. Validity of the Airborne APCS Technique

For defining passive ocean color algorithm spectralregions, the validity of the APCS technique rests solelyon the validity of the laser data as it relates to actualchlorophyll concentration. If the laser system doesnot yield fluorescence measurements which are some-how directly relatable to the actual extracted photo-pigments, the APCS technique will fail.

For the chlorophyll pigment, ship extraction/air-craft fluorescence comparison studies during the pastseveral years indicate that a linear relationship existsbetween the laser-induced chlorophyll fluorescencesignal and corresponding pigment extraction measure-ments determined onboard cooperating research ves-sels.39 10 Additionally, in data sets acquired duringthe shelf edge exchange processes (SEEP) experiment(and utilized for the bulk of the analysis for this paper)corroborating investigators from the University ofSouth Florida and Brookhaven National Laboratoryhave compared the airborne fluorescence measure-ments with surface measurements. For all four NASAairborne oceanographic lidar (AOL) SEEP missionsthe results26 indicated a linear regression correlationcoefficient of 0.93 between the aircraft and ship mea-surements of chlorophyll in the near-surface layer.

No comparisons have yet been made between air-borne phycoerythrin fluorescence and shipboard pig-

ment extractions. Phycoerythrin extraction and mea-surement methods have been reported.2 728 However,phycoerythrin measurements are not yet routinelyconducted onboard research vessels. Moreover, thelack of a strong cw light source in the -530-nm regionprevents monitoring the phycoerythrin fluorescencesignal from a standard flow-through fluorometer.

Inhibition of chlorophyll fluorescence by high-solarirradiance levels" could possibly invalidate the laserfluorescence results. Although the effects of solarphotoinhibition on airborne laser-induced chlorophyllfluorescence have been reported,2 9 we have never seenthe effect in our field data. Additionally, lidar receiv-er linearity must be carefully measured and operatingparameters established to avoid misinterpretation ofinstrument responses as photoinhibition effects. Theoperational settings of the AOL were established byground tests and validated further by field experi-ments.7 ' 8

111. Experiment Descriptions

The airborne data utilized to demonstrate the ac-tive-passive technique for passive ocean color algo-rithm band location were acquired primarily during aflight mission conducted in the New York Bight during1984. Some earlier data gathered during the NationalScience Foundation's warm core ring (WCR) field ex-periments in 1982 were additionally used to assess thepotential extendibility of the initial results obtainedduring the SEEP experiments for direct application inother water masses. Results from the above experi-ments were specifically chosen because (1) simulta-neous active and passive ocean color data were avail-able and (2) sufficient contrast between chlorophylland phycoerythrin patches existed to allow evaluationof the effectiveness of the technique for two differentpigments.

The Department of Energy (DOE) sponsored SEEPinvestigations along the east coast area of the U.S. aredesigned to assess the assimilative capacity of the Con-tinental Shelf to absorb energy by-products intro-duced into the near-shore ocean environment fromcoastal communities and marine activities such as en-ergy production plants and offshore oil operations.This capacity depends to a great extent on rates ofremoval by sinks in the marine ecosystem. According-ly, the distribution and abundance of marine phyto-plankton over the Shelf and in the adjacent slope watermasses are of fundamental importance in this process.The initial SEEP studies, conducted between Feb. andMay 1984, are part of a longer range program for themid-Atlantic region which is planned to continue every2 years over the next decade. Other institutions par-ticipating in these field studies are Brookhaven Na-tional Laboratory (BNL), Yale University, LaMontDoherty Geophysical Observatory, University ofSouth Florida, and Woods H-ole Oceanographic Insti-tution.

Oceanographic observations within the SEEP testarea were acquired from ship (RV Endeavor), mooredbuoy, satellite, and aircraft platforms. As previously


74 73 72 71 70LONGITUDE (Dog.)

Fig. 3. New York Bight location of the SEEP field test site wheremost of the airborne active-passive data were obtained. Line Hhad

the highest pigment variability and was selected for this study.

mentioned, high correlation has been obtained in com-parisons conducted between the laser-induced chloro-phyll fluorescence measurements and near-surfacechlorophyll data obtained from the research vessel.26

Similar agreement has been found between airbornesurface temperature measurements and surface layertemperature measurements acquired from the re-search vessel. The airborne ocean surface tempera-ture measurements were obtained with a Barnes PRT-5 IR radiometer and recorded by the lidar dataacquisition system. The Eulerean measurements ofchlorophyll fluorescence recovered from recording flu-orometers attached to the moored buoy lines haveyielded interesting results when compared to the near-ly synoptic wide-area airborne laser-induced fluores-cence values.2 6

Airborne lidar and passive ocean color data were alsoobtained during the 1982 warm core rings56 experi-ments. These warm core ring experiments were con-ducted to provide increased understanding of the bio-logical, physical, and chemical processes associatedwith warm core rings as they interact with surroundingshelf, slope, and Gulf Stream water masses.

