New approaches to removing cloud shadows and evaluating the 380 nm surface reflectance for improved aerosol optical thickness retrievals from the GOSAT/TANSO‐Cloud and Aerosol Imager
[1] A satellite aerosol retrieval algorithm was developed to utilize a near‐ultraviolet band of the Greenhouse gases Observing SATellite/Thermal And Near infrared Sensor for carbon Observation (GOSAT/TANSO)‐Cloud and Aerosol Imager (CAI). At near‐ultraviolet wavelengths, the surface reflectance over land is smaller than that at visible wavelengths. Therefore, it is thought possible to reduce retrieval error by using the near‐ultraviolet spectral region. In the present study, we first developed a cloud shadow detection algorithm that uses first and second minimum reflectances of 380 nm and 680 nm based on the difference in Rayleigh scattering contribution for these two bands. Then, we developed a new surface reflectance correction algorithm, the modified Kaufman method, which uses minimum reflectance data at 680 nm and the NDVI to estimate the surface reflectance at 380 nm. This algorithm was found to be particularly effective at reducing the aerosol effect remaining in the 380 nm minimum reflectance; this effect has previously proven difficult to remove owing to the infrequent sampling rate associated with the three‐day recursion period of GOSAT and the narrow CAI swath of 1000 km. Finally, we applied these two algorithms to retrieve aerosol optical thicknesses over a land area. Our results exhibited better agreement with sun‐sky radiometer observations than results obtained using a simple surface reflectance correction technique using minimum radiances.
1 Introduction
[2] Aerosols such as sulfate, sea salt, volcanic ash, soot, and mineral dust particles have a considerable influence on the earth's climatic system. Aerosols can have direct, indirect, and semidirect radiative effects; their net forcing is thought to be negative, although large uncertainties (i.e., more than ±0.5 Wm−2) remain on the magnitude [Solomon et al., 2007]. Therefore, it is important to determine both the distributions and characteristics of aerosols and long‐lived greenhouse gases (such as carbon dioxide and methane) in order to improve climate predictions under global change scenarios [Solomon et al., 2007].
[3] The Greenhouse gases Observing SATellite (GOSAT) was launched in January 2009 to measure carbon dioxide loading at a global scale, using the Thermal And Near infrared Sensor for carbon Observation Fourier Transform Spectrometer (TANSO‐FTS) for shortwave and longwave infrared spectral radiances [Yokota et al., 2004].
[4] The satellite also carries the Cloud and Aerosol Imager (TANSO‐CAI), a push‐broom scanning imager with four bands (380 nm, 674 nm, 870 nm, and 1600 nm) for the purposes of cloud screening and aerosol detection. The footprints of these bands are 0.5 km, 0.5 km, 0.5 km, and 1.5 km, respectively, with signal‐to‐noise ratios of 200. These signal‐to‐noise ratios are measured in 47, 45, 29, and 7 [W/m2/str/µm], respectively [Yoshida et al., 2008; Kuze et al., 2009]. These radiances are typical values for land surface. CAI was fitted with a 380 nm wavelength near‐ultraviolet band based on the recognition that this wavelength offers advantages for aerosol detection over bright land surfaces in the visible spectrum. This occurs because land surface reflectance at near‐ultraviolet wavelengths is less than that at visible wavelengths, as found by the retrieval of aerosol optical depth using TOMS [Torres et al., 1998, 2002] and OMI [Torres et al., 2007, 2013] near UV measurements and the MODIS Deep Blue algorithm [Hsu et al., 2004, 2006]. Here, we demonstrate an approach that applies the near‐ultraviolet band in GOSAT/TANSO‐CAI aerosol retrieval.
2 Methods
2.1 Aerosol Model
[5] The log‐normal binomial function has been reported to be a good approximation of observed size distributions for tropospheric aerosols and has been used in remote sensing from surface and space [WCP (World Climate Programme), 1983; Nakajima and Higurashi, 1998; Higurashi et al., 2000; Dubovik et al., 2002; Smirnov et al., 2003], as shown in equation 1:
(1)
where V is column volume loading as a function of particle radius r. In the present study, we assumed a log‐normal size distribution with c1 = 0.5, c2 = 0.5, rm,1 =0.148 um, rm,2 = 3.86 um, S1 = 1.56, and S2 = 2.00, based on the average of AERONET sites used in the validation. c1 and c2 represent the peak values of each mode; rm,1 and rm,2 represent mode radii; and S1 and S2 represent standard deviation. The “small mode” is the first mode and the “coarse mode” is the second mode in this equation. Then, we assigned the complex refractive index 1.503 – 7.16 × 10−8i to small mode particles, and 1.445 – 1.00 × 10−8i to coarse mode particles. These values are based on the sea spray and sulfate models [WCP (World Climate Programme), 1983]. The imaginary parts of these refractive indices are thought to be too small in areas of large anthropogenic aerosol loading, resulting in an underestimation of the retrieved aerosol optical thickness (AOT). Therefore, we also investigated more realistic aerosol models during validation of our satellite retrieval results with respect to ground observations, as we describe later.
