Vicarious Calibration of GLI by Ground Observation Data - IEEE Xplore

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 43, NO. 10, OCTOBER 2005

2167

Vicarious Calibration of GLI by Ground Observation Data Mayumi Yoshida, Hiroshi Murakami, Yasushi Mitomi, Masahiro Hori, Kurtis J. Thome, Dennis K. Clark, and Hajime Fukushima

Abstract—We conducted vicarious calibration of the Global Imager (GLI) in visible to near-infrared channels over different targets. For calibration over the ocean, we used the normalized water-leaving radiance derived from the Marine Optical Buoy (MOBY) and the aerosol optical properties (aerosol optical depth, size distribution, and refractive index) obtained through the Aerosol Robotic Network (AERONET). For calibration over land, we used the ground-based measurement data at Railroad Valley Playa. The following GLI characteristics are recognized from the calibration results. First, GLI underestimates the radiance in channels 1, 2, 4, and 5. Next, in the near-infrared channels, there is good agreement between the observed and simulated radiance over bright targets. On the other hand, it is suggested that the GLI overestimates the radiance over dark targets (e.g., on the order of 15% at 4.0 W/m2 / m/sr in channels 18 and 19). Furthermore, we evaluated these calibration results over different targets taking into account the difference in the target radiance and in the accuracy between the two results. This combined evaluation of vicarious calibration results suggests the possibility that the GLI-observed radiance has offset radiance versus the simulated radiance. Index Terms—Global Imager (GLI), Marine Optical Buoy (MOBY), radiative transfer, vicarious calibration.

I. INTRODUCTION

T

HE Global Imager (GLI) was launched onboard the Advanced Earth Observing Satellite II (ADEOS-II) on December 14, 2002. GLI has 36 spectral channels from visible to infrared wavelengths for monitoring and understanding global environmental changes. It was designed to observe the atmosphere, ocean, land, and cryosphere. Table I presents the general GLI channel characteristics, though the detailed specification is described in [1]. Postlaunch radiometric calibration is necessary to investigate any change in the optics during launch and to monitor the degradation of the optics. Onboard calibrations, such as lamp

Manuscript received November 18, 2004; revised March 18, 2005. M. Yoshida and Y. Mitomi are with the Remote Sensing Technology Center of Japan, Tokyo 104-6021, Japan (e-mail: [email protected]; [email protected]). H. Murakami and M. Hori are with the Earth Observation Research and Application Center, Japan Aerospace Exploration Agency, Tokyo 104-6023, Japan (e-mail: [email protected]; [email protected]). K. J. Thome is with the University of Arizona, Tucson, AZ 85721 USA (e-mail: [email protected]). D. Clark is with the National Oceanic and Atmospheric Administration, Washington, DC 20233 USA (e-mail: [email protected]. gov). H. Fukushima is with School of High-Technology for Human Welfare, Tokai University, Shizuoka 410-0395, Japan (e-mail: [email protected]). Digital Object Identifier 10.1109/TGRS.2005.856113

and solar calibrations, attempt to evaluate the change [2]. However, it has become obvious that these techniques alone are inadequate to provide good ocean-color measurement [3]. Therefore, vicarious calibration is important to evaluate the accuracy of the calibration. Here, in the vicarious calibration method, the satellite-observed radiance is compared with the radiance simulated by a radiative transfer model using ground-truth observation data. Many previous studies have been conducted on vicarious calibration over the ocean and land. Gordon [4] provided one of the major schemes of vicarious calibration for ocean-color sensors. In his scheme, the gain for the longest wave spectral channel in the near-infrared (NIR) band is fixed at the prelaunch value, and the shorter wave channels are then calibrated with respect to this channel. A similar vicarious calibration procedure has been used for the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) by Eplee et al. [5]. In those procedures, the longest wave spectral channel in the NIR is assumed to be perfectly calibrated, and therefore, retains its postlaunch calibration error. Wang and Gordon [6] demonstrated that as long as the calibration error in the longest NIR channel (865 nm) was less than 10%, the postlaunch radiances corrected by vicarious calibration are sufficiently accurate to retrieve useful water-leaving radiance at moderate aerosol optical depths. This assumption is rational for ocean-color sensors like SeaWiFS, because reducing the assumed error in the longest NIR channel yielded only a slight improvement in water-leaving radiance, in spite of its difficulty. However, GLI is used for observing not only the ocean, but also the atmosphere, land, and cryosphere. It is, therefore, important to evaluate the accuracy in all channels, including NIR channels. Bruegge et al. [7] validated that radiometric accuracy of the Multi-angle Imaging Spectroradiometer is maintained over dark ocean targets in all channels, including NIR channels. For vicarious calibration over land, Slater et al. [8] proposed the reflectance-based method for small-footprint sensors such as the Landsat Thematic Mapper (TM). This method relies on ground-based surface measurements of a selected target at the time of sensor overpass. Biggar et al. [9] evaluated the rootsum-of-squares total error for the method to be 4.9%. Thome et al. [10] applied the same method to a 1-km footprint sensor. As introduced above, there have been many studies about vicarious calibration for specific targets. However, fewer vicarious calibration studies have been over both ocean and land targets. The observed radiance differs from one target to the other. We believe the combined calibration results for differing radiance provides helpful information for characterizing satellite sensors. In addition, each result has its own uncertainties with respect to

