This document describes two algorithm families – Deep Blue (DB) and Dark Target (DT) – used to produce aerosol optical properties. The DB family consists of the DB algorithm over land, and the Satellite Ocean Aerosol Retrieval (SOAR) algorithm over water. The DT family also has separate algorithms for land and water-covered pixels, referred to as DT-land and DT-water when it is necessary to distinguish between them. All four algorithms are described here to streamline discussion of similarities and highlight differences.
The algorithms provide a subset of the PACE OCI required aerosol products: specifically, total aerosol optical depth (AOD) at 550 and 675 nm over land, total AOD at 550 and 675 nm over oceans, and fine mode fraction (FMF) of aerosol optical depth at 550 nm over oceans. The other OCI required aerosol products (total AOD over land at 380, 440, and 500 nm over land and ocean) can be obtained via interpolation or extrapolation from these, although this has not yet been implemented in the proxy processing code. Sam-could you implement this? Just use the retrieved Angstrom exponent for now. It will at least give us something for the 380 nm requirement.
The DB and DT families provide additional (i.e. not on the OCI required list) data products. Over land, DB provides AOD at 412 and 490 nm and DT-land provides AOD at 490 nm. Over water both SOAR and DT-ocean provide AOD at 490, 870, 1240, 1640, and 2250 nm. All also provide the Ångström Exponent (AE) across the visible spectral range. The current implementation is to VIIRS measurements as a proxy for OCI, although these algorithms have also previously been applied to MODIS and other sensors.
Satellite instruments do not measure aerosol optical properties directly, but rather reflected or emitted radiation at the top of atmosphere (TOA) at various wavelengths. For cloud-free pixels, the apparent color of the Earth as seen from TOA is dependent on the (spectral and directional) characteristics of atmospheric (i.e. aerosols and molecules) and surface scattering and absorption. The surface signal is often the strongest contributor, particularly over land. As such, for both DB and DT families, cloud-free pixels (determined using an internal cloud mask) are first assigned to the appropriate algorithm using a land-sea mask. These individual algorithms are described below.
The one major departure from previous DB/DT applications is that in heritage processing output is provided at nominal 10 km horizontal resolution for MODIS and nominal 6 km for VIIRS. This is achieved by averaging of full-resolution retrieval results (for DB) or of cloud-free TOA reflectances prior to retrieval (for DT). In the current implementation the DB and DT algorithms are processed at full sensor resolution.
The key references for this algorithm are Levy et al. (2007a,2013); it is the second generation of an algorithm developed originally by Kaufman et al. (1997a). The main physical principle is that in the shortwave infrared (swIR) the atmospheric contribution to the TOA signal is small, and for vegetated surfaces, there is a direct relationship between the surface reflectance at 2.1 μm and that at blue (470-490 nm) and red (650-675 nm) wavelengths (Kaufman et al., 1997b). At these wavelengths the reflectance of vegetated surfaces is also fairly dark (giving the algorithm its name) which increases the proportion of the TOA signal due to aerosols. While Kaufman et al. (1997a) adopted fixed ratios of surface reflectances between these bands based on Landsat observations, the main contribution of Levy et al. (2007a) was to develop this spectral relationship into to a dynamic ratio as a function of scattering angle and a swIR-based normalized difference vegetation index (NDVI). Fit coefficients were determined empirically using MODIS observations across the USA, and the same relationship is applied globally.
With this spectral land model as a constraint, aerosol loading is determined by varying the total amount and partition between two aerosol optical models in order to match the TOA reflectance. This provides the total AOD at 550 nm and a weighting parameter η; the AE (and AOD at other wavelengths) can be derived from these. The aerosol optical models are described in Levy et al (2007b). There are three potential fine-mode dominated optical models, which differ in size distribution and single scattering albedo (SSA), determined by global cluster analysis of AERONET data. For each pixel the fine-mode dominant model used depends on geographic location and season. There is also one coarse-mode dominated optical model representing desert dust, which is used globally. Both the fine-dominated and coarse-dominated models consist of bimodal distributions. Thus, η is not a fine mode optical depth fraction (FMF) here, but a fine vs. coarse model weighting parameter. Note that η and AE from DT-land have been shown to have limited skill and are not recommended for quantitative analysis (Levy et al., 2010, 2013).