[6] Our aerosol retrieval algorithm is based on a lookup table method [Higurashi and Nakajima, 1999] for CAI radiances with the above mentioned aerosol polydispersion model and uses the atmospheric radiative transfer code Rstar [Nakajima and Tanaka, 1986, 1988] and its polarization version, Pstar [Ota et al., 2009]. These software packages are described at and available from the OpenCLASTR website (http://ccsr.aori.u‐tokyo.ac.jp/~clastr/).
2.2 Surface Model
[7] Clear‐sky pixels were selected via the cloud detection algorithm of Ishida and Nakajima [2009] and Ishida et al. [2011]. Screening of cloud shadow is important in the estimation of ground reflectance using the minimum reflectance, because cloud shadow induces considerable error in the aerosol retrieval process. Luo et al. [2008] detected cloud shadows using geometric relationships between cloud top height, land surface, and a satellite. However, it is difficult to apply this method in the present study, because the CAI sensor is not equipped with a thermal infrared band, which is necessary to determine the height of the cloud top, even though IR observations may not always determine cloud height. Pinty et al. [2000] fitted the temporal variation of surface reflectance to a semiempirical function and rejected cloud shadow pixels from the resulting surface reflectance product. However, the observation frequency of the CAI sensor (once every 3 days) is insufficient to apply this method in the present study. Takeuchi and Yasuoka [2006] used first and second minima of visible and near‐infrared bands to reject cloud shadow pixels in MODIS data sets. Similarly, we developed a method of rejecting cloud shadow pixels using the difference between the first (Rmin,1st) and second (Rmin,2nd) minimum reflectances of both visible and near‐ultraviolet bands, as follows:
(2)
with
where the thresholds were set as follows: Rt,380nm = 0.10 and Rt,380nm = 0.06. In equation 2, subscript mod means modified. Figures 1 and 2 illustrate the minimum reflectance with and without cloud shadow screening for an area of northern China; Beijing is located slightly south of the image center. These figures indicate that most of the cloud shadow pixels were screened by the criteria given in equation 2. Figure 3 illustrates the fraction of pixels judged as cloud shadow according to this method for May, July, September, and November 2009, indicating that 5–15% of the minimum reflectance pixels are contaminated by cloud shadow in the region, except at high latitudes. The larger fraction (up to 45%) in high‐latitude areas is due to large solar zenith angles and the influence of snow and ice.
Distribution of Rmin,1st with false color imaging (band 1 = blue; band 2 = red; band 3 = green). A zoomed image, in which cloud shadows are visible, is shown in the bottom‐left corner.
Zonally averaged ratio of number of cloud‐shadow pixels to the total number of Rmin,1st.
[8] Applying the present cloud shadow screening algorithm, the minimum surface reflectance Rmin was obtained for each 0.5 × 0.5 km area for every 31 day period. Assuming moderate aerosol loading, the influence of aerosols in Rmin is approximately proportional to the AOT:
(3)
where Ag'(n) is the Rayleigh scattering‐corrected minimum reflectance in band n, Ag(n) the true ground reflectance, β(n) a proportionality coefficient, and τ(n) the AOT in band n, which remains within the minimum reflectance. The proportionality coefficient β(n) is a function of the size distribution and single‐scattering albedo of the aerosol system and Ag itself, and is usually positive unless the ground reflectance is greater than the neutral reflectance [Kaufman et al., 2001]. The positiveness of β(n) assures that Ag' is a good estimate for the ground reflectance Ag, because the satellite‐received radiance increases with increasing AOT when β(n) > 0 except in the case of strongly absorbing aerosols. Therefore, the simplest approximation is the conventional approximation of minimum reflectance:
(4)
[9] However, it is found that an additional correction is necessary to obtain the ground reflectance for Ag' in band 1 from CAI owing to the low observation frequency (i.e., a maximum of 11 observations within 31 days), relatively narrow CAI swath of 1000 km, and 3 day recurrent orbit of GOSAT. In practice, much fewer than 11 observations are typically obtained per month owing to the presence of clouds. Thus, Ag'(1) still includes a large aerosol influence. Accordingly, we used Ag'(2) instead of Ag'(1) to estimate Ag(1), as in the method proposed by Kaufman et al. [1997], by assuming the proportionality of the ground reflectances at near‐ultraviolet and visible wavelengths:
(5)
[10] The original Kaufman method substitutes a near‐infrared wavelength for a longer wavelength, e.g., Ag(4) in the case of CAI. However, we found no proportionality in the case of a band pair including the near‐ultraviolet (380 nm) and near‐infrared (1600 nm) bands, likely because the distance between the two wavelengths is too large to assume similar ground absorption mechanisms. Therefore, we used the relationship shown in equation 5 and assumed that the proportionality factor depends on the land type classified according to the NDVI, based on the fact that Murakami et al. [2007] found a good representation of 380 nm reflectance as a quadratic function of the NDVI in their analysis of Global Imager (GLI) data from the ADEOS‐II satellite. They defined the relationship between ground reflectance and the NDVI as Ag(1) = c1NDVI2 + c2NDVI2 + c3, where c1, c2, and c3 are empirically defined coefficients.
