Early phase analysis of OCTS radiance data for aerosol remote sensing

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 37, NO. 3, MAY 1999 1575

Early Phase Analysis of OCTS RadianceData for Aerosol Remote Sensing

Teruyuki Nakajima, Akiko Higurashi, Kazuma Aoki, Tatsuo Endoh, Hajime Fukushima,Member, IEEE,Mitsuhiro Toratani, Yasushi Mitomi, B. Greg Mitchell, and Robert Frouin

Abstract—An analysis of ocean color temperature scanner[(OCTS) on board the advanced earth observation satellite(ADEOS)] spectral radiance data was performed for retrievingglobal distributions of Angstrom factor and exponent, whichrepresent the aerosol optical thickness at a reference wavelength(500 nm in our study) and a spectral dependence of the opticalthicknesses, respectively, over ocean. Determination of calibrationconstants for OCTS-received radiances and development of anefficient look-up table method for synthesizing the radiances arekey issues for development of the present two-channel algorithmwith use of channel 6 and 8 radiances of OCTS. This algorithmhas been applied to Level-1B OCTS GAC data sets for producingthree month (April, May, and June 1997) global distributions ofAngstrom parameters. Geographical and seasonal distributionpatterns of Angstrom parameters suggest that anthropogenicsulfate aerosols in mid-latitudes and biomass burning aerosolsin the subtropical region are characterized by small particleshaving large Angstrom exponents, whereas mineral dustparticles from subtropical arid regions are characterized bylarge particles having small Angstrom exponents. There wasa fairly good agreement between satellite-retrieved values ofAngstrom parameters and values obtained by sky radiometerslocated on coasts.

I. INTRODUCTION

FOR more than two decades, efforts have been made toacquire the ocean color signature with satellite-borne ra-

diometers. Coastal Zone Color Scanner (CZCS) on NIMBUS-7was a successful ocean color imager that produced convincingglobal distributions of the chlorophyll-a pigment concentra-tion. One important experience from the CZCS data analysisis that correction of atmospheric path radiance due to aerosolsis necessary for successful inversion of water leaving radi-ances (e.g., [1], [2]). Hence, the CZCS algorithm includes an

Manuscript received May 31, 1998; revised December 20, 1998. Thiswork was supported by NASDA OCTS/GLI projects and the Grant-in-Aidfor Scientific Research on Priority Areas (Grant 08241104) of the JapaneseMinistry of Education, Science, Sports and Culture. The work of B. G.Mitchell and R. Frouin was supported, respectively, by NASA Grants NAGW-6559 and NAGSS-97135.

T. Nakajima is with the Center for Climate System Research, Universityof Tokyo, Tokyo 153-8904, Japan (e-mail: [email protected]).

A. Higurashi is with the National Institute for Environmental Studies,Ibaraki 305-0053, Japan.

K. Aoki and T. Endoh are with the Institute of Low Temperature Science,Hokkaido University, Sapporo 060-0819, Japan.

H. Fukushima and M. Toratani are with the School of High-Technology forHuman Welfare, Tokai University, Numazu 410-0395, Japan.

Y. Mitomi is with the Remote Sensing Technology Center of Japan, Tokyo,Japan.

B. G. Mitchell and R. Frouin are with Scripps Institution of Oceanography,University of California, San Diego, La Jolla, CA 92093-0218 USA.

Publisher Item Identifier S 0196-2892(99)03582-2.

estimation of wavelength dependence of path radiances due toaerosols using the two longest wavelengths, by which atmo-spheric path radiances at shorter wavelengths are estimated byextrapolation. This wavelength dependence index is also veryuseful information regarding the aerosol size distribution sothat modern ocean color imagers are useful for aerosol remotesensing. An ocean color imager may provide significantlybetter aerosol remote sensing than traditional methods basedon AVHRR and geostationary satellites radiometers [3], [4].Ocean color imagers have more sensitive channels in thevisible-near infrared spectral region with narrowband widthsas compared with the broad channels 1 and 2 of AVHRR.While limited, one-channel radiance analyses of AVHRRand geostationary satellite radiometers have provided usefulinformation of large-scale distribution of the aerosol opticalthickness, suggesting noticeable contributions from fossil fuel-burning aerosols, biomass burning aerosols, and mineral dustaerosols [4], [5]. The above discussion suggests that radiancedata analysis of an ocean color imager will add significantlynew information to our knowledge of global scale aerosoloptical properties.

In this paper, we report early phase study of global scaleaerosol remote sensing with channels 6 and 8 radiances of theocean color and temperature scanner (OCTS) radiometer onboard the ADEOS (Midori) polar orbiter that was launchedin August 1996. Data were obtained for approximately sevenmonths before shutdown of ADEOS in July 1997. This paperwill discuss determination of sensor calibration constants, thepresent algorithm for retrievingAngstrom parameters, andinterpretation of three monthly composites of OCTS-derivedaerosol data. Our approach for aerosol characterization is basedon the use of two-channel radiances in the visible-near infraredregion similar to Higurashi and Nakajima [6] for applicationto channels 1 and 2 radiances of NOAA/AVHRR. In the latterpart of the paper, we will discuss some comparison of theAVHRR retrievals with ground observation.