0196-2892/$20.00 © 2005 IEEE

2168


TABLE I GLI CHANNEL CHARACTERISTICS

channels. The calibration error decreases with decreasing wavelength over the ocean, in inverse proportion to the increase in Rayleigh scattering. However, the error magnitude has little dependence on the wavelength over land. Therefore, we expect that one result with less uncertainty can help to evaluate the other result. This study attempts to understand the characteristics of GLI in the visible to near-infrared channels, by evaluating the combination of the vicarious calibration results of different targets having different observed radiance. II. VICARIOUS CALIBRATION OVER OCEAN A. Methodology Following Gordon [11], the total radiance observed by the can be expressed as satellite sensor over the ocean

(1) Here, is the radiance resulting from multiple scattering by air molecules in the absence of aerosols, is the radiance resulting from multiple scattering by aerosols in the abis the interaction term between molecsence of the air, is the radiance of direct sunular and aerosol scattering, light reflected by the sea surface (sun glitter), is the radiis the water-leaving radiance. ance due to whitecaps, and is the direct transmittance of the atmosIn this equation, is the diffuse transmittance. phere, and To derive a vicarious calibration coefficient, which is correction factor of the applied calibration coefficient, the total radiance needs to be calculated by (1) and compared with the obcan be discarded when served value from GLI. First,

we select the data whose geometries do not allow for the efby limiting data with low surfect. We can also ignore can be theoretically estimated face-wind speed. Second, by the Rayleigh scattering theory, with ancillary data on ozone and and surface pressure. Finally, remain to be determined. can be derived from in situ measurement in the ocean. We can also simulate aerosol-related by a radiative transfer code, if we know terms including the aerosol optical properties in the area. In our calibrations, we used the aerosol optical properties derived from simultaneous in situ atmospheric measurements. For the simulation, we used the radiative transfer code RSTAR [12]–[14], developed at the University of Tokyo. We included the ozone absorption effect in the simulation, but did not include the absorption effect by oxygen and water vapor. B. Data For a quality check on the satellite data, we used the same exclusion criteria as that used for SeaWiFS [15], [16], except the wind speed threshold. We decided to accept all data with wind speed less than 7 m/s. The seven calibration points that passed the quality check are summarized in Table II, along with the surface wind speed and the aerosol optical depth. We used version 2 GLI Level-1B data to which prelaunch calibration and stripe noise correction had been applied. GLI employs 12 (1-km resolution channels) or 48 (250-m resolution channels) detectors, and gets image data by two-sided (sides A and B) scanning mirrors. Errors in normalizing the detector sensitivities and the mirror reflectances cause stripe pattern noise in the images [2], [17]. The version 2 GLI Level-1B processing corrects the noise based on detector 6 and scanning mirror side

YOSHIDA et al.: VICARIOUS CALIBRATION OF GLI BY GROUND OBSERVATION DATA

2169

TABLE II SEVEN DAYS SELECTED FOR THE VICARIOUS CALIBRATION: DATE, SURFACE WIND SPEED, V (m/s), AND AEROSOL OPTICAL DEPTH MEASURED AT 873 nm, 873

Fig. 2. Vicarious calibration coefficient for each channel except for channel 17. The dotted lines show the coefficients for the seven days in Table II. The solid line shows (open circle) the mean and (error bar) the standard deviation of the seven coefficients.

Fig. 1. (Solid lines) Normalized water-leaving radiances for GLI channels acquired by MOBY and (dotted lines) simulated total radiances on April 8 and September 22, 2003.

B. Consequently, the derived vicarious calibration coefficients are those of detector 6 and mirror side B. derived from the Marine Optical Buoy We used (MOBY) observation off the Hawaiian Islands. MOBY [18] is the primary calibration site for SeaWiFS and has been continuously collecting upwelling radiance data at fine spectral resolution over the whole visible spectrum since 1996. The measured radiances were convolved with the GLI spectral response functions [19]. Fig. 1 illustrates examples of the normalized water-leaving radiances for the GLI channels, along with the simulated total radiances for comparison. We utilized the aerosol optical properties derived from sun and sky-scanning spectral radiometer measurements at Lanai Island. The data are obtained through the aerosol robotic network (AERONET) [20], and it contains the aerosol optical depth, size distribution, and refractive index. We also applied the total ozone concentration derived from the Television Infrared Observation Satellite Program (TIROS) Operational Vertical Sounder and the sea-surface pressure from Japan Meteorological Agency objective analysis data. C. Results We simulated the top of atmosphere (TOA) radiance using the input data described in Section II-B. Fig. 2 presents the ratio of the TOA radiance simulated using ground-truth data to the radiance measured by GLI for each day and each channel, except for channel 17, which suffers from absorption of radiance by oxygen. The observed radiance in channels 14 and 15 would

