Multiangle implementation of atmospheric ... - Wiley Online Library

15 downloads 478 Views 3MB Size Report
spectral regression coefficient (SRC) relating surface bidirectional reflectance in the blue ...... at https://neptune.gsfc.nasa.gov/bsb/index.php?section=101).
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 116, D03211, doi:10.1029/2010JD014986, 2011

Multiangle implementation of atmospheric correction (MAIAC): 2. Aerosol algorithm A. Lyapustin,1,2 Y. Wang,1,2 I. Laszlo,3 R. Kahn,4 S. Korkin,1,2 L. Remer,4 R. Levy,5 and J. S. Reid6 Received 31 August 2010; revised 4 November 2010; accepted 2 December 2010; published 11 February 2011.

[1] An aerosol component of a new multiangle implementation of atmospheric correction (MAIAC) algorithm is presented. MAIAC is a generic algorithm developed for the Moderate Resolution Imaging Spectroradiometer (MODIS), which performs aerosol retrievals and atmospheric correction over both dark vegetated surfaces and bright deserts based on a time series analysis and image‐based processing. The MAIAC look‐up tables explicitly include surface bidirectional reflectance. The aerosol algorithm derives the spectral regression coefficient (SRC) relating surface bidirectional reflectance in the blue (0.47 mm) and shortwave infrared (2.1 mm) bands; this quantity is prescribed in the MODIS operational Dark Target algorithm based on a parameterized formula. The MAIAC aerosol products include aerosol optical thickness and a fine‐mode fraction at resolution of 1 km. This high resolution, required in many applications such as air quality, brings new information about aerosol sources and, potentially, their strength. AERONET validation shows that the MAIAC and MOD04 algorithms have similar accuracy over dark and vegetated surfaces and that MAIAC generally improves accuracy over brighter surfaces due to the SRC retrieval and explicit bidirectional reflectance factor characterization, as demonstrated for several U.S. West Coast AERONET sites. Due to its generic nature and developed angular correction, MAIAC performs aerosol retrievals over bright deserts, as demonstrated for the Solar Village Aerosol Robotic Network (AERONET) site in Saudi Arabia. Citation: Lyapustin, A., Y. Wang, I. Laszlo, R. Kahn, S. Korkin, L. Remer, R. Levy, and J. S. Reid (2011), Multiangle implementation of atmospheric correction (MAIAC): 2. Aerosol algorithm, J. Geophys. Res., 116, D03211, doi:10.1029/2010JD014986.

1. Introduction [2] The Earth Observing System [King and Greenstone, 1999] initiated high‐quality global Earth observations and operational aerosol retrievals over land. With the wide‐swath (2300 km) MODIS instrument, the MODIS Dark Target [Kaufman et al., 1997; Remer et al., 2005; Levy et al., 2007] and Deep Blue algorithms [Hsu et al., 2004] provide a daily global view of planetary atmospheric aerosol loading. The MISR algorithm [Martonchik et al., 1998; Diner et al., 2005; Kahn et al., 2005] makes high‐quality aerosol retrievals in 380 km swaths covering the globe in 8 days.

1 Goddard Earth Sciences and Technology Center, University of Maryland Baltimore County, Baltimore, Maryland, USA. 2 NASA Goddard Space Flight Center, Greenbelt, Maryland, USA. 3 STAR, NESDIS, NOAA, Camp Springs, Maryland, USA. 4 Laboratory for Atmospheres, NASA Goddard Space Flight Center, Greenbelt, Maryland, USA. 5 Science Systems and Applications, Inc., Lanham, Maryland, USA. 6 Aerosol and Radiation Section, Marine Meteorology Division, Naval Research Laboratory, Monterey, California, USA.

Copyright 2011 by the American Geophysical Union. 0148‐0227/11/2010JD014986

[3] With the general success of the MODIS aerosol program, the issue of surface characterization remains high on the list of priorities. Unlike MISR, which uses multiangle observations for simultaneous aerosol‐surface retrievals, the current MODIS processing is pixel‐based and relies on a single‐orbit data. For every pixel, this approach produces a single measurement characterized by two main unknowns, aerosol optical thickness (AOT) and surface reflectance (SR). This lack of information constitutes a fundamental problem of the remote sensing that cannot be resolved without a priori information. For this reason, the MODIS Dark Target algorithm makes spectral assumptions about surface reflectance, whereas the Deep Blue method uses an ancillary global surface reflectance database. Both algorithms use a Lambertian surface model and apply an empirical correction for surface anisotropy. The aerosol product errors from the surface model have been well documented [e.g., Drury et al., 2008; Kahn et al., 2009; Wang et al., 2007; H. J. Hyer et al., An over‐land aerosol optical depth data set for data assimilation by filtering, correction, and aggregation of MODIS Collection 5 optical depth retrievals, submitted to Atmospheric Chemistry and Physics, 2010]. [4] This and a companion paper (A. Lyapustin et al., Multiangle implementation of atmospheric correction

D03211

1 of 15

D03211

LYAPUSTIN ET AL.: ALGORITHM MAIAC, 2

D03211

Figure 1. Block diagram of MAIAC algorithm. The initial capital letters indicate spatial and temporal domains of operations, for example, at pixel (P) or/and block (B) level, and using the data of the last Tile only (T) or using the full time series of the queue (Q).

