Full report

Original page 1

NASA Technical Memorandum SeaWiFS Technical Report Stanford B. Hooker, Editor Goddard Space Flight Center Greenbelt, Maryland Elaine R. Firestone, Technical Editor General Sciences Corporation Laurel, Maryland Volume 29, The SeaWiFS Pigment Algorithm James Aiken and Gerald F. Moore Plymouth Marine Laboratory Plymouth, United Kingdom Charles C. Trees San Diego State University San Diego, California National Aeronautics and Space Administration eoddard Space Flight Center Greenbelt, Maryland 20771 1995 104566, Vol. 29 Series CZCS-Type Stanford B. Hooker NASA Goddard Space Flight Center Greenbelt, Maryland Dennis K. Clark NOAA/National Environmental Satellite Data Information Service Camp Springs, Maryland

Original page 2

This publication is available from the NASA Center for AeroSpace Information, 800 Elkridge Landing Road, Linthicum Heights, MD 21090-2934, (301) 621-0390.

Original page 3

J. Aiken, G.F. Moore, C.C. Trees, S.B. Hooker, and D.K. Clark ABSTRACT The Sea-viewing Wide Field-of-view Sensor (SeaWiFS) mission will provide operational ocean color that will be superior to the previous Coastal Zone Color Sensor (CZCS) proof-of-concept mission. An algorithm is needed that exploits the full functionality of SeaWiFS whilst remaining compatible in concept with algorithms used for the CZCS. This document describes the theoretical rationale of radiance band-ratio methods for determining chlorophyll a and other important biogeochemical parameters, and their implementation for the SeaWIFS mission. Pigment interrelationships are examined to explain the success of the CZCS algorithms. In the context where chlorophyll a absorbs only weakly at 520 nm, the success of the 520 nm to 550 nm CZCS band ratio needs to be explained. This is explained by showing that in pigment data from a range of oceanic provinces chlorophyll a (absorbing at less than 490 nm), carotenoids (absorbing at greater than 460 nm), and total pigment are highly correlated. Correlations within pigment groups particularly photoprotectant and photosynthetic carotenoids are less robust. The sources of variability in optical data are examined using the NIMBUS Experiment Team (NET) bio-optical data set and bio-optical model. In both the model and NET data, the majority of the variance in the optical data is attributed to variability in pigment (chlorophyll a), and total particulates, with less than 5% of the variability resulting from pigment assemblage. The relationships between band ratios and chlorophyll is examined analytically, and a new formulation based on a dual hyperbolic model is suggested which gives a better calibration curve than the conventional log-log linear regression fit. The new calibration curve shows the 490:555 ratio is the best single-band ratio and is the recommended CZCS-type pigment algorithm. Using both the model and NET data, a number of multiband algorithms are developed; the best of which is an algorithm based on the 443:555 and 490:555 ratios. From model data, the form of potential algorithms for other products, such as total particulates and dissolved organic matter (DOM), are suggested. 1. INTRODUCTION uct derived from CZCS imagery, termed the CZCS Pigment Algorithm product. There is no predetermined con- As a second-generation ocean color instrument, the Sea- sensus for the rationale or definition of this product (chloviewing Wide Field-of-view Sensor (SeaWiFS) offers a vari- rophyll a, photosynthetic pigments, or total pigments). A ety of design improvements over its predecessor the Coastal choice must be made at the outset, so a methodological Zone Color Scanner (CZCS). The design of the SeaWiFS approach can be determined and described. The proposed instrument was driven by science requirements as defined approach should be essentially empirical, and use band raby the SeaWiFS Prelaunch Science Working Group (SP- tios in common with the original CZCS algorithms. There SWG). The SPSWG was an ad hoc committee selected is a strong desire in the community, for example, the Bioby the National Aeronautics and Space Administration Optical Algorithm Working Group (BOAWG), that there (NASA) Headquarters for the purpose of providing NASA should be a SeaWiFS product compatible with the CZCS with guidance in the formulation of mission objectives, imagery, so retrospective processing can be applied and specifications, and goals. The SPSWG has specifically ex- comparability between CZCS and SeaWiFS data can be pressed a requirement for continuity between CZCS and achieved. SeaWiFS products. To derive global bio-mass and productivity trends on Consequently, the SeaWiFS Project Office (SPO) plans decadal time scales, it is possible the standard NASA CZCS to produce three groups of level-2 derived products: Sea- two-band algorithm, used for the global processing, could WiFS baseline, CZCS-type, and potential SeaWiFS prod- be used unaltered. It may be that an algorithm using ucts. A differentiation is made between CZCS-type pig- all three bands at 443, 520, and 550 nm, however, would ment and SeaWiFS baseline chlorophyll-like pigment con- give a more statistically robust relationship; these bands centrations. The SeaWiFS semianalytical algorithm for are close to the SeaWiFS 443, 510, and 555 nm bands and chlorophyll a will be developed using analytical and semi- continuity would seem likely in this case. analytical models. Chlorophyll a is the parameter that has This hypothesis is flawed logically in a number of rebeen chosen as it is regarded universally as the most ap- spects, primarily because of the differences between the propriate measure of viable phytoplankton biomass (i.e., CZCS and SeaWiFS instruments (Hooker et al. 1993). Seaphytoplankton which are growing actively and capable of WiFS will have precision radiometry (10 bit resolution), growth). precision calibration (prelaunch and onboard), stability of The CZCS-type pigment product is supposed to pro- calibration monitoring, and established vicarious calibravide some form of continuity with the total pigment prod- tion schemes--CZCS had nothing comparable. SeaWiFS

Original page 4

TheSeaWiFSCZCS-TypePigmentAlgorithm nm)which contribute to over 95% of the total pigment biomass. Most will havetwoinfrared(IR) bands(765and865 whereasof these have an important influence on water color at the will allowfor precisionatmosphericcorrection, CZCSatmosphericcorrectionwaslimitedandoftenfailed wavelengths of the SeaWiFS bands. An accurate assessatmosphericment would involve the sum of the optically weighted con- (manyoftheassumptionsnecessaryforCZCS bandwas tribution of the main pigments present for different natural correctionwereinvalid,e.g.,oftenthe 670nm not truly a zerowater-leavingradianceband);SeaWiFSphytoplankton assemblages in different geographical sites correctedand seasons. will haveat leastfiveprecisionatmospherically bands(whereasCZCShadthree)offeringa greaterpoten- A comment on pigment measurement techniques is aptial for multibandalgorithmsof widespreadapplicability.propriate here. Much of the insight into the composition In fact,the onlystrictly comparablefeaturebetweenthe and significance of differing phytoplankton pigments has CZCSandSeaWiFSinstrumentsisthecommonblueband come as a result of the development of analyses using at 443nm. the high performance liquid chromatography (HPLC) tech- Theradicaldifferencebetweenthe atmosphericcorrec-nique (Mantoura and Llewellyn 1983 and Trees et al. 1985), tionschemesmaymeanthat comparabilitywill belimited. which is the recommended methodology in the Ocean Op- It is likelythe bestSeaWiFSband-ratioalgorithmswill tics Protocols for SeaWiFS Validation (Mueller and Austin usethree,four,or fivevisiblebandsnot compatiblewith 1995). CZCS:threebandsusing412,443,and555nmor 443,490, Along with the insight on pigments for some has come and555nm;four bandsusing412,443,490(or 510),and confusion for others, with earlier reports that concentra- 555nm;or fivebandsusing412,443,490,510,and555nm. tions of pigments determined by HPLC and fluorescence TheBOAWGteamhasagreedthatbesidestheCZCS-type(Trees et al. 1985) differed markedly--much lower pigment pigmentalgorithm,thereshouldbecontinuedresearchto concentrations were obtained using the HPLC technique. identifythebestpossibleSeaWiFSpigmentalgorithm. A thorough investigation by Trees et. al (1995) has shown Theobjectiveof this studyis to providea band-ratiothat when each method is applied rigorously, each yields algorithmthathasthebestpossiblecontinuitywithCZCS about one-to-one (=h10%) relationships for chlorophyll a measurements.Thefocusof attention,however,is onthe in most bio-optical provinces. Errors can arise, however, derivationofthebestpossibleband-ratioalgorithmforthe if the protocols for sampling, filtration, extraction, and retrievalofphytoplanktonpigmentsfromSeaWiFSobser- calibration are not adhered to strictly; exceptions occur vations.Thedesireis to achievethesegoalsbasedon a if either chlorophyll b or chlorophyll c are atypically high soundtheoreticalbasisandrationale. 2. PIGMENTS REVISITED The product from the CZCS pigment algorithm was chlorophyll a plus phaeopigment concentration, Ca + Cp (as determined by the fluorometric method), which were considered the main absorbing agents of biogenic origin by the NIMBUS Experiment Team (NET). The reasons the choice of the Ca + Cp parameter was partly historical and partly methodological. Prior to 1980, the principle the 1) Chlorophylls a, b, and c; methods for the determination of chlorophyll a were (greater than about 5% of the total). Recently, there have been several studies investigating the contribution of the major phytoplankton pigments to light absorption in the oceans, including the independent analyses by Bidigare et al. (1990) as well as Hoepffner and Sathyendranath (1993). Although the latter study was restricted to samples from the Georges Bank area, both studies showed the major pigments that need to be included to for account for 95% of the light absorbed are relatively few in number: tri-chrometric spectrometric method (Strickland and Par- 2) The photosynthetic carotenoids (PSC); and sons 1972) and the fluorescence method (Yentsch and Menzel 1963). Both methods could be imprecise (e.g., at low concentrations) and frequently under- and over-estimated 3) The photoprotectant carotenoids (PPC). In the yellow-orange part of the spectrum (at around interfering 550nm), the phycobilin, phycoerythrin, and phycocyanin both Ca and Cp due to the presence of other pigments, notably chlorophyll b and chlorophyll c. It is now accepted by most biological oceanographers pigments are moderately important light absorbers. These pigments occur mostly in cyanobacteria, which are generthat phaeopigments rarely exceed 3-8% of the total pig- ally unimportant in the surface layers of the ocean relment concentration in the surface layer of the ocean, with evant to this study; the only known instances of signiffor icance correspond to blooms of cyanobacteria occurring the exception of a few well understood circumstances, localized. in upwelling regions. Other, taxa-specific, tag-pigments example, when zooplankton grazing is high and Furthermore, there are a number of other photosynthetic are also insignificant to the total light absorption in the and photoprotectant pigments which co-exist, co-vary, and ocean--individually or collectively they account for less absorb at the same wavelengths as chlorophyll a, and which than 5% of the absorption. Both Bidigare et al. (1990) occur in significant concentrations. Individually, they may and Hoepffner and Sathyendranath (1993) give tables of account for approximately 5-50% of the total pigment con- the specific absorption coefficients of these major pigment centration and, in combination with chlorophyll a, may groups. Although there is general agreement between these 2

Original page 5

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark Table 1. Totalpigmentconcentration(CTp) to chlorophyll a (Ca) relationships. CTp R 2 0.021 + 2.17Ca 95.5 -0.040 + 2.30Ca 95.9 0.031 + 2.16Ca 93.2 Table 2. Global total pigment regressions_. Region Ca R 2 C b R 2 Cc Antarctic 0.544 97.3 0.006 35.7 0.154 NEAT 89-90 0.422 99.6 0.016 76.6 0.134 NEAT 915 0.519 98.6 0.024 27.1 0.020 GIN Seas 0.367 96.6 0.085 74.6 0.103 Georges Bank 0.525 98.8 0.066 84.4 0.051 Bermuda (BATS) 0.499 99.6 O.026 52.6 0.036 EqPac 0.446 99.7 0.074 93.6 0.042 Global 0.475 0.042 0.077 Regression assumes no intercept. two different studies, there is some divergence concerning the coefficients in the blue spectral region. Figure la, from Bidigare et al. (1990), shows the weightspecific absorption coefficient for the major pigment groups, and Fig. lb shows the equivalent figure from Hoepffner and Sathyendranath (1993). Figure 2 shows the actual pigment absorption for the major pigments and the summed total for measurements from a cruise to the Northeast Atlantic pigment interrelationships for seven widely differing bio- (NEAT) in June 1991 (Holligan et al. 1993). For the latter, geochemical (bio-optical) provinces: the phytoplankton assemblage was dominated by coccolithophores and small flagellates, yet these data are similar to the earlier studies with the total pigment absorption dominated by chlorophyll a, PSC, and PPC--chlorophyll b and chlorophyll c have less than 5% significance. Bidigare et al. (1990) concluded accessory pigments do not always co-vary with chlorophyll a over depth and time. In this study, the relationships of accessory pigments to chlorophyll a for the surface layers only, sensed by satellite color imagers, are examined using data from a wide variety of sources (published and unpublished). An examination of the relationship between chlorophyll a and total pigments (sum of chlorophylls a, b, c, PSC, and PPC) shows a robust N Comments 670 All data. 630 Ca < 3mgm -3 416 z < 10m R 2 Cabc R 2 Cpp R 2 Cps R 2 79.6 0.295 98.8 0.043 45.4 0.252 96.2 99.5 0.427 99.3 0.049 83.5 0.377 99.5 86.3 0.436 97.4 0.122 80.5 0.313 92.5 88.8 0.443 95.0 0.096 68.8 0.346 94.9 98.2 0.358 98.5 0.104 96.9 0.253 97.7 89.7 0.437 99.4 0.231 98.7 0.207 98.8 91.4 0.436 99.2 0.249 94.5 0.186 97.1 0.407 0.128 0.276 :_NEAT 1991 is coccolithophore bloom data. ferent in the five SeaWiFS blue and green bands (412, 443, 490, 510, and 555 nm). 2. The relative composition of carotenoids (e.g., PSC) in different biogeographical (bio-optical) provinces is not constant and varies significantly from cruise to cruise and province to province. This is demonstrated here by a thorough analysis of all • Greenland, Iceland, and Norwegian (GIN) Seas 1986-87 (Trees), • Georges Bank (Hoepffner and Sathyendranath 1992), • NEAT 1989-90 (United Kingdom) Biogeochemical Ocean Flux Study (BOFS), • NEAT 1991 BOFS (Trees), • Equatorial Pacific (EqPac) 1992 (Spring and Fall), • Antarctica 1992 BOFS (Sterna), and • Bermuda Atlantic Time-Series Station (BATS). relationship (97% of the variance explained). Nearly 5,600 Note that the total pigment to chlorophyll a relationships pigment determinations from many bio-optical provinces for all combined data are highly correlated as shown in were used in the analysis (see Fig. 3). Province by province Table 1, with the major pigment relationships shown in and cruise by cruise, the ratio of total pigment to chloro- Fig. 4. Examining the data province by province, it is eviphyll a varied from 1.876-2.876 with a mean of 2.164. dent there are widely varying ratios of the concentrations of The conclusion from this analysis could be that it mat- chlorophylls a, b, c, PPC, and PSC (Ca, Cb, Co, Cpp, and ters little whether the algorithm product is chlorophyll a or Cps , respectively) to total pigment concentration (CTp) total pigments, since the relationship between these two as shown in Table 2 and Fig. 5a-g. measures of marine phytoplankton biomass, on a global This analysis shows that for chlorophyll a, chlorophyll c, level, are so tightly coupled. Two issues make this hy- total carotenoids, and PSC, the intraprovince covariance pothesis invalid: is extremely tight for all provinces (R 2 is 96-99% for chlo- 1. The optical influence of the different pigment groups rophyll a to CTp), although the coefficients of variance dif- (e.g., the chlorophylls and carotenoids) are quite dif- fer: notably, the fraction of chlorophyll a is lowest in the 3

