• Open Access

Suomi-NPP VIIRS aerosol algorithms and data products


  • John M. Jackson,

    Corresponding author
    1. Northrop Grumman Aerospace Systems, Redondo Beach, California, USA
    • Corresponding author: J. M. Jackson, Northrop Grumman Aerospace Systems, One Space Park, Redondo Beach, CA 90278, USA. (sid.jackson@ngc.com)

    Search for more papers by this author
  • Hongqing Liu,

    1. I. M. Systems Group, Inc., College Park, Maryland, USA
    2. Center for Satellite Applications and Research, National Environmental Satellite, Data, and Information Service, National Oceanic and Atmospheric Administration, College Park, Maryland, USA
    Search for more papers by this author
  • Istvan Laszlo,

    1. Center for Satellite Applications and Research, National Environmental Satellite, Data, and Information Service, National Oceanic and Atmospheric Administration, College Park, Maryland, USA
    2. Department of Atmospheric and Oceanic Science, University of Maryland, College Park, Maryland, USA
    Search for more papers by this author
  • Shobha Kondragunta,

    1. Center for Satellite Applications and Research, National Environmental Satellite, Data, and Information Service, National Oceanic and Atmospheric Administration, College Park, Maryland, USA
    Search for more papers by this author
  • Lorraine A. Remer,

    1. Joint Center for Earth Systems Technology, University of Maryland, Baltimore County, Baltimore, Maryland, USA
    Search for more papers by this author
  • Jingfeng Huang,

    1. Center for Satellite Applications and Research, National Environmental Satellite, Data, and Information Service, National Oceanic and Atmospheric Administration, College Park, Maryland, USA
    2. Earth System Science Interdisciplinary Center, University of Maryland, College Park, Maryland, USA
    Search for more papers by this author
  • Ho-Chun Huang

    1. Center for Satellite Applications and Research, National Environmental Satellite, Data, and Information Service, National Oceanic and Atmospheric Administration, College Park, Maryland, USA
    2. Earth System Science Interdisciplinary Center, University of Maryland, College Park, Maryland, USA
    Search for more papers by this author


[1] The Visible Infrared Imaging Radiometer Suite (VIIRS) instrument on board the Suomi National Polar-orbiting Partnership (S-NPP) spacecraft was launched in October 2011. The instrument has 22 spectral channels with band centers from 412 nm to 12,050 nm. The VIIRS aerosol data products are derived primarily from the radiometric channels covering the visible through the short-wave infrared spectral regions (412 nm to 2250 nm). The major components of the VIIRS aerosol retrieval process are data screening, land inversion, ocean inversion, suspended matter typing, and aggregation. The primary data product produced is the aerosol optical thickness (AOT) environmental data record. A higher resolution AOT intermediate product is also produced. These AOT products and their corresponding retrieval algorithms are described in detail, including theoretical basis, retrieval limitations, and data quality flagging. Preliminary evaluation of the data products has been undertaken by the VIIRS aerosol calibration/validation team using Aerosol Robotic Network ground-based observations to show that the performance of AOT retrievals meets the requirements specified in the Joint Polar Satellite System Level 1 requirements.

1 Introduction

[2] We live in a time of increasing awareness of the environmental challenges facing our present and future [Solomon et al., 2007; Karl et al., 2009; Biermann, 2012; Smith, 2013], and this awareness is coupled with an increasing commitment to provide the tools necessary to obtain the information to understand and meet those challenges [Anthes et al., 2007; Hartmann et al., 2012; Karl et al., 2009; Trenberth et al., 2013]. One important environmental challenge is to understand the role of atmospheric aerosols in meteorological processes on a broad spectrum of both spatial and temporal scales from weather to climate processes and in the degradation of air quality [Charlson and Pilat, 1969; Twomey, 1977; Rosenfeld and Lensky, 1998; Pope et al., 2002; Lau et al., 2006; Yu et al., 2006; Chin et al., 2007, 2009]. These small suspended liquid and solid particles in our atmosphere such as mineral dust, smoke, volcanic ash, particulate pollution, sea salt, and biogenic compounds play an active role in the Earth's energy balance [Chin et al., 2009], hydrological cycle [Koren and Feingold, 2011; Koren et al., 2012], and atmospheric chemistry [Martin et al., 2003]. At the same time, these particles can cause ill health ffects when breathed into the lungs [Pope et al., 2002] and degrade visibility [Hand et al., 2011], which can interfere with military field operations and civilian aviation safety [Schafer et al., 2004]. Because these particles and their precursors are emitted into the atmosphere from a variety of sources, undergo chemical transformation as they are transported globally, and are removed from the atmosphere by various processes on the order of days to weeks, their compositions and distributions are characterized by a high level of temporal and spatial variability [Kaufman et al., 2002; Chin et al., 2009].

[3] A global view of the aerosol system can be best achieved by spaceborne observations that complement not only the more detailed but also temporally and spatially limited ground-based and airborne observations [Diner et al., 2004]. There has been a long history of retrieving aerosol from space that began with localized attempts using Landsat sensors [Griggs, 1975; Kaufman and Sendra, 1988]. Following these first attempts, a more global perspective began to take shape using a variety of sensors and a variety of techniques (King et al. [1999] and Li et al. [2009] for reviews). The different sensors and techniques exhibit different strengths and weaknesses. For example, Dark Target methods that follow from the earlier Landsat attempts work best over dark ocean [Rao et al., 1989] or dark vegetated surfaces [Kaufman and Sendra, 1988], while Deep Blue or ultraviolet methods [Hsu et al., 1996, 2006] provide accurate retrievals over bright deserts, but require information on aerosol layer height, which is not always available. Other trade-offs exist between different wavelength ranges, single- or multiple-angle views of the scene, the availability of polarization, passive and active measurements, spatial resolution, etc., (again Li et al. [2009], for review).

[4] One such long-term global aerosol data set based on Dark Target methods has been produced for over a dozen years from the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the NASA Earth Observing System (EOS) satellites Terra and Aqua [Kaufman et al., 1997, 2005; Remer et al., 2005; Levy et al., 2007, 2013]. The new algorithm described in this paper, currently applied to the Visible Infrared Imaging Radiometer Suite (VIIRS) sensor onboard the Suomi National Polar-orbiting Partnership (S-NPP) satellite, is based on the MODIS Dark Target algorithm. Thus, it enjoys many of the same strengths and suffers from many of the same weaknesses as its MODIS predecessor. S-NPP was launched into Sun-synchronous orbit with a 13:30 Local Time of Ascending Node in October 2011. Like MODIS, the S-NPP VIIRS provides a key set of aerosol products based on daily global observations from space. These products are designed to match the precedent established by NASA's EOS MODIS and are expected to be used by the climate, air quality, research, applied, private, governmental, and military communities for their research, forecasts, communication, and policy decisions.

[5] General information about the VIIRS sensor and the aerosol products are presented in sections 2 and 3, respectively. As a focus of this paper, sections 4 and 5 describe the algorithms used for retrieving the aerosol optical thickness. Section 6 discusses how pixels appropriate for aerosol retrieval are selected and how an indication of the quality is assigned to the retrieved aerosol data. Section 7 describes how the environmental data records are constructed from the individual pixel retrievals through aggregation. Some validation results from the comparisons of the VIIRS aerosol optical thickness against ground measurements are presented in section 8. For the vast majority of potential readers who are familiar with the MODIS aerosol product, a brief comparison of the major features of the VIIRS and MODIS products and algorithms are provided in section 9. Section 10 gives a summary and some concluding remarks.

