Total Ozone Mapping Spectrometer measurements of aerosol absorption from space: Comparison to SAFARI 2000 ground-based observations



[1] The capability to detect the presence of absorbing aerosols in the atmosphere using space-based near-UV observations has been demonstrated in the last few years, as indicated by the widespread use by the atmospheric sciences community of the Total Ozone Mapping Spectrometer (TOMS) aerosol index as a qualitative representation of aerosol absorption. An inversion procedure has been developed to convert the unique spectral signature generated by the interaction of molecular scattering and particle absorption into a quantitative measure of aerosol absorption. In this work we evaluate the accuracy of the near-UV method of aerosol absorption sensing by means of a comparison of TOMS retrieved aerosol single scattering albedo and extinction optical depth to ground-based measurements of the same parameters by the Aerosol Robotic Network (AERONET) for a 2-month period during the SAFARI 2000 campaign. The availability of collocated AERONET observations of aerosol properties, as well as Micropulse Lidar Network measurements of the aerosol vertical distribution, offered a rare opportunity for the evaluation of the uncertainty associated with the height of the absorbing aerosol layer in the TOMS aerosol retrieval algorithm. Results of the comparative analysis indicate that in the absence of explicit information on the vertical distribution of the aerosols, the standard TOMS algorithm assumption yields, in most cases, reasonable agreement of aerosol optical depth (±30%) and single scattering albedo (±0.03) with the AERONET observations. When information on the aerosol vertical distribution is available, the accuracy of the retrieved parameters improves significantly in those cases when the actual aerosol profile is markedly different from the idealized algorithmic assumption.

1. Introduction

[2] The role of atmospheric aerosols in the global climate is one of the largest remaining sources of uncertainty in the assessment of global climate change. Aerosols directly affect the energy balance of the Earth-atmosphere system, through the processes of scattering of solar radiation, which redistributes the incoming solar energy in the atmosphere, and absorption (of both solar and infrared radiation), which transforms radiative energy into internal energy of the absorbing particles and heats up the atmosphere. In addition, aerosols, through their role as cloud condensation nuclei, have an indirect effect on climate by affecting the albedo and lifetime of clouds [Haywood and Boucher, 2000].

[3] The cooling effect of aerosols, associated with the backscattering to space of a fraction of the incoming solar energy, is considered to be a very important counteracting factor of the well known warming effect of the greenhouse gases. The absorption by aerosol particles of a fraction of the incident sunlight and, for certain aerosol types, infrared radiation, results in a heating of the atmosphere. Thus aerosol absorption reduces the cooling effect commonly associated with aerosol particles. Although, the impact of aerosol absorption on climate is still a subject of considerable debate [Penner et al., 2003], recently published theoretical analysis suggest that black carbon may be the second most important global warming substance (in terms of its direct radiative forcing effect) after carbon dioxide, and larger than methane [Jacobson, 2002]. The role of aerosol absorption effects on climate is, therefore, an issue that needs to be better understood in order to reduce the currently large uncertainties of its climatic effect.

[4] In this paper, we present and discuss the results of the application of the near-UV method to observations by the Total Ozone Mapping Spectrometer (TOMS), on board the Earth Probe (EP) satellite (also known as EP-TOMS), to retrieve aerosol single scattering albedo during the Southern African Regional Science Initiative (SAFARI 2000) campaign [Swap et al., 2002]. A brief discussion of some theoretical aspects associated with the quantification of aerosol absorption is carried out in section 2, followed by a short review of commonly used remote sensing approaches to measure aerosol absorption in section 3. In section 4, the physical basis of the near-UV approach is given, and a brief description of the TOMS aerosol algorithm is presented. In section 5, the TOMS retrieved single scattering albedo and optical depth are compared to collocated retrievals of the same parameters from observations by the Aerosol Robotic Network (AERONET). The last section includes additional discussion, a summary of results and conclusions of the analysis.

2. Absorbing Aerosol Components

[5] The fundamental microphysical property driving particle absorption is the imaginary component of the refractive index, k, of the aerosol components. The absorption coefficient of aerosol particles depends on k, the size of the particles, and the manner in which the different components are incorporated into the mixture, i.e., internal or external mixtures [Sokolik and Toon, 1999]. A commonly used macroscopic measure of aerosol absorption is the single scattering albedo, ω0, defined as the ratio of the scattering to extinction (absorption plus scattering) coefficients. For a detailed discussion of the physical aspects of absorption by particles, the reader is directed to the review paper by Horvath [1993].

[6] Atmospheric aerosols absorb radiation over a very wide region of the solar spectrum, from the ultraviolet (UV) to the near-infrared (IR). The aerosol component that contributes the most to particle absorption of solar radiation is black carbon or soot, found in many aerosol types of anthropogenic origin such as urban-industrial aerosols and biomass burning. Black carbon absorption depends only weakly on wavelength over the near-UV to near-IR spectral region, where the soot imaginary refractive index is relatively constant [Twitty and Weinman, 1971; Bergstrom et al., 2002].