Original page 6

TheSeaWiFSCZCS-Type Pigment Algorithm 0.12i===.q ! 0.1 E 04 E ,_ 0.08 C O e_ e_ 0.06 L_ O .O ,c( 0.04- 0 // e_mm if • 0.02o Q. 0:lz::_ 400 450 Wavelength [nm] 0.12 0.1 E 04 E 0.08 C O =i e_ 0.06 t._ 0 m 0.04 = J_ @ 0.02 Q. I 0 .J,. 400 450 Wavelength [nm] Key , _ C a cb c_ ....... Cps -- Cpp A ! 600 650 700 750 Key --ca -- cb _ c_ ....... Cos+ cpp \11 550 600 650 700 750 Fig. 1. Pigment-specific absorption versus wavelength from a) Bidigare et al. (1990) and b) Hoepffner and Sathyendranath (1993), top and bottom plots, respectively.

Original page 7

J. Aiken, G.F. Moore, C.C. Trees, I s :'. _ :. e E 0.04- .1 _ ••co•o•• " : • ._ 0.03- : ..°. °* .... "P% "_ "_ 0.01 ".',e 0.02- _ ""-.. ' 0 S.B. Hooker, and D.K. Clark Key C a Cps _ C b - - Cpp _ Cc .... Cre 400 450 500 550 600 650 700 750 Wavelength [nm] Fig. 2. Simulated pigment absorption. 12. 9-' 6. + O° 0 2 Ca [mg m "3] "1- ++ 4- 4- :t + +4+ +._ + % % :lq +++ + ++ 3 Fig. 3. Chlorophyll a versus total pigments.

Original page 8

The SeaWiFS CZCS-Type Pigment Algorithm a) I0.00 i r .' i • o,, i I • ? 1,oo ..........................i............... D: • : • t, b) 10.00 ? 1.00 0.10 ............................. o.1o 0.01 I I 0.01 0.10 1.00 10.00 -3 Chlorophyll • rng_ c) 10.00 I I * " E • J' ! !.." 1,00 _ q, • 0.10 l/ ........................................................... 0.01 / I I 0.01 0.10 1.00 10.00 -3 Oarotenolds morn e) 10.00 v i • • ._ , 1.00 .......................... i .................. _ .......................... E 0.10 ...... ",'.......................... ;".......................... el_o • . 0.01 I 1 001 0.10 1.00 1000 -3 Photosynthetic Carotenolda mg.m 2 0,01 0.01 0.10 1.00 10.00 -3 Chlorophyll a n_m d) 10.00 • • • 0:- 1,00 t ..; :. " 0.10 o /" i 0.01 0,01 0.10 1.00 10.00 -8 Total CNorophyll morn f) 1o.00 :i" ................... ?§ 1.oo •"-" 4:. O.lO # 0.01 I 0.01 0.10 1.00 1000 -8 Photolxotectant Carotenok_ mQm Fig. 4. Pigment regressions: a) CTp --: 0.134 1.96Ca (R2=93.2%), b) Cabc = 0.02+1.20Ca (R2=97.6%), 2 o c) CTp = 0.13 + 2.27[Cpp + Cp_I (R --93.7_), d) Cpp + Cps = 0.12 + 0.76Cabc (R2=77.5%), e) Cpp + Cps -- 0.03 + 1.15Cps (R =97.5%), and f) Cpp + Cps -_- 0.41 + 3.67Cpp (R2=56.2%).

Original page 9

L J. Aiken, G.F. Moore, C.C. Trees, S.B. Hooker, and D.K. Clark 0 0 0 0 0 0 o o o o d m m m _ n i E I I)l o 0 m m tO o If) o to 0 _ v- v- o d o d d o qg lal9 , )) m )) m +6 iN m +) m , m m m • m to 0 tO 0 tO 0 to 0 _ 0 to 0 d d o d d o o d d d Vg la, 9 9 ! d_D .... _ o m + d_ or) 2d m m_. : ! iN c_E I I u _- 0 I I _ "_Z o _ +6 -- im _'_z o -.z r2 0 0 0 0 0 0 •,- d o o o --< "'9/ (aa9+_a9) /

Original page 10

The SeaWiFS CZCS-Type Pigment Algorithm GIN Seas (0.367), and highest in Antarctica (0.514). The 3. BAND-RATIO ALGORITHMS low fraction in the GIN Seas is partly compensated by the high fraction of chlorophyll b (0.085) and chlorophyll (0.103). In contrast, the fraction of carotenoids, mostly PSC, is lowest in Antarctica (0.295) and relatively constant elsewhere; relative to chlorophyll a, there are substantially more carotenoids in the GIN Seas than anywhere else and only here are they the major pigment group and exceed the fraction for chlorophyll a. The global mean fraction of chlorophyll a is close to 0.5 (depressed only by the GIN Seas data), whereas, the global mean fraction of total carotenoids is about 0.4. The global mean fractions of chlorophylls b and c are 0.042 and 0.077, respectively, and are most significant in the GIN Seas. Chlorophyll b is practically insignificant in the Antarctica, NEAT, and the BATS data. The global mean fraction of PPC is 0.128, over 20% of the total pigment composition in BATS and EqPac data (both of which are high light) and relatively low in the Antarctica (0.043) and NEAT 89-90 data, both of which are low light environments. The approach adopted here, in line with the tried and c tested methods used for CZCS, is essentially empirical, although, as is the case with the mainstream semianalytical SeaWiFS algorithm for chlorophyll a, the methods are validated by recourse to standard bio-optical, hydrological models. By this means, semi-empirical algorithms are developed, whereby analytical methods are used to propose the form of the algorithm, while empirical methods are used to determine the numerical coefficients. The two-band (blue to green ratio) algorithms used for CZCS, developed by empirical methods (Clark 1981) or semianalytical models (Gordon et al. 1988), were in fact remarkably successful for processing and interpreting CZCS imagery, dependant as they were, on high accuracy in-water measurements of normalized water-leaving radiances, LWN(A), and concurrently determined pigment concentrations. Any errors in the historical CZCS pigment databases are a result of the sensor measurements (radiometric accuracy and stability, as well as atmospheric The extremes of the interpigment variance, from pro- correction procedures) and not a result of the algorithms vince to province, is highlighted in Table 3 and Figs. 5g-i, which shows the ratio of total carotenoids (Cpp + Cps) to chlorophyll a (Ca), and to total chlorophyll (Cabc). The latter being the sum of chlorophylls a, b, and c. The lowest ratio (0.522) is found in Antarctica and the highest ratio employed. The NET database of water-leaving radiances and in situ pigment measurements, although restricted in its geographical and seasonal coverage, is still the major reliable source of data for algorithm development and is used in this paper for these reasons. (1.129) is found in the GIN Seas. These extremes are re- Recently, the NET database has been reworked for the duced somewhat if the carotenoid to total chlorophyll ratio is considered (0.416 for Antarctica and 0.738 for the GIN Seas). Table 3. Global interpigment ratiost. Cpp'-Cps Cpp'+Cps Region R 2 R 2 C Cabc Antarctic 0.522 94.3 0.416 97.9 NEAT 89-90 0.835 94.8 0.735 97.2 NEAT 91 0.814 92.9 0.745 91.9 GIN Seas 1.129 86.3 0.738 84.5 Georges Bank 0.665 95.0 0.549 96.3 BATS 0.824 98.8 0.715 98.3 EqPac 0.911 98.2 0.711 97.2 Global 0.814 0.658 t Regression assumes no intercept. Using the radiances LWN(443) and LwN(550), these differences in the relative pigment concentrations would lead to significant differences in the coefficients of a standard two-band algorithm for each of these different biogeochemical (bio-optical) provinces, because of the different contributions to the optical absorption from each pigment at 443 and 550 nm. It is likely that multiband algorithms (3, 4, and 5 wavelengths) designed to account for the influences of different pigments on the absorption at each waveband, may produce algorithms that would have more widespread application. SeaWiFS wavelengths and bandwidths using binomial or polynomial curve fitting procedures to generate a range of empirical algorithms, which successfully explain high percentages of the variance between the variables. All of these are completely satisfactory as SeaWiFS algorithms, and with the greater precision of the SeaWiFS sensor, they should provide accurate interpretations of the imagery (after atmospheric correction). These algorithms are listed in Table 4. Using NASA Airborne Oceanographic Lidar (AOL) data from the North Atlantic Bloom Experiment (NABE) in May 1989 (Hoge and Swift 1993) to simulate SeaWiFS data, Aiken et al. (1992) demonstrated that combinations of two-band ratios could successfully explain a greater percentage of the variance between pigment and radiance ratios than any single two-band ratio on its own (Table 5). Similar results were obtained by Aiken and Moore (1995), using data from a bio-optical model with the values for absorption and scattering coefficients taken from the literature. Algorithms for the other constituents of the water column could be derived from the same synthetic data (see Table 6). In the following section, an attempt is made to justify the validity of band-ratio algorithms, employing simple two-band ratios or linear combinations of more than two ratios. These can account for a greater percentage of the variance between variables in the NET data, because each two-band ratio can be related to a specific property of the bio-optical assemblage, which can be demonstrated by recourse to simple bio-optical models.

Original page 11

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark Table4. AlgorithmsfromD. Clark. Parameter Case Ratio Ao A1 A2 A3 R2 CTp 1 2 + 3 + 4:5 8.73 -11.20 4.43 -0.62 97.0 1 + 2 2 + 3 + 4:5 4.76 -5.36 1.70 -0.21 93.6 CTP CT 1 2+3+45 8.00 -9.99 3.77 -0.50 95.8 4.74 -5.37 1.65 -0.18 92.2 CT 1+2 2+3+4:5 TSM 1 2 + 3 + 4:5 4.81 -7.56 3.32 -0.50 72.2 TSM 1 + 2 2 + 3 + 4:5 5.57 -8.64 3.83 -0.58 84.6 ISM 1 2 + 3 + 4:5 11.20 -18.70 8.77 -1.37 60.1 ISM 1 + 2 2 + 3 + 4:5 4.81 -7.56 3.32 -0.50 72.2 OSM 1 2 + 3 + 4:5 4.79 -8.02 3.75 -0.60 74.1 OSM 1 + 2 2 + 3 + 4:5 4.39 -7.29 3.35 -0.53 83.6 3.56 -4.53 1.44 0.15 86.6 c(535) 1 + 2 2 + 3 + 4:5 c(535) 1 + 2 4:5 -0.12 Note: All are log10 regressions. -1.74 81.6 ['able 5. AOL algorithms (Aiken et al. 1992' of the form a(Lu(A1):Lu(Ae))_(Lu(A3):Lu(A4)) . A1 A2 A3 A4 a 410 555 0.78 440 555 2.02 490 555 2.51 440 555 410 440 1.88 440 555 440 490 2.69 fl 3` R 2 N -2.90 64.1 753 -2.84 86.3 753 -2.77 92.8 753 -2.47 2.71 86.8 733 -2.64 4.04 93.7 733 Table 6. Model algorithms (Aiken and Moore 1994) of the form ((R(A1):R(A2))Z(R(Aa):R(A4)) _ Algorithm A1 A2 A3 A4 Chlorophyll 412 555 443 555 490 555 443 555 412 443 490 555 412 555 DOC 412 555 443 555 490 555 443 555 412 443 490 555 412 555 a /3 3' R 2 1.87 -1.87 69 1.53 -2.15 89 1.77 -2.71 95 0.52 -2.41 -5.20 95 1.05 -4.66 -5.37 69 1.44 -0.92 28 1.04 -0.56 12 0.97 -0.51 6 1.33 -1.10 -10.2 90 -1.67 3.51 -14.8 77 4. BIO-OPTICAL MODELS the Fresnel reflectance at normal incidence, _ is the Fresnel reflectance for sun and sky irradiance, r is the air-water reflectance for diffuse irradiance, and Q is the ratio of up- 4.1 Analytical Basis of Band-Ratio Models welling irradiance to radiance, which varies with the an- Water-leaving radiance is a function of the downwelling gular distribution of the upwelling light field, and is equal light field, the interface effects, and the inherent optical to r for an isotropic distribution. The (1 - p)(1 - 5)n-2 properties (IOPs) of the water column constituents inte- term gives the effect of the air-water interface, and shows grated over 1-2 optical depths. Analytically, it can be a weak relationship with wavelength, varying as the refractive index of water. The term 1 - rR can be assumed to expressed as be unity in Case-1 waters. [(1- p)(1- )R] Assuming the interface term is constant, the ratio of LWN = F0 [ n( ---rR)-Q J (1) remotely sensed water-leaving radiances at wavelengths Ai and )'5, respectively, is expressed as where F0 is the extraterrestrial irradiance, n is the refractive index of seawater, R is the irradiance reflectance, p is LO = R(A,)Q(£j)Fo(A,) (2) R(Aj)Q(A,)Fo(Aj)' 9

