* Present address: Center for Antarctic Environment Monitoring, Na-tional Institute of Polar Research, 9-10, Kaga 1, Itabasi-ku, Tokyo 173-8515, Japan.
Bio-Optical Relationship of Case I Waters:The Difference between the Low- andMid-Latitude Waters and the Southern Ocean
TORU HIRAWAKE1*, HIROO SATOH1, TAKASHI ISHIMARU1,YUKUYA YAMAGUCHI1 and MOTOAKI KISHINO2
1Department of Ocean Science, Tokyo University of Fisheries, 5-7, Konan 4, Minato-ku, Tokyo 108-8477, Japan2Institute of Physical and Chemical Research (RIKEN), 2-1, Hirosawa,Wako-shi, Saitama 351-0198, Japan
(Received 16 April 1999; in revised form 27 September 1999; accepted 27 September 1999)
Both historic and currently operational chlorophyll algorithms of the satellite-borneocean color sensors, such as SeaWiFS, were evaluated for in situ spectral radiationand chlorophyll data in some Case I waters, including the waters in the Indian Oceansector of the Southern Ocean. Chlorophyll a concentration of the data set (n = 73)ranged from 0.04 to 1.01 mg m–3. The algorithms had higher accuracy for the low-and mid-latitude waters (RMSE: 0.163–0.253), specifically the most recently devel-oped algorithms of OCTS and SeaWiFS showed 0.163 and 0.170 of Root Mean SquareErrors, respectively. However, these algorithms had large errors (0.422–0.621) forthe Southern Ocean data set and underestimated the surface chlorophyll by morethan a factor of 2.6. The absorption coefficients in the blue spectral region retrievedfrom remote sensing reflectance varied in a nonlinear manner with chlorophyll a con-centration, and the value in the Southern Ocean was significantly lower than that inthe low- and mid-latitude waters for each chlorophyll a concentration. The underes-timation of chlorophyll a concentration in the Southern Ocean with these algorithmswas caused by the lower specific absorption coefficient in the region compared withthe low- and mid-latitude waters under the same chlorophyll a concentration.
1992). A decade after the end of the CZCS mission, thesecond such sensor, the Ocean Color and TemperatureScanner (OCTS) on the ADEOS platform, was launchedin 1996 by NASDA and collected high-resolution imageryduring its ten months of operation. Moreover, the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) on theOrbview-2 platform was successfully launched in August1997. The new sensors have more visible bands and twoNIR bands to avoid the limitation of CZCS (Hooker etal., 1993). Therefore, atmospheric correction has madeprogress in terms of accuracy (Gordon and Wang, 1994)and the effects of accessory pigments (Sathyendranath etal., 1994), and colored dissolved organic matter (CDOM)(Aiken et al., 1992; DeGrandpre et al., 1996) can be alsoconsidered. This means that ocean color, which is a spec-trum of the water leaving upwelling radiance or reflect-ance at the sea surface (Hovis et al., 1980), and pigmentconcentration can be observed from space with increas-ingly high precision.
1. IntroductionOcean color remote sensing has developed as a pro-
cedure to estimate pigment concentration ofphytoplankton in the ocean for clarification of the globalcarbon cycle. The possibility of using ocean color remotesensing was mentioned by Clarke et al. (1970) and con-firmed by airplane experiments conducted by NASA.After the experiments, in 1978, the first satellite-borneocean color sensor, Coastal Zone Color Scanner (CZCS)which had only four visible bands, was launched on theNimbus-7. Although the design was limited, the CZCSprovided high quality imagery until 1986 (Hooker et al.,
246 T. Hirawake et al.
In-water algorithms to derive chlorophyll a and pig-ments (chlorophyll a + pheopigment) have also devel-oped with the progress of the ocean color sensors. Themost frequently utilized in-water algorithms are repre-sented empirically using bio-optical relationships betweenthe ratio of radiance at two or three different wavelengthsand phytoplankton pigment concentration. Clarke et al.(1970) proposed this concept, and Morel and Prieur (1977)illustrated the relationship in their figure. Gordon andClark (1980) developed some in-water algorithms withthe development of the CZCS. A bio-optical relationshipgiven by Gordon et al. (1983) was widely applied as astandard algorithm in processing the global CZCS dataset. After the CZCS mission, a number of improved algo-rithms were developed for the CZCS and new sensors (e.g.Morel, 1988; Sathyendranath et al., 1989; Müller-Kargeret al., 1990; Aiken et al., 1992, 1995; Mitchell, 1992;Kishino et al., 1998). Most of these algorithms, however,were based on data of limited geographic distribution. Inparticular, the standard algorithm of the CZCS (Gordonet al., 1983) was derived from the Nimbus ExperimentTeam (NET) data set (Acker, 1994) which contains radi-ance and chlorophyll data observed only around NorthAmerica.
The radiance spectrum at the sea surface used in anin-water algorithm depends on the radiance reflectance,which is related to inherent optical properties such as theabsorption and scattering coefficients (Gordon et al.,1988). For Case I water (Morel and Prieur, 1977), theoptical property depends on only the absorption coeffi-cient of phytoplankton and its degradaion products. How-ever, the chlorophyll a specific absorption coefficientvaries significantly with regions and seasons (Mitchelland Kiefer, 1988; Garver et al., 1994; Allai et al., 1997;Garver and Siegel, 1997) and the variation affects spec-tral radiance. Despite the variability, the suitability ofthese algorithms has not been evaluated sufficiently.Therefore the algorithms developed using regionally lim-ited data cause error in estimating global chlorophyll adistribution. For example the CZCS standard algorithm
underestimated chlorophyll a concentration in the South-ern Ocean (Arrigo et al., 1994), and it was postulated thatthe phytoplankton had manifested some sort of pigmentpackage effect (Mitchell and Holm-Hansen, 1991).