[7] Hematite and other iron oxides are the main absorbing components of desert dust aerosols. Unlike soot, the imaginary refractive index of desert dust is largest in the ultraviolet and decreases rapidly with wavelength [Patterson et al., 1977; Alfaro et al., 2004]. Early measurements of the imaginary refractive index of Saharan dust reported significant absorption from the UV to the visible [Patterson et al., 1977]. Recent results, however, using both remote sensing as well as in situ methods, indicate that Saharan mineral dust absorption in the UV is significantly less than previously thought [Colarco et al., 2002; Sinyuk et al., 2003; Alfaro et al., 2004]. In the visible, a combination of satellite and ground-based remote sensing [Kaufman et al., 2001], and airborne measurements over the Atlantic ocean [Haywood et al., 2003] produced single scattering albedo values of Saharan dust close to unity, significantly larger than the values associated with the Patterson et al. [1977] data on the imaginary component of refractive index. Airborne in situ measurements in the visible of mineral aerosol of Asian origin over the North Pacific [Anderson et al., 2003] also produced single scattering albedo estimates close to one.

[8] Organic material, a component of some natural aerosol types, has also been suggested as an important absorbing species, mainly in the ultraviolet [Jacobson, 1999]. A few measurements of the absorption effects of organic aerosol types have been conducted [Lyubovtseva, 2002]; however, current knowledge of their optical properties is very poor.

3. Measuring Aerosol Absorption

[9] Aerosol absorption can be directly measured by the application of a variety of optical techniques both in situ and in the laboratory. For a detailed description of commonly used direct methods of measuring aerosol absorption, the reader is referred to the review papers by Horvath [1993] and by Heintzenberg et al. [1997].

[10] Inverse ground-based methods to measure column-integrated single scattering albedo using solar radiation observations have been applied. Shadow band radiometer observations of the direct and diffuse components of the incident solar flux, in conjunction with radiative transfer calculations have been used to infer ω0 [Herman et al., 1975; King, 1979]. Eck et al. [1998], produced estimates of the single scattering albedo for biomass burning aerosols in Amazonia by matching measured to model computed irradiances using as input the observed aerosol optical depth and particle size distribution. Another method to derive aerosol absorption using a combination of direct Sun measurements and sky radiances has been developed as a part of the Aerosol Robotic Network [Holben et al., 1998]. The technique infers the aerosol particle size distribution and complex refractive index by fitting measurements of aerosol optical depth and sky radiances at four wavelengths (0.44, 0.67, 0.87 and 1.02 μm) to radiative transfer calculations [Dubovik and King, 2000].

[11] The use of satellite observations to measure aerosol absorption has also been attempted. Kaufman [1987] developed a method to retrieve aerosol single scattering albedo, based on the measurement of the radiance change in the visible between a clear and a hazy day over a varying surface reflectance. It was shown that for a certain critical value of the surface reflectance, the net atmospheric effect due to the competing processes of scattering and absorption of light is almost independent of aerosol optical thickness, and is mainly determined by the aerosol single scattering albedo. Although this method has been shown to work on a few cases, its practical applicability is hindered by the limited spatial variability of the actual surface reflectance, that makes the identification of the critical reflectance value a difficult task. Recently, another method to measure aerosol absorption from space in the visible spectral region has been suggested [Kaufman et al., 2002]. The method proposes the use of a satellite viewing configuration to measure the upwelling radiance on and off the Sun glint regions over the oceans. The off-glint measurements would be used to characterize the aerosol scattering properties, whereas the strong Sun glint background would be used to quantify the aerosol absorption effects. This approach would not measure particle absorption over land where most sources of anthropogenic aerosols are located.

[12] Aerosol absorption from space can also be measured using observations of backscattered near-UV radiation [Torres et al., 1998]. This approach works equally well over the oceans and the continents, and is the only currently available technique to globally detect and characterize aerosol absorption from space. A discussion of results of aerosol absorption measurements by the near-UV method is the central theme of this paper.

4. Satellite Measurement of Aerosol Absorption in the Near-UV

4.1. Physical Basis

[13] The near-UV method of aerosol absorption sensing from space derives its sensitivity from the interaction of the large molecular scattering component characteristic of this spectral range, and absorption by the aerosol particles of Rayleigh scattered photons. As shown in Figure 1, the Rayleigh scattered radiation emanating from the scattering effects of the molecular atmosphere, peaks in the near-UV spectral region, and decreases rapidly with wavelength. At 340 nm the Rayleigh component is about an order of magnitude larger than at 600 nm. Also shown is the calculated spectral dependence of the surface reflected radiation, for typical surface types. The well-known peak in the green associated with the reflectance of vegetated surfaces, does not show up here because of the coarse spectral resolution of the calculations in that spectral region (∼50 nm). Note that in the near-UV, unlike in the visible and near-IR spectral regions, the surface contribution is significantly smaller than the Rayleigh scattering component even over vegetated and desert surfaces. The low near-UV reflectance of land and water surfaces facilitates the sensing of aerosols over the oceans and the continents, including the arid and semiarid regions of the world.

Figure 1.

Model calculations illustrating the spectral dependence of the upwelling reflectance at the top of the atmosphere. The Rayleigh scattering contribution is shown as the solid line. The contribution of different surface types is also shown: ocean (dotted line), vegetation (dashed line), and desert (dot-dash line).

