lable at ScienceDirect

Estuarine, Coastal and Shelf Science 93 (2011) 396e402

Contents lists avai

Estuarine, Coastal and Shelf Science

journal homepage: www.elsevier .com/locate/ecss

An optical model for the remote sensing of coloured dissolved organic matterin coastal/ocean waters

S.P. Tiwari, P. Shanmugam*

Ocean optics and Imaging Group, Department of Ocean Engineering, Indian Institute of Technology Madras, Chennai 600036, India

a r t i c l e i n f o

Article history:Received 1 July 2010Accepted 5 May 2011Available online 18 May 2011

Keywords:coloured dissolved organic matteroceanographyoptical modellingremote sensingSeaWiFScoastal waters

* Corresponding author.E-mail address: pshanmugam@iitm.ac.in (P. Shanm

0272-7714/$ e see front matter � 2011 Elsevier Ltd.doi:10.1016/j.ecss.2011.05.010

a b s t r a c t

An optical model is developed for the remote sensing of coloured dissolved organic matter (CDOM) ina wide range of waters within coastal and open ocean environments. The absorption of CDOM (denotedas ag) is generally considered as an exponential form model, which has two important parameters e theslope S and absorption of CDOM at a reference wavelength ag(l0). The empirical relationships for derivingthese two parameters are established using in-situ bio-optical datasets. These relationships use thespectral remote sensing reflectance (Rrs) ratio at two wavelengths Rrs(670)/Rrs(490), which avoids theknown atmospheric correction problems and is sensitive to CDOM absorption and chlorophyll in coastal/ocean waters. This ratio has tight relationships with ag(412) and ag(443) yielding correlation coefficientsbetween 0.77 and 0.78. The new model, with the above parameterization applied to independentdatasets (NOMAD SeaWiFS match-ups and Carder datasets), shows good retrievals of the ag(l) withregression slopes close to unity, little bias and low mean relative and root mean square errors. Thesestatistical estimates improve significantly over other inversion models (e.g., Linear Matrix-LM andGarver-Siegel-Maritorena-GSM semi-analytical models) when applied to the same datasets. These resultsdemonstrate a good performance of the proposed model in both coastal and open ocean waters, whichhas the potential to improve our knowledge of the biogeochemical cycles and processes in thesedomains.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

Coloured dissolved organic matter (CDOM), defined as materialspassing through a 0.2 mm filter and often described as yellowsubstance or gelbstoff, plays a critical role in a broad range ofmarine aquatic ecosystems. It primarily absorbs ultraviolet andblue light radiation in 350e500 nm range and plays an importantrole in determining the underwater light fields. In addition, CDOMrepresents a significant component of ocean optical signals forsatellite-based measurements of ocean colour and can interfere inglobal and regional estimates of primary production. The opticallyactive fraction of CDOM affects the ocean colour, underwater lightfields and aquatic chemistry through a suite of sunlight-initiatedphotochemical processes. CDOM is a highly complex macromo-lecular material containing humic and fulvic [check fulvic at proof]substances. In the open ocean, where coastal runoff and riverineinput are negligible on annual time scales and Chlorophyll-a concentrations (Chl-a) are typically less than 0.5 mg m�3, CDOMexhibits a featureless absorption spectrum that decreases

ugam).

All rights reserved.

exponentially with increasing wavelength from ultraviolet (UV)into visible wavelength and influences the spectral distribution andlight availability in the water column. The CDOM spectral slope (S)indicates the rate at which the CDOM absorption decreases withincreasing wavelength. Changes in the shape of CDOM absorptionspectrum or S have been attributed to solar photo-bleaching orphoto-oxidation (increase in S) or to the differing nature of CDOMsources (Blough and Del Vecchio, 2002; Twardowski and Donaghay,2002), thus providing additional insights into the nature of CDOM.Moreover, absorption of sunlight by CDOM has important impli-cations to carbon cycling in the marine environment. For example,the process of photo-oxidation can result in photoproducts anda variety of organic compounds with low molecular weights (Gaoand Zepp, 1998). Absorption by CDOM can mitigate the damagingeffect of solar UV radiation in the aquatic system, while its loss dueto photo-oxidation can decrease absorption in the UV and visiblespectral regions. In the visible range of the electromagnetic spec-trum, CDOM absorption can reduce the amount of photosyntheti-cally active radiation (PAR) available to the phytoplankton and thuscan affect primary productivity or can interfere with satellitedeterminations of seawater constituents such as phytoplanktonpigments (Blough and Del Vecchio, 2002).