Any utilization of historic (CZCS and OCTS), cur-rent (SeaWiFS) and future sensors (e.g. Moderate-Reso-lution Imaging Spectrometer (MODIS) and Global Imager(GLI)) for the estimation of chlorophyll a and productiv-ity in the ocean requires a knowledge of bio-optical prop-erties and algorithms, both regionally and seasonally. Thedata sets corresponding to new sensors, however, are lim-ited (McClain et al., 1992) because they use differentwavelengths from the CZCS (Aiken et al., 1995). In par-ticular, the bio-optical data obtained from the SouthernOcean are limited due to the difficulty of access (Mitchell,1992), but such data are required to improve the preci-sion of the bio-optical algorithm and to understand themechanisms underlying regional differences (Mitchell andHolm-Hansen, 1991).
In this paper we report the results of an evaluationof historic and currently used algorithms for our in situspectral radiation and chlorophyll data in some Case Iwaters, including the Southern Ocean. For a more theo-retical interpretation of the difference in the algorithmbetween the low- and mid-latitude waters and the South-ern Ocean, we also present the difference in chlorophylla specific absorption coefficient between both watersusing a reflectance model, and we discuss their effectson the bio-optical algorithms.
2. Materials and Methods
2.1 Sampling locationSpectral radiation and chlorophyll data for this study
were collected in the Arabian Sea, Indian Ocean, JapanSea, Southern Ocean and the Pacific Ocean, during thecruises of T/V Umitaka-Maru III or Shinyo-Maru of To-kyo University of Fisheries, and the 38th Japanese Ant-arctic Research Expedition (JARE) on board the ice-breaker Shirase (Fig. 1). Details of the dates, and number
Location Date n Spectroradiometer
Eastern Indian Ocean, Arabian Sea Jan. 1994 16 (15)* MER-2020ASouthern Ocean Jan. and Feb. 1996 20 MER-2020AJapan Sea July 1996 15 MER-2040Pacific Dec. 1996 12 MER-2040Southern Ocean (JARE) Dec. 1996 and Mar. 1997 10 ( 9)* PRR-600Total 73 (71)
Table 1. Data sources.
*Spectral radiation data at one station was removed from the data set for optical analysis due to effect of squall or ship-shadow.
Bio-Optical Relationship of Case I Waters 247
of bio-optical stations for all observed locations are listedin Table 1. Bio-optical measurement was carried out at73 stations.
2.2 Profiling of spectral radiationDownwelling irradiance (Ed) and upwelling radiance
(Lu) were measured with an underwater spectroradiometer,MER-2020A or MER-2040 or PRR-600 (BiosphericalInstruments) (Table 1). Although the technical specifica-tions of the spectroradiometers were slightly differentfrom each other, they were compatible with the OCTSand/or SeaWiFS and some channels will be used for fu-ture ocean color sensors (Table 2). Data from the sensorswere stored on an internal memory (MER-2020A) ortransmitted to a computer on deck (MER-2040 and PRR-
600). The instruments were deployed from the sunny sideto avoid any ship-shading effect, and measured spectralradiation up to 150 m depth.
2.3 Chlorophyll a and pheopigmentsSeawater samples for the measurement of chlorophyll
a standing stock of phytoplankton were collected usingVan Dorn bottles or a Rossete multi water samplerequipped with an OCTOPUS system (OCTO-ParameterUnderwater Sensor; Ishimaru et al., 1984) at severaldepths from the sea surface to 150 m.
For the determination of phytoplankton chlorophylla and pheopigments, between 200 and 500 ml of watersample was filtered onto a glass fiber filter (WhatmanGF/F, ø25 mm). The filter was immediately soaked in
Fig. 1. Location of the sampling stations in this study.
248 T. Hirawake et al.
Tab
le 2
. T
echn
ical
spe
cifi
cati
ons
of o
cean
col
or s
enso
r on
sat
elli
tes
and
spec
tror
adio
met
ers.
Bio-Optical Relationship of Case I Waters 249
N,N-dimethylformamide (DMF) and pigments were ex-tracted in the dark (Suzuki and Ishimaru, 1990). The con-centration of chlorophyll a was determinedfluorometrically with a Turner Designs 10-005R fluor-ometer within a few days after the extraction (Parsons etal., 1984). The fluorometer was calibrated with a spec-trophotometer using pure chlorophyll a (SIGMA) and theconcentration of a standard solution was calculated us-ing the value of chlorophyll a specific absorption coeffi-cient in DMF (Porra et al., 1989).