II. CALIBRATION OF OCTS RADIANCES

The most important but difficult challenge in ocean colorand aerosol remote sensing with a satellite-borne radiometeris calibration, and, therefore, there have been extensive cal-ibration efforts for characterizing satellite-borne radiometers(e.g., [7], [8] for POLDER radiometer). OCTS is a crosstracking imager with spatial resolution of 700 m and with12 channel, as summarized in Table I [9]. Each scan ofthe cross tracking mirror captures ten line images by ten

0196–2892/99$10.00 1999 IEEE

1576 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 37, NO. 3, MAY 1999

TABLE ISPECTRAL RANGES AND SOLAR IRRADIANCES OFOCTS CHANNELS. BAND LIMITS

(�min��max) AND EFFECTIVE EXTRATERRESTIAL IRRADIANCE F0 ARE LISTED

ch �center �min � �max F0(w=m2=�m)1 0.412 0.402-0.422 17102 0.443 0.433-0.453 18823 0.490 0.480-0.500 19464 0.520 0.510-0.530 18575 0.565 0.555-0.575 18456 0.670 0.660-0.680 15317 0.765 0.745-0.785 12268 0.865 0.845-0.885 9869 3.72 3.55-3.88 —10 8.53 8.25-8.80 —11 10.9 10.3-11.4 —12 12.0 11.4-12.5 —

detectors for each wavelength. Sensor characterization issueswith such an imaging mechanism are pixel registration amongchannels, relative calibration among ten detectors, and absolutecalibration for converting digital counts to radiances. Therehas been a significant effort for the sensor characterization bythe OCTS team of National Space Development Agency ofJapan/Earth Observation Research Center (NASDA/EORC),which will be described elsewhere. Radiometrically and geo-metrically corrected Level-1B/local area coverage (LAC) datasets have been archived using calibration constants determinedby EORC. Collaborating with NASDA, we have constructedthree-month composite Level-1B/global area coverage (GAC)data sets from the LAC data sets by resampling every fivepixels.

At an early stage of OCTS data analysis, we had fewsimultaneous vicarious calibration efforts for ocean color andaerosol retrievals. The CalCOFI 9610 campaign, organized byScripps Institution of Oceanography [10], was the only earlyphase experiment with complete sets of vicarious calibrationand validation data for ocean color and aerosol remote sensingwith simultaneous OCTS overflight. In the measurements, ashipborne sky radiometer (PREDE SKR-01S prototype) andthe University of Lille SIMBAD polarization radiometer wereon board the research vessel Roger Revelle for obtainingdirect/circum solar radiances and upwelling radiances fromthe ocean, respectively. The sky radiometer, PREDE SKR-01S prototype, is a multispectral photometer that can measuredirect solar irradiance and diffuse sky radiances at wavelengthsof 315, 400, 500, 870, 940, and 1040 nm. From direct solarirradiance and sky radiance distribution, we have obtained thecolumnar volume size distribution as a functionof particle radius by the method of Nakajimaet al. [11]to solve the following integral equation for the normalizedcircum solar radiances:

(1)

where is the direct solar irradiance and is the radiantflux received by the sky radiometer in a sky direction of polarangles with a solar positionis the solid view angle of the radiometer, and is thecontribution of multiple scattering. The kernel function isgiven in this paper by the Mie theory with a complex refractive

Fig. 1. Volume size distributions retrieved from the normalized sky radiancesoff Santa Maria Bay on November 1, 1996.

index of aerosols as 1.45–0.005which is a suitable valuefor the observation situation. The aerosol volume distribution

is then obtained after iteratively removing themultiple scattering contribution Movement of theship was detected and compensated by a sun tracker, GlobalPositioning System (GPS) sensor and level sensor of thePREDE SKR-01S system. As discussed in Nakajimaet al.[12], the retrieved size distribution is not strongly affected bycalibration of the radiometer and by the assumption of theground albedo and the refractive index if forward scatteringangles are used for the inversion. In our case we use ascattering angle range of for inversion of (1).The retrieval also does not depend strongly on the verticaldistribution of aerosols in case of almucantar scan with thezenith angle of viewing direction equal to the solar zenithangle, which is our case.

Due to frequent malfunction of the prototype SKR-01Ssystem, we have only one complete measurement set of skyradiances and upwelling radiances at a location on November1, 1996, off Santa Maria Bay (33.75N, 120.17 W). Therewas a high pressure (in a range of 1021 hPa and 1023 hPa) overthe western U. S. Great Basin with a weak wind condition of3 m/s from the northeast direction. Fig. 1 shows volume sizedistributions inverted by the above-mentioned method withnormalized radiances at wavelengths ofand nm. Reconstructed errors of the measured normal-ized radiances with the volume size distributions are within7%. There were three modes in the observed volume sizedistributions with mode radii less than 0.2m, around 1 mand larger than 10 m, suggesting three aerosol sources ofaccumulation mode, sea salt mode, and sea-spray and/or dustparticles, respectively. The size distributions have a dominantsubmicrometer mode and less dominant coarse particle mode,suggesting a strong gas-to-particle conversion activity on thisday.

With the volume size distributions thus retrieved, we cansynthesize the optical thickness, phase function, and further,OCTS-received radiances. Fig. 2 shows a time series of theaerosol optical thickness from sky radiometry as described

NAKAJIMA et al.: OCTS RADIANCE DATA FOR AEROSOL REMOTE SENSING 1577

Fig. 2. Time series of the aerosol optical thickness retrieved by sunphotom-etry and sky radiometry off Santa Maria Bay on November 1, 1996.

Fig. 3. Spectral dependence of the aerosol optical thicknesses at around11:00 on November 1, 1996.

above and from usual sunphotometry with the SIMBADradiometer. The SIMBAD radiometer can measure direct solarirradiance and upwelling diffuse radiance in spectral bandscentered at 443, 560, 670, and 870 nm. Polarization filtersare used to reduce reflected skylight in the instrument’s field-of-view. Fig. 3 shows the spectral dependence of opticalthickness from PREDE and SIMBAD radiometers at closelocal times; 11:05 and 11:08, respectively. There is an overallsimilarity between optical thickness spectra obtained by thetwo methods, but with a slight difference in the range of 20%.Unnatural fluctuations in the optical thickness from the skyradiometry as a function of time suggest some problem withthe sky radiometer data for obtaining the optical thicknessdue to movement of the ship to which the sky radiancemethod is sensitive. Nonetheless, we see some advantagein the sky radiometry method for obtaining aerosol pathradiance, since 1) the inversion of the normalized radiancedoes not need a calibration of the radiometer, 2) the normalizedradiance is almost proportional to the aerosol amount andhence is sensitive to a small optical thickness typical for theocean environment, and 3) the synthesis of satellite-received

Fig. 4. Correction factors of OCTS radiometers obtained by various meth-ods.

radiances is performed so as to be consistent with sky radiancedistribution, which is a stronger constraint than the opticalthickness wavelength dependence in the usual sunphotometry.