S.P. Tiwari, P. Shanmugam / Estuarine, Coastal and Shelf Science 93 (2011) 396e402 397

The spectral variations of absorption and scattering of clear andturbid waters have been studied over the wavelength range from400 to 700 nm. The behaviour of S in natural waters is highlysensitive with respect to the wavelength range over which it isestimated. In CDOM studies, variability in S is associated withchanges in the composition of the CDOM pool (Carder et al., 1989),and in the open ocean it is usually associated with degradation(solar and bacterial) and mixing of different CDOM pools (Brown,1977; Gao and Zepp, 1998). However, the source of CDOM (inconcert with the blue colour reflected by ocean water) can betracked effectively. For instance, the largest source of CDOM is endmembers (riverine and marine) that are determined from theconservative mixing model (Kowalczuk et al., 2006). Stedmon andMarkager (2001) established an algorithm to distinguish CDOM ofterrestrial or marine origin based on the relationship betweenCDOM absorption and the spectral slope.

Modelling the spectral shape of the absorption coefficientspectra ag(l) (m�1) in the optical region has been reported byvarious researchers (Bricaud et al., 1981; Roesler et al., 1989;Kowalczuk et al., 2006), who have shown that it should followthe exponential decay function. Therefore, the spectral informationwith respect to ag(l) can only be obtained by measuring ag ata specific wavelength (e.g., 412 nm) and with known mean slope ofexponential decay constant. The S coefficients change with wave-length and season, and co-vary with CDOM concentration.

For developing remote sensing algorithms, a complete charac-terization of the water is required. For instance, understanding thevariability of inherent and apparent optical properties and waterconstituents is essential for correct retrievals of the CDOM, sus-pended sediment (SS) and Chl concentrations. Remote sensingalgorithms have been developed for many optical properties ofseawater, including spectral absorption and spectral backscat-tering, and are frequently evaluated in a number of regional fieldstudies. Spectral band ratios are widely used to interpret theremote sensing data. In coastal waters where the spatial distribu-tion of optically active constituents is variable, these ratios seem tohave local and regional characteristics (Kahru and Mitchell, 1998;Kutser et al., 2001). However, various empirical and semi-analyt-ical algorithms have been additionally proposed for the retrieval ofCDOM absorption coefficient (ag) from remote sensing data at thespecific wavelengths (Kahru and Mitchell, 1998; Kutser et al., 2001;Maritorena et al., 2002; Boss and Roesler, 2006). These algorithms/models fail to provide the spectral dependence of ag, which mayvary from region to region according to the specific biogeochemicalconstituents of CDOM.

0.0 0.2 0.4 0.6 0.8 1.0 1.20

5

10

15

20

Freq

uenc

y

Insitu_ag(443) (m-1)

a

Fig. 1. Histograms of the (a) ag(443) (m�1) and (b) Rrs(4

The objectives of this paper are (1) to develop a robust opticalmodel for the remote sensing of CDOM in coastal and open oceanwaters, and (2) to validate the proposed model using in-situ bio-optical datasets as well as satellite match-ups datasets.

2. Data and methods

2.1. Data sets

An updated NASA bio-Optical Marine Algorithm Dataset (here-after referred to as NOMAD) was obtained from the NASA OceanBiology Processing Group. It consists of two types of datasets e in-situ bio-optical dataset and satellite corresponding data withconcurrent SeaWiFS observations of remote sensing reflectance(Rrs) and in-situ ag at keywavelengths. The in-situ datasets are highquality data acquired over 4459 stations and stored in the systemfor use in algorithm development and validation (O’Reilly et al.,1998, 2000). Fig. 1 shows the frequency distribution of the in-situRrs(443) and ag(490) collected in coastal and ocean waters(N ¼ 223). The lower end of the histograms represents clear waterswhile the higher end represents the coastal waters. These datawereused to develop the empirical relationships for the new model.Another suite of NOMAD dataset (composing NOMAD in-situ data)includes SeaWiFS-remote sensing reflectances and coincidentlymeasured in-situ ag(l) data from the coastal and open oceanregions (NOMAD-2). These satellite corresponding data togetherwith the Carder bio-optical dataset (N ¼ 618) obtained duringcruises in the west coast of Florida in different seasons and yearsfrom 1999 to 2006 were used to validate the performance of theproposed model.

