Forgot your password?
Document Actions
A Research Paper
Summary of Ocean Color Remote Sensing methods
C O A S TA L O C E A N O P T I C S A N D D Y N A M I C S
The New Age of
Hyperspectral
Oceanogra
B Y G R A C E C H A N G , K E V I N M A H O N E Y, A M A N D A B R I G G S W H I T M I R E ,
D AV I D K O H L E R , C U R T I S M O B L E Y, M A R L O N L E W I S , M A R K M O L I N E ,
E M M A N U E L B O S S , M I N S U K I M , W I L L I A M P H I L P O T, A N D T O M M Y D I C K E Y
Oceanography
22 June 2004
Introduction
coastal and open-ocean studies. Advances in
A multispectral optical sensor collects data
computer technology in the last decade have
at select wavebands or channels. An example
enabled more rapid processing of hyperspec-
is the Sea-viewing Wide-Field-of-view Sen-
tral data and greatly improved the storing
sor (SeaWiFS) ocean color satellite, which
and archiving capability of these large, and
measures eight wavebands between 402 and
often difficult-to-manage data sets.
885 nm (20-40 nm bandwidth with peaks
phy
Hyperspectral technology has expanded
centered around 412, 443, 490, 510, 555, 670,
from hand-held radiometers to submerged
765, and 865 nm). Optical oceanographers
sensors for measurements of inherent op-
have been using multispectral sensors since
1
tical properties (IOPs), optical properties
the 1980s with great success .
that depend on only the aquatic medium
A hyperspectral sensor gives continuous
itself (e.g., absorption and scattering; Mo-
spectral coverage over a broad wavelength
bley, 1994) and apparent optical properties
range [at least over visible wavelengths,
(AOPs), which depend on the IOPs and the
and preferably from near ultraviolet (UV)
geometry of the light field. Recently, hy-
to near infrared (IR)] with better than 10
perspectral airborne detectors have been
nm resolution. The utility of hyperspectral
enhanced for high spectral and spatial reso-
measurements has long been recognized in
lution measurements of ocean radiance and
fields as diverse as geology and astronomy,
reflectance. Although multispectral sensors
and hyperspectral instruments have been
have a higher signal to noise ratio for the
used in oceanographic research for about 30
same quality of optical components (be-
years. However, most of these instruments
cause they integrate over a larger bandwidth
have been laboratory bench-top spectro-
and thus collect more photons each band),
photometers and radiometers that measure
the sensitivity and data quality of hyper-
absorption and radiance or irradiance at <10
spectral sensors are rapidly increasing and
nm continuous spectral resolution from the
costs are coming down. Thus the shift from
UV to IR wavelengths. These instruments
multispectral to hyperspectral systems will
were relatively slow with sample scan rates
continue. The availability of hyperspectral
on the order of minutes to maximize signal
sensors opens a new door for optical ocean-
to noise. Just a decade ago, computational
ography and related fields that make use of
limitations also made processing and stor-
optical remote sensing of the oceans. Here,
age of large amounts of hyperspectral data
we discuss a few of the scientific advantages
difficult. However, within the last five years,
to using high spectral resolution sensors and
high sample rate (less than seconds) in situ
describe valuable hyperspectral applications
and remote sensing hyperspectral sensors
in the marine environment.
have been developed and utilized for various
1
See special issues: “Hydrologic Optics” in Limnology and Oceanography, 34(8), 1989; “Ocean Color From Space: A Coastal
Zone Color Scanner Retrospective” in Journal of Geophysical Research, 99(C4), 7291-7270, 1994; and “Ocean Optics” in
Journal of Geophysical Research, 100(C7), 13,133-13,372, 1995).
Oceanography June 2004 23
Spectral Techniques populations are generally more abundant
Traditionally, multispectral remote sensors and less diverse, (2) terrestrial influences
have been utilized for characterizing open- [high concentrations of colored dissolved
ocean waters. Some results have shown that organic matter (CDOM) and particles]
a few, wide, carefully selected bands may cannot be ignored, (3) the influence of the
be all that is needed to monitor these water ocean bottom (bottom reflectance and sedi-
bodies whose optical signatures are domi- ment resuspension) is important, and (4)
nated by chlorophyll a and co-varying opti- high temporal and spatial variability collude
cally significant constituents. However, when to create an optically diverse environment.
these open ocean algorithms (O’Reilly et al., Not only do these influences complicate the
1998) are applied to coastal areas, the results characterization of the water and bottom
are less useful, if not altogether inapplicable types, but also make the atmospheric correc-
(Hu et al., 2000; Lee and Carder, 2002). The tion of these scenes difficult. Traditional blue
coastal ocean is an optically complex envi- water atmospheric corrections (e.g., “black
ronment. For example: (1) phytoplankton pixel” assumptions; Siegel et al., 2000) are
no longer valid. These correction methods
Grace Chang (grace.chang@opl.ucsb.edu) is As- assume that any remote sensing signal at the
sistant Researcher at Ocean Physics Laboratory, IR wavelengths is due to the atmosphere,
University of California, Santa Barbara, Goleta, but this assumption does not hold in high
CA 93117. Kevin Mahoney is Postdoctoral sediment or optically shallow coastal waters.
Fellow at Monterey Bay Aquarium Research Thus, the successful removal of the atmo-
Institute, Moss Landing, CA 95039. Amanda spheric interference in the water-leaving
Briggs-Whitmire is Graduate Student at radiance signal within the coastal environ-
College of Oceanic and Atmospheric Sciences, ment requires a priori knowledge of a host
Oregon State University, Corvallis, OR 97331. of atmospheric constituents (e.g., water col-
David Kohler is Senior Scientist at Florida Envi- umn vapor, aerosol type and density, ozone
ronmental Research Institute, Tampa, FL 33611. concentration). Without a priori knowledge,
Curtis Mobley is Vice President and Senior these constituents must be derived from the
Scientist at Sequoia Scientific, Inc., Bellevue, WA spectral data stream itself, decreasing the
98005. Marlon Lewis is Professor at Depart- degrees of freedom with which to resolve
ment of Oceanography, Dalhousie University, the water leaving radiance signal. Addition-
Halifax, Nova Scotia B3H 4J1 Canada. Mark ally, the increased development along the
Moline is Associate Professor at California Poly- world’s coastal boundaries adds a degree of
technic State University, San Luis Obispo, CA complexity in the determination of concen-
93465. Emmanuel Boss is Assistant Professor at tration and interactions between the ma-
School of Marine Sciences, University of Maine, rine and terrestrial aerosols, such that the
Orono, ME 04469. Minsu Kim is Postdoctoral atmospheric parameterization may change
Associate at School of Civil & Environmental dramatically within a single scene. Hyper-
Engineering, Cornell University, Ithaca, NY, spectral information provides optical ocean-
14853. William Philpot is Associate Professor ographers the potential to accurately correct
at School of Civil & Environmental Engineering, remote sensing images and classify complex
Cornell University, Ithaca, NY, 14853. Tommy oceanic environments, finer-scale features
Dickey is Professor at Ocean Physics Labora- (e.g., bottom type and characteristics and
tory, University of California, Santa Barbara, phytoplankton blooms), and depth-depen-
Goleta, CA 93117. dent IOPs.