also be slightly influenced by water-vapor absorption, and thus the ratios in these channels tend to show greater value. Although large variability is observed at longer wavelengths, we can generally see the following characteristics. First, the GLI-observed radiance in channels 1, 2, 4, and 5 seems to be too low. Second, GLI overestimates the radiance at longer wavelengths. This is on the order of 15% in channels 18 and 19. The relative channel gain characteristics agree well with that of Murakami et al. [21], although the absolute values should not be compared because Murakami et al. derived their calibration coefficients relative to two GLI channels. Our next step was to estimate the systematic errors of the derived vicarious calibration coefficients. The main contributions to TOA radiance are from molecular scattering, aerosol scattering, and water-leaving radiance. The molecular scattering radiance is precisely calculated by theory, even in consideration of multiple scattering and polarization [22]. Typically, in clear water, the contribution of water-leaving radiance to TOA radi% in the blue ( nm), 5% in the green ( ance is nm), and negligible in the near-infrared ( nm). The 5% error of water-leaving radiance causes only a maximal error of 0.5% on the TOA radiance. Therefore, the primary cause of uncertainties is the aerosol-scattering term, especially at longer wavelengths. For that reason, we evaluated the uncertainties of the vicarious calibration coefficient caused by the inaccuracy of the aerosol optical properties. Dubovik et al. [23] assessed the accuracy of aerosol optical properties (size distribution, complex refractive index, and single-scattering albedo) retrieved from radiances measured by sun–sky scanning radiometers of AERONET. They simulated the retrieval sensitivity for three typical aerosols (water-soluble, dust, and biomass burning) to the instrumental offset (i.e., systematic) errors. Using this result, we estimated the uncertainties of the vicarious calibration coefficient. In our case, the aerosol optical depth at 440 nm was less than 0.2, and the atmospheric conditions were clear. Therefore, we used their results for water-soluble aerosol and aerosol optical depth less than 0.2. The estimated uncertainties of the vicarious

2170


Fig. 3. Estimated calibration coefficient on April 8 in the presence of the aerosol optical properties errors in size distribution (r), real (n), and imaginary (k) part of the refractive index. The number in parenthesis shows the degree of the error based on the result of Dubovik et al. [23].

calibration coefficient on April 8 appear in Fig. 3. We see from the figure that the errors of the real part of the refractive index mainly affect the calibration coefficient. It leads to a maximal relative error of 8% in channels 18 and 19. It is important to note that the error decreases with decreasing wavelength because the contribution of Rayleigh scattering to total radiance increases with decreasing wavelength. III. VICARIOUS CALIBRATION OVER LAND A. Methodology This method is based on the reflectance-based method provided by Slater et al. [8]. The simulation procedure is similar to that over the ocean, but there are some differences. Over land, and in (1). In there are obviously no effects of addition, the land surface is considered to be rather Lambertian, although the sea surface is considered to be specular. Therefore, can be simulated by a radiative transfer code when we know the surface reflectance, aerosol optical properties, and the amount of absorption gas. We included successive reflections and scattering between the surface and atmosphere in our simulation. We also included not only ozone but also water vapor absorption effect, because we obtained column water vapor by the ground measurements. The surface reflectance is derived from ground-based measurements of the selected target at the time of the satellite overpass. In Section II-A, we used the aerosol optical properties derived from in situ measurements directly. For this calibration, however, we have only the aerosol optical depth measurements. Therefore, we employed the optical depth measurements to derive the aerosol optical properties through aerosol models. In practice, we first prepared a set of candidate aerosol models for the target area. Next, we selected the aerosol model whose Angstrom coefficient corresponded with that of the measurements and used the aerosol properties of the model for the calculation. We can derive the vicarious calibration coefficient by comparing the simulated total radiance with the satellite-observed radiance.

Fig. 4. Spectral surface reflectance measured at Railroad Valley Playa on May 27, July 22, August 19, and August 23, 2003. The line entitled 2003/05/27_alt shows the measurement collected by an alternative spectrometer 3 h after GLI overpass.