With a full radiative transfer package,P-Star (polarizationsubsystem of the system for transfer of atmospheric radia-tion) developed at the University of Tokyo, which includespolarization effects in a coupled atmosphere-ocean systemwith wind-roughened sea surface, we have simulated OCTS-received radiances with the optical parameters retrieved withthe sky radiometer data and the response functions of OCTSchannels provided by NASDA as follows:

(2)

where is the atmospheric path radiance plus surfacereflection contribution to the total radiance calculated with theatmosphere and wind-roughened sea surface system loadedwith aerosols having the size distribution retrieved by skyradiometry; is the water leaving radiance upwelling fromthe ocean body; and is the optical thickness ofthe atmosphere truncated so as to include diffuse transmission.The ocean surface reflection model is that of Nakajimaetal. [13] which is based on the Cox and Munk wave slopestatistics. We ignored the white cap effect discarding datawith wind velocity larger than 15 m/s. We also ignored up-welling radiation from ocean body because this effect is not sosignificant at wavelengths under consideration. Ozone amount,surface wind velocity, and water vapor profile were obtainedfrom TOMS archive, SSM/I archive, and NCEP objectiveanalysis, respectively. The simulation results are shown inFig. 4 and Table II in terms of correction factors labeledSD obs.in the following expression for the OCTS-receivedradiance in th channel:

(3)

where is the radiance estimated by the pre-flight calibration constants. According to NASDA, there isessentially no offset, so that we can assume In orderto study the effect of the assumed size distribution, two othercorrection factors are calculated using NASDA’s operationalalgorithm of CZCS type radiance simulation [14] with twoaerosol typesTR50andMT80, i.e., tropospheric aerosol model


TABLE IICORRECTION FACTORS OFOCTS RADIANCES OBTAINED BY VARIOUS METHODS

Fig. 5. Volume size distributions ofTR50, MT80, andSD.obs.

with relative humidity of 50% andmaritime aerosol modelwith relative humidity of 80%, respectively. Fig. 5 shows thatthe present retrieved size distribution is between these twosize distributions with respect to dominance of coarse particles.Correction factors corresponding to these two size distributionsare also shown in Fig. 4 and Table II with labelsTR50 andMT80. It is found that all the estimates agree within 1% forchannels 1–4, since the molecular scattering is dominant withthe small optical thickness less than 0.1 even in channel 1 inthe present case. On the other hand, the correction factors forchannels 5–8 depend on the methods with differences morethan 10% at maximum, with an increasing tendency withincreasing contribution of coarse particles. This observationsuggests that a reasonable retrieval of columnar size distribu-tion is essential for obtaining accurate calibration constants atlonger wavelengths with the present method.

In order to get independent estimates of the correction fac-tors, we have investigated OCTS radiances, plotting radiancesagainst aerosol-free theoretical radiances for each pixel on theglobe (Fig. 6). Since OCTS-received radiances in channels 1–8increase with increasing aerosol optical thickness outside thesun glint region and, without strong absorptivity of aerosols[15], all the observed radiances outside the sun glint region are

expected to be larger than the theoretical values for aerosol-free atmospheres

(4)

We have selected minimum radiances in eachlatitude–longitude box of 0.5 0.5 to compare with

which are calculated for aerosol-freeatmospheres with underlying sea surface. A suitablecalibration curve should be the upper bound of such aplot, excluding several pixels which cannot be explainedtheoretically. Based on a detailed examination of such plots,we have determined the coefficients of the calibrationcurves as W/m str m,

W/m str m, for channels 6and 8. We did not set to zero in this procedure becausewe feel there is an offset in the plots. For comparison,we drew several calibration lines with different calibrationmethods (Fig. 6). No difference was observed between theMT80 method and theSD-obsmethod. The convergence ofcalibration lines other than theTR50 line suggests that thistype of plot is useful to determine the bound for the region ofexistence of realistic calibration lines, at least for channel 6.Table II lists calibration coefficients obtained by this minimumradiance method shown in the column labeledRad.min.

Another two sets of the correction factor have been obtainedby Fukushima [16] as shown by values labeled MOBY andYBOM. The former is based on a tuning of simulated radiancesto AVIRIS overflight radiances and radiance values measuredby the NASA Marine Optical Buoy System [17] (MOBY) offHawaii. The latter is based on tuning of simulated radiancesso as to be consistent with radiance values measured by theNASDA Yamato Bank Optical Moored Buoy System [18](YBOM) in the Japan Sea with preset value of one forchannel 6. It is found in Table II and Fig. 4 that there is asignificant uncertainty in determined calibration coefficientsfor channel 8. In theSD-obsmethod, it is possible that thecoarse particle mode was underestimated in the retrieved sizedistribution of the CalCOFI 9610 campaign. In principle,vicarious calibration methods need many measurements toreduce measurement errors. For the radiance data analysisin the following sections,Rad.minvalues will be tentativelyused for calibration constants in channels 6 and 8, sincethe constants can generate reasonable aerosol distributionsconsistent with ground-based observation as shown in latersections.