Oceanography
24 June 2004
ly distinct backscattering spectra between
to the fourth derivative) together with simi-
With higher spectral resolution data (i.e.,
species (cultured) (Bricaud et al., 1983; Ahn
larity index analysis (e.g., Millie et al., 1997;
more wavelengths) come more degrees of
et al., 1992) whereas other microorgan-
Schofield et al., In press). These decomposi-
freedom for optical models and empirical
isms, such as bacteria and flagellates, show
tion analyses are techniques that separate
algorithms. Many ocean color algorithms
wavelength independent backscatter (Morel
pigment peaks and shoulders from troughs
in use today involve empirical relationships
and Ahn, 1990, 1991). These studies and the
in phytoplankton absorption curves of
between the property of interest (i.e., chloro-
results of numerous modeling efforts (see
mixed assemblages. The similarity index is
phyll a concentration, IOPs, etc.) and wave-
Stramski et al., 2001 and references therein)
typically used to correlate measured absorp-
band ratios of remote sensing reflectance or
demonstrate that backscatter is not spectral-
tion with known phytoplankton absorption
water-leaving radiance (O’Reilly et al., 1998).
Most of these algorithms are derived by re-
“Hyperspectral information provides optical oceanographers
gressions of radiance at select (or available)
wavebands or waveband ratios versus the
the potential to accurately correct remote sensing images and
property of interest. Naturally, the regression
classify complex oceanic environments, finer-scale features...,
results are maximized at the highest num-
ber of statistically independent wavelengths and depth-dependent [inherent optical properties].”
available. Also, the spectral resolution of
derived IOPs is limited by the number of
ly flat (as it is oftentimes modeled) or easily
curves for identification purposes by taking
wavebands of the ocean color remote sens-
predicted for all particles. Therefore, back-
into account the differences in shapes be-
ing data used in the regression.
scatter has the potential to provide a means
tween two spectra based on the peaks and
Multispectral measurements of absorp-
to identify phytoplankton by group or spe-
troughs of each spectrum. These identifica-
tion are useful for determining the relative
cies and to determine particle characteristics.
tion techniques usually cannot be applied
concentrations and variability of the differ-
This provides incentive for the development
to multispectral data because the required
ent constituents in the water column: water
of in situ hyperspectral backscatter sensors
features (i.e., peaks and troughs) are not well
itself, phytoplankton, CDOM, and inorgan-
and algorithms.
resolved.
ics (Schofield et al., In press and references
While the absorption properties of nu-
therein). Absorption peaks of chlorophyll a,
Examples of Hyperspectral
merous planktonic species and other water
non-pigmented troughs, and the exponen-
Analyses
column constituents have been studied ex-
tial slopes of CDOM and inorganic material
tensively, the same cannot be said for their
are well distinguished in absorption spec- Hyperspectral data used in combination
backscattering properties. Backscattering
tra collected by most multispectral sensors. with spectral techniques such as derivative
properties must be known in order to ac-
However, in order to identify phytoplankton analysis, spectral angle mapping, spectral
curately interpret ocean color measure-
by taxonomic group or species, quantifi- deconvolution, and similarity indices can aid
ments because the reflectance of the upper
cation of the absorption by accessory or in the characterization of marine ecosystems
ocean is directly related to the ratio of the
marker pigments beyond chlorophyll a is including the detection and identification of
backscattering coefficient to the absorption
oftentimes necessary. Some accessory pig- harmful algal blooms, an increasing prob-
coefficient. Hyperspectral backscattering
ments are unique to individual phytoplank- lem in the world’s coastal oceans (Millie et
measurements can be used to distinguish
ton taxa and usually cannot be discerned in al., 1997; Lohrenz et al., 1999). For example,
phytoplankton populations from co-varying
absorption spectra with a limited number of Figure 1 shows phytoplankton absorption
seawater constituents because the spectral
wavelengths or wavebands (accessory pig- spectra for a red tide species, Karenis bre-
dependence of backscattering by algal cells
ment peaks are generally narrow), but can vis, measured with a multispectral sensor,
is different from that of other particles (Bri-
be discriminated in hyperspectral data. This a hyperspectral sensor, and modeled us-
caud et al., 1983; Stramski et al., 2001). Also,
discrimination can be accomplished with ing Mie theory (following Mahoney, 2001).
hyperspectral backscatter measurements in
various methods such as spectral unmixing K. brevis can be identified by its accessory
the laboratory have revealed that some phy-
and deconvolution, Gaussian decomposi- pigment, Gyroxanthin–diester, which has
toplankton species may show complex, high-
tion, and derivative analysis (usually taken unique absorption peaks at 444 and 469 nm
Oceanography June 2004 25
(Örnólfsdóttir et al., 2003). As seen in Figure remote sensing reflectance spectra for two
1, the multispectral spectrum lacks detailed water types generated by the Hydrolight
absorption information, i.e., pigment peaks radiative transfer model (Mobley, 1994).
by distinguishing accessory pigments, due Water 1 is 6.5 m and has low chlorophyll a
to a limited number of wavebands. Hyper- and CDOM concentrations with a bottom
spectral data allow for the detection of spe- type of a mixture of soft coral and Sargas-
cies-discriminating accessory pigments and sum, while Water 2 is 13 m deep, “pure wa-
are more adequate for comparing measured ter” with a flat green sponge bottom type. By
spectra to a reference spectrum for similar- inspection of the hyperspectral spectra, the
ity index analysis (Figure 1). Wood et al. difference between the two curves is obvious
(2002) have also used these techniques and in the 500-600 nm range. However, spectra
presented evidence that distinctive hyper- for the two water types produced using only
spectral signatures are associated with Syn- the SeaWiFS wavebands appear almost iden-
echococcus blooms in upwelling and nutrient tical (note: the SeaWiFS spectra were derived
enrichment systems in the Gulf of Califor- by applying the SeaWiFS spectral response
nia. Cannizzaro et al. (2002) show that it is function to the hyperspectral signatures). A
possible to utilize multispectral techniques second example, Figure 3, shows 122 remote
(SeaWiFS) to detect K. brevis. However, their sensing reflectance spectra generated by Hy-
method works only for waters under certain drolight for various combinations of nine
optical conditions (low concentrations of different sets of IOPs, 32 different bottom
CDOM and suspended sediments relative to reflectances, and 22 depths between 5.5 and
chlorophyll a or low backscattering relative 50 m. These spectra are clearly unique. How-
to absorption) as different ocean color prod- ever, every spectrum has nearly the same
ucts (particulate backscattering and its rela- remote sensing reflectance wavelength ratio:
tionship to chlorophyll a) are used as proxies Rrs(490)/Rrs(555) = 1.71 ± 0.01. This ratio,
for K. brevis abundance. if used in the SeaWiFS Ocean Chlorophyll 2
In the past, multispectral techniques have (OC2) band-ratio algorithm (O’Reilly et al.,
been used for the derivation of water depth 1998, as revised on http://seawifs.gsfc.nasa.
and bottom bathymetry (e.g., Philpot, 1989; gov/SEAWIFS/RECAL/Repro3/OC4_repro-
Maritorena et al., 1994), and more recently cess.html), gives a chlorophyll concentra-
tion of 0.59 ± 0.01 mg Chl m-3. Thus these
for characterization of bottom type (see
“Light in Shallow Waters” in Limnology and simulated water bodies, which have IOPs
Oceanography, 48(2), 2003). These analy- corresponding to chlorophyll concentrations
ses generally involve empirical algorithms, between 0.0 (pure water) and 0.2 mg Chl
m-3, are all viewed as the same by the OC2
where reflectance waveband ratios are re-
gressed against water depth. Wavelength lim- algorithm. The OC2 algorithm fails here
itations and commonly employed assump- because of bottom effects in optically clear
tions that the water optical properties are waters simulated by Hydrolight.
vertically uniform and constant over the area While much of the interest in hyper-
being mapped can lead to inaccurate retriev- spectral approaches relates to the visible
als of bottom depth and characteristics un- wavebands, several oceanic constituents of
der certain conditions. These retrievals can interest have distinct spectral signatures
be improved with hyperspectral data (Lee in the UVA/UVB (e.g., Ogura and Hanya,
and Carder, 2002 and references therein). 1966). Chief among these is nitrate, a ma-
For example, Figure 2 shows hyperspectral jor plant nutrient that limits the primary
Oceanography
26 June 2004
0.3
Figure 1. Phytoplankton taxonomic group or species identification is now achievable
with the development of hyperspectral instruments; generally narrow accessory pig-
0.25 ment absorption wavelength peaks that are unique to specific species can be discerned.