B. Data For this calibration, we used the ground-based measurement data at Railroad Valley Playa observed by the Remote Sensing Group (RSG) at the University of Arizona. The Railroad Valley Playa is located in central Nevada, U.S., at an elevation of approximately 1.5 km. The center of the test site used for the calibration is located at 38.497 north and 115.691 west. The RSG has been using this site for radiometric calibration of large-footprint sensors such as the Moderate Resolution Imaging Spectrometer. Thome et al. [10] describe the site and the measurements in detail. The spectral measurements of surface reflectance were performed using a spectroradiometer. The retrieved surface reflectance on May 27, July 22, August 19, and August 23, 2003 are shown in Fig. 4. Regarding the measurement on July 22, RSG used the alternative spectrometer to collect the data because the regular spectrometer had to be sent back for maintenance. On May 27, in addition to the regular measurements, they also used the alternative spectrometer to collect the data 3 h after the GLI overpass (2003/05/27_alt in Fig. 4). The aerosol optical depth, column water vapor, and ozone optical depth are derived by the solar transmittance measurements, using the approach proposed in [24] and [25]. The derived aerosol optical depths are illustrated in Fig. 5. C. Results First, we selected the optimum aerosol model for the target area. Fig. 6 depicts the relative aerosol optical depth of the basic aerosol models (dust-like, water-soluble, oceanic, and soot models) and the measurements normalized at 500 nm. We see from the figure that most of the measured aerosol angstrom exponents are larger than those of the basic models. Consequently, the aerosol model in this area cannot be expressed by a mixture of the basic aerosol models. For this reason, we believed the aerosol properties in this area are basically expressed by the rural model (a composite of 70% of water-soluble and 30% of dust-like aerosol in volume) in RSTAR, which is the


2171

Fig. 7. Vicarious calibration coefficients for Railroad Valley Playa on May 27, July 22, August 19, and August 23, 2003. Fig. 5. Spectral aerosol optical depth derived by the solar transmittance measurement at Railroad Valley Playa on May 27, July 22, August 19, and August 23, 2003.

Fig. 8. Comparison of the mean vicarious calibration coefficients derived from (dotted line) Railroad Valley Playa and (solid line) MOBY measurements. Dashed line is the result from Barrow measurement in [26]. The error value for Railroad Valley Playa and Barrow data is based on [9] and [26], respectively.

IV. COMPARISON OF THE RESULTS Fig. 6. Normalized aerosol optical depth at 500 nm of (solid lines) the basic models and (dotted or dashed lines) the measurements at Railroad Valley Playa on May 27, July 22, August 19, and August 23, 2003.

basic aerosol model over land, since it can be considered that there is no effect of urban or ocean origin aerosols at Railroad Valley Playa. We then changed the volume size distribution so as to explain the measured aerosol angstrom exponent. The aerosol optical properties together with the data described in Section III-B were then used in a radiative transfer code to simulate the satellite-observed radiance. The simulated radiance was compared with the observed radiance from the GLI sensor to derive the vicarious calibration coefficient. Fig. 7 summarizes the derived coefficients for the GLI visible and near-infrared 1-km resolution channels, except the saturated channels. We also excluded the result of channel 17 because it suffers from the absorption of radiance by oxygen. There is good agreement between the observed and simulated radiances in channels 8, 13, and 15. The channels at shorter wavelengths (channels 1, 2, 4, and 5) seem to be too low in sensitivity, although large variability is observed depending on observation days.

To evaluate the coefficients and describe the characteristics of the GLI-observed radiance, we compared the vicarious calibration coefficients derived in Sections II-C and III-C (Fig. 8). We also compared our results with those of Nieke et al. [26] who obtained the GLI calibration coefficient by cross-calibration over snow fields at Barrow. Here, it should be noted that the satellite-observed radiance over the ocean (MOBY) is much lower than over land (Railroad Valley Playa and Barrow) at longer wavelengths (less than 10% in channels 13, 15, and 19) (Fig. 9). In contrast, the differences in the observed radiance between over the ocean and land at shorter wavelengths are relatively small compared to longer wavelengths. At shorter wavelengths, where the difference in target radiance is smaller, all results indicate that GLI underestimates the radiance in channels 1, 2, 4, and 5, although the calibration coefficients derived by Railroad Valley Playa measurements are higher than those derived by MOBY and Barrow measurements. At longer wavelengths, where the target radiance is significantly different, the two results over the bright targets (Railroad Valley Playa and Barrow) correspond with each other. However, the calibration coefficient over the dark target (MOBY) is much lower than over the bright targets (Railroad Valley Playa and Barrow). Further evaluation will be discussed in Section V.

2172


Fig. 9. Mean GLI-observed radiance over (dotted line) Railroad Valley Playa and (solid line) MOBY. Dashed line is the radiance over Barrow on April 14, 2003.