[11] Figure 4 illustrates the relationship of the frequency distribution of the proportionality factor a(NDVI) as classified by NDVI ranges of (0, 0.1), (0.2, 0.3), (0.4, 0.5), (0.6, 0.7), and (0.8, 0.9). This figure indicates that the mode value of the a‐factor increases as NDVI increases, and the distribution becomes narrower for smaller values of the NDVI. The mode value of a(NDVI) for each NDVI class was fitted as follows:
(6)
where c1 = − 0.192309, c2 = − 9.62693, and c3 = 0.3. In most cases, it was found that the ground reflectance at 380 nm is less than that at 670 nm by the a‐factor less than 0.5. In this case, the ground reflectance Ag(1) can be estimated using equations 3 and 5:
(7)
aβ(2)τ(2) is usually negligible, because a < 0.5 and β(2)τ(2) < β(1)τ(1) for most cases of light aerosol loading without coarse dust particles. From a practical perspective, it is also preferable to use instead of , because 670 nm observations are available from most recent satellite imagers (such as MODIS), whereas τ(1) is an unknown variable to be retrieved. Thus, we used in the present study to correct the remaining AOT in for better evaluation of Ag(1).
Frequency distributions of the proportionality factor a(NDVI) in equation (5) classified by NDVI ranges.
3 Results
[12] We analyzed the data from CAI for April–December 2009. Radiometric correction was conducted by Shiomi et al. [2010]. The AOT in band 1 was then retrieved from the radiance data using the land surface albedo Ag(1), assuming either equation 4 or equation 5. Hereafter, we refer to the former method as the minimum reflectance (MR) method and the latter as the modified Kaufman (MK) method.
[13] The upper panel of Figure 5 presents an RGB image derived from CAI for the area around Beijing, illustrated in Figures 1 and 2, on 30 August 2009. The blue, red, and green colors represent bands 1 (380 nm), 2 (680 nm), and 3 (870 nm), respectively. In this representation, the ocean, vegetated land, and arid areas are represented by blue, green, and yellow, respectively. Clouds are represented by white, because cloud reflectance depends little on wavelength.
A false color image (band 1 = blue; band 2 = red; band 3 = green) around for Beijing and surrounding areas on 30 August 2009. (a) Distribution of AOT380 retrieved by using the MK method for the same area for the same day. (b) SSA = 1.00 is assumed.
[14] The lower panel of Figure 5 illustrates the AOT at 380 nm (hereafter, referred to as AOT380) on 30 August 2009; this was calculated from AOT in band 1 according to the MK method and using the aerosol polydispersion model described in the preceding section. The AOT380 product clearly illustrates a dense aerosol plume spreading around Beijing, corresponding to a faint color modulation in the RGB image.
[15] The AOT380 retrieved in this study was then compared with AOT380 from the AERONET (AErosol RObotic NETwork) level 2.0 product [Smirnov et al., 2000]. AERONET is a network of ground observations of aerosol optical properties [Holben et al., 1998] and is managed by NASA. AOTs for 253 days were derived from a resampled CAI data set consisting of 2400 × 1441 pixels for the globe from 23 April 2009, to 31 December 2009, for comparison with the AERONET AOTs. Only data that satisfied the following five conditions were compared: (1) the observation sites were located between latitudes of 65°S and 65°N; (2) more than 10 pixels were analyzed successfully in a 5 × 5 pixel window around each AERONET site; (3) the standard deviation of AOT550 was less than 0.08; (4) the time difference between the AERONET observation site and the surrounding CAI observation was less than 30 min; (5) availability of AOD retrievals at 380 nm; and (6) there were more than three observations satisfying all of conditions 1–5 for the site in question. First, the calculations were performed using the nonabsorbing aerosol model and 14 selected AERONET stations. We also performed similar calculations using SSAs at the 380 nm wavelength, which were extrapolated from mean SSA values at the wavelengths used in the AERONET observations for these 14 sites. Table 1 summarizes the observation sites used in this comparison, average SSAs calculated from AERONET, and the number of matching times. Figures 6, 7, and 8 show scatterplots of correlation coefficients, biases and root mean square differences for the MR and MK methods, respectively. The statistics in these figures show that the MK method generally performs better. For example, for all cases except three sites (Egbert, Fresno, and Paris), RMK is larger than RMR. Sede Boker is not shown in this scatterplot, because there are only two observations that satisfy the criteria for both cases simultaneously. The RMK of all points is 0.59 and RMR of all points is 0.47 for nonabsorbing aerosol model case. On the other hand, the RMK of all points is 0.60 and the RMR of all points is 0.48 for AERONET's SSA case. The RMR of all points is larger than the RMK of all points in both cases. Biases of CAI‐retrieved AOT with respect to that of AERONET were calculated as follows:
(8)