Original page 12

TheSeaWiFSCZCS-TypePigmentAlgorithm wherethe expressionL,:j is a shorthand form for the ratio chlorophyll biomass absorption can be partitioned accord- Lw_v(X,) /LwN(Aj). ing to the functional groups of pigment. This includes Q shows a weak relationship with wavelength, which chlorophyll a, b, and c; the photosynthetic carotenoids; is analytically difficult to determine, the main factor be- and the photoprotectant carotenoids as per the following (Morel formulation: ing the relative change in scattering phase function and Gentili 1991 and 1993). The main determinant of the radiance ratio is the irradiance reflectance R. This may expressed as = (3) [ a(A) J' where G(#0, A) represents the effect of the downwelling be Ca¢ = a_Ca + %C b + acCc, (6) -[- apsCps + appCpp where Ca, C b, Co, aa, a b, and ac are the concentrations and specific absorptions of chlorophyll a, b, and c, respectively; Cps, Cpp, aps, and app are the concentrations and specific absorptions of the photosynthetic and photolight field; bb(A) is the backscatter coefficient; and a(A) the protectant carotenoids, respectively. absorption coefficient. The IOPs, a(A) and bb(A), are the It can be seen from this that band-ratio algorithms are sum of the optical properties of pure seawater and the opti- almost wholly dependant on the IOPs. The CZCS algocally active water column constituents, i.e., chlorophyll (a, rithm, or the SeaWiFS equivalent, can be taken as a case b, and c), carotenoids, dissolved organic matter (DOM), study. The A,:555nm algorithms (where A_ -- 412, 443, and detrital particulates. 490, or 510 nm) can be approximated by the following ex- Substituting for R, the normalized water-leaving radi- pression: ance ratio, Li:j, is expressed as (4) a_(A_) + ag(A_)G + aab_(Ai)C L,:j = g[a(Aj)bb(Ai)F°(A_) ] La( Adbb(Aj)Fo( A_)J ' the bbp(Aj)P F0(A3)" where g is assumed to be a constant that consists of ratios of the air-sea interface effects, the effects of the light field [the Morel and Gentili (1991) f factor], and the relative spectral variation of Q. Using Morel and Gentili's (1993) figures for the spectral variation of f/Q, g will be 1 + 3.5% with the remotely sensed viewing geometry. For the purposes of discussion of the effects of the water constituents, the factor g has been omitted (it is assumed be unity). By partitioning the IOPs of the constituents of the water into the sum of the parts, Li:j can be expressed as aw(Aj) + ag(Aj)G + acz(Aj)C L_:f a_(A,) + %(A,)G + a,(A,)C (5) sorption, and ag(443) is approximately 0.3 x a¢(443). If bbw(Ai) + bbp(Ai)P F0(Ai) × × -bbw()j) + bbp()j)P Fo()j)' where aw and bbw are the absorption and backscatter cocon- (8) efficients of water, respectively; P is the particulate centration including detrital material, and bbp is its specific backscatter coefficient (normally normalized to chlo- or rophyll a concentration); G is the concentration of DOM and + CA ' (9) and DOM-like absorbers and ag its specific absorption; C is the chlorophyll biomass concentration and a¢ is its specific absorption. In this formulation, the backscatter has been decoupied from the chlorophyll concentration, and it is assumed that the relationship between the backscatter and phytoa(Aj) + ag(Aj)G + aabc(Aj)C Li: j (7) bb (A,) + × × -- The particulate backscatter, bbp, will come from both detrital material and phytoplankton; this backscatter, although correlated with chlorophyll biomass, shows a highly variable relationship from province to province. For all but the most oligotrophic waters, bbp is greater than bbw at 555nm (or 550nm for CZCS). Bricaud et al. (1981) show to for typical oceanic particle distributions that the wavelength dependence for bbp is approximately A - 1; thus, the backscatter term in (7) can be approximated by an empirical constant. Bricaud et al. (1981) showed for most oceanic areas, DOM absorption is correlated with chlorophyll-specific abthis DOM dependence is used in (7), then the radiance ratio can be approximated by Li:j = Bb La,.(A,) + %(A,)CJ Fo(Aj)' where Bb, B, Aj, and A_ are arbitrary constants. In the case where Lwg(555) is the reference band of the twoband ratio and when ag is much less than aw (i.e., C less than 1.0mgm-3), the radiance ratio can be further approximated to plankton biomass depends on ecological, rather than op- B tical, correlates. The formulation of biomass absorption does not include the package effect (Duysens 1956). The 10 L_:j - I + CA (10)

Original page 13

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark In the analysespresentedhere,the hyperbolicequations(9) and(10)arethe basicmodelsforthe two-band ratioalgorithmsthatuseSeaWiFSband5ratherthanthe sponses (Barnes et al. 1994). Data for pure seawater scatconventionalmodelderivedfromalog-logfit. Thesehypertering were taken from Morel (1974) and for pure seawabolicmodelshavethe usefulpropertythat thecoefficient ter absorbancy from Smith and Baker (1981). Particulate B can be expressed in terms of the IOPs of pure seawater, backscatter was scaled to the backscatter from Gordon et i.e., al. (1988), at 443nm and 1 mgm -3 chlorophyll; backscat- 4.2.1 Model Parameterization All the data were integrated over the SeaWiFS band re- B = g[)__[bbw(A,) aw(Aj)a_(Ai)F0(Aj)F°(Ai)]j, (11) ter for other wavelengths was calculated using an a -n dependence, with n = 1. The exact value of n, however, will and that there is a lower limit to the ratio of normalized depend on the oceanic particle size distribution (Morel and water-leaving radiances Prieur 1977). Carder (pets. comm) indicates n depends on the R(443):R(490) ratio, which is dependent on DOM and the carotenoid to chlorophyll ratio. DOM absorption was Li:¢ = B determined using a curve of the form This lower limit is useful, since it determines the range of applicability of the algorithms, i.e., the point where the ag(A) = ag(375)e -S(-375) (14) radiance ratio does not give a meaningful estimate of chlo- using a slope S of 0.014 and a base value of 0.06 (Bricaud rophyll a or pigment. et al. 1981), the slope being identical with that used in The factor g in (11) corresponds to the ratio of the Carder et al. 1994. Detrital material was assumed to have Morel and Gentili f/Q ratio at bands i and j. Morel and the same basic spectral shape as DOM and was included Gentili (1993) found this ratio to be 0.804 for the 440 and in the model with a backscatter to absorption ratio of 565 nm wavelengths, over the whole range of water types 1:2.18 derived from the transmission and absorption data and #0. Comparisons of the clear water B values in Tables of Prieur and Sathyendranath (1981) and assuming the 11 and 12 show the f/Q ratio to be 0.799 for the 443:555 San Diego harbor scattering phase function (Petzold 1972). band ratio. This further validates the hyperbolic model, Specific pigment absorption was taken from Bidigare et al. since discrepancies between model values can be explained (1990). The model parameter values are summarized in in terms of light field effects. Tables 7 and 8. The estimated global averages for pigment are shown in Table 9, and compared with data for chloro- 4.2 Model Development phyll specific absorption from Prieur and Sathyendranath (1981). Two bio-optical models were developed to support the analysis of the various algorithms: the first to determine the differential effects of the bio-optical determinands on the radiance ratios; and the second to determine the ef- 4.2.2 Global Determinands This model was used to determine the differential effects of the biological variability and intercorrelation of fects of the biogeochemical parameters on the two-band rathe water column constituents on the radiance ratios. The tios. The values of these parameters were fixed for a chloropurpose of the second model was to test algorithm formu- phyll value of 1.0 mg m -3. The pigments were fixed at the lation on sets of simulated data that contained variability global ratios of C,:Cb=5.77; C_:C¢=5.01; Ca:Cps=l.28; found in a wider range of bio-optical provinces than the Ca:Cpp-5.32; determined from the global pigment data NET data. set for chlorophyll a in the range 0.8-1.2. The particulates Both models use (5) and (6) with a full pigment assem- were assumed to be phytoplankton only. The increase in blage. The package effect was not included in either model. scattering effectively increased the phytoplankton specific Raman emission was included in the second model, using scattering. Detrital material increased scattering and the an approximation of Marshall and Smith's (1990) expres- base DOM absorption in the ratio 1:2.18. sion for surface Raman reflectance: , bRE (13) LWN = 6.0Qa + a ' el (BSM) was chlorophyll a; in each run of the model, 4.2.3 BSM Data The driving variable for the Bio-Optical Synthetic Modwhere E is the downwelling irradiance at the Raman exci- 2,000 random data points were generated using a log unitation wavelength, bR is the Raman scattering coefficient, form random variate with values of chlorophyll from 0.018a _ is the absorption at the Raman excitation wavelength, 20.08. These chlorophyll values were used to determine the and a is the absorption at the Raman emission wavelength. detrital, DOM, and pigment concentrations, and, hence, Without the Raman term, the model does not give a rea- bio-optical parameters using (6) and (7). Raman stimusonable approximation to the optical properties of pure lated emission was simulated using a randomly varying F0 water. from 60-150#Wcm -2 nm -1 using (13). 11

Original page 14

TheSeaWiFSCZCS-TypePigmentAlgorithm Table 7. Inherentopticalpropertiesof bio-opticalconstituents. Band A bbw bbp 1 412 0.0034 0.0031 2 443 0.0024 0.0030 3 490 0.0016 0.0028 4 510 0.0013 0.0027 5 555 0.0009 0.0026 t Data derived from Prieur and Sathyendranath (1981). Table 8. Inherent optical properties of phytoplankton Band A a_ t aa 1 412 0.050 0.017 2 443 0.058 0.018 3 490 0.045 0.003 4 510 0.036 0.001 5 555 0.019 0.001 Data derived from Prieur and Sathyendranath (1981). Data derived from Bidigare et al. (1991). Table 9. Inherent optical properties of phytoplankton Band A aezt a_§ 1 412 0.050 0.032 2 443 0.058 0.053 3 490 0.045 0.046 4 510 0.036 0.032 5 555 0.019 0.009 Data derived from Prieur and Sathyendranath (1981). ato a_ a¢ t b R (A t ) 0.016 0.034 0.050 0.015 0.023 0.058 0.021 0.012 0.045 0.00048 0.036 0.009 0.036 0.00042 0.067 0.005 0.019 0.00029 pigments_;. a b ac a p s a p p 0.003 0.021 0.008 0.022 0.013 0.046 0.019 0.045 0.011 0.011 0.036 0.049 0.001 0.002 0.033 0.022 0.002 0.003 0.008 0.001 pigments weighted according to climatological ratios:. aa a b ac aps app 0.017 0.001 0.004 0.006 0.004 0.018 0.003 0.009 0.015 0.008 0.003 0.002 0.002 0.030 0.009 0.001 0.000 0.000 0.027 0.004 0.001 0.000 0.001 0.007 0.000 The absorbance values are calculated according to climatological ratios (C:Cb=5.77, C_:C_=5.01, C,:Cps=l.28, and Ca:Cpp=5.32). The data is from the GIN Seas, EqPac, NEAT, and the Antarctic. § ar_ = aa + a b q- ac + aps -[-app. The variance for the scattering was derived from the with chlorophyll a, but with a log variance of 0.21, derived UK-BOFS database using c(670) and chlorophyll, for data at or above the 10% light level. The variance of the c(670) to chlorophyll ratio was found to be log-normally distributfrom NABE DOM concentrations. The relationships described above resulted in the 95% ranges in the model input data of 0.023-15.02 for chlorophylla; 0.001-1.70 for ed with a variance of 0.823; since log variance is scale in- chlorophyll b; 0.002-1.87 for chlorophyll c; 0.015-9.69 for variate, this log variance was used to generate the detrital backscatter contribution. The variance for phytoplankton backscatter was arbitrarily set at 0.100. The ratio of detrital backscatter was twice the value of phytoplankton backscatter; this figure was derived from Prieur and was (Yentsch and Menzel 1963) of chlorophyll a determination Sathyendranath's (1981) brain(525). The pigment data similarly found to be log-normally distributed, with varib, following figures, derived from Trees et al. (1985), and valiances of 1.017, 0.683, 0.482, and 0.151 for chlorophyll chlorophyll c, PSC, and PPC, respectively. is chlorophyll a was calculated as The variation of DOM absorption with chlorophyll dependant on oceanic conditions. Bricaud et al. (1981) observe an almost constant background, a9(375 ) -- 0.06, whereas Prieur and Sathyendranath (1981) and Carder et al. (1989) show a weak correlation with chlorophyll a, and Carder et al. (1986) show a good relationship with chlorophyll a. The assumption applied here is that ag co-varies 12 PSC; 0.003-5.20 for PPC; and 0.024-13.62 for DOM relative absorption. In order to compare the results from the BSM data with the NET data, it was necessary to simulate pigment data. Trees et al. (1985) showed that the fluorometric method was affected by coexisting chlorophyll b and c. Using the dated by comparison with NEAT HPLC data, fluorometric Cs = 0.941C_ - 0.292C b + 0.371Cc (15) and pigment as Ps = 1.166Ca + 0.616C b + 0.544Cc. (16)

