new atmospheric correction technique to retrieve the … shanmugam... · new atmospheric correction...

New atmospheric correction technique to retrieve the ocean colour from SeaWiFS imagery in

complex coastal waters

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2007 J. Opt. A: Pure Appl. Opt. 9 511

(http://iopscience.iop.org/1464-4258/9/5/016)

Download details:

IP Address: 129.115.103.99

The article was downloaded on 07/10/2012 at 02:50

Please note that terms and conditions apply.

View the table of contents for this issue, or go to the journal homepage for more

Home Search Collections Journals About Contact us My IOPscience

http://iopscience.iop.org/page/terms

http://iopscience.iop.org/1464-4258/9/5

http://iopscience.iop.org/1464-4258

http://iopscience.iop.org/

http://iopscience.iop.org/search

http://iopscience.iop.org/collections

http://iopscience.iop.org/journals

http://iopscience.iop.org/page/aboutioppublishing

http://iopscience.iop.org/contact

http://iopscience.iop.org/myiopscience

IOP PUBLISHING JOURNAL OF OPTICS A: PURE AND APPLIED OPTICS

J. Opt. A: Pure Appl. Opt. 9 (2007) 511–530 doi:10.1088/1464-4258/9/5/016

New atmospheric correction technique toretrieve the ocean colour from SeaWiFSimagery in complex coastal watersPalanisamy Shanmugam1 and Yu-Hwan Ahn

Ocean Satellite Research Group, Korea Ocean Research and Development Institute, Ansan,PO Box 29, Seoul 425-600, Korea

E-mail: [email protected] and [email protected]

Received 22 November 2006, accepted for publication 27 March 2007Published 2 May 2007Online at stacks.iop.org/JOptA/9/511

AbstractThere exists a large demand for an accurate atmospheric correction of satellite ocean colour data overhighly turbid coastal waters, where the standard atmospheric correction (SAC) algorithms designed foropen ocean water turn out to be unsuccessful because of eventual interference of elevated radiance fromsuspended materials and perhaps the shallow bottom with the corrections based on the two near-infraredbands at 765 and 865 nm in which the water-leaving radiances are discarded (or modelled) in order toestimate aerosol radiative properties and extrapolate these into the visible spectrum in the atmosphericcorrection of the imagery. Furthermore, in the presence of strongly absorbing aerosols (e.g. Asian dustand Sahara dust) the SAC algorithms often underestimate water-leaving radiance values in the violet andblue spectrum or completely fail to deliver the desired biogeochemical products for coastal regions. Tomake the satellite ocean colour data offer unrivaled utility in monitoring and quantifying the componentsof ecologically important coastal waters, this study presents a more realistic and cost-effectiveimage-based atmospheric correction method to accurately retrieve water-leaving radiances andchlorophyll concentrations from SeaWiFS imagery in the presence of strongly absorbing aerosols overhighly turbid Northwest Pacific coastal waters. This method is a modified version of the spectral shapematching method (SSMM) previously developed by Ahn and Shanmugam (2004 Korean J. Remote Sens.20 289–305), re-treating the assumption of spatial homogeneity of the atmosphere using simple modelsfor assessing the contributions of aerosol and molecular scattering. Because of the difficulties in makingatmospheric measurements concurrently with each overpass of SeaWiFS the atmospheric diffusetransmittance values are dependent on a standard method with the SAC scheme designed for processingSeaWiFS ocean colour data. The new method is extensively tested under the presence of variousatmospheric conditions using SeaWiFS imagery and the results are compared with in situ (ship-borne)measurements in highly turbid coastal waters of the Korean Southwest Sea (KSWS). Such comparisondemonstrates the efficiency of SSMM in terms of removing the effects of strongly absorbing aerosols(Asian dust) and improving the accuracy of water-leaving radiance retrieval with an RMSE deviation of0.076, in contrast with 0.326 for the SAC algorithm which masked most of the sediment-laden andaerosol-dominated coastal areas. Further comparison in the Yellow Sea waters representing a massivephytoplankton bloom on 27 March 2002 revealed that the SAC algorithm caused an excessive correctionfor the visible bands, with the 412 nm band being affected the most, leading to severe overestimation ofchlorophyll concentrations in the bloom-contained waters. In contrast, the SSMM remained veryeffective in terms of reducing errors of both water-leaving radiance and chlorophyll concentrationestimates.

Keywords: SeaWiFS, ocean colour, atmospheric correction, SAC algorithm, SSMM, absorbing aerosols,Northwest Pacific coastal waters

1 Author to whom any correspondence should be addressed.

1464-4258/07/050511+20$30.00 © 2007 IOP Publishing Ltd Printed in the UK 511

http://dx.doi.org/10.1088/1464-4258/9/5/016

mailto:[email protected]

mailto:[email protected]

http://stacks.iop.org/JOptA/9/511

P Shanmugam and Y-H Ahn

1. Introduction

Coastal waters are ecologically very important and also amajor resource for human populations, contributing a largeshare of the world fisheries catch and supporting a rapidlygrowing mariculture industry. These waters are generallyrecognized as Case II waters [2] that contain constituents otherthan phytoplankton, such as suspended sediments (SS) anddissolved organic matter (DOM). Such additional componentsresulting from terrestrial inputs and bottom resuspension makethese waters optically more complex than Case I waters,where most of the quantitative applications of ocean colourremote sensing have focused in determining the abundance anddistribution of phytoplankton chlorophyll and understandingthe ocean biological and biogeochemical processes since thelate 1970s [3–9]. Interpreting satellite ocean colour datafor such applications required algorithms to first correctfor atmospheric effects using detailed models of radiativetransfer in the atmosphere and then retrieve information onthe biogeochemical variables in ocean waters using fairlysimple empirical relationships. Such algorithms specificallydeveloped for ocean waters are found inadequate in opticallycomplex turbid coastal waters (Case II), which require newalgorithms based on new approaches dealing with bothatmospheric correction and retrievals of ocean bio-opticalproperties from water-leaving radiance [10, 11]. The currentstudy concentrates on atmospheric correction of satellite oceancolour data in Case II waters.

The term ‘atmospheric correction’ is the key procedurein ocean colour data processing as it removes about 80–90% of the top-of-atmosphere (TOA) signal recorded bythe sensor [1, 12]. The remaining signal is the desiredwater-leaving radiance (Lw) that carries immense informationconcerning biogeochemical properties of the ocean waters.This demonstrates the necessity of an accurate atmosphericcorrection that eliminates radiance backscattered from theatmosphere (due to air molecules and aerosols) and possiblyreflected by the sea surface but never entering the ocean.For major satellite ocean colour missions such as CoastalZone Colour Scanner (CZCS) and Sea-viewing Wide Field-of-view Sensor (SeaWiFS), Gordon and Wang have proposedstandard atmospheric correction (SAC) algorithms to retrievewater-leaving radiance from the total radiance measured byCZCS and SeaWiFS at the TOA [13, 14]. Due to difficultiesin deriving information regarding aerosol properties whichvary in time and space, these atmospheric correction schemesassume the water-leaving radiances in the bands centred at 670,765 and 865 nm to be zero in order to estimate aerosol opticalproperties and extrapolate these into the visible spectrum.Such approaches, referred to as the dark-pixel atmosphericcorrection techniques, have been widely accepted for clearocean waters, where satisfactory results have been achievedwith uncertainties <20% in chlorophyll concentrations fromCZCS, <5% in water-leaving radiance and <35% inchlorophyll concentrations from SeaWiFS [3, 15].

In more optically complex turbid coastal waters, however,the SAC schemes do not work well because of theassumption of negligible water-leaving radiance at 670, 765and 865 nm, invalidated by the turbid water constituents(suspended sediments and possibly bottom reflection) that

contribute significant amounts of water-leaving radiance tothese bands. Such an assumption ultimately leads to thesatellite determination of water-leaving radiance in the violetand blue (e.g. SeaWiFS Bands B1 and B2) to severelyunderestimate the in situ observations for highly productiveand sediment-dominated waters. To overcome the black-pixelassumption, researchers have proposed alternate methods ofatmospheric correction for processing CZCS data [16, 17] andSeaWiFS data [18–20] whereby the water-leaving radiances at670, 765 and 865 nm are taken into account by estimatingvalues of water-leaving radiances at the shorter wavelengthsusing empirical band ratio relations in the case of CZCS orthe assumption of spatial homogeneity and nearest-neighbourmethod in the case of SeaWiFS. These modified algorithmshave been successfully applied and significant improvementshave been achieved over the SAC algorithm, particularly inregions of weakly absorbing aerosols. However, implementingthe SAC or modified algorithms in the presence of stronglyabsorbing aerosols would result in underestimation of water-leaving radiance values in the blue leading to unrealisticchlorophyll concentrations in the coastal waters [21, 22].

Instead of the SAC schemes comprising of morecomplex radiative transfer codes (RTCs), several researchershave attempted to develop the atmospheric correctionprocedures solely depending on the image propertiesthemselves. Such procedures are referred to as theimage-based atmospheric correction methods. It is worthmentioning some of these methods, namely the dark-objectsubtraction method [23], the invariant object method [24],the histogram matching method [25], the cosine estimation ofatmospheric transmittance method [26], the contrast reductionmethod [27], the path extraction method [28] and the spectralshape matching method (SSMM) [1]. Studies indicatethe performance of these methods to be satisfactory forland applications using Landsat TM [29, 30], AirborneVisible/Infrared Imaging Spectrometer (AVIRIS) [31] andQuickBird [32] and coastal and inland water applicationsusing Landsat TM [28, 33] and SeaWiFS [1]. Note thatseveral of the above methods, which retrieve land-surfacereflectance properties with good accuracy, are questionablein performing an accurate atmospheric correction of satelliteocean colour imagery over aquatic environments. Thisis because, for aquatic applications, the atmospheric pathsignal includes the surface (Fresnel) reflection of the skylightand specular reflection of the direct solar beam, which issubsequently scattered to the sensor after reflection. For landapplications, the path radiance represents the light withoutreaching the surface. This is because the surface-reflected lightgenerally does not contain information of a water body [34].Furthermore, the water-leaving radiance signal contributesonly 10–20% of the TOA signal which is much smaller than thecontribution of land surface-reflected light to the TOA signal.

Since SeaWiFS has been designed to provide relativelyhigh frequency synoptic information over large areas, datafrom this sensor can be very useful in monitoring sedimenttransport, detecting various spatial extents of phytoplanktonblooms including the highly toxic red tides and delineatingcirculation patterns around coastal zones. In order to allowthe application of SeaWiFS in these areas this study presentsa new atmospheric correction method which is modified from

512

New atmospheric correction technique to retrieve the ocean colour from SeaWiFS imagery in complex coastal waters

Table 1. SeaWiFS performance specifications. (Noise (NEdR—noise equivalent differential spectral radiance) = Locean/SNR.Signal-to-noise ratio (S/N) = Locean/NEdR.)

