Algorithm for retrieval of aerosol optical properties over the ocean from the Geostationary Ocean Color Imager

Remote Sensing of Environment 114 (2010) 1077–1088

Contents lists available at ScienceDirect

Remote Sensing of Environment j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / r s e

Algorithm for retrieval of aerosol optical properties over the ocean from the Geostationary Ocean Color Imager Jaehwa Lee a,d, Jhoon Kim a,d,⁎, Chul H. Song b, Joo-Hyung Ryu c, Yu-Hwan Ahn c, C.K. Song e a

Institute of Earth, Astronomy, and Atmosphere, Brain Korea 21 Program, Department of Atmospheric Sciences, Yonsei University, Seoul, Republic of Korea Department of Environmental Engineering, GIST, Gwangju, Republic of Korea Ocean Satellite Research Group, Korea Ocean Research and Development Institute, Ansan, Republic of Korea d Aerosol and Cloud Group, NASA JPL, Pasadena, and JIFRESSE, UCLA, L.A., CA, USA e National Institute of Environmental Research, Inchon, Korea b c

a r t i c l e

i n f o

Article history: Received 8 April 2009 Received in revised form 28 December 2009 Accepted 29 December 2009 Keywords: Remote sensing Algorithm Aerosol optical depth Fine-mode fraction Aerosol type Geostationary

a b s t r a c t An aerosol retrieval algorithm for the first Geostationary Ocean Color Imager (GOCI) to be launched in March 2010 onboard the Communication, Ocean, and Meteorological Satellite (COMS) is presented. The algorithm retrieves aerosol optical depth (AOD), fine-mode fraction (FMF), and aerosol type in 500 m × 500 m resolution. All the products are retrieved over clear water which is defined by surface reflectance ratio between 640 nm and 860 nm (SRR) less or equal to 2.5, while only AOD is retrieved over turbid water (SRR N 2.5) due to high surface reflectance. To develop optimized algorithm for the target area of GOCI, optical properties of aerosol are analyzed from extensive observation of AERONET sunphotometers to generate lookup table. Surface reflectance of turbid water is determined from 30-day composite of Rayleighand gas corrected reflectance. By applying the present algorithm to MODIS top-of-the atmosphere reflectance, three different aerosol cases dominated by anthropogenic aerosol contains black carbon (BC), dust, and non-absorbing aerosol are analyzed to test the algorithm. The algorithm retrieves AOD, and size information together with aerosol type which are consistent with results inferred by RGB image in a qualitative way. The comparison of the retrieved AOD with those of MODIS collection 5 and AERONET sunphotometer observations shows reliable results. Especially, the application of turbid water algorithm significantly increases the accuracy in retrieving AOD at Anmyon station. The sensitivity study between MODIS and GOCI instruments in terms of relative sensitivity and scattering angle shows promising applicability of the present algorithm to future GOCI measurements. © 2010 Elsevier Inc. All rights reserved.

1. Introduction The Geostationary Ocean Color Imager (GOCI) is the first multichannel ocean color sensor in visible (VIS) and near infrared (NIR) from geostationary orbit onboard the Communication, Ocean, and Meteorological Satellite (COMS) planned to be launched in March 2010 to observe ocean color around the Korean Peninsula. The GOCI has eight spectral channels at 412, 443, 490, 555, 660, 680, 745, and 865 nm. The spatial coverage of GOCI extends to 2500 km × 2500 km, centered at 36°N and 130°E with 500 m resolution (Kang et al., 2006). By taking advantage of geostationary platform, GOCI can provide hourly spectral image which can be used for continuous monitoring of aerosols over cloud-free areas. Aerosol optical properties can be retrieved from reflectance measured by GOCI. There have been numerous studies to retrieve ⁎ Corresponding author. Institute of Earth, Astronomy, and Atmosphere, Brain Korea 21 Program, Department of Atmospheric Sciences, Yonsei University, Seoul, Republic of Korea. E-mail address: [email protected] (J. Kim). 0034-4257/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2009.12.021

aerosol optical properties from ‘GOCI-like’ sensors with multi-VIS/NIR channels aboard low Earth orbit (LEO) satellites, such as Sea-viewing Wide Field-of-view Sensor (SeaWiFS), Moderate Resolution Imaging Spectroradiometer (MODIS), and Medium-spectral Resolution Imaging Spectrometer (MERIS), etc. There have been studies to retrieve aerosol optical depth (AOD) by using single-channel (e.g. Kim et al., 2008; Knapp et al., 2002; Wang et al., 2003), and using dual-channels to retrieve both AOD and Angstrom exponent (AE) (e.g. Higurashi and Nakajima, 1999; Mishchenko et al., 1999). Higurashi and Nakajima (2002) developed aerosol type detection algorithm over ocean by using four channels of SeaWiFS at 412, 443, 670, and 865 nm. In their algorithm, not only AOD and AE but also aerosol type is inferred by taking advantage of four-channel information. Jeong and Li (2005) developed an aerosol classifying algorithm by utilizing Total Ozone Mapping Spectrometer (TOMS) and Advanced Very High Resolution Radiometer (AVHRR). The consistency of the aerosol type detection has been compared and evaluated by Kim et al. (2007) using the MODIS and OMI (Ozone Monitoring Instrument). From the MODIS, Remer et al. (2005) retrieved spectral AOD, fine-mode fraction (FMF), AE, and aerosol model which contain aerosol type