V. DISCUSSION In this section, we further evaluate the results by taking into account the difference in the target radiance and in the accuracy between the two results. We first discuss the coefficients at shorter wavelengths. In these bands, the coefficients should correspond with each other because the target radiance is not very different, but the calibration coefficients derived by Railroad Valley Playa measurements are higher than those by MOBY and Barrow measurements. When we take into consideration that the uncertainty of the result by MOBY measurement is small at these wavelengths, this comparison would suggest that the vicarious calibration coefficient derived by Railroad Valley Playa measurements is too high at shorter wavelengths, especially in channels 1 and 2. The possible causes of this discrepancy are the following. The first possible factor is a measurement error (mainly ground reflectance) over Railroad Valley Playa. This effect is estimated to be within 4.9% by Biggar et al. [9] as we mentioned in Section I. This factor may explain the larger discrepancy in channels 1 and 2 because the signal-to-noise ratio of the ground reflectance measurement is low due to the low solar irradiance and low surface reflectance in these channels. The second possible factor is misregistration of the ground-based measurement of reflectance to the GLI sensor data. The third possible factor is that vicarious calibration coefficients vary with the observing conditions, possibly by degree of contrast between the target and the surrounding radiance. These factors should be investigated in future studies to determine the cause of the discrepancy, possibly by comparing the vicarious calibration results with those of other sensors over Railroad Valley Playa. Next, we discuss the coefficients at longer wavelengths, where the target radiance differs greatly. The agreement between the two results over the bright targets (Railroad Valley Playa and Barrow) suggests the validity of the results at longer wavelengths. On the other hand, the coefficient over the dark targets is significantly lower than over the bright targets, even if we consider the greater calibration-coefficient errors over ocean (8% in channels 18 and 19) at these wavelengths.

We now consider the cause of the difference of the vicarious calibration coefficients concerning the target radiance. At first, we suspected the nonlinearity of the sensor is the cause, since the difference depends on thetarget radiance. This is unlikely because the observed radiances over MOBY and Railroad Valley Playa at longer wavelengths (e.g., less than 5 and 150 W/m /sr/ m in channel 19) are much lower than the maximum radiance for linear response by the prelaunch evaluation (e.g., 211 W/m /sr/ m in channel 19) [19]. This suggests the possibility that the observed radiance has an offset value because only the observed radiance over dark targets exceeds the simulated radiance. To investigate this possibility, we examined whether or not the derived calibration coefficients over the MOBY site depend on the simulated radiance. Fig. 10 illustrates the ratio of the observed radiance to the simulated radiance (i.e., inverse of the calibration coefficient) versus the simulated radiance in channels 3, 4, 5 (the examples at shorter wavelengths), 16, 18, and 19 (the examples at longer wavelengths). The ratio on September 19 (indicated by the arrow in the figure) obviously deviates from those on the other days even at shorter wavelengths (channels 3, 4, and 5). Since the observed radiance is high enough not to be influenced by the possible offset at shorter wavelengths, it is believed that the deviation for the day is caused by other factors. On the other days, the ratios increase with decreasing simulated radiance at longer wavelengths, although the ratios do not depend on the simulated radiance at shorter wavelengths. This supports the possibility that the observed radiance has the offset radiance versus the simulated radiance. Some might object, however, that the ratios depend on thesimulatedradiancebecausethesimulatedradiancedependson the aerosol radiance, which has primary uncertainties. A glance at the aerosol ratio (indicated by the shading in Fig. 10) reveals that the coefficients do not depend on the aerosol ratio. Thus, the possibility that the observed radiance has an offset value was suggested. We therefore estimated the offset using the results over dark and bright targets. This estimation was possible in channels 10–19, except channels 14, 15, and 17, which suffer from the absorption of radiance by water vapor and oxygen. In these channels, the observed radiance over bright targets is high enough not to be influenced by the offset, and the observed radiance over dark targets is low enough to be significantly influenced by the offset. In this condition, it is reasonable to suppose that the derived calibration coefficient over a bright target is the slope of the calibration curve. Using the slope, we can estimate the quantity of the offset by curve fitting of the simulated and observed radiance over a dark target. Based on the derived vicarious calibration coefficients over the bright targets (Railroad Valley Playa and Barrow) in channels 13, 15, and 19, and the relation between the saturation (10, 11, 12, 16, and 18) and nonsaturation (13, 15, and 19) channels over MOBY, we assumed the slopes in all channels from 10–19, except 14, 15, and 17, to be 1.0 for approximation. Fig. 11 illustrates examples of the observed radiance versus simulated radiance over the MOBY site and the estimated fitting curve. The fitting without the offset is also displayed for comparison. We see that the curve with the offset corresponds with each point better than the curve without the offset. Two factors are considered as possible cause of the offset. The first is the influence of surrounding land or clouds. We simulated the satellite radiance over the MOBY, assuming a uni-


2173

Fig. 10. Ratio of the observed radiance to the simulated radiance versus the simulated radiance in channels 3, 4, 5, 16, 18, and 19. Shading represents the ratio of the simulated aerosol radiance to the total simulated radiance.