2 VIIRS Sensor

[6] VIIRS is a cross-track scanning radiometer sensor that measures reflected and emitted radiation from the Earth-atmosphere system in 22 spectral bands, spanning from 412 nm to 12,050 nm. It also features dual gain bands which allow a greater dynamic range while retaining high signal-to-noise ratio (SNR) at low radiance values making the bands usable for land, ocean, and atmospheric applications [Cao et al., 2013]. VIIRS provides single-angle observation and does not measure polarization. It has a wide swath (~3000 km), which allows it to fully sample the Earth every day (Figure 1). An on-board pixel trimming algorithm eliminates redundant views of the same Earth scenes, which mitigates the “bowtie” effect at swath edges [Cao et al., 2013]. In addition, the pixel aggregation and geometric strategy reduces the growth of the pixel sizes toward the swath edge. VIIRS pixel ground-projected instantaneous field of view (GIFOV) grows by approximately a factor of 4 from nadir to edge of scan, in contrast to the eightfold increase of the MODIS GIFOV [Cao et al., 2013]. VIIRS has three types of bands: imagery bands, moderate resolution bands (M-bands), and the day-night band. The spatial resolution of a VIIRS observation depends on the VIIRS bands used. The M-bands, many of which are used in the aerosol retrieval, have 0.742 × 0.776 km nadir resolution and 1.60 × 1.58 km at the edge of scan. Table 1 summarizes the 16 VIIRS M-bands. The reflective M-bands (M1, M2, M3, M5, M6, M7, M8, M10, and M11) are used by the VIIRS aerosol algorithm for aerosol retrievals. Bands M4, M9, M12, M15, and M16 are used in the internal screening tests for data quality assurance. Some of these VIIRS bands (M1–M5, M7) have dual gains which allow a greater dynamic range while retaining high signal-to-noise ratio (SNR) at low radiance values making the bands usable for land, ocean, and atmospheric applications [Cao et al., 2013].

Figure 1.

Global true color (RGB) image from VIIRS daytime granules on 22 May 2013 (red channel: M5; green channel: M4; blue channel: M3).

Table 1. Characteristics of the 16 VIIRS M-Bands and Their Use in the VIIRS Aerosol Retrievals
Band NameWavelength (nm)Bandwidth (nm)Use in Aerosol Algorithm
  1. aDual gain bands.
M1a41220Inversion over land
M2a44514Inversion over land
M3a48819Inversion over land Internal tests land and ocean
M4a55521Internal tests ocean
M5a67220Inversion over land and ocean Internal tests land
M674615Inversion over ocean
M7a86539Inversion over ocean Internal tests land
M81,24027Inversion over ocean Internal tests land and ocean
M91,37815Internal tests land
M101,61059Inversion over ocean Internal tests land and ocean
M112,25047Inversion over land and ocean Internal tests land and ocean
M123,700191Internal tests land
M1510,763989Internal tests land and ocean
M1612,016864Internal tests land and ocean

[7] The signal-to-noise ratio (SNR) of these bands is high with large margin versus design specifications, ~200–400 for bands M1–M7 and ~10–300 for bands M8–M11. All bands are expected to meet SNR design specifications throughout the lifetime of the instrument (~7 years) even after considering the post–launch degradation that was caused by the presence of tungsten oxides on the surfaces of the mirrors of the rotating telescope assembly. The exposure of this contaminant to solar ultraviolet light causes a progressive darkening throughout the visible, near-infrared, and short-wave infrared wavelengths [Cao et al., 2013]. This time-dependent change in the telescope throughput is handled in the sensor data record (SDR) radiometric calibration algorithm; therefore, the impact on the downstream environmental data records (EDRs) is insignificant. All of the VIIRS reflective solar bands meet or exceed the absolute radiometric accuracy requirement of 2% [Cao et al., 2013]. More information on the uncertainties in the measured reflectances in the VIIRS sensor data records (SDRs) can be found in Cao et al. [2013].

3 VIIRS Aerosol Products and Availability

[8] The raw VIIRS measurements of reflected and thermal radiances are presented as dimensionless “counts” and stored in raw data records (RDRs). The RDRs are processed with calibration and geolocation information and are presented with engineering units of radiance and quality flags as sensor data records (SDRs). SDRs are used to derive geophysical parameters including aerosol optical thickness. Processing of SDRs for aerosol retrieval is done on a granule by granule basis. One VIIRS granule typically consists of 768 × 3200 (along track by cross track) 0.75 km pixels. Aerosol parameters are derived from the M-band SDRs in the 412 to 2250 nm range. Other bands are used to create the VIIRS Cloud Mask (VCM) [Baker, 2013], which is used as input to aerosol algorithms, as well as in internal tests to characterize environmental conditions. The M-bands used to derive optical thickness are all within window regions, and their bandwidths are narrow to minimize gas absorption (e.g., O2, O3, and H2O).

[9] VIIRS aerosol retrievals are performed at the M-band pixel level and produce a full set of aerosol parameters called the intermediate product (IP). It consists of aerosol optical thickness at 550 nm, Ångström exponent, and aerosol model information including a single-aerosol model selected over land and three parameters retrieved over ocean. The three parameters of fine mode index, coarse mode index, and fine mode fraction define the bimode aerosol mixture. The pixel resolution IP provides critical information to downstream algorithms in the VIIRS processing and can be useful in its own right by providing publicly available aerosol products at higher resolution. However, caution should be exercised by understanding the limitations of the higher-resolution products and utilizing associated quality flags.

[10] The environmental data record (EDR) is the official level 2 product of VIIRS. There are three aerosol EDRs: aerosol optical thickness (AOT), aerosol particle size parameter (APSP), and suspended matter (SM). The AOT and APSP EDRs are generated by aggregating 8 × 8 IP retrievals, which leads to 96 × 400 (along track by cross track) EDR cells with the resolution of ~6 × 6 km at nadir (~12.8 × 12.8 km at the edge of scan) in a typical VIIRS granule. The VIIRS AOT EDR consists of aerosol optical thickness reported at 10 M-Band wavelengths from 412 nm to 2250 nm as well as at 550 nm. The APSP EDR is reported as the Ångström exponent calculated from AOTs at 445 nm (M2) and 672 nm (M5) over land and from 865 nm (M7) to 1610 nm (M10) over ocean. The SM EDR is provided at the M-band pixel resolution and includes a classification of the aerosol type and smoke concentration. The categories of aerosol types consist of volcanic ash, dust, smoke, and sea salt, which are either inferred from the IP aerosol model information or directly read from the VIIRS Cloud Mask (volcanic ash). All of the VIIRS aerosol products are only reported during daytime and only over dark land and nonsunglint ocean surfaces. All three EDRs include quality flags to enable users to better utilize VIIRS products. The high-quality EDR retrievals, once fully validated, are the only retrievals with specific performance requirements defined in the Joint Polar Satellite System (JPSS) Level 1 requirements [Kennedy et al., 2013].

[11] In this paper, only the AOT product and the algorithms used to retrieve it are discussed. It has been shown that success of reliable Ångström exponent retrieval from multispectral satellite measurements alone like MODIS and VIIRS is limited, especially over land [Levy et al., 2010]. Analysis of the VIIRS APSP product has also shown the lack of value of this parameter over land and having only a limited value over ocean. Thus, these products will not be addressed further. The suspended matter EDR is also beyond the scope of this paper. Further information on all VIIRS aerosol products can be found in the VIIRS aerosol optical thickness and particle size parameter algorithm theoretical basis document [Baker, 2011b] and the VIIRS suspended matter algorithm theoretical basis document [Baker, 2011b].

[12] An example global image of the VIIRS high-quality AOT EDR is shown in Figure 2. Comparison of this to the red-green-blue (RGB) image in Figure 1 shows where cloud, sunglint, and bright surface prevented aerosol retrieval.

Figure 2.

Global high-quality EDR aerosol optical thickness at 550 nm on 22 May 2013.

[13] The VIIRS aerosol EDRs are available at the National Oceanic and Atmospheric (NOAA) Comprehensive Large Array-data Stewardship System (CLASS) at http://www.nsof.class.noaa.gov. A Beta maturity level AOT and APSP product is available at this site to the public starting from 2 May 2012. Users should not use the data between 15 October and 27 November 2012 due to a processing error (H. Liu et al., manuscript in preparation, 2013). The Beta maturity product is an early release product with minimal validation. It is provided to users to gain familiarity with data formats and parameters, but it is generally not recommended for scientific studies and applications. The Beta AOT product has a known significant bias (compared to Aerosol Robotic Network (AERONET) and MODIS AOT) over land. An improved product, at provisional maturity level, with a much reduced global bias is available starting from 23 January 2013. Although product quality may still not be optimal, and improvements are still occurring, the research community is encouraged to participate in the evaluation of the products at provisional maturity.