1078

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

information used in inversion by using seven channels from 490 to 2130 nm of MODIS over ocean. In this study, we present an aerosol retrieval algorithm that has been developed for the ocean color sensor, GOCI onboard the geostationary satellite, COMS. Lookup tables (LUT) are constructed based on extensive analysis of aerosol optical properties obtained from AERONET sunphotometer observations (Holben et al., 1998) over the East Asia. Details of algorithm are provided, and tested by using the MODIS data to retrieve aerosol optical properties. 2. Aerosol optical properties over the East Asia To investigate aerosol optical properties over the East Asia, inversion products from AERONET sunphotometer observations (Dubovik and King, 2000) are analyzed. Daily mean Level 1.5 cloud-screened products for three years from 2005 to 2007 are used to analyze aerosol optical properties. To ensure reliability of inversion products, only AOD greater than 0.4 at 550 nm are used. Three AERONET stations are selected by considering their location representativeness and data abundance: Beijing, Anmyon, and Shirahama stations, with large number of data, located at urban area of China, western coastal region of Korea, and southern coastal region of Japan, respectively. For aerosol retrieval algorithms, optical properties including aerosol size and single scattering albedo (SSA) are very important. Fig. 1 represents real part of refractive index (REFR hereafter) of aerosol over FMF and SSA space obtained from AERONET at Beijing, Anmyon, and Shirahama, respectively. The vertical lines intercepting FMF at 0.4 and 0.6 denote that from fine-mode AOD of bi-modal lognormal volume size distribution are 40% and 60% of the total, respectively. FMF less than 0.4 and greater than 0.6 represent coarse and fine-modedominant case, respectively. As shown in Fig. 1(a), aerosols at Beijing cover large portion of FMF versus SSA domain. It indicates that aerosols in Beijing originate from various sources from fine- to coarse-mode and from highly-absorbing to non-absorbing aerosols. For FMF less than 0.4, SSA is usually less than 0.95, which indicates that coarse-mode aerosols in this region absorb radiation at blue wavelengths. The upper envelope of data points representing decrease of SSA with decreasing FMF is largely due to strongly absorbing dust aerosols at wavelengths from ultraviolet (UV) to blue (cf. Dubovik et al., 2002; Sokolik et al., 1993). On the other hand, fine-mode aerosols have wide ranges of SSA values indicating mixed sources of anthropogenic aerosols in Beijing. High SSA values correspond to non-absorbing anthropogenic aerosols such as sulfate and nitrate, whereas low SSA values imply the presence of black carbon (BC). Both SSA and FMF of aerosols can be changed with the presence of water soluble particles and hygroscopic growth (Wang and Martin, 2007). One thing to note is that the REFR decreases with increasing FMF and SSA for fine-mode aerosols, approaching to the value of water, 1.33. This can be attributed to hygroscopic growth of water soluble aerosol, as relative humidity becomes very high. The lower value of REFR at low SSA in fine-mode can be explained by relatively high content of BC which has REFR of about 1.75 (Hess et al., 1998) mixed with water soluble aerosols. The aerosol optical properties at Anmyon and Shirahama are similar to those in Beijing except for the absence of extremely low FMF and SSA, and lower REFR, which can be explained by the location of these two sites at coastal region, thus are affected by maritime conditions. The upper envelope of data points show relatively high SSA at these locations and the REFR is low compared to the case of Beijing. The absence of extremely low SSA implies lower BC loading at these locations due to long distance from large cities and influence of

Fig. 1. Real part of refractive index over FMF and SSA space to represent aerosol optical properties from AERONET sunphotometer observations located at (a) Beijing, (b) Anmyon, and (c) Shirahama. The reference wavelengths used in the figure are 440 nm for refractive index and SSA, while it is 550 nm for FMF. Data with AOD at 550 nm greater than 0.4 are used for reliability of inversion products.

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

1079

Table 1 Refractive index of three fine-mode and two coarse-mode basic aerosol models and of turbid water aerosol model. HA, MA, and NA represent highly-absorbing, moderatelyabsorbing, and non-absorbing fine-mode aerosols, respectively. Aerosol model

Refractive index

400 nm

440 nm

675 nm

870 nm

HA

Real Imaginary Real Imaginary Real Imaginary Real Imaginary Real Imaginary Real Imaginary