2.2. Performance assessment

To assess the performance of the model, two basic statisticalmeasures were used: the root mean square error (RMSE, randomerror) and mean normalized bias (MNB, systematic error). Inaddition, the slope (S), intercept (I) and correlation coefficient (R2)were also obtained from regression analysis for further assessment.Because the in-situ and model data are independent and are bothsubject to errors, systematic and random errors were calculated asfollows:

RMSE ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1n

Xni¼1

ðMi � IiÞ2vuut (1)

0.000 0.005 0.010 0.015 0.020 0.025 0.030 0.0350

5

10

15

Freq

uenc

y

Insitu Rrs(490) (sr-1)

b

90) (sr�1) from NOMAD in-situ datasets (N ¼ 233).

S.P. Tiwari, P. Shanmugam / Estuarine, Coastal and Shelf Science 93 (2011) 396e402398

MNB ¼ 1XnððMi � IiÞ=IiÞ (2)

ni¼1

where Mi is the model-derived values, Ii is the in-situ measure-ments, and n is the number of valid retrievals. All these statisticalmeasures for the derived ag were calculated for the wavelengths400, 412, 443, 490, 510 and 555 nm.

3. Results

3.1. CDOM absorption

3.1.1. Spectral absorption modelThe CDOM absorption spectrum is obtained using a simple

single exponential model:

agðlÞ ¼ agðlrÞe�sðl�lrÞ (3)

where ag(l) and ag(lr) are the absorption coefficients at a desiredwavelength and a reference wavelength, respectively, and S is thespectral slope coefficient (nm�1) of an exponent that determinesthe shape of the absorption curve (Lundgren, 1976; Bricaud et al.,1981; D’Sa et al., 1999; Stedmon et al., 2000). Moreover, S is oftenused as a proxy for CDOM composition including the ratio of fulvicand humic acids and molecular weight. The CDOM absorptionspectrum could be determined by measuring the absorption at twowavelengths and subsequently by fitting a straight line betweenthe absorption measurements plotted at logarithmic scale. Whenthe slope of the exponential is fixed then even a single absorptionmeasurement would be sufficient. The average S values chosen forthe remote sensing bio-optical models, which incorporate thesingle exponential model (Carder et al., 1999), are between 0.014and 0.015 nm�1 from the historical precedence.

3.1.2. Empirical relationships for deriving ag (412) and ag (443) andestimating S coefficient

To describe the CDOM absorption as a function of wavelength l,the slope parameter S and ag at a reference wavelength (especiallywithin the blue domain, 412 and 443 nm, where high absorptionoccurs) needed to be determined. Based on the analysis usingNOMAD in-situ bio-optical dataset, satellite-detecting bands forCDOM absorption were chosen as 490 and 670 nm because thesebands are free from atmospheric correction errors (Johannessenet al., 2003; Belanger et al., 2008). To estimate absorption coeffi-cients of CDOM at short wavelengths, relationships between the

1E-3 0.01 0.1 1 101E-3

0.01

0.1

1

10

a g(412

) (m

-1)

Rrs

(670)/Rrs(490)

-1

a

Fig. 2. Relationships between the ag(412) and ag(443) and remote sensing refle

log-transformed ag at 412 and 443 nm and log-transformed spec-tral ratio of the remote sensing reflectance Rrs(670)/Rrs(490) wereestablished using the NOMAD in-situ bio-optical dataset (Fig. 2). Agoodness of fit was found between ag 412 and 443 nm and Rrs(670)/Rrs(490), with a correlation coefficient (R2) of 0.78 for the Rrs(670)/Rrs(490) vs. ag(412), and 0.77 for the Rrs(670)/Rrs(490) vs. ag(443).Regression analysis demonstrated the best-fit linear polynomialequations with constant coefficients that describe a variation inreflectance band ratios Rrs(670)/Rrs(490) and variation in the ag at412 and 443 nm, as follows.

agð412Þ ¼ 0:00411þ 2:0��Rrsð670ÞRrsð490Þ

�; R2 ¼ 0:78 (4)

agð443Þ ¼ 0:00129þ 0:6543��Rrsð670ÞRrsð490Þ

�; R2 ¼ 0:77 (5)

where ag (412) and ag (443) represent the CDOM absorption coef-ficients at 412 and 443 nm, constants are the regression coeffi-cients, and Rrs is the remote sensing reflectance.