The main limitation of the DT-land algorithm is that the spectral surface relationship only holds for vegetated surfaces and is trained empirically on data from limited areas (Kaufman et al., 1997b; Levy et al., 2007a). As a result it is systematically in error for other land surface types where these `dark target’ and spectral ratio assumptions are invalid, including bare soil, deserts, snow, and urban areas. Internal TOA brightness tests attempt to identify such pixels, which are not processed by the algorithm.
The key references for this algorithm are Hsu et al. (2013, 2019); it was originally developed (Hsu et al., 2004) to fill some of the aforementioned bright surface coverage gaps in the DT product, and later extended to include vegetated pixels as well. DB has three different land surface reflectance models, and a combination of location and checks on the TOA reflectance determines which is used for a given pixel. The key spectral ranges used by the algorithm are the deep blue (412 nm, giving the algorithm its name), blue (470-490 nm), and red (650-675 nm).
Over traditionally `bright’ non-vegetated scenes such as deserts, bare soil, and some urban areas, the physical principle behind DB is that the surface reflectance is fairly stable in time at all three spectral ranges, and dark in the deep blue spectral region (Hsu et al., 2004). Thus a global data base of surface reflectance (at all 3 wavelengths) was created as a function of season, scattering angle, and NDVI, based on atmospherically-corrected satellite observations on low-AOD days with some assumed background aerosol loading. This data base is used to look up surface reflectance for suitable pixels. Note that snow surfaces are not dark in the blue and deep blue and so snow-covered pixels are excluded from processing.
Over vegetated surfaces, empirical ratio-based models similar to the DT-land algorithm are used. The main difference between these and DT are that DB uses an external land cover data base to separate between croplands and areas covered by predominantly natural vegetation, with separate spectral models developed for these two vegetated surface sub-types (Hsu et al., 2019). The 412 nm band is not used for these pixels.
A `hybrid’ method is applied over some regions, most commonly large urban areas or semi-vegetated transitional regions (Hsu et al., 2013). This uses the surface reflectance data base to obtain the surface reflectance, and then scales this using empirical polynomial bidirectional reflectance distribution functions (BRDFs) obtained offline by atmospheric correction of TOA reflectance near Aerosol Robotic Network (AERONET, Holben et al., 1998) sites. This essentially takes the magnitude and spectral variation from the data base method and imposes the angular variation (BRDF) observed from a nearby AERONET site.
Regardless of surface model, AOD retrieval proceeds the same way. Aerosol optical models are determined based on location, season, and spectral tests on the TOA reflectance (Hsu et al., 2019). For each individual wavelength, AOD is retrieved by finding the value needed to match TOA reflectance at that wavelength given the modeled surface reflectance. This is a non-iterative one-dimensional lookup of precalculated TOA reflectances. The AOD at 550 nm and AE are then determined by spectral interpolation and fitting of the AOD retrieved at the individual two (for vegetated surfaces) or three (for other surfaces) wavelengths.
A commonality between DB and DT-land is that the surface reflectance models are empirical and require training using satellite and AERONET data. A difference is that DT mixes two bimodal aerosol models to determine AOD for each pixel, while DB performs independent retrievals on each wavelength and interpolates spectrally. This increases the robustness of AE estimates from DB compared to DT, but means that DB aerosol optical models are not necessarily consistent between wavelengths (Sayer et al., 2013).
The SOAR (Sayer et al., 2012a, 2018) and DT-water (Tanré et al., 1997; Levy et al., 2013) algorithms share more similarities than the land algorithms. They are both multispectral weighted least-squares fits of precalculated radiative transfer results to TOA reflectance measurements, and both retrieve AOD and FMF.