4 VIIRS Aerosol Algorithms

[14] The science version of the VIIRS aerosol algorithm was initially developed by Raytheon and updated and extensively modified by Northrop Grumman Aerospace Systems over several years before the launch of the S-NPP satellite. Because of the strong similarities between the sensors and the long history of MODIS in the aerosol community, the VIIRS aerosol algorithm is based on the MODIS heritage. The algorithm started out very similar to the MODIS collection 3 [Kaufman et al., 1997; Tanré et al., 1997] aerosol algorithm. As more lessons were learned from MODIS over the years, the features of the VIIRS algorithm also changed over time. But not all of the lessons gained from MODIS and other satellite-based aerosol retrievals have been incorporated (e.g., the realization of a lack of sensitivity to particle size over land [Levy et al., 2010]). The changes introduced into the VIIRS algorithm mostly affected the retrieval over land. As a result, the current VIIRS aerosol algorithm over land is significantly different from the MODIS overland aerosol algorithm. The over-ocean aerosol algorithm also involves some differences with its MODIS counterpart. A summary of the similarities and differences is provided in section 9.

[15] The requirements of the VIIRS aerosol algorithm also changed over the years. Certain choices made along the line of development are currently being reevaluated by the JPSS VIIRS aerosol calibration/validation team (e.g., the choice of optical thickness range considered and the choice of the wavelength pair to calculate the Ångström exponent over ocean), and changes may be implemented in a future update of the algorithm. Therefore, this section describes the algorithm in its present form (software version I1.5.07.01).

[16] The VIIRS aerosol algorithm simultaneously retrieves the AOT and aerosol model from reflected solar radiation measured in multiple visible and near to shortwave IR channels. In this process, by iterating through increasing values of AOT and various candidate aerosol models, the algorithm searches for the optimal solution that best matches the reflectance measurements to the theoretical values. Over ocean, the goodness of fit is judged by the closeness between computed and measured top-of-atmosphere (TOA) reflectance in the selected channels. Over land, it is the comparison between the derived and expected spectral surface reflectance that determines the best solution.

[17] To compute the reflectances needed in the retrieval, the VIIRS algorithm follows the approximation used in the vector radiation transfer model (RTM) Second Simulation of the Satellite Signal in the Solar Spectrum, Vector version 1.1 (6S-V1.1) [Kotchenova et al., 2006; Kotchenova and Vermote, 2007; Vermote et al., 2006]. According to 6S-V1.1 RTM, the spectral reflectance measured at satellite level ρtoa(τA) is

display math(1)

where τA is the aerosol optical thickness and θs and θv are the solar and view zenith angles, respectively. P is the surface pressure. ρR and ρA are the atmospheric path reflectances due to molecular (Rayleigh) and aerosol scattering, respectively. The quantity TR + A is the total (direct + diffuse) transmittance due to molecules and aerosols. inline image and inline image are the two-way (down and up) ozone and water vapor transmittances, with inline image representing the total column amount of water vapor. Tgog is the two-way transmittance due to gases other than water vapor and ozone (O2, CO2, CH4, and N2O). ρsurf is the Lambertian surface reflectance, while SR + A is the spherical albedo of the atmosphere (molecules and aerosols). Contribution by nonisotropic reflectance at the surface is included in the non-Lambertian terms. Several assumptions are made in equation (1) to analytically decouple the aerosol scattering, Rayleigh scattering, and gaseous absorption such that the variation of each component can be efficiently accounted for. The gas absorption, apart from that by water vapor, is assumed to occur above the scattering layers of the atmosphere. As water vapor absorption is considered to be negligible at short wavelengths where Rayleigh scattering is significant, the Rayleigh reflectance is not attenuated by the water vapor transmittance. Because aerosol is assumed to be well mixed with water vapor, the aerosol path reflectance is, on average, affected by half of the total column amount of water vapor.

[18] The combined molecular and aerosol path reflectance ρR + A are precalculated with 6S-V1.1 RTM for the sea level standard atmospheric pressure P0 (1013.25 hPa). The aerosol path reflectance needed in equation (1) is then approximated as

display math(2)

The Rayleigh reflectance and transmittance is analytically calculated following the scheme described in Appendix A.

[19] The ozone and water vapor amounts and the actual surface pressure needed to calculate the atmospheric terms in equation (1), as well as vector surface winds and 2 m air temperature, are obtained from the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) model [Environmental Modeling Center, 2003]. If NCEP GFS model data is not available, backup options include the Fleet Numerical Meteorology and Oceanography Center Navy Global Environmental Model data [Rosmond, 1992]. All of the gaseous transmittance terms, Tgog, inline image, and inline image, are analytically computed based on the empirical relationships described in Appendix B. Band dependent coefficients for these empirical relationships were computed using the 6S-V1.1 RTM by varying the concentration of the respective species over six different atmospheric profiles (tropical, midlatitude summer, midlatitude winter, subarctic summer, subarctic winter, and 1962 U. S. standard atmosphere) [Mc Clatchey et al., 1971]. The prelaunch measurements of the VIIRS bands relative spectral response functions (RSRs) were used in the RTM calculations. Although updated RSRs accounting for on orbit changes in rotating telescope assembly transmission due to tungsten oxide contamination have been made available by the VIIRS SDR Cal/Val team, there are currently no plans to update these coefficients due to the expected negligible impact on the aerosol retrievals.

[20] The band-dependent total transmittance TR + A, the atmospheric spherical albedo SR + A, and atmospheric path radiance ρR + A, as well as some of the non-Lambertian surface reflectance terms to be discussed later, are precomputed for an extensive set of geometries, multiple values of the AOT, and candidate aerosol models using the 6S-V1.1 RTM and stored in look-up tables (LUTs) which are described in detail in section 5.

[21] Because the reflective properties of ocean and land are very different, separate retrieval approaches are used for over-land and over-ocean retrievals, respectively. A schematic illustration of the major components of VIIRS aerosol algorithm is shown in Figure 3.

Figure 3.

Flowchart of the VIIRS aerosol algorithm processing.

4.1 Land Inversion

[22] The aerosol retrieval over land uses the VIIRS reflectances in the M1 (412 nm), M2 (445 nm), M3 (488 nm), M5 (672 nm), and M11 (2250 nm) bands. The land inversion is a modified version of the dark dense vegetation approach [Kaufman et al., 1997]. In this approach, aerosol optical thickness is retrieved using an expected (empirically derived) relationship between the surface reflectances in the blue (488 nm) and red (672 nm) bands. Reflectances in other channels are used for selecting the most appropriate aerosol model from a set of candidates [Vermote and Kotchenova, 2008]. The five candidate aerosol models used by VIIRS land inversion are described in section 4.1.1.

[23] The basis of the AOT retrieval over land is that the ratio of TOA reflectances in the blue and red bands is changed from that of the surface value in the presence of aerosols. If the expected blue-to-red surface reflectance ratio for a given target is known, the AOT for a specific aerosol model can be derived by retrieving the blue and red surface reflectances by varying the value of AOT until the ratio of retrieved blue and red surface reflectances is close to the expected ratio.

[24] In practice, for a given AOT and aerosol model, surface reflectances at the blue and red channels are first retrieved from the VIIRS TOA measurements followed by the prediction of the blue channel surface reflectance from the retrieved red channel reflectance multiplied the expected blue-to-red ratio. The predicted and retrieved blue reflectances are then compared. This process is repeated with a different value of AOT until the retrieved blue reflectance equals the predicted blue reflectance. The AOT value for which this happens is considered the solution for the particular aerosol model considered.

[25] The process is repeated for each of the five aerosol models. For each aerosol model, the surface reflectance at 412 nm, 445 nm, 488 nm, 672 nm, and 2250 nm is computed using the retrieved AOT for that model. A residual is then computed based on deviations from the 412 nm, 445 nm, 488 nm, and 2250 nm surface reflectances predicted from the 672 nm surface reflectance.

display math(3)

[26] The model with the lowest residual is selected. The entire process is schematically shown in Figure 3.

[27] The VIIRS land inversion assumes a Lambertian surface when retrieving the surface reflectances. Neglecting the non-Lambertian terms, equation (1) is solved for the surface reflectance ρsurf for each value of AOT and each aerosol model in the LUT:

display math(4)

[28] The LUT values are indexed by AOT at 550 nm, and thus, the AOT that is only directly retrieved is at 550 nm. The AOT values at the VIIRS bands are then calculated using the normalized aerosol extinction coefficients from the atmospheric LUT for the aerosol model selected.

[29] The expected surface reflectance band ratios, which were updated on 22 January 2013, were computed using VIIRS/AERONET matchup data as described in section 4.1.2.