2.4 Optical data processing and algorithmsEd(λ) and Lu(λ) were averaged every 1 m. Values of
Ed(λ) and Lu(λ) at just below the sea surface, Ed(λ , 0–)and Lu(λ , 0–), were defined as coefficients fitting the databetween 0 and 10 m according to the following equations(Gordon et al., 1983):
Ed(λ , z) = Ed(λ , 0–) × exp[–Kd(λ) × z], (1)
and
Lu(λ , z) = Lu(λ , 0–) × exp[–Ku(λ) × z], (2)
where λ is wavelength, z is depth, Kd is diffuse attenua-tion coefficient of downwelling irradiance, and Ku is dif-fuse attenuation coefficient of upwelling radiance. Waterleaving radiance (Austin, 1974), Lw(λ), was calculatedfrom:
Lw(λ) = (t/nw2) × Lu(λ , 0–), (3)
where t is surface transmittance from sea to air, and nw isrefractive index of sea water. t and nw are approximately0.98 and 1.341 (Austin, 1974), respectively, and they arerelatively independent of wavelength (Gordon and Morel,1983). The normalized water leaving radiance used forthe analysis of algorithms presented here, Lwn(λ), is de-fined as follows:
L L F Ewn w o dsλ λ λ λ( ) = ( ) × ( ) ( ) ( )/ , 4
where Fo (λ) is the mean extraterrestrial solar irradiance(Neckel and Labs, 1984), and Eds(λ) is incident irradi-ance at the sea surface. The value of Fo (λ) was averagedover bandwidths of 10 nm, which is sufficient in this studysince the bandwidths of the spectroradiometer are approxi-mately 10 nm (Mobley, 1994). The remote sensing re-flectance used in the test of SeaWiFS algorithm and re-flectance model, Rrs(λ), can be written
Rrs(λ) = Lw(λ)/Eds(λ). (5)
The algorithms evaluated in this study were fiveempirical bio-optical relationships and their equations areas follows:
[C + P] = 1.130 × [Lwn(443)/Lwn(550)]–1.705, (6)
C = 5.56 × {[Lwn(443) + Lwn(520)]/Lwn(550)}–2.252, (7)
C = 0.2818 × {[Lwn(520) + Lwn(565)]/Lwn(490)}3.497, (8)
where C and P are chlorophyll a and pheopigments con-centration (mg m–3) at the sea surface, and R =log10[Rrs(490)/Rrs(555)]. Equation (6) is the standard al-gorithm of the CZCS for [C + P] < 1.5 (mg m–3) (Gordonet al., 1983), and Eq. (7) is Clark’s improved algorithmfor the CZCS, which avoids the switch of the coefficientsin Eq. (6) at 1.5 (mg m–3) (Müller-Karger et al., 1990).Equation (8) is the OCTS standard chlorophyll a algo-rithm version 1.0 (Kishino et al., 1998). Equation (9) isthe SeaWiFS algorithm, updated in 1998 (McClain et al.,1998). Equation (10) is one of the algorithms developedby Mitchell (1992) for the waters in the Antarctic Ocean.Here we have assumed that the wavelengths 550 nm inEqs. (6) and (7), 441 and 560 nm in Eq. (10) are equal to555, 443, and 565 nm, respectively.
2.5 Semi-analytic modelThe semi-analytic model developed by Carder et al.
(1999) was used for calculating absorption coefficients.The model is an algorithm based on the following equa-tion:
Rft
Q n
b
a brsw
b
b
λλ
λλ λ
( ) = ( )( )
( ) + ( )[ ] ( )2
2 11,
where f is an empirical factor, Q(λ) is the upwelling irra-diance to radiance ratio, and t and nw are as described inthe previous section. The ratio f/Q(λ) is relatively inde-pendent of wavelength, λ , and solar zenith angle (Gordonet al., 1988; Morel and Gentili, 1993). bb(λ) is the totalbackscattering coefficient, a(λ) is the total absorptioncoefficient. The a(λ) and bb(λ) terms were expanded as
bb(λ) = bbw(λ) + bbp(λ), (12)
and
250 T. Hirawake et al.
a(λ) = aw(λ) + aph(λ) + ad(λ) + ag(λ), (13)
where the subscripts w, p, ph, d, and g refer to water,particles, phytoplankton, detritus, and gelbstoff (alsocalled colored dissolved organic matter, CDOM). bbw(λ)and aw(λ) are constant and their values were taken fromSmith and Baker (1981) for bbw(λ) and Pope and Fry(1997) for aw(λ).
Carder et al. (1999) represented the shapes of thebbp(λ), the aph(λ), and the ad(λ) + ag(λ) spectra as func-tions of Rrs(λ) at three wavelengths (443, 490 and 555nm), aph(675), and ad(400) + ag(400), respectively.aph(675) and ad(400) + ag(400) were computed using spec-tral ratios of Rrs(λ) at three wavelengths (412, 443, and555 nm) which allow one to cancel the ratio, f·t2/Q(λ)·nw
2
in Eq. (11). The aph(λ) can be calculated by inserting thecomputed aph(675) value into the function representingthe aph(λ) spectrum. In this paper, however, a(λ) was cal-culated by Eq. (11) using the values f/Q(λ) = 0.0949(Gordon et al., 1988), t = 0.98, and nw = 1.341 (Austin,1974) to remove the error caused by approximation ofthe aph(λ) spectrum. The aph(λ) was then calculated byEq. (13). The algorithm code introduced in Carder et al.(1999) was used for the computation of ad(400) + ag(400).
2.6 Evaluation of algorithmRoot-mean-square error (RMSE) was computed as
an index by which to evaluate algorithm performance(O’Reilly et al., 1998; Carder et al., 1999). The RMSE isdefined as follows:
Table 3. Definition of the symbols.
Bio-Optical Relationship of Case I Waters 251
RMSEMod Obs
=( ) − ( )[ ] ( )∑ log log
,10 102
14n
where Mod is the modeled value, Obs is observed (in situ)data and n is the number of data.
The symbols that are used in this paper are listed inTable 3.
3. Results
3.1 Chlorophyll a dataFrequency distributions of latitude, chlorophyll a
concentration (C), and concentration of chlorophyll a +pheopigments ([C + P]) in our data set are shown in Fig.2. Although stations were somewhat concentrated on the
latitude from 60°S to 70°S (23 stations), the observed sta-tions were scattered over a vast geographical region from50°N to 70°S and a ratio of the number of data in theSouthern Ocean to those in the low- and mid-latitudewaters was 30:43. The C and [C + P] ranged over twoorders of magnitude in the low range from 0.04 to 1.01and from 0.04 to 1.17, respectively. The mean value ofchlorophyll a concentration for all stations was 0.27 mgm–3. The Southern Ocean means of C and [C + P] (0.33and 0.38 mg m–3) were greater than those in the low- andmid-latitude waters (0.22 and 0.25). Although the resultsin the Southern Ocean described below are mentionedseparately from the low- and mid-latitude waters, the dif-ference in the means of C and [C + P] are not statisticallysignificant.