Wavelength (centre) Locean Locean Noise S/N ratio for Es

Band FWHM (nm) (counts) (mW cm−2 μm−1 sr−1) (counts) Locean (mW cm−2 μm−1 sr−1)

1 402–422 (412) 638.4 9.10 0.645 990 172.812 433–453 (443) 618.2 8.41 0.566 1091 190.203 480–500 (490) 613.0 6.56 0.524 1170 196.264 500–520 (510) 612.2 5.64 0.531 1152 188.025 545–565 (555) 607.1 4.57 0.568 1069 183.066 660–680 (670) 584.9 2.46 0.749 781 151.157 745–785 (765) 531.6 1.61 0.619 859 122.298 845–885 (865) 513.5 1.09 0.707 726 96.19

the previously developed spectral shape matching method(SSMM) by Ahn and Shanmugam [1]. In the followingsections the theoretical basis of this method is describedand its efficiency in removing the atmospheric effects andretrieving water-leaving radiance and pigment concentrationsfrom SeaWiFS imagery in the presence of various atmosphericconditions (including typical Asian dust aerosols in the springof 2000 and 2001) over the Northwest Pacific waters off theKorean and Chinese coasts is assessed. The method is furthertested on SeaWiFS images of highly turbid coastal waters inthe Korean Southwest Sea (KSWS) and waters with a massivephytoplankton bloom in the Yellow Sea (YS). The results arecompared with field measurements and those from standardatmospheric correction and bio-optical algorithms includedin the NASA SeaWiFS Data Analysis System version 4.4(SeaDAS v.4.4).

2. Methods

2.1. SeaWiFS characteristics and the standard atmosphericcorrection (SAC) algorithm

The SeaWiFS ocean colour instrument was successfullylaunched on 1 August 1997, measuring the top-of-atmospheresignal in eight narrow bands spanning the visible andnear-infrared (henceforth NIR) parts of the electromagneticspectrum. Table 1 describes the radiometric specifications forSeaWiFS which includes the bands centred at 412, 443, 490,510, 555, 670, 765 and 865 nm (corresponding to bands 1, 2,3, 4, 5, 6, 7 and 8, respectively) with full width half-maximumbandwidths (FWHM ∼ nm) of 20 nm (λ < 700 nm) and 40 nm(λ > 700 nm). It also describes the values of the total signalfor typical ocean water (Locean), noise equivalent signal andS/N ratio for Locean and the mean values of extraterrestrialsolar irradiance [35, 36].

For processing data from the SeaWiFS sensor, the standardatmospheric correction algorithm developed by Gordon andWang [14] is generally employed in order to estimate thewater-leaving radiance and subsequently derive the desiredgeophysical products. Implementation of this algorithminto the SeaWiFS data processing system is accomplishedthrough the use of lookup tables that were generated with>25 000 radiative transfer simulations for Rayleigh scattering,aerosol contributions and effects of the atmospheric diffusetransmittance. To describe the SAC algorithm and itsimplementation in the data processing system, the observed

radiance is converted to the dimensionless reflectance ρ(λi )

using equation (1)

ρ(λi ) = π L(λi )

F0(λi ) cos θ0(1)

where L(λi ) is the radiance, F0(λi ) is the mean extraterrestrialsolar irradiance and θ0 is the solar zenith angle. The SeaWiFSmeasured reflectance at the TOA system can be written as,

ρT(λi ) = ρra(λi ) + ρas(λi ) + to−snρw(λi ) (2)

where ρT(λi ) is the total reflectance measured by SeaWiFS,ρra(λi ) is the reflectance resulting from Rayleigh scattering(which also includes the Fresnel reflectance at the sea surface),ρas(λi ) is the reflectance resulting from aerosol scattering(also includes the interaction term between aerosol–Rayleighscattering), to−sn(λi ) is the atmospheric diffuse transmittanceand ρw(λi) is the desired water-leaving reflectance to bederived from ρT(λi ).

In the above equation, computation of Rayleigh re-flectance ρra(λi ) in the visible and near-infrared wavelengths israther straightforward because of its dependence on the molec-ular composition of the atmosphere [37]. Thus this term canbe achieved with good accuracy without use of the remotelysensed data. However, the problem is deriving aerosol re-flectance ρas(λi ) which cannot be easily estimated withouthaving prior knowledge because it varies significantly overtime and space scales. To achieve this term, Gordon andWang [14] created a set of aerosol lookup tables consistingof 12 candidate aerosol models (Maritime, Coastal, Tropo-sphere, and Oceanic aerosols with various relative humidities)for different solar and viewing geometries. Assuming the ρw

to be equal to zero or estimating this term in the two NIRbands at 765 and 865 nm (from an in-water model includedin SeaDAS version 4.4) ρas(765) and ρas(865) can be derivedfrom the Rayleigh-corrected TOA reflectance in these bands.Based on the ratio of these two, the aerosol type defined byε = ρas(765)/ρas(865) is determined and ρas and diffuse trans-mittance (to−sn) in the visible wavelengths (Bands 1–6) arecomputed from the pre-computed lookup tables, in order to es-timate the desired water-leaving reflectance in these bands.

This correction procedure enables accurate retrieval ofwater-leaving radiance (within ±5% error) in open oceanwaters [6], but eventually fails in turbid coastal waters wherean increased water-leaving radiance from suspended sedimentsof both organic and inorganic origin and perhaps a shallowbottom contribute to the atmospheric correction bands 765 and

513


TOA radiance in SeaWiFS B841 21 March 2001

SAC Lw in SeaWiFS B841 21 March 2001

a b

ES

YS

ECS

KOREACHINA

RUSSIA

JAPAN

Figure 1. (a) Colour composite image of 21 March 2001 (a typical Asian dust event day) from the top-of-atmosphere (TOA) radiance(mW cm−2 μm−1 sr−1) in SeaWiFS bands (B841) centred at 865 nm (red), 490 nm (green) and 412 nm (blue). (b) The corresponding imagefrom water-leaving radiances Lw (mW cm−2 μm−1 sr−1) retrieved from the standard atmospheric correction (SAC) algorithm. The inset onthe lower left corner of the TOA radiance image shows the aerosol sample collected on filter by Yuan et al [52] during this Asian dust event.Analysis of this sample indicated the presence of ferro-magnesium and quartzo felspathic minerals that dominated the aerosol plume. Therectangular box in the TOA image is made for figures 2(a) and (b). ECS—East China Sea, YS—Yellow Sea, ES—East Sea.

865 nm [18, 38]. Comparison of near-simultaneous match-upsof SeaWiFS and field observations of water-leaving reflectanceshowed that in turbid coastal waters the SAC algorithm oftenretrieved negative water-leaving reflectance in 412 and 443 nmbands and its retrievals in these bands became even worse inwaters with chlorophyll concentrations >2 mg m−3 [19, 20].The large errors toward lower wavelengths would resultprimarily owing to excessive aerosol path radiance removalfrom an inappropriate extrapolation over a greater wavelengthrange. Validation studies demonstrated that these errorswere significantly reduced after the NIR signals were takeninto account in the atmospheric correction of SeaWiFS/CZCSimagery in turbid coastal waters [16, 19, 20, 39].

Hu and his co-workers evaluated a turbid wateratmospheric correction scheme in moderately turbid Gulf ofMaine waters where they found that SeaWiFS estimates ofnormalized water-leaving radiance in 412 and 443 nm bandswere still lower than the in situ observations [18]. Thiswas attributed to another problem that the SAC algorithmis facing without inclusion of the lookup tables for stronglyabsorbing aerosols over coastal areas. Figures 1(a) and (b)are good examples of how the strongly absorbing aerosols(referred to as Asian dust or Yellow dust by Iwasaka and hisco-workers [40]) of the Spring 2001 are transported by windfrom the Chinese desert to Northwest Pacific ocean waters,where the SAC algorithm severely overestimated the aerosolreflectance ratio, yielding very low or improbable negativewater-leaving radiances (at 412 and 443 nm) in relatively lowaerosol areas (green colour signature in the ES) and showing acomplete failure of atmospheric correction for the high aerosolareas off the Chinese and Korean coasts. The SAC algorithm

retrieves ρw at 443 nm with an error <0.001–0.002 for non-and weakly absorbing aerosols, which is usually the case foropen ocean regions where aerosols occur locally. However,for the coastal regions large errors or complete failure of thisalgorithm occur because these regions usually have both CaseII waters in which the ρw at the two NIR bands are discardedor the estimated ρw in these bands are not sufficient to dealwith turbid water and strongly absorbing aerosols [41]. Thisunderlines the necessity of a set of more realistic aerosolmodels with representative characteristics of absorbing aerosolto be used in the aerosol lookup tables.

2.2. The spectral shape matching method (SSMM)

The modified version of the SSMM atmospheric correctionscheme starts with radiance instead of reflectance. Thereforethe total signal recorded by SeaWiFS at the TOA in a spectralband centred at a wavelength λi can be defined by

LT(λi ) = Lpath(λi) + t(s−o)Lsg(λi ) + t(o−sn)Lwc(λi )

+ t(o−sn)Lw(λi) (3)

where LT(λi) is the total radiance at the TOA, Lpath(λi ) isthe path radiance resulting from scattering in the atmosphereand from specular reflection of atmospherically scattered light(skylight) from the sea surface, Lsg(λi ) is the radiance ofdirect sunglint from the sea surface, Lwc(λi ) is the radiance ofwhitecaps at the sea surface and Lw(λi) is the desired water-leaving radiance. The terms ts−o(λi ) and to−sn(λi) are direct(sun → ocean) and diffuse (ocean → sensor) transmittances ofthe atmospheric column, respectively. The ts−o is appropriatefor sun glitter Lsg(λi ) that is highly directional except at high

514


wind speeds, while to−sn is pertinent for water-leaving radianceand the whitecap radiance as they have a near-uniform angulardistribution [42]. Because SeaWiFS has been designed to havea provision for tilting the scan plane away from the specularimage of the sun, the Lsg may be disregarded for brevity.The whitecap contribution Lwc in SeaWiFS imagery can beestimated by using a previously established reflectance modelwith the input of sea surface wind speed [43]. However,SeaWiFS observations show that the whitecap reflectancemodel used in the SAC algorithm produces unacceptable errors(overestimation of the whitecap contribution) when the seasurface wind speed is greater than 7–8 m s−1 [14]. This resultsin low estimates of water-leaving radiance values. To avoidsuch a miscalculation, the current SSMM ignores the whitecapcontribution Lwc in the atmospheric correction of SeaWiFSimagery. Thus, equation (3) becomes

LT(λi) = Lpath(λi ) + t(o−sn)Lw(λi ). (4)

Here the path radiance Lpath(λi ) can be decomposed into twobroad components as follows

Lpath(λi) = L ra(λi ) + Las(λi) ≈ γra(λi ) + γas(λi ) (5)

where L ra(λi ) is the Rayleigh radiance due to single andmultiple scattering by air molecules in the absence of aerosols,Las(λi ) is the aerosol radiance due to single and multiplescattering by aerosol particles in the absence of air molecules,γra(λi ) is the Rayleigh factor and γas(λi) is the aerosol factor(γ factor is relative to radiance). For simplicity, the complexterm between molecular and aerosol scattering is neglected inthe image-based atmospheric correction scheme.