A total of four missions was flown with the AirborneOceanographic Lidar (AOL)3-6 and its integral passiveocean color subsystem (POCS)7,8 during the SEEPexperiments. For detailed analysis, we focused ourattention primarily on the SEEP mission flown 2 Apr.This mission was chosen because calibrated ocean col-or data acquired with the NASA Multichannel OceanColor Sensor (MOCS)' 6"17 were also available for com-parison and analysis. Active and passive ocean colorspectra obtained with the AOL during the 8 Apr. mis-sion were also analyzed to determine general agree-ment with the 2 Apr. results. However, no corrobora-tive data from the MOCS were available from thelatter mission for comparison. In addition, both mis-sions were flown following moderately strong westerlywind events which resulted in a general increase inboth the level and spatial variability of chlorophyll a

concentration within the surface layer, especially inthe inner portion of the SEEP study site.7 8

The location of flight lines occupied during the 2Apr. mission is shown in Fig. 3. Only the uppermostline is provided with a letter designation H, which willbe referred to during subsequent discussions. (Thelocations of the 8 Apr. flight lines are virtually identicalto those in Fig. 3 and accordingly will not be shown.)The perimeter of the flight lines essentially defines theSEEP study site. The flight lines were designed inconjunction with BNL to complement their samplingstrategy with the moored fluorometer arrays. Theaircraft data provided periodic synoptic assessment ofthe distribution of chlorophyll a and ocean surfacetemperature between the buoys, especially along-shelf, which was expected to be the dominant directionof surface current flow and, therefore, the net trajec-tory for entrained particulate matter.

IV. Application of the Band Location Technique toAirborne Field Data

Line H from the 2 Apr. flight was chosen from theSEEP missions for detailed analysis for this paper.This line was primarily selected because the chloro-phyll and phycoerythrin concentrations were found tobe both elevated and variable. Spatial variability anda degree of noncoherence in the relative concentrationsof the two photopigments were considered essential fordevelopment and demonstration of the active-passivetechnique for ocean color algorithm spectral regionlocation development. As will become evident in thesucceeding discussion, both chlorophyll and phycoery-thrin have considerable effect over wide sections of thepassive ocean color spectrum, thus the spatial variabil-ity and noncoherence of the photopigments permit anypassive chlorophyll or phycoerythrin algorithm to betested for functionality throughout all regions of aflight line regardless of the relative loading of the otherphotopigment.

Profiles of laser-induced chlorophyll and phycoery-thrin fluorescence are plotted in Fig. 4(a) as a functionof distance along the flight line H. As mentioned inthe previous section, the data, in Fig. 4 and in subse-quent figures in this section have been subjected to aten-point simple average. Both photopigments havebeen normalized with the water Raman backscattersignal to remove effects on the fluorescence signal lev-els due to variability of attenuation properties in thenear-surface layer along the flight track. (A discus-sion of the normalization technique using the waterRaman backscatter signal can be found in Refs. 4,9,23,and 24.) A profile of the water Raman backscattersignal is plotted in Fig. 4(b). Profiles of ocean surfacetemperature, and the passive ocean color signal at 746nm, are plotted, respectively, in Figs. 4(c) and (d).These latter profiles have been included to provide thereader with a more complete picture of the attendanttemperature distribution and relative ambient lightlevels along the flight line.

The most prominent feature of the laser-inducedfluorescence profiles shown in Fig. 4(a) is the large


aWN!4j

zz44

0

:4

KK

2

z

I4r

r

in-=

0..

co;D

0

0

8-

7-

0 10 20 30 40 50 60 70 80 90 100 110 120 130

DISTANCE (KM)

Fig. 4. Along track profiles of line H showing (a) laser-inducedchlorophyll a and phycoerythrin fluorescence as normalized by thewater Raman backscatter. The low correlation of chlorophyll andphycoerythrin is easily identified in the first 40-km section of theline. (b) Laser-induced water Raman backscatter (note the slightdepression in the first 40-km segment of the line). (c) Sea surfacetemperature along line H. (d) Radiance received in the passivechannel centered at 746 nm. The spectral bandwidth is 11.25 nm.

(This figure is reproduced from Hoge et al.8 )

chlorophyll patch located near the eastern end of lineH (between 10 and 40 km). Also worth noting are thegeneral coherence between the two photopigment pro-files over most of the remaining flight line and thepresence of a number of 1-3-km patches of both photo-pigments. Sufficient detail existed in this flight line(as well as in other flight lines of this SEEP mission) toallow rigorous assessment and evaluation of variouscombinations of spectral bands for passively determin-ing photopigment concentration.

A. Passive Detection of Chlorophyll and Phycoerythrin

Figure 5 shows the results obtained on flight line Hwhen the curvature values -logG,(Xj) (computed withthe symmetrical; three-band algorithm as described inthe preceding section) were linearly regressed for eachXj against laser-induced chlorophyll [Fig. 5(a)] andphycoerythrin [Fig. 5(b)] fluorescence measurements.In both figures, the linear correlation coefficient PC orpp has been plotted as a function of the wavelength ofthe center band of the three-band curvature algorithm

a) It

0.5-

CHLOROPHYLL AOL- POCSLINE H/SEEP C

0.0-

-0.5-

-1.0 . . . . . .460 480 500 520 540 560 580 600 620 640 660 680 700

WAVELENGTH j(nm)

PHYCOERYTHRI LIDNE H /SEEP C

0. 0 t5- 0.0__.A_____AA^___AAjAA

b)

460 480 500 520 540 560 580 600 620WAVELENGTH kj(nm)

640 660 680 700

Fig. 5. (a) Chlorophyll spectral correlation function p(Xj). Thisfunction yields spectral regions at 460-510, 645-660, and 680-695nm where the upwelled spectral radiance variability of the ocean isdirectly related to chlorophyll pigments. (b) Phycoerythrin spec-tral correlation p,(Xj). This function gives narrow spectral regionsat 600 and -670 where the upwelled spectral radiance variability isstrongly related to phycoerythrin pigment fluorescence. By com-paring (a) and (b) it is evident that chlorophyll and phycoerythrinocean color spectral variability contributions are inversely related attheir principal correlation wavelengths and that the ocean color

spectrum is essentially invariant to these pigments at 630 nm.