[14] To illustrate the interaction of the radiative transfer effects between the molecular atmosphere, the surface, and the aerosols, as measured by a satellite borne sensor, it is useful to express the upwelling radiance at the top of the atmosphere (TOA) in terms of the molecular scattering and surface reflection components. In the absence of aerosols and clouds, the TOA “background” radiance, Iλ, is given by the combination of the Rayleigh scattering by the molecular atmosphere image and the reflection by the underlying surface, image

equation image

When the atmosphere contains aerosol particles, the TOA radiance is modified by the aerosols scattering and absorption effects, as shown by the approximation

equation image

The first term on the right side of equation (2) is the singly scattered radiance added by the aerosol layer, a function of the incoming solar flux, πF0, the aerosol scattering phase function, P(θ), single scattering albedo, ω0, and extinction optical depth, τ. The parameters μ and μ0 represent the cosines of the satellite and solar zenith angles respectively. The second term represents the attenuation of the Rayleigh scattered and surface reflected radiance by the absorption effects of the aerosol layer. Since the molecular scattering component, depends on atmospheric pressure, the attenuation of image by aerosol absorption also depends on the height of the aerosol layer. Therefore the second term of equation (2) includes only the attenuation of that fraction of molecular scattering originating from within and underneath the aerosol layer. The third term accounts for the molecular scattering above the aerosol layer where the attenuation effect of the particles is significantly reduced. The terms pa and ps represent respectively the pressure at the level of the aerosol layer, and at the surface. (A previously published version of equation (2) [Torres et al., 2002a], incorrectly extended the pressure scaling of the aerosol absorption effect to the surface term image

[15] In this approximation, particle multiple scattering, as well as other second-order terms (reflected and then scattered radiation and vice versa), have been neglected. Therefore, strictly speaking, this approximation is only valid when the aerosol optical depth is very small, and particle multiple scattering can be ignored. We use it here just to illustrate the Rayleigh-scattering particle absorption interaction that is the backbone of the near-UV method to characterize aerosol absorption.

[16] The net aerosol induced reflectance change, ΔIλ, is given by the combined effect of the competing processes of aerosol scattering and the attenuation by aerosol absorption effects of the molecular and surface components, as shown below

equation image

Figure 2 shows the spectral dependence of the scattering and attenuation terms as calculated using equation (3) for an absorbing aerosol layer at 3 km above the surface. The aerosol properties used in the calculations are representative of carbonaceous particles resulting from biomass burning. For the aerosol model used, the calculated aerosol single scattering contribution is a weak function of wavelength resulting from the opposite spectral dependence of the optical depth and scattering phase function terms (not shown). The calculated spectral dependence is not a general result, and may be different for other aerosol models. The attenuation term is very large in the near-UV and converges rapidly to near zero as the wavelength increases. The large near-UV multiple molecular scattering component, increases the length of photon paths through the aerosol, and provides a “medium” in which aerosol absorption effects are clearly observed and can be measured. Owing to the mutual cancellation by the scattering and absorption processes, the resulting net effect (solid line in Figure 2) is small in the in the near-UV (it can actually be negative for strongly absorbing aerosols). The observed spectral contrast at two near-UV wavelengths, is the quantity commonly known as the absorbing aerosol index, routinely derived from TOMS observations [Herman et al., 1997]. At visible wavelengths, where the attenuation effect is negligible, the net result is just the single scattering contribution. Note that for nonabsorbing aerosols (i.e., ω0 = 1), the second term of equation (5) becomes zero and the net aerosol effect is just the single scattering term. A similar attenuation effect can be observed, at all wavelengths, for an enhanced contribution of the surface term (Is), as when the absorbing aerosol layer lies above an ice/snow covered surface, a cloud deck, or against the highly reflective background provided by Sun glitter effects [Kaufman et al., 2002].

Figure 2.

Spectral dependence of the aerosol single scattering contribution (dashed line), the attenuation of the Rayleigh component by aerosol absorption (dot-dashed line), and the net aerosol effect (solid line).

4.2. TOMS Algorithm

[17] On the basis of the previously discussed physical basis, an approach to retrieve aerosol properties using TOMS measurements in the near-ultraviolet spectral region has been developed [Torres et al., 1998, 2002a]. In the TOMS near-UV method, measurements of the backscattered radiance (I) at two wavelengths λ1 and λ2, (λ2 > λ1) in the range 330–380 nm are used. Aerosol particles are characterized by examining the variability of the relationship between the spectral contrast (Iλ1/Iλ2) and the radiance at the longer wavelength (Iλ2). The inversion procedure uses a set of precomputed look up tables (LUTs) of radiances emerging at the top of an aerosol-laden atmosphere. The LUTs were generated using the University of Arizona radiative transfer code [Herman et al., 1995] that fully accounts for multiple scattering and polarization effects.

[18] The TOMS aerosol algorithm uses a combination of spectral and geographical location considerations to select, for each pixel, one of four possible aerosol types: desert dust, carbonaceous, urban-industrial and sulfate aerosols. Each aerosol type is represented by a set of aerosol models characterized by an assumed particle size distribution and fixed real component of the refractive index (spectrally independent). The relative spectral dependence of the imaginary component of the refractive index in the range λ1–λ2 is also prescribed. The imaginary refractive index at the longer wavelength is allowed to vary in the retrieval process, to capture the temporal and spatial variability of the aerosol absorption capacity.