4.1.1 Land Aerosol Models

[30] The VIIRS aerosol algorithm uses a set of five predefined aerosol microphysical models in the land inversion. The models are dust (Cape Verde), high-absorption smoke (African savanna, Zambia), low-absorption smoke (Amazonian forest, Brazil), clean urban (Goddard Space Flight Center, Greenbelt, MD), and polluted urban (Mexico City) aerosols, based on Dubovik et al. [2002]. All models have size distributions defined by bimodal lognormal distributions of spherical particles.

display math(5)

[31] Where rv is the volume mean radius, σv is the standard deviation of the radius, and modes m = 1 and m = 2 are the fine and coarse modes, respectively.

[32] Table 2 contains the physical parameters of the models which are inputs to a Mie scattering code for calculating the modeled optical properties. These optical properties are inputs into the 6S-V1.1 RTM [Kotchenova and Vermote, 2007] which is used to compute the atmospheric parameters stored in the atmospheric LUT (section 5).

Table 2. Land Aerosol Model Physical Properties
 DustSmoke, Low AbsorptionSmoke, High AbsorptionUrban, CleanUrban, Polluted
  1. aRefers to AOT at 0.44 µm.
  2. bRefers to AOT at 1.02 µm.
  3. cThe imaginary part of the index of refraction for dust is defined for the following (wavelength, index of refraction) pairs: (0.350 µm, 0.0025), (0.400 µm, 0.0025), (0.412 µm, 0.0025), (0.443 µm, 0.0025), (0.470 µm, 0.0023), (0.488 µm, 0.0021), (0.515 µm, 0.0019), (0.550 µm, 0.0016), (0.590 µm, 0.0013), (0.633 µm, 0.0010), (0.670 µm, 0.0007), (0.694 µm, 0.0007), (0.760 µm, 0.0007), (0.860 µm, 0.0006), (1.240 µm, 0.0006), (1.536 µm, 0.0006), (1.650 µm, 0.0006), (1.950 µm, 0.0006), (2.250 µm, 0.0006), and (3.750 µm, 0.0006).
Refractive Indices
Real part1.481.471.511.41–0.03a1.47
Imaginary partc0.00930.0210.0030.014
Size Parameter Fine Mode
Volume mean radius (µm)0.120.13 + 0.04a0.12 + 0.025a0.12 + 0.11a0.12 + 0.04a
Standard deviation0.49 + 0.10b0.400.400.380.43
Volume concentration (µm3/µm2)0.02 + 0.02b0.12a0.12a0.15a0.12a
Size Parameter Coarse Mode
Volume mean radius (µm)1.903.27 + 0.58a3.22 + 0.71a3.03 + 0.49a2.72 + 0.60a
Standard deviation0.63–0.10a0.790.730.750.63
Volume concentration (µm3/µm2)0.9b0.05a0.09a0.01 + 0.04a0.11a

4.1.2 Computing Surface Reflectance Ratios

[33] The latest updated spectral surface reflectance ratios used in the retrieval over land were derived from a matchup database composed of AERONET AOT data from the level 1.5 inversion product [Dubovik et al., 2002] from 99 globally distributed AERONET sites, 51 by 51 moderate resolution patches of VIIRS M-band reflectances, and ancillary data collected over these AERONET sites. The database included ~60,000 matchups.

[34] The 6S-V1.1 RTM was run in atmospheric correction mode [Kotchenova and Vermote, 2007] to produce “true” spectral surface reflectances. For this run, matchups with stable AOT (changes in AOT are less than 0.02 over the hour that contains the VIIRS matchup) were selected. Only pixels flagged as high-quality land pixels by the AOT IP were processed. Pixels with an elevation difference of more than 100 m from the AERONET station were also screened out before computing the surface reflectance band ratios. To screen out outliers, the highest and lowest 10% ratio values were discarded before computing the mean and standard deviation of the surface reflectance ratios presented in Table 3.

Table 3. VIIRS Surface Reflectance Band Ratios
Band RatioMeanSD

[35] Note that the Beta maturity level over-land AOT product was obtained from the algorithm running with prelaunch values of the band ratios. These prelaunch values were calculated using the MODIS/AERONET matchup data (originally derived for the MOD09 MODIS surface reflectance product). Because of the differences between the VIIRS and MODIS spectral channels, these band ratios were not fully applicable to VIIRS and resulted in a high bias in the Beta level VIIRS AOT product over land in comparison to MODIS and AERONET values.

[36] Even in the latest updated band ratios as listed in Table 3, the standard deviation of the band ratios is a substantial fraction (10% to 30%) of the mean values indicating that using global values for the expected spectral surface reflectance relationship is a major source of error in the VIIRS land aerosol inversion. As illustrated in Figure 4, which plots the spectral surface reflectances of typical vegetation normalized to those in the red (M5) band, the reflectance in the blue (M3) band is quite variable. Considering seasonal and regional changes in vegetation cover, a global constant value of the surface reflectance ratio is not expected to work everywhere and every time.

Figure 4.

Spectral reflectance of vegetation (normalized to reflectance at 672 nm) from Advanced Spaceborne Thermal Emission and Reflection Radiometer spectral library version 2.0 [Baldridge et al., 2009].

4.2 Ocean Inversion

[37] Following the MODIS approach [Tanré et al., 1997], the VIIRS aerosol retrieval over ocean is carried out by searching for the AOT and aerosol model that most closely reproduces the VIIRS-measured TOA reflectance in multiple bands. In order to account for the large variation of aerosol properties with LUTs of manageable size, the algorithm adopts the approximation of decomposing the TOA reflectance ρtoa(τa) as a sum of contributions associated with a fine mode inline image and a coarse mode inline image aerosol model weighted by the fine mode fraction parameter η [Wang and Gordon, 1994]:

display math(6)

[38] Four coarse mode and five fine mode models [Kaufman and Tanré, 1996] are considered in the inversion over ocean, and η is varied from 0 to 1 in steps of 0.01 leading to 2020 candidate aerosol models. For each of the 2020 candidate aerosol model combinations, matching the measured TOA reflectance at M7 (865 nm) is performed first to retrieve AOT at 550 nm (τ550), then the TOA reflectances in the M5 (672 nm), M6 (746 nm), M8 (1,240 nm), M10 (1610 nm), and M11 (2250 nm) bands are computed using the retrieved τ550, and the associated residual is calculated as the sum of the squared difference between the calculated (inline image) and measured (inline image) reflectances:

display math(7)

[39] The best solution with minimum residual among the 2020 retrievals is selected as the final retrieval containing the AOT at 550 nm, indices of fine and coarse modes, and the fine mode fraction parameter η. A diagram of the aerosol retrieval over ocean is shown in Figure 3.

[40] Calculation of the TOA reflectance over ocean follows the non-Lambertian simulation scheme in the 6S-V1.1 RTM with the simplification introduced by using LUTs:

display math(8)

[41] The TOA reflectance is estimated as the sum of two components: path radiance from atmosphere and contribution from surface with the atmospheric interaction. For the path radiance, atmospheric reflection from aerosol and molecules is precalculated and saved in the LUT for an extensive coverage of geometries, all fine and coarse aerosol models, and various amounts of τ550 at the sea level standard atmospheric pressure. Variation of surface pressure is handled by correcting the LUT reflectance with the difference of the Rayleigh reflection analytically calculated for the input and standard pressure (Appendix A). The surface term consists of the contributions from water-leaving radiance, whitecap, and specular reflection [Koepke, 1984]. The water-leaving radiance and whitecap reflection ρw + wc are considered Lambertian and are functions of pigment concentration and wind speed, respectively. The retrieval algorithm deliberately avoids using the first four VIIRS M-bands due to the uncertain water-leaving reflectance. The water-leaving reflectance at band M5 (672 nm) is assigned to be 0.001, while the water-leaving reflectance is assumed to be zero for the other channels. Calculation of the whitecap reflectance is described in Appendix C.

[42] The sunglint reflection ρG is computed analytically as a function of wind speed and direction using the Cox and Munk [1954] model for wave slope distribution and Fresnel's reflection coefficients described in Born and Wolf [1975] and is attenuated by the two-way direct transmittance inline image. The coupling of sunglint reflection with the atmosphere involves the scenarios of single atmosphere-surface interaction (direct-down-direct-up, diffuse-down-direct-up, direct-down-diffuse-up, and diffuse-down-diffuse-up) and multiple atmosphere-surface interactions and is attenuated by the diffuse transmittance inline image in one direction and the direct transmittance in the other direction (or in the case of inline image, the two-way diffuse transmittance). The coupling terms inline image, inline image, and inline image, described in section 5, represent various angular integrals of ρG; they are precalculated and saved in a separate sunglint LUT (see section 5).