Furthermore, root mean square differences (RMSD) between CAI‐retrieved AOT and that of AERONET were calculated as follows:
(9)
where N is the number of samples, AOTCAI the AOT obtained by CAI, and AOTAERONET the AOT obtained by AERONET. The average bias of method MK was 0.032 and that of method MR was −0.094 in nonabsorbing aerosol case. On the other hand, the average bias of method MK was 0.061 and that of method MR was −0.080 in AERONET's SSA case. Here, “average” means the average of all data of each site. The MK was found to be closer to zero than that of method MR, indicating that the underestimation in the MR method is improved by use of the MK method. Moreover, the average RMSD for the MK was 0.118, and that for MR was 0.132 in nonabsorbing aerosol case. On the other hand, the average RMSD for the MK was 0.147 and that for MR was 0.135 in AERONET's SSA case. The average RMSD for the MK method was found to be smaller than that for the MR method in nonabsorbing aerosol case. On the other hand, there is small difference between the average RMSD for MK method and that for MR method in AERONET's SSA case. The left panels in Figures 6, 7, and 8 show the scattering plot of correlation coefficients, RMSD, and bias in the nonabsorbing aerosol case, and the right panels in Figures 6, 7, and 8 show those in the AERONET's SSA case. The distribution pattern in left panels and right panels are similar, indicating that the impact of errors in the assumed ground surface albedo is greater than that in the assumed SSA.
Table 1.
Observation Sites, Latitude, Longitude, Average SSA at 380 nm, and Number of Observations (NMK, NMR, Nboth)aa
Prime means that SSA fixed at 1.0.
RMSD between CAI‐retrieved and AERONET‐observed AOT for MR and MK. (a) SSA is fixed as 1.0 and (b) AERONET‐observed SSAs are used.
[16] Scatterplots of CAI‐retrieved AOT380 vs. AERONET‐observed AOT380 are presented in Figures 9-15, where crosses and circles represent the MR and MK results, respectively. The left and right ones present AOT380s retrieved assuming the nonabsorbing and absorbing aerosol models, respectively, with SSAs calculated from AERONET (see Table 1). Different AERONET sites are characterized by different aerosol species and different surface types. For example, Tomsk, Yaktusk, and Skukuza can be influenced by different types of biomass burning aerosols to various degrees. Sede Boker can be influenced by mineral dust aerosols, while urban sites such as Paris and GSFC have an increased contribution from sulfate aerosols. Furthermore, Sede Boker is in arid terrain while the other sites mentioned are in more vegetated regions. As illustrated by these figures, most of the AOT380 values retrieved from the CAI data are smaller than those from AERONET for the MR method; this indicates that there is some AOT contribution remaining in the . Conversely, the MK method improves the comparison overall, but produced overestimations at Bonanza Creek and Laegeren and an underestimation at Paris. Retrieval of AOTs is generally more difficult in cities and areas of complex terrain owing to errors in the assumed ground reflectance and/or aerosol optical properties. It makes little difference for the sites with SSA larger than 0.95 whether using SSA 1.0 or the actual SSA in Table 1. On the other hand, the MK retrievals agree better with AERONET for the sites where SSA is less than 0.95 such as Paris, Sevastopol, Tomsk, and Sede Boker when using the actual SSA than when assuming SSA = 1.0. This is especially true at the Tomsk and Skukuza sites. These two sites can exhibit very high aerosol absorption optical depths [Kajii et al., 2002; Eck et al., 2003].
Comparison of AOT380 from CAI and from AERONET at Halifax (a) with nonabsorption aerosol model and (b) with AERONET absorption model. Comparison of AOT380 from CAI and from AERONET at GSFC (c) with nonabsorption aerosol model and (d) with AERONET absorption model. Blue circles show MK method, and pink crosses show MR method.
As for Figure 9, but at the observation sites of (a, b) Sede Boker and (c, d) Skukuza.
4 Sensitivity Analysis
[17] In the present study, we adopted SSAs that were derived from values observed by AERONET. We also used the nonabsorption aerosol model for sensitivity analysis. Additionally, it is essential that the influences of the aerosol model and other factors used in this study be investigated. The aerosol's size distribution in the nominal case is same with previous section, (c1 = 0.5, c2 = 0.5, rm,1 =0.148 µm, rm,2 = 3.86 µm, S1 = 1.56, and S2 = 2.00) and SSA's nominal case is nonabsorption model (SSA = 1.0). The aerosol height of nominal case is that an aerosol layer exists at 0–2 km as a box function. Ground reflectance of nominal case is 0.05. Solar and satellite zenith angles are both assumed to be 30°, and the relative azimuth angle is assumed to be 0° (forward scattering), 90°, and 180° (backscattering). In this sensitivity analysis, first we calculated the radiance for each specific condition. Simulations were performed for AOT of 0.2 and 0.5 at 550 nm (= AOT of 0.35 and 0.88 at 380 nm). Then, we retrieved the AOT with nominal condition.