[19] The assumed vertical distribution varies based on the selected aerosol type. For carbonaceous and desert dust particles, the aerosol load is assumed to be vertically distributed following a Gaussian function characterized by a peak (aerosol layer height) and a half-width (aerosol layer geometric thickness) values. For sulfate and urban-industrial aerosols, the aerosol concentration is largest at the surface and decreases exponentially with height. When the aerosols are identified as of the carbonaceous type, the aerosol layer is assumed to be at 3.0 km sea level. A global climatology of aerosol layer height derived from transport model calculations [Ginoux et al., 2004] is used for desert dust aerosols.

[20] Surface effects are characterized making use of a climatology of minimum near-UV reflectivity compiled from Nimbus7/TOMS observations [Herman and Celarier, 1997]. No spectral dependence is assumed in the 330–380 nm range. A recent analysis using Global Ozone Monitoring Experiment (GOME) data shows that for most land surface types, the near-UV reflectivity is only weakly wavelength-dependent [Koelemeijer et al., 2003]. A weak spectral dependence in the near-UV prevails over coastal waters. Over the open oceans, however, a larger spectral dependence is observed associated with the effects of clear water absorption in the near-UV [Litjens et al., 1999].

[21] The retrieved parameters are the total optical depth and the imaginary component of the refractive index at the longer wavelength. For historical reasons, and to facilitate validation studies using AERONET data, the retrieved parameters are reported at 380 nm. The aerosol single-scattering albedo associated with the retrieved imaginary refractive index and the other assumed aerosol properties, is calculated.

[22] The near-UV algorithm has been applied to the multiyear-long record of observations by the TOMS sensor onboard the Nimbus-7 spacecraft (1979–1993) and the Earth Probe platform (1996 to present). The aerosol optical depth product has been validated using ground-based Sun photometer measurements [Torres et al., 2002a, 2002b], and compared to airborne Sun photometer observations during the SAFARI 2000 [Schmid et al., 2003] and the PRIDE [Livingston et al., 2003] field campaigns. Comparison of aerosol optical depth to model calculations [Chin et al., 2002], and to other satellite retrievals [Myhre et al., 2004] have also been carried out. The long-term TOMS record on atmospheric aerosol load has been used to identify aerosol increases in regions of China and India [Massie et al., 2004]. In this paper results of the absorption optical depth retrievals are discussed and compared to independent observations.

5. Evaluation of TOMS Retrievals During SAFARI 2000

5.1. Biomass Burning Aerosols Observations

[23] The SAFARI 2000 field campaign [Swap et al., 2002] took place over a vast region of central and southern Africa covering nine African countries during the period 13 August to 25 September 2000. Several ground-based [Eck et al., 2003], airborne [Schmid et al., 2003], and satellite observations were carried out to measure the properties of the dense carbonaceous aerosol layer that during the dry season covers this region of the world as a result of long established land clearing agricultural practices. In this section we present results of aerosol absorption measurements using TOMS observations, and evaluate their accuracy by comparison with AERONET's ground-based measurements. We also examine in detail the sensitivity to the aerosol vertical distribution, by incorporating into the analysis actual lidar measured aerosol profiles.

5.1.1. TOMS Retrievals

[24] Aerosol optical depth and single scattering albedo were obtained by inversion of the TOMS observations at 331 and 360 nm using the algorithm described in section 4. On the basis of the geographical considerations used by the algorithm to differentiate between desert dust and biomass burning particles, the aerosol load was identified as composed of carbonaceous particles over the entire area of the campaign. The particle size distribution used by the algorithm to characterize this aerosol type is a bimodal function with mode radii of 0.087 and 0.567, and standard deviations of 1.537 and 2.203 for the fine and coarse modes respectively. These parameters were compiled from AERONET's long-term statistics for biomass burning aerosols [Dubovik et al., 2002]. The real refractive index is 1.5, and the imaginary component varies between 0.0 and 0.05. Both refractive index components are assumed to be spectrally independent.

[25] Figure 3 shows the aerosol spatial distribution on 5 September, as obtained from TOMS observations. The dark gray shaded areas correspond to scenes affected by severe cloud contamination. The data gap observed in the middle of the map (white areas) is due to the lack of coverage by contiguous orbits. The horizontal extent of the smoke layer as given by the aerosol index is shown in Figure 3a. The aerosol index is positive for elevated (about 2 km or higher) absorbing particles, and negative for low altitude weakly absorbing aerosols.

Figure 3.

Geographical distribution of aerosol properties over southern Africa on 5 September 2000 as derived from TOMS observations. Shown quantities are: (a) aerosol index, (b) extinction optical depth, (c) single scattering albedo, and (d) the absorption optical depth. The aerosol layer height was assumed to be 3 km above sea level. Dark gray shading indicates severe cloud contamination. White areas correspond to missing data, mostly due to lack of interorbital coverage.