The main conceptual difference between the two approaches is in aerosol optical models. DT-ocean has 5 (monomodal) fine-mode aerosol models and 4 coarse-mode models. It determines the combination of AOD and FMF best matching TOA reflectance for each of the potential (5x4=20) combinations of these, and reports AOD and FMF for not only the best-fitting combination but also average results for all combinations which match the TOA reflectance significantly well. Optical properties for these modes are given in Levy et al. (2009). In contrast, SOAR performs retrievals for each of four distinct aerosol models, representing conditions dominated by maritime (Sayer et al., 2012b), desert dust (Lee et al., 2017), smoke or haze (Sayer et al., 2012a), and mixed (Sayer et al., 2018a) aerosol types. These are all bimodal models based on climatologies of AERONET inversion products (Dubovik and King, 2000) and have their own valid range of AOD and FMF given by Sayer et al. (2018a). The model providing the best match to TOA reflectance measurements is reported in the product. Further, the SOAR dust model assumes spheroidal particles (Lee et al., 2017), while all DT modes assume spherical particles; consideration of nonsphericity is important for accurate retrievals of dust aerosols (e.g. Mishchenko et al., 1997), and spheroids are often used as an approximation for these effects. Neither algorithm imposes geographic constraints on where given optical models can be selected.A second difference is in the surface reflectance models. Both retrievals assume a water surface including wind-speed dependent whitecaps and Sun glint, and both include a contribution by scattering from within the water body; for DT-ocean this is a constant term while for SOAR it uses a Case I optical model driven by an ancillary data base of chlorophyll concentration (Sayer et al., 2018a). These models break down in turbid and shallow waters, which are brighter than assumed at visible wavelengths. In this case, DT-ocean checks for and does not process such pixels (Li et al., 2003), while SOAR reverts to a backup retrieval using only swIR bands which are less strongly affected by these issues (Sayer et al., 2018a). This backup retrieval has proven to be fairly robust (Sayer et al., 2018b).
TBD
Direct-Sun AOD observations from stationary Sun photometers within AERONET are used routinely to validate satellite AOD retrievals. This has been done many times for both the DT (e.g. Levy et al., 2010, 2013) and DB (e.g. Sayer et al., 2013, 2018a) algorithm families. The same approach will be applied for evaluation of PACE OCI aerosol retrievals. AERONET is a sparse but global network of instruments providing regular coverage suitable for validation of retrievals over land, and limited over-water validation from coastal or island sites. Typically, AERONET data are averaged in time and satellite retrievals in space (e.g. Ichoku et al., 2002). The uncertainty of AERONET AOD is sufficiently small (~0.02 in the UV and short visible, ~0.01 from the midvisible to swIR; Holben et al., 1998, Eck et al., 1999) to be able to meet PACE requirements. AE is also evaluated in this way, although strictly speaking this is not a direct validation because AERONET AE, as a spectral derivative measure, can be highly uncertain when AOD is low (Wagner and Silva, 2008).
AERONET direct-Sun AOD is also an input to the AERONET spectral deconvolution algorithm (SDA), which provides fine and coarse-mode AOD as well as FMF (O’Neill et al., 2003, 2006). The uncertainty of SDA-based FMF estimates depends on AOD and aerosol optical properties, but is around 0.1 for low to moderate aerosol loadings and decreases as AOD increases (O’Neill et al., 2001). This is also sufficient for PACE requirements.
Over water, essentially the same analysis is performed except using ship-based measurements collected on cruises as part of the Maritime Aerosol Network (Smirnov et al., 2009, 2011). These are made with hand-held Sun photometers and so uncertainty is slightly larger compared to AERONET land sites (~0.02 in the midvisible; Knobelspiesse et al., 2004) but still sufficient. The same suite of data products relevant to PACE OCI validation (i.e. spectral AOD, AE, and FMF) are available from MAN, and will be used to validate OCI aerosol products over open water.
Comparisons with other satellite data products (e.g. DB/DT algorithms applied to other sensors and retrievals from the PACE polarimeters), as well as relevant ground, ship, or airborne data that may be available, will also be performed.