[18] First, we investigated the influence of the imaginary refractive index of aerosols. Aerosols in urban areas can contain absorbing particles such as soot, which would create errors in our retrieval. In this sensitivity test, we used SSA = 0.95 and SSA = 0.90 case. The sensitivity analysis for SSA is presented in Table 2. Negative value of ΔAOT indicates underestimation. It is clear from Table 2 that the influence of SSA is considerable. The error in AOT is largest for relative azimuth angles around 180°, i.e., the backscattering geometry. In the backscattering geometry, the error is about 30% for SSA = 0.95 and about 60% for SSA = 0.90. This result indicates that determination of the SSA is important.
Table 2.
Sensitivity Analysis for SSAaa
ϕ is relative azimuth angle, respectively. SSA is assumed as 0.95 and 0.90 in forward calculation.
ϕ
0°
90°
180°
0°
90°
180°
ΑΟΤ
0.2
0.2
0.2
0.5
0.5
0.5
ΔAOT0.95
−0.05
−0.05
−0.06
−0.13
−0.14
−0.16
ΔAOT0.90
−0.09
−0.10
−0.11
−0.25
−0.27
−0.29
a
ϕ is relative azimuth angle, respectively. SSA is assumed as 0.95 and 0.90 in forward calculation.
[19] The influence of the mode radii is presented in Table 3. The results suggest that the influence of mode radius of small particle is considerable about 20%, but that of large particle is negligible.
Table 3.
Sensitivity Analysis for Mode Radiiaa
ϕ is relative azimuth angle. In the second and third lines, r1 is assumed as 0.20 µm, and in the fourth and fifth lines, r2 is assumed as 3.00 µm in forward calculation.
ϕ
0°
90°
180°
0°
90°
180°
ΑΟΤ
0.2
0.2
0.2
0.5
0.5
0.5
ΔAOT
−0.05
−0.05
−0.04
−0.13
−0.12
−0.11
ΑΟΤ
0.2
0.2
0.2
0.5
0.5
0.5
ΔAOT
0.0
0.0
0.0
−0.1
−0.1
−0.1
a
ϕ is relative azimuth angle. In the second and third lines, r1 is assumed as 0.20 µm, and in the fourth and fifth lines, r2 is assumed as 3.00 µm in forward calculation.
[20] We also studied the influence of surface reflectance. We used surface reflectance = 0.03 for perturbation. This is because the half width of surface reflectance is about 0.03 (see Figure 16). Our results are shown in Table 4. The errors caused by uncertainty of surface reflectance have a nonnegligible influence on AOT retrieval especially absorption aerosol case. In the absorption aerosol case, the error is at most 50%.
The histogram of global surface reflectance at 380 nm, in May 2009.
Table 4.
Sensitivity Analysis for Surface Reflectanceaa
We assumed that surface reflectance is 0.03 in forward calculation as perturbation. ϕ is relative azimuth angle, respectively. The second and third lines are for SSA = 1.00 case, the fourth and fifth lines are for SSA = 0.95 case, and the sixth and seventh lines are for SSA = 0.90 case.
ϕ
0°
90°
180°
0°
90°
180°
ΑΟΤ
0.2
0.2
0.2
0.5
0.5
0.5
ΔAOT
−0.07
−0.08
−0.07
−0.06
−0.07
−0.07
ΑΟΤ
0.2
0.2
0.2
0.5
0.5
0.5
ΔAOT
−0.09
−0.11
−0.10
−0.08
−0.09
−0.10
ΑΟΤ
0.2
0.2
0.2
0.5
0.5
0.5
ΔAOT
−0.14
−0.16
−0.16
−0.11
−0.13
−0.15
a
We assumed that surface reflectance is 0.03 in forward calculation as perturbation. ϕ is relative azimuth angle, respectively. The second and third lines are for SSA = 1.00 case, the fourth and fifth lines are for SSA = 0.95 case, and the sixth and seventh lines are for SSA = 0.90 case.
[21] Finally, we investigated the sensitivity test for aerosol height. Large molecular scattering at 380 nm can result in significant error in retrieved AOT if the assumed aerosol vertical distribution is incorrect. We assumed that an aerosol layer exists at 2–4 km as a box function, and we assumed both nominal SSA case and absorbing case. Our results are shown in Table 5. This results show that the height assumption is not really important for very weakly or nonabsorbing aerosols. However, it is important for absorbing aerosols.
