Journal of Geophysical Research: Atmospheres

Extinction-to-backscatter ratios of Saharan dust layers derived from in situ measurements and CALIPSO overflights during NAMMA


  • 17 March 2009


[1] We determine the extinction-to-backscatter (Sa) ratios of dust using (1) airborne in situ measurements of microphysical properties, (2) modeling studies, and (3) the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) observations recorded during the NASA African Monsoon Multidisciplinary Analyses (NAMMA) field experiment conducted from Sal, Cape Verde during August to September 2006. Using CALIPSO measurements of the attenuated backscatter of lofted Saharan dust layers, we apply the transmittance technique to estimate dust Sa ratios at 532 nm and a two-color method to determine the corresponding 1064 nm Sa. This method yielded dust Sa ratios of 39.8 ± 1.4 and 51.8 ± 3.6 sr at 532 and 1064 nm, respectively. Second, Sa at both wavelengths is independently calculated using size distributions measured aboard the NASA DC-8 and estimates of Saharan dust complex refractive indices applied in a T-Matrix scheme. We found Sa ratios of 39.1 ± 3.5 and 50.0 ± 4 sr at 532 and 1064 nm, respectively, using the T-Matrix calculations applied to measured size spectra. Finally, in situ measurements of the total scattering (550 nm) and absorption coefficients (532 nm) are used to generate an extinction profile that is used to constrain the CALIPSO 532 nm extinction profile and thus generate a stratified 532 nm Sa. This method yielded an Sa ratio at 532 nm of 35.7 sr in the dust layer and 25 sr in the marine boundary layer consistent with a predominantly sea-salt aerosol near the ocean surface. Combinatorial simulations using noisy size spectra and refractive indices were used to estimate the mean and uncertainty (one standard deviation) of these Sa ratios. These simulations produced a mean (± uncertainty) of 39.4 (±5.9) and 56.5 (±16.5) sr at 532 and 1064 nm, respectively, corresponding to percentage uncertainties of 15% and 29%. These results will provide a measurements-based estimate of the dust Sa for use in backscatter lidar inversion algorithms such as CALIOP (Cloud-Aerosol Lidar With Orthogonal Polarization).

1. Introduction

[2] Lidar is a powerful tool for studying the vertical distribution of aerosols and clouds in the atmosphere. Of particular importance is the distribution and transport of Saharan dust systems. The deployment of CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations), a joint NASA-CNES satellite mission, has enabled vertically resolved measurements of Sahara air layer(s) (SAL), which will provide significant insights into properties of Sahara dust aerosols. CALIPSO is designed to provide measurements to advance our understanding of the role of aerosols and clouds in the climate system [Winker et al., 2009]. The Cloud-Aerosol LIdar With Orthogonal Polarization (CALIOP) [Winker et al., 2007] is the primary instrument on the CALIPSO satellite. CALIOP is designed to acquire vertical profiles of elastic backscatter at two wavelengths (1064 and 532 nm) from a near-nadir-viewing geometry during both day and night phases of the orbit. In addition to the total backscatter at the two wavelengths, CALIOP also provides profiles of linear depolarization at 532 nm. Accurate aerosol and cloud heights and retrievals of extinction coefficient profiles are derived from the total backscatter measurements [Vaughan et al., 2009]. The depolarization measurements enable the discrimination between ice clouds and water clouds [Hu et al., 2009] and the identification of nonspherical aerosol particles [Liu et al., 2009]. Additional information, such as estimates of particle size for the purpose of discriminating between clouds and aerosols, are obtained from the ratios of the signals obtained at the two wavelengths. On 28 April 2006, the CALIPSO satellite was launched into a low Earth Sun-synchronous orbit at a 705 km altitude and an inclination of 98.2°. A few months later, in August 2006, the NASA African Monsoon Multidisciplinary Analyses (NAMMA) campaign commenced at the Cape Verde Islands, 350 miles off the coast of Senegal in West Africa. NAMMA was designed to study the evolution of precipitating convective systems largely as this evolution pertained to the SAL and its role in the tropical cyclogenesis. Several aircraft flights were dedicated to nearly coincident measurements with NASA's orbiting satellites (including Aqua, TRMM, and CloudSat/CALIPSO). For this study, we use data collected aboard NASA's DC-8 medium altitude research aircraft outfitted with, among other instruments, a full suite of sensors and probes designed to measure aerosol microphysical and optical properties. Relevant parameters include high spatial-resolution scattering and absorption coefficients at multiple wavelengths in the visible spectrum and dry particle size distributions over the 0.08–10 μm diameter range (G. Chen et al., Observations of Saharan dust microphysical and optical properties from the eastern Atlantic during NAMMA airborne field campaign, submitted to Atmospheric Chemistry and Physics, 2010).