3.2 Spectral adjustments for Lw(λ) and Eds(λ)Wavelength constructions of spectroradiometers used
in this study were different from each other (Table 2).The MER-2020A is not equipped with a channel at 520nm, while the PRR-600 has no channels at 510 nm and555 nm. Because the bandwidth of these channels was 10nm, differences of wavelengths between 510 and 520 nm,555 and 565 nm, were considered to be non-negligiblefor algorithms and a correction between wavelengths forLw(λ) and Eds(λ) was required. The relationships betweenboth channels were calculated with least square fittingfor possible Lw(λ) and Eds(λ) data, measured with MER-2040 for 510 to 520 nm adjustment (n = 27) and withMER-2020A and MER-2040 for 555 to 565 nm adjust-ment (n = 62) (Table 1). That gave linear relationships(Fig. 3) with a strong correlation of more than R2 = 0.98as follows:
Although the optical data from some regions were usedfor the adjustments, these spectral relationships were sig-nificant (p < 0.0001). Therefore, the results will be use-ful within the range of the chlorophyll a concentration inthis study to compare the ocean color sensors that havedifferent spectral specifications.
Fig. 2. Frequency distributions of latitude of sampling station,sea surface chlorophyll a concentration and concentrationof chlorophyll a plus pheopigments in our data set.
252 T. Hirawake et al.
3.3 Evaluation of historic and currently operated algo-rithmsThe distribution of the bio-optical data set was iden-
tified by plotting, the relationships between Rrs(490)/Rrs(555) ratio and chlorophyll a concentration for the dataset (n = 71) and for each sampling location (Fig. 4). Al-though the ratio Rrs(490)/Rrs(555) is used in the SeaWiFSstandard algorithm, a widely applied formulation, suchas Eq. (6), is used for regression for all data as a refer-ence, and represented by following equation:
C (mg m–3) = 1.277[Rrs(490)/Rrs(555)]–1.577, (19)
with a low correlation (R2 = 0.465). The Rrs(490)/Rrs(555)ratios were widely scattered and the subsets in the polarregion and the others were distributed symmetrically withrespect to the regression line.
Scatter plots of band ratio vs. C adopted in the his-toric and currently operated algorithms and the bio-opti-cal relationship (solid line) represented by from Eq. (6)to (10) are shown in Fig. 5. Regression lines were calcu-lated separately for low- and mid-latitude waters andSouthern Ocean using the CZCS-type formulation:
C or [C + P] = a × [band ratio]b. (20)
These are also shown in Fig. 5. Root Mean Square Errors(RMSE) produced by Eq. (6) to (10) and regression coef-ficients for the low- and mid-latitude waters, the South-ern Ocean including the JARE data, and for the all dataare summarized in Tables 4 and 5, respectively.
The lines of the CZCS, Clark’s, OCTS and SeaWiFSalgorithms were close to the regression lines for the low-and mid-latitude waters (short dashed line). These algo-rithms showed a relatively high accuracy for the low- andmid-latitude waters in comparison with the SouthernOcean, and their RMSE ranged from 0.170 to 0.253. Onlythe OCTS and SeaWiFS algorithm satisfied the criteriaof RMSE (<0.185) reported by O’Reilly et al. (1998).However, these algorithms yielded a large error (0.422–0.621) and underestimated the surface C and [C + P] by afactor of more than 2.6 for the Southern Ocean data set.The underestimation appeared in the significantly largevalue of the regression coefficient, a, in the SouthernOcean (Table 5). Even if the regression for the entire dataset was used, which had low coefficient of determinationR2 (0.465–0.618), the data distribution was inherentlysymmetric with respect to the regression line and the prob-lem cannot be resolved. Mitchell’s algorithm (Mitchell,1992), which was developed using the bio-optical datafrom cruises to the Antarctic Peninsula, gave a better re-
Fig. 3. Interrelationships between 510 and 520 nm, 555 and 565 nm for Lw(λ) and Eds(λ). The equations of regression lines withleast square fitting are shown in Eqs. (15), (16), (17) and (18).
Bio-Optical Relationship of Case I Waters 253
Location RMSE
Algorithm
CZCS Clark OCTS SeaWiFS MitchellC + P C + P C C C + P
Fig. 4. The relationship between the band ratio used in theSeaWiFS standard algorithm, Rrs(490)/Rrs(555), and chlo-rophyll a concentration for all data of the final data set (n =71) and its subset. The solid line shows regression for allthe data. Solid circle (�) and open circle (�) represent thedata in the low- and mid-latitude waters and the SouthernOcean, respectively.
Table 4. Root Mean Square Error (RMSE) for the low- and mid-latitude waters (Indian Ocean, Japan Sea and Pacific), theSouthern Ocean (included JARE data) and all data produced by historic and current bio-optical algorithm.
sult for the Southern Ocean data (RMSE = 0.235) in com-parison with the other algorithms.
O’Reilly et al. (1998) showed that the SeaWiFS al-gorithm using modified cubic polynomial (MCP) formu-lations represented by Eq. (9) performed particularly wellstatistically. In this study, however, RMSE of the MCPalgorithm (0.170) was not significantly different from theerror of the CZCS-type algorithm represented by Eq. (20)(0.193).