(MAIAC): 3. Atmospheric correction, manuscript in preparation, 2011) present a new generic algorithm, multiangle implementation of atmospheric correction (MAIAC). It retrieves aerosol information over land simultaneously with parameters of a surface bidirectional reflectance factor (BRF) model using the time series of MODIS measurements and image‐based rather than pixel‐based processing. This approach ensures that the number of measurements exceeds the number of unknowns, a necessary condition for solving the generic problem. The time series accumulation also provides the multiangle coverage for every surface grid cell required for the BRF retrievals from MODIS measurements. [5] MAIAC has been in development over the past 5 years [Lyapustin and Wang, 2008, 2009]. Figure 1 gives an overview of MAIAC processing. It starts by gridding the received L1B data, splitting them into tiles, and placing the new data in the queue with the previous data. Given the selected projection, the gridding process translates sensor’s geolocated swath observations into grid cells of fixed latitude‐longitude coordinates. The queue implements a sliding window algorithm and holds from 5 (poles) to 16 (equator) days of imagery, depending on latitude. In order to limit variation of the footprint with changing view zenith angle (VZA), the resolution of MODIS 500 m channels B1–B7 is coarsened to 1 km. We use the MODIS land gridding algorithm [Wolfe et al., 1998] with minor modifications that allow us to better preserve the signal anisotropy in the gridded data when the measured reflectance is high, for example over snow, thick clouds or water with glint. The processing uses both 1 km grid cells, which are also called pixels below, and fixed 25 × 25 km2 areas called blocks. The blocks are used when the spatial imagery information is required, for example in the cloud detection or in the surface characterization part of the aerosol algorithm. [6] Next, the column water vapor is retrieved for the last tile using MODIS near‐IR channels B17–B19, located in the water vapor absorption band at 0.94 mm. This algorithm is a modified version of Gao and Kaufman [2003]. It is fast and has the average accuracy of ±5–10% over the land surface [Lyapustin and Wang, 2008]. The water vapor retrievals are

implemented internally to exclude dependence on other MODIS processing streams and unnecessary data transfers. [7] The MAIAC cloud mask (CM) and snow grain size algorithms were described earlier [Lyapustin et al., 2008, 2009]. In brief, the CM algorithm is based on the notion that the surface spatial pattern is stable and reproducible for short time periods under cloud‐free conditions, whereas clouds randomly disturb this pattern. The algorithm uses covariance analysis to identify cloud‐free regions. On this basis, it builds a reference clear‐sky image of the surface, which is used for pixel‐level cloud masking. The algorithm has an internal land‐water‐snow dynamic classification, which detects surface changes and guides the algorithm flow. The updated version of the MAIAC CM algorithm is described in the Algorithm Theoretical Basis Document [Lyapustin and Wang, 2008]. [8] Steps 4–6 of Figure 1 represent MAIAC aerosol algorithm which includes surface characterization (computation of spectral regression coefficient, SRC) in the MODIS blue band B3 (0.47 mm) and aerosol retrieval (optical thickness and fine mode fraction). Finally, the atmospheric correction (AC) algorithm uses the accumulated time series of MODIS data to derive parameters of the BRF model and surface albedo for each grid cell, as well as spectral BRF for the last observation. The BRF model parameters are stored in the memory and are used with the next MODIS observations to help cloud masking (surface change detection) and aerosol‐ surface retrievals. [9] The goal of current paper is to give a systematic description of MAIAC aerosol algorithm (the AC algorithm will be described separately (Lyapustin et al., manuscript in preparation, 2011)). Sections 2 and 3 of this paper describe the general idea of MAIAC and its implementation for the aerosol retrieval. The aerosol and AC MAIAC components are deeply interrelated. On one hand, aerosol retrievals rely on knowledge of the spectral surface BRF from the previous cycle of atmospheric correction. On the other hand, the current AC results depend on the aerosol retrievals. Validation analysis based on AERONET [Holben et al., 1998] measurements over dark and bright surfaces is presented

2 of 15

D03211

LYAPUSTIN ET AL.: ALGORITHM MAIAC, 2

Figure 2. A queue of K‐day blue band MODIS measurements for SRC retrievals. The image schematically shows a 25 × 25 km2 block of gridded geometrically corrected and calibrated (L1B) TOA data. in sections 4 and 5, respectively. Section 6 gives several examples illustrating large‐scale MAIAC performance. The paper concludes with a summary.