Shown here are three different methods used to measure phytoplankton absorption
spectra for a red tide species, Karenis brevis, on the west Florida shelf. Closed circles
0.2
symbolize absorption measured with a multispectral sensor (ac-9). Open circles signify
data modeled using Mie theory (following Mahoney, 2001), and plus signs represent
Gyroxanthin−diester
a (m −1 )
0.15 data measured with a hyperspectral sensor (HiStar). It is apparent in this figure that
the multispectral spectrum lacks the distinguishing accessory pigment peaks due to a
limited number of wavebands. Hyperspectral data, however, allow for the detection of
0.1 species-discriminating accessory pigments and are more adequate for comparing mea-
sured spectra to a reference spectrum and thus phytoplankton species identification.
K. brevis can be identified by its accessory pigment, Gyroxanthin–diester, which has
0.05
unique absorption peaks at 444 and 469 nm (Örnólfsdóttir et al., 2003). (Multispectral
ac9 data were provided by Oscar Schofield and John Kerfoot, Rutgers University and hyper-
0 Mi e spectral data were provided by Steven Lohrenz, University of Southern Mississippi.)
HiStar
−0.05
350 400 450 500 550 600 650 700 750
Wavelength (nm)
Figure 2. Bottom effects in shallow coastal waters may lead to inaccurate remote
sensing retrievals of bottom depth if limited spectral bands are utilized for analysis.
This figure shows modeled hyperspectral (solid lines) and multispectral (SeaWiFS
wavebands; circles) spectra for two water types, generated by the Hydrolight radia-
tive transfer model (Mobley, 1994). Water 1 (blue) is 6.5 m deep and has low chloro-
phyll-a and CDOM concentrations with a bottom type of a mixture of soft coral and
Sargassum, while Water 2 (green) is 13 m deep, “pure water” with a flat green sponge
bottom type. By inspection of the hyperspectral spectra, the difference between the
two curves is obvious in the 500-600 nm range. However, spectra for the two water
types produced using only the SeaWiFS wavebands appear almost identical. (Sea-
WiFS spectra in this figure were derived by applying the SeaWiFS spectral response
function to the hyperspectral signatures).
Figure 3. Chlorophyll concentration algorithms designed for multispectral instrumenta-
tion may not be useful for shallow, optically clear waters. Shown here are one hundred
twenty two Hydrolight-generated remote sensing reflectance (Rrs) spectra for Bahamian
waters using various combinations of nine different sets of IOPs, 32 different bottom
reflectances, and 22 depths between 5.5 and 50 m. These spectra are clearly unique.
However, every spectrum has nearly the same remote sensing reflectance wavelength
ratio: Rrs(490)/Rrs(555) = 1.71 ± 0.01 (490 and 555 nm are indicated by the vertical black
dashed lines). If this ratio were applied to the commonly used SeaWiFS band-ratio
algorithm (OC2; O’Reilly et al., 1998), it would give a chlorophyll concentration of 0.59
± 0.01 mg Chl m-3. In other words, the same chlorophyll concentration would be deter-
mined for all 122 spectra despite the fact that these simulated water bodies have IOPs
corresponding to chlorophyll concentrations between 0.0 (pure water) and 0.2 mg Chl
m-3. The OC2 algorithm fails here because of bottom effects in optically clear waters.
Oceanography June 2004 27
production of organic matter in many re- is now achievable with the development of
gions of the world’s oceans. The net vertical hyperspectral instruments; generally narrow
transport of nitrate, for example, constrains accessory pigment absorption wavelength
the export flux of organic matter from the peaks that are unique to specific species can
surface ocean in a steady-state sense. Nitrate be discerned. High spectral resolution back-
dissolved in seawater exhibits a broad ab- scattering spectra are unique to some phy-
sorption maximum centered at ~210 nm; it toplankton species and can aid in the char-
competes with the absorption of bromide, acterization of oceanic particles. One other
a conservative component of sea-salt, and exciting aspect of hyperspectral technology
to a lesser extent, the carbonate ion (Figure is the development of optically based chemi-
4). In anaerobic areas, sulphide also absorbs cal sensors. These sensors allow for long-
in a band around 220 nm and various dis- term monitoring of ecologically important
solved organic compounds of oceanographic nutrients and potentially harmful pollutants
and practical purposes (e.g., TNT) exhibit at unprecedented time and space scales.
absorption maxima in the UV. Past attempts Hyperspectral instrumentation is becom-
to estimate the concentration of nitrate and ing increasingly important to oceanographic
other compounds with multispectral instru- research as coastal and open ocean observ-
ments have been met with equivocal suc- ing systems are rapidly developing into key
cess. The introduction of a field-deployable elements for scientific research, monitoring,
hyperspectral UV absorption spectrometer, decision-making, science education, and
i.e., the In Situ Ultraviolet Spectrometer outreach. Some concerns of these observa-
(ISUS), coupled with advanced spectro- tories are that autonomous sampling plat-
scopic deconvolution techniques, has made forms can be limited by weight and volume
routine spectral measurements of nutrients and data bandwidth capabilities. The incor-
possible (Johnson and Coletti, 2002; Figure poration of hyperspectral sensors to autono-
4). Oceanographers are now able to resolve mous sampling platforms of an observing
ABOVE AND PRECEDING THREE SPREADS: Three bands
(RGB= 666, 547, 439 nm) from a March 23, 1996 Air-
nitrate concentrations in the ocean at tem- system can expand the amount of informa-
borne Visible/Infrared Imaging Spectrometer (AVIRIS)
poral and spatial scales consistent with mea- tion gained from one instrument without
image taken over the Florida Keys from an ER-2 aircraft
surements of temperature and salinity and compromising platform payload. High
at 20 km above ground. The top of the image is near
the eastern end of the Keys; the bottom of the image is
to an accuracy and precision more than ac- spectral resolution sensors provide a greater
near the western end. The rough heading is 260 degrees
ceptable for oceanographic biogeochemical number of wavelengths for various analysis
(clockwise from north) top to bottom (i.e., just south of
west). AVIRIS is an optical sensor that delivers calibrated investigations, as a direct result of a hyper- techniques, particularly in optically complex
images of upwelling spectral radiance in 224 contiguous
spectral approach to the problem. coastal environments. In addition, emerging
spectral channels (bands) with wavelengths from 400
cabled observatories offer exceptional power
to 2500 nanometers. Note that this image is not atmo-
Summary and Conclusions
spherically corrected. It is pi*radiance/[mean_solar_ir- and data bandwidth for hyperspectral sen-
radiance_at_the_top_of_the_atmosphere * cos(solar_
sors.