3.4 Absorption coefficientIn order to confirm and to lead to a more obvious
relationship between the difference of the bio-optical al-gorithms and absorption coefficient for our data set, wecalculated the absorption coefficient from the remote-sensing reflectance, Rrs, based on the semi-analytic modelreported by Carder et al. (1999). Relationships betweenabsorption coefficients at four wavelengths and chloro-phyll a concentration are shown in Figs. 6, 7 and 8. Thenegative values of absorption calculated from the modelwere removed from the plots. The variation of the ab-sorption coefficients were represented in a nonlinearmanner with chlorophyll a concentration, C, as well asthe report of Bricaud et al. (1998), which is a power func-tion:
absorption = A × CB, (21)
where absorption is aph(λ) + ad(λ) + ag(λ), ad(λ) + ag(λ),and aph(λ). For example, the approximations with leastsquare fits at 443 nm for the low- and mid-latitude wa-ters and the Southern Ocean were as follows:
Fig. 5. Scatter plots of band ratio vs. C or C + P for the low- and mid-latitude waters (�) and the Southern Ocean data (�). Thebio-optical relationship adopted in the historic and current algorithms, Eq. (6)–Eq. (10), are represented by solid line. Shortdashed lines (- - -) and dashed lines (— — —) show regression results for the low- and mid-latitude waters and the SouthernOcean, respectively.
Other results of the least square fitting, A and B, are listedin Table 6.
aph(λ) + ad(λ) + ag(λ) (Fig. 6), which represents thetotal absorption coefficient with the effect of pure wateritself removed, in the blue visible region (412, 443, 490nm) in the Southern Ocean, was significantly lower thanthe low- and mid-latitude waters for the same value of C.In particular, the difference at 412 nm was significant(more than 2-fold), as shown in the value of A (Table 6).The aph(555) + ad(555) + ag(555) value was almost con-stant in the low- and mid-latitude waters, and increasedslightly with C in the Southern Ocean.
The sum of absorption coefficients due to detritusand gelbstoff, ad(λ) + ag(λ) (Fig. 7), was independent ofC in the Southern Ocean. In the low- and mid-latitude
waters, however, it co-varied obviously with the concen-tration at 412, 443, and 490 nm. ad(λ) + ag(λ) at 555 nmwas constant and low over the region studied.
aph(λ) (Fig. 8) at the wavelengths of 412, 443, and490 nm in both regions increase with C. The regressionlines at 412 and 443 nm of the Southern Ocean were al-most linear with C, as shown in the value of B ≅ 1 for Eq.(21) (Table 6). The difference of aph(λ) at these wave-length between the Southern Ocean and the low- and mid-latitude waters seems to be smaller than that of ad(λ) +ag(λ). Although the variation of aph(555) in the SouthernOcean was large, it is somewhat dependent on C as wellas the aph(555) + ad(555) + ag(555). aph(555) values inthe low- and mid-latitude waters were almost constant.
Table 5. Results of the regression of the relationships between ratio of normalized water leaving radiance or remote sensingreflectance at two or three wavelengths and the chlorophyll a concentration illustrated in Fig. 5.
Fig. 6. Variations of the absorption coefficients aph(λ) + ad(λ) + ag(λ) at 412, 443, 490 and 555 nm retrieved from remote sensingreflectance, Rrs, as a function of the chlorophyll a concentration for the low- and mid-latitude waters (�) and the SouthernOcean data (�). The curves represent the power function fitted to the data, and the fitting results were shown in Table 6.
256 T. Hirawake et al.
Fig. 8. As Fig. 6, but for aph(λ).
Fig. 7. As Fig. 6, but for ad(λ) + ag(λ).
Bio-Optical Relationship of Case I Waters 257
Table 6. Results of the regression of the relationships between chlorophyll a concentration and absorption coefficients at fourwavelengths shown in Figs. 6, 7 and 8.
4. DiscussionThe bio-optical data used in this study were obtained
from oceanic regions which are considered to be typicalCase I waters. The mean value of chlorophyll a concen-tration for all stations, 0.27 mg m–3, was similar to thevalue in the SeaBAM (SeaWiFS Bio-optical AlgorithmMini-Workshop) data set of NASA (O’Reilly et al., 1998).Furthermore, the mean value of C in the Southern Oceandata set (0.33 mg m–3) was slightly less than the value inthe eastern Indian sector of the Southern Ocean (0.38 mgm–3) reported by Fukuchi (1980). According to the clas-sification of trophic level in the world ocean suggestedby Antoine et al. (1996), it was considered that our dataset corresponded to oligotrophic and mesotrophic water.Furthermore, Antoine et al. (1996) also reported that eu-trophic water comprised only 2.4% of the world ocean,and Fukuchi (1980) reported a similar percentage of chlo-rophyll a concentrations (2%) greater than 2 mg m–3 inthe Southern Ocean. Although the number of our data waslimited, it may be possible to evaluate the oligotrophicand mesotrophic Case I waters from low to high latitude.
Case I water is elementary to deal with optically(Morel and Prieur, 1977; Gordon and Morel, 1983). How-ever, the bio-optical relationship in the Southern Oceanwas different from the low- and mid-latitude waters (Figs.4 and 5). The trend of the distribution was similar to theresults illustrated by Mitchell and Holm-Hansen (1991)and Mitchell (1992), and indicated the difficulty of deal-ing with bio-optical data in the Southern Ocean and low-and mid-latitude waters as a data set having the samecharacteristics. Moreover, it was suggested that any bio-optical relationships based on the data sets of low- andmid-latitude waters are not suitable for application to theSouthern Ocean. Although Mitchell (1992) and Arrigo etal. (1994) reported the underestimation of pigment con-centrations by the standard algorithm for the CZCS, ourresult suggests that this propensity holds not only for theCZCS algorithm but also for the other band ratio algo-rithms developed using data in the low- and mid-latitudewaters. The underestimation cannot be avoided due to theparallel distribution (Fig. 5) of the Southern Ocean datato lines of the four algorithms (CZCS, Clark’s, OCTS and
258 T. Hirawake et al.
SeaWiFS). Even if the improved algorithms for all datawere applied, they estimate an incorrect value of chloro-phyll a concentration in both waters (RMSE: 0.298–0.442).