2. Statement of the Problem [10] The scales of global spatiotemporal variability of the Earth’s land surface and aerosols are significantly different. Indeed, the land surface is highly variable in space and changes little over short time intervals, whereas aerosol loading can change rapidly over time but vary spatially on scales of ∼50–60 km [Anderson et al., 2003]. These scale differences suggest that image‐based processing of the multiday MODIS data can be used for generic aerosol‐surface retrievals. [11] MAIAC defines the elementary processing area as a block with the side of N = 25 pixels (25 km), where variability of AOT is assumed to be small. If the surface BRF changes little during the accumulation period of K days, then the number of measurements in any given spectral channel exceeds the number of unknowns KN 2 > K þ 3N 2 if K > 3:

ð1Þ

Above, K is the number of AOT values for K different days, and 3 is the number of free parameters of the Ross–Thick Li–Sparse (RTLS) BRF model [Lucht et al., 2000]. Schematically, the accumulated data set is shown in Figure 2. [12] To simplify the inversion problem, MAIAC initially retrieves BRF in the shortwave infrared (SWIR) MODIS band B7 (2.1 mm), which is usually atmospherically transparent, and assumes that the BRF shape is similar between the SWIR and the blue bands of MODIS, 0:47 ð0 ; ; 8Þ ¼ b2:1 ð0 ; ; 8Þ:

ð2Þ

The MODIS Dark Target Algorithm uses an empirical parameterization of spectral regression coefficient (SRC) b,

D03211

whereas MAIAC retrieves this parameter for each 1 km grid cell. [13] The principle of spectral similarity of the BRF shape was extensively tested and implemented in the ATSR‐2 [Veefkind et al., 1998] and MISR [Diner et al., 2005] operational aerosol retrievals. This physically well‐based approach reduces the total number of unknown parameters to K+N 2. [14] The spectral similarity assumption (2) for the BRF shape works well for relatively dark surfaces, including vegetation and darker soils, for two main reasons. First, the surface absorption coefficient, or inversely, surface brightness, is similar in the visible and SWIR spectral regions, and second, the scale of macroscopic surface roughness, which defines shadowing, is much larger than the wavelength. If the difference in surface reflectance between the blue and the SWIR channels becomes large, e.g., over bright soils and deserts, the accuracy of approximation (2) deteriorates. The case of bright surfaces requires a special treatment and it will be addressed later in section 5. [15] Once SRCs are derived, the full surface BRF becomes known from equation (2) at a grid resolution of 1 km. At this step, the coarse 25 km resolution AOT is discarded, and the high‐resolution (1 km) aerosol retrievals are performed using the last day of MODIS measurements. [16] MAIAC is based on minimization of an objective function, so it directly controls the assumptions used (see section 3). For example, the objective function remains high if the surface has changed or if aerosol variability within the block was high on one of the days. Such days are identified and are excluded from processing. [17] From a historical prospective, MAIAC’s development was strongly influenced by the extensive heritage of the MISR and MODIS retrieval algorithms, from employing a rigorous radiative transfer model with a non‐Lambertian surface for simultaneous aerosol/surface retrievals [Diner et al., 1999, 2001] to the concept of using the image spatial structure for aerosol retrievals over land [Martonchik et al., 1998]. The last idea, with a different implementation, was proposed in the Contrast Reduction method by Tanre et al. [1988], who showed that consecutive images of the same surface area, acquired on different days, can be used to evaluate AOT differences between days.

3. Aerosol Algorithm [18] MAIAC retrievals are based on the look‐up tables (LUT) precomputed for a set of aerosol models. The detail of the LUT algorithm, which restores the top of atmosphere (TOA) reflectance for a given geometry, wavelength, AOT, water vapor and surface pressure, are provided in the companion paper [Lyapustin et al., 2011] which will be further referred to as part 1. Because MODIS provides only a spectral slice of information, MAIAC does not attempt MISR‐like retrievals for multiple aerosol models with different absorption and nonsphericity of particles [Kahn et al., 2010; Martonchik et al., 2009]. Instead, it uses regionally prescribed background aerosol models following the MODIS Dark Target approach [Levy et al., 2007]. In parts of the world systematically affected by mineral dust, MAIAC uses an additional regional dust aerosol model. The regional aerosol model is currently specified according to

3 of 15

D03211

D03211

LYAPUSTIN ET AL.: ALGORITHM MAIAC, 2 Table 1. Aerosol Model Parameters: Volumetric Radius Rv and Standard Deviation s (mm) for the Fine (F) and Coarse (C) Fractionsa