Fig. 11. Scatter diagrams between the simulated and observed radiance over the MOBY site in channels (a) 10 and (b) 19 on all days except September 19 in Table II. Estimated fitting curve with (solid line) offset and (dashed line) without offset are also shown.

form oceanic surface. In practice land and clouds surround the site. The photons reflected by the environmental surface reach the sensor by scattering in the atmosphere, which contributes to the observed radiance. We evaluated the effect using the 6S radiative transfer code [27], which is convenient for simulating the environmental radiance. For the simulation, we considered a circular target of radius and reflectance surrounded by a homogeneous surface of reflectance . We set to 0 assuming an ocean target, and set to the standard vegetation reflectance assuming land environment and to 0.5 assuming cloud environ-

ment. We assumed the cloud height to be 5 km. The sensor geometries and aerosol optical depth were set to that of the actual measurements at MOBY on April 8, 2003. Figs. 12 and 13 illustrate the estimated land and cloud effects versus the target radius. We see from the figure that the influence of the environment decreases significantly with increasing radius. The mean distance from MOBY to environment land and clouds in the data used for the calibration exceeds 30 km. Assuming the worst situation (target radius of 30 km), the effect of land and clouds is less than 0.2 W/m / m/sr (land) and

2174


Fig. 12. (a) Estimated radiance by the environmental land and (b) the ratio of the environmental to the total radiance from channels 1 to 19, except channel 17. Horizontal axis shows the target radius.

Fig. 13.

Same as Fig. 12 but for clouds.

0.1 W/m / m/sr (clouds) at 865 nm, which is much lower than the estimated offset. The other possible factor is stray light. GLI calibration team suspects that stray light may influence the GLI sensor [2]. They suspect that the stray light influence on the side A mirror is more than on side B. However, our results suggest the stray light may also influence the side B mirror. We estimated the magnitude of the possible stray light in this case (Fig. 14), although the magnitude probably depends on the observing conditions, especially the surrounding radiance. The possible stray light is considered to be the difference between the total offset radiance and the environmental radiance. The estimated possible stray light in this case was about

0.4 W/m / m/sr in channel 19 and 1.2 W/m / m/sr in channel 10. Future study is needed to confirm these results and possibilities by comparison with other satellite sensors. VI. CONCLUSION We derived the vicarious calibration coefficients over the ocean and land using ground-truth measurements. The following GLI characteristics are recognized from the calibration results. First, GLI underestimates the radiance in channels 1, 2, 4, and 5. Next, in the near-infrared channels, there is good agreement between the observed and simulated radiance over bright targets. On the other hand, it is suggested that


Fig. 14. (Solid line) Total offset radiance, (dashed line) radiance from land, and (dotted line) clouds assuming target radius is 30 km, and (dashed–dotted line) the residual radiance calculated by subtracting environmental radiances from total offset radiance.

GLI overestimates the radiance over dark targets. Analysis of the two vicarious calibration results suggests the possibility that the GLI-observed radiance has offset radiance versus the simulated radiance. The combined evaluation of the vicarious calibration results over different targets with different observed radiance proved helpful in understanding the satellite sensor characteristics. ACKNOWLEDGMENT This work was conducted with the JAXA GLI calibration team and the GLI calibration working group. The authors express their thanks to S. Flora (Moss Landing Marine Laboratories) and C. Cattrall (University of Arizona) for providing the MOBY and the Railroad Valley Playa ground-truth data. The authors thank C. McClain (SIMBIOS Project Science Team) for his effort in establishing and maintaining the Lanai AERONET site. Thanks are also extended to the National Oceanic and Atmospheric Administration and to the Japan Meteorological Agency for providing TOVS ozone data and objective analysis data. The authors are grateful to the Open CLASTER project for allowing us to use the RSTAR (system for transfer of atmospheric radiation) package in this research. The authors wish to thank K. Tanaka, Y. Senga, J. Nieke, and all members of the GLI calibration team for their support and suggestions. The authors also wish to acknowledge helpful advice from M. Toratani, H. Kobayashi, T. Y. Nakajima, and R. Hoeller. R. Frouin and the members of Scripps Institution of Oceanography offered useful comments. REFERENCES [1] T. Y. Nakajima, T. Nakajima, M. Nakajima, H. Fukushima, M. Kuji, A. Uchiyama, and M. Kishino, “Optimization of the Advanced Earth Observing Satellite II Global Imager channels by use of radiative transfer calculations,” Appl. Opt., vol. 37, no. 15, pp. 3149–3163, 1998. [2] S. Kurihara, H. Murakami, K. Tanaka, T. Hashimoto, I. Asanuma, and J. Inoue, “Calibration and instrument status of ADEOS-II Global Imager,” Proc. SPIE, vol. 5234, pp. 11–19, 2003. [3] R. H. Evans and H. R. Gordon, “Coastal zone color scanner “system calibration:” A retrospective examination,” J. Geophys. Res., vol. 99, pp. 7293–7308, 1994. [4] H. R. Gordon, “In-orbit calibration strategy for ocean color sensors,” Remote Sens. Environ., vol. 63, pp. 265–278, 1998.