III. D EVELOPMENT OF AN AEROSOL

REMOTE SENSING ALGORITHM

An efficient way of reconstructing theoretical values ofOCTS-received radiances is an important issue for realizingsuccessful retrievals. For this purpose, we have developed twotypes of look-up tables. One is adopted for NASDA’s oper-ational algorithm to synthesize radiances. For the discussionwe define the apparent reflectance as

(5)


Fig. 6. Plot of observed minimum radiances versus theoretical molecular scattering radiances in channel 6. Solid lines are calibrations obtained byvarious methods shown in Fig. 4 and Table II.

where is the extraterrestrial solar irradiance in thethchannel. The apparent reflectance is further decomposed intoseveral components as follows:

(6)

where

correction factor of gaseous absorption;linearized singly scattered component of aerosolsincluding gaseous absorption along the ray;aerosol-molecule interaction component;molecular scattering component;singly scattered sun-glint component including at-tenuation by molecules and aerosols.

and are, respectively, given as

(7)

and

(8)

where and are single scattering albedo and optical thick-ness of the airmass, is the scattering phase function, and

is the aerosol optical thickness at wavelength of 500nm. and include the interaction between sea surfaceand atmosphere but with a fixed mean wind velocity at 7m/s, because these components depend weakly on the windvelocity at 10 m above the sea surface. These approximationsare similar to those of Gordon and Wang [2] that expandthe aerosol-molecule interaction component by the linearized

aerosol single scattering since they observed the followingapproximation:

(9)

A benefit of (9) is that the aerosol optical thickness is directlyobtained by an inversion of the equation. Furthermore, thecoefficients are weak functions of angular coordinates. InNASDA’s operational algorithm, mainly due to computationalresource limit, the maximum order of expansion in (9) isset as or depending on angular condition of solarinsolation and emergent angles. We have developed a moreaccurate look-up table method for research use based onHigurashi and Nakajima [6]. The following formula is usedfor decomposition of the reflectance into singly and multiplescattered components:

(10)

(11)

and

(12)

The exponential term in (12) is introduced to compensatesome exponential dependence in (11), taking into account the


(a) (b)

Fig. 7. Errors involved in the synthetic apparent reflectances constructed by look-up tables.(�; �0) plane with� = 0:

observation of Gordon and Wang [2] that a linear dependenceis important in the total reflectance given by (10).

Gaseous absorption correction factors are given as

(13)

and

(14)

where is the absorption optical thickness of ozone. Absorp-tion by other gaseous components is included in the opticalthickness and single scattering albedo in (11) and (12). Thecorrection factor for water vapor is tabulated with angularcoordinates and water vapor amount as gridding parameters.Since the band widths of OCTS are small enough and band lo-cations are designed so as to avoid strong gaseous absorption,the gaseous absorption correction is not a significant issue inthe radiometry of OCTS, as compared with the situation ofAVHRR. For our use of channels 6 and 8 in this study, thewater vapor correction is not needed, so that we set

Fig. 7 shows an example of error analysis of the abovetwo formulas associated with (8) and (12) in the principalplane In the simulation,P-Star has been used forevaluating true radiances in the coupled-atmosphere oceansystem with wind roughened surface including polarizationeffects in the atmosphere and sea surface. It is found from thefigure that large errors exceeding0.001 appear in the regionof with (8) and whereas (12) with

gives the region of errors exceeding0.0001 only inTherefore, we will adopt (12) with for

data analysis shown in the next section.

IV. A NALYSIS OF OCTS RADIANCE DATA

In order to obtain the aerosol signature, we have applied thealgorithm of Higurashi and Nakajima [6] to spectral radiancesof channels 6 and 8 of OCTS. The retrieval parameters areAngstrom parameters in the following expression forthe spectral optical thickness as a function of wavelength

, i.e., Angstrom’s law:

(15)

Hereafter, and are referred to as theAngstrom factorand exponent. TheseAngstrom parameters are also observablein sunphotometry and sky radiometry and are suitable for val-idation by ground-based sunphotometry and sky radiometry.Furthermore, a scatter plot between and is useful foraerosol characterization as shown by ground-based radiationmeasurements (e.g., [19] and [20]) and from recent AVHRRretrievals by a two-channel algorithm [6], [15], [21]. Thecombination of channels 6 and 8 of OCTS is similar to that ofchannels 1 and 2 of AVHRR in terms of wavelength ratio, i.e.,

m and m versusm and m , but better

in terms of spectral purity of radiances without significantgaseous absorption in the bands. Especially the insignificanteffect of water vapor absorption greatly improves the reliabilityof the retrieval with OCTS radiances.

As for the aerosol model, we assume a bimodal log-normalsize distribution with prescribed mode radii and dispersion

in order to constrain the two-channel retrieval,

(16)

In our analysis, m, m,and which have been determined so as to re-alize the best fit of AVHRR-derivedAngstrom parametersto ground-based values [6]. The complex refractive index ofaerosols is assumed to be 1.5-0.005for both channels. Withprescribed mode radii and dispersions, we are left with twoundetermined coefficients to be determined from twospectral radiances in channels 6 and 8. Since there are uniquerelationships between and and between and ,we can directly search using look-up tables by aniterative procedure. In the present study, we adoptdefined by a log-linear regression fit to in (16), which


Fig. 8. Three monthly composites for retrieval ofAngstrom parameters (April, May, and June 1997).

can be calculated with given at five sunphotometerwavelengths , and nm. We takethis indirect definition of theAngstrom parameters, not definedat satellite wavelengths, in order to find values effective withina broad spectral range from 400–1000 nm.

Fig. 8 depicts three monthly composites ofAngstrom pa-rameters from Level-1B GAC data (April, May, and June1997). The OCTS spatial resolution of 700 m 700 mis advantageous for cloud screening compared with AVHRR(1 km 4 km for GAC) and POLDER (6 km 7 km). Wesegmented the globe into 0.5 0.5 latitude–longitude boxesand applied the following selection rule of a cloud-free pixelfor the analysis:

1) discard the segment if it contains more than 80% of thetotal pixels with ;

2) discard the segment if the lowest 15% reflectances are;

3) discard the segment if the root-mean square deviation ofthird through fifteenth lowest reflectances exceeds 0.02;

4) select the third lowest reflectance in channel 6.

Angular condition for the analysis is limited to andA gap around the equator in the figure is the area

where no data were taken during mirror tilting of OCTS toavoid sun-glitter reflection.