The value of the spectral slope parameter (S) was derived fromlinear least-squares (LLS) fit of the absorption spectra to an expo-nential function over the range from 412 to 670 nm, as follows:

S ¼ 1ðlr � lÞ Log10

�agð412Þagð443Þ

�(6)

The slope values are consistent with those found in the previousstudies (Morel and Prieur, 1977; Zepp and Schlotzhaue, 1981;Carder et al., 1999; Twardowski et al., 2004). However, the rangeof ag (412) and ag (443) (m�1) for most of the sites largely variesfrom lower values in open ocean waters to higher values in coastalwaters. Therefore, the proposed exponential model with newparameterization needs to be validated using the in-situ CDOMabsorption measurements.

3.2. Modelling and describing the CDOM absorption spectra

In order to assess the performance of the new model, themodelled spectra are compared with the measured spectra (fromNOMAD match-ups data, independent of those used in the algo-rithm development) at the wavelengths of 412e670 nm (Fig. 3). Forcomparison, two semi-analytical models such as the constrainedLinear Matrix (LM) (Boss and Roesler, 2006), and Garver-Siegel-Maritorena (GSM) (Maritorena et al., 2002) models were alsoapplied to the same datasets (Fig. 3). Briefly, these models use the

1E-3 0.01 0.1 1 101E-3

0.01

0.1

1

10

a g(443

) (m

)

Rrs(670)/Rrs(490)

b

ctance ratio Rrs(670)/Rrs(490) from the NOMAD in-situ dataset (N ¼ 233).

400 420 440 460 480 500 520 540 5600.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

0.50

LM_a

g (m

-1)

Wavelength (nm)

400 420 440 460 480 500 520 540 5600.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

0.50

NO

MAD

-2_a

g (m

-1)

Wavelength (nm)400 420 440 460 480 500 520 540 560

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

0.50

NM

_ag (m

-1)

Wavelength (nm)

400 420 440 460 480 500 520 540 5600.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

0.50

GSM

_ag (m

-1)

Wavelength (nm)

In-situ This study

GSM LM

Fig. 3. Spectral variations in the absorption coefficient of coloured dissolved organic matter ((ag) (m�1)) derived using the NOMAD SeaWiFS match-ups dataset. Dots representmeasurements at the SeaWiFS wavelengths, while lines correspond to the exponential regression functions.

S.P. Tiwari, P. Shanmugam / Estuarine, Coastal and Shelf Science 93 (2011) 396e402 399

single exponential model with an assumption of slope0.01 � S � 0.02 (LM) or 0.0206 (GSM) to produce the absorptioncoefficients of coloured dissolved and detrital organic matter(acdm). Note that the detrital absorption values are much smallerthan the coloured dissolved absorption values, and the combinedabsorption represented by acdm should be slightly larger than theabsorption due to coloured dissolved substances. Interestingly, allthe three models show a near-exponential decay at the wave-lengths from 412 to 555 nm. However, the new model provides agvalues closest to the measured (in-situ) values although showingslight differences in the blue domain. The small differences mayarise due to the discrepancies in the absorption measurementsprovided by different instruments with different calibration stan-dards, correction and analysis methods (Twardowski et al., 2004).On the contrary, both the LM and GSM models tend to haveproduced high values at short wavelengths (412e443 nm).Although a small part of the absorption may be attributed to thedetrital matter, the differences are owing to the inadequate range ofthe slope coefficients used in these models to determine the CDOMabsorption curves in the considered waters.

3.3. Model validation

To gain further insight into the difference between model esti-mates and in-situ measurements of ag at each wavelength, themodels were applied to two independent datasets e NOMAD Sea-WiFS match-ups dataset (for wavelengths 412e555 nm) and Carderdataset (400 nm). Fig. 4 compares the model estimates of ag(l) within-situ measurements of ag(l) (NOMAD match-ups dataset). Themodel performance evaluation is summarized in Table 1. Note thatthe results of new model are closely consistent with in-situ ag(l)

across a considerable range of environments and generally follow the1:1 line as shown in Fig. 4. The results of GSM model are quitecomparable with in-situ ag at 412 and 443 nm, but are significantlyunderestimated at the wavelengths of 490e555 nm. In contrast, theLM shows pronounced overestimations throughout thewavelengths(most data points above the 1:1 line). Looking more closely at theirstatistics, the newmodel performs generally equal or better than theGSM model depending on the statistics considered. For instance, itsRMSE, MNB, slope (S) and intercept (I) have improved over the GSMmodel at the wavelengths of 412e443 nm. However, these statisticsare equal or slightly better for the GSM model at the wavelengthsfrom490 to555nm. Considering theband-average statistics, thenewmodel has slightly lower RMSE and intercept values and higher slopevalues. But, MNB and R2 have deteriorated for the new model. Inprinciple, the GSM should have produced slightly higher absorptioncoefficients than the LM, since these models enables the retrievals ofacdm instead of ag. However, the GSM model resulted in underesti-mations of ag values in this study. As expected, the LM consistentlyshow higher ag values at all the wavelengths, thereby yielding poorstatistics (except R2) when compared to other two models.