Table 5.
Sensitivity Analysis for Aerosol Heightaa
We assumed that an aerosol layer exists at 2–4 km as a box function in forward calculation. ϕ is relative azimuth angle, respectively. The second and third lines are for nominal SSA case (SSA = 1.00), the fourth and fifth lines are for SSA = 0.95, and the sixth and seventh lines are for SSA = 0.90 case.
ϕ
0°
90°
180°
0°
90°
180°
ΑΟΤ
0.2
0.2
0.2
0.5
0.5
0.5
ΔAOT
0.01
0.00
0.00
0.02
0.00
0.00
ΑΟΤ
0.2
0.2
0.2
0.5
0.5
0.5
ΔAOT
0.03
0.03
0.02
0.10
0.09
0.08
ΑΟΤ
0.2
0.2
0.2
0.5
0.5
0.5
ΔAOT
0.09
0.08
0.09
0.30
0.30
0.32
a
We assumed that an aerosol layer exists at 2–4 km as a box function in forward calculation. ϕ is relative azimuth angle, respectively. The second and third lines are for nominal SSA case (SSA = 1.00), the fourth and fifth lines are for SSA = 0.95, and the sixth and seventh lines are for SSA = 0.90 case.
[22] The sensitivity analyses described above suggest that mode radius of large particle and aerosol height for nonabsorbing aerosol case do not have a considerable influence on AOT retrieval, but SSA assumption, mode radius of small particle, surface reflectance, and aerosol height for absorbing aerosol case have a considerable influence on AOT retrieval. Especially, the AOT error caused by an error in surface reflectance causes up to 80% error. Therefore, it is important to estimate the surface reflectance correctly, and the method of estimation of the surface reflectance described in this paper is useful to estimate better AOT values. However, further work will be required in future to ensure the development of more reasonable aerosol and surface reflection models, particularly with regard to the imaginary index of reflection.
5 Conclusions
[23] In the present study, we developed two new algorithms for aerosol remote sensing using GOSAT/TANSO‐CAI observations. The first of these algorithms provides a method to detect cloud shadow and utilizes band 1 and band 3 reflectances. Large Rayleigh scattering by atmospheric molecules at the 380 nm wavelength (band 1) was found to be effective for detecting cloud shadow when combined with reflectances at longer wavelengths. We rejected 5–15% of the imager pixels owing to contamination by cloud shadow for middle‐ to low‐latitude land surfaces.
[24] Then, we developed the modified Kaufman method to estimate the ground reflectance in band 1 (380 nm), as described by equations 5 and 6. This method seems to be more accurate than the minimum reflectance (MR) method described by equation 4. We found that a large negative bias in the AOT (approximately 0.1) when using the MR method was corrected by the modified Kaufman (MK) method. This result also suggests that the minimum reflectance for a near‐ultraviolet wavelength over a 31 day period is still contaminated by aerosols when the imager swath is small (1000 km), likely because the satellite orbit is not oriented for frequent observation at a single point of the globe.
[25] We have included several assumptions that must be examined in further detail in future studies. For example, the ground type classification used to determine the a‐factor in equation 5 is derived only from the NDVI. However, the NDVI is not the only index available to characterize ground surface optical properties. Both desert and urban areas have small NDVIs, but urban areas cannot be differentiated using this method. We also assumed a simple nonabsorbing aerosol model for the demonstration of the 380 nm band utility, but this assumption typically leads to underestimation of the AOT in urban regions. Despite the limitations imposed by these assumptions, this is a unique study of near‐ultraviolet remote sensing at a resolution of 0.5 km, particularly because no other high‐resolution near‐ultraviolet band imagery is available. Future work will conduct retrieval using more realistic aerosol optical properties for each region of the globe.
[26] We used several assumptions when the AOT is retrieved. We made sensitivity studies to evaluate AOT retrieval errors caused by these assumptions and to investigate the range of tolerance of for appropriate parameter setting. The result of the sensitivity studies showed that the largest influences on the AOT retrieval are caused by errors in SSA and surface reflectance. Although we found a considerably large AOT error as 30% for an error of 0.05 in SSA, it is still smaller than what we expect in the aerosol remote sensing for longer wavelengths, say 0.55 µm, at which the surface reflectance is larger than in the case of 0.38 µm. The value of the critical SSA, at which the satellite‐received radiance becomes independent of the AOT value [Kaufman, 1987], is lower than 0.9 for such small surface reflectance at 0.38 µm, and the AOT error is nearly proportional to 1 ‐ SSA in the assumed range of SSA in the sensitivity tests.