Accurate estimation of L ra(λi ) and Las(λi ) requires insitu field measurements of aerosol type, optical thickness, airpressure, ozone optical thickness, wind, water vapour, etc.,at the time of each image acquisition. These measurementsare frequently unavailable or are of questionable qualitywhich makes routine and accurate atmospheric correctionof images with radiative transfer models difficult [44].Furthermore, scattering and absorption by aerosols are difficultto characterize due to their variation in time and space [14],thus constituting the most severe limitation to the atmospherecorrection of satellite data [45]. Thus, instead of modellingthe atmospheric optical properties, the image-based SSMMscheme takes advantages of the two SeaWiFS NIR bands inorder to estimate the magnitude of Las in the NIR2 using thefollowing expression

γas(λNIR2) = Q(λNIR2) × (LT(λNIR2),�(λNIR1, λNIR2))

− (C × (SS)(λNIR2)) (6)

where γas(λNIR2) is the aerosol factor in the spectral bandλNIR2, �(λNIR1, λNIR2) is the ratio of (LT) at λNIR1 to(LT) at λNIR2, (LT(λNIR2), �(λNIR1, λNIR2)) is relatedto LT(λNIR2) and the ratio �(λNIR1, λNIR2) at the twoNIR bands, LT(λNIR2) is the total radiance at λNIR2,and (SS)(λNIR2) is related to the backscattered signal bysuspended sediments (SS) at λNIR2. Two unknown constantsQ(λNIR2) and C are to be determined for (LT(λNIR2), �(λNIR1,λNIR2)) and for (SS)(λNIR2), respectively. These can beaccomplished by using an iterative convergence method. Theterm Q(λNIR2) is spectrally dependent on the aerosol type

Figure 2. Examples of the histogram of (a) TOA radiance and(b) aerosol-corrected radiance (mW cm−2 μm−1 sr−1) from thepixels of SeaWiFS imagery collected on 21 March 2001 (rectangularbox in figure 1(a)). The aerosol corrected radiance provides us thepossibility of distinguishing the turbid water pixels (0.6–0.68) fromthe relatively clear water pixels (0.58–0.6) around the KoreanSouthwest Sea.

(LT(λNIR2),�(λNIR1, λNIR2)), while C is stable with respectto suspended sediments. Since γas(λNIR2) contains a smalleramount of backscattered signal by SS in shallow coastalwaters, (SS)(λNIR2) must be estimated and eliminated fromγas(λNIR2) before extrapolating to the visible bands. This willresult in a correct retrieval of water-leaving radiance in thevisible bands. To do this, first the aerosol-corrected radiance isobtained by subtracting the first estimate of the aerosol factor(for which C × (SS)(λNIR2) = 0) from the TOA radiance atλNIR2. From the computed histogram of the aerosol-correctedradiance image, for example, figures 2(a) and (b) show thehistograms of TOA radiance and aerosol-corrected radiance(mW cm−2 μm−1 sr−1) from the pixels of SeaWiFS imagerycollected on 21 March 2001, the values of (SS) are discernedto be variable for turbid water pixels and set to nearly zero forclear waters. This is achieved by the determination of C in theequation. This leads to the estimated values of (SS)(λNIR2)

being reapplied in equation (6) to obtain new values of theaerosol factor γas(λNIR2). The resulting aerosol factor froma pixel of SeaWiFS image is proportional to the aerosolradiance of that pixel, thus accounting for additive effects of

515


Table 2. Coefficients for computing the Rayleigh factor γra(λi ) in the SeaWiFS spectral bands.

Wavelength (nm)

Coefficients 412 443 490 510 555 670 765 865

α −0.1020 −0.1296 −0.1524 −0.1200 −0.0775 −0.0217 −0.0040 −0.0020β 8.1103 7.3896 5.4256 4.4509 2.9344 1.1817 0.5240 0.270

atmospheric aerosol scattering. The estimated aerosol factorγas(λNIR2) is then extrapolated and removed in the visible. Thisis achieved by using the lookup table of aerosol spectral modelsgenerated from the iterative convergence method. The aerosolfactor γas(λNIR2) simply multiplied by the term Q(λi) givesrise to the aerosol factor γas(λi ) of a visible spectral bandcentred at a wavelength λi according to,

γas(λi) = Q(λi ) × γas(λNIR2). (7)

If the aerosol type is fixed over the spatial scale of 100–1000 km [18] or some time over 1000 km, then the aerosolfactor may vary from one pixel to another pixel with respectto the aerosol concentrations observed by SeaWiFS in eightspectral bands. Using equation (7), the aerosol radianceis computed and subtracted from LT(λi ), which derives theaerosol-corrected radiance Lac(λi) as follows,

Lac(λi ) = (LT(λi )− Las(λi )) = L ra(λi )+ t(o−sn)Lw(λi ). (8)

Unlike the aerosol, the Rayleigh scattering component isspatially nearly stable [14] and can be approximated fromLac(λi ) using simple linear equations. Here SSMM makesan assumption that the spectral shape of the water-leavingradiance of clear ocean waters is stable compared to thatof turbid coastal waters whose constituents with respect totidal currents and bottom circulation significantly modifythe spectral shape of the water-leaving radiance in thesewaters [28]. With the known/measured water-leaving radianceof ocean waters this assumption allows estimation of L ra(λi )

from Lac(λi ) as follows,

L ra(λi ) = Lac(λi ) − t(o−sn)Lw(clear)(λi ). (9)

Here a match-up of the in situ water-leaving radiance ofrelatively clear ocean waters is used to achieve the first estimateof L ra(λi ) at one point using equation (9). Note that the insitu water-leaving radiance values are translated to the topof the atmosphere using the to−sn(λi ) values derived from astandard method as discussed later. Gould and Arnone [39]proposed a similar equation taking the difference betweenLT(λi ) and Lw(clear)(λi ) to assess the path radiance Lpath(λi )

which was kept constant and applied over the entire image,with the assumption that the atmosphere (Rayleigh and aerosolscattering, and possibly surface reflection) does not varyover the spatial scale of that image. It may be valid for alow altitude sensor like the compact airborne spectrographicimager (CASI), but it does not hold for high altitude/polarorbiting sensors observing a wide area. To assess contributionsof the L ra component for the pixels of a SeaWiFS image aset of simple linear equations is derived relating the L ra(λi )

to Lac(λi ) through

γra(λi ) = α[Lac(λi)] + β ≈ L ra(λi ). (10)

The Rayleigh factor γra(λi ) is proportional to the Rayleighradiance and varies from one pixel to another depending onthe magnitude of the signal observed at λi . The empiricalcoefficients α and β for each SeaWiFS band are given intable 2. Note that these coefficients are better consistent forSeaWiFS measurements of near noon time and different watertypes similar to those off the Korean and Chinese coasts. Now,substituting equations (5)–(10) into equation (4) yields thedesired water-leaving radiance as follows,

Lw(λi ) = [LT(λi ) − L ra(λi ) − Las(λi )]t(o−sn)(λi )

. (11)

In equations (3) and (11), the term atmospheric diffusetransmittance to−sn(λi ), which propagates the water-leavingradiance from the ocean surface to the sensor (o–sn) andhas a multiplicative effect caused by both scattering andabsorption, needs to be assessed accurately in order to recoverthe water-leaving radiance just above the ocean surface Lw(λi )

from to−sn(λi )Lw(λi ). However, accurate correction for themultiplicative effect due to transmittance usually requires insitu field measurements of atmospheric optical depth [30],which is not always practical during each satellite overpass,and therefore, the average transmittance values computed forthe different spectral bands in the visible wavelength and forthe different atmospheric conditions were taken into accountin atmospheric correction of satellite data [26, 32]. Thecurrent SAC scheme devised for SeaWiFS enables a goodapproximation of to−sn(λi ) values from tables containing dataof ∼25 000 simulations using exponential fitting [46] as givenbelow

t(o−sn)(λi , θ) = X (λi , θ) exp[−Y (λi , θ)τas(λi )] (12)

where X (λi , θ) and Y (λi , θ) are fitting coefficients for theeight SeaWiFS spectral bands, 12 aerosol models, and varioussolar and viewing angle geometries. τas(λi ) is the aerosoloptical thickness. The t(o−sn)(λi , θ) values from the tables areaccurate with an accuracy within 0.1% [41].

The derived water-leaving radiance can be convertedto the reflectance (using equation (1)) which is divided byπ steradians to compare to the remote-sensing reflectance:1/πρ(λ) ≈ Rrs(λ) [2]. Figure 3 summarizes the overallcorrection scheme of the SSMM to retrieve water-leavingradiance and chlorophyll concentrations from SeaWiFSimagery in coastal waters.

2.2.1. Aerosol spectral models. Developing the aerosolspectral models for image-based atmospheric correction ofSeaWiFS data necessitates better understanding of the type,characteristic, source and composition of the aerosols ofthe regions, either from the historical data sets from insitu measurements or from previously demonstrated results

516


Figure 3. A flowchart of the SSMM atmospheric correction approach to retrieve the water-leaving radiance and chlorophyll concentrationsfrom SeaWiFS imagery in coastal waters.

from satellite data. It has been found that both absorbingand non-absorbing aerosols often occur above the NorthwestPacific waters off the Korean and Chinese coasts. Themost commonly observed aerosols of these regions areAsian dust, carbonaceous (biomass burning/urban), sulfate(Continental/Maritime origin) and sea salt (Maritime) aerosols.Asian dust. which is often referred to as Yellow dust [40], isstrongly absorbing in the violet and blue because it containsan abundance of ferro-magnesium minerals in the 0.5–17 μmsize range. These mineral particles are blown by windin the Chinese desert regions and are transported over theNorthwest Pacific ocean waters during spring [47–49]. Incontrast, sulfate aerosol particles are generally small in size andweakly absorbing, while carbonaceous aerosol particles aremore complicated in their composition and optical propertiesbut recognized as moderately or strongly absorbing aerosolswith inclusion of soot particles [50]. Sea salt or maritimeaerosols are weakly absorbing and are commonly found in the

oceanic boundary layer of the atmosphere, with sizes mostlybelow 0.05 μm [12].