General features of theAngstrom parameter distributionsare similar to those of AVHRR-retrieved distributions reportedby Higurashi and Nakajima [6] and Nakajima and Higurashi[21]. The Saharan dust layer was abnormally small in 1997as compared with other years [4], [5]. On the other hand,aerosol activities look very active in the central Americaand the Indo-China regions. The central American aerosollayer is characterized by largeAngstrom exponents. Since theAngstrom exponent increases with decreasing volume peakratio large Angstrom exponents suggest an existenceof small particles in this region, and, therefore, suggest thatgas-to-particle conversion was vigorous by forest fires activein this season [22]. It is important to note that outflowof aerosols from China to the Pacific Ocean is character-ized also by largeAngstrom exponents showing an existenceof small particles. As studied by many investigators (e.g.,[23]–[26]), the April–May season has frequent outflow eventsof Chinese mineral dust particles carried by travelling lowpressure systems to the areas of Japan and adjacent PacificOcean. One expects, therefore, that coarse particles shouldbe dominant in the region during that season, contrary to thepresent satellite observation. Although validation experimentsare needed to understand this contradiction, it is useful to knowthat there is evidence of long-range transport of anthropogenic


Fig. 9. Comparison between satellite-retrieved and ground-based measurements ofAngstrom parameters.

aerosols from the continent [27], [28]. This suggests thatsulfate aerosols are also transported by migrating low-pressuresystems from East Asia into the Pacific Ocean region.

It is interesting to note that the Indian Ocean region wasdominated by small particles in April, whereas large particleswere dominant in June. According to Hermanet al. [29], UV-absorbing aerosols were dominant in the April–August seasonof 1988 in this region, but the distribution of aerosols wasdifferent in April–May from that in June–August. In the formerperiod, the entire Indian region had a high concentrationof UV-absorbing aerosols, whereas semiarid areas of thePakistan-Middle East region was dominated by UV-absorbingaerosols in the latter period. Those two satellite observationssuggest that the April–May period in this region had a mixtureof large mineral dust particles and small sulfate and/or biomassburning aerosols, whereas the June–August period had largemineral aerosols. LargeAngstrom exponents around Indonesiaand Australia in May–June can be attributed to biomassburning aerosols triggered by early onset of the large El Ninoevent in 1997.

Validation of the present satellite retrievals is important inorder to support the interesting findings of aerosol distributionsdiscussed in this section. As part of NASDA’s OCTS project,we set PREDE skyradiometers at a coast site of Niigatacity (37.92N, 139.00E) and Hedo Point, Okinawa (26.86N,128.25E). Fig. 9 compares the satellite-retrievedAngstromparameters and those measured by sky radiometers. Satellitedata are averages over 150 km150 km areas and ground-based data are daily averages. Although scatter is large,there is good correspondence between satellite-retrieved andground-based values ofAngstrom parameters, especially forthe Angstrom exponent. In this regard, it is interesting tocompare the result with those obtained with the POLDERradiometer [30]. They adopted 670 and 865 nm radiances ofPOLDER for retrievingAngstrom parameters, which are verysimilar to ours. Twelve aerosol models with mono-modal log-normal size distributions were prepared with different mode

radii to cover a range of theAngstrom exponent, insteadof our tuning of two volume peaks of the bimodal log-normal size distribution. The real part of the aerosol complexrefractive index had values of 1.33, 1.40, or 1.50 dependingon the aerosol model, and the imaginary part was fixedat zero, whereas we assumed one refractive index of 1.50-0.005 Angstrom parameters for validation were evaluatedat the wavelengths of satellite channels, whereas we usedan averaged value at regular sunphotometer wavelengths.Despite of several small algorithm differences, both algorithmsproduced a good correlation ofAngstrom parameters withground-based validation values, indicating the ability of satel-lite remote sensing to retrieve realistic values ofAngstromparameters. In this regard, it should be noted that there is avery good correlation of the POLDER results with ground-basedAngstrom exponents. Optical thickness retrievals withPOLDER also showed a good correlation: better than ourresults. It is interesting to note, however, that POLDER resultsshowed a large variety of the optical thickness at 865 nmlarger than 0.3, which are almost all our cases shown in Fig. 9.This indicates that aerosol layers with large optical thicknessesare spatially and/or temporally highly inhomogeneous and arenot so suitable for validation, and, hence, difference in thedegree of correlation between the two studies may be attributedto difference in the degree of turbidity at validation sites.Unfortunately, we do not have such remote sites with lessturbidity for our study. Another important difference betweenthe two studies is selection of ground data, i.e., POLDERvalidation was based on data collected within30 min ofADEOS overflight time. Spatial average of 100 km100 kmis also better than that of 150 km 150 km in the presentstudy.

V. DISCUSSION AND CONCLUSIONS

It is clear from the preceding sections that the present OCTSdata analysis is encouraging and useful for depicting the globaldistributions of aerosol optical properties. In spite of uncer-


Fig. 10. Latitudinal distribution of� and �500 derived from OCTS andAVHRR.

tainties in the calibration of OCTS radiances, we successfullyretrieved global distributions ofAngstrom parameters whichindicate realistically the aerosol characteristics on a globalscale. Such confidence comes from validation with ground-based sky radiometers as shown in Fig 9 and also indirectlyby examination of POLDER results. Although there are similarproducts ofAngstrom parameters derived from AVHRR two-channel radiances, such remote sensing is difficult due tostrong water vapor contamination in channel 2 of AVHRR.Therefore, it is interesting to find similar features in Fig. 8in terms of spatial distributions ofAngstrom parameters. Itshould be recognized that those two channels used in this studyhave practically no contamination by water vapor absorption.Therefore, it can be concluded that the present retrievals sup-port AVHRR two-channel aerosol characterization as well asusefulness of ocean color sensors for aerosol characterization.