Acknowledgments
[27] We are grateful to the GOSAT teams of the National Institute for Environmental Studies and the Japan Aerospace Exploration Agency for providing CAI data and technical support for the present analysis. We are thankful to the AERONET teams at the stations used in this paper, such as Glen Lesins, Brent Holben, Adrian Hauser, Stefan Wunderle, Bruce McArthur, Norm O'Neill, Carol J. Bruegge, Mikhail Panchenko, John R. Vande Castle, Francois Ravetta, Jacques Pelon, Arnon Karnieli, and Stuart Piketh. Some of the authors were supported by projects of JAXA/EarthCARE and GCOM/C, MEXT/VL for Climate System Diagnostics, MOE/Global Environment Research Fund A‐1101, NIES/GOSAT, NIES/CGER, and MEXT/RECCA/SALSA.
References
Dubovik, O.,
B. N. Holben,
T. F. Eck,
A. Smirnov,
Y. J. Kaufman,
M. D. King,
D. Tanré, and
I. Slutsker (2002), Variability of absorption and optical properties of key aerosol types observed in worldwide locations, J. Atmos. Sci., 59, 590– 608.
Eck, T. F., et al. (2003), Variability of biomass burning aerosol optical characteristics in southern Africa during the SAFARI 2000 dry season campaign and a comparison of single scattering albedo estimates from radiometric measurements, J. Geophys. Res., 108(D13, 8477), doi:10.1029/2002JD002321.
Higurashi, A., and
T. Nakajima (1999), Development of a two channel aerosol retrieval algorithm on global scale using NOAA / AVHRR, J. Atmos. Sci., 56, 924– 941.
Higurashi, A.,
T. Nakajima,
B. N. Holben,
A. Smirnov,
R. Frouin, and
B. Chatenet (2000), A study of global aerosol optical climatology with two channel AVHRR remote sensing, J. Clim., 13, 2011– 2027.
Hsu, N. C.,
S.‐C. Tsay,
M. D. King, and
J. R. Herman (2004), Aerosol properties over bright‐reflecting source regions, IEEE Trans. Geosci. Remote Sens., 42, 557– 569.
Hsu, N. C.,
S.‐C. Tsay,
M. D. King, and
J. R. Herman (2006), Deep Blue retrievals of Asian aerosol properties during ACE‐Asia, IEEE Trans. Geosci. Remote Sens., 44, 3180– 3195.
Ishida, H., and
T. Y. Nakajima (2009), Development of an unbiased cloud detection algorithm for a spaceborne multispectral imager, J. Geophys. Res., 114, D07206, doi:10.1029/2008JD010710.
Ishida, H.,
T. Y. Nakajima,
T. Yokota,
N. Kikuchi, and
H. Watanabe (2011), Investigation of GOSAT TANSO‐CAI cloud screening ability through an inter‐satellite comparison, J. Appl. Meteorol. Clim., 50, 1571– 1586.
Kajii, Y., et al. (2002), Boreal forest fires in Siberia in 1998: Estimation of area burned and emissions of pollutants by advanced very high resolution radiometer satellite data, J. Geophys. Res., 107(D24, 4745), doi:10.1029/2001JD001078.
Kaufman, Y. J.,
A. E. Wald,
L. A. Remer,
B.‐C. Gao,
R.‐R. Li, and
L. Flynn (1997), The MODIS 2.1‐um channel‐correlation with visible reflectance for use in remote sensing of aerosol, IEEE Trans. Geosci. Remote Sens., 35, 1286– 1298.
Kaufman, Y. J.,
D. Tanré,
O. Dubovik,
A. Kamieli, and
L. A. Remer (2001), Absorption of sunlight by dust as inferred from satellite and ground‐based remote sensing, Geopys. Res. Lett., 28, 1479– 1482.
Kuze, A.,
H. Suto,
M. Nakajima, and
T. Hamazaki (2009), Thermal and near infrared sensor for carbon observation Fourier‐transform spectrometer on the Greenhouse Gases Observing Satellite for greenhouse gases monitoring, Appl. Opt., 48(35), 6716– 6733.
Luo, Y.,
A. P. Trishchenko, and
K. V. Khlopenkov (2008), Developing clear‐sky, cloud and cloud shadow mask for producing clear‐sky composites at 250‐meter spatial resolution for the seven MODIS land bands over Canada and North America, Remote Sens. Environ., 112, 4167– 4185.
Nakajima, T., and
M. Tanaka (1986), Matrix formulation for the transfer of solar radiation in a plane‐parallel scattering atmosphere, J. Quant. Spectrosc. Radiat. Transfer, 35, 13– 21.
Nakajima, T., and
M. Tanaka (1988), Algorithms for radiative intensity calculations in moderately thick atmospheres using a truncation approximation, J. Quant. Spectrosc. Radiat. Transfer, 40, 51– 69.
Ota, Y.,
A. Higurashi,
T. Nakajima, and
T. Yokota (2009), Matrix formulations of radiative transfer including the polarization effect in a coupled atmosphere‐ocean system, J. Quant. Spectrosc. Radiat. Transfer, 111, 878– 894.