Mitchell (1992) hypothesized that a lower pigment-specific absorption coefficient, a*(λ) (m2 mg chl–1), holdsin the polar region compared with low- and mid-latitudewaters, due to significant packaging effect (Mitchell andHolm-Hansen, 1991; Brody et al., 1992; Sosik et al.,1992). This and a low detrital concentration, were thesource of the underestimation of surface chlorophyll aconcentration in the Southern Ocean. The region we sur-veyed was different from the area studied by the aboveauthors, which was concentrated in the relatively coastalwater off the Antarctic Peninsula. However, our resultsalso demonstrated a similar difference in the bio-opticalrelationships and the underestimation using the historicand currently operated algorithms (Fig. 5). Furthermore,the trends were attributed to low absorption coefficientsof detritus or gelbstoff and lower chlorophyll a specificabsorption coefficient of phytoplankton in the SouthernOcean compared with the low- and mid-latitude waters(Figs. 6, 7 and 8). Because the values were arrived at semi-analytically by using the reflectance model, it is a moreanalytical demonstration that the regional difference inabsorption coefficients has an important effect on bio-optical algorithms, such as underestimating the chloro-phyll a concentration in the Southern Ocean.
The absorption coefficients, especially aph(λ ) +ad(λ) + ag(λ), varied in a nonlinear manner with chloro-phyll a concentration, and power function, Eq. (21), areappropriate to represent these variations, as observed inthe results by Bricaud et al. (1995) and Bricaud et al.(1998). aph(443) + ad(443) + ag(443) in the SouthernOcean was lower than the value in the low- and mid-lati-tude waters (Fig. 6), as already mentioned. However, thefitting results for the Southern Ocean (Eq. (23)) were al-most same as the result, of Bricaud et al. (1998) in vari-ous oceanic Case I waters,
Although the constant A (0.1066) in Eq. (22) was approxi-mately twice as great as the results of Bricaud et al.(1998), Eq. (22) was close to the relationship in the CaseI waters reported by Morel (1988):
Kd(440) – Kw(440) = 0.1041 × C0.707, (25)
where Kw is the diffuse attenuation coefficient of purewater. Although Kd(λ) – Kw(λ) ≅ aph(λ) + ad(λ) + bbp(λ) +ag(λ ), where bbp(λ ) is a backscattering coefficient,Kd(λ) – Kw(λ) ≅ aph(λ) + ad(λ) + ag(λ) because ap(λ) +ag(λ) >> bbp(λ). Therefore the difference between Eq. (24)
and Eq. (25) mainly depends on ag(440). Similarly, ourresults in Eqs. (22) and (23) may be influenced by ag(443).The values of ag(443) in the Southern Ocean were smallerthan in the low- and mid-latitude waters. However, thedifference in aph(443) between the low- and mid-latitudewaters and the Southern Ocean (Fig. 8) was also one ofthe factors underlying the difference between the resultsin Eqs. (22) and (23).
While at 555 nm, the constancy of the value of ab-sorption coefficients in the low- and mid-latitude waterscoincided with the trend of Bricaud et al. (1995), whichis revealed in the constants A and B of Eq. (21) close to 0in the green part of the spectrum. In the Southern Ocean,aph(555) had a little dependence on C using the powerfunction. However, the values of aph(555) were widelyscattered with C. If linear relationships between absorp-tion at 555 nm and C were used, they were not signifi-cant. Therefore, the difference of the chlorophyll algo-rithms using the 555 nm band depends almost entirely onthe difference of absorption in the blue part of the spec-trum.
The nonlinear relationships, which represent the de-crease of specific absorption coefficient with increasingchlorophyll a concentration (Stambler et al., 1997), sug-gest that the low specific absorption coefficient will notbe observed in the overall chlorophyll range of the South-ern Ocean. In other words, high values are expected inthe low chlorophyll a concentration regions. Therefore,the cause of the underestimation of the chlorophyll a con-centration in the Southern Ocean with bio-optical algo-rithms could be caused by the low specific absorptioncoefficient in the region compared with the low- and mid-latitude waters under the same chlorophyll a concentra-tion.
The data in this study are consistent with the use ofhistoric and current bio-optical algorithms for the low-and mid-latitude Case I waters. However, those modelscould not be used for the data from the Southern Ocean.The SeaWiFS (490/555) algorithm is statistically supe-rior to any other two-band-ratio algorithm (O’Reilly etal., 1998), and this is supported by the evidence that the490 nm wavelength is less affected by accessory pigmentand CDOM absorption than any other band in the blue(Aiken et al., 1995). However, it was still not possible toavoid the underestimation in the Southern Ocean. Thedifference between the low- and mid-latitude waters andthe Southern Ocean is essentially attributed to the differ-ence of absorption characteristics in the blue spectral re-gion. Thus, an algorithm using a blue-green band ratiomay not cancel the difference. To avoid the underestima-tion, one requires another set of coefficients in the algo-rithms, suitable for the Southern Ocean, such as those ofMitchell (1992). However, switching the coefficients in-duces a discontinuity on the satellite chlorophyll maps
Bio-Optical Relationship of Case I Waters 259
(Denman and Abbott, 1988).Therefore the coefficients of the algorithms should
be switched smoothly between the low- and mid-latitudewater and the Southern Ocean. The algorithm will prob-ably have three dimension, chlorophyll a, band ratio, andany other parameters which can predict a*(λ). a*(λ) can-not be used as a third dimension because it needs the valueof chlorophyll a concentration. We will have to find aparameter that is independent of chlorophyll a concen-tration for accurate estimation in the Southern Ocean fromsatellite ocean color sensors.