East Coast, USA West Coast, USA Dust, Solar Village

RFv

sF

RCv

sC

nre

nim

0.12 + 0.05t, ≤ 0.2 0.16 0.14

0.35 + 0.05t, ≤ 0.45 0.4 2.2

2.8 + 0.4t, ≤ 3.2 2.4 0.5

0.6 + 0.2t, ≤ 0.8 0.6 0.55

1.42 − 0.03t, 1.45 at t >1.2 1.45 1.50

0.0045, 0.002 0.0055, 0.002 0.0015, 0.0015

a

The real (nre) and imaginary (nim) refractive indices are common for both modes. The first (East Coast) is a dynamic model with parameters depending on aerosol optical thickness at 0.47 mm (t). The imaginary refractive index is assumed a constant at l 0.25, (b) >0.5, and (c) >0.85.

scatterplots in the red band (0.67 mm). Figure 6a shows results for four East Coast United States sites, including GSFC, Stennis, Walker Branch and Wallops. In this region, the MOD04 product historically has had a best performance. The overall agreement is good; the correlation coefficient (R2) exceeds 0.9, the slope of the regression line is within a few percent of unity, and the offset is near zero. Overall, the accuracy of MAIAC and MOD04 is very similar over the East coast, which has a significant amount of vegetation and

4.2. Angstrom Exponent [47] MAIAC’s aerosol optical thickness at 0.47 mm and a fine mode fraction define spectral dependence of AOT as well as single scattering albedo and phase function. The last row of Figures 6a and 6b shows correlation of computed MAIAC AOT in the red band (0.67 mm) with measured AERONET AOT0.67. Over the East Coast sites, correlation at 0.67 mm is almost as good as that at 0.47 mm, only the regression coefficient is insignificantly lower by 0.02–0.03. In general, this indicates a high quality of aerosol retrievals for the atmospheric correction of MODIS measurements. [48] On the other hand, a good independent agreement between MAIAC and AERONET AOT at two wavelengths does not automatically translates into agreement of an Angstrom parameter. Figure 7 shows a scatterplot of MAIAC‐ AERONET Angstrom parameter, which in both cases was computed using the blue and red wavelengths. The data are shown for the GSFC site which has the largest set of AERONET data and highest R2 for AOT. Three different plots shown in Figure 7 were obtained with increasing values of MAIAC AOT0.47, exceeding 0.25 (Figure 7a), 0.5 (Figure 7b) and 0.85 (Figure 7c). As expected, with growing atmospheric

9 of 15

D03211

LYAPUSTIN ET AL.: ALGORITHM MAIAC, 2

D03211

Figure 8a. (left) Scatterplot of MAIAC‐AERONET AOT and (right) a time series of MAIAC and AERONET AOT for the Solar Village site (Saudi Arabia). The data represent days 100–280 of 2003.

opacity from Figure 7a to Figure 7c the impact of surface‐ related errors decreases and correlation in the Angstrom parameters improves. This effect is observed for all East Coast sites considered here. On the other hand, the level of correlation is insignificant and usually it is less than 0.1. These numbers agree with statistics for Collection 5 MOD04 algorithm (R2 = 0.066 [see Levy et al., 2007, Figure 11b]). One needs to keep in mind that the variation of the size parameter over the East Coast, United States, is rather limited, whereas the MOD04 statistics was based on a large set of MODIS test granules including dust‐affected regions with much larger range of the size parameter and aerosol type variability. Currently, it can be concluded that the MAIAC Angstrom parameter is in the right range of values with an uncertainty comparable to that of MOD04 C5 algorithm.

5. Retrievals Over Bright Surfaces [49] MAIAC was developed as a generic algorithm designed to work over both dark and bright surfaces. It controls the assumptions used in the SRC retrievals with the exception of an assumption of spectral similarity of the BRF shape between the blue and SWIR spectral bands (equation (2)). This assumption does not hold when surface reflectance, and hence BRF shape, are very different between the blue and SWIR spectral regions. One obvious example is snow, which is very bright in the visible and dark at 2.1 mm. Due to significant multiple scattering of light inside snowpack at visible wavelengths and lack thereof at 2.1 mm, where snow is strongly absorbing, the anisotropy of snow BRF is significantly lower at 0.47 mm as compared to 2.1 mm [e.g., Hudson et al., 2006; Lyapustin et al., 2010]. Salt pans or playas, such as those found southwest of Salt Lake City, Utah, are in the same category; their spectral reflectance is similar to that of snow, and they cannot be treated using the current version of MAIAC. [50] The reflectance of bright deserts (sand) grows with wavelength in the VIS‐NIR spectral range, causing spectral dependence of the BRF shape. Figure 8a (left) shows a scatterplot of MAIAC‐AERONET AOT for the Solar Village AERONET site during 2003, starting from day 100. It shows scatter of ±0.25 for AOT 0 for backscattering angles, where BRF2.1 is higher (e.g., 0.55–0.6 for Solar Village), and D < 0 at forward scattering angles, where the BRF2.1 is lower (e.g., 0.4–0.45 for the Solar Village). Thus, for all available observation geometries in the queue, MAIAC first finds the minimal and maximal SWIR reflectance, typically representing the forward and backscattering directions, and evaluates D as a function of the current measurement r2.1,    max  av  max  min min D ¼  2:1  av 2:1 = 2:1  2:1 ; 2:1 ¼ 2:1 þ 2:1 =2: ð12Þ