To build the aerosol spectral models and understandthe spectral behaviour of different aerosols, SeaWiFS datacollected from the high resolution picture transmission (HRPT)station at the Korea Ocean Research and Development Institute(KORDI) during 1999–2005 were processed and six aerosoltypes were discerned based on a visual interpretation anddigital classification of SeaWiFS data [50] and in situdata [47, 49, 51]. These are the bright aerosol 1 (BA1), brightaerosol 2 (BA2), mixture aerosol, yellowish brown aerosol1 (YB1), yellowish brown aerosol 2 (YB2) and yellowishbrown aerosol 3 (YB3) (table 3, figure 4). The Asian dustaerosols represented by YB were detected over the East ChinaSea (ECS), Yellow Sea (YS), Korean Seas (KS), East Sea(ES) and often the North Pacific during spring 1999–2005,while the bright aerosols of continental/maritime origin werefound to influence the oceanic boundary layer during spring–

517


Table 3. Classification of the most commonly observed aerosols over Northwest Pacific waters (ECS, YS, KS, ES, and neighbouring waters).

Aerosol type Characteristic Source Origin/composition Reference

Bright aerosol 1 Weakly/moderately absorbing Continental Biomass burning smoke/urban [61]

Bright aerosol 2 Weakly/moderately absorbing Continental/Maritime Complex haze/urban [12, 42]

Mixture aerosol Absorbing Continental/Maritime Mixture of continental and maritimeaerosols

[12, 50]

Yellowish brownaerosol 1 (YB1)

Strongly absorbing Desert Asian dust with mineral particles [48, 62]


Strongly absorbing Desert Asian dust with ferro-magnesium andquartzo-felspathic minerals

[49, 51]


Extremely absorbing Desert Asian dust with abundance of stronglyabsorbing ferro-magnesium minerals

[47]

Figure 4. The γas(λi ) value as a function of wavelength (nm) for thesix aerosol spectral models used as candidates for retrieving theaerosol contributions in the SeaWiFS image. In the violet and bluespectral regions the magnitude of the aerosol factor diminishes as theaerosol type changes from bright (reflecting) to yellowish brown(YB) aerosols (strongly absorbing). The downward arrow in thisfigure represents the decrease of the aerosol factor (particularly in theviolet and blue) with increasing vertical structure of Asian dustaerosols from 6 April 2000 to 21 March 2001 images.

fall seasons. Using these data, the aerosol spectral modelswere generated from equations (6) and (7) by the iterativeconvergence method. Figure 4 illustrates variations of theγas as a function of wavelength (nm), for six aerosol typesfrom BA to YB. Clearly, the γas for YB aerosols increasesfrom 865 to 510 nm and decreases from 510 to 412 nmprimarily due to strong absorption caused by mineral particlescontained in these aerosols. Unlike the bright aerosols thatare weakly/moderately absorbing, the magnification of verticalstructure and abundance of YB aerosol concentrations notablydiminished the γas with decreasing wavelength (figure 4).This suggests the impact of this absorption to be especiallydependent on the vertical distribution of the aerosol [22, 48].In contrast, the bright aerosol spectral models show increasedγas with decreasing wavelength and no remarkable changes dueto vertical intensity of the BA were found in the violet–greenwavelengths. This study evaluated another way of estimatingthe γas(λi ) by ignoring the Q(λi ) and replacing the LT(λNIR2)

with LT(λi) in equation (6). This approach yielded unrealistic

spectral forms of aerosols leading to an excessive correction inthe violet–green wavelengths, which in turn resulted in largeerrors in chlorophyll retrieval from SeaWiFS data.

2.2.2. Atmospheric diffuse transmittance table (to−sn). Theatmospheric diffuse transmittance is one of the inputs to theatmospheric correction of remotely sensed data. This can beobtained from in situ or simulated atmospheres using radiativetransfer models [26, 32]. Because of the difficulties in makingatmospheric measurements concurrently with each overpassof SeaWiFS, the present study adopts the to−sn(λi ) values asindependent of field measurements derived from a previouslyestablished method as equation (12). Table 4 describes thevalues of to−sn for each spectral band of SeaWiFS representingvarious atmospheric conditions (clear, BA and Asian dust(AD)) off the Chinese and Korean coasts during 2000–2005. Itis believed that the computed values are close to those of in situfield measurements because equation (12) generates acceptableresults with an error of <0.1%. These values in each spectralband are averaged for each atmospheric condition observed bySeaWiFS. Note that the average to−sn values of each band forthree aerosol types of 15 different dates appear to be nearlysimilar to the average of to−sn values of each band for eachaerosol type. Adopting these values in the SSMM atmosphericcorrection scheme will substantially improve the accuracy ofwater-leaving radiance retrieval better than can be achievedwith the previous method that assumes the to−sn value to bea constant regardless of the spectral bands [1].

3. Results and discussions

3.1. Performance of atmospheric correction by SAC andSSMM

The performance of the SAC and SSMM algorithms wasassessed on SeaWiFS images in the presence of variousatmospheric conditions (including clear, bright and typicalAsian dust aerosols) over Northwest Pacific waters off theKorean and Chinese coasts. Prior to the application of SSMM,the digital counts of Level 1A SeaWiFS raw data productobtained from the HRPT station at KORDI were convertedto total radiance (mW cm−2 μm−1 sr−1) at the level of theTOA by using the SeaWiFS band sensitivity and pre-launchcalibration coefficients given in NASA SeaWiFS technicalreport series [36]. Figures 5(a) and (b) show examples of

518


Table 4. Atmospheric diffuse transmittances (to−sn) computed from SeaWiFS images for the period 2000–2003.

Atmospheric diffuse transmittances (to−sn) in the SeaWiFS bands

Image data Atmospheric conditions 412 443 490 510 555 670 765 865

06/04/2000 Asian dust 0.8035 0.8413 0.8822 0.8943 0.9161 0.9479 0.9621 0.969821/03/2001 Asian dust 0.7783 0.8289 0.8828 0.8984 0.9265 0.9646 0.9795 0.986319/03/2002 Asian dust 0.7979 0.8345 0.8745 0.8869 0.9091 0.9380 0.9555 0.963016/04/2004 Asian dust 0.8225 0.8592 0.8980 0.9092 0.9296 0.9575 0.9693 0.974515/04/2005 Asian dust 0.7993 0.8381 0.8801 0.8922 0.9147 0.9472 0.9615 0.9693

Mean transmittance (AD) 0.8003 0.8404 0.8835 0.8962 0.9192 0.9511 0.9656 0.9726

04/09/2002 Bright aerosol 0.8200 0.8595 0.9023 0.9146 0.9367 0.9668 0.9794 0.985006/08/2003 Bright aerosol 0.8296 0.8643 0.9015 0.9123 0.9303 0.9584 0.9704 0.975911/08/2004 Bright aerosol 0.8109 0.8451 0.8825 0.8937 0.9142 0.9435 0.9575 0.964526/03/2005 Bright aerosol 0.7948 0.8349 0.8787 0.9203 0.9140 0.9465 0.9618 0.969008/05/2005 Bright aerosol 0.8271 0.8618 0.8987 0.9060 0.9288 0.9565 0.9680 0.973504/09/2002 Bright aerosol 0.8200 0.8595 0.9023 0.9145 0.9367 0.9668 0.9794 0.9850

Mean transmittance (BA) 0.8165 0.8531 0.8927 0.9094 0.9248 0.9543 0.9674 0.9736

28/08/2001 Clear 0.7776 0.8156 0.8582 0.8710 0.8945 0.9295 0.9450 0.954024/02/2002 Clear 0.7828 0.8239 0.8700 0.8835 0.9095 0.9450 0.9615 0.969508/10/2003 Clear 0.7867 0.8255 0.8715 0.8814 0.9050 0.9383 0.9538 0.961410/10/2003 Clear 0.8233 0.8565 0.8925 0.9025 0.9235 0.9515 0.9645 0.9705

Mean transmittance (Clear) 0.7950 0.8328 0.8753 0.8873 0.9102 0.9428 0.9577 0.9652

Overall mean transmittance 0.8039 0.8421 0.8838 0.8976 0.9180 0.9494 0.9635 0.9704

the total radiance images of 6 April 2000 and 21 March2001 generated from the three SeaWiFS bands centred at 865(red), 490 (green) and 412 nm (blue). Note that the blackareas denote the non-water mask (land and cloud) based ona threshold value at NIR2. Inevitably, there are a large numberof intense-aerosol covered ocean pixels also included in thatmask in figure 5(b), which will be dealt with later. These twoimages clearly illustrate the spatial distribution of the yellowishbrown Asian dust aerosols, which were transported widely overEast China, Korea and Japan during spring 2000 and 2001.The additive effects induced by these aerosols significantlymodified the spectral shape of the total radiance signals incoastal and ocean waters. This is primarily because thedetected aerosols contained average concentrations of crustalelements (Ca, Al, Fe, Mg, Mn) accounting for 65–86% ofeach of their loadings, when the wind speed and aerosolmass loadings severely increased [47, 49, 51, 52]. Such highconcentrations of ferro-magnesium mineral particles absorbedlight much better than they reflected it (figure 4).

Figures 5(c) and (d) present the results of SSMMapplied on these two SeaWiFS images. Note that theatmospheric effects were perfectly removed and retrievedwater-leaving radiances demonstrate a cross-shelf transport ofsuspended sediments from ECS coastal waters to Korea waters(figure 5(c)) and turbid algal matter-containing water massmovement along the east coast of Korea (figure 5(d)). In therelatively clear ES waters where the TOA radiance signalswere strongly influenced by the Asian dust aerosols, SSMM-retrieved water-leaving radiances are more consistent withour field measurements off the Korean east coast. However,the accuracy of water-leaving radiance retrieval with SSMMbecame slightly worse in the straylight affected areas andscan edges where the satellite estimates of water-leavingradiances in the violet–green bands were 1.5–2 times higherthan the usually observed water-leaving radiances in the ES(figure 5(d)).

Figures 6(a)–(f) provide a better comparison of the resultsin eight bands (bands 1–8) before and after the SAC andSSMM algorithms were employed on SeaWiFS images of 6April 2000 and 21 March 2001. The brighter areas representthe Asian dust aerosols in the TOA radiance images andsuspended sediment distribution in the water-leaving radianceimages excluding the western boundary areas of the ES infigure 6(d). While SSMM was capable of retrieving the water-leaving radiance values for the intense aerosol pixels of theECS, YS, ES and Korean Sea, the SAC algorithm showeda complete failure of the scheme in these areas (indicatedby arrows), yielding low and negative water-leaving radiancevalues at 412 and 443 nm in the less-intense aerosol pixelsof the ES, ECS and YS (figures 6(c) and (f)). This may beowing to an extensive overcorrection of aerosol effects in theviolet and blue bands [22]. The SSMM was also successful inrecovering most of the turbid coastal areas previously maskedby the SAC algorithm. The inappropriate in-water modelto estimate Lw at the two NIR bands with this algorithmultimately resulted in very low water-leaving radiances inturbid waters and nearly zero water-leaving radiance values inrelatively clear waters (figures 6(c) and (f)).