2175

[5] R. E. Eplee, Jr., W. D. Robinson, S. W. Bailey, D. K. Clark, P. J. Werdell, M. Wang, R. A. Barnes, and C. R. McClain, “The calibration of seawifs: Part 2. Vicarious techniques,” Appl. Opt., vol. 40, pp. 6701–6718, 2001. [6] M. Wang and H. R. Gordon, “Calibration of ocean color scanners: How much error is acceptable in the near infrared?,” Remote Sens. Environ., vol. 82, pp. 497–504, 2002. [7] C. J. Bruegge, W. A. Abdou, D. J. Diner, B. J. Gaitley, M. C. Helmlinger, R. A. Kahn, and J. V. Martonchik, “Validating the MISR radiometric scale for the ocean aerosol science communities,” in Post-Launch Calibration of Satellite Sensors. Leiden, The Netherlands: A. A. Balkema, 2004, pp. 103–115. [8] P. N. Slater, S. F. Biggar, R. G. Holm, R. D. Jackson, Y. Mao, M. S. Moran, J. M. Palmer, and B. Yuan, “Reflectance- and radiance-based methods for the in-flight absolute calibration of multispectral sensors,” Remote Sens. Environ., vol. 22, pp. 11–37, 1987. [9] S. F. Biggar, P. N. Slater, and D. I. Gellman, “Uncertainties in the in-flight calibration of sensors with reference to measured ground sites in the 0.4–1.1 m range,” Remote Sens. Environ., vol. 48, pp. 245–252, 1994. [10] K. J. Thome, J. Czapla-Myers, and S. F. Biggar, “Vicarious calibration of Aqua and Terra MODIS,” Proc. SPIE, vol. 5151, pp. 395–405, 2003. [11] H. R. Gordon, “Atmospheric correction of ocean color imagery in the Earth Observation System era,” J. Geophys. Res., vol. 102, no. D14, pp. 17 081–17 106, 1997. [12] T. Nakajima and M. Tanaka, “Matrix formulation for the transfer of solar radiation in a plane-parallel scattering atmosphere,” J. Quant. Spectrosc. Radiat. Transf., vol. 35, pp. 13–21, 1986. [13] T. Nakajima and M. Tanaka, “Algorithms for radiative intensity calculations in moderately thick atmospheres using a truncation approximation,” J. Quant. Spectrosc. Radiat. Transf., vol. 40, pp. 51–69, 1988. [14] K. Stamnes, S. C. Tsay, W. Wiscombe, and K. Jayaweera, “Numerically stable algorithm for discrete-ordinate-method radiative transfer in multiple scattering and emitting layered media,” Appl. Opt., vol. 27, pp. 2502–2509, 1988. [15] R. E. Eplee, Jr. and C. R. McClain, “MOBY data analysis for the vicarious calibration of SeaWiFS bands 1–6,” NASA Goddard Space Flight Center, Greenbelt, MD, NASA Tech. Memo. 2000-206892, 2000. [16] A. B. Franz, E. J. Ainsworth, and S. Baily, “SeaWiFS vicarious calibration: An alternative approach utilizing simultaneous in situ observations of ocean and atmospheric optical properties,” NASA Goddard Space Flight Center, Greenbelt, MD, NASA Tech. Memo. 2001-209982, 2001. [17] H. Murakami, K. Tanaka, S. Kurihara, Y. Okamura, J. Inoue, J. Nieke, I. Asanuma, H. Yatagai, Y. Mitomi, M. Yoshida, R. Higuchi, S. Kawamoto, K. Isono, and Y. Senga, “GLI early calibration results for oceanographic applications,” Proc. SPIE, vol. 5155, pp. 73–82, 2003. [18] D. K. Clark, H. R. Gordon, K. K. Voss, Y. G. W. Brokenow, and C. Trees, “Validation of atmospheric correction over oceans,” J. Geophys. Res., vol. 102, pp. 17 209–17 217, 1997. [19] GLI Calibration Team. (2002, Dec.) GLI user’s guide. JAXA, Tokyo, Japan. [Online]. Available: http://suzaku.eorc.jaxa.jp/GLI/cal/guide/ index.html. In Japanese. [20] B. N. Holben, T. F. Eck, I. Slutsker, D. Tanre, J. P. Buis, A. Setzer, E. Vermote, J. A. Reagan, Y. J. Kaufman, T. Nakajima, F. Lavenu, I. Jankowiak, and A. Smirnov, “AERONET—A federated instrument network and data archive for aerosol characterization,” Remote Sens. Environ., vol. 66, pp. 1–16, 1998. [21] H. Murakami, M. Yoshida, K. Tanaka, H. Fukushima, M. Toratani, A. Tanaka, and Y. Senga, “Vicarious calibration of ADEOS-2 GLI visible to shortwave infrared bands using global datasets,” IEEE Trans. Geosci. Remote Sens., vol. 43, no. 7, pp. 1571–1584, Jul. 2005. [22] H. R. Gordon, J. W. Brown, and R. H. Evans, “Exact Rayleigh scattering calculations for use with the Nimbus-7 Coastal Zone Color Scanner,” Appl. Opt., vol. 27, no. 5, pp. 862–871, 1988. [23] O. Dubovik, A. Smirnov, B. N. Holben, M. D. King, Y. J. Kaufman, T. F. Eck, and I. Slutsker, “Accuracy assessment of aerosol optical properties retrieved from AERONET sun and sky radiance measurements,” J. Geophys. Res, vol. 105, pp. 9791–9806, 2000. [24] S. F. Biggar, D. I. Gellman, and P. N. Slater, “Improved evaluation of optical depth components from Langley plot data,” Remote Sens. Environ., vol. 32, pp. 91–101, 1990. [25] K. J. Thome, B. M. Herman, and J. A. Reagan, “Determination of precipitable water from solar transmission,” J. Appl. Meteorol., vol. 31, pp. 157–165, 1992. [26] J. Nieke, T. Aoki, T. Tanikawa, H. Motoyoshi, and M. Hori, “A satellite cross-calibration experiment,” IEEE Geosci. Remote Sens. Lett., vol. 1, no. 3, pp. 215–219, Jul. 2004. [27] E. F. Vermote, D. Tanre, J. L. Deuze, M. Herman, and J. J. Morcrette, “Second simulation of the satellite signal in the solar spectrum, 6S: An overview,” IEEE Trans. Geosci. Remote Sensing, vol. 35, no. 3, pp. 675–686, May 1997.