Dubovik, O., and M. D. King (2000), A flexible inversion algorithm for retrieval of aerosol optical properties from Sun and sky radiance measurements, Journal of Geophysical Research, 105, 20,673–20,696, doi: 10.1029/2000JD900282
Eck, T. F., B. N. Holben, J. S. Reid, O. Dubovik, A. Smirnov, N. T. O'Neill, I. Slutsker, and S. Kinne, (1999), Wavelength dependence of the optical depth of biomass burning, urban, and desert dust aerosols. Journal of Geophysical Research, 104(D24), 31,333–31,349, doi: 10.1029/1999JD900923
Holben, B. N., et al. (1998), AERONET—A Federated Instrument Network and Data Archive for Aerosol Characterization, Remote Sens. Environ., 66(1), 1-16, doi: 10.1016/S0034-4257(98)00031-5
Hsu, N. C., S. C. Tsay, M. D. King, and J. R. Herman (2004), Aerosol properties over bright‐reflecting source regions. IEEE Transactions on Geoscience and Remote Sensing, 42(3), 557–569, doi: 10.1109/TGRS.2004.824067
Hsu, N. C., M.-J. Jeong, C. Bettenhausen, A. M. Sayer, R. Hansell, C. S. Seftor, J. Huang, and S.-C. Tsay (2013), Enhanced Deep Blue aerosol retrieval algorithm: The second generation, J. Geophys. Res. Atmos., 118, 9296–9315, doi: 10.1002/jgrd.50712
Hsu, N. C., J. Lee, A. M. Sayer, W. Kim, C. Bettenhausen, and S.-C. Tsay (2019). VIIRS Deep Blue aerosol products over land: Extending the EOS long‐term aerosol data records. Journal of Geophysical Research: Atmospheres, 124, 4026–4053, doi: 10.1029/2018JD029688
Ichoku, C., D. Chu, S. Mattoo, Y. Kaufman, L. Remer, L., Tanré, I. Slutsker, and B. Holben (2002), A spatio-temporal approach for global validation and analysis of MODIS aerosol products, Geophys. Res. Lett., 29, 1616, doi: 10.1029/2001GL013206
Kaufman, Y. J., D. Tanré, L. A. Remer, E. F. Vermote, A. Chu, and B. N. Holben (1997a), Operational remote sensing of tropospheric aerosol over land from EOS moderate resolution imaging spectroradiometer, J. Geophys. Res., 102(D14), 17051–17067, doi: 10.1029/96JD03988
Kaufman, Y. J., A. E. Wald, L. A. Remer, B.-C. Gao, R.-R. Li, and L. Flynn (1997b), The MODIS 2.1-μm channel-correlation with visible reflectance for use in remote sensing of aerosol, IEEE Trans. Geosci. Remote Sens., 35(5), 1286-1298, doi: 10.1109/36.628795
Li, R. R., Y. J. Kaufman, B.-C. Gao, and C. O. Davis (2003), Remote sensing of suspended sediments and shallow coastal waters, IEEE Trans. Geosci. Remote Sens., 41, 559–566, doi: 10.1109/TGRS.2003.810227
Lee, J., N. C. Hsu, A. M. Sayer, C. Bettenhausen, and P. Yang (2017), AERONET‐based nonspherical dust optical models and effects on the VIIRS Deep Blue/SOAR over water aerosol product. Journal of Geophysical Research: Atmospheres, 122, 10,384–10,401, doi:
Levy, R. C., L. A. Remer, S. Mattoo, E. F. Vermote, and Y. J. Kaufman (2007a), Second‐generation operational algorithm: Retrieval of aerosol properties over land from inversion of Moderate Resolution Imaging Spectroradiometer spectral reflectance, J. Geophys. Res., 112, D13211, doi: 10.1029/2006JD007811
Levy, R. C., L. A. Remer, and O. Dubovik (2007b), Global aerosol optical properties and application to Moderate Resolution Imaging Spectroradiometer aerosol retrieval over land, J. Geophys. Res., 112, D13210, doi: 10.1029/2006JD007815
Levy, R. C., L. A. Remer, D. Tanré, S. Mattoo, and Y. J. Kaufman (2009), Algorithm for remote sensing of tropospheric aerosol over dark targets from MODIS, collections 005 and 051, algorithm theoretical basis document, revision 2, available online at https://atmosphere-imager.gsfc.nasa.gov/sites/default/files/ModAtmo/ATBD_MOD04_C005_rev2_0.pdf
