2.4.2.1 Satellite Retrievals

Satellite retrievals of aerosol optical depth in cloud-free regions have improved via new generation sensors (Kaufman et al., 2002) and an expanded global validation program (Holben et al., 2001). Advanced aerosol retrieval products such as aerosol fine-mode fraction and effective particle radius have been developed and offer potential for improving estimates of the aerosol direct radiative effect. Additionally, efforts have been made to determine the anthropogenic component of aerosol and associated direct RF, as discussed by Kaufman et al. (2002) and implemented by Bellouin et al. (2005) and Chung et al. (2005). However, validation programs for these advanced products have yet to be developed and initial assessments indicate some systematic errors (Levy et al., 2003; Anderson et al., 2005a; Chu et al., 2005), suggesting that the routine differentiation between natural and anthropogenic aerosols from satellite retrievals remains very challenging.

2.4.2.1.1 Satellite retrievals of aerosol optical depth

Figure 2.11 shows an example of aerosol optical depth τ_aer (mid-visible wavelength) retrieved over both land and ocean, together with geographical positions of aerosol instrumentation. Table 2.2 provides a summary of aerosol data currently available from satellite instrumentation, together with acronyms for the instruments. τ_aer from the Moderate Resolution Imaging Spectrometer (MODIS) instrument for the January to March 2001 average (Figure 2.11, top panel) clearly differs from that for the August to October 2001 average (Figure 2.11, bottom panel) (Kaufman et al., 1997; Tanré et al., 1997). Seasonal variability in τ_aer can be seen; biomass burning aerosol is most strongly evident over the Gulf of Guinea in Figure 2.11 (top panel) but shifts to southern Africa in Figure 2.11 (bottom panel). Likewise, the biomass burning in South America is most evident in Figure 2.11 (bottom panel). In Figure 2.11 (top panel), transport of mineral dust from Africa to South America is discernible while in Figure 2.11 (bottom panel) mineral dust is transported over the West Indies and Central America. Industrial aerosol, which consists of a mixture of sulphates, organic and black carbon, nitrates and industrial dust, is evident over many continental regions of the NH. Sea salt aerosol is visible in oceanic regions where the wind speed is high (e.g., south of 45°S). The MODIS aerosol algorithm is currently unable to make routine retrievals over highly reflective surfaces such as deserts, snow cover, ice and areas affected by ocean glint, or over high-latitude regions when the solar insolation is insufficient.

Figure 2.11. Aerosol optical depth, τ_aer, at 0.55 μm (colour bar) as determined by the MODIS instrument for the January to March 2001 mean (top panel) and for the August to October 2001 mean (bottom panel). The top panel also shows the location of AERONET sites (white squares) that have been operated (not necessary continuously) since 1996. The bottom panel also shows the location of different aerosol lidar networks (red: EARLINET, orange: ADNET, black: MPLNET).

Table 2.2. Periods of operation, spectral bands and products available from various different satellite sensors that have been used to retrieve aerosol properties.

Notes: ^a DRE is the direct radiative effect and includes both natural and anthropogenic sources (see Table 2.3). The Angstrom exponent, a, is the wavelength dependence of τ_aer and is defined by a = –ln(τ_aerλ1/τ_aerλ2) / ln(λ_1 / λ_2) where λ_1 = wavelength 1 and λ_2 = wavelength 2.

^{ b} TOMS followed up by the Ozone Monitoring Instrument (OMI) on the Earth Observing System (EOS) Aura satellite, launched July 2004.