Since the LM model is restricted to certain wavelengths(412e670 nm), it could not be applied to the Carder in-situ dataset(ag at 400 nm) which covers a wide range of waters of the coastaloceanic environments. Thus, performance of the other two models(new model and GSM) was evaluated using these data. Fig. 5compares the model-derived ag with those from in-situ measure-ments. Table 2 describes the statistics for these two models. Notethat the new model has achieved excellent ag retrievals especiallyin waters with high CDOM contents, although showing noticeableunderestimations in waters with low CDOM contents. As a result,RMSE, MNB, S and I have significantly improved for the newmodel.

0.01 0.1 1 100.01

0.1

1

10 GSM LM NM

Sate

llite_

a g(412

) (m

-1)

In situ_ag(412) (m-1)

1E-3 0.01 0.1 1 101E-3

0.01

0.1

1

10 GSM LM NM

Sate

llite_

a g(443

) (m

-1)

In situ_ag(443) (m-1)

1E-3 0.01 0.1 1 101E-3

0.01

0.1

1

10 GSM LM NM

Sate

llite_

a g(490

) (m

-1)

In situ_ag(490) (m-1)

1E-3 0.01 0.1 1 101E-3

0.01

0.1

1

10 GSM LM NM

Sate

llite_

a g(510

) (m

-1)

In situ_ag(510) (m-1)

1E-4 1E-3 0.01 0.1 1 101E-4

1E-3

0.01

0.1

1

10 GSM LM NM

Sate

llite_

a g(555

) (m

-1)

In situ_ag(555) (m-1)

Fig. 4. Comparisons of the model-derived ag(l) with those from the in-situ dataset (NOMAD SeaWiFS match-ups dataset) at the wavelengths from 412 to 555 nm (N ¼ 55).

S.P. Tiwari, P. Shanmugam / Estuarine, Coastal and Shelf Science 93 (2011) 396e402400

This demonstrates successful retrieval of ag within a reasonableaccuracy and shows significant improvements from the newmodelover the semi-analytical model. Although the results of GSMmodelare consistent with in-situ ag data in waters with low CDOMcontent, it show high scatters of points below and above the 1:1line (indicating large underestimation and overestimation).Further, the GSM model significantly overestimated ag in highCDOM areas (Fig. 5). This caused the deterioration of its statisticscompared to those of the new model. The best statistics with little

errors of the validation satellite match-ups and in-situ datasets forthe newmodel imply that it has more potential to not only improveag(l) retrievals in clear oceanic waters but also reliably retrieve thiscomponent in the optically complex coastal environments.

3.4. Implications for the optical remote sensing

The CDOM study is important for coastal waters, which isgenerally considered unimportant while studying open ocean

Table 1Statistical comparisons between the model-derived ag(l) and in-situ ag(l) (fromNOMAD SeaWiFS match-ups dataset) in the blue-green wavelength region. NM e

New model, GSM e Garver-Siegel-Maritorena model, and LM e constrained LinearMatrix model (LM).

IOP’s RMSE MNB S I R2 N

NMag(412) 0.199 0.116 0.990 0.005 0.757 55ag(443) 0.117 0.175 0.897 0.004 0.756 55ag(490) 0.060 0.228 0.807 0.003 0.743 55ag(510) 0.048 0.284 0.753 0.003 0.733 55ag(555) 0.028 0.374 0.670 0.002 0.700 55Average 0.090 0.235 0.823 0.003 0.738 55

GSMag(412) 0.283 0.363 0.629 0.009 0.866 55ag(443) 0.129 0.184 0.692 0.007 0.863 55ag(490) 0.048 �0.086 0.834 0.004 0.844 55ag(510) 0.036 �0.157 0.881 0.003 0.830 55ag(555) 0.022 �0.323 1.036 0.002 0.781 55Average 0.106 0.0002 0.814 0.005 0.837 55

LMag(412) 0.370 0.675 1.255 0.023 0.777 55ag(443) 0.203 0.642 1.232 0.013 0.796 55ag(490) 0.081 0.565 1.137 0.006 0.806 55ag(510) 0.058 0.586 1.121 0.004 0.801 55ag(555) 0.029 0.604 1.035 0.002 0.769 55Average 0.148 0.614 1.156 0.009 0.790 55

Table 2Statistical comparisons between the model-derived ag(400) and in-situ ag(400)(from Carder in-situ dataset). NM e New model, GSM e Garver-Siegel-Maritorenamodel, and LM e constrained Linear Matrix model (LM).