Hyperspectral technology provides a means
zenith_angle)]. Original data courtesy of NASA/JPL.
Optical oceanographers have been posing
for optical oceanographers to classify and
Caption courtesy of Marcos Montes of Naval Research
Laboratory, Washington D.C.
hyperspectrally-related questions since the
quantify complex oceanic environments (in
popularity of ocean exploration expanded
situ and remotely): bottom depth and type,
in the 1950s. However, technological and
particle characteristics, depth-dependent
computing constraints limited us to the
IOPs, and specific chemical compounds.
use of multispectral or even single wave-
Hyperspectral data enable, for the first time,
length sensors in our field studies. Now that
a real attempt at environmental spectros-
computing power has become more than
copy. In situ and remote phytoplankton
adequate to handle large quantities of data
taxonomic group or species identification
Oceanography
28 June 2004
Figure 4. Nitrate is a major plant nutrient that limits the primary production of organic matter
in many regions of the world’s oceans. Nitrate dissolved in seawater exhibits a broad absorption
maximum centered at ~210 nm. The introduction of a field-deployable hyperspectral UV absorp-
tion spectrometer, known as In Situ Ultraviolet Spectrometer (ISUS), coupled with advanced spec-
troscopic deconvolution techniques, has made routine spectral measurements of nutrients pos-
sible at unprecedented time and space scales. The specific molar absorption of bromide (black,
dotted line) and nitrate (black, solid line) are shown with the absorption spectrum of whole
water (red line; 1 nm resolution) measured with the ISUS (MBARI/Satlantic Inc.) deployed on a
Conductivity-Temperature Depth profiler (CTD) at 150 m depth in the western Equatorial Pacific.
Most of the variance in absorption is explained by bromide; advanced deconvolution techniques
are required to extract the concentration of nitrate (here 14.9 µM) based on its absorption.
west Florida shelf: A novel classification technique, plankton upon the optical properties and particulate
and technology has allowed miniaturiza-
Ocean Optics XVI, CD-ROM. organic carbon in oceanic waters, J. Mar. Res., 48,
tion of in situ and remotely sensed optical
Hu C., K.L. Carder, and F.E. Muller-Karger, 2000: Atmo- 145-175.
sensors, we have been able to utilize hyper- spheric Correction of SeaWiFS Imagery over Turbid Morel, A., and Y.-H. Ahn, 1991: Optics of heterotrophic
Coastal Waters: A Practical Method, Rem. Sens. Envi- nanoflagellates and ciliates: A tentative assessment
spectral instruments in the field to answer a
ron., 74(2), 195-206. of their scattering role in oceanic waters compared
host of scientific questions that were never Johnson, K.S., and L.J. Coletti, 2002: In situ ultraviolet to those of bacterial and algal cells, J. Mar. Res., 49,
before possible. This is an exciting time for spectrophotometry for high resolution and long- 177-202.
term monitoring of nitrate, bromide and bisulfide in Ogura, N., and T. Hanya, 1966: Nature of ultra-violet
oceanographers. New spectral algorithms
the ocean, Deep-Sea Res. I, 49, 1291-1305. absorption in sea water, Nature, 212, 758-759.
and techniques will be further developed Lee, Z.P., and K.L. Carder, 2002: Effect of spectral band Örnólfsdóttir, E.B., J.L. Pinckney, and P.A. Tester, 2003:
and refined as high spectral resolution tech- numbers on the retrieval of water column and bot- Quantification of the relative abundance of the toxic
tom properties from ocean color data, Appl. Opt., 41, dinoflagellate, Karenia brevis (Dinophyta), using
nologies are improved for IOP sensors for
2191-2201. unique photopigments, J. of Phycol., 39(2), 449-457.
measurements of absorption, attenuation, Lohrenz, S.E., G.L. Fahnenstiel, G.J. Kirkpatrick, C.L. O’Reilly, J.E., S. Maritorena, B.G. Mitchell, D.A. Siegel,
Carroll, and K.A. Kelly, 1999: K.L. Carder, S.A. Garver, M. Kahru, and C. McClain,
scattering, backscattering, and the volume
Microphotometric assessment of spectral absorption 1998: Ocean color chlorophyll algorithms for Sea-
scattering function. Potentially, a suite of
and its potential application for characterization of WiFS, J. Geophys. Res., 103(C11), 24,937-24,953.
relatively small, robust hyperspectral in- harmful algal species, J. Phycol., 35, 1438-1446. Philpot, W.D., 1989: Bathymetric mapping with passive
Mahoney, K., 2001: Optical properties of Karenis bre- multispectral imagery, Appl. Opt., 28, 1569-1578.
struments will be available to measure and
vis and implications for remote sensing reflectance, Siegel, D., M. Wang, S. Maritorena, and W. Robinson,
resolve bottom depth and type, particle size Ph.D. Dissertation, University of Southern Missis- 2000: Atmospheric correction of satellite ocean color
and type (including phytoplankton species sippi. imagery: The black pixel assumption, Appl. Opt.,
Maritorena, S., A. Morel, and B. Gentili, 1994: Diffuse 39(21), 3582-3591.
identification), and micro- and macro-nu-
reflectance of oceanic shallow waters: Influence of Schofield, O., T. Bergmann, M.J. Oliver, A. Irwin, P.W.
trient concentrations in situ in the very near water depth and bottom albedo, Limnol. Oceanogr., Bissett, M.A. Moline, and C. Orrico, In press: Inver-
future. 39, 1689-1703. sion of the bulk absorption in the Mid-Atlantic Bight
Millie, D.F., O.M. Schofield, G.J. Kirkpatrick, G. John- and its utility for water mass analysis in optically
sen, P.A. Tester, and B.T. Vinyard, 1997: Detection complex coastal waters, J. Geophys. Res.
References of harmful algal blooms using photopigments and Stramski, D., A. Bricaud, and A. Morel, 2001: Modeling
Ahn, Y.H., Bricaud, A., and A. Morel, 1992: Light back- absorption signatures: A case study of the Florida red the inherent optical properties of the ocean based on
scattering efficiency and related properties of some the detailed composition of planktonic community,
tide dinoflagellates, Gymnodinium breve, Limnol.
phytoplankters, Deep-Sea Res., 39, 1835-1855. Appl. Opt., 40, 2929-2945.
Oceanogr., 42(5, part 2), 1240-1251.
Bricaud, A., Morel, A., and L. Prieur, 1983: Optical ef- Wood, M., W.S. Pegau, W.K. W. Li, C.C. Trees, J.L. Muel-
Mobley, C.D., 1994: Light and Water: Radiative Transfer
ficiency factors of some phytoplankters, Limnol. ler, and H. Maske, 2002: Optical signatures of a Syn-
in Natural Waters, Academic Press, San Diego, 592
Oceanogr., 28, 816-832. echococcus bloom in the Gulf of California, Ocean
pp.
Cannizzaro, J.P., K.L. Carder, F.R. Chen, and C.A. Heil, Optics XVI, CD-ROM.