1.46 0.018 1.45 0.010 1.44 0.0040 1.57 0.0031 1.36 0.001 1.47 0.011

1.46 0.018 1.45 0.010 1.44 0.0040 1.57 0.0028 1.36 0.001 1.47 0.011

1.47 0.017 1.46 0.011 1.45 0.0046 1.58 0.0012 1.36 0.001 1.49 0.010

1.48 0.018 1.46 0.012 1.45 0.0050 1.57 0.0011 1.36 0.001 1.49 0.010

MA NA Dust Sea salt Turbid water

maritime air mass for hygroscopic growth. The lower REFR is partly due to the effect of marine aerosol which shows small values (Dubovik et al., 2002). From the analysis of optical properties of aerosol from the AERONET inversion products over the East Asia, it is found that various types of aerosol result in very broad optical properties in this region. To improve the accuracy of aerosol retrieval from satellites, it is essential to take these optical properties into account, for example SSA and REFR. Especially, appropriate aerosol model is critical for high AOD case such as severe dust storm and wild fire events. Furthermore, aerosol type information is required to continuously monitor different aerosol types in this region. To define basic fine- and coarse-mode aerosols in calculating LUT, refractive index is obtained by averaging AERONET products with AOD N 0.4. For fine-mode aerosols, FMF N 0.8 with three different SSA ranges i.e. SSA N 0.95 (non-absorbing, NA), 0.90 b SSA ≤ 0.95 (moderately-absorbing, MA), and 0.85 b SSA ≤ 0.90 (highly-absorbing, HA) are used to define basic models. For coarse-mode aerosols, dust is defined by FMF b 0.2 with SSA b 0.95, and seasalt model is taken from Dubovik et al. (2002). For the retrieval of AOD over turbid water, aerosol model is obtained by averaging all AERONET inversions without constraint to represent mean state. The spectral refractive indices for basic aerosol models are summarized in Table 1. 3. Aerosol retrieval algorithm The important advantage of the GOCI lies in the measurements at number of visible channels in both high temporal and spatial resolutions from GEO, which enable us to retrieve hourly aerosol optical depth with size and chemical type information, for the first time. The disadvantage is the lack of IR channels to detect clouds and retrieve surface reflectance. Thus, clouds need to be detected either from the visible channels of GOCI or from the MI (Meteorological Imager) onboard the same satellite (e.g. Choi and Ho, 2009; Choi et al., 2007). Although the MI has IR channels to provide cloud information, its operation is not synchronized with GOCI to measure the same pixel at the same time, and furthermore its spatial resolution is 4 km in IR.

Fig. 2. Flowchart of aerosol remote sensing algorithm from GOCI.

Thus, considering the spatial and time scales of cloud imagery, clouds must be detected by visible channels of GOCI. Lack of IR channels in GOCI adds another difficulty in retrieving surface reflectance. In the aerosol algorithm of MODIS (Levy et al., 2007; Remer et al., 2005) for example, surface reflectance is retrieved using the 2.1 μm channel, which does not see the aerosol but is sensitive to surface reflectance. Thus, clear-sky composite method is adopted, which has been used widely for aerosol retrieval with limited channels from geostationary orbit (e.g. Kim et al., 2008; Knapp et al., 2002; Wang et al., 2003). Basically, LUT approach is adopted to retrieve AOD and FMF. A radiative transfer model (RTM) is used in calculating TOA reflectance at spectral channels of GOCI. In this study, Rstar5b RTM is used, which simulates radiation fields in the atmosphere–land–ocean system assuming plane parallel atmosphere (Nakajima and Tanaka, 1986). The dimensions of calculated LUT are summarized in Table 2. To test the algorithm, the closest channels of MODIS are used. The retrieved product has spatial resolution of 1 km × 1 km for MODIS proxy and 500 m × 500 m for real GOCI data. Fig. 2 represents the flowchart of the GOCI algorithm to retrieve aerosol optical properties. In the algorithm, cloud pixels are detected and masked out first since the accuracy of AOD retrieval is subject to cloud masking due to its strong signal. The algorithm adopts cloud detection procedure by using spatial variability and threshold tests used in the MODIS operational algorithm introduced by Remer et al. (2005) and Martins et al. (2002). The spatial variability test detects clouds by using a threshold for standard deviation of TOA reflectances in 3 × 3 pixels. Since the cloud top is inhomogeneous compared to aerosol layer or Rayleigh atmosphere over ocean, the algorithm efficiently detects clouds except for homogeneous clouds. In the

Table 2 Dimensions of LUT for absorbing and non-absorbing aerosol models. Variable name

No. of entries

Entries

Wavelength SZA SAZA RAA AOD FMF

6 (MODIS) 8 8 19 10 11

412, 443, 470, 555, 640, 860 nm 0, 10,…, 70 0, 10,…, 70 0, 10,…, 180 0.0, 0.1, 0.3, 0.6, 1.0, 1.5, 2.1, 2.8, 3.6, 5.0 0.0, 0.1,…, 1.0

SZA: solar zenith angle, SAZA: satellite zenith angle, RAA: relative azimuth angle.

Table 3 Aerosol type classification by using radiation absorptivity and FMF retrieved from the GOCI aerosol remote sensing algorithm. FMF

FMF N 0.7 FMF ≤ 0.7

Absorptivity Absorbing

Non-absorbing

BC (HA, MA) Dust

NA Sea salt

1080

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

reflection assuming wind speed of 6 m/s at 10 m height above sea level. Wind speed of 6 m/s is selected as mean wind speed over the ocean within the coverage. For the clear water case, AOD and FMF are retrieved simultaneously by comparing simulated- and observed TOA reflectances at 660 and 865 nm for 6 combinations of fine- and coarsemode (3 fine-mode multiplied by 2 coarse-mode). In this algorithm the AOD at 550 nm is retrieved because we simulated the spectral TOA reflectances with respect to AOD at 550 nm. After that, appropriate LUT group is selected by matching spectral TOA reflectance between observation and calculation at 412, 443, 490, and 555 nm. Since clear ocean surface is sufficiently dark at all GOCI channels, all of the rest channels, not used in the AOD retrieval, can be used for selecting proper LUT group by SSA. In this study, 412, 443, 490, and 555 nm of MODIS are used for selecting appropriate aerosol model. Since MODIS ocean channels at 680 and 745 nm are usually saturated for aerosol case due to their narrow dynamic range, they cannot be used to select LUT. Aerosol type is then derived from retrieved FMF and LUT group for AOD greater than 0.4 for the reliability of the products. The major criteria to classify aerosol type are summarized in Table 3. Compared to clear water case, only AOD is retrieved for turbid water by using 490 nm at which the reflectance shows higher sensitivity to aerosol than at the longer wavelengths (cf. Hsu et al., 2006) with turbid water aerosol model. 3.1. Detection of turbid water Fig. 3. (upper) Areal average of minimum cloud fraction (CF) and (lower) median of minimum AOD in 0.1° × 0.1° grid box for different time windows over the Yellow Sea. The data are from MODIS operational algorithm (“MYD04_L2”) in 10 km × 10 km resolution at nadir. For AOD, median instead of mean is used to exclude high AOD over turbid water.