Original page 15

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K Clark 5. DATA ANALYSES 5.1 Algorithm Development In the previous section, a theoretical basis for bandratio algorithms using standard hydrological, bio-optical models was established. In this section, these models are used with selected data sets to derive empirical fits of the data to produce the algorithms for SeaWiFS. The inherent statistical properties of the NET data are explored to establish a baseline for comparative reference. Using the conventional power law model and the pigment definition for the CZCS algorithms Ca + Cp = a'(L,:j) _' (17) or ln(Ca + Cp) = ln(a') + /Yln (Li:j). (18) Table 10 shows the coefficients of the ln-ln regression and the percentage of variance explained (R 2) for pigment (Ca + Cp) regressed against all two-band combinations for the NET data (SeaWiFS band set), and Fig. 6 shows the scatterplots for the major ratios. The regressions fall into four groups, which depend on the wavelength of the reference radiance. The first group, reference LWN(555), shows very high R 2 values for all two-band ratios, indicating that Ca + Cp is highly correlated in all cases. The increasing value of R 2 from 412 to 443 to 490 to 510 is surprising since the 490 and 510 bands are at longer wavelengths than the chlorophyll a (or phaeopigment) abfour groups of band ratios as the NET data (Table 10). The R 2 values are considerably higher for the BSM data (up to 96%) where the only source of variability is from the bio-optical determinands. The higher R 2 values are explained by the fact that no measurement error was added to the BSM data. The inference from this observation is that since the synthetic pigment and radiance data are so highly correlated, the basic bio-optical model used in the synthesis and the parameter values employed are both reliable. Using simple bio-optical models, a hyperbolic function of chlorophyll was found to be the simplest expression for the relationship for the band ratio, see (10), with an asymptotic coefficient (B) as C --* 0, which relates to the IOPs of pure water given by (11). This simpler model is only robust where chlorophyll is low, and can be used to empirically determine the clear water B. Using the clear water B, the full range of data (Case-1 and Case-2) can be fitted using the full model (9). Tables 11 and 12 show the coefficients for the full model with Ca + Cp and Ca and band ratios fitted for the NET data. Tables 13 and 14 show the coefficients for chlorophyll a and simulated pigment for the BSM data. In all cases, the percentage of variance explained is high, up to 91% for the NET data and up to 96% for the key band ratios for the BSM data. In all cases, this model provides a superior fit compared with the ln-ln regressions. Again, the primary conclusion from these findings is the basic bio-optical models and parameters used are sound, as demonstrated by their utility to generate relationships sorption peaks. This is due to the highly correlated co- with high confidence. A secondary conclusion is that these occurrence and co-variance of chlorophyll and carotenoids (which absorb at 490 and 510nm), demonstrated in Section 2, which means carotenoids effectively and accurately behave as surrogates for chlorophyll at the longer wavelengths. Coupled with this effect, the longer wavelengths are least affected by the absorption of light in the blue spectral region by DOM, which contaminates the 412 and 443nm wavebands to the maximum extent. The second and third groups of band ratios (reference bands 510 and 490 nm, respectively) also show increasing R 2 values with increasing wavelength, 412 to 443 to 490 nm. The same explanations of the effects of co-existing DOM and accessory pigments (carotenoids) apply. In each case, the percentage of variance explained is smaller than for the first group as the reference wavelengths are influenced somewhat more by carotenoid absorption. Only the LWN(412):LwN(443) ratio is poorly correlated with pigment; (this is understandable since these wavebands are influenced most significantly by the absorption of DOM and detritus which have similar absorption spectra). Table 10 shows the ln-ln regression coefficients and R 2 values for the BSM data set. The scatterplots for the LWN(555) base ratios compared to NET data are shown in Fig. 7. Interestingly, these regressions show exactly the same patterns of coefficients and R 2 values for the methods, using synthetic data, are suitable for the generation of algorithms for parameters where there are few in situ calibration measurements, but where there is a good knowledge of the IOPs of the parameters (e.g., DOM, pigments, etc.). It is suggested here that good in situ data sets with a limited range of parameter values (e.g., NET data which has no accessory pigment measurements) can be bootstrapped to synthetic data to infer a greater number of parameters. It can be shown in these and other analyses that there is potential for closure between apparent optical measurements and synthetic measurements that use IOPs. 5.2 Sensitivity Analysis of Ratio Models Figure 8 shows the sensitivity of retrieved pigment concentration for each of the primary ratios (412:555, 443:555, 490:555, and 510:555). Table 15 shows symbolically the effects on the remaining secondary band ratios, together with the main ratios. At the level of chlorophyll a considered, detritus, DOM, and chlorophyll-specific scattering all change the apparent chlorophyll retrieved and the ratios by 5-10%, the effect being greatest in the blue part of the spectrum and decreasing towards the green. It is surprising that scattering should depress the ratio and result in an 13

Original page 16

TheSeaWiFSCZCS-TypePigmentAlgorithm Table 10. Logregressions of NET and synthetic simulated pigments with band ratios. Band NET Ca + Cp Ratio Intercept Slope 412:555 -0.339 -1.095 443:555 0.051 -1.284 490:555 0.696 -2.085 510:555 0.688 -2.864 412:510 -0.505 -1.533 443:510 -0.464 -2.209 490:510 0.627 -7.035 412:490 -0.791 -1.733 443:490 -0.924 -2.983 412:443 -0.601 -2.913 Table 11. NET data pigment curve fits. Band Ratio C BSM Ps R 2 Intercept Slope R 2 72.2 -0.199 -1.387 90.9 81.7 0.028 -1.562 95.1 86.4 0,322 -2.262 95.4 86.8 0.151 -3.541 95.9 58.7 -0.418 -2.193 84.4 74.4 -0.073 -2.702 87.8 81.5 0,604 -6.117 92.3 46.5 -0.967 -3.315 77.5 65.9 -0.605 -4.745 82.6 20.9 -1.629 -9.761 58.2 A1 A2 R 2 C_t 412:555 13.14 ± 1.86 0.059 ± 0.07 16.79 ± 0.76 84.4 13.80 443:555 9.55 ± 1.08 0.068 ::t=0.05 9.68 ± 0.40 87.9 11.94 490:555 5.29 ± 0.37 0.232 ± 0.04 3.82 ± 0.16 89.8 5.67 510:555 3.07 ± 0.15 0.262 ± 0.03 1.94 ± 0.09 91.4 2.80 C_ adjusted for Raman scattering in model. Table 12. NET data chlorophyll curve fits. Band Ratio C A1 A2 R 2 Cwt 412:555 13.14 ± 1.86 0.081 ± 0.09 20.51 ± 0.94 83.3 13.80 443:555 9.55 ± 1.08 0.115 ± 0.07 11.04 ± 0.52 86.5 11.94 490:555 5.29 ± 0.37 0.328 ± 0.05 4.27 ± 0.21 88.5 5.67 510:555 3.07 ± 0.15 0.350 ± 0.04 2.44 ± 0.12 90.5 2.80 C_ adjusted for Raman scattering in model. Table 13. BSM chlorophyll a curve fits. Band Ratio C A1 A2 R 2 Cw t 412:555 13.82 ± 0.84 0.167 ± 0.03 23.09 =t=0.34 90.3 13.80 443:555 11.74 ± 0.52 0.279 ± 0.03 18.58 ± 0.24 92.7 11.94 490:555 5.35 ± 0.10 0.499 ± 0.02 7.42 ± 0.09 95.4 5.67 510:555 2.63 ± 0.02 0.662 ± 0.01 3.61 ± 0.04 95.8 2.80 C,_ adjusted for Raman scattering in model. Table 14. BSM pigment curve fits. Band Ratio C A1 A2 R 2 Cw_ 412:555 13.82 :t=0.84 0.134 -4-0.03 17.74 ± 0.26 90.0 13.80 443:555 11.74 ± 0.52 0.219 ± 0.02 14.26 ± 0.19 92.6 11.94 490:555 5.35 ± 0.10 0.385 ± 0.01 5.69 ± 0.07 95.2 5.67 510:555 2.63 ± 0.02 0.510 ± 0.01 2..76 ± 0.03 95.6 2.80 C'_ adjusted for Raman scattering in model. 14

Original page 17

J. Aiken, G.F. Moore, C.C. Trees, I0,0 a) o o tO _J 1.0 ¢q J I o i 0.1 0,01 0,10 1.00 10.00 100.00 Plgrnentmg" 10.0 o c) LO LO .--I .............................................?.e........................... 1.0 0 __1 0.1 0.01 0.10 I.{)(} I0.00 100.00 P'o'nent mgm" 10.0 I ! i e) o °o,,,! i 0 i o , :: o i 0 ! o 0 ; 1.0 "'ii ....... _ .................. v- oi o ! 0.1 0.01 0.10 1.00 10.00 100,00 Pigmentmgm" Fig. 6. NET data (weighted pigment) scatterplots: S.B. Hooker, and D.K. Clark 10.0 O,o b) o tO i o tO i °o i i . 0 .J 1.0 ................... _ ............. ..-.., .-.--.o ....... 4 .................. i o o E 0.1 0.01 0.10 1.00 10.00 100.00 PWnentrng.m4 10.0 d) oo i o tO tO " 1.0 ...................[............N- i.................. 0 .J ! ioo° 0.1 0.10 1.00 10.00 100.00 0.01 Pigment mgln" 10.0 f) 0 3 • 1.0 o rn 5 .1 0.1 0,01 0.10 1.oo lO.0O 100.00 Plgmnt mg.m" a) LWN(412):LwN(555), b) LWN(443):LwN(555), c) LwN (490) :LwN (555), d) LwN (510): LwN (555), e) Lw (412):Lw (510), and f) Lw (490):Lw (510). 15

Original page 18

The SeaWiFS CZCS-Type Pigment Algorithm lOO . a ** • tF.'L.O • - . •; .. "el lO -"'::.. ,.'. I,,]N,AgJPO• . "t_ll*. _ =o_e I b,,,,, ."-" ..0 ." 0.1 _ ........ j ......... n ......... 0.01 0.10 1.00 10.00 Pigment mg.m 4 100 • "..'.'l... - ;:',.,.:2. • 10 • "..l_t_o , "tO, _ • .% 0.1 0.01 0.10 1.00 1000 Rgment m_m 4 100 b) UO •":.:,t.o, 1.0 • _ =.R-';. • " 0.1 ......... i ......... i ........ 0.01 0.10 1.00 10.00 Rgment mg.m 4 10.0 .................... d)l 1.0 o 0.1 ......... _ ......... ' 0.01 0.10 1,00 10.00 Pigment mg.m 4 Fig. 7. NET data (circles) compared with the synthetic d_ta set (diamonds): a) LWN(412):LwN(555), b) Lw(490):LwN(555), c) LWN(443):LwN(555), 16 and d) LwN(510):LwN(555).

Original page 19

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark 412:555 443:555 490:555 510:555 Band Ratio 412:555 443:555 490:555 510:555 Band Ratio 412:555 443:555 490:555 510:555 Band Ratio g 412:555 443:555 490:555 510:555 Band Ratio 1.20- b 1.15- .m Q,. .¢_ 1.10- 1.05- J_ 0 1.00- 412:555 443:555 490:555 510:555 Band Ratio d 412:555 443:555 490:555 510:555 Band Ratio f O"J °_ °_ 1.05 " 412:555 443:555 490:555 510:555 Band Ratio ,- 1.20e, h _ ! 1.15 -if ° ¢ 1.10-. m¢ 1.05 1.oo 412:555 443:555 490:555 510:555 Band Ratio Fig. 8. The effects on retrieved pigment of 20% (10% for detritus) change in concentration relative to a climatoligical value: a) scattering, b) Cb, c) DOM, d) Cc, e) detritus, f) Cps, g) Ca, and h) Cpp. All values are calculated for a chlorophyll concentration of 1.0 mg m -3. 17