To further assess the efficiency of SAC and SSMMalgorithms chlorophyll concentrations were retrieved from theatmospherically corrected SeaWiFS images of 6 April 2000and 21 March 2001 using the standard NASA OC2v4 bio-optical algorithm included in SeaDAS v4.4 (table 5). Thisalgorithm is purely empirical and uses a simple band ratio,Rrs(490)/Rrs(555), to estimate chlorophyll concentrationsfrom SeaWiFS data [53]. Figures 7(a)–(d) compare retrievalsof the chlorophyll concentrations with SAC-OC2v4 andSSMM-OC2v4 algorithms on SeaWiFS data. One canobserve that the results from these two algorithms are verydifferent in terms of the coverage and magnitude of thechlorophyll concentrations. SSMM appears to be successfulin removing much of the influence of the Asian dust aerosols

519


TOA radiance in SeaWiFS B841

6 April 2000

TOA radiance in SeaWiFS B841

21 March 2001

SSMM Lw in SeaWiFS B841

21 March 2001

SSMM Lw in SeaWiFS B841

6 April 2000

b

a

d

c

Figure 5. (a), (b) Colour composite images of 6 April 2000 and 21 March 2001 (typical Asian dust event days) from the top-of-atmosphere(TOA) radiance (mW cm−2 μm−1 sr−1) in SeaWiFS bands (B841) centred at 865 nm (red), 490 nm (green) and 412 nm (blue). (c) and (d) Thecorresponding images from water-leaving radiances Lw (mW cm−2 μm−1 sr−1) retrieved from the SSMM atmospheric correction scheme.Note that SSMM retrieves water-leaving radiances from SeaWiFS image over the regions of dense Asian aerosol dusts off the Korean andChinese coasts. The lines denote the 500 km S/N transects running across the East Sea.

Table 5. Standard SeaWiFS chlorophyll retrieval algorithms and their coefficients [53, 63]. (OC2v4〈Chl〉 = 10(a0+a1 R+a2 R2+a3 R3) + a4.OC4v4〈Chl〉 = 10(a0+a1 R+a2 R2+a3 R3+a4 R4).)

SeaWiFS algorithm a0 a1 a2 a3 a4 R

OC2v4 0.319 −2.336 0.879 −0.135 −0.071 Log10(Rrs(490)/Rrs(555))OC4v4 0.366 −3.067 1.930 0.649 −1.532 Log10((Rrs(443) > Rrs(490) > Rrs(510))/Rrs(555))

and recovering large previously unprocessed areas from theApril 2000 and March 2001 images. The newly derived water-leaving radiances allow more accurate estimates of chlorophyllconcentrations in turbid coastal waters as well as areas severely

affected by the aerosols (figures 7(c) and (d)). Several complexanticyclonic eddy features in the vicinity of the ES andan intense chlorophyll pattern along the North Korean andRussian coastal areas are evident in these images. These might

520


Figure 6. Comparison of atmospheric correction by the SAC and SSMM algorithms applied to two Asian aerosol dusty images fromSeaWiFS sensor on 6 April 2000 ((a)–(c)) and 21 March 2001 ((d)–(f)). (a) and (d) the top-of-atmosphere (TOA) radiance(mW cm−2 μm−1 sr−1) in the SeaWiFS spectral bands, (b) and (e) water-leaving radiance (mW cm−2 μm−1 sr−1) from SSMM, and (c) and (f)water-leaving radiance from the SAC algorithm.

be related to frequent spring blooms caused by a combinationof enhanced irradiance with the effect of vertical mixing andelevated nutrients in spring [54, 55]. In contrast, the SACalgorithm shows particularly high chlorophyll concentrationsin the ECS and YS waters, primarily because of a backgroundof the Asian dust aerosols (figure 7(a)), creating masks overhighly turbid coastal waters and the areas dominated by intenseaerosols off the Korean and Chinese coasts (indicated bythe arrows in figures 7(a) and (b)). Nevertheless the lowchlorophyll concentrations inferred from the SAC algorithmalong the Russian coast does not significantly identify theintense phytoplankton blooms from other water types. Suchresults demonstrate an efficiency of the SAC algorithm forprocessing SeaWiFS imagery of highly turbid coastal watersin the presence of intense mineral dust aerosols. For severalimages of the Asian dust aerosols of spring, SSMM retrievedapproximately the same concentrations in the YS and ES asit did in the absence of dust, proving to be a very effectivemethod in deriving information on biogeochemical variablesand allowing SeaWiFS applications to turbid coastal waters.

To understand why the SSMM scheme was unable toretrieve the chlorophyll concentrations for the pixels of a verythick aerosol plume layer (encircled areas in figure 7(d)),the signal-to-noise ratio (SNR) was computed as the ratio ofγas(λNIR2) to TOA radiance at NIR2 using SeaWiFS images of6 April 2000 and 21 March 2001. Figures 8(a) and (b) illustratevariations of SNR and TOA radiance along the 500 kmtransects running across the ES from North Korean coastal

areas to ES offshore in the case of the 6 April 2000 image andfrom Korean east coastal areas to ES offshore in the case ofthe 21 March 2001 image (transects in figures 5(a) and (b);profiles in figures 8(a) and (b)). Clearly, the SNR diminishedwhen the concentration and vertical depth of the Asian dustaerosol plume increased to influence the TOA radiance signalfrom the offshore to coastal areas of the ES. On 6 April 2000,the SNR appeared to be higher than the TOA radiance beforeit converged close to the North Korean coast area, where theaerosol intensity was rather high. In contrast, on 21 March2001 the SNR values were higher than the TOA radiancesover a dispersed phase of the Asian dust aerosols from 500to 130 pixels, but abruptly decreased with increasing aerosolplume intensity in the transect area 130–0 pixels. This suggeststhat magnification of the vertical structure and abundance ofAsian dust aerosols might have hindered emergence of thewater signal out of the thick layer of the aerosol plume in thisregion. Under these circumstances, retrieval of water-leavingradiances with any atmospheric correction algorithms remainslimited, and therefore, such areas should be probably maskedwhile processing the satellite ocean colour data.

3.2. Validation with in situ data

A first validation of the methods with in situ measurementswas performed in highly turbid coastal waters of the KoreanSouthwest Sea. This region is relatively shallow (5–30 m)where strong semidiurnal tides combine with frequent strongwinds to cause relatively high concentrations of suspended

521


SeaWiFS Chl from SSMM-OC2algorithms(6 April 2000)

c

thick aerosol ring

SeaWiFS Chl from SSMM-OC2algorithms(21 March 2001)

d

SeaWiFS Chl from SAC-OC2 algorithms(6 April 2000)

a

SeaWiFS Chl from SAC-OC2 algorithms(21 March 2001)

b

ES

ECS

YS

Figure 7. SeaWiFS chlorophyll (mg m−3) images of 6 April 2000 and 21 March 2001 processed using the SAC-OC2 bio-optical algorithm((a) and (b)) and SSMM-OC2 bio-optical algorithm ((c) and (d)). Arrow marks indicate the masked areas due to failure of the SAC algorithmin the presence of the intense Asian dust aerosols. Even in the aerosol less-intense areas, OC2 chlorophyll was overestimated primarilybecause of blue-overcorrection of the SAC algorithm. SSMM retrieved more realistic estimates in these areas.

sediments (SS = 2–120 g m−3) by means of resuspension ofbottom sediments [1]. Around these areas we conducted fieldmeasurements of water samples for 20 and 23 October 1998,representing good weather conditions with clear sky and windspeed (5–10 m s−1). Analysis of the water samples indicatedconcentrations of chlorophyll (Chl), suspended sediments(SS) and dissolved organic matter (DOM) varying from 0.6–2.5 mg m−3, 2 to 120 g m−3 and 0.38 to 1.8 m−1, respectively.Concurrently, the above-water radiance spectra were collected

in these waters with an ASD FieldSpec Pro Dual UV/VNIRSpectroradiometer with spectral range from 350–1050 nm andspectral sample interval of 1.4 nm. This instrument wascalibrated every year and several inter-calibrations with otherinstruments were performed to confirm its stability. In order tochoose more appropriate measurement data for the validationprocess, the time difference for radiometric measurementswas kept ±2 h, the areas partly or fully covered by cloudswere disregarded, the measured radiances in a spectral sample

522


Figure 8. Horizontal profiles of SNR and TOA radiance (mWcm−2 μm−1 sr−1) at NIR2 along the 500 km transects from SeaWiFSimages of 6 April 2000 and 21 March 2001 in the East Sea(figures 5(a) and (b)). Note that to maximize variations the S/N ratiovalues (Y -axis in (b)) are scaled.

interval of 1.4 nm were averaged over bandwidth of eachSeaWiFS band, and some measurements which had occurredaround very shallow and small/narrow tidal channel areas werediscarded from the validation because of possible discrepanciesfrom sub-pixel variability and surface adjacency effects causedby the spatial resolution (1.1 km) of SeaWiFS sensor.

SeaWiFS images of the above periods were processed byuse of the SAC and SSMM algorithms and the retrieved water-leaving radiances, for example, at 412, 555 and 865 nm, for23 October 1998 are compared in figure 9. This image isshown because the effect of ignoring water signal or its poorestimates in the NIR1 and NIR2 is greatest at the shorterwavelengths of the spectrum. Though the SAC algorithmretrieved the water-leaving radiances in relatively less turbidwaters off the complex islands, it showed a complete failureto retrieve these quantities in shallow waters (<15 m) withvery high SS concentrations around the islands and tidalchannel areas. This may be attributable to the single scatteringaerosol reflectance ratio ε = ρas(765)/ρas(865) deducedfrom the image by the SAC algorithm, which reached highervalues causing error flagging of the algorithm. Allowingthese calculations to process the data would have resulted inlarge errors of the algorithm associated with negative water-leaving radiances in the violet and blue bands. Note thatthe turbid water pixels around the island areas previouslymasked by the SAC algorithm were successfully recoveredby the SSMM algorithm, which exhibits a highly reflectivewater mass movement with a clear spatial structure towardsthe offshore. The retrieved water-leaving radiances correlatedwell with the known SS distributions. In contrast, the invalidassumption/inappropriate in-water model associated with SACalgorithm led to an overestimated ε = ρas(765)/ρas(865)

which removed excessive aerosol path radiance in the violetand blue bands, resulting in very low and improbable negativewater-leaving radiances for these waters.