Morel, A.,and Y.H. Ahn, 1990: Optical efficiency factors
2002: Remote detection of red tide blooms on the of free-living marine bacteria: Influence of bacterio-
Oceanography June 2004 29
The New Age of
Hyperspectral
Oceanogra
B Y G R A C E C H A N G , K E V I N M A H O N E Y, A M A N D A B R I G G S W H I T M I R E ,
D AV I D K O H L E R , C U R T I S M O B L E Y, M A R L O N L E W I S , M A R K M O L I N E ,
E M M A N U E L B O S S , M I N S U K I M , W I L L I A M P H I L P O T, A N D T O M M Y D I C K E Y
Oceanography
22 June 2004
Introduction
coastal and open-ocean studies. Advances in
A multispectral optical sensor collects data
computer technology in the last decade have
at select wavebands or channels. An example
enabled more rapid processing of hyperspec-
is the Sea-viewing Wide-Field-of-view Sen-
tral data and greatly improved the storing
sor (SeaWiFS) ocean color satellite, which
and archiving capability of these large, and
measures eight wavebands between 402 and
often difficult-to-manage data sets.
885 nm (20-40 nm bandwidth with peaks
phy
Hyperspectral technology has expanded
centered around 412, 443, 490, 510, 555, 670,
from hand-held radiometers to submerged
765, and 865 nm). Optical oceanographers
sensors for measurements of inherent op-
have been using multispectral sensors since
1
tical properties (IOPs), optical properties
the 1980s with great success .
that depend on only the aquatic medium
A hyperspectral sensor gives continuous
itself (e.g., absorption and scattering; Mo-
spectral coverage over a broad wavelength
bley, 1994) and apparent optical properties
range [at least over visible wavelengths,
(AOPs), which depend on the IOPs and the
and preferably from near ultraviolet (UV)
geometry of the light field. Recently, hy-
to near infrared (IR)] with better than 10
perspectral airborne detectors have been
nm resolution. The utility of hyperspectral
enhanced for high spectral and spatial reso-
measurements has long been recognized in
lution measurements of ocean radiance and
fields as diverse as geology and astronomy,
reflectance. Although multispectral sensors
and hyperspectral instruments have been
have a higher signal to noise ratio for the
used in oceanographic research for about 30
same quality of optical components (be-
years. However, most of these instruments
cause they integrate over a larger bandwidth
have been laboratory bench-top spectro-
and thus collect more photons each band),
photometers and radiometers that measure
the sensitivity and data quality of hyper-
absorption and radiance or irradiance at <10
spectral sensors are rapidly increasing and
nm continuous spectral resolution from the
costs are coming down. Thus the shift from
UV to IR wavelengths. These instruments
multispectral to hyperspectral systems will
were relatively slow with sample scan rates
continue. The availability of hyperspectral
on the order of minutes to maximize signal
sensors opens a new door for optical ocean-
to noise. Just a decade ago, computational
ography and related fields that make use of
limitations also made processing and stor-
optical remote sensing of the oceans. Here,
age of large amounts of hyperspectral data
we discuss a few of the scientific advantages
difficult. However, within the last five years,
to using high spectral resolution sensors and
high sample rate (less than seconds) in situ
describe valuable hyperspectral applications
and remote sensing hyperspectral sensors
in the marine environment.
have been developed and utilized for various
1
See special issues: “Hydrologic Optics” in Limnology and Oceanography, 34(8), 1989; “Ocean Color From Space: A Coastal
Zone Color Scanner Retrospective” in Journal of Geophysical Research, 99(C4), 7291-7270, 1994; and “Ocean Optics” in
Journal of Geophysical Research, 100(C7), 13,133-13,372, 1995).
Oceanography June 2004 23
Spectral Techniques populations are generally more abundant
Traditionally, multispectral remote sensors and less diverse, (2) terrestrial influences
have been utilized for characterizing open- [high concentrations of colored dissolved
ocean waters. Some results have shown that organic matter (CDOM) and particles]
a few, wide, carefully selected bands may cannot be ignored, (3) the influence of the
be all that is needed to monitor these water ocean bottom (bottom reflectance and sedi-
bodies whose optical signatures are domi- ment resuspension) is important, and (4)
nated by chlorophyll a and co-varying opti- high temporal and spatial variability collude
cally significant constituents. However, when to create an optically diverse environment.
these open ocean algorithms (O’Reilly et al., Not only do these influences complicate the
1998) are applied to coastal areas, the results characterization of the water and bottom
are less useful, if not altogether inapplicable types, but also make the atmospheric correc-
(Hu et al., 2000; Lee and Carder, 2002). The tion of these scenes difficult. Traditional blue
coastal ocean is an optically complex envi- water atmospheric corrections (e.g., “black
ronment. For example: (1) phytoplankton pixel” assumptions; Siegel et al., 2000) are
no longer valid. These correction methods
Grace Chang (grace.chang@opl.ucsb.edu) is As- assume that any remote sensing signal at the
sistant Researcher at Ocean Physics Laboratory, IR wavelengths is due to the atmosphere,
University of California, Santa Barbara, Goleta, but this assumption does not hold in high
CA 93117. Kevin Mahoney is Postdoctoral sediment or optically shallow coastal waters.
Fellow at Monterey Bay Aquarium Research Thus, the successful removal of the atmo-
Institute, Moss Landing, CA 95039. Amanda spheric interference in the water-leaving
Briggs-Whitmire is Graduate Student at radiance signal within the coastal environ-
College of Oceanic and Atmospheric Sciences, ment requires a priori knowledge of a host
Oregon State University, Corvallis, OR 97331. of atmospheric constituents (e.g., water col-
David Kohler is Senior Scientist at Florida Envi- umn vapor, aerosol type and density, ozone
ronmental Research Institute, Tampa, FL 33611. concentration). Without a priori knowledge,
Curtis Mobley is Vice President and Senior these constituents must be derived from the
Scientist at Sequoia Scientific, Inc., Bellevue, WA spectral data stream itself, decreasing the
98005. Marlon Lewis is Professor at Depart- degrees of freedom with which to resolve
ment of Oceanography, Dalhousie University, the water leaving radiance signal. Addition-
Halifax, Nova Scotia B3H 4J1 Canada. Mark ally, the increased development along the
Moline is Associate Professor at California Poly- world’s coastal boundaries adds a degree of
technic State University, San Luis Obispo, CA complexity in the determination of concen-
93465. Emmanuel Boss is Assistant Professor at tration and interactions between the ma-
School of Marine Sciences, University of Maine, rine and terrestrial aerosols, such that the
Orono, ME 04469. Minsu Kim is Postdoctoral atmospheric parameterization may change
Associate at School of Civil & Environmental dramatically within a single scene. Hyper-
Engineering, Cornell University, Ithaca, NY, spectral information provides optical ocean-
14853. William Philpot is Associate Professor ographers the potential to accurately correct
at School of Civil & Environmental Engineering, remote sensing images and classify complex
Cornell University, Ithaca, NY, 14853. Tommy oceanic environments, finer-scale features
Dickey is Professor at Ocean Physics Labora- (e.g., bottom type and characteristics and
tory, University of California, Santa Barbara, phytoplankton blooms), and depth-depen-
Goleta, CA 93117. dent IOPs.