Original page 20

TheSeaWiFSCZCS-TypePigmentAlgorithm Table 15. Sensitivityof ratioto modelsingleparameterchange. Band Ratio Scat. Gelb. Detr. 412:555 ÷÷ ÷+÷÷ 4.÷÷÷ 443:555 4.÷ ÷÷ ÷÷÷ 490:555 ÷ ÷ ÷÷ 510:555 4. 412:443 ÷ ÷÷ 4,÷÷÷÷ 443:490 4. ÷÷÷÷ 490:510 ÷4,÷ 412:490 4. ÷÷4. ÷÷4.÷4. 443:510 4. ÷ ÷4,4,÷÷ 412:510 4-4- 4.4.÷ 4.4.4.÷÷ Chl. a Chl. b Chl. c PSC PPC ÷4- 4-4- ÷ 4. ÷ ÷÷÷ 4- ÷4.4. ÷÷ ÷ ÷÷ 4.÷ 4. ÷÷ Note: < 3% +; 3 - -5% 4-+; 5 - -7% +++; 7 - -20% ++++; > 20% +++4-+ (and similarly for negative values). Table 16. Principal component factor loadings. Band Ratio F1-NET F1-BSM 412:555 ÷ ÷ 443:555 ÷4- ÷4. 490:555 4-4-÷ ÷÷4. 510:555 ÷4.÷ 4.÷4.÷ 412:443 443:490 490:510 ÷÷ ÷ 412:490 ÷ 443:510 412:510 F2-NET F2-BSM F3-NET F3-BSM ÷ ÷÷ ÷ 4-4-4-4-4- 4.4.÷÷4. ÷÷4-4-4- ÷ ÷4-4- ÷÷4,÷÷ ÷÷4.÷ 4-÷÷ ÷4-÷ ÷ ÷÷ ÷+÷÷÷ ÷4- ÷÷÷ Note: < 3% +; 3 - -5% ++; 5- -7_ +++; 7- -20% ++++; > 20% +4.+4.+ (and similarly for negative values). increase in apparent chlorophyll retrieved, but it should be remembered that the ratios have been explored in a region where backscatter from water is dominant. Chlorophyll a has a dominant effect on the 412:555 and 443:555 ratios, contrasted with photosynthetic carotenoids which affect the 490:555 and 510:555 ratios. The photoprotectant carotenoids effect most of the band ratios with the dominant affect being on the 490:555 ratio. Chlorophylls b and c have only minor affects on the band ratios, and major province difference would be needed for their affect on band ratios to become important. The only two-band ratios affected by all pigments are the 443:555 and 490:555, with the later being less affected by DOM or detritus. Of all the secondary ratios, the most useful may be 412:443, which is not influenced by pigment but is strongly influenced by DOM and detritus. For determining accessory pigments, the 412:490, 443:510, and 412:510 ratios seem most promising, since they show a differential influence between photosynthetic carotenoids and chlorophyll a. The response of the different ratios to differing biogeochemistry, would imply that most, if not all, of the optical and biogeochemical information found in upwelled spectra is contained in the primary and secondary ratios considered in this study. 18 5.3 Analysis of NET and BSM Data The NET data represent high quality optical data, with limited measures of biogeochemical parameters other than chlorophyll a and phaeopigment. In order to determine the bio-optical variability of the data, principal components analysis was used, and the results were compared with the BSM data set containing variability from a wider range of provinces than those present in the NET data. The aim was to determine if the NET data contained sufficient variability to generate global algorithms, and to determine how far bio-optical variability reflects ecosystem variability. Analysis of the NET data log radiance ratios, by principal components analysis using varimax rotation, showed that three factors could be extracted explaining 99.6% of the total variance. These three factors explained 57.5, 41.2, and 0.9 percent of the variance, respectively. The relative factor loadings are shown in Table 16, with the normalized factor loadings for the LWN(555) based ratios shown in Fig. 9. The first column of each factor pair shows the factor pattern for the NET data. By comparison with the previous analysis of radiance ratio patterns, the first two factors, F1 and F2, can be ascribed to pigment biomass and detritus concentration, respectively, with the first factor

Original page 21

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark 100- 75 5o- 25- 412:555 443:555 490:555 510:555 Band Ratio 100m 75 8 • m 50 8 25 0-" 443:555 490:555 510:555 412:555 Band Ratio 100- 8, m_ 75 50 a. ° 0 w 412:555 443:555 490:555 510:555 Band Ratio Fig. 9. The normalized spectral change in the factors F1 (top), F2 (middle), and /:3 (bottom). The NET data is shown as the white bars and the simulated data as the hatched bars. 19

Original page 22

TheSeaWiFSCZCS-TypePigmentAlgorithm (R2_-89.5%)the NET data, a factor that was not included in the model. showinga highcorrelationwith logpigment ofthethird Figure 5i shows such a shift in pigment ratios; high light or logchlorophyll(R2=89.2%).Themeaning factor,F3, is more difficult to determine, but the principal loadings are on those ratios which correlate with carotenoid concentration, notably 490:510 and 510:555. This factor could be related to absolute carotenoid concentration or relative pigment abundance (PSC:PPC ratio). The BSM data showed a similar factor pattern to the NET data with 99.9% of the total variance explained, and the three factors explaining 52.3%, 43.2%, and 4.4% of the variance, respectively. The change in the total variance explained is probably attributable to the BSM data assuming a perfect radiometer; the relatively small change in total variance explained indicates the quality of the NET optica] data. The higher percentage variance explained by the third factor, reflects the greater range of pigment and detrital variance introduced into the synthetic data. The factor profiles are shown in Table 16, and the patterns for F1 and F2 are almost identical to the NET data. The comparison with Fa, is less clear, but both NET and BSM data show loadings on ratios that relate to carotenoid concentration. areas, e.g., EqPac and BATS, show a higher PPC relative to PSC ratio or PPC to carotenoid ratio. It is surprising that the information in the NET and BSM data sets can be reduced to three factors, whereas there are five factors in the biogeochemica] variables that generate the BSM optical data. This may be due to the limited range of bands chosen, but the reduction of optical data to three factors is supported by a number of other studies. Sathyendranath (1981), using principal components analysis on log reflectance data from 400-650nm with a resolution of 10nm, found three factors explaining 57.6%, 42.1%, and 0.2%, respectively, of the variance. Garver et al. (1995), using empirical orthogonal factor analysis on absorbance spectra from 400-700nm, found detrital and phytoplankton components explaining 54% and 44%, respectively, of the variance; the residual 2% of variance was assumed to be pigment variability. Mueller (1976), using principal components analysis on airborne spectrometer data from 400-750 nm with a resolution of 5- The difference in the precise loadings of the third factor is 7.5 nm, found four factors explaining 77.6%, 17.21%, 2.4%, likely to be caused by the lower pigment variability in the NET data compared with the BSM data. This hypothesis needs to be tested further using data sets that have concurrent radiance measurements and HPLC pigments. These will become available as part of the SeaWiFS calibration and validation activity. and 0.9%, respectively, of the variances; the first three factors were related to pigment, showing similar patterns to the NET factors, but the fourth factor was unrelated to pigment and may have been due to residual atmospheric effects. From this and other studies, it can be concluded that In the BSM data, the underlying biogeochemistry is for oceanic waters, the SeaWiFS band set provides suffiknown a priori; thus, the factors can be related to their principal driving variable. F1 shows the best relationship with total pigment, chlorophyll (a, b, and c) , and carotenoids, with the best correlation being with total pigment (R2=89.9%). F2 shows little correlation with pigment, correlating with the total particulate concentration and total DOM-like absorbance (i.e., detrital absorbance), but with a poor R 2 of 58.1% and 62.3%, respectively. F3 shows low correlations with all variables, the best being with DOM-like absorbance and photosynthetic carotenoids, with R 2 of 30.6% and 27.6% respectively. This would still seem to indicate F3 represents the chlorophyll to carotenoid ratio, since an increase in DOM has almost the same effect on a normalized SeaWiFS spectrum as an increase in carotenoids. Figure 9 shows the spectral differences in the factors, derived by selecting data where each factor was lower than cient information to adequately specify the upwelling spectrum, and that the full compliment of ratios contain the full variance of the original upwelling radiance. The total variability in the optical signal can be reduced to three factors, although there are more factors in the underlying biogeochemistry. This fact has two implications for algorithm development: first, it is unlikely that more than three suitably chosen ratios are needed to retrieve any parameter to maximum accuracy; and second, different biogeochemical signals are not uniquely converted to optical signatures, implying that the perfect single biogeochemica] parameter algorithm may not exist. 5.4 Multiband Algorithms Multiband algorithms were developed using two methodologies, the first used empirical multiple regression of log the factor average. F1 and F2 show similar patterns in ratios to log pigment, i.e., fitting curves of the form both the NET data and the BSM data, although for F2 the typical DOM curve is more pronounced in the NET data than in the BSM data. F3 shows a greater disparity between the NET data and the BSM data; the main affect in the NET data is the 490:555 ratio whereas in the BSM data, it is the 510:555 ratio. The spectral patterning of the change in the NET data would suggest a shift in PPC, compared with a shift in total carotenoid in the BSM data. The change in PPC is probably due to light adaptation in 20 In(Co+cp) = -o + .lln(L,:j) (19) -'1- a21n(Lm:n) which is equivalent to k ca÷cp -- aII (20)

Original page 23

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark wherek represents an index to two vectors of band ratios kl and k2. The second method used linear combinations of the hyperbolic estimates. Before examining these algorithms, it is appropriate to make a statistical comment. First, the R 2 values returned by multiple log regression are not comparable with those returned by either nonlinear curve fitting or by linear multiple regression. Second, a small improvement in R 2 can be obtained by shifting the position of one outlier, and may not be a real improvement in the quality of calibration. With these caveats in mind, log-log multiple regressions are considered hereafter. Table 17 shows multiple regression algorithms for the NET data. The best R 2 achieved is 87.3% (for the 443:490 and 443:555 combination), which represents a marginal improvement of the variance explained compared with 86.8% (510:555 in Table 10). Compared to the R 2 for the 443:555 ratio, however, an improvement of 6.6% is achieved. Figure 10a shows comparison of this algorithm with the single ratio hyperbolic model; the difference between the two are slight, but a few points are pulled closer to the 1:1 line. The robustness of this algorithm was tested with some data collected in the Antarctic (BOFS Sterna), where there are known differences in the optical properties of the water (Mitchell and Holm-Hansen 1991 and Sullivan et al. 1993). Figure 10b shows a comparison between the NASA CZCS chlorophyll algorithm and the multiple regression algorithm. The retrieval shows a less biased estimate of chlorophyll than the NASA algorithm, with a slope closer to truth data in three important aspects: to 1:1. The retrieval is noisier, and this may be due or chlorophyll are linear. Such algorithms are more attractive since the intermediate estimates can be interpreted in terms of pigment. Table 20 gives the results of linear combinations of pairs of estimates. In all cases, except the 443:555 and 510:555 combination, there was no significant intercept for the regression. The best retrieval was for the 443:555 and 490:555 combination, with an R 2 of 95.3%; this algorithm is shown in Fig. lla, compared with the 443:555 hyperbolic fit. Compared with the ln-ln multiple regression in Fig. 10a, there is less bias at low pigment concentrations, whilst retaining the 1:1 relationship at high pigment concentrations. The results for the Antarctic data shown in Fig. llb are encouraging, with points pulled closer to the 1:1 line; however, there may be a slight tendancy to overestimate pigment in this data. The 443:555 and 490:555 combination has advantages for implementation as a remotely sensed algorithm; it avoids the 412 nm band where atmospheric correction may be a problem, and tends to the 490:555 algorithm in high pigment waters where LwN(443) tends toward zero. 6. IMPLEMENTATION The algorithms shown in the previous sections are all valid for the data and models considered. The development on an interim CZCS compatible algorithm involves consideration of the sensor aspects and state-of-the-art atmospheric correction. The data to be obtained from the SeaWiFS instrument is different from modeled and searadiometric problems, since the noise is additive and in- 1) The data is digitized on a limited scale---10 bits creases as more bands are used. It is not known if this will at the top of the atmosphere compared with 16 be a problem for SeaWiFS. It is noteworthy that a retrieval developed on one data set, should show an improvement with another from a different bio-optical province. Table 18 shows the algorithms developed from the BSM data, with the R 2 being in the range 93.3-98.1%. Since the synthetic model assumes no error in measurements, these are the best retrievals to be expected unless Raman emission is modeled. The difference between these R 2 and the NET data R 2 is best explained by the pigment analysis methodology, since an error of 5-10% is expected in the pro- To address the first two points, digitization of the data Yentsch and Menzel (1963) chlorophyll. Table 18 also vides a list of algorithms for new biogeochemical variables that may be tested when data sets with more extensive bit resolution for the NET data and infinite resolution for the model data; 2) Atmospheric attenuation of the water-leaving signal will reduce the effective level still further, especially at the shorter wavelengths, i.e., bands 1 and 2; and 3) The model and NET data are both based on nadir viewing geometry, whereas the SeaWiFS instrument will view up to 58.3 ° off nadir. has been simulated by assuming a sun angle of 50 ° and an atmospheric transmission of 0.5. This assumption results sea-truth measurements than the NET data become avail- in a dynamic range of about 160 counts for the waterable. Table 19 shows a test of multiple regression algorithms restricted to the SeaWiFS bands which are compatible with the CZCS. Except for total pigment, no improvelist ranges. For bands 1 and 2, there is no detectible waterment could be found taking a band combination. The be leaving radiance above 10mgm -3, compared to bands 3 demonstrates the alternative interpretations that can derived from CZCS imagery; in particular it emphasizes the fact the 510:555 ratio is a carotenoid retrieval. for retically at present, since there are limited data, e.g., Po- Multiple regression is a more appropriate technique the hyperbolic estimates, since the estimates of pigment leaving radiance, and corresponds to fairly high atmospheric turbidity in the early spring, e.g., the North Atlantic bloom. Figure 12 shows the mean water-leaving radiance for bands 1-4, compared with pigment averaged into log and 4 where there is significant water-leaving radiance. The issue of viewing angle can only be addressed theolarization and Directionality of the Earth's Reflectances 21