[54] The amplitude d can be evaluated from the ASRVN data (0.1–0.15 in this case; see Figure 8b). The improved aerosol retrieval for Solar Village is shown in Figure 8c. The correction (12) reduced the bias by 0.05 and improved the correlation from 0.47 to 0.72. Figure 8c (right) shows the MAIAC‐AERONET AOT scatterplot for 5 years of MODIS Terra data with R2∼0.66. Currently, the correction (12) with d = 0.1 is used for all surfaces for which the maximum SWIR reflectance exceeds 0.2.

6. Examples of Large‐Scale Aerosol Retrievals [55] The performance of the MAIAC algorithm has been evaluated for large‐scale regions covering different parts of the world, and for one consecutive year of observations, or longer. Two examples of large‐scale AOT retrievals from MODIS TERRA are shown in Figures 9 and 10. Figure 9 shows smoke from biomass burning during the 2005 dry season over 1200 ×

Figure 8c. (left and middle) The same as Figure 8a but with angular correction. (right) MAIAC‐ AERONET AOT scatterplot for 5 years of MODIS Terra data (2002–2006). 11 of 15

D03211

LYAPUSTIN ET AL.: ALGORITHM MAIAC, 2

D03211

Figure 9. Fires during dry biomass burning season in Zambia, Africa, for day 205 of 2005 (area 1200 × 1200 km2). (left) The 1 km gridded MODIS TERRA TOA RGB image, (right) MAIAC AOT at 0.47 mm, and (inset at bottom right) MOD04. The high resolution (1 km) of MAIAC AOT allows detecting individual fire plumes. 1200 km2 in Zambia, Africa. The TOA image for day 205 shows dozens of small to large fires. The 1 km resolution of MAIAC makes it possible to resolve and trace individual fire plumes. Interestingly, these data also show the wind direction in the boundary layer. The fire plumes disappear at the coarse 10 km resolution of MOD04 shown in the inset. The comparison illustrates that the AOT magnitude and spatial distribution from MOD04 and MAIAC are similar, although there are certain differences depending on the surface type and viewing geometry. This example also demonstrates that the significantly higher spatial resolution of MAIAC offers new information about aerosol distribution, and possibly the strength of aerosol sources as well. The AOT gradient at 1 km resolution is high enough for automated smoke plume detection, so these data could be used for applications such as air quality monitoring. [56] Figures 10a and 10b give an example of MAIAC aerosol retrievals over southeast Asia and part of the bright Arabian Peninsula (3000 × 1500 km2). Figure 10a is the MAIAC RGB NBRF image for this area. Black pixels correspond to either open water or salt pans, where MAIAC is not applicable. The arrow points to an active dust storm source region in the Khash desert of southern Afghanistan. One such storm is traced in the MODIS Terra data through 8–11 August 2004 (Figure 10b). Figure 10b (top) shows MAIAC AOT, with

red color indicating optical depths above 1. The MODIS TOA RGB images at the bottom show the active storm area on 8–9 August, clearly visible in contrast with the background NBRF image. The storm intensity decreases on 10 August, with atmospheric dust covering a large area of Pakistan and the Indian Ocean. The storm largely abates on 11 August, leaving only two small areas active, as indicated by the arrows. At the same time, the winds moved the bulk of the remaining atmospheric dust westward. Figure 10b (bottom) shows another dust storm occurring between 11 and 13 September 2004, originating in Arabian Peninsula. The AOT image shows the epicenter and expanse of the storm. The ovals overlaying the AOT image mark an arm of the storm stretching across the Persian Gulf. This dust transport is visible in the MODIS RGB image over the dark water, providing qualitative confirmation of the MAIAC retrievals. A more detailed validation analysis of MAIAC results during the United Arab Emirates (UAE) 2004 campaign [Reid et al., 2008] will be given elsewhere, along with comparison to the MODIS Deep Blue algorithm.

7. Summary [57] This paper presents the aerosol component of a new MAIAC algorithm based on a time series analysis and image‐ based processing of MODIS data. MAIAC is a generic algo-

12 of 15

D03211

LYAPUSTIN ET AL.: ALGORITHM MAIAC, 2

Figure 10a. MAIAC NBRF RGB image of southeast Asia (area ∼3000 × 1500 km2). The arrow points at the area of Khash desert in southern Afghanistan.