Fig. 10 showsAngstrom parameters zonally averaged overocean in April of 1990 derived from AVHRR and of 1997derived from OCTS. Although some of the differences be-tweenAngstrom parameter distributions in 1990 and 1997 canbe attributed to technical differences in AVHRR and OCTSalgorithms, we can find several interesting differences whichlook realistic. First of all, it is found that the northern hemi-sphere has larger optical thickness andAngstrom exponentthan those in the southern hemisphere by about 0.08 and0.3, respectively both in 1990 and 1997, indicating significanteffects of human activities and biomass burning, consistentwith previous reports [5], [22], [29]. Zonal distributions ofthe turbidity factor in 1990 and 1997 are very similarto each other showing two peaks at around 7N and 40 Nin the northern hemisphere, other than the broad peak around

20 S. Although not shown in a figure, close investigation ofthe derived global distributions indicates that biomass burningaerosols seemed to be prevailing around South Africa andWestern Australia in April 1990, producing large AVHRRAngstrom exponent values around 20S, as shown in Fig. 10.The turbidity peaks around 40N is associated with largeAngstrom exponent values, so that the peak can be reasonablyattributed to industrial activities and biomass burning in popu-lated areas in the northern hemisphere. On the other hand, theaerosol composition responsible to the turbidity peaks around7 N is rather complicated. In April 1997, the northern halfof the peak consisted of small particles and the southern halfwith large particles as compared with the situation in April1990. Such a characteristic distribution of small and largeparticles is possible by year-to-year concentration variationof mineral dust aerosols and biomass burning aerosols. InApril 1997, thick small particle layers were observed byOCTS around the Philippine, Indian, and Central Americanregions, while Saharan dust layer was weak in this year.Moulin et al. [28] showed a large intra- and inter-annualvariation of the mean Saharan aerosol optical thickness inthe Mediterranean–Atlantic region correlating with the NorthAtlantic Oscillation (NOA) index. Saharan aerosol opticalthickness can fluctuate by more than 50% for the same monthin the year, because there are significant differences in theamplitude and phase of general circulation in the region. Astudy of NOA index of 1997 has indicated a weak Saharandust storm development in 1997. It may be concluded that themagnitude of radiative effect is of the same order betweenmineral dust aerosols and biomass burning aerosols in thetropical region. This conclusion is important for energy budgetstudies of the earth-atmosphere system, since it can be changedsignificantly depending on the seasonal and annual variationof the emission strength of the two types of aerosols.

There are several important tasks we have to completebefore obtaining final aerosol products from OCTS radiancedata. Further validation of the products is necessary, since thequality of the products is very sensitive to the calibration ofOCTS spectral radiances. Comparison among products fromOCTS, AVHRR, and POLDER will be especially important,since characterization of the sensors and cloud screeningprocedures are different for those products. Calibration ofOCTS radiances should be improved with more data analysisand validation data. It is also important to produce Level-1GAC data sets for IR radiances, with which we can developimproved cloud screening procedures. Rejection of thin cirrusclouds and invisible small clouds will be improved withIR radiances. In conclusion, our experience with AVHRRand OCTS suggests that simultaneous use of visible-NIR-IRspectral radiances has an advantage for aerosol remote sensingover traditional use of visible-NIR radiances in ocean colorremote sensing, since aerosol remote sensing needs absolutevalues of radiances rather than relative color information usedin ocean color remote sensing. In this respect, it is satisfyingto know that future satellite-borne radiometers, such as EOS-AM1/MODIS, ENVISAT-1/MERIS, and ADEOS-2/ GLI, willprovide global observations of high-quality spectral radiancesfor use in aerosol remote sensing (e.g., [32] and [33]).


ACKNOWLEDGMENT

The authors are grateful to Prof. S. Ohta of HokkaidoUniversity, Y. Kinjo of the Okinawa Environmental andSanitary Research Institute, and H. Nagata and T. Watanabe ofJapan Sea National Fisheries Research Institute for supportingthe sky radiometer measurements in Okinawa and Niigata.They want to extend their gratitude to the Captain, crew,and research staff of R/V Roger Revelle for supporting thesky radiometer measurements on the research vessel; and Dr.P.-Y. Deschamps and Dr. B. Fougnie of the Universite desSciences et Technologies de Lille, France, for making availablethe SIMBAD data. Profs. H. Shimoda, H. Kawamura, and Mr.M. Shimada, and K. Okuzumi of the Earth Observing ResearchCenter of the National Space Development Agency of Japan(NASDA) are gratefully acknowledged for preparing OCTSL1B-GAC data sets.

REFERENCES

[1] H. R. Gordon and D. K. Clark, “Clear water radiances for atmosphericcorrection of coastal zone color scanner imagery,”Appl. Opt., vol. 24,pp. 4175–4180, 1981.

[2] H. R. Gordon and M. Wang, “Retrieval of water-leaving radiance andaerosol optical thickness over the oceans with SeaWiFS: A preliminaryalgorithm,” Appl. Opt., vol. 33, pp. 443–452, 1994.

[3] L. L. Stowe, R. M. Carey, and P. P. Pellegrino, “Monitoring the Mt.Pinatubo aerosol layer with NOAA/11 AVHRR data,”Geophys. Res.Lett., vol. 19, pp. 159–162, 1992.

[4] C. Molin, F. Dulac, C. E. Lambert, P. Chazette, I. Jankowiak, B.Chatenet, and F. Lavenu, “Long-term daily monitoring of Saharandust load over ocean using Meteosat ISCCP-B2 data 2. Accuracy ofthe method and validation using sun photometer measurements,”J.Geophys. Res., vol. 102, pp. 16 959–16 969, 1997.

[5] R. B. Husar, J. M. Prospero, and L. L. Stowe, “Characterization oftropospheric aerosols over the oceans with the NOAA advanced veryhigh resolution radiometer optical thickness operational product,”J.Geophys. Res., vol. 102, pp. 16 889–16 910, 1997.