[Eq. (4)]. Notice that in both figures there are sectionsof the spectral correlation function where the correla-tion is high (positive and negative) as well as sectionswhere the computed curvature -logG,(Xj) is apparent-ly uncorrelated with the photopigment fluorescenceF'(Xi).

In Fig. 5(a) there is a broad area of high correlationbeginning at -460 nm and extending to -510 nm. It isthus not surprising that Grew16"17 and Campbell andEsaias18 obtained good agreement between spectralcurvature measurements and laser-induced chloro-phyll fluorescence (and surface truthing chlorophyllmeasurements) using a three-band algorithm centeredat 490 nm. In fact these results suggest that locationof the three-band curvature algorithm is not critical inthe nominal 460-510-nm region. In this blue segmentthe observed reduction in curvature and resulting highcorrelation with chlorophyll fluorescence are probablydriven strongly by phytoplankton pigment absorption.

Smaller sections of high correlation are found in Fig.5(a) in the 645-660- and 680-695-nm regions. Spec-


IBD0

of

ECHLOROPHYLL

RYTHRIN

I

0 I \1

°l

V

I �

tral regions where the correlation is strongly negativemay be interpreted as a region where the curvature ofthe spectrum is reduced with increased chlorophyllloading as is the case for the three-band curvaturealgorithm centered at 490 nm as discussed by Camp-bell and Esaias.18 Conversely, strong positive correla-tion is an indication of increased spectral curvaturewith higher chlorophyll concentration. The high neg-ative correlation in the 645-660-nm segment isthought to be attributable to absorption by chloro-phyll pigments. 3 0

The positive correlation seen in the 680-695-nmregion may be due in part to observed solar-inducedchlorophyll fluorescence. This high correlation is inagreement with other researchers'9 -2' who showed thatin spite of the lower signals in the red region as muchinformation is present there as in the blue-green re-gion. Increased curvature could also result from addi-tional reflectance in a particular spectral band(s).Moreover, we recognize that strong negative (or posi-tive) curvature in one band will necessarily influencethe apparent curvature in adjacent regions as the mid-point of the three-band algorithm progressively ap-proaches the actual spectral feature. Thus, at thispoint in our investigations, we cannot absolutely deter-mine the physical basis for the apparent curvaturevariations and resulting high correlations in several ofthe spectral regions. Finally, the results in Fig. 5(a)suggest that improved chlorophyll measurementmight result from combining (or even contrasting)bands from the three high correlation regions. How-ever, for widely separated band segments, the watercolumn sensing depth will probably be different.

In Fig. 5(b) strong positive correlation is found be-tween -logGJ(Xj) and laser-induced phycoerythrinfluorescence when the middle band of the algorithmwas centered around 600 nm. At this time a physicalbasis for this 600-nm peak is difficult to identify sincethe in vivo phycoerythrin fluorescence occurs at 580-585 nm,31 and its absorption (excitation) maximum isfound in the 530-nm region.27 A second less pro-nounced area exhibiting high negative correlation canbe seen in the band centered near 670 nm. Its physicalbasis is also puzzling, but like the companion 600-nmpeak, it may be due to covariability with other phyto-plankton pigments whose functions are coupled tothose of phycoerythrin.

Comparison between the two plots of Figs. 5(a) and(b) reveals several other potentially important aspects.First, the correlation with phycoerythrin fluorescencediminishes to low values in the 460-510-nm spectralregion where there is high correlation with chlorophyllfluorescence. Low correlation (pp < 0.5) with phycoer-ythrin fluorescence is also found in the 645-660- and680-695-nm regions where the correlation with chloro-phyll fluorescence is also high. Conversely, in thespectral region near 600 nm, where the correlation withphycoerythrin fluorescence is strongest, there is abroad spectral region where the correlation with chlo-rophyll fluorescence is at a minimum. Second, andperhaps more important to the sensor band and algo-

rithm selection process, the plots of the spectral corre-lation functions in Fig. 5 provide a clear indication ofthe profound influence of both photopigments at sev-eral specific but often widely separated spectral re-gions. This aspect is an especially critical consider-ation for satellite-borne passive ocean color sensorswhere a spectral band suitable for removing atmo-spheric effects must be selected and where budget andlogistical constraints limit the number of availablechannels. Current techniques for processing CZCSimagery utilize the spectral channel centered at 670nm to correct atmospheric effects on the bluer chan-nels involved in the in-water algorithm for determin-ing chlorophyll concentration. Third, the 630-nm re-gion of the ocean color spectrum appears to berelatively invariant to both chlorophyll and phycoery-thrin pigments since the spectral correlation functionshave very low values there. These type regions maythus be useful for atmospheric correction.

The presence of three high-correlation peaks alsostrongly suggests that significant information contentresides within the ocean color spectrum and that usefulinformation can be found outside the most obviousslope, curvature, and fluorescence emission regions ofchlorophyll (refer again to Fig. 1). While these specif-ic findings are new, this general information-contentfinding is in agreement with Lin et al.'9 They foundchlorophyll information within numerous regions ofthe ocean color spectrum using eigenvector analysis ofairborne data together with ship truth. The APCStechnique appears to be easier to utilize than eigenvec-tor techniques. Furthermore, the correlation methodvividly shows the participating spectral bands withouthand-selecting the study spectral regions. Finally, thealgorithm can be specifically applied to bands of inter-est and thus is not limited to eigenvectors whose physi-cal meaning is somewhat obscure.