4.2.1 Ocean Aerosol Models

[43] The VIIRS aerosol algorithm uses a dynamic set of models in the ocean inversion. As mentioned above, there are four fine mode aerosol models and five coarse mode aerosol models. All models have size distributions defined by single-mode lognormal distributions:

display math(9)

[44] Table 4 lists the complex refractive index as a function of wavelength, volume mean radius (rv), width of the mode (σv), and a generic description of the aerosol type represented by each mode (1–4 are fine mode, 5–9 are coarse mode). These physical parameters of the models are inputs into a Mie scattering code which computes the model optical properties. These optical properties are the inputs into the 6S-V1.1 RTM for computing the atmospheric parameters stored in the atmospheric LUT (see section 5).

Table 4. Ocean Aerosol Model Physical Properties
 λ = 0.47 to 0.86 µm1.24 µm1.65 µm2.25 µmrv (µm)σv (µm)Comments
11.45–0.0035i1.45–0.0035i1.43–0.0035i1.40–0.001i0.10.40Wet water soluble type
21.45–0.0035i1.45–0.0035i1.43–0.0035i1.40–0.001i0.150.60Wet water soluble type
31.40–0.0020i1.40–0.0020i1.39–0.0005i1.36–0.0003i0.20.60Water soluble with humidity
41.40–0.0020i1.40–0.0020i1.39–0.0005i1.36–0.0003i0.250.60Water soluble with humidity
51.45–0.0035i1.45–0.0035i1.43–0.0035i1.43–0.0035i0.980.60Wet sea salt type
61.45–0.0035i1.45–0.0035i1.43–0.0035i1.43–0.0035i1.480.60Wet sea salt type
71.45–0.0035i1.45–0.0035i1.43–0.0035i1.43–0.0035i1.980.60Wet sea salt type
81.53–0.003i (0.47 µm) 1.53–0.001i (0.55 µm) 1.53–0.000i (0.66 µm) 1.53–0.000i (0.86 µm)1.46–0.000i1.46–0.001i1.46–0.000i1.480.60Dust-like type
91.53–0.003i (0.47 µm) 1.53–0.001i (0.55 µm) 1.53–0.000i (0.66 µm) 1.53–0.000i (0.86 µm)1.46–0.000i1.46–0.001i1.46–0.000i2.50.80Dust-like type

5 Look-Up Tables

[45] There are two look-up tables used in the VIIRS aerosol algorithm. The atmospheric LUT is used over both land and ocean and contains precomputed values of one-way total transmittance, atmospheric spherical albedo, atmospheric reflectance, and normalized aerosol extinction coefficients (ratio of optical thickness at the VIIRS band center wavelengths to optical thickness at 550 nm). All parameters are functions of spectral band, aerosol model, and optical thickness. The transmittance is also a function of solar or sensor zenith angle depending on path direction. The path radiance is also a function of both zenith angles and scattering angle. The dimension information of the atmospheric LUT is listed in Table 5. All parameters are computed using the 6S-V1.1 RTM at sea level standard atmospheric pressure without molecular absorption.

Table 5. Dimension Information of the Atmosphere LUT
LUT ParametersNumber of DimensionsDimension
VIIRS BandaAerosol ModelbAOT at 550 nmcZenith AngledScattering Anglee
  1. aTen VIIRS bands: M1–M11 except M9.
  2. bNine aerosol models over ocean: four fine modes and five coarse modes; five aerosol models over land: dust, high-absorbing smoke, low-absorbing smoke, clean urban aerosol, and polluted urban aerosol.
  3. cFifteen AOTs at 550 nm: 0.01, 0.05, 0.10, 0.15, 0.20, 0.30, 0.40, 0.60, 0.80, 1.00, 1.20, 1.40, 1.60, 1.80, and 2.00.
  4. dTwenty-one zenith angles: 0°–80° in steps of 4°.
  5. eThe 5527 scattering angles: in steps of 4° for each pair of solar and viewing zenith angles.
  6. fOver ocean, the normalized extinction coefficient has two dimensions without the dependence on AOT.
One-way transmittance4XXXX 
Reflectance4XXX X
Spherical albedo3XXX  
Normalized aerosol extinction coefficientf3XXX  

[46] The sunglint LUT contains precomputed values of inline imagewhich represents the hemispherical-directional sunglint reflectance used to calculate the diffuse-down-direct-up atmosphere-sunglint interaction. It is computed as the integral of wind roughened ocean surface bidirectional reflectance distribution function (BRDF, ρG) over all incident directions weighted by the downwelling diffuse radiance:

display math(10)

[47] Based on the assumption of reciprocity [Van de Hulst, 1957], the directional-hemispherical sunglint reflectance, inline image, is considered equal to the hemispherical-directional reflectance by switching the incident and viewing direction, i.e., inline image, and hence not stored in LUT as an extra parameter. The reflectance is a function of spectral band, aerosol model, optical thickness at 550 nm, solar and sensor zenith angles, and relative azimuth angle (Table 6). LUT also contains sunglint spherical albedo (inline image), which is a function of spectral band only and computed as an integral of the wind-roughened ocean BRDF over both the incident and viewing directions:

display math(11)
Table 6. Dimension Information of the Sunglint LUT
LUT ParametersNumber of Dimensions Dimension
VIIRS BandaAerosol ModelbAOT at 550 nmcSolar Zenith AngledViewing Zenith AngledRelative Azimuth Anglee
  1. aSix VIIRS bands: M5, M6, M7, M8, M10, and M11.
  2. bNine aerosol models: 4 fine modes and 5 coarse modes.
  3. cFifteen AOTs at 550 nm: 0.01, 0.05, 0.10, 0.15, 0.20, 0.30, 0.40, 0.60, 0.80, 1.00, 1.20, 1.40, 1.60, 1.80, and 2.00.
  4. dTwenty-one zenith angles: 0°–80° in steps of 4°.
  5. eTwenty-one relative azimuth angles: 0°–180° in steps of 9°.
Hemispherical-directional reflectance6XXXXXX

[48] The sunglint LUT parameters are computed using the 6S-V1.1 RTM with a fixed input wind speed of 2 m/s and direction of 0° azimuth angle; therefore, the wind dependence of the atmosphere-sunglint interaction terms is neglected to keep the LUT size manageable. However, since the direct sun glint term is computed analytically using the NCEP Global Forecast System ancillary wind speed, the majority of the surface BRDF effects is included in the computed TOA reflectances.

6 Data Screening

[49] Not all pixels within a VIIRS granule are suitable for aerosol retrieval, and not all retrievals have the same level of quality. Pixels with clouds, cloud shadows, snow, ice, subpixel water, bright land surface, fire, sunglint, suspended sediments or shallow water, and large solar zenith angle are deemed inappropriate pixels because the contribution of these nonaerosol “constituents” to the radiation received by the satellite sensor are either too large relative to the aerosol contribution (they mask the aerosol signal) or their contribution can only be poorly characterized (resulting in large uncertainty in AOT retrieval). The VIIRS algorithm screens out these pixels before proceeding with the aerosol retrieval. Table 7 defines how the results of the different tests determine the pixel quality and whether the tests are performed in the VIIRS aerosol algorithm itself or are performed by the VCM.

Table 7. VIIRS Aerosol Algorithm Data-Screening Criteria
ConditionPixel Quality LevelApplies toDetected by
DegradedExcludedNot ProducedLandOceanVCMInternal Tests
Out of spec range X XX X
Coastal or inland water  X  X 
Cloud contamination  XXXX 
Cloud adjacencyX  XX X
CirrusX  XXXX
Invalid SDR data  XXX X
Sun glint  XXXXX
Cloud shadowX  XXX 
Snow/ice  XXXXX
Fire  XX XX
Soil dominatedX  X  X
Bright surface  XX  X
Turbid water  X X X
Ephemeral water  XX  X
65° ≤ solar zenith angle < 80°X  XX X
solar zenith angle ≥ 80°  XXX X
Large retrieval residualX  XX X

[50] The VCM provides the aerosol algorithm with the land/water classification and majority of data-screening information. If the VCM flags identify a coastal or inland water surface type, indicate a high probability of cloud, sunglint, snow/ice, or fire, the aerosol algorithm does not retrieve for that pixel, and the quality flag is set to “not produced.”