Figure 10b. Dust storms in (top) southeast Asia and (bottom) Arabian peninsula in August–September of 2004. MAIAC AOT at 0.47 mm is shown at the top of each plot, and MODIS Terra RGB TOA images are shown at the bottom of each plot. 13 of 15

D03211

D03211

LYAPUSTIN ET AL.: ALGORITHM MAIAC, 2

rithm that performs aerosol retrievals and atmospheric correction over both dark vegetated surfaces and bright deserts with the current exceptions of bright salt pans and snow. The aerosol products include AOT and fine mode fraction at 1 km resolution. Following the MODIS operational aerosol algorithm (MOD04), the models of the fine and coarse aerosol fractions are assumed, and are specified regionally. [58] The aerosol algorithm derives the spectral regression coefficient (SRC), which relates surface BRF between the blue and SWIR bands. Due to this component, the aerosol retrievals can be made over both dark and bright surfaces at 1 km resolution. Once the SRC is computed, the surface BRF in the blue band is calculated from equation (2), and AOT0.47 is found by matching measured reflectance with a simulated value. Simultaneously, the fine mode fraction is evaluated from the best match to the measured reflectance in the red and SWIR channels, which, in turn, relies on the knowledge of spectral surface BRF from the previous cycle of atmospheric correction. For this process to be successful, the surface reflectance must be relatively stable over the 16 day time window. It works well over surfaces whose reflectance does not change much or when it changes relatively slowly over time, so that MAIAC is able to track the change through the atmospheric correction. On the other hand, high‐amplitude, rapid surface changes, such as rapid green‐up over bright soils, can cause systematic errors in both aerosol retrievals based on a previous knowledge of surface properties, and in the atmospheric correction. For this reason, the MAIAC AC algorithm has a change‐detection component and combines multiday BRF retrievals with the single‐day reflectance assessment. A detailed description of the AC algorithm and of its interrelation with aerosol algorithm will be given in the third paper of this series (Lyapustin et al., manuscript in preparation, 2011). [59] The above discussion reveals a complex mix of benefits and certain drawbacks of the time series approach. MAIAC removes the artificial spatial correlation between surface brightness and derived AOT (section 3.1), and quite dramatically improves quality of both aerosol retrievals and atmospheric correction by detecting thin or subpixel clouds (section 3.2). However, MAIAC may be prone to systematic errors when applied to periods of rapid change over vegetated bright soils, and can generate higher retrieval noise over brighter surfaces with its 1 km resolution than the 10 km MOD04 algorithm. The latter is explained by the factor of 8 variation of the MODIS footprint with the view angle; over heterogeneous surfaces, this causes higher spectral surface BRF uncertainty at 1 km resolution, which, in turn, increases aerosol retrieval uncertainty. [60] At the same time, AERONET validation shows that the MAIAC and MOD04 algorithms have similar accuracy over dark and vegetated surfaces, and that MAIAC generally improves accuracy over brighter surfaces due to the SRC retrieval and explicit BRF characterization, as demonstrated for several U.S. West Coast AERONET sites (section 4). Due to its generic nature and angular correction capability (section 5), MAIAC performs aerosol retrievals over bright deserts. The high‐resolution MAIAC aerosol product brings new information about distribution of aerosol sources, and potentially, their strength. These high‐resolution data are requested in different applications including air quality.

D03211

[61] A more extensive validation of MAIAC results and detailed comparisons with the MODIS operational Dark Target and Deep Blue algorithms will be given in the follow‐up papers. [62] Acknowledgments. The research of A. Lyapustin, Y. Wang, and S. Korkin was funded by the NASA Terrestrial Ecology Program (D. Wickland) and NASA Applications Program (L. Friedl and B. Doorn) and in part by the NOAA GOES‐R program (M. Goldberg). The work of R. Kahn is supported in part by NASA’s Climate and Radiation Research and Analysis Program, under H. Maring, NASA’s Atmospheric Composition Program, and the EOS‐MISR Project. The contribution of R. Levy and L. Remer to this work is supported by the NASA Radiation Science Program (H. Maring). J. Reid’s contribution was supported by the Office of Naval Research Code 322. This work strongly benefited from multiple discussions with our AERONET and NASA GSFC colleagues (A. Marshak, B. Holben, A. Sinuyk, A. Smirnov, I. Slutsker, and M. Sorokin). We are grateful to AERONET team for use of their data.