MODIS operational algorithm, threshold of 0.0025 for the standard deviation of reflectance at 550 nm is used to separate cloudy from cloud-free pixels. The threshold test detects bright clouds with the reflectance threshold of 0.4 at 470 nm. After the cloud masking, turbid water is detected based on clearsky composite data over 30 day window, which is described in Section 3.1. Then, the surface reflectance is determined by minimum reflectance from 30-day composite of Rayleigh- and gas corrected reflectance for the turbid water. The reason for using second minimum reflectance instead of the first minimum is to avoid cloud shadows. For clear water, it is determined by modeling the Fresnel

Retrieval of aerosol optical properties over turbid water is important for GOCI target area since large portion of the Yellow Sea is covered by turbid water which is upwind region of the Korean Peninsula and important pathway of significant aerosol events from China. To detect turbid water and to determine its surface reflectance, clear-sky composite method (cf. Kim et al., 2008; Knapp et al., 2002) is adopted to take advantage of geostationary orbit. In applying the composite method, it is better to take search window over shorter time period in order to minimize the error, considering the variability of turbid water. However, high aerosol loading over the East Asia also interferes to determine surface reflectance accurately. To determine optimal composite window, areal average and median of minimum cloud fraction and AOD, respectively, are analyzed over the Yellow Sea for three different time windows as shown in Fig. 3. The areal average of minimum cloud fraction and median of minimum AOD for 20-day composite window show distinctly larger value compared with the

Fig. 4. Cumulative histogram of surface reflectance ratio (SRR) for 640 and 860 nm over (a) the Yellow Sea and (b) clear water (130°E–140°E in longitude, and 25°N–30°N in latitude) in 2006, calculated by using MODIS data. Each gray-colored layer from bottom to top represents each month in 2006 from January to December. Surface reflectance (SR) is determined by 30-day second minimum Rayleigh- and gas corrected TOA reflectance.

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

Fig. 5. (left) RGB image composed from surface reflectance and (right) turbid water detected by SRR method in March, June, September, and December in 2006.

1081

1082

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

results for longer time windows, however the results for 30-day window are similar to those for 40-day window except in June and July for cloud fraction and June for AOD. Since the area coverage of turbid water is smaller in summer in the Yellow Sea and composite window with shorter time is better to detect variability of turbid water, the 30-day is selected as optimal period for the composite window to detect turbid water and to determine surface reflectance. In this study, the turbid water is assumed to be brighter at 660 nm than at 865 nm due to more absorption by water at 865 nm and deeper penetration depth at 660 nm (Li et al., 2003). To detect turbid water in the Yellow Sea, histogram analysis is performed to find out the differences of spectral reflectance characteristics for turbid and clear water. The surface reflectance at 660 nm is larger than that at 865 nm since absorption by water increases with wavelength. Furthermore, the suspended particle matter (SPM) such as sediment increases the reflectance at 660 nm more than at 865 nm since water absorption is too strong to interact with SPM at 865 nm. As a result, the ratio of surface reflectance at 660 nm to 865 nm is larger for turbid water compared to the clear water case. Thus, threshold method can be used to separate turbid and clear water by histogram analysis. Fig. 4 compares cumulative histogram of ratio of surface reflectance (SRR hereafter) at 640 to 860 nm over the Yellow Sea (see Fig. 4 (a)) with those over clear, remote ocean (see Fig. 4(b)) for a year in 2006 calculated by using the MODIS data. Each gray-colored layer from bottom to top represents each month in 2006 from January to December. As expected, values of SRR in Fig. 4(a) and (b) are greater than 1 for almost all data points due to stronger water absorption at 860 nm. The distribution of SRR shown in Fig. 4(b) represents clear water scene, thus the difference between Fig. 4(a) and (b), that is the high SRR tails, is due to the turbid water. Thus, clear water pixels belong to SRR from 1.0 to 2.5. As SRR goes beyond 2.5 in Fig. 4(a), decreasing slope slows down with increasing SRR, thus the threshold value of 2.5 is selected to detect turbid water in this region. The turbid water detected by SRR threshold method is shown for different seasons in Fig. 5 with the corresponding RGB image. From the RGB image, it is evident that the turbid water is expanded in winter related to mixed layer breaking of the Yellow Sea by strong convection induced by cold air temperature. The algorithm detects turbid water region reasonably well especially reddish turbid water due to selected wavelengths in the histogram method. The undetected greenish region does not cause significant error in clear water algorithm to retrieve aerosol optical properties which use 660 and 865 nm channels. The current test in this study is using MODIS data which is onboard the LEO satellites. The composite method to determine surface reflectance by using GEO satellite has advantages compared to LEO satellite. First of all, hourly measurements can provide significantly larger number of data for composite image. In addition, at given local time, fixed viewing geometry of satellite for each observation point with slowly changing solar geometry can avoid anisotropic effect of surface caused by Bidirectional Reflectance Distribution Function (BRDF). Finally, fixed observation point can avoid errors caused during gridding process of data compared to those from LEO satellites. Thus, it is expected that surface reflectance database and turbid water detection can be improved when GOCI is in orbit. 3.2. Retrieval of aerosol optical properties As the algorithm adopts LUT approach to retrieve optical properties of aerosol, three fine-mode and two coarse-mode aerosol models are used in calculating LUTs as summarized in Table 1. Then, the LUT is constructed by the combination of fine and coarse-mode aerosols by changing FMF for respective aerosol model. Fig. 6 shows an example of LUT for the combination of nonabsorbing fine-mode and dust by considering the spectral response function of MODIS to apply the algorithm to the real case. Since