Figure 9. Comparison of the water-leaving radiances (mW cm−2 μm−1 sr−1) at three wavelengths 412, 555 and 865 nm retrieved fromSeaWiFS imagery (23 October 1998) using the SAC and SSMM algorithms. The upper panels indicate failure of the SAC algorithm toretrieve the Lw which led to the masking of the largest part of the turbid coastal areas. The bottom panels indicate efficiency of SSMM inrecovering the previously masked turbid coastal waters. The rectangular box in the bottom panel (412 nm image) indicates the ship transectarea covering the coastal areas of the Korean Southwest Sea in October 1998.

523


Table 6. Mean relative error (MRE) and RMSE deviation (band averaged). Note that MRF and RMSE decrease by a significant amount afterthe SSMM atmospheric correction scheme was applied to SeaWiFS image data of 20 and 23 October 1998.

Wavelength (nm)

Method 412 443 490 510 555 670 765 865 RMSE

SAC −0.52 −0.33 −0.28 −0.26 −0.3 −0.5 −0.79 −0.81 0.326SSMM 0.16 0.42 0.16 0.08 −0.08 −0.003 −0.044 0.46 0.076

Figure 10. Comparison of in situ measurements (from ASDFieldSpec Pro Dual UV/VNIR Spectroradiometer) of water-leavingradiance (a) with coincident spectra (b) from SeaWiFS image on 23October 1998 in highly turbid coastal waters of the KoreanSouthwest Sea. Satellite Lw are from the SSMM.

Figures 10(a) and (b) compare the in situ water-leavingradiance spectra (see the ship transect area in figure 9) withconcurrent spectra retrieved from the SeaWiFS image (of 23October 1998) by use of the SSMM in highly turbid Koreansouthwest coastal waters (SS = 3.5–28 g m−3). While theSAC algorithm masked these areas, SSMM appeared to haverecovered the water-leaving radiances which closely match themeasured spectra in these waters. The water-leaving radiancevalues across the whole spectrum markedly increased withincreasing SS concentrations that augmented backscatteringmore than absorption towards the longer wavelengths. Thisdrew the attention of the researchers and encouraged them toexploit the usefulness of single band radiance or reflectance atthese wavelengths for remote estimation of SS from satellitese.g. [28, 56, 57]. This also suggests the invalidity of the

assumption of negligible water-leaving radiance at 765 and865 nm in these waters. Notice that the retrieved water-leavingradiance values are slightly lower than the measured values,probably due to sub-pixel variability of the sediment featuresobserved by the SeaWiFS sensor.

To better illustrate the merit of atmospheric correction,satellite estimates of water-leaving radiances were comparedwith the corresponding in situ measurements in each band ofthe SeaWiFS images collected on 20 and 23 October 1998.Statistical analyses were also performed to provide additionalinformation on how accurately the satellite retrieval agreeswith in situ ship measurements. Thus, the mean relative error(MRE) to characterize the bias of the algorithms (negative, ifthe algorithm underpredicts; positive, if it overpredicts) wascomputed using equation (13)

MRE =1n

∑n1 (retrieved − measured)

1n

∑n1 (measured)

(13)

and the root mean square error (RMSE) to characterize thebias of the algorithms in absolute terms was computed usingequation (14)

RMSE =√

1n

∑n1 (retrieved − measured)2

1n

∑n1 (measured)

. (14)

Scatter plots of satellite verses in situ measurements ofwater-leaving radiances in each band were also generated(figure 11), for which the coefficient of determination (r 2),slope (S) and intercept (I ) were inspected for both the SACand SSMM algorithms. The in situ spectra of Lw collected on20 October 1998 were used for the SAC algorithm (becauseit showed a complete failure for the ship transect area on 23October 1998) and 20 and 23 October 1998 for the SSMMalgorithm.

Table 6 depicts the values of the MRE and RMSE ofwater-leaving radiance retrievals with the SAC and SSMMalgorithms. In turbid coastal waters, the large MRE (inall the SeaWiFS bands) coupled with the SAC algorithmwere significantly reduced and the band average RMSEwas decreased from 0.326 to 0.076 with SSMM. Similarly,scatter plots in figure 11 show that the water-leaving radianceretrievals with SSMM are more consistent with ship data thanthose obtained with the SAC algorithm. In particular, theregression slopes of the correlation between retrieved and ship-measured water-leaving radiances were close to 1.0, excludingbands 1, 2 and 7 which yielded slightly lower values: 0.42,0.50 and 0.44, respectively. However, these values are stillbetter than those of the SAC algorithm. SSMM also revealedhigher r 2 values of the regressions than the SAC algorithm.The high values of S and r 2 indicate water-leaving radiances

524


SSMM SAC

n = 19 9

r2 = 0.86 0.83

S = 0.72 0.047

I = 0.02 0.0015

SAC failure

865nm0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35

Ship Data

Sa

tellite

Da

ta

Figure 11. Comparison of in situ (ship-measured) and satellite water-leaving radiances (mW cm−2 μm−1 sr−1) from the SAC and SSMMalgorithms on SeaWiFS images of 20 and 23 October 1998 in the coastal bays (Jin-do and Wan-do) of the Korean Southwest Sea. Match-up ofLw on 20 October 1998 was used for the SAC algorithm and 20 and 23 October 1998 for the SSMM algorithm. This is because the SACalgorithm was unable to retrieve water-leaving radiance values (circled area in the 864 m image) from the SeaWiFS imagery of 23 October1998 in highly turbid coastal waters of the KSS. ��—SSMM (relatively clear waters), —SAC (relatively clear waters), ◦—SSMM (turbidwaters).

retrieved from SSMM agree well with in situ data. In contrast,the SAC algorithm had poor estimates, exhibiting a clear biastoward underestimation in all the SeaWiFS bands. This couldbe partly explained by the expressions in equation (6) in which

the water-leaving radiance values at the NIR bands are assumedto be negligible (in the past version of SeaDAS) or modelledvalues by the SAC algorithm are not sufficient to deal withturbid waters.

525


Figure 12. Comparison of the chlorophyll concentrations (mg m−3) retrieved from (a) SAC-OC2 algorithms (notice the atmosphericcorrection failure which necessitates masking of the turbid coastal areas and also leads to high chlorophyll values in the aerosol-dominatedregions off the North Korean and Chinese coasts) and (b) SSMM-OC2 algorithms (success of atmospheric correction recovering thepreviously masked turbid coastal areas and showing reliable estimates of chlorophyll concentrations in the aerosol-dominated areas) usingSeaWiFS imagery on 27 March 2002 over the YS environment.

The second validation of atmospheric correction tookplace in the YS, where the in situ measurements of chlorophyll,suspended sediments, nutrients (NO3–N, PO4–P, SiO2–Si),salinity and temperature were performed before and duringthe decomposition phase of a massive phytoplankton bloomon 27 March 2002. SeaWiFS imagery collected on this daywas atmospherically corrected by using the SAC and SSMMalgorithms and subsequently processed to chlorophyll by usingthe OC2v4 bio-optical algorithm. Figures 12(a) and (b) showthe distribution of chlorophyll concentrations in and aroundthe YS waters. First, notice the large discrepancies betweenthese images, particularly in waters along the west coastalareas of Korea (north), where the SAC algorithm actuallyprocessed the image but yielded concentrations that are toohigh because of the high SS and DOM concentrations and abackground Asian dust aerosol. In addition, the SAC algorithmmasked most of SS-dominated waters along the Chinese andKorean southwest coastal areas (indicated by arrows). Thisputs forward the failure of both the black-pixel assumptionfor the NIR and the iterative NIR water-leaving radiancecorrection scheme [22, 58]. This is in contrast with the resultsof SSMM recovering the previously masked coastal areas andallowing chlorophyll estimates that are consistent with the fieldobservations.

To better exemplify the differences between the SACand SSMM algorithms a transect (in figure 12(a)) of thechlorophyll concentrations and water-leaving radiances wasestablished running across the bloom from Korean westcoastal waters to Chinese coastal waters. Figures 13(a)and (b) display the horizontal variations of chlorophyllconcentrations and water-leaving radiances in the SeaWiFSbands centred at 412, 443, 490 and 865 nm along thistransect area. Both atmospheric correction schemes retrievedmarkedly elevated water-leaving radiance values (in thesebands) in the vicinity of coastal areas and reduced values from45–105 pixels in the central YS. Such strong/elevated water-leaving radiances can be related to high reflective SS materials

Figure 13. Comparison of chlorophyll concentrations andwater-leaving radiances at 412, 443, 490 and 865 nm (from SeaWiFSimagery of 27 March 2002) along a transect area with the sampleinterval of 2 pixels (figure 12(a)) running across the bloom frombright (high relative reflectance) cold coastal waters of Korea in theeast to bright coastal waters of China in the west.

while the reduced/weaker signal to profound absorption bypigments. However, one can observe that the SAC algorithm

526


Figure 14. Comparison of in situ chlorophyll concentrations (mg m−3) along the ship-transect lines (A–E in figure 12(b)) with the SeaWiFSmeasurements on 27 March 2002 in the Yellow Sea. Field measurements of chlorophyll indicate the absence of the bloom in February 2002and occurrence of the bloom in April 2002.

(This figure is in colour only in the electronic version)

retrieved comparatively low water-leaving radiance valuesin all the bands throughout the transect area, particularlyyielding negative water-leaving radiances at 412 nm in thebloom-dominated waters (from 45 to 105 pixels) of the YS(figure 13(a)). The negative and nearly zero water-leavingradiance values at 412 and 865 nm may result in part from useof the inappropriate in-water model to estimate the contributedsignal in the NIR bands. In contrast, the effect of the SSMMin raising water-leaving radiances at these bands is clearly seenthroughout the transect area in figure 13(b). This improvementin terms of increasing water-leaving radiances in the violet andblue bands led to better estimates of chlorophyll concentrationsthan those overestimated by the SAC algorithm in bloom-dominated and turbid coastal waters.