Computed -logGJ(690) and -logG,,(600) valuesfrom the three-band algorithm, which were found fromFig. 5 to have the highest correlation with chlorophylland phycoerythrin, were next plotted in profile form inFig. 6. These plots allow visual assessment of thedetailed characteristics of the -logG,,(Xj) values bydirect comparison with the respective photopigmentfluorescence signals. The estimated chlorophyll con-centration computed from the curvature algorithmwith the middle band centered at 690 nm is plotted inFig. 6(a) as a function of distance along flight line Htogether with laser-induced chlorophyll fluorescenceoriginally shown in Fig. 4(a). For convenience of han-dling and plotting, both the active and passive concen-tration estimates have been autoscaled. As expectedfrom the high-correlation coefficient found in Fig. 5(a),the two profiles show good agreement over the entireflight line even down to relatively small scale features.Similarly, a profile of -logG,,(600) is plotted in Fig.6(b) together with phycoerythrin fluorescence [origi-nally shown in Fig. 4(a)]. Again good agreement isapparent over the entire flight line. Note particularlyhow well the phycoerythrin algorithm performs in thehigh-chlorophyll region (10-40 km).


2-

._X2

I-0

-J

U

O ..... ...... .......\.\ \\4......... ....... \4 \ \ ....... \\\l\ 4\\\\\\ \!!i0 20 40 60 80 1OO J20

DISTANCE (km)

zI

a:

I)

20 40 60 ' 80DISTANCE (km)

100

Fig. 6. Passively derived (a) chlorophyll at 690 nm and (b) phycoer-ythrin at 600 nm using a curvature algorithm of the form given in Eq.(4). The water Raman normalized laser-induced chlorophyll andphycoerythrin fluorescence are, respectively, plotted in each sectionfor comparison purposes. The data were obtained along line H (Fig.3) using the active and passive capabilities of the airborne oceano-

graphic lidar.

This initial application of our technique to selectpassive spectral bands utilizing data acquired with theAOL operated in a dual active-passive mode was veryencouraging. In short, these results showed that

(1) several regions of high correlation were deter-mined for chlorophyll measurement;

(2) one region was identified which showed particu-larly strong correlation with phycoerythrin fluores-cence;

(3) spectral areas of high correlation in the chloro-phyll regression plot correspond to the lower bands ofcorrelation in the phycoerythrin regression plot andvice versa; and

(4) a spectral segment which is relatively invariantto both chlorophyll and phycoerythrin pigments wasobserved.

Although agreement between the results from theband selection algorithm and the photopigment fluo-rescence measurements appears to provide consider-able evidence of the potential utility of the techniquefor passive ocean color band selection, verification ofthe curvature results with a second independent pas-sive sensor would assure that there was no instrumentrelated artifacts associated with the results. We,therefore, undertook examination of some data setscontaining concurrent passive ocean color measure-ments acquired with the NASA MOCS during theSEEP experiment and WCR studies (prior to imple-mentation of the passive ocean color subsystem intothe AOL).

Spectra collected with the MOCS were processedthrough the band selection algorithm using essentiallythe same procedure described for the AOL passivespectra. The resulting -logG,,(Xj) values were thenregressed independently against the AOL active laser-induced photopigment fluorescence measurements.The resulting spectral correlation function plots (notshown) produced quite similar results, although thepeak correlation coefficients were slightly lower, espe-cially for phycoerythrin estimation. Slightly lowerpeak regression coefficients would be expected sincethe MOCS passive and AOL active observations werenot obtained from the same exact footprints and be-cause of the differences in methods of real-time datacapture and handling between the two sensors. Forexample, in its present configuration, the MOCS isessentially locked in a nadir viewing position and thuscannot be pointed away from the solar glint pattern.Moreover, the MOCS integrates approximately fifteenocean color spectra per second from near-nadir pixels.Accordingly, individual observations containing sunglint could not be effectively filtered from surroundingobservations which are glint-free.

Profiles of estimated chlorophyll and phycoerythrinconcentration developed from MOCS ocean colorspectra using the three-band curvature algorithm areplotted in Fig. 7. Respective estimates of each of thephotopigments computed from the AOL passive dataare included for comparison purposes. The computed-logG>,,(490) values were selected for plotting the pro-files of estimated chlorophyll in Fig. 7(a) rather thanthe -logGj(690) values previously shown in Fig. 6(a).This change in band selection was necessary becausespectral coverage of the MOCS extends only to .700nm, thus precluding accurate computation of a-logG(Xj) value at 695 nm. Moreover, the MOCSsignal levels were too low in the red spectral region toyield satisfactory comparative results even at 660 nm.Other potentially useful passes were flown earlier inthe mission in conditions of higher solar illumination.The presence of sun glint during these earlier passes(>20% on initial flight lines compared with 3% on thefinal line H) probably accounts for much of the degra-dation in the overall agreement between the two sen-sors.