[26] The inversion algorithm allows the quantitative characterization of the aerosol load even for those conditions when the aerosol index is negative. The increased sensitivity to low altitude and weakly absorbing aerosols in the retrieval algorithm, is the result of using the single channel measurement (see section 4.2) in addition to the spectral ratio (associated with the aerosol index). Thus, for clear conditions, the results of the inversion procedure allows for a better characterization of weakly absorbing aerosols, and aerosol layers at lower altitudes, than possible when only the aerosol index is used [Torres et al., 1998]. Since cloud contamination was not a major retrieval obstacle on this day, the optical depth and single scattering albedo maps in Figures 3b and 3c, show a larger spatial coverage than the aerosol index as discussed above. The retrieved aerosol optical depth (AOD) and single scattering albedo (SSA) maps show data over the Atlantic Ocean as well as over the eastern part of the continent where the aerosol index is negative and fails to positively detect the absorbing aerosol signal. Optical depth values as high as 3.0, were retrieved in the densest part of the layer. SSA values as low as 0.82 are observed; however, the most predominant values are 0.86–0.90. Figure 3d shows the derived optical depth of absorption. Values as high as 0.5 are clearly observed.

5.1.2. AERONET Measurements

[27] During SAFARI 2000, the AERONET project increased the density of CIMEL Sun photometers to ensure a good coverage over the entire area of the campaign. The geographical coordinates and elevation of the sites used in this analysis are listed in Table 1. AERONET uses direct-Sun irradiance measurements at a 15 min interval to measure aerosol optical depth at 340, 380, 440, 500, 670, 870 and 1020 nm. In addition to the solar flux observations, the AERONET instrumentation also measures almucantar radiances at 440, 670, 870 and 1020 nm, which are used as input to an inversion algorithm [Dubovik et al., 2000] to retrieve column averaged particle size distribution, and the complex refractive index of the tropospheric aerosol load. The AERONET-retrieved negligible spectral dependence of the imaginary component of refractive index for carbonaceous aerosols in several different regions of the world, including the African Savanna [Dubovik et al., 2002], is consistent with the TOMS algorithmic assumption of wavelength-independent refractive index at 331 and 360 nm. AERONET reported results indicate no spectral dependence in the range 440–1020 nm, consistent with Twitty and Weinman [1971], who also show that the lack of spectral dependence of the imaginary refractive index of soot extends to the near-UV region.

Table 1. Summary of TOMS-AERONET Comparison of Single Scattering Albedo Retrievals During SAFARI 2000
AERONET SiteLatitude, LongitudeElevation, m aslAERONET TotalTOMS TotalCoincidenceAERONET AverageTOMS AverageDifference
Mongu−15, 2311074844360.910.900.01
Mwinilunga−11, 2414301835130.900.91−0.01
Ndola−12, 2812702223130.870.91−0.04
Senanga−16, 2310253543240.880.90−0.02
Solwezi−12, 2613331835120.870.90−0.03
Zambezi−13, 231040244290.870.91−0.04
Inhaca−20, 3273151050.890.92−0.03
Skukuza−24, 3115020880.880.91−0.03

5.1.3. MPLNET Backscatter Profiles

[28] The NASA Micropulse Lidar Network (MPLNET) [Welton et al., 2001] consists of micropulse lidar (MPL) sites colocated with AERONET Sun photometers. During SAFARI 2000, two MPL systems were deployed to Mongu, Zambia and Skukuza, South Africa. The MPL system is a single channel (523 nm), autonomous, eye-safe lidar system that is used to determine the vertical structure of clouds and aerosols. Raw MPL data were acquired at 1-min time resolution, and 75 m vertical resolution. The raw data were converted into uncalibrated lidar signals and used to infer the altitude of aerosol and cloud heights. Aerosol extinction profiles were calculated for times coincident with AERONET aerosol optical depth measurements using the algorithm described by Welton et al. [2000].

5.2. Validation Analysis

[29] The simultaneous availability of AERONET measurements of aerosol optical depth and single scattering albedo, and the MPLNET information on aerosol vertical distribution, a source of uncertainty in the near-UV method, provided an excellent opportunity to evaluate the sensitivity of the TOMS retrieval products to the assumed aerosol profile.

5.2.1. Optical Depth Comparison

[30] Figure 4 shows a comparison of the TOMS retrieved optical depth to AERONET measurements at the eight sites listed in Table 1 for the standard assumption on the aerosol vertical distribution, i.e., aerosol layer at 3 km above sea level. The dotted lines indicate the expected accuracy (the larger of 0.1 and 30%) according to a previously published sensitivity analysis [Torres et al., 1998]. Of the total 127 comparison points, 104 (or 82%) are within the expected accuracy limits. This level of agreement is the same as reported for different aerosol types in other published validation analysis [Torres et al., 2002a, 2002b].

Figure 4.

Comparison of TOMS retrieved to AERONET measured aerosol optical depth at 380 nm during SAFARI 2000. Measurements at eight sites were used in the analysis. The one-to-one correspondence is shown as the solid line. The dotted lines indicate the TOMS estimated level of uncertainty: largest of 0.1 or 30% of the ground-based measurement.