IOP’s RMSE MNB S I R2 N

NMag(400) 1.686 �0.071 0.832 �0.004 0.829 581

GSMag(400) 2.406 0.388 1.422 �0.0001 0.418 581

S.P. Tiwari, P. Shanmugam / Estuarine, Coastal and Shelf Science 93 (2011) 396e402 401

waters. The current semi-analytical models are found successful inrelatively clear waters which are mainly determined by phyto-plankton and its detritus (Morel and Gentili, 2009). In coastalwaters, these models tend to produce large errors mainly becauseof the inadequate datasets collected from the limited areas andused for the algorithm development. When applying these modelsto other regions, the empirical coefficients derived using a rela-tively small dataset obtained within the limited areas should bechanged and the results would be biased by the influence of highconcentrations of chlorophyll/suspended sediments or otherproperties of waters with the different environments. The presentmodel parameterized with the new relationships using in-situ bio-optical datasets covering a wide range of waters, is very simple butmore efficient in terms of remotely estimating the CDOM absorp-tion coefficients in coastal as well as oceanic waters. The modelvalidations with in-situ datasets and coincidently collected satellite

1E-3 0.01 0.1 1 101E-3

0.01

0.1

1

10 GSM NM

Mod

el_a

g(400

) (m

-1 )

Carder_a g (400) (m ) -1

Fig. 5. Comparisons of the model-derived ag(400) with those from the Carder in-situdataset (N ¼ 581).

data showed relatively small errors that are acceptable on globallevel. Since the model relies on a Rrs(670) and Rrs(490) ratio atwhich the determination of accurate remote sensing reflectancesfrom top-of-the-atmosphere is highly possible with the standardatmospheric algorithm embedded in SeaDAS, it can be effective forretrieving the optical properties of CDOM and describing theirspatial and temporal variability in coastal and oceanic waters usingsatellite remote sensing data.

4. Discussion and conclusions

The importance of CDOM for the upper ocean optics, biogeo-chemical processes and structure and function of ecosystems hasbeen emphasized in previous studies (Stedmon and Markager,2001; Twardowski and Donaghay, 2002; Hu et al., 2006; Ahnet al., 2008; Morel and Gentili, 2009). Therefore, a method thatcould estimate the amount of CDOM in surface waters over largegeographic areas would be highly desirable. Satellite remotesensing has the potential to CDOM observation with high spatialand temporal resolution and enables scaling up to the level of largeecosystems and biomes (Kutser et al., 2005; Zhao et al., 2009). Thesatellite-detection bands for CDOM, 412 and 443 nm, are often usedto form the remote sensing reflectance or normalizedwater-leavingradiance ratios with 555 nm band in the development of itsretrieval algorithms. However, it is well known that the atmo-spheric correction is a difficult task in coastal zones, where theassessment of aerosol optical properties via the near-infrared bandsis questionable because of the interference with high dissolved andsuspended particulate materials (Morel and Gentili, 2009). Withinthis in mind and the capabilities offered by spectral channels in thevisible domain with current ocean colour sensors, this study hasdeveloped a straightforward tool which is applicable to oceancolour remote sensing data to drive the CDOM optical properties inopen ocean and coastal waters.

Currently, few semi-analytical models (e.g., GSM, LM) are usedto derive the absorption coefficients of coloured dissolved anddetrital organic matter (CDM) in open ocean waters. Althoughthesemodels describe the CDM variability in relatively clear waters,they suffer from large uncertainty in coastal waters due to severallimitations. For instance, spectral slopes are assumed constant inthe semi-analytical models but it is likely that this parameter variesin the world ocean based on the characteristics of the constituentsof the water (Wang et al., 2005). Further, these models are notusable with any suite of wavelengths. Refinement of these modelswith new parameterization allowing slopes to vary, possibly asa function of chlorophyll-a or other parameter, may be required toimprove CDOM retrievals in coastal waters. Considering theselimitations, we have improved a single exponential model withappropriate definitions of parameters of the functions describingthe spectra of CDOM and its absorption coefficients. Theseparameters were estimated based on the empirical relationshipsobtained from the NOMAD in-situ bio-optical datasets. The