Profiles of estimated phycoerythrin concentrationare shown plotted in Fig. 7(b) using -logG,,(Xj) valueslocated at 595 and 600 nm for the MOCS and AOL/POCS, respectively. Although general agreement be-tween the two phycoerythrin concentration profiles inFig. 7(b) is reasonable over much of the flight line, thelevel of agreement is certainly below that found for thechlorophyll profiles in Fig.7(a). This is especially thecase in the initial 45 km of the flight line where consid-erable influence from chlorophyll appears to be exhib-ited in the MOCS phycoerythrin profile when it iscompared with chlorophyll profiles shown in Figs. 6(a)or 7(a). Note particularly that the features locatedbetween 15 and 35 km along the flight line in Fig. 7(b)appear to be noncoherent. The reason for the loweragreement of the passive MOCS phycoerythrin profile


ACTIVE CHLOROPHYLL AOL-POCSm PASSIVE CHLOROPHYLL LINE H /SEEP C

IH 10~Aj

EM ACTIVE PHYCOERYTHRIN AOL-POCS= PASSIVE PHYCOERYTHRIN LINE H/SEEP C

2

K \~\\KY

-

a)

.J

a.0

0-J=)

b)

0 20 40 60 80 l00 120DISTANCE (km)

Fig. 7. Passively derived (a) chlorophyll and (b) phycoerythrinusing an algorithm of the form in Eq. (4). The data were obtainedwith the MOCS and the AOL-POCS. The chlorophyll profile wasgenerated using a curvature algorithm centered at 490 nm for bothsensors. The phycoerythrin profile was obtained from curvaturealgorithms centered at 595 nm (MOCS) and 600 nm (AOL-POCS).

is unknown at this time. It may be due in part todifferences in the spectral bandwidth of the sensorchannels or to the lower sensitivity of the MOCS in theredder spectral region. The bandwidth of the MOCSchannels is -15 nm, while the AOL spectrometer chan-nels are 11.25 nm in width. The influence of chloro-phyll in the MOCS spectra in the vicinity of 600 nmand the differing channel bandwidth may also attri-bute to the slight difference in the wavelengths where-logG,,(Xj) values from the two sensors were found tohave the highest correlation with laser-induced phyco-erythrin fluorescence.

The potential utility of the technique for locatingpassive ocean color bands appears successful and wor-thy of continued development. The actual chloro-phyll and phycoerythrin results (i.e., the algorithm andbands themselves), however, require further study.The potential of these findings would be considerableif the results were found to be reasonably universal,i.e., if similar results are obtained at other oceaniclocations and under differing environmental condi-tions. This is especially the case for the passive phyco-erythrin algorithm since no alternative passive methodfor estimating this pigment has been previously pub-lished. To validate the technique further and to po-tentially establish more of a basis for universality forthe particular bands thus far identified for determin-ing photopigment concentration, flight lines A and G(not shown) from the same SEEP mission were alsoexamined in the same way. Qualitatively, the results

were found to be essentially the same. The spectralregions found to produce the highest correlation withthe respective photopigments were consistent with thefindings for flight line H. Quantitatively, however,the values of the highest correlation coefficients werefound to be somewhat lower. The remaining threemissions flown in connection with the SEEP experi-ments were not as suitable for performing this type ofanalysis. During the first two missions which wereflown in early March, the concentration levels andspatial variability of chlorophyll and phycoerythrinwere considerably lower, thus yielding less consistentresults when processed with a linear regression pro-gram. The 8 Apr. mission was flown under overcastconditions where sky contributions to the passiveocean color signal are higher. Nonetheless, the flightlines from the 8 Apr. mission did show results whichwere in good general agreement with those from 2 Apr.However, the peak correlation coefficient values weresomewhat lower.

Finally, we evaluated a data set from the 1982 WCRexperiment which contained both active and passiveocean color data. Since the WCR experiments predat-ed the relatively recent addition of the passive subsys-tem to the AOL, we utilized passive ocean color obser-vations which were acquired with the MOCS for directcomparison with simultaneous AOL laser-inducedphotopigment fluorescence measurements. TheWCR data set was regarded as particularly valuablesince it contained observations from several new watermasses including the Atlantic outer shelf, slope, GulfStream, and Sargasso Sea water types and since it wasacquired nearly 2 months later in the year. Thus thereexisted a reasonable potential for the presence of en-tirely different suites of phytoplankton.

The flight lines of the 24 June WCR mission werearranged to cross the ring at various angles with eachline passing through the ring center. The generallevels of photopigment concentration were relativelylow. Moreover, the highest levels of chlorophyll (1.5yg/liter maximum) were found near the center of thering while pockets of higher phycoerythrin concentra-tion were found in the ring boundary region, bothnorth and south of the ring. There was little spatialvariability of phycoerythrin levels in the east-westdirection. As a result, only flight line 6 (flown fromnorth to south) was found to contain sufficient con-trast between the two photopigments to allow evalua-tion of the bands identified and selected using subse-quent SEEP data.

Figure 8(a) shows a comparison between laser-in-duced chlorophyll fluorescence and passive (MOCS)chlorophyll values as a function of distance along theflight line. Similarly, Fig. 8(b) shows a comparisonbetween laser-induced phycoerythrin fluorescenceand estimates of the relative level of the phycoerythrinconcentration using passive ocean color spectra fromthe MOCS. The passive ocean color spectra from theMOCS were processed through the same three-bandalgorithms previously described for the data presentedin Fig. 7. The -10-km gap appearing in both the


a)

b)

-' 6 ' 20 ' 40 ' 60 ' 80 ' 0o 120 ' 4oDISTANCE (km)

Fig. 8. Passively derived (a) chlorophyll at 490 nm and (b) phycoer-ythrin at 600 nm using an algorithm of the form in Eq. (4). The datawere obtained with the MOCS during a warm core ring mission in1982. The AOL water-Raman normalized laser-induced chloro-phyll and phycoerythrin fluorescence are, respectively, plotted

(shaded) in each section for comparison purposes.

MOCS chlorophyll and phycoerythrin profiles begin-ning near the 105-km point in the flight line resultedwhen the data from the previous portion of the flightline were dumped from memory to tape.