2176


Mayumi Yoshida received the B.S. and M.S degrees in earth and planetary science from Hokkaido University, Sapporo, Japan, in 1998 and 2000, respectively. Since 2000, she has been working at the Remote Sensing Technology Center of Japan, Tokyo, and has participated in the in-flight calibration phase of the ADEOS2/GLI instrument.

Hiroshi Murakami received the B.S., M.S., and Ph.D. degrees in geophysics from Tohoku University, Sendai, Japan, in 1995, 1997, and 2002, respectively. Since 1997, he has been working on ADEOS OCTS and ADEOS-2 GLI at the NASDA Earth Observation Research Center (JAXA since 2003), Tokyo. His current work is algorithm development, validation of oceanographic products, and satellite sensor calibration.

Yasushi Mitomi received the B.E. and M.E. degrees in ocean engineering from Tokai University, Tokyo, Japan, in 1991 and 1993, respectively. Since 1993, he has been working at the Remote Sensing Technology Center of Japan, Tokyo. His current research theme is the development of the atmospheric correction method for ocean color remote sensing.

Masahiro Hori received the B.S., M.S., and Ph.D. degrees from the Hokkaido University, Sapporo, Japan, in 1993, 1995, and 1998, respectively. He is currently an Engineer at the Earth Observation Research and Application Center, Japan Aerospace Exploration Agency, Tokyo. His areas of research include aerosol chemical characterization, calculation of aerosol optical properties, laboratory characterization of activation capability of organic aerosols as cloud condensation nuclei, and more recently, optical remote sensing of snow, ice, and clouds in the cryosphere.

Kurtis J. Thome received the B.S. degree in meteorology from Texas A&M University, College Station, and the M.S. and Ph.D. degrees in atmospheric sciences from the University of Arizona, Tucson. He is currently an Associate Professor of optical sciences at the University of Arizona, where he is the head of the Remote Sensing Group. He has served as a member of the Landsat-7, ASTER, and MODIS science teams. His current research is focused on the vicarious calibration of earth-imaging sensors and related studies in atmospheric remote sensing, radiative transfer, and satellite atmospheric correction

Dennis K. Clark received the B.S. degree in meteorology from Pennsylvania State University, University Park, in 1964. He is currently serving as the Marine Observing Systems Team Leader, Office of Research and Applications, National Environmental Satellite, Data, and Information Service, National Oceanic and Atmospheric Administration (NOAA),Washington, DC. He began his career with the U.S. Naval Oceanographic Office as a Physical Oceanographer developing current measurement techniques and predictions for use in Vietnam. In 1969, he transferred to NOAA and began conducting research in the field of marine optics and their application to remote sensing. Since then, he has served as a member on all of the U.S. ocean color instrument science teams and has supported several Japanese sensor teams. He has developed bio-optical algorithms for CZCS and MODIS Terra and Aqua ocean color sensors, marine and atmospheric optical instrumentation (which includes the Marine Optical Buoy), calibration measurement protocols, and ocean color product validations utilizing at-sea experimental measurements. He has authored/coauthored over 50 peer-reviewed and other publications. Mr. Clark received the Department of Commerce Gold Medal in 2000 for his scientific contributions to the field of ocean color science.

Hajime Fukushima received the B.S. degree in electrical engineering and the M.S. and Ph.D. degrees in computing science from Tohoku University, Sendai, Japan, in 1969, 1972, and 1978, respectively. He is currently a Professor of information and communication technology with the school of High-Technology for Human Welfare, Tokai University, Numazu, Japan. He has been interested in atmospheric correction on ocean color remote sensing and has been involved in Japanese satellite ocean color observation programs. He is also interested in aerosol remote sensing over the ocean and land, with special interest on Asian dust and other anthropogenic aerosols.