S.P. Tiwari, P. Shanmugam / Estuarine, Coastal and Shelf Science 93 (2011) 396e402402

performance of the new model was assessed using the NOMADSeaWiFS match-ups (simultaneous measurements of reflectancefrom SeaWiFS and in-situ CDOM) and Carder in-situ datasets. Theretrievals of CDOM absorption by this model were promising andhad an excellent quantitative agreement with the measuredabsorption coefficients at 400, 412, 443, 490, 510 and 555 nm inboth open ocean and coastal waters. A slight deviation was foundbetween the model-derived and in-situ CDOM absorption coeffi-cients, likely due to the different sample measurements methodsadopted by different researchers or due to errors associated withthe measurements data. These results suggest that the new modelwhich uses 670 and 490 nm bands can accurately estimate low tohigher CDOM concentrations in a range of waters within themarine environments. The introduction of such a spectral bandratio (i.e., Rrs(670)/Rrs(490)) provides a significant advantage forSeaWiFS and other similar sensors to demonstrate with thefundamental information about the nature of CDOM in a variety ofwaters. In addition, the new model will also allow us to validateIOPs and the remote sensing estimates of phytoplankton (i.e.,chlorophyll concentration) and productivity, and may open up newpossibilities for using ocean colour remote sensing with studies inareas such as photochemistry, the photobiology of ultravioletradiation and even ocean circulation.

Acknowledgements

This work was supported by grants from the Space ApplicationCenter (SAC), Ahmadabad to Indian Institute of Technology (IIT)Madras, Chennai, India (Project No: OEC0809089SACXPSHA). Theauthors would like to thank the NASA Ocean Biology ProcessingGroup for making available the global, high quality bio-optical(NOMAD) data set to this study. The authors would also like tothank Dr. C. Hu, J.P. Cannizzaro and Prof. K.L. Carder for providingthe bio-optical dataset for the model validation. We are indebted tothe anonymous reviewers for their valuable comments andrecommendations, which greatly helped to improve the quality andstructure of this manuscript.

References

Ahn, Y.H., Shanmugam, P., Moon, J.E., Ryu, J.H., 2008. Satellite remote sensing ofa low-salinity water plume in the East China Sea. Annales Geophysicae 26,2019e2035.

Belanger, S., Babin, M., Larouche, P., 2008. An empirical algorithm for estimating thecontribution of chromophoric dissolved organic matter to total light absorptionin optically complex waters. Journal of Geophysical Research 113 (C04027).doi:10.1029/2007JC004436.

Blough, N.V., Del Vecchio, R., 2002. Distribution and dynamics of chromophoricdissolved organic mater (CDOM) in the coastal environment. In: Hansell, D.A.,Carlson, C.A. (Eds.), Biogeochemistry of Marine Dissolved Organic Matter.Academic Press, pp. 509e546.

Boss, E., Roesler, C., 2006. Over constrained linear matrix inversion with statisticalSelection. In: Lee, Z. (Ed.), IOCCG Report Number 5, Remote Sensing of InherentOptical Properties: Fundamentals, Tests of Algorithms, and Applications.

Bricaud, A., Morel, A., Prieur, L., 1981. Absorption by dissolved organic matter of thesea (yellow substance) in the UV and visible domains. Limnology and Ocean-ography 26, 43e53.

Brown, M., 1977. Transmission spectroscopy examinations of natural waters: C.Ultraviolet spectral characteristics of the transition from terrestrial humus tomarine yellow substance. Estuary Coastal Marine Science 5, 309e317.

Carder, K.L., Steward, R.G., Harvey, G.R., Ortner, P.B., 1989. Marine humic and fulvicacids: their effects on remote sensing of ocean chlorophyll. Limnology andOceanography 34, 68e81.

Carder, K.L., Chen, F.R., Lee, Z.P., Hawes, S.K., Kamykowski, D., 1999. Semi-analyticmoderate resolution imaging spectrometer algorithms for chlorophyll a andabsorption with bio-optical domains based on nitrate-depletion temperatures.Journal of Geophysical Research 104, 5403e5422.

D’Sa, E.J.R., Steward, G., Vodacek, A., Blough, N.V., Phinney, D., 1999. Opticalabsorption of seawater coloured dissolved organic matter determined usinga liquid capillary waveguide. Limnology and Oceanography 44, 1142e1148.

Gao, H., Zepp, R.G., 1998. Factors influencing photoreactions of dissolved organicmatter in a coastal river of the southeastern United States. EnvironmentalScience and Technology 32, 2940e2946.