References Anderson, T. L., R. J. Charlson, D. M. Winker, J. A. Ogren, and K. Holmen (2003), Mesoscale variations of tropospheric aerosols, J. Atmos. Sci., 60, 119–136, doi:10.1175/1520-0469(2003)0602.0.CO;2 Diner, D. J., et al. (1999), MISR level 2 surface retrieval algorithm theoretical basis, Rev. D, JPL D‐11401, 104 pp., Jet Propul. Lab., Pasadena, Calif. Diner, D. J., et al. (2001), MISR level 2 aerosol retrieval algorithm theoretical basis, Rev. E, JPL D‐11400, 104 pp., Jet Propul. Lab., Pasadena, Calif. Diner, D. J., J. V. Martonchik, R. A. Kahn, B. Pinty, N. Gobron, D. L. Nelson, and B. N. Holben (2005), Using angular and spectral shape similarity constraints to improve MISR aerosol and surface retrievals over land, Remote Sens. Environ., 95, 155–171, doi:10.1016/j.rse.2004.09.009. Drury, E., D. J. Jacob, J. Wang, R. J. D. Spurr, and K. Chance (2008), Improved algorithm for MODIS satellite retrievals of aerosol optical depths over western North America, J. Geophys. Res., 113, D16204, doi:10.1029/2007JD009573. Dubovik, O., B. Holben, T. F. Eck, A. Smirnov, Y. J. Kaufman, M. D. King, D. Tanre, and I. Slutzker (2002), Variability of absorption and optical properties of key aerosol types observed in worldwide locations, J. Atmos. Sci., 59, 590–608, doi:10.1175/1520-0469(2002)059 2.0.CO;2. Gao, B. C., and Y. J. Kaufman (2003), Water vapor retrievals using Moderate Resolution Imaging Spectroradiometer (MODIS) near‐infrared channels, J. Geophys. Res., 108(D13), 4389, doi:10.1029/2002JD003023. Holben, B. N., et al. (1998), AERONET—A federated instrument network and data archive for aerosol characterization, Remote Sens. Environ., 66, 1–16, doi:10.1016/S0034-4257(98)00031-5. Hsu, N. C., et al. (2004), Aerosol properties over bright‐reflecting source regions, IEEE Trans. Geosci. Remote Sens., 42, 557–569, doi:10.1109/ TGRS.2004.824067. Hudson, S. R., S. G. Warren, R. E. Brandt, T. C. Grenfell, and D. Six (2006), Spectral bidirectional reflectance of Antarctic snow: Measurements and parameterization, J. Geophys. Res., 111, D18106, doi:10.1029/2006JD007290. Ichoku, C. D., A. Chu, S. Mattoo, Y. J. Kaufman, L. A. Remer, D. Tanre, I. Slutsker, and B. N. Holben (2002), A spatial‐temporal approach for global validation and analysis of MODIS aerosol products, Geophys. Res. Lett., 29(12), 8006, doi:10.1029/2001GL013206. Kahn, R. A., B. J. Gaitley, J. V. Martonchik, D. J. Diner, K. A. Crean, and B. Holben (2005), Multiangle Imaging Spectroradiometer (MISR) global aerosol optical depth validation based on 2 years of coincident Aerosol Robotic Network (AERONET) observations, J. Geophys. Res., 110, D10S04, doi:10.1029/2004JD004706. Kahn, R. A., et al. (2009), MISR aerosol product attributes and statistical comparisons with MODIS, IEEE Trans. Geosci. Remote Sens., 47, 4095–4114, doi:10.1109/TGRS.2009.2023115. Kahn, R. A., B. J. Gaitley, M. J. Garay, D. J. Diner, T. Eck, A. Smirnov, and B. N. Holben (2010), ultiangle Imaging SpectroRadiometer global aerosol product assessment by comparison with Aerosol Robotic Network, J. Geophys. Res., 115, D23209, doi:10.1029/2010JD014601. Kaufman, Y. J., D. Tanré, L. A. Remer, E. F. Vermote, A. Chu, and B. N. Holben (1997), Operational remote sensing of tropospheric aerosol over land from EOS moderate resolution imaging spectroradiometer, J. Geophys. Res., 102, 17,051–17,067, doi:10.1029/96JD03988.