Original page 24

TheSeaWiFSCZCS-TypePigmentAlgorithm Table 17. NETdatalogmultipleregressionalgorithms. Variable Constant Coeff. 1 Ratio 1 Coeff. 2 Ratio 2 R 2 C_ 0.45 1.57 412:510 -2.46 443:555 84.7 C_ 0.94 3.47 443:490 -2.58 443:555 86.4 Ca + Cp 0.59 1.48 412:510 -2.41 443:555 85.4 Ca + Cp 1.09 3.39 443:490 -2.55 443:555 87.3 Table 18. BSM data log multiple regression algorithms. Variable Constant Coeff. 1 Ratio 1 Cs -0.040 4.60 412:510 Cs 0.345 4.60 443:490 Ps 0.194 4.99 412:510 P 0.918 2.64 443:510 P 0.229 10.77 490:510 G 0.616 2.44 412:443 G 0.657 12.49 412:510 Cpp + Cps -0.133 3.93 412:443 Cpp + Cps -0.174 0.61 443:490 DOM -0.012 4.69 412:443 DOM -0.021 0.39 443:490 C_bc 0. 724 0.58 412:443 CTp 1.063 --2.24 490:555 Table 19. BSM data CZCS compatible log multiple Coeff. 2 Ratio 2 Coeff. 3 Ratio 3 R 2 -7.22 443:555 3.46 490:555 97.2 -2.89 443:555 96.1 -7.98 443:555 4.16 490:555 94.2 -2.77 412:555 97.7 -4.58 443:555 98.1 --4.43 412:555 11.12 490:510 98.2 --12.50 443:555 97.7 --1.55 443:490 -3.71 510:555 97.1 --3.91 510:555 95.9 -1.41 412:490 --3.61 510:555 98.6 -3.77 510:555 94.7 -2.32 490:555 93.3 93.3 regression algorithms. Variable Constant Coeff. 1 Ratio 1 Coeff. 2 Ratio 2 R 2 Ca -0.157 -1.59 443:555 93.3 Ps 0.113 - 1.60 443:555 93.2 P 1.024 -1.69 443:555 96.7 Cpp -_- Cp S -0.251 -3.55 510:555 95.7 G 1.579 -1.70 443:555 93.2 DOM -0.069 -3.54 443:555 94.2 CTp 13.920 -40.60 510:555 20.76 553:510 92.7 Table 20. NET datl multiple regression hyperbolic fit pigment. Constant H(Al:555) Coeff. 1 412 0.188 ± 0.03 412 0.519±0.15 412 0.954 ± 0.07 443 -0.461 ± 0.03 0.757 ± 0.24 443 -0.659 ± 0.02 490 1.903 ± 0.13 22 H(A2:555) Coeff. 2 R 2 443 1.188 ± 0.08 84.8 490 1.121 ± 0.05 92.5 510 0.759 ± 0.04 91.4 490 1.821 ± 0.08 95.3 510 1.101 ± 0.29 66.0 510 -0.653 ± 0.06 91.6

Original page 25

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark 10.00 CO ) 1.00 "D "_ 0,10 rr 0,01 0.01 O,10 Pigment rng.m 10,00 03 I 1.00 E oy y : • "13 0.10 rr 0.01 0.01 O.10 Pigment mg.m 0 MULT2 o PIG44 1,00 10.00 -3 oi o ! o MULT2 o PIG44 1.00 10.00 -3 Fig. 10. Multiple regression log-log pigment [In(Ca + Cp) = 0.59 + 1.481n(LwN(412):LwN(510)) -- 2.411n(LwN(443):LwN(555))] (circles and upper line) compared with hyperbolic pigment H(443:555) (diamonds and lower line): a) NET data (top) and b) Sterna data (bottom). 23

Original page 26

TheSeaWiFSCZCS-TypePigmentAlgorithm 10.00 CO 1 ,°°..........................i.............._ .....:........... : o _f :o • ! /2/o .w 2:J cD {D 0.10 ...................... , ......................... -:........................... rr 0.01 i 0.01 O.10 Pigment mg.m 10.00 03 I 1.00 E E t : :F, 0.10 cr 0.01 0.01 O.10 Pigment mg.m Fig. 11. Hyperbolic multiple regression pigment (circles and upper line) compared with H(443:555) data (top) and b) Sterna data (bottom). 24 o MULT3 o PIG44 1.00 10.00 -3 i oO • o o: o MULT3 O PIG44 1.00 10.00 -3 [ln(Ca + Cp) = -0.461H(443:555) + 1.821H(490:555) derived pigment (diamonds and lower line): a) NET

Original page 27

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark 4O G ) 3O C O 4\ O "O 2O m E im {D 10 o 0.01 0.1 Pigment Key 0 Lw(412) - --CY- - Lw(443) ---A--- LW(490) ..... _ ..... Lw(510 ) 1 10 100 [mg m-3] Fig. 12. Mean simulated SeaWiFS counts as a function of pigment, log(Ca + Cp). 1000 100 "- I0 E 0.1 " 0.01 0.001 0.0001 H(412:555) H(443:555) H(490:555) H(510:555) °_ 0.01 0.1 I 10 100 Pigment [mg m -3] Fig. 13. Derivatives of the hyperbolic model versus Ps. 25

Original page 28

TheSeaWiFSCZCS-TypePigmentAlgorithm (POLDER)airborneimagery,to test publishedmodels.11, and 12 indicated there is no significant difference be- MorelandGentili(1993)indicatethat for ratios,theef- tween the quality of fits of the ratios for either the log or thanforradiancehyperbolic model fit. The pigment relationship shown in fectsof viewingoffnadirarelesssevere inversionmethods;the viewingangleeffectis not seenin Fig. 14 indicates there are biological reasons for consider- CZCS images, but some effects may be observed with the greater radiometric precision of SeaWiFS. The precision of the retrieval can be addressed using both the NET data and the hyperbolic model. The hyperbolic model can be differentiated analytically to show how dL,:j/dPs varies with pigment. The rate of change of A model constrained with high chlorophyll simulated in ratio per unit pigment gives the inherent accuracy the retrievals. Figure 13 shows this rate of change is high for low pigment concentrations, and also shows a log-log decrease with increasing pigment concentration. At low pigment dL,:j/dP s is higher for the 412:555 and 443:555, but the rate of change is similar for all band ratios at pigments greater the 1 mg m -3. The log-log relationship also indicates that the error structure of the pigment retrievals will be log normal, even if local variability in pigment is normally distributed. The relative sensitivities shown in Fig. 8 indicate the 412:555 and 443:555 ratios respond to chlorophylls and almost as strongly to DOM, and the 490:555 and 510:555 ratios respond principally to carotenoids, with the 490:555 ing the 490:555 ratio superior, since its response to chlorophyll will alleviate potential problems in oligotrophic waters. Both the hyperbolic model and log-log fits represent adequate models for the data, but the hyperbolic model shows erroneous retrievals at high chlorophyll where there is little data. data has been used to produce a revised set of coefficients for the hyperbolic model. The constrained fits shown in Tables 21 and 22 give a better R 2 for the 490:555 ratio compared with the 510:555 ratio. Using these revised tables, the error structure can be examined. Figure 15 shows the mean residual error for both the hyperbolic model and log-log regression. Although the hyperbolic model performs better at pigments less than 2mgm -3, the log-log regression covers the whole range of pigment. The final algorithm uses the hyperbolic estimates at low pigment to account for the deviation from log linearity at low chlorophyll (see Fig. 6), and the log-log regression at pigments greater than this. ratio showing some response to chlorophylls. Figure 14 Table 21. Revised NET data pigment curve fits, shows the relationship between total chlorophyll and total carotenoid, showing the ranges of pigment where each band ratio may be used. The carotenoid chlorophyll ratio seems to break down at total chlorophylls below 0.1 mg m -3. This effect may be due to different species assemblages in highly oligotrophic waters (possibly due to phycobiliproteins replacing carotenoids), or it may be an analytical artifact, since the total carotenoids are a sum of a greater number of components, each of which may be below the detection limits of HPLC. More data will be needed to resolve the problem, but the present data suggest some caution is required when using the 490:555 (and 510:555) ratio to determine chlorophyll in oligotrophic waters. These considerations limit the choice of final algorithm, from those suggested in the previous sections. The following conclusions can be made. 1) If algorithm switching is to be avoided, only combinations of the 490, 510, and 555 bands can be used. Table 17 shows that no multiple regression log algorithm could be derived using these bands. In essence, the 490:555, 490:410, and 510:555 contained no extra information to determine pigment. 2) If algorithm switching can be implemented, then Table 17 indicates the 443:490 and 443:555 ratios can be used. It will not be necessary to use the 412 nm band where atmospheric correction may be a problem. For the single algorithm, it is a matter of choosing between the 490:555 and 510:555 combinations. Tables 10, 26 where bands refer to the band:555 (band 5) ratio. Band B A1 A2 R 2 412 13.14 0.019 12.48 88.6 443 9.55 0.045 8.59 89.1 490 5.29 0.112 3.48 87.5 510 3.17 0.140 1.79 84.9 Table 22. Revised NET data chlorophyll curve fits, where bands refer to the band:555 (band 5) ratio. Band B A1 A2 R 2 412 13.14 0.022 14.45 86.6 443 9.55 0.054 10.31 87.5 490 5.29 0.136 4.23 85.7 510 3.17 0.169 2.16 82.7 This combined method gives an R 2 of 90.9% for pigment compared with R 2 values of 87.5% and 89.6% for the hyperbolic and log regressions, respectively. Although this is only a small improvement over the whole range, for pigment concentrations less than 2 mg m -3 the hyperbolic model gives an R 2 of 82.3% compared with an R 2 of 74.1% for the log regression. This improvement adds considerably to the accuracy of pigment retrievals in oligotrophic waters. Explicitly, the algorithms for chlorophyll and pigment are computed as follows: 1) The log regressions are determined as Ca : exp [0.464- .....1.9m[_ (21)

Original page 29

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark 10 Chlorophyll vs Carotenoid Suspect E E t....J 0.10 + 0.01 o o o 0 o 0 o o o o o o 0 o LWN (443) Suspect "% LWN (443) and LWN (490) Valid 0.001 i i i u ,,t ........ i ........ 0.001 0.01 Cabc [mg m-3] 0.1 1 10 Fig. 14. The global variation of total chlorophyll (C,_bc) versus carotenoids (Cpp + Cps), showing the range of validity for LWN(443) and LWN(490); the ranges where the relationship is suspect are to the right and left. 27

Original page 30

TheSeaWiFSCZCS-Type Pigment Algorithm 200 150 100 5O e_ 0 - -50 - -100 - Log - Log ' 41_- H(490:555) -150 - • ............• -200 - ' ' ' ' ''"] ' ' ' 'P'"I , , , , ,i,, , , , , ,,,, 0.01 0.1 C= + Cp [mg m"3] 1 10 100 Fig. 15. Percent error in retrieval for the log-log and hyperbolic fits of Ca + Cp for the 490:555 ratio• and ; (22) on a correction to the 490:555 ratio algorithm shown above. Ca --}- Cp = exp [0.696 - 2.085 In ()]LwN(490) and 2) if Ca or Ca + Cp are less than 2.0 mg m-3, the inversion of the hyperbolic model,i.e., C' = (L,:j - B)/(AIB- A2L:j), is used to calculate Cp and Ca as LWN(490) -- 5.29 can produce negative retrievals. The final multiband algorithm that was developed avoids these problems; it is based By using the same band ratios as the hyperbolic multiple pigment algorithms shown in Fig. 11 (see Table 19), this algorithm can be normalized to the pigment retrievals as described below. The multiple regression algorithm for chlorophyll is: C_, = LWN(555) (23) C_ = A1H(490:555) + A2H(443:555) (25) 0.719 -- 4 23_ • LWN(555) and L_N(490) _ 5.29 Ca q'- Cp = LWN(555) (24) 0•592 - 3 48 LWN(49°)' • LwN (555) It should be emphasized that the split represents a curve fitting method, rather than a split algorithm. With a more extensive high pigment data set, the hyperbolic model can be better constrained to cover the entire dynamic range of pigment. Figure 16 shows a comparison of the 490:555 ratio algorithm with the OCTS algorithm, the Clark combined ratio model, and the 443:555 ratio algorithm. The algorithm is, as expected, highly sensitive to carotenoids; it shows the least response to Gelbstoff, scattering, and detritus. Although a number of the multiband algorithms shown in previous sections show high R 2 values, they also display problems when used as inversion algorithms in that they 28 or c. H(443:555) ] (26) H(490:555) = A' +A2L_ j , which may be approximated by ca - At H(490:555) "LwN(443)LwN(555) + A_ ] (27) ½ = "LwN(443)LwN(555) + A_ ] where the square root is an arbitrary scaling of the LWN ratio, and A1 and A2 are arbitrary constants. The use of the LWN ratio rather than the ratio of the hyperbolic pigments has the advantage of reducing noise, since only two bands are required for the final correction.