Oceanography
24 June 2004
ly distinct backscattering spectra between
to the fourth derivative) together with simi-
With higher spectral resolution data (i.e.,
species (cultured) (Bricaud et al., 1983; Ahn
larity index analysis (e.g., Millie et al., 1997;
more wavelengths) come more degrees of
et al., 1992) whereas other microorgan-
Schofield et al., In press). These decomposi-
freedom for optical models and empirical
isms, such as bacteria and flagellates, show
tion analyses are techniques that separate
algorithms. Many ocean color algorithms
wavelength independent backscatter (Morel
pigment peaks and shoulders from troughs
in use today involve empirical relationships
and Ahn, 1990, 1991). These studies and the
in phytoplankton absorption curves of
between the property of interest (i.e., chloro-
results of numerous modeling efforts (see
mixed assemblages. The similarity index is
phyll a concentration, IOPs, etc.) and wave-
Stramski et al., 2001 and references therein)
typically used to correlate measured absorp-
band ratios of remote sensing reflectance or
demonstrate that backscatter is not spectral-
tion with known phytoplankton absorption
water-leaving radiance (O’Reilly et al., 1998).
Most of these algorithms are derived by re-
“Hyperspectral information provides optical oceanographers
gressions of radiance at select (or available)
wavebands or waveband ratios versus the
the potential to accurately correct remote sensing images and
property of interest. Naturally, the regression
classify complex oceanic environments, finer-scale features...,
results are maximized at the highest num-
ber of statistically independent wavelengths and depth-dependent [inherent optical properties].”
available. Also, the spectral resolution of
derived IOPs is limited by the number of
ly flat (as it is oftentimes modeled) or easily
curves for identification purposes by taking
wavebands of the ocean color remote sens-
predicted for all particles. Therefore, back-
into account the differences in shapes be-
ing data used in the regression.
scatter has the potential to provide a means
tween two spectra based on the peaks and
Multispectral measurements of absorp-
to identify phytoplankton by group or spe-
troughs of each spectrum. These identifica-
tion are useful for determining the relative
cies and to determine particle characteristics.
tion techniques usually cannot be applied
concentrations and variability of the differ-
This provides incentive for the development
to multispectral data because the required
ent constituents in the water column: water
of in situ hyperspectral backscatter sensors
features (i.e., peaks and troughs) are not well
itself, phytoplankton, CDOM, and inorgan-
and algorithms.
resolved.
ics (Schofield et al., In press and references
While the absorption properties of nu-
therein). Absorption peaks of chlorophyll a,
Examples of Hyperspectral
merous planktonic species and other water
non-pigmented troughs, and the exponen-
Analyses
column constituents have been studied ex-
tial slopes of CDOM and inorganic material
tensively, the same cannot be said for their
are well distinguished in absorption spec- Hyperspectral data used in combination
backscattering properties. Backscattering
tra collected by most multispectral sensors. with spectral techniques such as derivative
properties must be known in order to ac-
However, in order to identify phytoplankton analysis, spectral angle mapping, spectral
curately interpret ocean color measure-
by taxonomic group or species, quantifi- deconvolution, and similarity indices can aid
ments because the reflectance of the upper
cation of the absorption by accessory or in the characterization of marine ecosystems
ocean is directly related to the ratio of the
marker pigments beyond chlorophyll a is including the detection and identification of
backscattering coefficient to the absorption
oftentimes necessary. Some accessory pig- harmful algal blooms, an increasing prob-
coefficient. Hyperspectral backscattering
ments are unique to individual phytoplank- lem in the world’s coastal oceans (Millie et
measurements can be used to distinguish
ton taxa and usually cannot be discerned in al., 1997; Lohrenz et al., 1999). For example,
phytoplankton populations from co-varying
absorption spectra with a limited number of Figure 1 shows phytoplankton absorption
seawater constituents because the spectral
wavelengths or wavebands (accessory pig- spectra for a red tide species, Karenis bre-
dependence of backscattering by algal cells
ment peaks are generally narrow), but can vis, measured with a multispectral sensor,
is different from that of other particles (Bri-
be discriminated in hyperspectral data. This a hyperspectral sensor, and modeled us-
caud et al., 1983; Stramski et al., 2001). Also,
discrimination can be accomplished with ing Mie theory (following Mahoney, 2001).
hyperspectral backscatter measurements in
various methods such as spectral unmixing K. brevis can be identified by its accessory
the laboratory have revealed that some phy-
and deconvolution, Gaussian decomposi- pigment, Gyroxanthin–diester, which has
toplankton species may show complex, high-
tion, and derivative analysis (usually taken unique absorption peaks at 444 and 469 nm
Oceanography June 2004 25
(Örnólfsdóttir et al., 2003). As seen in Figure remote sensing reflectance spectra for two
1, the multispectral spectrum lacks detailed water types generated by the Hydrolight
absorption information, i.e., pigment peaks radiative transfer model (Mobley, 1994).
by distinguishing accessory pigments, due Water 1 is 6.5 m and has low chlorophyll a
to a limited number of wavebands. Hyper- and CDOM concentrations with a bottom
spectral data allow for the detection of spe- type of a mixture of soft coral and Sargas-
cies-discriminating accessory pigments and sum, while Water 2 is 13 m deep, “pure wa-
are more adequate for comparing measured ter” with a flat green sponge bottom type. By
spectra to a reference spectrum for similar- inspection of the hyperspectral spectra, the
ity index analysis (Figure 1). Wood et al. difference between the two curves is obvious
(2002) have also used these techniques and in the 500-600 nm range. However, spectra
presented evidence that distinctive hyper- for the two water types produced using only
spectral signatures are associated with Syn- the SeaWiFS wavebands appear almost iden-
echococcus blooms in upwelling and nutrient tical (note: the SeaWiFS spectra were derived
enrichment systems in the Gulf of Califor- by applying the SeaWiFS spectral response
nia. Cannizzaro et al. (2002) show that it is function to the hyperspectral signatures). A
possible to utilize multispectral techniques second example, Figure 3, shows 122 remote
(SeaWiFS) to detect K. brevis. However, their sensing reflectance spectra generated by Hy-
method works only for waters under certain drolight for various combinations of nine
optical conditions (low concentrations of different sets of IOPs, 32 different bottom
CDOM and suspended sediments relative to reflectances, and 22 depths between 5.5 and
chlorophyll a or low backscattering relative 50 m. These spectra are clearly unique. How-
to absorption) as different ocean color prod- ever, every spectrum has nearly the same
ucts (particulate backscattering and its rela- remote sensing reflectance wavelength ratio:
tionship to chlorophyll a) are used as proxies Rrs(490)/Rrs(555) = 1.71 ± 0.01. This ratio,
for K. brevis abundance. if used in the SeaWiFS Ocean Chlorophyll 2
In the past, multispectral techniques have (OC2) band-ratio algorithm (O’Reilly et al.,
been used for the derivation of water depth 1998, as revised on http://seawifs.gsfc.nasa.
and bottom bathymetry (e.g., Philpot, 1989; gov/SEAWIFS/RECAL/Repro3/OC4_repro-
Maritorena et al., 1994), and more recently cess.html), gives a chlorophyll concentra-
tion of 0.59 ± 0.01 mg Chl m-3. Thus these
for characterization of bottom type (see
“Light in Shallow Waters” in Limnology and simulated water bodies, which have IOPs
Oceanography, 48(2), 2003). These analy- corresponding to chlorophyll concentrations
ses generally involve empirical algorithms, between 0.0 (pure water) and 0.2 mg Chl
m-3, are all viewed as the same by the OC2
where reflectance waveband ratios are re-
gressed against water depth. Wavelength lim- algorithm. The OC2 algorithm fails here
itations and commonly employed assump- because of bottom effects in optically clear
tions that the water optical properties are waters simulated by Hydrolight.
vertically uniform and constant over the area While much of the interest in hyper-
being mapped can lead to inaccurate retriev- spectral approaches relates to the visible
als of bottom depth and characteristics un- wavebands, several oceanic constituents of
der certain conditions. These retrievals can interest have distinct spectral signatures
be improved with hyperspectral data (Lee in the UVA/UVB (e.g., Ogura and Hanya,
and Carder, 2002 and references therein). 1966). Chief among these is nitrate, a ma-
For example, Figure 2 shows hyperspectral jor plant nutrient that limits the primary
Oceanography
26 June 2004
0.3
Figure 1. Phytoplankton taxonomic group or species identification is now achievable
with the development of hyperspectral instruments; generally narrow accessory pig-
0.25 ment absorption wavelength peaks that are unique to specific species can be discerned.