[6] A. Higurashi and T. Nakajima, “Development of a two channel aerosolretrieval algorithm on global scale using NOAA/AVHRR,”J. Atmos.Sci., to be published.

[7] O. Hagolle, P. Goloub, P. Y. Deschamps, H. Cosnefroy, X. Briottet, T.Bailleul, J. M. Nicolas, F. Parol, B. Lafrance, and M. Herman, “Resultsof POLDER in-flight calibration,”IEEE Trans. Geosci. Remote Sensing,to be published.

[8] B. Fougnie, P. Y. Deschamps, and R. Frouin, “Vicarious calibration ofthe POLDER ocean color spectral bands usingin situ measurements,”IEEE Trans. Geosci. Remote Sensing, to be published.

[9] NASDA/EOC (1997): Advanced earth observing satellite(ADEOS): OCTS data processing algorithm description, Ver.2.0.1, NASDA Earth Observation Center. [Online.] Availablehttp://www.eorc.nasda.go.jp/ADEOS/UsersGuide/.

[10] B. G. Mitchell and M. Kahru, “Algorithms for SeaWiFS standardproducts developed with the CalCOFI data set,”CalCOFI Reports, vol.39, pp. 133–147, 1988.

[11] T. Nakajima, G. Tonna, R. Rao, Y. Kaufman, and B. Holben, “Use of skybrightness measurements from ground for remote sensing of particulatepolydispersions,”Appl. Opt., vol. 35, pp. 2672–2686, 1996.

[12] T. Nakajima, M. Tanaka, and T. Yamauchi, “Retrieval of the opticalproperties of aerosols from aureole and extinction data,”Appl. Opt.,vol. 22, pp. 2951–2959, 1983.

[13] T. Nakajima and M. Tanaka, “Effect of wind-generated waves on thetransfer of solar radiation in the atmosphere-ocean system,”J. Quant.Spectrosc. Radiat. Transfer, vol. 29, pp. 521–537, 1983.

[14] H. Fukushima, private communication for their determination of thecorrection factor, 1998.

[15] T. Nakajima and A. Higurashi, “AVHRR remote sensing of aerosoloptical properties in the Persian Gulf region, the summer 1991,”J.Geophys. Res., vol. 102, pp. 16 935–16 946, 1997.

[16] H. Fukushima, A. Higurashi, Y. Mitomi, T. Nakajima, T. Noguchi,T. Tanaka, and M. Toratani, “Correction of atmospheric effect ofADEOS/OCTS ocean color data: Algorithm description and evaluationof its performance,”J. Oceanogr., vol. 54, pp. 417–430.

[17] D. K. Clark, H. R. Gordon, K. J. Voss, Y. Ge, W. Broenkow, andC. Trees, “Validation of atmospheric correction over the oceans,”J.Geophys. Res., vol. 102, pp. 17 209–17 217, 1997.

[18] M. Kishino, J. Ishizaka, S. Saitoh, Y. Senga, and M. Utashima, “Ver-ification plan of ocean color and temperature scanner atmosphericcorrection and phytoplankton pigment by moored optical buoy system,”J. Geophys. Res., vol. 102, pp. 17 197–17 208, 1997.

[19] T. Nakajima, M. Tanaka, M. Yamano, M. Shiobara, K. Arao, and Y.Nakanishi, “Aerosol optical characteristics in the yellow sand eventsobserved in May, 1982 in Nagasaki—Part II model,”J. Meteorol. Soc.Jpn., vol. 67, pp. 279–291, 1989.

[20] Y. J. Kaufman, A. Gitelson, A. Karnieli, E. Ganor, R. S. Fraser, T.Nakajima, S. Mattoo, and B. N. Holben, “Size distribution and scat-tering phase function of aerosol particles retrieved from sky brightnessmeasurements,”J. Geophys. Res., vol. 99, pp. 10 341–10 356, 1994.

[21] T. Nakajima and A. Higurashi, “A use of two-channel radiances for anaerosol characterization from space,”Geophys. Res. Lett., vol. 25, pp.3815–3818.

[22] M. O. Andreae, “Global distribution of fires seen from space,”EOSTrans., vol. 74, pp. 129–144, 1993.

[23] K. Arao, Y. Makino, and Y. Nagaki, “Some statistical aspects of yellowsand,” Sci. Bull. Fac. Edu., Nagasaki Univ., vol. 30, pp. 65–74, 1979(in Japanese).

[24] K. Arao and Y. Ishizuka, “Volume and mass of yellow sand dust inthe air over Japan as estimated from atmospheric turbidity,”J. Metorol.Soc. Jpn., vol. 64, pp. 79–94, 1986.

[25] K. Kai, Y. Okada, O. Uchino, I. Tabata, H. Nakamura, T. Takasugi,and Y. Nikaidou, “Lidar observation and numerical simulation of aKosa (Asian Dust) over Tsukuba, Japan during the spring of 1986,”J. Meteorol. Soc. Jpn., vol. 66, pp. 457–472, 1988.

[26] J. T. Merrill, M. Uematsu, and R. Bleck, “Meteorological analysis oflong range transport of mineral aerosols over the North Pacific,”J.Geophys. Res., vol. 94, pp. 8584–8598, 1989.

[27] M. Uematsu, T. Sugita, V. V. Anikiev, and A. N. Medvedev, “Large-scale transport of pollution aerosol over the east coast of Asia,”Geophys.Res. Lett., vol. 19, pp. 2219–2221, 1992.

[28] N. Niimura, K. Okada, X.-B. Fan, K. Kai, K. Arao, G.-Y. Shi, and S.Takahashi, “Formation of Asian dust-storm particles mixed internallywith sea salt in the atmosphere,”J. Meteorol. Soc. Jpn., vol. 76, pp.275–288, 1998.