Figure 14 compares the in situ measurements ofchlorophyll concentrations with satellite estimates fromthe SAC-OC2 bio-optical algorithm, SAC-OC4 bio-opticalalgorithm (table 5) and SSMM-OC2 bio-optical algorithm onSeaWiFS imagery of 27 March 2002 in the YS. Note thatvery low chlorophyll concentrations were observed duringthe absence of the phytoplankton bloom in February andhigh chlorophyll concentrations during the decompositionphase of the bloom in April 2002 (one–two weeks timedifference with satellite overpass). Whilst the atmosphericcorrection was performed by the SAC algorithm on SeaWiFSimagery, two different bio-optical algorithms such as theOC2 and OC4 behaved differently in terms of yieldingchlorophyll concentrations: i.e. the OC4 algorithm gaveimproved estimates of chlorophyll concentrations over theOC2 algorithm at all stations. This is primarily because theOC4 uses the maximum band ratios compared to the OC2algorithm (table 5). Nevertheless, both the algorithms showeda clear tendency of an overestimation of the chlorophyllconcentrations in the bloom-dominated waters (A3, A7,B3, B5, B7, C5, C7, C9, D7, D9, E9) off the Koreanwest coast. Such large errors (overestimation) might resultfrom a pronounced overcorrection of atmospheric effects inthe violet and blue wavelengths, particularly in productivewaters with chlorophyll concentrations >2 mg m−3 [10, 20].

Note that at most of the stations the previously estimatedchlorophyll concentrations were brought down closer to thefield measurements by SSMM-OC2 algorithms. Due to thetime difference between the in situ and satellite observations,there was some disagreement resulting from spatial andtemporal variabilities of the spring bloom in the YS.

4. Conclusions

Coastal ocean colour observations specifically require accurateatmospheric correction algorithms that take into account thenon-zero water-leaving radiances in the NIR and includespecialized aerosol spectral models for correcting the effects ofstrongly absorbing aerosols over the coastal regions. However,the current SAC algorithm developed for SeaWiFS does notadequately address these conditions, thus limiting the SeaWiFSapplications mostly to the open ocean waters [11] wherethe assumption of zero water-leaving radiances in the NIRbands is valid [14]. In turbid coastal waters this assumptionis invalidated by the additional particulate scattering due tosuspended sediments. Occurrence of non-zero water-leavingradiances in the NIR ultimately leads to an overestimationof atmospheric effects (typically derived from NIR bands)which results in an atmospheric overcorrection in the visiblespectrum. The tendency of such an overcorrection is stillwell pronounced in the presence of absorbing aerosols froma thick vertical structure [21]. This leads to the water-leavingradiances of the violet and blue wavelengths to be severely,negatively or completely affected [10, 58].

More recent atmospheric correction algorithms accountfor the non-zero water-leaving radiances in the NIRbands by applying a combined approach to separationof the atmospheric and turbid water contributions to thesignals [19, 20, 34, 38, 59, 60]. Although these approachesdemonstrated a significant improvement over the SACalgorithm, an extensive research was required to providemore reliable correction procedures because the retrieval ofwater-leaving radiances and derivation of geophysical products

527


were not possible from satellite ocean colour imagery in thepresence of strongly absorbing aerosols over the coastal oceans(e.g. dust from Sahara desert in North Africa and Gobi desertin Northern China). More recent investigations [22, 42] useda spectral matching algorithm, with the assumption that thewater-leaving radiances in the NIR are negligible, for retrievingpigment concentrations from SeaWiFS imagery acquired in theregion of Sahara dust transport off the coast of Africa. Resultsof such specific studies revealed a considerable promise forprocessing SeaWiFS imagery in the presence of mineral dust,but pigment retrievals were not fully achieved and/or notfeasible when the aerosol optical thickness observed was ratherhigh for the dusty pixels. This suggests a likely error inthe estimation of the magnitude of aerosol radiance by thisoptimization technique.

To overcome these limitations associated with the SACalgorithm, a simple image-based atmospheric correctionmethod (SSMM) has been described and tested using theSeaWiFS imagery in this paper. The development of thismethod is in response to a large demand from the oceancolour community for generating the desired geophysicalproducts for the coastal regions. SSMM shows a goodpromise in delivering such products for these regions, withreduced time and cost when compared to the SAC algorithm.For example, a preliminary validation in turbid coastalwaters of the Korean Southwest Sea demonstrated that theSAC algorithm tended to have underestimated the water-leaving radiance values with large errors toward the shorterwavelengths and completely failed when regions of extremelyturbid water were identified in the top-of-atmosphere radiancesfor the NIR. SSMM reduces these errors and provides a goodreproduction of in situ water-leaving radiance spectra in thesewaters. The second validation, which was conducted usingSeaWiFS images of spring 2000 and 2001 (periods of stronglyabsorbing aerosols) off the Korean and Chinese coasts (regionsof great interest for the northeast Asian coastal managers,fishing community and naval operations), demonstrated thatthe SAC algorithm was unable to process the pixels of theaerosol-dominated areas and severely underestimated water-leaving radiances in the violet and blue spectral regions,leading to very high chlorophyll concentrations over the lessintense aerosol areas in the ES. This limited the potentialutility of SeaWiFS imagery for these regions. In contrast,the SSMM has been successful in removing much of theinfluence of Asian dust aerosols from SeaWiFS imageryand yielding reasonable estimates of both water-leavingradiance and chlorophyll concentrations for a greater coveragearea. The results, with an improved accuracy, highlightexisting anticyclonic eddy features, phytoplankton blooms andsuspended sediment transport off the Korean and Chinesecoasts. A further comparison of the match-ups of in situand SeaWiFS observations of chlorophyll in highly productivewaters of the YS ascertained an improved performance ofSSMM over the SAC algorithm. In these waters, theatmospheric overcorrection by the SAC algorithm increaseddramatically with chlorophyll concentrations, reducing theactual water-leaving radiance values of the violet and bluespectrum by about at least 50–75% when the chlorophyllexceeded 4 mg m−3.

Implementation of the SSMM on SeaWiFS images sug-gests that accurate retrievals of the water-leaving radiances andchlorophyll concentrations are doubtful in the regions wherethe effects of straylight and bright signal at the scan edges arestrong enough to influence the SeaWiFS measurements at theTOA. Our results indicated an overestimation of the water-leaving radiance retrieval by about 1.5–2 times for such ar-eas which might be disregarded. In the future refinement ofSSMM, it is planned to include a procedure to remove thewhitecaps effects, additional aerosol spectral models for boththe weakly and strongly absorbing aerosols, and conduct ex-tensive validation with numerous match-ups of SeaWiFS andin situ data in turbid coastal waters.

Acknowledgments

This research has been a part of the Development of CaseII water Bio-optical and Atmospheric correction Algorithmsfor the Geostationary Ocean Colour Sensor (GOCI) onCommunication Ocean Meteorological Satellite (COMS) in2008 and Multispectral Camera (MSC) on Korea MultipurposeSatellite 2 (KOMPSAT-2) in 2006, receiving support from theMinistry of Maritime Affairs and Fisheries (MOMAF) andMinistry of Science and Technology (MOST) through KORDIcontracts PM 294-00 and PN 524-00. We thank the anonymousreviewers for their valuable comments and suggestions.

References

[1] Ahn Y H and Shanmugam P 2004 New methods for correctingthe atmospheric effects in Landsat imagery over turbidwaters Korean J. Remote Sens. 20 289–305

[2] Morel A and Prieur L 1977 Analysis of variations in oceancolour Limnol. Oceanogr. 22 709–22

[3] Gordon H R, Clark D K, Muller J L and Hovis W A 1980Phytoplankton pigments from the Nimbus-7 coastal ZoneColor Scanner: comparisons with surface measurementsScience 210 63–6

[4] Gordon H R, Brown O B, Evans R H, Brown J W, Smith R C,Baker K S and Clark D K 1988 A semianalytic radiancemodel of ocean colour J. Geophys. Res. 93 10909–24

[5] Gregg W W and Conkright M E 2001 Global seasonalclimatologies of ocean chlorophyll: blending in situ andsatellite data for the Coastal Zone Color Scanner eraJ. Geophys. Res. 106 2499–515

[6] McClain C R, Cleave M L, Feldman G C, Gregg W W,Hooker S B and Kuring N 1998 Science quality SeaWiFSdata for global biosphere research Sea Technol. 39 10–6

[7] Morel A and Andre J M 1991 Pigment distribution and primaryproduction in the Western Mediterrnean as derived andmodeled from Coastal Zone Color Scanner observationsJ. Geophys. Res. C 96 12685–98

[8] Sathyendranath S and Morel A 1983 Light emerging from thesea—interpretation and uses in remote sensing RemoteSensing Applications in Marine Science and Technologyed A P Cracknell (Dordrecht: Reidel) pp 323–57

[9] Smith R C and Baker K S 1981 Optical properties of theclearest natural waters (200–800 nm) Appl. Opt. 20 177–84

[10] Darecki M and Stramski D 2004 An evaluation of MODIS andSeaWiFS bio-optical algorithms in the Baltic Sea RemoteSensing Environ. 89 326–50

[11] Sathyendranath S (ed) 2000 Remote sensing of ocean colour incoastal, and other optically-complex, waters IOCCG Reportvol 3 (Dartmouth, Nova Scotia, Canada: IOCCG ProjectOffice)

528

http://dx.doi.org/10.1126/science.210.4465.63

http://dx.doi.org/10.1029/1999JC000028

http://dx.doi.org/10.1016/j.rse.2003.10.012


[12] Antoine D and Morel A 1999 A multiple scattering algorithmfor atmospheric correction of remotely sensed ocean colour(MERIS instrument): principle and implementation foratmospheres carrying various aerosols including absorbingones Int. J. Remote Sens. 20 1875–916

[13] Gordon H R 1978 Removal of atmospheric effects fromsatellite imagery of the oceans Appl. Opt. 17 1631–6

[14] Gordon H R and Wang M 1994 Retrieval of water-leavingradiance and aerosol optical thickness over the oceans withSeaWiFS: a preliminary algorithm Appl. Opt. 33 443–52

[15] Hooker S B, Esaias W E, Feldman G C, Gregg W W andMcClain C R 1992 SeaWiFS Technical Report Series vol 1An overview of SeaWiFS and Ocean Colour (NASATechnical Memorandum 104566, July 1992)

[16] Bricaud A and Morel A 1987 Atmospheric corrections andinterpretation of marine reflectances in CZCS imagery: useof a reflectance model Oceanol. Acta 7 33–50

[17] Viollier M, Tanre D and Deschamps P Y 1980 An algorithm forremote sensing of water colour from space Bound. LayerMeteorol. 18 247–67

[18] Hu C, Carder K L and Muller-Karger F E 2000 Atmosphericcorrection of SeaWiFS imagery over turbid waters: apractical approach Remote Sens. Environ. 74 195–206

[19] Ruddick K G, Ovidio F and Rijkeboer M 2000 Atmosphericcorrection of SeaWiFS imagery for turbid and inland watersAppl. Opt. 39 897–913

[20] Siegel D A, Wang G M, Maritorena S and Robinson W 2000Atmospheric correction of satellite ocean colour imagery:the black pixel assumption Appl. Opt. 39 3582–91