Fig. 6. An example of computed LUT by using rstar5b RTM for different aerosol loadings (AOD at 550 nm) and FMF assuming combination of non-absorbing fine-mode and dust aerosols.

channels 1 and 2 of MODIS centered at 640 and 860 nm, are broader compared with those of the ocean color sensor, these cover GOCI spectral channels to be used for aerosol retrieval which is centered at 660 and 865 nm. Thus, these can be used as proxy data of GOCI before launch. For a given geometry, TOA reflectances at 640 and 860 nm increase with AOD (τ) at 550 nm since aerosol increases the TOA reflectance over dark surface. The increment is larger at 640 nm than at 860 nm for fine-mode case since AOD is larger at the shorter wavelength. From the LUTs and TOA reflectance at 640 and 860 nm, AOD and FMF for combination of fine- and coarse-mode are retrieved simultaneously by iterative method. After that, calculated TOA reflectance at 412, 443, 490, and 555 nm (channels 3, 4, 8, and 9 for MODIS data) from retrieved AOD, FMF, and LUT are compared with those of observation to select appropriate aerosol models. Finally, the aerosol model minimizing the χ 2 parameter, defined as below, between modeled and observed TOA reflectances are selected. 2

n

χ = ∑

i=1

2

ðρi;m  ρi;o Þ ρi;o

ð1Þ

where i is the wavelength and ρm and ρo represent modeled and observed TOA reflectances, respectively. 4. Retrieval results To evaluate the performance of the developed algorithm, three different aerosol cases are selected and retrieval of aerosol optical properties is performed by using TOA reflectances from MODIS onboard Aqua satellite. Figs. 7–9 show RGB image, AOD, FMF, and aerosol type from the GOCI algorithm on January 27, April 8, and August 12 in 2006, respectively. On each day, dominant aerosol type is inferred as anthropogenic aerosol contains BC mixed with dust (Fig. 7), dust (Fig. 8), and non-absorbing anthropogenic (NA) aerosol (Fig. 9) from the respective RGB image. These aerosol types are shown as different colors in RGB images related to inherent optical properties, so that the type can be inferred by RGB image, especially over dark surfaces such as ocean. On January 27, thick aerosol layer with AOD up to 1.2 is observed over the Yellow Sea with grayish and brownish color in RGB image. The FMF represents the dominance of anthropogenic aerosol showing about 0.7 to 1.0, and absorptivity test shows dominant aerosol type to be anthropogenic aerosol containing BC (HA, MA) on this day. From

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

1083

Fig. 7. Retrieved products from GOCI aerosol remote sensing algorithm on January 27, 2006.

the FMF, the existence of dust as well as BC can be inferred consistently with analysis of RGB image. Fig. 8 shows dust outbreak case on April 8, 2006 from continents over the Korean Peninsula reaching to Japan with thick yellowish aerosol layer in the RGB image. The AOD increases over 3, and the FMF ranges about 0.5 to 0.6 inferring mixed case with coarse-mode. Aerosol type classification procedure results in thick dust type aerosol. Fig. 9 shows a case with whitish colored aerosol over the sea between Korea and Japan on August 12, 2006. Reduced fossil fuel combustion, high relative humidity and insolation in summer provide favorable condition to form non-absorbing anthropogenic aerosol over this region. The aerosol layer shows AOD of 0.8 and FMF of 0.9, that is, fine-mode dominance. In this case, aerosol type classification

shows somewhat erroneous result that HA is detected as well as NA over the aerosol layer. Since the sensitivity of TOA reflectance to aerosol type is less compared to that for AOD and possibly FMF, the aerosol type information should be used with caution. Further investigation is needed to validate and to improve aerosol type classification to maximize the ability of the GOCI instrument for continuous environmental monitoring. To estimate the accuracy of retrieved AOD from the GOCI algorithm, comparison of AOD from the GOCI algorithm with that from MODIS collection 5 and AERONET sunphotometer observations in 2006 are shown in Figs. 10 and 11, respectively. Recently, Remer et al. (2008) validated the collection 5 MODIS AOD products from Aqua satellite, which underestimated the AERONET values at 550 nm by

1084

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

Fig. 8. Same as in Fig. 7 except for the case on April 8, 2006.