Original page 31

J. Aiken,G.F.Moore,C.C.Trees,S.B.Hooker,andD.K.Clark Cpp "///Y/////////////////////////////////'A \\\\\\\_ Cps Z//////////////////////////2"///////////////////////////2,"///////.Y//////////////////2".'/////////////////////Y//A Cc cb },_Y,_._ \\\\\_-\\_'\\"_ ca "///////,4Y////(S///////////2_/////////)I Detritus Scattering III///////IIIII/I/IIII/i.S_I/////////////////,"////A Gelbstoff y////,///////////y.@y////'s/////////////////////////////,y,, ] 0.00 0.02 0.04 ] H(490:555) H(443:555) OCTS Clark 0.06 0.080 0.10 0.12 Error in Pigment Retrieved per Onit Pigment Fig. 16. Relative error in retrieval of Ca + Cp at unit chlorophyll for the recommended 490:555 algorithm, the H(443:555) algorithm, the Ocean Color Temperature Sensor (OCTS) algorithm, and the Clark four-band algorithm at unit pigment for individual biogeochemical measures. The final algorithm adjustment is r LwN( 443) 7 (2s) c: =1455co 02v9! and (29) c"+ = 12s0[co+cp] 0.163] ½ The correction is applied where pigment concentration is less than 2 mg m -3, i.e., when LWN (443) is valid. This corrected pigment concentration results in an overall (0.02- 50 mg m -3) R 2 of 91.1%, and an R 2 of 90.9% when pigment concentration is less than 2mgm -3. The improvement is 8.6% when compared with the composite single-band algorithm. It is envisaged that when the algorithm is used for biogeochemical provinces with greater variation in pigment type, the algorithm will show a greater improvement in the explained variance. 7. CONCLUSIONS 1. Bio-optical models, show that there is a sound theoretical basis for band-ratio algorithms with explicit solutions which relate to the IOPs of water and its constituents. This indicates that the measurement of IOPs for algorithm development may be as appropriate as the measument of apparent optical properties (AOPs), e.g., R(,). 2. The BSM developed for this study has proved a powerful tool for the development of algorithms, and has the potential for further parameterization, so as to achieve closure with in situ data. 3. Biological coupling, as indexed by the major pigment groups, constrains and restricts the bio-optical variability (i.e., there is less bio-optical variability than would be expected); thus, there are sound grounds for an unbiased composite photosynthetic pigment algorithm. 4. Although globally robust, there is interprovince variability in the chlorophyll to carotenoid ratio. This variability implies that the accuracy of two band single pigment algorithms, e.g., chlorophyll a or PSC will be limited. This fact does not affect the utility of global photosynthetic pigment algorithms, however, and provides a basis for constructing high accuracy province-specific aigorithms. 5. Factor analysis of band ratio combinations (all Sea- WiFS two-band combinations) demonstrates that discrete band ratios contain the same variance as similar studies using complete spectra. 29

Original page 32

The SeaWiFS CZCS-Type Pigment Algorithm 6. Optimal band ratio combination algorithms provide some measure of confidence that a universal algorithm can be derived (within the bounds of current data sets), i.e., both empirical methods (conclusion 1) and bio-optical models (conclusion 2) point to a similar conclusion with respect to a universal photosynthetic pigment algorithm. the Sterna Not an acronym, but a BOFS Antarctic research 7. Within the radiometric constraints of SeaWiFS, 490:555 band combination provides the most robust retrieval of pigment or chlorophyll over five decades of pigment level. 8. With the exception of the total pigment regression (Table 19), the BSM data indicates there is no extra information to be gained by using multiband algorithms with CZCS data. The hyperbolic 443:555 PML Plymouth Marine Laboratory (United Kingdom) POLDER Polarization and Directionality of the Earth's Reflectances (France) PPC Photoprotectant Carotenoids PSC Photosynthetic Carotenoids SeaWiFS Sea-viewing Wide Field-of-view Sensor SPO SeaWiFS Project Office SPSWG SeaWiFS Prelaunch Science Working Group project. SYMBOLS a(A) The absorption coefficient. a I The absorption at the Raman excitation wavelength. aa The specific absorption of chlorophyll a. aabc The specific absorption of chlorophylls a, b, and c. ab The specific absorption of chlorophyll b. and 510:555 (with ammended coefficients) algo- ac The specific absorption of chlorophyll c. rithms can be used with CZCS data, as an alternative to log-log regression." ACKNOWLEDGMENTS The authors thank Jim Mueller of CHORS for advice and suggestions, especially for Section 4. In addition, Ray Barlow a9 The DOM/detritus specific absorbance. app The specific absorption of PPC. aps The specific absorption of PSC. aw The absorption coefficient of water. The DOM/chlorophyll combined absorbance. A_ An arbitrary constant. Aj An arbitrary constant. of A_ An arbitrary constant. PML, is acknowledged for supplying unpublished pigment data A; An arbitrary constant. and comments on Section 3. GLOSSARY AOL Airborne Oceanographic Lidar AOP Apparent Optical Property BATS Bermuda Atlantic Time-Series Station BOAWG Bio-Optical Algorithm Working Group BOFS Biogeochemical Ocean Flux Study BSM Bio-Optical Synthetic Model Case-1 Water whose reflectance is determined solely by absorption. Case-2 Water whose reflectance is significantly influenced by scattering. CZCS Coastal Zone Color Scanner DOM Dissolved Organic Matter EqPac Equatorial Pacific GIN Greenland, Iceland, and Norwegian (Seas) HPLC High Performance Liquid Chromatography IOP Inherent Optical Property IR Infrared NABE North Atlantic Bloom Experiment NASA National Aeronautics and Space Administration NEAT Northeast Atlantic NET NIMBUS Experiment Team NIMBUS Not an acronym, but a series of NASA experimental weather satellites containing a wide variety atmosphere, ice, and ocean sensors. OCTS Ocean Color Temperature Sensor (Japan) 3O The backscatter coefficient. bbp The particle specific backscatter coefficient (usually normalized to chlorophyll a concentration). bbw The backscatter coefficient of water. brain Scattering associated with phytoplankton (Prieur and Sathyendranath, 1981). bn The Raman scattering coefficient. B An empirical constant. Bb An empirical constant dependant on the backscatter ratio. c(A) The spectral attenuation coefficient. C Chlorophyll concentration. Ca The concentration of chlorophyll a. Cabc The concentration of chlorophylls a, b, and c. Cb The concentration of chlorophyll b. Cc The concentration of chlorophyll c. Cp Phaeopigment concentration. Cpp PPC concentration. Cps PSC concentration. Cs Simulated C. CTp Total pigment concentration. E_ The downwelling irradiance at the Raman excitation wavelength. F0 Extra terrestrial irradiance. F1 Pigment biomass loading factor. F2 Detritus concentration loading factor. F3 Carotenoid concentration (or relative pigment abundance) loading factor. g A constant that consists of the ratios of the air-sea of interface effects, the effects of the light field, and the relative spectral variation of Q. The concentration of DOM and DOM-like absorbers. The effect of the downwelling light field.

Original page 33

J. Aiken, G.F. Moore, C.C. Trees, S.B. Hooker, and D.K. Clark Pigment calculated from the hyperbolic transform Bidigare, R.R., M.E. Ondrusek, J.H. Morrow, and D.A. Kiefer, of L,:j. k An index to two vectors of band ratios kl and k2. Bricaud, A., A. Morel, and L. Prieur, 1981: Absorption by kl A band ratio vector. k2 A band ratio vector. L_:j The ratio of normalized water-leaving radiances at Carder, K.L., R.G. Steward, J.H. Paul, and G.A. Vargo, 1986: wavelengths i ()) to j (Aj): LWN())/LwN(,kj). Lw_(_) Normalized water-leaving radiance. L'N Normalized water-leaving radiance at the Raman excitation wavelength. --, R.G. Steward, G.R. Harvey, and P.B. Ortner, 1989: Man The index of refraction. P The particulate concentration including detrital ma- 1990: In vivo absorption properties of algal pigments. SPIE Ocean Optics, 1302, 290-302. dissolved organic matter of the sea (yellow substance) in the UV and visible domains. Limnol. Oceanol., 26, 45-53. Relationships between chlorophyll and ocean color constituents as they affect remote-sensing reflectance models. Limnol. Oceanogr., 31,403-413. rine humic and fulvic acids: Their effects on remote sensing of ocean chlorophyll. Lirnnol. Oeeanogr., 34, 38-81. terial. Clark, D.K., 1981: Phytoplankton pigment algorithms for the Ps Simulated C + Cp (q.v.). Q The ratio of upwelling irradiance to radiance, which Nimbus-7 CZCS. Oceanography from Space, J.F.R. Gower, Ed., Plenum Press, 227-238. varies with the angular distribution of the upwelling Duysens, L.N.M., 1956: The flattening of the absorption speclight field, and is rr for an isotropic distribution. r The air-water reflectance for diffuse irradiance. R 2 The regression coefficient. R(A) The irradiance reflectance at a particular wavelength. S The slope of a line. a' A power law constant. a0 A curve fitting constant. al A curve fitting constant. (_2 A curve fitting constant. ' A power law constant. A' Raman excitation wavelength. A Wavelength of light at a particular band. Aj Wavelength of light at a particular band. p The Fresnel reflectance at normal incidence. _5 The Fresnel reflectance for sun and sky irradiance. REFERENCES Aiken, J., and G. Moore, 1995: Special requirements for the validation of ocean colour information. Proceedings of the WMO/IOC Conference on Space-Based Ocean Observation, WMO/PD-No. 649, 93-101. trum of suspensions as compared with that of solutions. Biochim. Biophys. Acta., 19, 255, 257, 261. Garver, S.A., D.A. Siegel, and B.G. Mitchell, 1995: Variability in near surface particulate absorption spectra: What can a satellite imager see? Limnol. Oceanogr., 39, 1,349-1,367. Gordon, H.R., O.B. Brown, R.H. Evans, J.W. Brown, R.C. Smith, K.S. Baker, and D.K. Clark, 1988: A semianalytic radiance model of ocean color. J. of Geophys. Res., 93, 10,909-10,924. Hoepffner, N., and S. Sathendranath, 1992: Bio-optical characteristics of coastal waters: Absorption spectra of phytoplankton and pigment distribution in the Western North Atlantic. Limnol, Oceanogr., 37, 1,660-1,679. Hoepffner, N., and S. Sathyendranath, 1993: Determination of the major groups of phytoplankton pigments from the absorption spectra of total particulate matter. J. Geophys. Res., 98, 22,789-22,803. Hoge, F.E., and R.E. Swift, 1993: The influence of chlorophyll pigment upon the upwelling spectral radiances from the North Atlantic Ocean. Deep-Sea Res., 40, 265-278. Holligan, P.M., E. Fernandez, J. Aiken, W.B. Balch, P. Boyd, PH. Burkill, M. Finch, S.B. Groom, G. Malin, K. Muller, D.A. Purdie, and C. Robinson, 1993: A biogeochemical study of Emiliania huxleyi, in the North Atlantic. Global Biogeochemical Cycles, 7, 879-900. --, G. Moore, and P. Holligan, 1992: Remote sensing of Hooker, S.B., C.R. McClain, and A. Holmes, 1993: Ocean color oceanographic biology in relation to global climate change. J. Phyeol., 28, 579-590. Barnes, R.A., A.W. Holmes, W.L. Barnes, W.E. Esaias, C.R. McClain, and T. Svitek, 1994: SeaWiFS Prelaunch Radiometric Calibration and Spectral Characterization. NASA Tech. Memo. 104566, Vol. 23, S.B. Hooker, E.R. Firestone, and J.G. Acker, Eels., NASA Goddard Space Flight Center, Greenbelt, Maryland, 55 pp. imaging: CZCS to SeaWiFS. Marine. Tech. Soc. J., 27, 3- 15. Mantoura, R.F.C., and C.A. Llewellyn, 1983: The rapid determination of algal chlorophyll and carotenoid pigments and their breakdown products in natural waters by reversephase high-performance liquid chromatography. Anal. Chim. Acta., 151, 297-314. Marshall, B.R., and R.C. Smith, 1990: Raman scattering and in-water ocean optical properties. Appl. Opt., 29, 71-84. 31

Original page 34

TheSeaWiFSCZCS-TypePigmentAlgorithm prow THE SEAWIFS TECHNICAL REPORT SERIES Mitchell,B.G.,andO. Holm-Hansen,1991:Bio-optical ertiesof AntarcticPeninsulawaters:differentiationfrom temperateoceanmodels.Deep-Sea Res., 39, 1,009-1,028. Morel, A., 1974: Optical properties of pure water and sea water. Optical Aspects of Oceanography, N.G. Jerlov and S. Nielsen, Eds., Academic Press, 1-24. , and L. Prieur, 1977: Analysis of variation in ocean colour. Limnol. Oceanogr., 22, 709-722. --, and B. Gentili, 1991: Diffuse reflectance of oceanic waters. I. Its dependence on sun angle as influenced by the molecular scattering contribution. Appl. Opt., 30, 4,427- 4,438. , and B. Gentili, 1993: Diffuse reflectance of oceanic waters. II. Bidirectional aspects. Appl. Opt., 32, 6,864-6,879. Mueller, J.L., 1976: Ocean color spectra measured off the Oregon coast: characteristic vectors. Appl. Opt., 15, 394- 402. , and R.W. Austin, 1995: Ocean Optics Protocols for Sea- WiFS Validation, Revision 1. NASA Tech. Memo. 104566, Vol. 25, S.B. Hooker, E.R. Firestone, and J.G. Acker, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 66 pp. Petzold, T.J., 1972: Volume scattering functions for selected ocean waters. Scripps Inst. Oceanogr. Rep. 72-78, 79 pp. Prieur, L., and S. Sathyendranath, 1981: An optical classification of coastal and oceanic waters based on the specific spectral absorption curves of phytoplankton pigments, dissolved organic matter, and other particulate materials. Limnol. Oceanogr., 26,671-689. Sathyendranath, S., 1981: Influence des substances en solution Vol. 1 Hooker, S.B., W.E. Esaias, G.C. Feldman, W.W. Gregg, and C.R. McClaln, 1992: An Overview of SeaWiFS and Ocean Color. NASA Tech. Memo. 104566, Vol. 1, S.B. Hooker and E.R. Firestone, Eds., NASA Goddaxd Space Flight Center, Greenbelt, Maryland, 24 pp., plus color plates. Vol. 2 Gregg, W.W., 1992: Analysis of Orbit Selection for SeaWiFS: Ascending vs. Descending Node. NASA Tech. Memo. 104566, Vol. 2, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 16pp. Vol. 3 McClain, C.R., W.E. Esaias, W. Barnes, B. Guenther, D. Endres, S. Hooker, G. Mitchell, and R. Barnes, 1992: Calibration and Validation Plan for SeaWiFS. NASA Tech. Memo. 104566, Vol. 3, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 41 pp. Vol. 4 McClain, C.R., E. Yeh, and G. Fu, 1992: An Analysis of GAC Sampling Algorithms: A Case Study. NASA Tech. Memo. 104566, Vol. 4, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 22 pp., plus color plates. Vol. 5 Mueller, J.L., and R.W. Austin, 1992: Ocean Optics Protocols for SeaWiFS Validation. NASA Tech. Memo. 10,4566, Vol. 5, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 43 pp. et en suspension dans les eaux de met sur l'absorption et la Vol. 6 reflectance. Modelisation et application a la teledetection. Ph.D. Thesis, University of Paris, 123 pp (in French). Smith, R.C., and K.S. Baker, 1981: Optical properties of the clearest natural waters (200-800nm). Appl. Opt., 20, 177- 184. Strickland, J.D.H., and T.R. Parsons, 1972: A Practical Handbook of Sea Water Analysis. Fish. Res. Board. Canada, 310 pp. Sullivan, C.W., K.R. Arrigo, C.R. McClain, J.C. Comiso, and J.K. Firestone, 1993: Distributions of phytoplankton blooms in the Southern Ocean. Science, 262, 1,832-1,837. Trees, C.C., M.C. Kenicutt, and J.M. Brooks, 1985: Errors associated with the standard fluorometric determination of chlorophylls and phaeopigments. Mar. Chem., 17, 1-12. --, D.K. Clark, R. Bidigare, and M. Ondrusek, 1995: Chlorophyll a versus accessory pigment concentrations within the euphotic zone: A ubiquitous relationship? Science, (submitted). Yentsch, C.S., and D.W. Menzel, 1963: A method for determination of phytoplankton, chlorophyll and phaeophytin by fluorescence. Deep-Sea Res., 10, 221-231. 32 Firestone, E.R., and S.B. Hooker, 1992: SeaWiFS Technical Report Series Summary Index: Volumes 1-5. NASA Tech. Memo. 104566, Vol. 6, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 9 pp. Vol. 7 Darzi, M., 1992: Cloud Screening for Polar Orbiting Visible and IR Satellite Sensors. NASA Tech. Memo. 104566, Vol. 7, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 7pp. Vol. 8 Hooker, S.B., W.E. Esaias, and L.A. Rexrode, 1993: Proceedings of the First SeaWiFS Science Team Meeting. NASA Tech. Memo. 104566, Vol. 8, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 61 pp. Vol. 9 Gregg, W.W., F.C. Chen, A.L. Mezaache, J.D. Chen, J.A. Whiting, 1993: The Simulated SeaWiFS Data Set, Version 1. NASA Tech. Memo. 104566, Vol. 9, S.B. Hooker, E.R. Firestone, and A.W. Indest, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 17 pp.