Early retrievals for estimating τ_aer include the Advanced Very High Resolution Radiometer (AVHRR) single channel retrieval (e.g., Husar et al., 1997; Ignatov and Stowe, 2002), and the ultraviolet-based retrieval from the TOMS (e.g., Torres et al., 2002). A dual channel AVHRR retrieval has also been developed (e.g., Mishchenko et al., 1999; Geogdzhayev et al., 2002). Retrievals by the AVHRR are generally only performed over ocean surfaces where the surface reflectance characteristics are relatively well known, although retrievals are also possible over dark land surfaces such as boreal forests and lakes (Soufflet et al., 1997). The TOMS retrieval is essentially independent of surface reflectance thereby allowing retrievals over both land and ocean (Torres et al., 2002), but is sensitive to the altitude of the aerosol, and has a relatively low spatial resolution. While these retrievals only use a limited number of spectral bands and lack sophistication compared to those from dedicated satellite instruments, they have the advantage of offering continuous long-term data sets (e.g., Geogdzhayev et al., 2002).

Early retrievals have been superseded by those from dedicated aerosol instruments (e.g., Kaufman et al., 2002). Polarization and Directionality of the Earth’s Reflectance (POLDER) uses a combination of spectral channels (0.44–0.91 µm) with several viewing angles, and measures polarization of radiation. Aerosol optical depth and Ångstrom exponent (α) over ocean (Deuzé et al., 2000), τ_aer over land (Deuzé et al., 2001) and the direct radiative effect of aerosols (Boucher and Tanré, 2000; Bellouin et al., 2003) have all been developed. Algorithms for aerosol retrievals using MODIS have been developed and validated over both ocean (Tanré et al., 1997) and land surfaces (Kaufman et al., 1997). The uncertainty in these retrievals of τ_aer is necessarily higher over land (Chu et al., 2002) than over oceans (Remer et al., 2002) owing to uncertainties in land surface reflectance characteristics, but can be minimised by careful selection of the viewing geometry (Chylek et al., 2003). In addition, new algorithms have been developed for discriminating between sea salt, dust or biomass burning and industrial pollution over oceans (Bellouin et al., 2003, 2005; Kaufman et al., 2005a) that allow for a more comprehensive comparison against aerosol models. Multi-angle Imaging Spectro-Radiometer (MISR) retrievals have been developed using multiple viewing capability to determine aerosol parameters over ocean (Kahn et al., 2001) and land surfaces, including highly reflective surfaces such as deserts (Martonchik et al., 2004). Five typical aerosol climatologies, each containing four aerosol components, are used in the retrievals, and the optimum radiance signature is determined for nine viewing geometries and two different radiances. The results have been validated against those from the Aerosol RObotic NETwork (AERONET; see Section 2.4.3). Along Track Scanning Radiometer (ATSR) and ATSR-2 retrievals (Veefkind et al., 1998; Holzer-Popp et al., 2002) use a relatively wide spectral range (0.56–1.65 µm), and two viewing directions and aerosol climatologies from the Optical Parameters of Aerosols and Clouds (OPAC) database (Hess et al., 1998) to make τ_aer retrievals over both ocean and land (Robles-Gonzalez et al., 2000). The Ocean Colour and Temperature Scanner (OCTS) retrieval has a basis similar to the dual wavelength retrieval from AVHRR and uses wavelengths over the range 0.41 to 0.86 µm to derive τ_aer and α over oceans (e.g., Higurashi et al., 2000) using a bi-modal aerosol size distribution. The Sea-Viewing Wide Field-of-View Sensor (SeaWiFs) uses 0.765 µm and 0.856 µm radiances to provide τ_λ=0.856 and α over ocean using a bi-modal aerosol size distribution (M. Wang et al., 2005). Further SeaWiFs aerosol products have been developed over both land and ocean using six and eight visible channels, respectively (e.g., von Hoyningen-Heune et al., 2003; Lee et al., 2004).