10% with Pearson correlation coefficient of 0.907 over global ocean. The comparison result shows that performance of the GOCI algorithm is good with determination coefficient of 0.84 and regression slope of 1.17. The increase of standard deviation with AOD reflects the difference of assumed aerosol model in each algorithm which strongly affects retrieval results in high AOD condition, in particular. For the AERONET observations in Fig. 11 compared by using the method of Ichoku et al. (2002), the coefficient of determination is 0.51, 0.92, and 0.78 for Anmyon, Gosan, and Shirahama, respectively. Compared with other stations, the agreement at Anmyon station located close to the turbid water, i.e. yellow sea is poor, where the current algorithm overestimates AOD significantly. On the other hand, the algorithm produces reliable AOD product at Gosan and Shirahama, which are located near the clear ocean. This suggests that the effect of

turbid water should be considered. The statistics for AOD and FMF are compared in Tables 4 and 5, respectively, with those of MODIS collection 5. Since the MODIS operational algorithm masks out the turbid water before retrieval, the number of data at Anmyon station is small. The statistics shows better accuracy for the GOCI algorithm compared to the MODIS collection 5 over the coverage of GOCI except for the Anmyon station. Fig. 12 compares result of the GOCI algorithm after turbid water correction at Anmyon station. The determination of coefficient becomes better from 0.51 to 0.89 with increase of regression slope (from 0.69 to 1.09) and decrease of y-intercept (0.31 to 0.015). From this result, it is noteworthy that turbid water correction should be adopted over the Yellow Sea for better accuracy thus better coverage although this procedure may cause discontinuity at the boundary of turbid water.

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

1085

Fig. 9. Same as in Fig. 7 except for the case on August 12, 2006.

Since the algorithm is optimized for geostationary platform, especially turbid water algorithm, there are possibilities in improving the comparison results for real data after the launch of GOCI. Extensive inter-comparison among observed dataset of GOCI, AERONET and other satellites for long-term period should be carried out to diagnose error sources and to improve the algorithm in the near future. 5. Applicability and sensitivity comparison between GOCI and MODIS In this study, the performance of the GOCI algorithm is evaluated with the MODIS data. Thus, the sensitivity of the algorithm considering real GOCI data after launch should be investigated, in

terms of spectral response function and scattering angle. Fig. 13 shows spectral response function of GOCI and MODIS used in this study. The GOCI has eight channels centered at 412, 443, 490, 555, 660, 680, 745, and 865, whereas MODIS channels used in this study are centered at 412, 443, 470 (instead of 490), 555, 640 (instead of 660 and broader), and 860 nm. Since the main mission of GOCI focuses on ocean color retrieval, requirement specification of signal to noise ratio is very high ranging from 750 to 1200 (Kang et al., 2006) compared with MODIS aerosol channels (470, 555, 640, 860 nm). Thus, GOCI is expected to produce aerosol products at least comparable accuracy with those in this study after the effects of spectral response function and scattering angle are evaluated. The relative percentage sensitivity of the GOCI channels with reference to the MODIS is shown in Fig. 14. The sensitivity represents

1086

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088 Table 5 Same as in Table 4 except for the case of FMF. The data with AOD greater than 0.2 are used in this analysis. Station

Anmyon Gosan/ Shirahama

Fig. 10. Comparison of AOD retrieved from GOCI algorithm by using MODIS data with those of MODIS collection 5 over GOCI coverage in 2006. The dot and bar represent mean and standard deviation in each bin divided by same number of data. The regression line and determination coefficient are calculated by using full cloud of data.

GOCI algorithm

MODIS collection 5 algorithm

n

R2

Slope

y-intercept

n

R2

Slope

y-intercept

24 24

0.004 0.33

− 0.02 0.52

0.97 0.25

2 16

n/a 0.25

n/a 0.44

n/a 0.30

better by up to 8% than MODIS, whereas other channels show similar sensitivity. Fig. 15 compares frequency distribution of scattering angles in the observing geometry between GOCI and MODIS. MODIS covers wider scattering angle range due to its cross-track scanning observation. On the other hand, the scattering angle of GOCI is limited to narrower range due to small observation area, and the peak shifts toward backscattering direction compared with the case of MODIS. Note that the scattering angle of GOCI varies with observation time due to variation of solar angle. As shown in Fig. 16, the relative error of AOD from GOCI algorithm depends less on scattering angle compared with the MODIS. Especially, the relative errors are small where the frequency of scattering angles (130°–170°) for GOCI observations is high. From the comparison shown in Figs. 14–16, it is expected that

Fig. 11. Comparison of AOD retrieved from GOCI algorithm by using MODIS data with those of AERONET observation at Anmyon (without turbid water correction), Gosan, and Shirahama in 2006. The statistics at each location are summarized in Table 4.

difference in TOA reflectance between aerosol and aerosol-free case and the relative sensitivity is calculated by averaging LUT for difference in aerosol optical properties i.e. FMF and refractive index. Data with glint angle less or equal to 40° are excluded in this analysis. In the figure, relative percentage of 0% implies the same performance, while positive values indicate better performance based on calculations. The sensitivity of GOCI channels centered at 490 and 660 is

Fig. 12. Comparison of AOD retrieved from GOCI turbid water algorithm with those of AERONET observation at Anmyon.

Table 4 Statistics of AOD comparison between satellite algorithms and AERONET observation in 2006. n represents number of data used in the comparison. Station

Anmyon Gosan Shirahama

GOCI algorithm 2

n

R

39 31 59

0.51 0.92 0.78

MODIS collection 5 algorithm

Slope

y-intercept

n

R2

Slope

y-intercept

0.69 0.93 0.72

0.31 0.02 0.04

5 22 47

0.90 0.87 0.70

1.01 0.86 0.74

− 0.02 0.08 0.09

Fig. 13. Relative spectral response of GOCI and MODIS channels. Solid- and dashed lines represent those of GOCI and MODIS, respectively. GOCI channels are centered at 412, 443, 490, 555, 660, 680, 745, and 865 nm, whereas MODIS channels are centered at 412, 443, 470, 555, 640, and 860 nm.