Hu, C., Lee, Z.P., Muller-Karger, F.E., Carder, K.L., Walsh, J.J., 2006. Ocean colourreveals phase shift between marine plants and yellow substance. IEEE Geo-science and Remote Sensing Letters 3, 262e266.

Johannessen, S.C., Miller, W.L., Cullen, J.J., 2003. Calculation of UV attenuation andcoloured dissolved organic matter absorption spectra from measurements ofocean colour. Journal Geophysical Research 108 (C9). doi:10.1029/2000JC000514.

Kahru, M., Mitchell, B.G., 1998. Spectral reflectance and absorption of massive redtide off southern California. Journal of Geophysical Research 103, 21601e21609.

Kowalczuk, P., Stedmon, C.A., Markager, A.S., 2006. Modelling absorption by CDOMin the Baltic Sea from season, salinity and chlorophyll. Marine Chemistry 101,1e11.

Kutser, T., Herlevi, A., Kallio, K., Arst, H., 2001. Hyperspectral model for interpre-tation of passive optical remote sensing data from turbid lake. Science of TotalEnvironment 268, 47e48.

Kutser, T., Pierson, D.C., Tranvik, L., Reinart, A., Sobek, S., Kallio, K., 2005. Usingsatellite remote sensing to estimate the coloured dissolved organic matterabsorption coefficient in Lakes. Ecosystems 8, 709e720.

Lundgren, B., 1976. Spectral Transmittance Measurements in the Baltic. Report 30.Institute of Physical Oceanography, University Copenhagen.

Maritorena, S., Siegel, D.A., Peterson, A.R., 2002. Optimization of a semi-analyticalocean colour model for global-scale applications. Applied Optics 41, 2705e2714.

Morel, A., Gentili, B., 2009. A simple band ratio technique to quantify the coloureddissolved and detrital organic material from ocean colour remotely sensed data.Remote Sensing of Environment 113, 998e1011.

Morel, A., Prieur, L., 1977. Analysis of variations in ocean colour. Limnology ofOceanography 22, 709e722.

O’Reilly, J.E., Moritorena, S., Mitchell, B.G., Seigel, D.S., Carder, K.L., Garver, S.A.,Kahru, M., McClain, C., 1998. Ocean colour chlorophyll algorithms for SeaWiFS.Journal of Geophysical Research 103, 24,937e24,953.

O’Reilly, J.E., et al., 2000. SeaWiFS postlaunch calibration and validation analyses,part 3. NASA/TM-2000-206892. In: Hooker, S.B., Firestone, R.R. (Eds.), SeaWiFSPostlaunch Technical Report Series, Vol. 11. NASA, Greenbelt, Maryland, p. 49.2000.

Roesler, C.S., Perry, M.J., Carder, K.L., 1989. Modeling in-situ phytoplanktonabsorption from total absorption spectra in productive inland marine waters.Limnology and Oceanography 34, 1510e1523.

Stedmon, C.A., Markager, S., 2001. The optics of chromophoric dissolved or0ganicmatter (CDOM) in the Greenland Sea: an algorithm for differentiation betweenmarine and terrestrial derived organic matter. Limnology and Oceanography 46,2087e2093.

Stedmon, C.A., Markager, S., Kass, H., 2000. Optical properties and signatures ofchromophoric organic dissolved matter (CDOM) in Danish coastal waters.Estuarine, Coastal and Shelf Science 51, 267e278.

Twardowski, M.S., Donaghay, P.L., 2002. Photobleaching of aquatic dissolvedmaterials: absorption removal, spectral alteration, and their interrelationship.Journal of Geophysical Research 107 (C8), 6.1e6.12.

Twardowski, M.S., Boss, E., Sullivan, J.M., Donaghay, P.L., 2004. Modeling the spec-tral shape of absorption by chromophoric dissolved organic matter. MarineChemistry 89, 69e88.

Wang, P., Boss, E., Roesler, C., 2005. Uncertainties of inherent optical propertiesobtained from semi-analytical inversions of ocean colour. Applied Optics 44,4074e4085.

Zepp, R.G., Schlotzhaue, F., 1981. Comparison of photochemical behavior of varioushumic substances in water: III spectrometric properties of humic substances.Chemosphere 10, 479e486.

Zhao, J., Cao, W., Wang, G., Yang, D., Yang, Y., Sun, Z., Zhou, W., Liang, S., 2009. Thevariations in optical properties of CDOM throughout an algal bloom event.Estuarine, Coastal and Shelf Science 82, 225e232.