5.2.2. Single Scattering Albedo

[31] A comparison of the aerosol single scattering albedo from AERONET and TOMS observations during SAFARI 2000 was carried out. Because of their reliability, AERONET sky radiance measurements at large solar zenith angles are preferentially used in the retrieval algorithm. For this reason, most single scattering albedo retrievals use either morning or afternoon measurements and, therefore, an exact collocation at the time of the satellite overpass (about 1030 local time (LT)) is not always possible. The evaluation of the space and ground-based retrievals is carried out by comparing the daily average single scattering albedo from the AERONET retrieval to the TOMS retrievals averaged over a 1° × 1° box centered at the AERONET site. A total of 120 coincident single scattering albedo observations were compiled (see Table 1). Because of the lack of a precise time collocation, the diurnal variability in single scattering albedo and/or planetary boundary layer height may affect the comparison. The TOMS 380 nm single scattering albedo value is converted to 440 nm, the shortest wavelength for which AERONET single scattering albedo retrievals are reported. The adjustment (∼0.01 reduction) was derived performing Mie calculations at 440 nm for the particle size distribution and real refractive index assumed in the look up tables for carbonaceous aerosols, and the imaginary component retrieved by the TOMS algorithm. The underlying assumption of no spectral dependence of the imaginary refractive index in the 380–440 nm range is supported by direct observations [Twitty and Weinman, 1971] and observations-based estimates [Bergstrom et al., 2002].

[32] Figure 5 shows the difference of AERONET and TOMS retrieved single scattering albedo as a function of time, at the eight sites used in this analysis. The differences are reported for three aerosol layer height assumptions (1.5, 3.0 and 6.0 km above sea level) used in the TOMS algorithm. The thin solid lines at ±0.03 indicate the expected accuracy level of the AERONET retrieval [Dubovik et al., 2000]. The 3 km TOMS retrieved single scattering albedo at Mongu, agrees with AERONET result within 0.03. The 3 km standard assumption seems to work reasonably well from the end of August up to about 25 September. A height value increasingly higher than 3 km and approaching 6 km would be required to get a good match of the two retrievals between the end of September and mid-October. Again, an altitude between 3 and 6 km would produce a good match toward the end of the comparison period. The 3 km assumption also works reasonably well at Mwinilunga, yielding retrieval differences within the AERONET uncertainty. At the Ndola and Senanga sites, an aerosol layer height between 1.5 and 3 km would yield an excellent match. The standard 3 km assumption, however, still produces agreement with AERONET within 0.03, although the TOMS retrieval is systematically larger the AERONET result. At the last two sites (Solwezi and Zambezi), it is clear than the aerosol layer at 1.5 km does a much better job than the 3 km assumption. At these sites, the TOMS-AERONET difference using the 3 km standard assumption is within 0.05 of the AERONET value. The few retrievals at the Inhaca and Skukuza sites, show again a better agreement with the TOMS retrieval at 1.5 km than when using the standard 3 km assumption.

Figure 5.

Difference between TOMS retrieved single scattering albedo (using different aerosol layer height assumptions) and AERONET inferred values at the eight sites used in the analysis. Dotted line, 1.5 km; solid line, 3.0 km; dashed line, 6.0 km.

[33] The observed retrieval differences at Mongu and Senanga are an intriguing feature of this analysis. In spite of being the closest two AERONET sites (less than 100 km apart), the comparison suggests 3 km as the most likely aerosol height at Mongu, whereas a lower aerosol layer height seems to prevail at Senanga. However, the AERONET single scattering albedo value at Mongu is systematically higher, by about 0.03, than the value measured at Senanga, in spite of the proximity of the two locations. In a recently published analysis of the variability of AERONET derived optical properties during SAFARI 2000, the Mongu data was excluded from the analysis due to some uncertainty on the sky radiometer calibration data [Eck et al., 2003]. Thus there is the possibility that the apparent difference in inferred aerosol layer height at these two sites may be a result of a calibration offset of the AERONET sky radiance measurements at Mongu. It should be pointed out, however, that the 3 km assumption used in the TOMS retrieval, also yields excellent agreement in the TOMS-AERONET optical depth comparison at Mongu, as discussed in the previous section. The AERONET optical depth observation uses direct Sun measurements and, therefore, it is not affected by any calibration problem of the sky radiance measurements.

[34] Figure 6 shows a scatter diagram of the AERONET and TOMS retrieved single scattering albedo values at the eight sites, for the TOMS standard retrieval (i.e., aerosol layer at 3 km). An underestimation of the TOMS retrieved single scattering albedo value occurs when the actual aerosol layer altitude is higher than the assumed value. However, when the actual aerosol height is lower than the assumption, the satellite derived single scattering albedo is overestimated. The observed retrieval difference between the two techniques is also a function of the single scattering albedo magnitude (as given by the AERONET retrieval). This is a confirmation of previously published results [Torres et al., 1998], showing that the sensitivity of the TOMS single scattering albedo retrieval to aerosol layer height increases as the aerosol absorption increases. A total of 63% of the comparison cases produce differences within the uncertainty of the AERONET retrieval (0.03) whereas 87% fall within 0.05.

Figure 6.

Comparison of single scattering albedo retrieved using the near-UV method to ground-based retrievals using AERONET's Almucantar measurements. Solid lines indicate TOMS-AERONET differences of ±0.03. Dashed lines correspond to TOMS-AERONET differences of ±0.05. See text for details.