Despite the increased sophistication and realism of the aerosol retrieval algorithms, discrepancies exist between retrievals of τ_aer even over ocean regions (e.g., Penner et al., 2002; Myhre et al., 2004a, 2005b; Jeong et al., 2005; Kinne et al., 2006). These discrepancies are due to different assumptions in the cloud clearing algorithms, aerosol models, different wavelengths and viewing geometries used in the retrievals, different parametrizations of ocean surface reflectance, etc. Comparisons of these satellite aerosol retrievals with the surface AERONET observations provide an opportunity to objectively evaluate as well as improve the accuracy of these satellite retrievals. Myhre et al. (2005b) showed that dedicated instruments using multi-channel and multi-view algorithms perform better when compared against AERONET than the simple algorithms that they have replaced, and Zhao et al. (2005) showed that retrievals based on dynamic aerosol models perform better than those based on globally fixed aerosol models. While some systematic biases in specific satellite products exist (e.g., Jeong et al., 2005; Remer et al., 2005), these can be corrected for (e.g., Bellouin et al., 2005; Kaufman et al., 2005b), which then enables an assessment of the direct radiative effect and the direct RF from an observational perspective, as detailed below.

2.4.2.1.2 Satellite retrievals of direct radiative effect

The solar direct radiative effect (DRE) is the sum of the direct effects due to anthropogenic and natural aerosol species while the direct RF only considers the anthropogenic components. Satellite estimates of the global clear-sky DRE over oceans have advanced since the TAR, owing to the development of dedicated aerosol instruments and algorithms, as summarised by Yu et al. (2006) (see Table 2.3). Table 2.3 suggests a reasonable agreement of the global mean, diurnally averaged clear-sky DRE from various studies, with a mean of –5.4 W m^–2 and a standard deviation of 0.9 W m^–2. The clear-sky DRE is converted to an all-sky DRE by Loeb and Manalo-Smith (2005) who estimated an all-sky DRE over oceans of –1.6 to –2.0 W m^–2 but assumed no aerosol contribution to the DRE from cloudy regions; such an assumption is not valid for optically thin clouds or if partially absorbing aerosols exist above the clouds (see Section 2.4.4.4).

Table 2.3. The direct aerosol radiative effect (DRE) estimated from satellite remote sensing studies (adapted and updated from Yu et al., 2006).

Notes:

^a SSM/I: Special Sensor Microwave/Imager; VIRS: Visible Infrared Scanner; ERBE: Earth Radiation Budget Experiment.

^b NESDIS: National Environmental Satellite, Data and Information Service.

Furthermore, use of a combination of sensors on the same satellite offers the possibility of concurrently deriving τ_aer and the DRE (e.g., Zhang and Christopher, 2003; Zhang et al., 2005), which enables estimation of the DRE efficiency, that is, the DRE divided by τ_aer (W m^–2 τ_aer^–1). Because the DRE efficiency removes the dependence on the geographic distribution of τ_aer it is a useful parameter for comparison of models against observations (e.g., Anderson et al., 2005b); however, the DRE efficiency thus derived is not a linear function of τ_aer at high τ_aer such as those associated with intense mineral dust, biomass burning or pollution events.

2.4.2.1.3. Satellite retrievals of direct radiative forcing

Kaufman et al. (2005a) estimated the anthropogenic-only component of the aerosol fine-mode fraction from the MODIS product to deduce a clear sky RF over ocean of –1.4 W m^–2. Christopher et al. (2006) used a combination of the MODIS fine-mode fraction and Clouds and the Earth’s Radiant Energy System (CERES) broadband TOA fluxes and estimated an identical value of –1.4 ± 0.9 W m^–2. Bellouin et al. (2005) used a combination of MODIS τ_aer and fine-mode fraction together with data from AeroCom (see Section 2.4.3) to determine an all-sky RF of aerosols over both land and ocean of –0.8 ± 0.2 W m^–2, but this does not include the contribution to the RF and associated uncertainty from cloudy skies. Chung et al. (2005) performed a similar satellite/AERONET/model analysis, but included the contribution from cloudy areas to deduce an RF of –0.35 W m^–2 or –0.50 W m^–2 depending upon whether the anthropogenic fraction is determined from a model or from the MODIS fine-mode fraction and suggest an overall uncertainty range of –0.1 to –0.6 W m^–2. Yu et al. (2006) used several measurements to estimate a direct RF of –0.5 ± 0.33 W m^–2. These estimates of the RF are compared to those obtained from modelling studies in Section 2.4.4.7.