[21] Chomko R M and Gordon H R 1998 Atmospheric correction ofocean colour imagery: use of the Junge power-law aerosolsize distribution with variable refractive index to handleaerosol absorption Appl. Opt. 37 5560–72

[22] Moulin C, Gordon H R, Chomko R M, Banzon V F andEvans R H 2001 Atmospheric correction of ocean colourimagery through thick layers of Saharan dust Geophys. Res.Lett. 28 5–8

[23] Chavez P S 1988 An improved dark-object subtractiontechnique for atmospheric scattering correction ofmultispectral data Remote Sens. Environ. 24 459–79

[24] Hall F G, Strebel D E, Nickeson J E and Goetz S J 1991Radiometric rectification: toward a common radiometricresponse among multidate, multisensor images Remote Sens.Environ. 35 11–27

[25] Richter R 1996 A spatially adaptive fast atmospheric correctionalgorithm Int. J. Remote Sens. 17 1201–14

[26] Chavez P S 1996 Image-based atmosphericcorrections—revisited and improved Photogramm. Eng.Remote Sens. 62 1025–36

[27] Tanre D, Deschamps P Y, Devaux C and Herman M 1998Estimation of Saharan aerosol optical thickness fromblurring effects in Thematic Mapper data J. Geophys. Res.93 15955–64

[28] Ahn Y H, Shanmugam P and Moon J E 2006 Retrieval of oceancolour information from Multi-spectral Camera onKompsat-2 for monitoring highly dynamic ocean featuresInt. J. Remote Sens. 27 367–92

[29] Liang S, Fang H and Chen M 2001 Atmospheric correction ofLandsat ETM +land surface imagery—Part 1: methodsIEEE Trans. Geosci. Remote Sens. 39 2490–8

[30] Moran M S, Jackson R D, Slater P N and Teillet P M 1992Evaluation of simplified procedures for retrieval of landsurface reflectance factors from satellite sensor outputRemote Sens. Environ. 41 169–84

[31] Farrand W H, Singer R B and Merenyi E 1994 Retrieval ofapparent surface reflectance from AVIRIS data—acomparison of empirical line, radiative transfer and spectralmixture methods Remote Sens. Environ. 47 311–21

[32] Wu J, Wang D and Bauer M E 2005 Image-based atmosphericcorrection of QuickBird imagery of Minnesota croplandRemote Sens. Environ. 99 315–25

[33] Brivio P A, Giardiano C and Zilioli E 2001 Determination ofconcentration changes in Lake Garda using image-based

radiative transfer code for Landsat TM images Int. J. RemoteSens. 22 487–502

[34] Hu C, Muller-Karger F E, Andrefouet S and Carder K L 2001Atmospheric correction and cross-calibration ofLandsat-7/ETM+imagery over aquatic environments: Amultiplatform approach using SeaWiFS/MODIS RemoteSens. Environ. 78 99–107

[35] Barnes R A and Zalewski E F 2003 Reflectance-basedcalibration of SeaWiFS. II. Conversion to radiance Appl.Opt. 42 1648–60

[36] Hooker S B, Firestone E R and Acker J C 1994 SeaWiFSPre-launch Radiometric Calibration and SpectralCharacterization, (SeaWiFS technical report series) NASATechnical Memorandum 104566 vol 22 (Greenbelt, MD:NASA Goddard Space Flight Centre)

[37] Deschamps P Y, Herman M and Tanre D 1983 Modeling of theatmospheric effects and its application to the remote sensingof ocean colour Appl. Opt. 22 3751–8

[38] Arnone R A, Martinolich P, Gould R W, Sydor M andStumpf R P 1998 Coastal optical properties using SeaWiFS.Ocean Optics XIV Conf., (Kailua-Kona, Hawaii, Nov. 1998);Ocean Optics XIV CD-ROM (Washington, DC: Office ofNaval Research)

[39] Gould R W and Arnone R A 2002 Coastal optical propertiesestimated from airborne sensors. Reply to the comments byHu and Carder Remote Sens. Environ. 79 138–42

[40] Iwasaka Y, Yamato M, Imasu R and Ono A 1988 Transport ofAsian dust (kosa) particles; importance of weak kosa eventson the geochemical cycles of soil particles Tellus B 40495–503

[41] Wang M 1999 A sensitivity study of the SeaWiFS atmosphericcorrection algorithm: effects of spectral band variationsRemote Sens. Environ. 67 348–59

[42] Gordon H R 1997 Atmospheric correction of ocean colourimagery in the Earth Observing Era J. Geophys. Res. D102 17081–106

[43] Koepke P 1984 Effective reflectance of oceanic whitecaps Appl.Opt. 23 1816–24

[44] Song C, Woodcock C E, Seto K C, Lenney M P andMacomber S A 2000 Classification and change detectionusing Landsat TM data: when and how to correctatmospheric effects? Remote Sens. Environ. 75 230–44

[45] Coppin P R and Bauer M E 1994 Processing of multi-temporalLandsat TM imagery to optimize extraction of forest coverchange features IEEE Trans. Geosci. Remote Sens.32 918–27

[46] Yang H and Gordon H R 1997 Remote sensing of ocean colour:assessment of water-leaving radiance bidirectional effects onatmospheric diffuse transmittance Appl. Opt. 36 7887–97

[47] Cheng T, Lu D, Wang G and Xu Y 2005 Chemicalcharacteristics of Asian dust aerosol from HunshanDake Sandland of Northern China Atmos. Environ.39 2903–11

[48] Fukushima H and Toratani M 1997 Asian dust aerosol: opticaleffect on satellite ocean colour signal and a scheme of itscorrection J. Geophys. Res. 102 17119–30

[49] Liu X, Zhu J, Van Espen P, Adams F, Xiao R, Dong S andLi Y 2005 Single particle characterization of spring andsummer aerosols in Beijing: formation of composite sulfateof calcium and potassium Atmos. Environ. 39 6909–18

[50] Higurashi A and Nakajima T 2002 Detection of aerosol typesover the East China Sea near Japan from four-channelsatellite data Geophys. Res. Lett. 29 1–4

[51] Cheng M T, Lin Y C, Chio C P, Wang C F and Kuo C Y 2005Characteristics of aerosols collected in central Taiwanduring an Asian dust event in spring 2000 Chemosphere61 1439–50

[52] Yuan C-S, Sau C-C, Chen M-C, Huang M H, Chang S-W,Lin Y-C and Lee C G 2004 Mass concentration andsize-resolved chemical composition of atmospheric aerosolssampled at the Pescadores Islands during Asian dust storm inthe years of 2001 and 2002 TAO 15 857–879

529

http://dx.doi.org/10.1080/014311699212533

http://dx.doi.org/10.1007/BF00122023

http://dx.doi.org/10.1016/S0034-4257(00)00080-8

http://dx.doi.org/10.1029/2000GL011803

http://dx.doi.org/10.1016/0034-4257(88)90019-3

http://dx.doi.org/10.1016/0034-4257(91)90062-B

http://dx.doi.org/10.1080/01431160500300479

http://dx.doi.org/10.1109/36.964986

http://dx.doi.org/10.1016/0034-4257(92)90076-V

http://dx.doi.org/10.1016/0034-4257(94)90099-X


http://dx.doi.org/10.1080/014311601450059

http://dx.doi.org/10.1016/S0034-4257(01)00252-8

http://dx.doi.org/10.1016/S0034-4257(01)00233-4

http://dx.doi.org/10.1016/S0034-4257(98)00095-9

http://dx.doi.org/10.1029/96JD02443

http://dx.doi.org/10.1016/S0034-4257(00)00169-3

http://dx.doi.org/10.1109/36.298020

http://dx.doi.org/10.1016/j.atmosenv.2004.12.045

http://dx.doi.org/10.1029/96JD03747

http://dx.doi.org/10.1016/j.atmosenv.2005.08.007

http://dx.doi.org/10.1029/2002GL015357

http://dx.doi.org/10.1016/j.chemosphere.2005.04.120


[53] O’Reilly J E, Maritorena S, Mitchell B G, Seigel D A,Carder K L, Garver S A, Kahru M and McClain C 1998Ocean colour chlorophyll algorithms for SeaWiFSJ. Geophys. Res. 103 24937–53

[54] Ahn Y H, Shanmugam P, Chang K I, Moon J E and Ryu J H2005 Spatial and temporal aspects of phytoplankton bloomsin complex ecosystems off the korean coast from satelliteocean color observations Ocean Sci. J. 40 67–78

[55] Zuenko Y, Selina M and Stonik I 2006 On conditions ofphytoplankton blooms in the coastal waters of theNorth-Western East/Japan Sea Ocean Sci. J. 41 31–41

[56] Ahn Y H, Moon J E and Gallegos S 2001 Development ofsuspended particulate matter algorithms for ocean colourremote sensing Korean J. Remote Sens. 17 285–95

[57] Stumpf R P and Pennock J R 1991 Remote estimation of thediffuse attenuation coefficient in a moderately turbid estuaryRemote Sens. Environ. 38 182–91

[58] Bailey S W and Werdell P J 2006 A multi-sensor approach forthe on-orbit validation of ocean colour satellite data productsRemote Sens. Environ. 102 12–23

[59] Gould R W and Arnone R A 1994 Extending Coastal ZoneColor Scanner estimates of the diffuse attenuation coefficientinto Case II waters Ocean Optics XII; Proc. SPIE2258 342–56

[60] Lavender S J, Pinkerton M H, Moore G F, Aiken J andBlondeau-Patissier D 2005 Modification to the atmosphericcorrection of SeaWiFS ocean colour images over turbidwaters Cont. Shelf Res. 25 539–55

[61] Vermote E F, Saleous N Z and Justice C O 2002 Atmosphericcorrection of MODIS data in the visible to middle infrared:first results Remote Sens. Environ. 83 97–111

[62] Murayama T et al 2001 Ground-based network observation ofAsian dust events of April 1998 in east Asia J. Geophys. Res.106 18345–59

[63] O’Reilly J E et al 2000 Ocean Colour Chlorophyll aAlgorithms for SeaWiFS OC2 and OC4: version 4, SeaWiFSPostlaunch Calibration and Validation Analyzes: Part 3(SeaWiFS Postlaunch Technical Report Series vol 11) edS B Hooker and E R Firestone (Greenbelt, MD: NASAGoddard Space Flight Centre) pp 9–23

530

http://dx.doi.org/10.1029/98JC02160

http://dx.doi.org/10.1016/0034-4257(91)90088-N


http://dx.doi.org/10.1117/12.190077

http://dx.doi.org/10.1016/j.csr.2004.10.007

http://dx.doi.org/10.1016/S0034-4257(02)00089-5

http://dx.doi.org/10.1029/2000JD900554

new atmospheric correction technique to retrieve the … shanmugam... · new atmospheric correction...

Documents