AcknowledgementsWe thank the officers and crews of the T/V Umitaka-
maru III and Shinyo-maru, Tokyo University of Fisher-ies, for their cooperation during the cruises. We also thankProf. M. Fukuchi, Dr. K. Watanabe and Dr. T. Odate forproviding the precious data of the 38th JARE. We aregrateful to Dr. K. L. Carder and Dr. F. R. Chen for pro-viding their algorithm code. Dr. G. W. Hosie is duly ac-knowledged for correction of English and improvementof the manuscript, and three anonymous referees for theirhelpful comment and criticism of the manuscript.
This research was financially supported by the ShowaShell Sekiyu Foundation for Promoting of Environmen-tal Research, and the Fund for research in Commemora-tion of the Centennial Anniversary of Tokyo Universityof Fisheries.
ReferencesAcker, J. G. (1994): The heritage of SeaWiFS: A retrospective
on the CZCS NIMBUS Experiment Team (NET) Program.NASA Tech. Memo. 104566, Vol. 21, 44 pp.
Aiken, J., G. F. Moore and P. M. Holligan (1992): Remote sens-ing of oceanic biology in relation to global climate change.J. Phycol., 28, 579–590.
Aiken, J., G. F. Moore, C. C. Trees, S. B. Hooker and D. K.Clark (1995): The SeaWiFS CZCS-type pigment algorithm.NASA Tech. Memo. 104566, Vol. 29, 34 pp.
Allai, K., A. Bricaud and H. Claustre (1997): Spatial variationsin the chlorophyll-specific absorption coefficients ofphytoplankton and photosynthetically active pigments in theequatorial Pacific. J. Geophys. Res., 102, 12413–12423.
Antoine, D., J.-M. André and A. Morel (1996): Oceanic pri-mary production, 2, Estimation at global scale from satel-lite (CZCS) chlorophyll. Global Biochem. Cycles, 10, 57–69.
Arrigo, K. R., C. R. McClain, J. K. Firestone, C. W. Sullivanand J. C. Comiso (1994): A comparison of CZCS and insitu pigment concentrations in the Southern Ocean. p. 30–34. In Case Studies for SeaWiFS Calibration and Valida-tion, Part 1, NASA Tech. Memo. 104566, Vol. 13, ed. by S.B. Hooker and E. R. Firestone.
Austin, R. W. (1974): The remote sensing of spectral radiancefrom below the ocean surface. p. 317–344. In Optical As-pects of Oceanography, ed. by N. G. Jerlov and E. S.
Nielsen, Academic Press, London.Bricaud, A., M. Babin, A. Morel and H. Claustre (1995): Vari-
ability in the chlorophyll-specific absorption coefficientsof natural phytoplankton: Analysis and parameterization.J. Geophys. Res., 100, 13321–13332.
Bricaud, A., A. Morel, M. Babin, K. Allali and H. Claustre(1998): Variations of light absorption by suspended parti-cles with chlorophyll a concentration in oceanic (case 1)waters: analysis and implications for bio-optical models. J.Geophys. Res., 103, 31033–31044.
Brody, E., B. G. Mitchell, O. Holm-Hansen and M. Vernet(1992): Species-dependent variations of the absorption co-efficient in the Gerlache Strait. Antarctic J. U.S., 27, 160–162.
Carder, K. L., F. R. Chen, Z. P. Lee, S. K. Hawes and D.Kamykowski (1999): Semianalytic Moderate-ResolutionImaging Spectrometer algorithms for chlorophyll a andabsorption with bio-optical domains based on nitrate-de-pletion temperatures. J. Geophys. Res., 104, 5403–5421.
Clarke, G. L., G. C. Ewing and C. J. Lorenzen (1970): Spectraof backscattered light from the sea obtained from aircraftas a measure of chlorophyll concentration. Science, 167,1119–1121.
DeGrandpre, M. D., A. Vodacek, R. K. Nelson, E. J. Bruce andN. V. Blough (1996): Seasonal seawater optical propertiesof the U.S. Middle Atlantic Bight. J. Geophys. Res., 101,22727–22736.
Denman, K. L. and M. R. Abbott (1988): Time evolution ofsurface chlorophyll patterns from cross-spectrum analysisof satellite color images. J. Geophys. Res., 93, 6789–6798.
Fukuchi, M. (1980): Phytoplankton chlorophyll stocks in theAntarctic Ocean. J. Oceanogr. Soc. Japan, 36, 73–84.
Garver, S. A. and D. A. Siegel (1997): Inherent optical prop-erty inversion of ocean color spectra and its biogeochemicalinterpretation 1. Time series from the Sargasso Sea. J.Geophys. Res., 102, 18607–18625.
Garver, S. A., D. A. Siegel and B. G. Mitchell (1994): Variabil-ity in near-surface particulate absorption spectra: What cana satellite ocean color imager see? Limnol. Oceanogr., 39,1349–1367.
Gordon, H. R. and D. K. Clark (1980): Atmospheric effects inthe remote sensing of phytoplankton pigments. BoundaryLayer Meteorol., 18, 299–313.
Gordon, H. R. and A. Morel (1983): Remote Assessment ofOcean Color for Interpretation of Satellite Visible Imagery.Springer-Verlag, New York, 114 pp.
Gordon, H. R. and M. Wang (1994): Retrieval of water-leavingradiance and aerosol optical thickness over the oceans withSeaWiFS: a preliminary algorithm. Appl. Opt., 33, 443–452.
Gordon, H. R., D. K. Clark, J. W. Brown, O. B. Brown, R. H.Evans and W. W. Broenkow (1983): Phytoplankton pigmentconcentrations in the Middle Atlantic Bight: comparison ofship determinations and CZCS estimates. Appl. Opt., 22,20–36.