Original page 35

J. Aiken,G.F.Moore,C.C.Trees, S.B. Hooker, and D.K. Clark Vol. 10 Woodward, R.H., R.A. Barnes, C.R. McClain, W.E. Esaias, W.L. Barnes, and A.T. Mecherikunnel, 1993: Modeling of the SeaWiFS Solar and Lunar Observations. NASA Tech. Memo. 104566, Vol. 10, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 26 pp. Vol. 11 Patt, F.S., C.M. Hoisington, W.W. Gregg, and P.L. Coronado, 1993: Analysis of Selected Orbit Propagation Models for the SeaWiFS Mission. NASA Tech. Memo. 104566, Vol. 11, S.B. Hooker, E.R. Firestone, and A.W. Indest, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 16pp. Vol. 12 Firestone, E.R., and S.B. Hooker, 1993: SeaWiFS Technical Report Series Summary Index: Volumes 1-11. NASA Tech. Memo. 104566, Vol. 12, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 28 pp. Vol. 13 McClain, C.R., K.R. Arrigo, J. Comiso, R. Fraser, M. Daxzi, J.K. Firestone, B. Schieber, E-n. Yeh, and C.W. Sullivan, 1994: Case Studies for SeaWiFS Calibration and Validation, Part 1. NASA Tech. Memo. 104566, Vol. 13, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 52pp., plus color plates. Vol. 14 Mueller, J.L., 1993: The First SeaWiFS Intercalibration Round- Robin Experiment, SIRREX-1, July 1992. NASA Tech. Memo. 104566, Vol. 14, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 60 pp. Vol. 15 Gregg, W.W., F.S. Patt, and R.H. Woodward, 1994: The Simulated SeaWiFS Data Set, Version 2. NASA Tech. Memo. 104566, Vol. 15, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 42pp., plus color plates. Vol. 16 Mueller, J.L., B.C. Johnson, C.L. Cromer, J.W. Cooper, J.T. McLean, S.B. Hooker, and T.L. Westphal, 1994: The Second SeaWiFS Intercalibration Round-Robin Experiment, SIRREX-2, June 1993. NASA Tech. Memo. 104566, Vol. 16, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 121 pp. Vol. 17 Abbott, M.R., O.B. Brown, H.R. Gordon, K.L. Carder, R.E. Evans, F.E. Muller-Karger, and W.E. Esaias, 1994: Ocean Color in the 21st Century: A Strategy for a 20-Year Time Series. NASA Tech. Memo, 104566, Vol. 17, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 20 pp. Vol. 18 Firestone, E.R., and S.B. Hooker, 1995: SeaWiFS Technical Report Series Summary Index: Volumes 1-17. NASA Tech. Memo. 104566, Vol. 18, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 47 pp. Vol. 19 McClain, C.R., R.S. Fraser, J.T. McLean, M. Darzi, J.K. Firestone, F.S. Patt, B.D. Schieber, R.H. Woodward, E-n. Yeh, S. Mattoo, S.F. Biggar, P.N. Slater, K.J. Thome, A.W. Holmes, R.A. Barnes, and K.J. Voss, 1994: Case Studies for SeaWiFS Calibration and Validation, Part 2. NASA Tech. Memo. 104566, Vol. 19, S.B. Hooker, E.R. Firestone, and J.G. Acker, Eels., NASA Goddard Space Flight Center, Greenbelt, Maryland, 73 pp. Vol. 20 Hooker, S.B., C.R. McClain, J.K. Firestone, T.L. Westphal, E-n. Yeh, and Y. Ge, 1994: The SeaWiFS Bio-Optical Archive and Storage System (SeaBASS), Part 1. NASA Tech. Memo. 104566, Vol. 20, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 40 pp. Vol. 21 Acker, J.G., 1994: The Heritage of SeaWiFS: A Retrospective on the CZCS NIMBUS Experiment Team (NET) Program. NASA Tech. Memo. 104566, Vol. 21, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 43 pp. Vol. 22 Barnes, R.A., W.L. Barnes, W.E. Esaias, and C.R. McClain, 1994: Prelaunch Acceptance Report for the SeaWiFS Radiometer. NASA Tech. Memo. 104566, Vol. 22, S.B. Hooker, E.R. Firestone, and J.G. Acker, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 32 pp. Vol. 23 Barnes, R.A., A.W. Holmes, W.L. Barnes, W.E. Esaias, C.R. McClain, and T. Svitek, 1994: SeaWiFS Prelaunch Radiometric Calibration and Spectral Characterization. NASA Tech. Memo. 104566, Vol. 23, S.B. Hooker, E.R. Firestone, and J.G. Acker, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 55 pp. Vol. 24 Firestone, E.R., and S.B. Hooker, 1995: SeaWiFS Technical Report Series Summary Index: Volumes 1-23. NASA Tech. Memo. 104566, Vol. 24, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 36 pp. Vol. 25 Mueller, J.L., and R.W. Austin, 1995: Ocean Optics Protocols for SeaWiFS Validation, Revision 1. NASA Tech. Memo. 104566, Vol. 25, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 66 pp. Vol. 26 Siegel, D.A., M.C. O'Brien, J.C. Sorensen, D.A. Konnoff, E.A. Brody, J.L. Mueller, C.O. Davis, W.J. Rhea, and S.B. Hooker, 1995: Results of the SeaWiFS Data Analysis Round-Robin (DARR), July 1994. NASA Tech. Memo. 104566, Vol. 26, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 58 pp. Vol. 27 J.L. Mueller, R.S. Fraser, S.F. Biggar, K.J. Thome, P.N. Slater, A.W. Holmes, R.A. Barnes, C.T. Weir, D.A. Siegel, D.W. Menzies, A.F. Michaels, and G. Podesta, 1995: Case Studies for SeaWiFS Calibration and Validation, Part 3. NASA Tech. Memo. 104566, Vol. 27, S.B. Hooker, E.R. Firestone, and J.G. Acker, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 46 pp. 33

Original page 36

TheSeaWiFSCZCS-TypePigmentAlgorithm vo/. 2s McClain, C.R., K.R. Arrigo, W.E. Esaias, M. Darzi, F.S. Patt, R.H. Evans, J.W. Brown, C.W. Brown, R.A. Barnes, and L. Kumar, 1995: SeaWiFS Algorithms, Part 1. NASA Tech. Memo. 104566, VoL 28, S.B. Hooker, E.R. Firestone, and J.G. Acker, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 38 pp., plus color plates. 34 Vol. 29 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, S.B. Hooker and E.R. Firestone, Eds., NASA Goddard Space Flight Center, Greenbelt, Maryland, 34 pp.

Original page 37

Original page 38

Form Approved REPORT DOCUMENTATION PAGE ousNo,o7o4-o188 Public reporting burden for this collection of information is eslimated to average 1 hour per response, including he time for rev ew ng nstructions, searching existing data sources gathering and mainlaining the data needed, and completing and reviewing the collection of information Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway. Suite 1204, Arlington, VA 22202-4302, and to the Office of Mana_ement and Budget. Paperwork Reduction Proiecl (0704-0188), Washin_lton, D,C 20503. 1. AGENCY USE ONLY (Leave blank) 2. REPORT DATE June 1995 4. TITLE AND SUBTITLE SeaWiFS Technical Report Series 3. REPORT TYPE AND DATES COVERED Technical Memorandum 5. FUNDING NUMBERS Volume 29-The SeaWiFS CZCS-Type Pigment Algorithm Code 970.2 6. AUTHOR(S) James Aiken, Gerald F. Moore, Charles C. Trees, Stanford and Dennis K. Clark Series Editors: Stanford B. Hooker and Elaine R. Firestone 7. PERFORMING ORGANIZATION NAME(S) AND ADDRESSEES) Laboratory for Hydrospheric Processes Goddard Space Flight Center Greenbelt, Maryland 20771 B. Hooker, 8. PERFORMING ORGANIZATION REPORT NUMBER 95B00096 NAME(S) AND ADDRESS(ES) 10. SPONSORING/MONITORING 9. SPONSORING/MONITORING AGENCY National Aeronautics and Space Administration Washington, D.C. 20546--0001 11. SUPPLEMENTARY NOTES Elaine R. Firestone: General AGENCY REPORT NUMBER TM-104566, Vol. 29 Sciences Corporation, Laurel, Maryland; James Aiken and Gerald Moore: Plymouth Marine Laboratory, Plymouth, United Kingdom; Charles C. Trees: San Diego State University, San Diego, California; Dennis K. Clark: NOAA/National Environmental 1211. DISTRIBUTION/AVAILABILITY STATEM E NT Unclassified-Unlimited Subject Category 48 Satellite Data Information Service, Camp Springs, Maryland DISTRIBUTION CODE Report is available from the Center for AeroSpace Information (CASI), 2b. 800 Elkridge Landing Road, Linthicum Heights, MD 21090; (301)621-0390 13. ABSTRACT (Max/_ 200 words) The Sea-viewing Wide Field-of-view Sensor (SeaWiFS) mission will provide operational ocean color that will be superior to the previous Coastal Zone Color Sensor (CZCS) proof-of-concept mission. An algorithm is needed that exploits the full functionality of SeaWiFS whilst remaining compatible in concept with algorithms used for the CZCS. This document describes the theoretical rationale of radiance bandratio methods for determining chlorophyll a and other important biogeochemical parameters, and their implementation for the SeaWlFS mission. Pigment interrelationships are examined to explain the success of the CZCS algorithms. In the context where chlorophyll a absorbs only weakly at 520 nm, the success of the 520 nm to 550 nm CZCS band ratio needs to be explained. This is explained by showing that in pigment data from a range of oceanic provinces chlorophyll a (absorbing at less than 490 nm), carotenoids (absorbing at greater than 460 rim), and total pigment are highly correlated. Correlations within pigment groups particularly photoprotectant and photosynthetic carotenoids are less robust. The sources of variability in optical data are examined using the NIMBUS Experiment Team (NET) bio-optical data set and bio-optical model. In both the model and NET data, the majority of the variance in the optical data is attributed to variability in pigment (chlorophyll a, and total particulates, with less than 5% of the variability resulting from pigment assemblage. The relationships between band ratios and chlorophyll is examined analytically, and a new formulation based on a dual hyperbolic model is suggested which gives a better calibration curve than the conventional log-log linear regression fit. The new calibration curve shows the 490:555 ratio is the best single-band ratio and is the recommended CZCS-type pigment algorithm. Using both the model and NET data, a number of multiband algorithms are developed; the best of which is an algorithm based on the 443:555 and 490:555 ratios. From model data, the form of potential algorithms for other products, such as total particulates and dissolved organic matter (DOM), are suggested. 14. _LIBJECT TERMS SeaWiFS, Oceanography, Algorithms, Pigments, CZCS, Data Analyses, NET Data 17. SECURITY _TION llL SECURITY CLASSIFICATION OF REPORT OF THIS PAGE Unclassified Unclassified 15. NUMBER OF PAGES Band-Ratio, Bio-Optical Models 34 16. PRICE CODE 19. SECURITY CLASSIFICATION 20. LIMITATION OF ABSTRACT OF ABSTRACT Unlimited Unclassified