Pinty, B.,
F. Roveda,
M. Verstraete,
N. Gobron,
Y. Govaerts,
J. Martonchik,
D. Diner, and
R. Kahn (2000), Surface albedo retrieval from Meteosat 2. Applications, J. Geophys. Res., 105(D14, 18113), doi:10.1029/2000JD900114.
Shiomi, K.,
S. Kawakami, and
T. Kina (2010), Operation results of initial calibration and validation of “IUBKI”, Aeronautical Space Sci. Jpn., 58, 158– 163 (in Japanese).
Smirnov, A.,
B. N. Holben,
T. F. Eck,
O. Dubovik, and
I. Slutsker (2000), Cloud‐screening and quality control algorithms for the AERONET database, Remote Sens. Environ., 73(3), 337– 349.
Smirnov, A.,
B. N. Holben,
O. Dubovik,
R. Frouin,
T. F. Eck, and
I. Slutsker (2003), Maritime component in aerosol optical model derived from Aerosol Robotic Network data, J. Geophys. Res., 108(D1, 4033), doi:10.1029/2002JD002701.
Solomon, S., et al. (2007), Intergovernmental Panel on Climate Change (IPCC), Climate Change 2007: The Physical Science Basis, Cambridge Univ. Press, Cambridge, United Kingdom and New York, NY, USA.
Takeuchi, W., and
Y. Yasuoka (2006), Development of cloud and shadow free compositing technique with MODIS QKM ASPRS 2006, Annual Conference, Reno, Nevada, USA, 2006.
Torres, O.,
P. K. Bhartia,
J. R. Herman, and
J. Gleason (1998), Derivation of aerosol properties from satellite measurements of backscattered ultraviolet radiation: Theoretical basis, J. Geophys. Res., 103, 17,099– 17,110.
Torres, O.,
P. K. Bhartia,
J. R. Herman,
A. Sinyuk,
P. Ginoux, and
B. Holben (2002), A long‐term record of aerosol optical depth from TOMS observations and comparison to AERONET measurements, J. Atmos. Sci., 59, 398– 413.
Torres, O.,
A. Tanskanen,
B. Veihelmann,
C. Ahn,
R. Braak,
P. K. Bhartia,
P. Veefkind, and
P. Levelt (2007), Aerosols and surface UV products from Ozone Monitoring Instrument observations: An overview, J. Geophys. Res., 112, D24S47, doi:10.1029/2007JD008809.
Torres, O.,
C. Ahn, and
Z. Chen (2013), Improvements to the OMI near UV aerosol algorithm using A‐train CALIOP and AIRS observations, Atmos. Meas. Tech. Discuss., 6, 5621– 565,652.
WCP (World Climate Programme) (1983), Report of the experts meeting on aerosols and their climatic effects, Williamsburg, Virginia, 28‐30 March 1983 (Edited by Deepak, A. and H. E. Gerber) WMO‐ICSU WCP‐55.
Yokota, T.,
H. Oguma,
I. Morino,
A. Higurashi,
T. Aoki, and
G. Inoue (2004), Test measurements by a BBM of the nadir‐looking SWIR FTS aboard GOSAT to monitor CO2 column density from space, Proc. SPIE, 5652, 182.
Yoshida, J.,
T. Kawashima,
J. Ishida,
K. Hamada,
J. Tanii,
Y. Katsuyama,
H. Suto,
A. Kuze,
M. Nakajima, and
T. Hamazaki (2008), Prelaunch performance test results of TANSO‐FTS and CAI on GOSAT, Proc. SPIE 7082.
Please check your email for instructions on resetting your password.
If you do not receive an email within 10 minutes, your email address may not be registered,
and you may need to create a new Wiley Online Library account.
Request Username
Can't sign in? Forgot your username?
Enter your email address below and we will send you your username
If the address matches an existing account you will receive an email with instructions to retrieve your username
(1)
(2)
(3)
(4)
(5)
(6)
(7)
instead of 
, because 670 nm observations are available from most recent satellite imagers (such as MODIS), whereas τ(1) is an unknown variable to be retrieved. Thus, we used 
in the present study to correct the remaining AOT in 
for better evaluation of Ag(1).
(8)
(9)
. Conversely, the MK method improves the comparison overall, but produced overestimations at Bonanza Creek and Laegeren and an underestimation at Paris. Retrieval of AOTs is generally more difficult in cities and areas of complex terrain owing to errors in the assumed ground reflectance and/or aerosol optical properties. It makes little difference for the sites with SSA larger than 0.95 whether using SSA 1.0 or the actual SSA in Table 1. On the other hand, the MK retrievals agree better with AERONET for the sites where SSA is less than 0.95 such as Paris, Sevastopol, Tomsk, and Sede Boker when using the actual SSA than when assuming SSA = 1.0. This is especially true at the Tomsk and Skukuza sites. These two sites can exhibit very high aerosol absorption optical depths [Kajii et al., 2002; Eck et al., 2003].