Shown here are three different methods used to measure phytoplankton absorption
spectra for a red tide species, Karenis brevis, on the west Florida shelf. Closed circles
0.2
symbolize absorption measured with a multispectral sensor (ac-9). Open circles signify
data modeled using Mie theory (following Mahoney, 2001), and plus signs represent
Gyroxanthin−diester
a (m −1 )
0.15 data measured with a hyperspectral sensor (HiStar). It is apparent in this figure that
the multispectral spectrum lacks the distinguishing accessory pigment peaks due to a
limited number of wavebands. Hyperspectral data, however, allow for the detection of
0.1 species-discriminating accessory pigments and are more adequate for comparing mea-
sured spectra to a reference spectrum and thus phytoplankton species identification.
K. brevis can be identified by its accessory pigment, Gyroxanthin–diester, which has
0.05
unique absorption peaks at 444 and 469 nm (Örnólfsdóttir et al., 2003). (Multispectral
ac9 data were provided by Oscar Schofield and John Kerfoot, Rutgers University and hyper-
0 Mi e spectral data were provided by Steven Lohrenz, University of Southern Mississippi.)
HiStar
−0.05
350 400 450 500 550 600 650 700 750
Wavelength (nm)
Figure 2. Bottom effects in shallow coastal waters may lead to inaccurate remote
sensing retrievals of bottom depth if limited spectral bands are utilized for analysis.
This figure shows modeled hyperspectral (solid lines) and multispectral (SeaWiFS
wavebands; circles) spectra for two water types, generated by the Hydrolight radia-
tive transfer model (Mobley, 1994). Water 1 (blue) is 6.5 m deep and has low chloro-
phyll-a and CDOM concentrations with a bottom type of a mixture of soft coral and
Sargassum, while Water 2 (green) is 13 m deep, “pure water” with a flat green sponge
bottom type. By inspection of the hyperspectral spectra, the difference between the
two curves is obvious in the 500-600 nm range. However, spectra for the two water
types produced using only the SeaWiFS wavebands appear almost identical. (Sea-
WiFS spectra in this figure were derived by applying the SeaWiFS spectral response
function to the hyperspectral signatures).
Figure 3. Chlorophyll concentration algorithms designed for multispectral instrumenta-
tion may not be useful for shallow, optically clear waters. Shown here are one hundred
twenty two Hydrolight-generated remote sensing reflectance (Rrs) spectra for Bahamian
waters using various combinations of nine different sets of IOPs, 32 different bottom
reflectances, and 22 depths between 5.5 and 50 m. These spectra are clearly unique.
However, every spectrum has nearly the same remote sensing reflectance wavelength
ratio: Rrs(490)/Rrs(555) = 1.71 ± 0.01 (490 and 555 nm are indicated by the vertical black
dashed lines). If this ratio were applied to the commonly used SeaWiFS band-ratio
algorithm (OC2; O’Reilly et al., 1998), it would give a chlorophyll concentration of 0.59
± 0.01 mg Chl m-3. In other words, the same chlorophyll concentration would be deter-
mined for all 122 spectra despite the fact that these simulated water bodies have IOPs
corresponding to chlorophyll concentrations between 0.0 (pure water) and 0.2 mg Chl
m-3. The OC2 algorithm fails here because of bottom effects in optically clear waters.
Oceanography June 2004 27
production of organic matter in many re- is now achievable with the development of
gions of the world’s oceans. The net vertical hyperspectral instruments; generally narrow
transport of nitrate, for example, constrains accessory pigment absorption wavelength
the export flux of organic matter from the peaks that are unique to specific species can
surface ocean in a steady-state sense. Nitrate be discerned. High spectral resolution back-
dissolved in seawater exhibits a broad ab- scattering spectra are unique to some phy-
sorption maximum centered at ~210 nm; it toplankton species and can aid in the char-
competes with the absorption of bromide, acterization of oceanic particles. One other
a conservative component of sea-salt, and exciting aspect of hyperspectral technology
to a lesser extent, the carbonate ion (Figure is the development of optically based chemi-
4). In anaerobic areas, sulphide also absorbs cal sensors. These sensors allow for long-
in a band around 220 nm and various dis- term monitoring of ecologically important
solved organic compounds of oceanographic nutrients and potentially harmful pollutants
and practical purposes (e.g., TNT) exhibit at unprecedented time and space scales.
absorption maxima in the UV. Past attempts Hyperspectral instrumentation is becom-
to estimate the concentration of nitrate and ing increasingly important to oceanographic
other compounds with multispectral instru- research as coastal and open ocean observ-
ments have been met with equivocal suc- ing systems are rapidly developing into key
cess. The introduction of a field-deployable elements for scientific research, monitoring,
hyperspectral UV absorption spectrometer, decision-making, science education, and
i.e., the In Situ Ultraviolet Spectrometer outreach. Some concerns of these observa-
(ISUS), coupled with advanced spectro- tories are that autonomous sampling plat-
scopic deconvolution techniques, has made forms can be limited by weight and volume
routine spectral measurements of nutrients and data bandwidth capabilities. The incor-
possible (Johnson and Coletti, 2002; Figure poration of hyperspectral sensors to autono-
4). Oceanographers are now able to resolve mous sampling platforms of an observing
ABOVE AND PRECEDING THREE SPREADS: Three bands
(RGB= 666, 547, 439 nm) from a March 23, 1996 Air-
nitrate concentrations in the ocean at tem- system can expand the amount of informa-
borne Visible/Infrared Imaging Spectrometer (AVIRIS)
poral and spatial scales consistent with mea- tion gained from one instrument without
image taken over the Florida Keys from an ER-2 aircraft
surements of temperature and salinity and compromising platform payload. High
at 20 km above ground. The top of the image is near
the eastern end of the Keys; the bottom of the image is
to an accuracy and precision more than ac- spectral resolution sensors provide a greater
near the western end. The rough heading is 260 degrees
ceptable for oceanographic biogeochemical number of wavelengths for various analysis
(clockwise from north) top to bottom (i.e., just south of
west). AVIRIS is an optical sensor that delivers calibrated investigations, as a direct result of a hyper- techniques, particularly in optically complex
images of upwelling spectral radiance in 224 contiguous
spectral approach to the problem. coastal environments. In addition, emerging
spectral channels (bands) with wavelengths from 400
cabled observatories offer exceptional power
to 2500 nanometers. Note that this image is not atmo-
Summary and Conclusions
spherically corrected. It is pi*radiance/[mean_solar_ir- and data bandwidth for hyperspectral sen-
radiance_at_the_top_of_the_atmosphere * cos(solar_
sors.