The agreement between the profiles of chlorophyllconcentration in Fig. 8(a) is in much lower agreementthan seen in the results from the later SEEP missionsor from results presented previously by Grew'6"17 andCampbell and Esaias.18 Although both profiles ap-pear to exhibit similar trends over much of the flightline, an apparent scale difference exists between the 0-80-km section and the remaining portion of the flightline. In addition, there is a considerable offset be-tween the peak area of the observed chlorophyll patchseen between 80 and 105 km. By contrast, the generallevel of coherence between the laser-induced phycoer-ythrin fluorescence profile and the phycoerythrin con-centration profile estimated from the MOCS oceancolor data appears to be in quite reasonable agreementover much of the flight line. It is also very evident thatthe large patch of chlorophyll located between 80 and105 km [Fig. 8(a)] does not appear to influence theMOCS phycoerythrin profile which remains essential-ly parallel to the laser-induced phycoerythrin profilein that area of the flight line.

V. Summary, Discussion, and Conclusions

We have described in detail a new active-passivecorrelation spectroscopy technique which allows re-gions of ocean color spectral reflectance variability tobe directly related to the presence of specific fluores-cent waterborne constituents. Subtle variations inthe solar-induced ocean color reflectance spectrum areshown to be vividly revealed by correlating spectralcurvature changes with laser-induced fluorescence ofchlorophyll and/or phycoerythrin. The three-bandcurvature correlations against chlorophyll fluores-cence are sequentially executed in 1-nm intervalsacross the ocean color spectrum to generate a so-calledspectral correlation function. Correlation values ex-ceeding ±0.8 then signify important spectral regions ofocean color covariability with the known fluorescentconstituent(s). In turn, the identified color spectralsegments are regarded as the most optimum availablefor placement of passive sensor bands and resultingapplication of their computational algorithms. It hasalso been shown that the described method differsconsiderably from conventional correlation spectros-copy in that the known correlation data (laser) areactually obtained simultaneously with the airbornepassive ocean color spectra. While only applied here-in to ocean color spectra, the technique is also expectedto have important application to remotely sensed landand/or atmospheric radiances.

The recent addition of a passive ocean color subsys-tem into the AOL has allowed the acquisition of activeand passive ocean color measurements from essential-ly the same footprint. The relative mobility affordedby the aircraft platform, coupled with the 6.25-ppslaser transmitter, provides a relatively high samplingdensity of paired observations over wide areas anddiffering water types. This capability has in turn fa-cilitated development of a technique for identificationand verification of algorithm spectral regions for re-motely determining chlorophyll concentration. Thistechnique may also be applicable for the developmentof algorithms for passively measuring other phyto-plankton photopigments as well as other waterborneconstituents that can be sensed with an active lidarsystem.

The approach has been to select initially a form forthe algorithm. A number of forms are possible, and inthis initial investigation we selected a three-band cur-vature algorithm with sidebands that are separatedfrom the center band wavelength by 30 nm. Althoughthis choice was in part arbitrary, it was based on previ-ous success demonstrated by an algorithm of this type.This three-band algorithm is convoluted with (or ap-plied stepwise in 1-nm intervals to) the upwelled radi-ance spectrum. The logarithm of the resulting spec-tral curvature values are then linearly regressedagainst the water-Raman-normalized laser-inducedchlorophyll (or phycoerythrin) fluorescence measure-ments obtained concurrently during the flight mission.The spectral correlation functions thus generated arethen examined to reveal spectral regions where accept-able correlation between the -logG,(Xj) values and the


laser-induced fluorescence values has been achieved.Useful spectral regions more narrow than -15 nm werenot expected since the AOL-POCS instrument band-width is only 11.25 nm. Finally, the -logG,,(Xj) valuesshowing the highest correlation are graphed in a profileform along with the fluorescence measurements to al-low more detailed visual examination and to revealadditional characteristics of the resulting passive con-stitutent estimation.

The initial application of the technique to data ob-tained in the New York Bight during the 1984 SEEPexperiment was extremely successful. In particular,the processed passive ocean color data obtained onflight line H (2 Apr. 1984), which exhibited relativelyhigh chlorophyll concentration variability (0.5-6.0,ug/liter) and areas of noncoherence between chlorophylland phycoerythrin fluorescence levels, resulted in re-gression coefficients >0.95 and revealed few areas ofinconsistency with the concurrent chlorophyll andphycoerythrin laser-induced fluorescence measure-ments. These results were essentially verified or vali-dated by processing spectra obtained by an indepen-dent passive ocean color instrument (MOCS) alsoflown simultaneously onboard the NASA P-3A air-craft.

Attempts to apply the same techniques on otherflight lines flown during the same mission were alsosuccessful, although somewhat lower correlation val-ues were obtained, potentially due to the greateramounts of sun glint observed on these other passes.Similar results were obtained when the technique wasapplied to data sets obtained during a subsequentflight which occupied the same flight lines -1 weeklater. In an attempt to ascertain the potential univer-sality of the spectral bands identified during the SEEPexperiment, passive ocean color data obtained with theMOCS on a mission flown 2 years earlier in connectionwith the WCR studies were also processed through thethree-band curvature algorithm. The resulting pro-files of estimated chlorophyll and phycoerythrin wereshown to be in reasonable agreement with respectivelaser-induced fluorescence measurements of thoseconstituents.

Although the correlation between the -logG,,(Xj)values computed from the passive data (using the spec-tral bands determined to be most optimal) producedresults which varied from highly correlated to moremoderate levels of correlation, the band location tech-nique itself appears to have been satisfactorily demon-strated. Both the placement of sidebands in thethree-band curvature algorithm relative to the centerband and the form of the algorithm itself were notoptimized during correlation processing. Numerousseparations of the sidebands are possible, and the 30-nm symmetric sideband arrangement may not yieldthe most optimum results. Moreover, other formsneed to be examined. These include in-water two-band ratio algorithms such as those currently utilizedin processing CZCS imagery.