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

1087

Fig. 14. Relative sensitivity of TOA reflectances for AOD of GOCI channels with reference to MODIS channels. The sensitivity represents TOA reflectance difference between aerosol and aerosol-free case and the relative sensitivity is calculated by averaging LUT for difference in aerosol optical properties i.e. FMF and refractive index. Data with glint angle less or equal to 40° are excluded in this analysis.

the developed algorithm is expected to perform well with the real GOCI data although it is tested by using data from MODIS. 6. Conclusion The main scientific objective of aerosol products from GOCI lies in environmental monitoring with its advantage of continuous observation capability as well as the assessment of climate change when the mission provide data for long term. The environment and climate

Fig. 16. Relative error of AOD retrieved from (a) GOCI algorithm and (b) MODIS collection 5 algorithm with respect to scattering angle at Gosan and Shirahama stations. In this comparison, cloud-masked mean reflectance data in 10 km × 10 km resolution are used for long-term retrieval for GOCI algorithm. Gray and black colors consist of all data point and AOD greater than 0.2, respectively.

Fig. 15. Frequency distribution of scattering angle of MODIS and GOCI over GOCI coverage. The frequency is calculated by using actual scattering angle for MODIS and by using computer code for sun- and satellite-geometry in 1° × 1° resolution in 2006. Each color for GOCI frequency distribution represents frequency at each observation time of GOCI.

of East Asia, in particular is affected by aerosols not only their large amount but also various types causing different impacts. The information on aerosol type with AOD based on continuous observation will be valuable in many fields, especially monitoring environment and climate change to investigate the effect of different aerosol types over the region. In this study, the aerosol retrieval algorithm of GOCI is described and its application capability is demonstrated by using MODIS TOA reflectances. The algorithm provides AOD and FMF, and derives aerosol type from retrieved FMF and absorptivity from inversion procedure over clear water and also retrieves AOD over turbid water. East Asia, one of the most polluted regions over the globe, is affected by aerosols with different optical properties from fine- to coarsemode as well as from strongly absorbing to non-absorbing. From this result, it is inferred that accurate inversion procedure to select appropriate aerosol model is essential to improve the accuracy of aerosol retrieval in this region. The LUT considers basic aerosol types

1088

J. Lee et al. / Remote Sensing of Environment 114 (2010) 1077–1088

of both fine- and coarse-mode and their combinations. As the retrieval of aerosol information from remote sensing is ill-posed problem, not only types of aerosol but also the surface type including turbid water should be considered carefully in the algorithm. It is noteworthy that large proportion of the Yellow Sea consists of turbid water over which the long range transport of aerosols from China occurs. The SRR threshold method by using the 30-day composite data adopted in the algorithm detects turbid water reasonably well when compared with RGB image. By applying the developed algorithm to the TOA reflectance of MODIS, the three different aerosol cases dominated by anthropogenic aerosol containing black carbon (BC), dust, and non-absorbing anthropogenic aerosol are analyzed to evaluate the performance of the algorithm. The algorithm reliably retrieves AOD, where size information and aerosol type are consistent with aerosol type inferred by RGB image in a qualitative way. Results from further quantitative inter-comparison for the detection of aerosol type are underway. The comparison of retrieved AOD with those of MODIS collection 5 and AERONET sunphotometer observations shows reliable results although the algorithm is developed for the geostationary. The relative sensitivity of GOCI with reference to MODIS and scattering angle dependence of retrieved AOD are compared and show promising results to apply the developed algorithm to real GOCI data. Since GOCI has higher signal to noise ratio, more channels for inversion, and small scattering angle dependence, the developed algorithm is expected to produce reliable results on hourly basis. The most prominent capability of GOCI lies in observation of aerosol information in higher resolution in temporal, spatial and spectral domains. The aerosol products from GOCI can be utilized in many ways, especially to monitor environment and climate change, and to assess the effect of aerosols in the East Asia, one of the most important regions in understanding pollution and climate change in global scale. The algorithm introduced in this study has been updated to improve its accuracy. The obtained results with the developed algorithm will be analyzed over long term in the future to diagnose various error sources in the algorithm. Acknowledgments We thank the Korean Ocean Research and Development Institute (KORDI) for the development and application of GOCI in this work. This work was funded by the Korea Meteorological Administration Research and Development Program under grant CATER 2006-3203. This research was partially supported by the Brain Korea 21 (BK21) program for J. Kim and J. Lee. We also thank the principal investigators and their staff for establishing and maintaining the AERONET sites. References Choi, Y. -S., & Ho, C. -H. (2009). Validation of the cloud property retrievals from the MTSAT-1R imagery using MODIS observations. International Journal of Remote Sensing, 30, 5935−5958. Choi, Y. -S., Ho, C. -H., Ahn, M. -H., & Kim, Y. -M. (2007). An exploratory study of cloud remote sensing capabilities of the Communication, Ocean and Meteorological Satellite (COMS) imagery. International Journal of Remote Sensing, 28, 4715−4732.