Hyperspectral technology provides a means
zenith_angle)]. Original data courtesy of NASA/JPL.
Optical oceanographers have been posing
for optical oceanographers to classify and
Caption courtesy of Marcos Montes of Naval Research
Laboratory, Washington D.C.
hyperspectrally-related questions since the
quantify complex oceanic environments (in
popularity of ocean exploration expanded
situ and remotely): bottom depth and type,
in the 1950s. However, technological and
particle characteristics, depth-dependent
computing constraints limited us to the
IOPs, and specific chemical compounds.
use of multispectral or even single wave-
Hyperspectral data enable, for the first time,
length sensors in our field studies. Now that
a real attempt at environmental spectros-
computing power has become more than
copy. In situ and remote phytoplankton
adequate to handle large quantities of data
taxonomic group or species identification
Oceanography
28 June 2004
Figure 4. Nitrate is a major plant nutrient that limits the primary production of organic matter
in many regions of the world’s oceans. Nitrate dissolved in seawater exhibits a broad absorption
maximum centered at ~210 nm. The introduction of a field-deployable hyperspectral UV absorp-
tion spectrometer, known as In Situ Ultraviolet Spectrometer (ISUS), coupled with advanced spec-
troscopic deconvolution techniques, has made routine spectral measurements of nutrients pos-
sible at unprecedented time and space scales. The specific molar absorption of bromide (black,
dotted line) and nitrate (black, solid line) are shown with the absorption spectrum of whole
water (red line; 1 nm resolution) measured with the ISUS (MBARI/Satlantic Inc.) deployed on a
Conductivity-Temperature Depth profiler (CTD) at 150 m depth in the western Equatorial Pacific.
Most of the variance in absorption is explained by bromide; advanced deconvolution techniques
are required to extract the concentration of nitrate (here 14.9 µM) based on its absorption.
west Florida shelf: A novel classification technique, plankton upon the optical properties and particulate
and technology has allowed miniaturiza-
Ocean Optics XVI, CD-ROM. organic carbon in oceanic waters, J. Mar. Res., 48,
tion of in situ and remotely sensed optical
Hu C., K.L. Carder, and F.E. Muller-Karger, 2000: Atmo- 145-175.
sensors, we have been able to utilize hyper- spheric Correction of SeaWiFS Imagery over Turbid Morel, A., and Y.-H. Ahn, 1991: Optics of heterotrophic
Coastal Waters: A Practical Method, Rem. Sens. Envi- nanoflagellates and ciliates: A tentative assessment
spectral instruments in the field to answer a
ron., 74(2), 195-206. of their scattering role in oceanic waters compared
host of scientific questions that were never Johnson, K.S., and L.J. Coletti, 2002: In situ ultraviolet to those of bacterial and algal cells, J. Mar. Res., 49,
before possible. This is an exciting time for spectrophotometry for high resolution and long- 177-202.
term monitoring of nitrate, bromide and bisulfide in Ogura, N., and T. Hanya, 1966: Nature of ultra-violet
oceanographers. New spectral algorithms
the ocean, Deep-Sea Res. I, 49, 1291-1305. absorption in sea water, Nature, 212, 758-759.
and techniques will be further developed Lee, Z.P., and K.L. Carder, 2002: Effect of spectral band Örnólfsdóttir, E.B., J.L. Pinckney, and P.A. Tester, 2003:
and refined as high spectral resolution tech- numbers on the retrieval of water column and bot- Quantification of the relative abundance of the toxic
tom properties from ocean color data, Appl. Opt., 41, dinoflagellate, Karenia brevis (Dinophyta), using
nologies are improved for IOP sensors for
2191-2201. unique photopigments, J. of Phycol., 39(2), 449-457.
measurements of absorption, attenuation, Lohrenz, S.E., G.L. Fahnenstiel, G.J. Kirkpatrick, C.L. O’Reilly, J.E., S. Maritorena, B.G. Mitchell, D.A. Siegel,
Carroll, and K.A. Kelly, 1999: K.L. Carder, S.A. Garver, M. Kahru, and C. McClain,
scattering, backscattering, and the volume
Microphotometric assessment of spectral absorption 1998: Ocean color chlorophyll algorithms for Sea-
scattering function. Potentially, a suite of
and its potential application for characterization of WiFS, J. Geophys. Res., 103(C11), 24,937-24,953.
relatively small, robust hyperspectral in- harmful algal species, J. Phycol., 35, 1438-1446. Philpot, W.D., 1989: Bathymetric mapping with passive
Mahoney, K., 2001: Optical properties of Karenis bre- multispectral imagery, Appl. Opt., 28, 1569-1578.
struments will be available to measure and
vis and implications for remote sensing reflectance, Siegel, D., M. Wang, S. Maritorena, and W. Robinson,
resolve bottom depth and type, particle size Ph.D. Dissertation, University of Southern Missis- 2000: Atmospheric correction of satellite ocean color
and type (including phytoplankton species sippi. imagery: The black pixel assumption, Appl. Opt.,
Maritorena, S., A. Morel, and B. Gentili, 1994: Diffuse 39(21), 3582-3591.
identification), and micro- and macro-nu-
reflectance of oceanic shallow waters: Influence of Schofield, O., T. Bergmann, M.J. Oliver, A. Irwin, P.W.
trient concentrations in situ in the very near water depth and bottom albedo, Limnol. Oceanogr., Bissett, M.A. Moline, and C. Orrico, In press: Inver-
future. 39, 1689-1703. sion of the bulk absorption in the Mid-Atlantic Bight
Millie, D.F., O.M. Schofield, G.J. Kirkpatrick, G. John- and its utility for water mass analysis in optically
sen, P.A. Tester, and B.T. Vinyard, 1997: Detection complex coastal waters, J. Geophys. Res.
References of harmful algal blooms using photopigments and Stramski, D., A. Bricaud, and A. Morel, 2001: Modeling
Ahn, Y.H., Bricaud, A., and A. Morel, 1992: Light back- absorption signatures: A case study of the Florida red the inherent optical properties of the ocean based on
scattering efficiency and related properties of some the detailed composition of planktonic community,
tide dinoflagellates, Gymnodinium breve, Limnol.
phytoplankters, Deep-Sea Res., 39, 1835-1855. Appl. Opt., 40, 2929-2945.
Oceanogr., 42(5, part 2), 1240-1251.
Bricaud, A., Morel, A., and L. Prieur, 1983: Optical ef- Wood, M., W.S. Pegau, W.K. W. Li, C.C. Trees, J.L. Muel-
Mobley, C.D., 1994: Light and Water: Radiative Transfer
ficiency factors of some phytoplankters, Limnol. ler, and H. Maske, 2002: Optical signatures of a Syn-
in Natural Waters, Academic Press, San Diego, 592
Oceanogr., 28, 816-832. echococcus bloom in the Gulf of California, Ocean
pp.
Cannizzaro, J.P., K.L. Carder, F.R. Chen, and C.A. Heil, Optics XVI, CD-ROM.
Morel, A.,and Y.H. Ahn, 1990: Optical efficiency factors
2002: Remote detection of red tide blooms on the of free-living marine bacteria: Influence of bacterio-
Oceanography June 2004 29