Various environmental factors may have influencedthe results. Cloud cover and ambient light levels were

not routinely monitored on the initial missions flownwith the AOL passive ocean color subsystem. Recentmodifications to the AOL will ensure that these pa-rameters will be continuously monitored. Questionsneed to be resolved regarding the effects on these re-sults of other water types, high sediment, and/or dis-solved organic material loading (Gelbstoff), andchanges in phytoplankton assemblages. For example,different phytoplankton species could contain a con-siderably different ratio of photopigments than wereencountered along line H, and this in turn could affectthe spectral location of the bands and degree of corre-lation achievable with the form of a particular algo-rithm.

We are, however, somewhat limited in the extent towhich these issues could be addressed with availabledata sets. The instrument had only been configuredto operate in the dual active-passive ocean color modesince spring 1984, restricting the investigations to arelatively few data sets. In general, the AOL has beenflown in support of large-scale oceanographic experi-ments that have multi-institutional participation. Inthese experiments, the flight patterns as well as theactivities onboard the research vessel are generallydesigned to provide data required by the participatingoceanographers, and information essential to improv-ing remote sensors is often obtained more incidentallythan by design. Accordingly, information on specia-tion, auxiliary photopigments (other than chlorophylla), phaeophytin, dissolved organic material concentra-tion, and the attenuation properties of the near surfaceocean layer, all of which would be useful in analyzingresults, are essentially unavailable in our existing datasets.

Nonetheless, the results thus far obtained using thepassive ocean color band selection technique indicatethat photopigments contained in chlorophyll exertprofound influences in specific regions of the entireocean color spectrum. These areas extend well intothe red region beyond 670 nm. Currently the 670-nmCZCS band is assumed to be free of contributions fromwithin the water column and thus is being utilized tocorrect CZCS imagery for atmospheric effects.

The results herein also suggest the presence of win-dows within the ocean color spectrum where a particu-lar photopigment appears to exhibit little influence onthe spectrum. It is in these spectral regions that an-other photopigment may be most optimally deter-mined. At present too little data exist to evaluate thepotential of this latter finding. It is possible that thethree-band curvature algorithm given here is only indi-rectly responding to phycoerythrin. For example, ifphycoerythrin is highly correlated to another pigment,say phycocyanin, the algorithm may actually be mea-suring the latter pigment. Thus the high correlationof the active and passive phycoerythrin measurementsmay be occurring via the covariability of another pig-ment.

The algorithm efforts described herein were initiat-ed to identify algorithms which could recover plank-tonic chlorophyll accessory pigments from upwelled


spectral radiances. The ultimate objective is to devel-op passive techniques for detection and mapping ofchlorophyll and accessory pigments from satellite plat-forms. This should in turn greatly assist in develop-ment and production of global plankton species maps.In the very long term it is hoped that spaceborne lidartechniques can complement passive systems much asthey presently do on airborne platforms.7 8

The authors wish to extend their personal thanks tothe many persons involved with the scientific fieldexperiments, the AOL project, and aircraft operations.We are particularly indebted to the Instrument Elec-tro-Optics Branch for the loan of the frequency-dou-bled Nd:YAG laser used to obtain the active data. Wealso thank the Ocean Processes Branch of NASAHeadquarters for continued support and encourage-ment. AOL system calibration assistance by the Sen-sor Evaluation Branch of GSFC and by Warren Hovisand Jack Knoll of the NOAA National EnvironmentSatellite Data and Information Service is gratefullyacknowledged. Finally, the authors thank Gary Grewfor participation in the field experiments and for theMOCS data.

References1. W. Strumm and J. Morgan, Aquatic Chemistry (Wiley, New

York, 1981).2. H. R. Gordon and A. Y. Morel, "Remote Assessment of Ocean

Color for Interpretation of Satellite Visible Imagery, A Review,"in Lecture Notes on Coastal and Estuarine Studies (Springer-Verlag, New York, 1983).

3. F. E. Hoge and R. N. Swift, "Application of the NASA AirborneOceanographic Lidar to the Mapping of Chlorophyll and OtherOrganic Pigments," in Chesapeake Bay Plume Study Super-flux 1980, NASA Conf. Publ. 2188, 349 (1981).

4. F. E. Hoge and R. N. Swift, "Airborne Simultaneous Spectro-scopic Detection of Laser-Induced Water Raman Backscatterand Fluorescence from Chlorophyll a and Other Naturally Oc-curring Pigments," Appl. Opt. 20, 3197 (1981).

5. F. E. Hoge and R. N. Swift, "Airborne Dual Laser Excitation andMapping of Phytoplankton Photopigments in a Gulf StreamWarm Core Ring," Appl. Opt. 22, 2272 (1983).

6. F. E. Hoge and R. N. Swift, "Airborne Mapping of Laser-In-duced Fluorescence of Chlorophyll a and Phycoerythrin in aGulf Stream Warm Core Ring," paper 18, in Mapping Strate-gies in Chemical Oceanography, A. Zirino, Ed. (AmericanChemical Society, Washington, DC, 1985), pp. 353-372.

7. F. E. Hoge, R. E. Berry, and R. N. Swift, "Active-Passive Air-borne Ocean Color Measurement. 1: Instrumentation," Appl.Opt. 25, 39 (1986).