Dubovik, O., & King, M. D. (2000). A flexible inversion algorithm for retrieval of aerosol optical properties from Sun and sky radiance measurements. Journal of Geophysical Research, 105, 20,673−20,696. Dubovik, O., Holben, B. N., Eck, T. F., Smirnov, A., Kaufman, Y. J., King, M. D., et al. (2002). Variability of absorption and optical properties of key aerosol types observed in worldwide locations. Journal of Atmospheric Sciences, 59, 590−608. Hess, M., Koepke, P., & Schult, I. (1998). Optical properties of aerosols and clouds: The software package OPAC. Bulletin of the American Meteorological Society, 79, 831−844. Higurashi, A., & Nakajima, T. (1999). Development of a two-channel aerosol retrieval algorithm on a global scale using NOAA AVHRR. Journal of Atmospheric Sciences, 56, 924−941. Higurashi, A., & Nakajima, T. (2002). Detection of aerosol types over the East China Sea near Japan from four-channel satellite data. Geophysical Research Letters, 29(17), 1836. doi:10.1029/2002GL015357. Holben, B. N., Eck, T. F., Slutsker, I., Tanré, D., Buis, J. P., Setzer, A., et al. (1998). AERONET— A federated instrument network and data archive for aerosol characterization. Remote Sensing of Environment, 66, 1−16. Hsu, N. C., Tsay, S. C., King, M. D., & Herman, J. R. (2006). Deep blue retrievals of Asian aerosol properties during ACE-Asia. IEEE Transactions on Geoscience and Remote Sensing, 44(11), 3180−3195. Ichoku, C., Chu, D. A., Mattoo, S., Kaufman, Y. J., Remer, L. A., Tanré, D., et al. (2002). A spatio-temporal approach for global validation and analysis of MODIS aerosol products. Geophysical Research Letters, 29(12), 8006. doi:10.1029/2001GL013206. Jeong, M. -J., & Li, Z. (2005). Quality, compatibility, and synergy analyses of global aerosol products derived from the advanced very high resolution radiometer and Total Ozone Mapping Spectrometer. Journal of Geophysical Research, 110, D10S08. doi:10.1029/2004JD004647. Kang, G., Youn, H. S., Choi, S. B., & Coste, P. (2006). Radiometric calibration of COMS Geostationary Ocean Color Imager. Proceedings of SPIE, 6361, 636112. doi:10.1117/12.689888. Kim, J, Lee, J., Lee, H. C., Higurashi, A., Takemura, T., & Song, C. H. (2007). Consistency of the aerosol type classification from satellite remote sensing during the ABC EAREX campaign. Journal of Geophysical Research, 112, D22S33. doi:10.1029/2006JD008201. Kim, J., Yoon, J. -M., Ahn, M. H., Sohn, B. J., & Lim, H. S. (2008). Retrieving aerosol optical depth using visible and mid-IR channels from geostationary satellite MTSAT-1R. International Journal of Remote Sensing, 29(21), 6181−6192. Knapp, K. R., Vonder Haar, T. H., & Kaufman, Y. J. (2002). Aerosol optical depth retrieval from GOES-8: Uncertainty study and retrieval validation over South America. Journal of Geophysical Research, 107(D7), 4055. doi:10.1029/2001JD000505. Levy, R. C., Remer, L. A., Mattoo, S., Vermote, E. F., & Kaufman, Y. J. (2007). Secondgeneration operational algorithm: Retrieval of aerosol properties over land from inversion of Moderate Resolution Imaging Spectroradiometer spectral reflectance. Journal of Geophysical Research, 112, D13211. doi:10.1029/2006JD007811. Li, R. R., Kaufman, Y. J., Gao, B. -C., & Davis, C. O. (2003). Remote sensing of suspended sediments and shallow coastal waters. IEEE Transactions on Geoscience and Remote Sensing, 41(3), 559−566. Martins, J. V., Tanré, D., Remer, L. A., Kaufman, Y. J., Mattoo, S., & Levy, R. (2002). MODIS cloud screening for remote sensing of aerosol over oceans using spatial variability. Geophysical Research Letters, 29, 8009. doi:10.1029/2001GL013252. Mishchenko, M. I., Geogdzhayev, I. V., Cairns, B., Rossow, W. B., & Lacis, A. A. (1999). Aerosol retrievals over the ocean by use of channels 1 and 2 AVHRR data: sensitivity analysis and preliminary results. Applied Optics, 38(36), 7325−7341. Nakajima, T., & Tanaka, M. (1986). Matrix formulations for the transfer of solar radiation in a plane-parallel scattering atmosphere. Journal of Quantitative Spectroscopy & Radiative Transfer, 35, 13−21. Remer, L. A., Kaufman, Y. J., Tanré, D., Mattoo, S., Chu, D. A., Martins, J. V., et al. (2005). The MODIS aerosol algorithm, products and validation. Journal of Atmospheric Sciences, 62, 947−973. Remer, L. A., Kleidman, R. G., Levy, R. C., Kaufman, Y. J., Tanre´, D., Mattoo, S. J., et al. (2008). Global aerosol climatology from the MODIS satellite sensors. Journal of Geophysical Research, 113, D14S07. doi:10.1029/2007JD009661. Sokolik, I., Andronova, A., & Johnson, T. C. (1993). Complex refractive index of atmospheric dust aerosols. Atmospheric Environment, 27(A), 2495−2502. Wang, J., & Martin, S. T. (2007). Satellite characterization of urban aerosols: Importance of including hygroscopicity and mixing state in the retrieval algorithms. Journal of Geophysical Research, 112, D17203. doi:10.1029/2006JD008078. Wang, J., Christopher, S. A., Brechtel, F., Kim, J., Schmid, B., Redemann, J., et al. (2003). Geostationary satellite retrievals of aerosol optical thickness during ACE-Asia. Journal of Geophysical Research, 108, 1−14.