[51] In addition to using the information in the VCM input, the VIIRS algorithm also performs its own tests for the presence of thin cirrus, fire, subpixel water, snow, ice, sunglint, and turbid or shallow water. These internal tests at present are mostly redundant with the information provided by the VCM but can be important in some circumstances to meet more conservative screening criteria of the aerosol products. The necessity of these internal tests will be reevaluated once the VCM has been fully tuned and is deemed stable by the VIIRS Cloud Mask Cal/Val team.

[52] The internal cloud test attempts to identify cirrus cloud over land. This test is a reflectance threshold test on band M9 (1.38 µm) and surface temperature just like the test in the VCM, but the threshold can be set tighter since the regions where this test is likely to produce false alarms (high, snow-covered mountains, Greenland and Antarctica) are not suitable regions for aerosol retrievals. Adjacency to cloud is detected for each pixel by checking the VCM cloud confidence flags of the neighboring 3 × 3 pixels. Near-cloud AOT retrievals are degraded due to the cloud 3-D effect unaccounted for by the assumption of 1-D radiative transfer in the aerosol algorithm [Wen et al., 2006].

[53] The aerosol algorithm uses the most conservative of three sun glint masks. The first two are computed by the VCM. These tests are a static 36° cone around the specular direction and a specular facet coverage of 1.5% computed from a one-dimensional Cox and Munk distribution based on the NCEP wind speed. An additional internal test for sunglint over ocean is computed by the aerosol algorithm from a two-dimensional Cox and Munk distribution based on the NCEP wind speed and direction and using the index of refraction of seawater for band M8 (1240 nm) to compute the Fresnel reflectance. A threshold of 3% is used to eliminate any glint ocean pixels undetected by VCM. Sunglint is also detected over land. This is important because land-based water bodies such as rivers, lakes, ponds, and irrigated farmland may not be tagged as water in the land/water mask. In the specular reflection direction, these inland water pixels, even large puddles, can be misinterpreted to be fires, clouds, or clear land pixels. A land pixel is assumed to be affected by sunglint if two reflectance anomalies (constructed from the visible and short-wave infrared TOA reflectances) and the surface temperature are greater than specified thresholds.

[54] The presence of snow/ice over land is internally detected by the threshold tests on the visible reflectance anomaly [Baker, 2011a], ratio of TOA reflectance at M8 and M7, and the surface temperature derived from the split window technique [Walton et al., 1998]. The detection of sea ice over ocean requires the surface temperature being less than 275 K. Given that current evaluation shows undetected snow over high-latitude land and too many false alarms for sea ice over high-latitude ocean, improving the snow/ice mask is an ongoing issue for the algorithm development.

[55] The MODIS heritage turbid/shallow water test described in Li et al. [2003] is used to identify turbid and/or shallow water which is excluded from processing. Turbid water contains suspended sediments, while the reflection from the sea bottom is visible in shallow water. Neither of these conditions is sufficiently well characterized to be used in the aerosol retrieval and must be avoided.

[56] The fire test over land is tested on the midinfrared reflectance anomaly [Baker, 2011a] and retrieved surface temperature higher than the threshold values. The ephemeral water test computes TOA normalized difference vegetation index (NDVI) and excludes values which are too low (< 0.1) from processing. This test will be triggered by desert and semiarid regions which will also fail the bright pixel test, so the quality flag for this test should not be taken as a definitive indication of ephemeral water. For the VIIRS instrument, the NDVI is computed as

display math(12)

[57] Bright land pixels unsuitable for processing by the dark target algorithm are screened out if the shortwave NDVI (NDVISWIR) is less than 0.05, and TOA reflectance at band M11 (2250 nm) is higher than 0.3 [Baker, 2011a]. The NDVISWIR is computed as

display math(13)

[58] The algorithm also makes a distinction between vegetation-dominated pixels and those that are less vegetated and assigns quality flags differently for each category. A vegetation-dominated pixel requires NDVISWIR > 0.2 [Baker, 2011a]. This condition is necessary for the retrieval to obtain a quality flag of “high” quality.

[59] The VIIRS aerosol algorithm also sets the quality of the pixel level retrieval based on a series of tests for various adverse conditions. The quality of the pixel level retrieval is used in the aggregation phase of the algorithm to determine which pixels to use to create the EDR and set the overall quality level of the EDR. The algorithm classifies each pixel as not produced, excluded, degraded, or high quality. An “excluded” quality flag indicates the retrieved AOT is out of the specified range (currently between 0 and 2). When retrieval conditions are not favorable (e.g., the pixel adjacent to cloud and soil-dominated surface) or retrieval performance is not optimal (e.g., large retrieval residual), a “degraded” quality flag is assigned to the AOT. If none of these conditions are present, the pixel obtains a “high-quality” flag.

[60] The quality flag values used for the AOT IP are as follows: 0 = high quality, 1 = degraded, 2 = excluded, and 3 = not produced. The quality flag values used for the AOT EDR are reversed: 3 = high quality, 2 = medium quality, 1 = low quality, and 0 = not produced.

7 Outlier Removal and Aggregation

[61] The AOT EDR is aggregated from 8 × 8 IP pixel level data. Only high-quality and degraded quality pixels are used in the IP to EDR aggregation. The overall quality of the EDR depends upon the number of high-quality pixels available within the 64 pixel EDR cell. Each EDR cell is either reported as land or ocean or is set to a fill value. If more than half of the available (non-bowtie deleted) pixels are land (or ocean), the EDR cell is reported as land (or ocean). The EDR cell is set to a fill value if neither condition applies, such as coastal regions or inland water. Only land IP level retrievals are used to construct a land EDR, and only ocean pixels are used to construct an ocean EDR. If more than a quarter of the available (non-bowtie deleted) pixels in the EDR cell are high quality, the EDR is reported as high quality and only high-quality pixels are averaged in the aggregation after outliers (highest 40% and lowest 20% IP retrievals) are removed. If the high-quality condition is not met but more than a quarter of the available (non-bowtie deleted) pixels in the EDR cell are either high or degraded quality, the EDR is reported as medium quality, both high-quality and degraded quality pixels are averaged in the aggregation after outliers are removed. If neither the high- nor medium-quality criteria are met but at least one pixel in the EDR cell is of high or degraded quality, EDR is reported as low quality and both high-quality and degraded quality pixels are averaged in the aggregation after outliers are removed. Low-quality EDRs usually have large errors and are not suitable to use for scientific investigations. This aggregation procedure is applied to all the spectral AOT at each wavelength.

8 Preliminary Evaluation of AOT

[62] The VIIRS AOT product has been compared with aerosol products derived from observations of the Aerosol Robotic Network (AERONET) [Holben et al., 1998]. Because the VIIRS evaluation is proceeding in near real time, the matchups used in this evaluation are level 1.5. As most AERONET stations are land based, stations are spatially limited and do not offer a complete global evaluation, especially over open ocean which is the reason for also evaluating VIIRS retrievals with the Maritime Aerosol Network (MAN) data set (Liu et al., manuscript in preparation, 2013).

[63] The VIIRS aerosol/AERONET matchup data set is accumulated automatically on the NOAA Science Investigator Processing System (NSIPS) Cal/Val system via a Product Generation Executable (PGE) which saves all inputs to and outputs from the VIIRS aerosol algorithm. This enables very rapid reprocessing of the matchup data in order to test proposed changes to the algorithm. The in situ data source for the matchups was the AERONET level 1.5 (hourly) inversion product initially. On 17 February 2013, the PGE was updated to use the AERONET level 1.5 direct Sun product which has all of the direct Sun measurements on a time interval of 15 min. The quality assurance criteria used with the aerosol/AERONET matchup data set are as follows. A matchup time window of plus or minus 1 h is used. If two or more AERONET observations within the time window are available, they are averaged. The matchup spatial area for the EDR product consists of the 5 × 5 EDR cells centered around the cell that includes the AERONET station, while the matchup spatial area for the IP product contains the 51 × 51 pixels centered around the AERONET site. For a match-up to be included in the statistics, at least 25% of retrievals in the matchup area must be high quality. Then the spatial average of all high-quality retrievals is calculated and used for comparison with the AERONET temporal average. AERONET AOT at 550 nm is interpolated from the AERONET values at 440 nm and 870 nm. The interpolation is linear in log-log space.