8. F. E. Hoge, R. N. Swift, and J. K. Yungel, "Active-PassiveAirborne Ocean Color Measurements. 2: Applications," Appl.Opt. 25, 48 (1986).

9. M. Bristow, D. Nielsen, D. Bundy, and F. Furtek, "Use of WaterRaman Emission to Correct Airborne Laser Fluorosensor Datafor Effects of Water Optical Attenuation," Appl. Opt. 20, 2889(1981).

10. R. C. Smith, U. California, Santa Barbara; personal communica-tion.

11. D. A. Kiefer, "Chlorophyll a Fluorescence in Marine CentricDiatoms: Responses of Chloroplasts to Light and NutrientStress," Mar. Biol. 23, 39 (1973).

12. R. H. Wiens and H. H. Zwick, "Trace Gas Detection by Correla-

tion Spectroscopy," in Infrared, Correlation and FourierTransform Spectroscopy, J. S. Mattson, H. B. Mark, Jr., and H.C. MacDonald, Jr., Eds. (Marcel Dekker, New York, 1977),Chap. 3, pp. 119-190.

13. R. A. O'Neil, L. Buja-Bijunas, and D. M. Rayner, "Field Perfor-mance of a Laser Fluorosensor for the Detection of Oil Spills,"Appl. Opt. 19, 863 (1980).

14. P. G. Hasell, Jr., L. M. Peterson, F. J. Thomson, E. A. Work, andF. J. Kriegler, "Active and Passive Multispectral Scanner forEarth Resources Applications," Final Report under NASA Con-trol NAS9-14594 to Environmental Research Institute of Michi-gan, ERIM Report 115800-49-F (June 1977).

15. D. R. Lyzenga, "Remote Bathymetry Using Active-PassiveTechniques," in IEEE Digest of the International Geoscienceand Remote Sensing Symposium, Vol. 2 (June 1981), pp. 779-786.

16. G. W. Grew and L. S. Mayo, "Ocean Color Algorithm for RemoteSensing of Chlorophyll," NASA Tech. Paper 2164 (LangleyResearch Center, Hampton, VA, 1983).

17. -G. W. Grew, "Real-Time Test of MOCS Algorithm During Su-perflux 1980," in Chesapeake Bay Plume Study Superflux1980, NASA Conf. Publ. 2188 (1981).

18. J. W. Campbell and W. E. Esaias, "Basis for Spectral CurvatureAlgorithms in Remote Sensing of Chlorophyll," Appl. Opt. 22,1084 (1983).

19. S. Lin, G. A. Borstad, and J. F. R. Gower, "Remote Sensing ofChlorophyll in the Red Spectral Region," in Remote Sensing ofShelf Sea Hydrodynamics, J. C. J. Nihoul, Ed. (Elsevier, Am-sterdam, 1984), pp. 317-336.

20. R. A. Neville and J. F. R. Gower, "Passive Remote Sensing ofPhytoplankton via Chlorophyll a Fluorescence," J. Geophys.Res. 82, 3487 (1977).

21. J. F. R. Gower, "Observations of in situ Fluorescence of Chloro-phyll in Saanich Inlet," Boundary Layer Meteorol. 18, 235(1980).

22. H. R. Gordon, D. K. Clark, J. W. Brown, 0. B. Brown, R. H.Evans, and W. W. Broenkow, "Phytoplankton Pigment Concen-tration in the Middle Atlantic Bight: Comparison of Ship De-terminations and CZCS Estimates," Appl. Opt. 22, 20 (1983).

23. L. R. Poole and W. E. Esaias, "Water Raman Normalization ofAirborne Laser Fluorosensor Measurements: A Computer Stu-dy," Appl. Opt. 21, 3756 (1982).

24. A. F. Bunkin, D. V. Vlaso, D. M. Mirkamilov, and V. P.-Slobo-dyanin, "Aerial Laser Sounding of the Turbidity Profile andMapping of the Distribution of Phytoplankton," Dokl. Akad.Nauk SSSR 279,335 (1984). [NASA Tech. Memo. 77851 (1985)].

25. R. C. Smith and K. S. Baker, "Optical Classification of NaturalWaters," Limnol. Oceanogr. 23, 260 (1978).

26. J. J. Walsh, U. South Florida; private communication (manu-script in preparation).

27. C. M. Moreth and C. S. Yentsch, "A Sensitive Method for theDetermination of Open Ocean Phytoplankton PhycoerythrinPigments by Fluorescence," Limnol. Oceanogr. 15, 313 (1970).

28. D. E. Stewart, "A Method for the Extraction and Quantitation ofPhycoerythrin from Algae," NASA Contractor Report 165996by Bionetics Corp. under contract NAS1-16978, Hampton, VA.

29. M. P. F. Bristow, D. H. Bundy, C. M. Edmonds, P. E. Ponto, B.E. Frey, and L. F. Small, "Airborne Laser Fluorosensor Surveyof the Columbia and Snake Rivers: Simultaneous Measure-ments of Chlorophyll, Dissolved Organics and Optical Attenua-tion," Int. J. Remote Sensing 6, 1707 (1985).

30. S. W. Jeffrey, "Algal Pigment Systems," in Primary Productivi-ty in the Sea, P. G. Falkowski, Ed. (Plenum, New York, 1980).

31. R. J. Exton, W. M. Houghton, W. Esaias, R. C. Harriss, F. H.Farmer, and H. H. White, "Laboratory Analysis of Techniquesfor Remote Sensing of Estuarine Parameters Using Laser Exci-tation," Appl. Opt. 22, 54 (1983).


Documents

Active-passive correlation spectroscopy: a new technique for identifying ocean color algorithm spectral regions