[29] J. R. Herman, P. K. Bhartia, O. Torres, C. Hsu, C. Seftor, and E. Celarier,“Global distributions of UV-absorbing aerosols from Nimbus 7/TOMSdata,” J. Geophys. Res., vol. 102, pp. 16 911–16 923, 1997.

[30] P. Goloub, D. Tanre, J. L. Deuze, M. Herman, A. Marchand, and F. M.Breon, “Validation of the first algorithm applied for deriving the aerosolproperties over the ocean using the POLDER/ADEOS measurements,”this issue, pp. 1586–1596.

[31] C. Moulin, C. E. Lambert, F. Dulac, and U. Dayan, “Control ofatmospheric export of dust from North Africa by the North Atlanticoscillation,” Nature, vol. 387, pp. 691–694, 1987.

[32] Y. J. Kaufman, D. Tanre, L. A. Remer, E. F. Vermote, A. Chu, and B. N.Holben, “Operational remote sensing of tropospheric aerosol over landfrom EOS moderate resolution imaging spectroradiometer,”J. Geophys.Res., vol. 102, pp. 17 051–17 068, 1997.

[33] D. Tanre, Y. J. Kaufman, M. Herman, and S. Mattoo, “Remote sensingof aerosol properties over oceans using the MODIS/EOS spectralradiances,”J. Geophys. Res., vol. 102, pp. 16 971–16 988, 1997.

Teruyuki Nakajima was born August 13, 1950.He received the B.A. degree in physics and theM.S. and Sci.D. degrees in geophysics from TohokuUniversity, Sendai, Japan, in 1973, 1975, and 1981,respectively.

From 1977 to 1991, he worked at Tohoku Univer-sity, first as a Teaching Assistant at the GeophysicalInstitute, then as an Assistant Professor in the UpperAtmosphere and Space Research Laboratory, andan Associate Professor at the Research Center forAtmospheric and Oceanic Variation. In 1991, he was

an Associate Professor at the Geophysical Institute, University of Tokyo, and,since 1991, he has been an Associate Professor and now Professor at theCenter for Climate System Research, University of Tokyo.


Akiko Higurashi was born on April 12, 1969. Shereceived the B.A. degree in physics from the ScienceUniversity of Tokyo, Tokyo, Japan in 1993, theM.S. degree in geophysics from Tohoku University,Sendai, Japan, in 1995 and the Sci.D. degree in earthand planetary physics from the University of Tokyoin 1998.

Since 1998, she has been a Research Scientistin the Atmospheric Environment Division, NationalInstitute for Environmental Studies, Ibaraki, Japan.

Kazuma Aoki was born July 23, 1969. He receivedthe B.S. degree from the Science University ofTokyo, Tokyo, Japan, in 1994, and the M.S. de-gree in environmental earth science from HokkaidoUniversity, Sapporo, Japan, in 1997. He is currentlya graduate student in the Division of Oceanic andAtmospheric Sciences, Graduate School of Environ-mental Earth Science, Hokkaido University.

Tatsuo Endoh graduated in 1962 from HokkaidoEducational University, Sapporo, Japan. Hereceived the M.S. and Sci.D. degrees from theScience School of Hokkaido University in 1964and 1976, respectively.

From 1968 to 1975, he was a Research Assistantand from 1975 to 1981 a Lecturer in the Faculty ofScience, Hokkaido University. Since 1981, he hasbeen an Associate Professor in Low TemperatureScience at Hokkaido University.

Hajime Fukushima (M’83) received the Ph.D. de-gree in computing science from Tohoku University,Japan, in 1978.

From 1978 to 1983, he was a Research Associateat the School of Information Engineering, TohokuUniversity. From 1983 to 1991, he was an AssociateProfessor at the School of Marine Science and Tech-nology, Tokai University, Numazu, Japan. Since1992, he has been a Professor at the Department ofInformation and Communication Technology, TokaiUniversity.

Mitsuhiro Toratani received the B.E., M.E., andDr.E. degrees in marine science and technologyfrom Tokai University, Shizuoka, Japan, in 1986,1989, and 1994.

He has been with the Department of Informationand Communication Technology, Tokai University,Shizuoka, Japan, since 1992, where he is an Asso-ciate Professor. He is engaged in the study of oceancolor satellite remote sensing. He is working ona study of atmospheric correction, KOSA (yellowsand) transportation, and the ocean phenomena in

East and South China Sea using ocean color satellite (OCTS, SeaWiFS) data.

Yasushi Mitomi received the B.E. and M.E. degreesin ocean engineering in 1991 and 1993 from TokaiUniversity, Shizuoka, Japan.

Since 1993, he has been working at the RemoteSensing Technology Center of Japan, Tokyo. Hiscurrent research theme is the development of the at-mospheric correction method for ocean color remotesensing.

B. Greg Mitchell was born in 1954 in Houston, TX. He received the B.S.degree in aquatic biology (special honors in botany) from the Universityof Texas, Austin, in 1977, then studied biological oceanography at theUniversity of Southern California, Los Angeles, and received the Ph.D.degree in 1987. His dissertation research was concerned with developingmethods to characterize the absorption coefficient of phytoplankton and marineparticulates and applications to ocean optics and photosynthesis research.

He has continued his interest in ocean photosynthesis and optics at theScripps Institution of Oceanography, University of California, San Diego,La Jolla, where he is presently an Associate Research Biologist. From 1990to 1992, he took a leave of absence from his university appointment tomanage the NASA Headquarters Ocean Biogeochemistry Program and serveas the Program Scientist for SeaWiFS. While at NASA, he coordinatedthe establishment of the SeaWiFS Project at Goddard Space Flight Center,Greenbelt, MD, and the selection of the SeaWiFS Science Team. He serveson NASA’s SeaWiFS and SIMBIOS Science Teams and NASDA’s GLIAlgorithm Development Team.

Robert Frouin , for a photograph and biography, see this issue, p. 1574.

Early phase analysis of OCTS radiance data for aerosol remote sensing

Documents