[64] The time period used in this evaluation is 1 May 2013 through 30 April 2013. Because the operational land AOT was originally processed using the prelaunch surface reflectance ratios based on MODIS band passes through 22 January 2013 and a processing error occurred in the operational system in October and November 2012, the VIIRS matchup data was reprocessed through version I1.5.07.01 of a standalone version of the operational code known as the Algorithm Development Library using the updated surface reflectance ratios based on the VIIRS band passes for the time period 1 May 2012 through 23 January 2013. The remaining data used is the actual operational data product. Figure 5 shows the scatter plots of VIIRS AOT EDR against AERONET measurements. Table 8 lists the statistics of retrieval performance for both EDR and IP products. Over land, the mean VIIRS AOT at 550 nm EDR is 0.02 smaller than the mean AOT at 550 nm measured by AERONET. The standard deviation of the differences is 0.11. The AOT IP has a near-zero bias and a slightly larger standard deviation (0.12) than the EDR. Over ocean, both EDR and IP AOT are less than ~0.02 higher than AERONET with the standard deviation of the error is 0.08 and 0.05, respectively.

Figure 5.

Scatter plots of retrieved VIIRS aerosol optical thickness at 550 nm EDR versus AERONET level 1.5 measurements. (left) Over land: May 2012 through April 2013. (right) Over ocean: May 2012 through April 2013.

Table 8. VIIRS Aerosol Optical Thickness Performance Relative to AERONET
  May 2012 to April 2013
AOT EDRLand4056−0.0230.1120.742
AOT IPLand30260.0000.1230.723

[65] The VIIRS AOT EDR has also been evaluated with a different set of VIIRS-AERONET matchup data and the MODIS collection 5.1 AOT products (Liu et al., manuscript in preparation, 2013). The results from these studies show that the performance of the VIIRS AOT retrieval is comparable to that of Aqua MODIS on a global scale. VIIRS global mean AOT differs from that of MODIS by −0.01 over ocean and 0.03 over land. Compared to the AERONET measurements, VIIRS retrievals exhibit a slight positive bias of 0.013 over ocean and negative bias of −0.01 over land. The VIIRS overocean AOT EDR agrees with MAN ship-based measurements very well with a difference of 0.01.

9 Differences Between VIIRS and MODIS Aerosol Algorithms

[66] While the VIIRS aerosol algorithm was developed based on the MODIS aerosol heritage, there are significant differences between the final products, attributable to both sensor differences and algorithm differences. These differences are summarized in Table 9.

Table 9. VIIRS Versus MODIS Orbit, Algorithm, and Product Characteristics
Orbit altitude824 km690 km
Equator crossing time13:30 UTC13:30 UTC (Aqua)
Swath width3000 km2300 km
Pixel resolution (nadir)0.75 km0.5 km
Pixel resolution (swath edge)1.5 km2 km
Algorithm (General)
Main source of data screeningExternal VCMInternal tests
Aggregation onOutputsInputs
Residual calculated asAbsolute differenceRelative difference
Overocean Algorithm
Channel used0.67, 0.74, 0.86, 1.24, 1.61, 2.25 µm0.55, 0.66, 0.86, 1.24, 1.61, 2.12 µm
Surface reflectionNon-Lambertian, function of wind speed and directionLambertian, independent on wind (will change in C6)
Aerosol modelCombination of fine and coarse modesCombination of fine and coarse modes
Match toTOA reflectancesTOA reflectances
Retrieval over inland waterNoYes
Over land Algorithm
Channel used0.41, 0.44, 0.48, 0.67, 2.25 µm0.47, 0.66, 2.12 µm
Aerosol modelSelect one from five predefined modelsMix two assigned fine and coarse mode-dominated models
Spectral surface reflectanceConstant ratios of 0.41, 0.44, 0.48, 2.25 µm over 0.66 µmLinear relationship between 0.66 and 2.12 µm as a function of NDVI and scattering angle; constant linear relationship between 0.47 and 0.66 µm
Match toSurface reflectancesTOA reflectances
Nominal spatial resolution0.75 km (IP) 6 km (EDR)10 km (C5) 3 km (C6)
Granule size86 s5 min
AOT range[0, 2][−0.05, 5]
Main product landSpectral AOTSpectral AOT
Main product oceanSpectral AOT Ångström exponentSpectral AOT fine mode fraction

[67] As listed in Table 9, VIIRS flies at a higher orbit than MODIS and has a broader swath. The broad swath permits full global coverage every day, without the gaps between swaths at the equator that occur in MODIS data, but it also introduces larger sensor view angles and longer slant paths that may have an effect on the aerosol retrievals. VIIRS has similar wavelengths and spectral band widths as MODIS, but they are not exactly the same, and those differences require different gas corrections and different surface reflectance ratios. Finally, VIIRS spatial resolution is significantly different from MODIS, which has a higher resolution at nadir but lower resolution at the edge of scan. Due to the VIIRS aggregation scheme, its GIFOV only grows by a factor of 4 while the MODIS GIFOV grows by a factor of 8 from the nadir resolution. These sensor differences imply that even if the MODIS algorithm were to be applied unchanged to the VIIRS radiances, the resulting aerosol product would be different.

[68] As previously described in section 4, the VIIRS and MODIS algorithms are different. The VIIRS algorithm uses different criteria in choosing pixels for retrieval. For example, it does not retrieve over inland water such as the Great Lakes. MODIS does retrieve over large bodies of inland water but avoids most small-scale lakes, rivers, and ponds. VIIRS depends heavily on the external VIIRS cloud mask, which divorces much of the cloud and improper surface identification from the internal aerosol algorithm. The MODIS algorithm does most of the cloud masking and pixel selection internally [Martins et al., 2002].

[69] The VIIRS over-land algorithm differs from the MODIS operational Dark Target algorithm [Kaufman et al., 2005; Remer et al., 2008; Levy et al., 2007, 2010]. Instead, it is based on the MODIS atmospheric correction algorithm [Vermote and Kotchenova, 2008], which retrieves aerosol information as a by-product in deriving the land surface reflectance. Thus, there is no publicly available MODIS aerosol product that is a direct predecessor to the VIIRS over-land aerosol product. The VIIRS land product sets up an atmospheric correction problem that solves for the surface reflectance: Algorithm searches for the solution of aerosol such that the resulting surface reflectance matches the expected surface reflectance determined from the expected surface reflectance ratios. On the other hand, the MODIS over-land algorithm applies a forward radiative transfer calculation: The algorithm finds the amount and mixture of aerosols such that the computed TOA reflectance matches the direct measurements. Both algorithms must model the surface reflectance using assumptions of the spectral signature of the surface reflectance. The VIIRS algorithm primarily relies on a ratio between two near wavelengths: blue and red. The MODIS algorithm relies equally on both a blue/red band ratio and a ratio between the red band and the 2.13 µm band. The more spectrally distant the bands are, the more uncertainty is introduced into the assumption [Remer et al., 2009]. Other important differences between the VIIRS and MODIS over-land retrievals are that VIIRS uses information from five channels, including two “Deep Blue” channels to choose between five static multimodal aerosol models. There is no model mixing in the VIIRS over-land algorithm. MODIS uses only three channels with the shortest wavelength equal to 0.466 µm. The MODIS retrieval mixes two assigned multimodal models to obtain the aerosol model used in the retrieval. Finally, the VIIRS AOT range is [0, 2.0], while MODIS permits small negative AOT values down to −0.05 and high AOT values up to 5.0.

[70] The VIIRS over-ocean retrieval resembles the operational MODIS over-ocean retrieval closely. The only differences are that the VIIRS retrieval largely avoids assumption of water-leaving radiance value by confining the retrieval to wavelengths greater than or equal to 0.67 µm, while the MODIS retrieval also includes 0.55 µm. VIIRS also includes a 0.746 µm channel that MODIS does not. The VIIRS uses the most conservative of three glint masks, while the MODIS glint mask is a static 40° from specular reflection. VIIRS also derives its rough ocean surface reflectance as a function of wind speed and direction. The traditional MODIS over-ocean retrieval set global wind speed to be 6 m/s, but that will change in the next MODIS collection 6 [Levy et al., 2013]. Just like over land, the VIIRS over-ocean AOT retrievals only cover [0, 2.0], a smaller range in comparison to MODIS.

[71] The VIIRS and MODIS algorithms, over both land and ocean, differ in how each comes to solution, meaning they minimize differently weighted cost functions. We refer the reader to the individual ATBDs for further information [Baker, 2011a; Remer et al., 2009]. VIIRS reports only the best solution, even when solutions are not necessarily unique. MODIS over-ocean reports both the best solution and the average of all acceptable solutions.