Levy, R. C., L. A. Remer, R. G. Kleidman, S. Mattoo, C. Ichoku, R. Kahn, and T. F. Eck (2010), Global evaluation of the Collection 5 MODIS dark-target aerosol products over land, Atmos. Chem. Phys., 10, 10399–10420, doi: 10.5194/acp-10-10399-2010
Levy, R. C., S. Mattoo, L. A. Munchak, L. A. Remer, A. M. Sayer, F. Patadia, and N. C. Hsu (2013), The Collection 6 MODIS aerosol products over land and ocean, Atmos. Meas. Tech., 6, 2989–3034, doi: 10.5194/amt-6-2989-2013
Mishchenko, M. I., L. D. Travis, R. A. Kahn, and R. A. West (1997), Modeling phase functions for dustlike tropospheric aerosols using a shape mixture of randomly oriented polydisperse spheroids, Journal of Geophysical Research, 102, 16,831–16,847, doi: 10.1029/96JD02110
O'Neill, N. T., O. Dubovik, and T. F. Eck, (2001), Modified ångström coefficient for the characterization of submicrometer aerosols. Applied Optics, 40(15), 2368–2375. doi: 10.1364/AO.40.002368
O'Neill, N. T., T. F. Eck, A. Smirnov, B. N. Holben, and S. Thulasiraman, (2003), Spectral discrimination of coarse and fine mode optical depth. Journal of Geophysical Research, 108(D17), 4559–4573, doi: 10.1029/2002JD002975
O'Neill, N., T. Eck, A. Smirnov, B. Holben, B. and Thulasiraman, S. (2006), Spectral deconvolution algorithm technical memo (Tech. rep.). Greenbelt, MD: NASA Goddard Space Flight Center, revision April 26, 2006, version 4, available online from http://aeronet.gsfc.nasa.gov/new_web/PDF/tauf_tauc_technical_memo1.pdf
Sayer, A. M., N. C. Hsu, C. Bettenhausen, Z. Ahmad, B. N. Holben, A. Smirnov, G. E. Thomas, and J. Zhang (2012a), SeaWiFS Ocean Aerosol Retrieval (SOAR): Algorithm, validation, and comparison with other data sets, J. Geophys. Res., 117, D03206, doi: 10.1029/2011JD016599
Sayer, A. M., A. Smirnov, N. C. Hsu, and B. N. Holben (2012b), A pure marine aerosol model, for use in remote sensing applications, Journal of Geophysical Research, 117, D05213. doi: 10.1029/2011JD016689
Sayer, A. M., N. C. Hsu, C. Bettenhausen, and M.-J. Jeong (2013), Validation and uncertainty estimates for MODIS Collection 6 "Deep Blue" aerosol data, J. Geophys. Res. Atmos., 118, 7864– 7872, doi: 10.1002/jgrd.50600
Sayer, A. M., N. C. Hsu, J. Lee, C. Bettenhausen, W. V. Kim, and A. Smirnov (2018a), Satellite Ocean Aerosol Retrieval (SOAR) algorithm extension to S‐NPP VIIRS as part of the "Deep Blue" aerosol project. Journal of Geophysical Research: Atmospheres, 123, 380–400 doi: 10.1002/2017JD027412
Sayer, A. M., et al. (2018b), Validation of SOAR VIIRS over‐water aerosol retrievals and context within the global satellite aerosol data record, Journal of Geophysical Research: Atmospheres, 123, 13,496–13,526, doi: 10.1029/2018JD029465
Smirnov, A., et al. (2009), Maritime aerosol network as a component of aerosol robotic network. Journal of Geophysical Research, 112, D06204, doi: 10.1029/2008JD011257
Smirnov, A., et al. (2011), Maritime aerosol network as a component of AERONET – first results and comparison with global aerosol models and satellite retrievals, Atmos. Meas. Tech., 4, 583–597, doi: 10.5194/amt-4-583-2011
Tanré, D., Y. J. Kaufman, M. Herman, and S. Mattoo (1997), Remote sensing of aerosol properties over oceans using the MODIS/EOS spectral radiances, J. Geophys. Res., 102(D14), 16971–16988, doi: 10.1029/96JD03437
Wagner, F., and A. M. Silva (2008), Some considerations about Ångström exponent distributions, Atmospheric Chemistry and Physics, 8, 481–489, doi: 10.5194/acp-8-481-2008
Sample data products are available on request.