14 of 15

D03211

LYAPUSTIN ET AL.: ALGORITHM MAIAC, 2

King, M., and R. Greenstone (Eds.) (1999), EOS Reference Handbook: A Guide to Earth Science Enterprise and the Earth Observing System, 355 pp., NASA GSFC, Greenbelt, Md. Levy, R. C., L. Remer, S. Mattoo, E. Vermote, and Y. J. Kaufman (2007), Second‐generation algorithm for retrieving aerosol properties over land from MODIS spectral reflectance, J. Geophys. Res., 112, D13211, doi:10.1029/ 2006JD007811. Lucht, W., C. B. Schaaf, and A. H. Strahler (2000), An algorithm for the retrieval of albedo from space using semiempirical BRDF models, IEEE Trans. Geosci. Remote Sens., 38, 977–998, doi:10.1109/36.841980. Lyapustin, A., and Y. Wang (2008), MAIAC—Multi‐angle implementation of atmospheric correction for MODIS: Algorithm theoretical basis document, v1.0, 78 pp., report, NASA GSFC, Greenbelt, Md. (Available at https://neptune.gsfc.nasa.gov/bsb/index.php?section=101) Lyapustin, A., and Y. Wang (2009), The time series technique for aerosol retrievals over land from MODIS, in Satellite Aerosol Remote Sensing Over Land, edited by A. Kokhanovky and G. De Leeuw, pp. 69–99, Springer Praxis, Berlin, doi:10.1007/978-3-540-69397-0_3. Lyapustin, A., D. Williams, B. Markham, J. Irons, B. Holben, and Y. Wang (2004), A method for unbiased high resolution aerosol retrieval from Landsat, J. Atmos. Sci., 61, 1233–1244, doi:10.1175/1520-0469(2004) 0612.0.CO;2. Lyapustin, A., Y. Wang, and R. Frey (2008), An automatic cloud mask algorithm based on time series of MODIS measurements, J. Geophys. Res., 113, D16207, doi:10.1029/2007JD009641. Lyapustin, A., M. Tedesco, Y. Wang, T. Aoki, M. Hori, and A. Kokhanovsky (2009), Retrieval of snow grain size over Greenland from MODIS, Remote Sens. Environ., 113, 1976–1987, doi:10.1016/j.rse.2009.05.008. Lyapustin, A., et al. (2010), Analysis of snow bidirectional reflectance from ARCTAS spring-2008 campaign, Atmos. Chem. Phys., 10, 4359–4375, doi:10.5194/acp-10-4359-2010. Lyapustin, A., J. V. Martonchik, Y. Wang, I. Laszlo, and S. Korkin (2011), Multiangle implementation of atmospheric correction (MAIAC): 1. Radiative transfer basis and look‐up tables, J. Geophys. Res., 116, D03210, doi:10.1029/2010JD014985. Martonchik, J. V., et al. (1998), Techniques for the retrieval of aerosol properties over land and ocean using multiangle imaging, IEEE Trans. Geosci. Remote Sens., 36, 1212–1227, doi:10.1109/36.701027. Martonchik, J. V., R. A. Kahn, and D. J. Diner (2009), Retrieval of aerosol properties over land using MISR observations, in Satellite Aerosol Remote Sensing Over Land, edited by A. A. Kokhanovsky and G. de Leeuw, pp. 267–293, Springer Praxis, Berlin, doi:10.1007/978-3-540-69397-0_9.

D03211

Reid, J. S., et al. (2008), An overview of UAE(2) flight operations: Observations of summertime atmospheric thermodynamic and aerosol profiles of the Southern Arabian Gulf, J. Geophys. Res., 113, D14213, doi:10.1029/ 2007JD009435. Remer, L. A., and Y. J. Kaufman (1998), Dynamic aerosol model: Urban/ Industrial aerosol, J. Geophys. Res., 103, 13,859–13,871, doi:10.1029/ 98JD00994. Remer, L., et al. (2005), The MODIS aerosol algorithm, products, and validation, J. Atmos. Sci., 62, 947–973, doi:10.1175/JAS3385.1. Smirnov, A., B. N. Holben, T. F. Eck, O. Dubovik, and I. Slutsker (2000), Cloud screening and quality control algorithms for the AERONET data base, Remote Sens. Environ., 73, 337–349, doi:10.1016/S0034-4257(00) 00109-7. Tanre, D., P. Y. Deschamps, C. Devaux, and M. Herman (1988), Estimation of Saharan aerosol optical thickness from blurring effect in Thematic Mapper data, J. Geophys. Res., 93, 15,955–15,964, doi:10.1029/JD093iD12p15955. Veefkind, J. P., G. de Leeuw, and P. Durkee (1998), Retrieval of aerosol optical depth over land using two‐angle view satellite radiometry during TARFOX, Geophys. Res. Lett., 25, 3135–3138, doi:10.1029/98GL02264. Wang, L., J. Xin, Y. Wang, Z. Li, G. Liu, and J. Li (2007), Evaluation of the MODIS aerosol optical depth retrieval over different ecosystems in China during EAST‐AIRE, Atmos. Environ., 41, 7138–7149, doi:10.1016/j. atmosenv.2007.05.001. Wang, Y., A. Lyapustin, J. L. Privette, J. T. Morisette, and B. Holben (2009), Atmospheric correction at AERONET sites: A new science and validation dataset, IEEE Trans. Geosci. Remote Sens., 47, 2450–2466, doi:10.1109/ TGRS.2009.2016334. Wolfe, R. E., D. P. Roy, and E. Vermote (1998), MODIS land data storage, gridding, and compositing methodology: Level 2 grid, IEEE Trans. Geosci. Remote Sens., 36, 1324–1338, doi:10.1109/36.701082. R. Kahn and L. Remer, Laboratory for Atmospheres, NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA. S. Korkin, A. Lyapustin, and Y. Wang, NASA Goddard Space Flight Center, MC 614.4, Greenbelt, MD 20771, USA. (alexei.i.lyapustin@ nasa.gov) I. Laszlo, STAR, NESDIS, NOAA, Camp Springs, MD 20746, USA. R. Levy, Science Systems and Applications, Inc., Lanham, MD 20706, USA. J. S. Reid, Aerosol and Radiation Section, Marine Meteorology Division, Naval Research Laboratory, Monterey, CA 93943, USA.

15 of 15