[72] The algorithms also differ in how they are aggregated and the products offered. VIIRS makes 0.75 km pixel level retrievals and then aggregates AOT to produce the 6 km EDR product. MODIS first aggregates reflectances to produce a set of representative spectral reflectances for the retrieval box and then makes the retrieval from that single set of reflectances. The VIIRS product spatial resolution is 0.75 km for the IP and 6 km for the EDR at nadir; the MODIS product spatial resolution is 10 km at nadir, though a 3 km product will be available in the upcoming collection 6 [Levy et al., 2013].

10 Summary

[73] The VIIRS aerosol products are derived from the spectral radiances measured by VIIRS on the polar-orbiting Suomi-NPP satellite. As a multispectral approach, VIIRS aerosol algorithm is developed to find an optimal solution of aerosol loading (optical thickness) and intrinsic properties (aerosol model) that best match the measurements in several channels.

[74] The algorithm is applied at the pixel level with 0.75 km resolution. This pixel level product is referred to as an intermediate product (IP). The IP aerosol retrievals are aggregated to produce the environmental data record (EDR) at 6 km resolution with associated quality flags. The EDR data include spectral aerosol optical thickness, Ångström exponent, and suspended matter. At the time of this writing, only the aerosol optical thickness and Ångström exponent products are at the provisional maturity stage, meaning no major issues are believed to be present in the products and in the algorithm used to produce them, and the general research community is encouraged to participate in the evaluation. However, product quality may still not be optimal, and incremental product improvements are still occurring. The aerosol community is strongly advised to consult the product status documents prior to use of data in publications. It should be noted that even though the Ångström exponent product is provisional, the product over land should not be used as it does not have any quantitative value; the over-land Ångström exponent product will likely not be generated in the future. The suspended matter product is at the beta maturity level, and it is not appropriate for quantitative scientific research; the algorithm to produce it will be significantly changed in the future.

[75] VIIRS aerosol products are publicly accessible from NOAA's Comprehensive Large Array-data Stewardship System (CLASS at http://www.class.ngdc.noaa.gov). Users can also find the VIIRS aerosol products readme file from http://www.nsof.class.noaa.gov/saa/products/welcome. This document along with the VIIRS aerosol products users' guide, available online at http://www.star.nesdis.noaa.gov/jpss/ATBD.php, provides the essential documentation for accessing and using the VIIRS aerosol data products. Further algorithm changes will be described in the aforementioned documentation.

[76] Preliminary evaluation of the VIIRS aerosol product against collocated AERONET is carried out in the NSIPS Cal/Val system. It shows the VIIRS AOT product meeting its requirements [Kennedy et al., 2013], with mean accuracy (bias) against AERONET of ~0.02 and precision (standard deviation of the error) ~0.05 over ocean and ~0.12 over land. The current JPSS Level 1 threshold requirements for VIIRS aerosol optical thickness are shown in Table 10. Further evaluation shows similar performance on a global scale by evaluating VIIRS retrieval with AERONET, MAN, and MODIS and is described in detail by Liu et al. (manuscript in preparation, 2013).

Table 10. VIIRS Aerosol Optical Thickness Threshold Requirements
StratificationThreshold Requirement
Measurement accuracy 
Over ocean0.08 (Tau < 0.3); 0.15 (Tau ≥ 0.3)
Over land0.06 (Tau < 0.1); 0.05 (0.1 ≤ Tau ≤ 0.8); 0.2 (Tau > 0.8)
Measurement precision 
Over ocean0.15 (Tau < 0.3); 0.35 (Tau ≥ 0.3)
Over land0.15 (Tau < 0.1); 0.25 (0.1 ≤ Tau ≤ 0.8); 0.45 (Tau > 0.8)

[77] The Suomi-NPP VIIRS aerosol algorithm and products offer the community a continuation of access of global monitoring of aerosol that MODIS currently provides with the added benefit of broader swaths and finer spatial resolution. The products and algorithm are being rigorously evaluated, and currently, the AOT EDR is meeting requirements on a global basis. As the VIIRS products mature, we expect the VIIRS aerosol products to become an important tool for a wide variety of research and applied science applications.

Appendix A: Calculation of Rayleigh Reflectance and Transmittance

[78] The molecular reflectance and transmission are functions of the amount of molecules traversed by the incoming and outgoing light, which is dependent on the Sun-satellite geometry (θs, θv, φ) and the atmospheric pressure at the target surface (Psfc). In the VIIRS aerosol algorithm, the molecular reflectance, inline image, is calculated based on the analytical expression developed by Vermote and Tanré [1992] which accounts for impact of multiple scattering and polarization on intensity and depends on the molecular (Rayleigh) optical thickness τR corresponding to the target surface pressure. The molecular reflectance for the actual pressure Psfc is obtained by adjusting the molecular optical thickness:

display math(A.1)

where P0 is the standard atmospheric pressure and inline image is precomputed with spectral values listed in Table A1. These values were computed using prelaunch measurements of the VIIRS relative spectral response functions using the Hansen and Travis [1974] approximation for the Rayleigh optical depth of air. Note that molecular scattering has a strong spectral signature, which requires accurate spectral registration of the VIIRS bands to maintain a high accuracy in the aerosol retrieval.

Table A1. Spectral Values of Coefficients Used by VIIRS Aerosol Algorithm
CoefficientVIIRS Bands

[79] Molecular reflection is calculated as the sum of single-scattering contribution and the correction for higher orders of scatterings:

display math(A.2)

where δ0,m is the Kronecker delta and inline image and Δm(τR) are the single- and multiple-scattering terms corresponding to the mth term of the phase function Pm(μs, μv).

[80] The Rayleigh transmittance TR(μ) and spherical albedo SR are used to correct the optical functions stored in LUT to the local pressure, which are computed at standard surface pressure. Specifically, the LUT transmittances TR + A in equation (1) need to be multiplied by the ratio of Rayleigh transmission TR(μ) at actual pressure to that at standard pressure; the LUT spherical albedo SR + A is corrected by adding the difference between the Rayleigh spherical albedo SR at actual pressure and that at standard pressure. The pressure correction is realized through the adjustment of Rayleigh optical thickness.

[81] The analytical expression of Rayleigh transmittance function is based on the two-stream method,

display math(A.3)

where μ is the cosine of the solar or viewing zenith angle.

[82] For conservative molecular scattering, the spherical albedo SR is given by

display math(A.4)

where En is the exponential integral inline image

Appendix B: Calculation of Gaseous Transmittance

[83] Ozone transmittance (inline image) for each band (λ) is calculated from the total column concentration (inline image) in units of atm cm:

display math(B.1)

where M is the air mass defined as

display math(B.2)

θs and θv are the solar and viewing zenith angles. The constants inline image are the absorption coefficients and are functions of wavelength.

[84] Once the total column water vapor concentration (inline image) in units of cm is obtained, the transmission function (inline image) for each band (λ) is calculated from

display math(B.3)

[85] The other absorbing gases in the atmosphere included in the aerosol retrieval are carbon dioxide (CO2), oxygen (O2), nitrous oxide (N2O), and methane (CH4). These are well mixed in the atmosphere and do not vary temporally or spatially. Computing their transmission function does not require ancillary data sets that explicitly provide their concentrations. Instead, their transmission functions are linked to the amount of atmosphere traversed by the incoming and outgoing light, which is quantified by the air mass, M, and scaled atmospheric pressure P (ratio of instantaneous to standard pressure). The transmission of other gases (Tgog) is given by

display math(B.4)

[86] Table A1 lists the spectral values of the coefficients used in the VIIRS aerosol algorithm.

Appendix C: Calculation of Whitecap Reflectance

[87] The whitecap (foam) reflectance ρwc is calculated as

display math(C.1)

Where V is the wind speed in m s−1. The ρwc is the product of an effective reflectance ρwc − eff and whitecap coverage [Koepke, 1984]. The current algorithm assumes a spectrally constant value of 0.22 for ρwc − eff; the spectral variation shall be introduced in an algorithm update.


[88] The authors would like to acknowledge the support from the following organizations for their assistance on the development of the VIIRS aerosol algorithm and the associated VIIRS aerosol calibration and validation work: NOAA JPSS program office, VIIRS SDR and VCM calibration and validation teams, NASA MODIS Dark Target and Deep Blue aerosol teams, NASA AERONET and MAN teams, NASA Land PEATE at GSFC, and Atmosphere PEATE at the University of Wisconsin.