[35] A summary of the AERONET-TOMS comparison of single scattering albedo during SAFARI 2000 for the eight sites used in the analysis is shown in Table 1. The number of AERONET and TOMS retrievals available during the campaign are given in columns 3 and 4 respectively. The number of coincidences, days when both retrievals were performed, is given in column 5. The time-averaged value of the AERONET retrievals is listed in column 6, and column 7 shows the corresponding TOMS average value. The TOMS retrievals associated with the standard 3 km layer height assumption was used in the averaging. The difference between the AERONET and TOMS average values is shown in column 8.

5.3. Evaluation of Sensitivity to Aerosol Profile

[36] Backscatter profiles measured by the MPLNET lidar at Mongu were used to evaluate the performance of the TOMS retrieval algorithm when an accurate representation of the aerosol vertical distribution is available. On four days, measured aerosol profiles were available within 30 min of the satellite overpass time. The normalized actual profiles, shown in Figure 7, were used in the calculation of special lookup tables for the retrieval algorithm, instead of the standard tables using idealized vertical distributions. Note that, except for the case of 15 September, the MPLNET measured profiles, show a well-defined single aerosol layer. On 9–15 September, however, the measured aerosol distribution shows a two-layer distribution, markedly different from the one-layer profile assumed in the TOMS aerosol algorithm.

Figure 7.

Normalized MPLNET profiles (solid lines) at Mongu on (a) 2 September, (b) 4 September, (c) 15 September, and (d) 17 September used in the radiative transfer calculations. The dashed lines depict the idealized Gaussian vertical distribution used in the TOMS retrieval algorithm. The normalized profiles are truncated at the terrain elevation above sea level indicated by the horizontal dotted lines.

[37] A comparison of TOMS and AERONET retrieved values of single scattering albedo on the four days when the MPLNET profiles were available is shown in Table 2. The AERONET measurements are listed in column 2. Columns 3, 4, and 5 show the absolute AERONET-TOMS difference for the algorithm assumptions of 1.5, 3.0 and 6.0 km respectively, in aerosol layer height. The AERONET-TOMS difference when the actual profile (as given by the lidar measurements) was used in the retrieval procedure is given in column 6. The space-based retrieval results when the standard 3 km assumption is used, yield values within 0.01 of the ground-based method. The use of the actual profiles, marginally improves the comparison to an almost perfect agreement. A graphical depiction of these results is shown in Figure 8.

Figure 8.

Difference in AERONET and TOMS retrievals of single scattering albedo at Mongu as a function of time during SAFARI 2000. The solid line shows the difference for the TOMS standard assumption of an aerosol layer at 3 km. The asterisks indicate the differences when the actual MPLNET profiles (see Figure 7) are used in the retrieval.

Table 2. Evaluation of Sensitivity of TOMS Retrieved Single Scattering Albedo to Aerosol Vertical Distributiona
DateAERONETTOMS Zaer = 1.5 kmTOMS Zaer = 3.0 kmTOMS Zaer = 6.0 kmTOMS Actual Profile
  • a

    Absolute AERONET-TOMS differences for different assumptions on aerosol layer height are shown in columns 3–6.

2 Sept. 20030.880.02−0.01−0.05−0.01
4 Sept. 20030.880.040.01−0.030.00
15 Sept. 20030.900.01−0.01−0.030.00
17 Sept. 20030.920.030.01−0.020.01

[38] Table 3 shows the performance of the algorithm in the retrieval of the aerosol optical depth using the assumed and the actually measured aerosol vertical distribution. Column 2 shows the AERONET observed optical depth. The relative AERONET-TOMS difference for the algorithm assumptions of 1.5, 3.0 and 6.0 km in aerosol layer height, are shown in columns 3, 4, and 5 respectively. The obtained differences when the actual profiles are used, are shown in column 6. On two days (4 and 15 September), the use of the actual profile produced an obvious improvement of the accuracy of the TOMS retrieved aerosol optical depth. Not surprisingly, the largest improvement (from −45% to 13%) was observed on 15 September, the day when the actual profile was very different from the idealized one-layer distribution used in the standard retrieval. On 4 September, when the actual profile was wider that the Gaussian assumption, the AERONET-TOMS difference improved from −13% to 2%. The excellent AERONET-TOMS agreement found for the 3 km assumption on 2 and 17 September, when the assumed profiles resembled more the actual distributions, deteriorated slightly when the actual profiles were used in the retrieval. This effect could be the result of the elimination of cancellation of errors from other sources when the actual profile is used.

Table 3. Evaluation of Sensitivity of TOMS Retrieved Optical Depth to Aerosol Vertical Distributiona
DateAERONETTOMS Zaer = 1.5 kmTOMS Zaer = 3.0 kmTOMS Zaer = 6.0 kmTOMS Actual Profile
  • a

    The AERONET-TOMS differences in percent (relative to AERONET) for different assumptions on aerosol layer height are shown in columns 3–6.