Gordon, H. R., O. B. Brown, R. H. Evans, J. W. Brown, R. C.Smith, K. S. Baker and D. K. Clark (1988): A semianalyticradiance model of ocean color. J. Geophys. Res., 93, 10909–10924.
260 T. Hirawake et al.
Hooker, S. B., W. E. Esaias, G. C. Feldman, W. W. Gregg andC. R. McClain (1992): An overview of SeaWiFS and oceancolor. NASA Tech. Memo. 104566, Vol. 1, 24 pp.
Hooker, S. B., C. R. McClain and A. Holmes (1993): Oceancolor imaging: CZCS to SeaWiFS. Mar. Technol. Soc., 27,2–15.
Hovis, W. A., D. K. Clark, F. Anderson, R. W. Austin, W. H.Wilson, E. T. Baker, D. Ball, H. R. Gordon, J. L. Mueller,S. Z. El-Sayed, B. Sturm, R. C. Wrigley and C. S. Yentsch(1980): Nimbus-7 Coastal Zone Color Scanner: Systemdescription and initial imagery. Science, 210, 60–63.
Ishimaru, T., H. Otobe, T. Saino, H. Hasumoto and T. Nakai(1984): OCTOPUS, an octo parameter underwater sensor,for use in biological oceanography studies. J. Oceanogr.Soc. Japan, 40, 207–212.
Kishino, M., T. Ishimaru, K. Furuya, T. Oishi and K. Kawasaki(1998): In-water algorithms for ADEOS/OCTS. J.Oceanogr., 54, ADEOS field campaign off Sanriku/NorthPacific, 431–436.
McClain, C. R., W. E. Esaias, W. Barnes, B. Guenther, D.Endres, S. B. Hooker, B. G. Mitchell and R. Barnes (1992):SeaWiFS calibration and validation plan. NASA Tech. Memo.104566, Vol. 3, 41 pp.
McClain, C. R., M. L. Cleave, G. C. Feldman, W. W. Greg, S.B. Hooker and N. Kuring (1998): Science Quality SeaWiFSdata for global biosphere research. Sea Technol., Sep., 10–16.
Mitchell, B. G. (1992): Predictive bio-optical relationships forpolar oceans and marginal ice zones. J. Mar. Sys., 3, 91–105.
Mitchell, B. G. and O. Holm-Hansen (1991): Bio-optical prop-erties of Antarctic Peninsula waters: differentiation fromtemperate ocean models. Deep-Sea Res., 38, 1009–1028.
Mitchell, B. G. and D. A. Kiefer (1988): Chlorophyll a specificabsorption and fluorescence excitation spectra for light-lim-ited phytoplankton. Deep-Sea Res., 35, 639–663.
Mobley, C. D. (1994): Light and Water; Radiative Transfer inNatural Waters. Academic Press, San Diego, 592 pp.
Morel, A. (1988): Optical modeling of the upper ocean in rela-tion to its biogenous matter content (Case I waters). J.Geophys. Res., 93, 10749–10769.
Morel, A. and Gentili, B. (1993): Diffuse reflectance of oce-anic waters. II. Bidirectional aspects. Appl. Opt., 32, 6864–6879.
Morel, A. and L. Prieur (1977): Analysis of variations in oceancolor. Limnol. Oceanogr., 22, 709–722.
Müller-Karger, F. E., C. R. McClain, R. N. Sambrotto and G.C. Ray (1990): A comparison of ship and Coastal Zone ColorScanner mapped distribution of phytoplankton in the south-eastern Bering Sea. J. Geophys. Res., 95, 11483–11499.
Neckel, H. and D. Labs (1984): The solar radiation between3300 and 12500 Å. Solar Phys., 90, 205–258.
O’Reilly, J. E., S. Maritorena, B. G. Mitchell, D. A. Siegel, K.L. Carder, S. A. Garver, M. Kahru and C. McClain (1998):Ocean color chlorophyll algorithms for SeaWiFS. J.Geophys. Res., 103, 24937–24953.
Parsons, T. R., Y. Maita and C. M. Lalli (1984): A Manual ofChemical and Biological Methods for Seawater Analysis.Pergamon Press, New York, 173 pp.
Pope, R. M. and E. S. Fry (1997): Absorption spectrum (380–700 nm) of pure water. II. Integrating cavity measurements.Appl. Opt., 36, 8710–8723.
Porra, R. J., W. A. Thompson and P. E. Kriedemann (1989):Determination of accurate extinction coefficients and simul-taneous equations for assaying chlorophylls a and b ex-tracted with four different solvents: verification of the con-centration of chlorophyll standards by atomic absorptionspectroscopy. Biochim. Biophys. Acta, 975, 384–394.
Sathyendranath, S., L. Prieur and A. Morel (1989): A three-component model of ocean colour and its application to re-mote sensing of phytoplankton pigments in coastal waters.Int. J. Remote Sens., 10, 1373–1374.
Sathyendranath, S., F. E. Hoge, T. Platt and R. N. Swift (1994):Detection of phytoplankton pigments from ocean color:improved algorithms. Appl. Opt., 33, 1081–1089.
Smith, R. C. and K. S. Baker (1981): Optical properties of theclearest natural waters (200–800 nm). Appl. Opt., 20, 177–184.
Sosik, H. M., M. Vernet and B. G. Mitchell (1992): A compari-son of particulate absorption properties between high- andmid-latitude surface waters. Antarctic J. U.S., 27, 162–164.
Stambler, N., C. Lovengreen and M. M. Tilzer (1997): The un-derwater light field in the Bellingshausen and AmundsenSeas. Hydrobiol., 344, 41–56.
Suzuki, R. and T. Ishimaru (1990): An improved method forthe determination of phytoplankton chlorophyll using N,N-Dimethylformamide. J. Oceanogr. Soc. Japan, 46, 190–194.