2 Sept. 20031.18−2752212
4 Sept. 20032.11−78−13172
15 Sept. 20032.66−90−45−1313
17 Sept. 20032.21−2521917

6. Summary and Conclusions

[39] The performance of the near-UV approach of measuring aerosol absorption from space has been evaluated making use of AERONET ground-based observations of optical depth and single scattering albedo during SAFARI 2000. The availability of lidar profiles during the campaign allowed a detailed analysis of the sensitivity of the retrieval technique to the aerosol vertical distribution.

[40] The optical depth validation indicates that the TOMS retrieval result is within 30% of the Sun photometer measurements for 82% of the 127 comparison cases, when the aerosol profile is represented as a single layer at 3 km above the ground. This conclusion is consistent with previous algorithm performance assessments [Torres et al., 2002a, 2002b]. Sources of uncertainty are the aerosol vertical distribution, subpixel cloud contamination effects, and, to a lesser extent, the surface reflectivity.

[41] A total of 120 TOMS-AERONET coincidences at 8 sites were used to evaluate the accuracy of the TOMS derived single scattering albedo. The evaluation of the single scattering albedo retrieval with reference to the AERONET results, shows than in 63% of the cases the satellite retrieval agrees within 0.03 with the AERONET results, whereas the agreement is within 0.05 in 87% of the coincidences used in the analysis. This results indicate that the near-UV measurements can be used to measure the aerosol single scattering albedo from space with an accuracy equivalent to the one reported by the AERONET project of ±0.03 [Dubovik et al., 2002] for aerosol optical depth values of about 0.4. It should be said that the accuracy of the AERONET single scattering albedo product improves as the aerosol layer becomes optically thicker. Thus, at the large optical depth values (0.6 and higher) encountered during SAFARI 2000, the accuracy of the AERONET absorption measurements should be better than 0.03. However, no estimates of the AERONET accuracy for large aerosol loads are available in the literature.

[42] On four days during the campaign, MPLNET measurements of the aerosol vertical distribution were available within 30 min of the satellite overpass. The lidar profiles were used as input to generate look up tables for the retrieval in lieu of the standard vertical distribution assumption. When using the actual aerosol profiles the accuracy of the retrieved aerosol optical depth improved significantly in the cases when the measured and assumed profiles were very different. As it would be expected, no significant effect was observed when the assumed profile resembled the measured one. The evaluation of the accuracy of the single scattering albedo showed that even in the one case that the measured vertical distribution deviated significantly from the simple one-layer assumption the retrieved ssa using the standard 3 km profile was still very close (within 0.01) to the ground-based measurement. The use of the actual profiles resulted in an additional improvement of the accuracy of the space-based measurements.

[43] The reported results support the use of the 3 km aerosol layer height for the retrieval of smoke properties in the absence of accurate vertical distribution information over the continental areas. As shown by Anderson et al. [1996], using lidar measurements during TRACE-A, the aerosol vertical distributions over Africa and South America are not too different. Elevated aerosol layers, however, may result downwind of the source areas. As shown in this work, the use of actual vertical profile information enhances the accuracy of the near-UV method.

[44] To date the near-UV technique is the only tested and validated method of measuring aerosol absorption from space. The multidecade long TOMS record has been used to produce the only available satellite global data set on aerosol absorption, from 1979 to present. The Ozone Monitoring Instrument (OMI) on the EOS-Aura satellite, launched in July 2004, will continue the record of aerosol absorption into the future. The Aura mission is one of several satellites flying in formation in the so-called A-train, that includes the AQUA and Cloud Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) platforms. The OMI aerosol algorithm [Torres et al., 2003] takes advantage of the OMI extended spectral coverage (270 to 500 nm), the smaller-than-TOMS footprint (3 × 13 km for aerosol retrieval), and the simultaneous availability of aerosol related information from other A-train sensors, to enhance the science value of the aerosol information. For instance, the nadir aerosol vertical distribution measured by CALIPSO will undoubtedly benefit the quality of the OMI aerosol products.

[45] In order to reduce the uncertainty associated with the absorption effects of aerosols in both climate and air quality applications, it is necessary to continue and improve current aerosol absorption sensing capabilities. Because of the weak wavelength dependence of the imaginary refractive index of soot (the aerosol component responsible for most of the absorption effect of anthropogenic aerosols), near-UV absorption retrievals of the imaginary refractive index can be extrapolated to the visible and near-IR.

[46] The TOMS measurements have played a fundamental role in the development and testing of the near-UV method, and OMI is certainly an improvement over TOMS capabilities. The Advanced Earth Orbiting Satellite II (ADEOS II) Global Imager (GLI) sensor [Nakajima et al., 1998] was the first satellite instrument to combine near-UV, visible and near-IR channels at a spatial resolution suitable for accurate aerosol retrieval. Unfortunately, the satellite failed just a few months after launch. To take full advantage of the potential of the near-UV method, future satellite sensing missions should include measurements in the near-UV to derive aerosol absorption information, that combined with measurements in the visible and near-infrared will significantly contribute to improve the understanding of the radiative transfer effect of aerosols on the Earth-atmosphere system.


[47] We thank David Larko from SSAI for his assistance with the color graphics. E. J. Welton and MPLNET are funded by the NASA Earth Observing System and NASA Radiation Sciences Program.