[3] Depending on the mineralogical composition, the SAL can have a significant impact on both the radiation balance and cloud processes. Dust particles scatter in the shortwave regime (cooling the planet) and absorb both shortwave and longwave radiation (heating the planet). By some estimates, the anthropogenic forcing due to dust, i.e., dust generated by human activity such as land clearing, is comparable to the forcing by all other anthropogenic aerosols combined [Sokolik and Toon, 1996]. Saharan dust influences cyclone activity and convection in the region off the west coast of Africa and air quality as far west as the U.S. east coast and Gulf of Mexico. There have been reports of causal links between cyclone activity and dust loading suggesting that perhaps the Sahara dust layer acts to inhibit cyclone development [Dunion and Velden, 2004] and more generally convection [Wong and Dessler, 2005]. Sahara dust is unique in its ability to maintain layer integrity as it is transported over long distances (∼7500 km) to the Americas [Liu et al., 2008; Maring et al., 2003; Savoie and Prospero, 1976]. The presence of Sahara dust layers have been found to perturb ice nuclei (IN) concentrations as far away as in Florida. During CRYSTAL-FACE (Cirrus Regional Study of Tropical Anvils and Cirrus Layers-Florida Area Cirrus Experiment), DeMott et al. [2003] found that IN concentrations were significantly enhanced in heterogeneous ice nucleation regimes warmer than −38°C, when Saharan dust layers are present. It is therefore important to study the distribution and optical properties of Sahara dust.

[4] In order to estimate the optical depth of the Sahara dust layers from elastic backscatter lidar measurements, the Sa ratio must be known or prescribed. Given aerosol free regions above and below a lofted dust aerosol layer, Sa can be calculated from the attenuated backscatter profile of a space-based lidar return [Young, 1995]. Sa for dust aerosols is dependent on the mineral composition, size distribution, and shape parameters (e.g., aspect ratio and complexity factor). All of these are highly variable and for the most part not well known. For these reasons, Sa obtained from scattering models have larger uncertainties than models of the nearly spherical urban pollution or marine aerosols.

[5] There have been several studies and measurements of dust Sa at 532 nm [Ackermann, 1998; Anderson et al., 2000; Berthier et al., 2006; Di Iorio et al., 2003, 2009; Muller et al., 2007; Müller et al., 2000; Tesche et al., 2009] and relatively few such measurements or studies of dust Sa at 1064 nm [Ackermann, 1998; Liu et al., 2008; Tesche et al., 2009]. Prior to NAMMA, there were a number of vertically resolved measurements of Saharan dust microphysical and optical properties including African Monsoon Multidisciplinary Analysis (AMMA) [Chazette et al., 2007; Haywood et al., 2008]. NAMMA studies along with CALIPSO measurements provide a unique opportunity to compare extinction measurements derived from in situ profile measurements of total scattering and absorption aboard the NASA DC-8 and CALIPSO extinction profiles estimated from two wavelength retrieval methods. These profiles by extension provide the constraints from which the lidar ratios can be determined as explained in the following sections. Section 2 discusses the CALIPSO lidar data and its analysis. The NAMMA data and analyses are discussed in section 3, and coincident CALIPSO-NAMMA measurements are presented in section 4. In section 5, the size distributions measured during NAMMA aboard the DC-8 are implemented in a T-Matrix scheme to estimate profiles of Sa ratios. Section 6 discusses the uncertainty in Sa using a combinatorial method.

2. CALIPSO Lidar Data and Extinction-to-Backscatter Ratio Retrieval Methods

[6] The CALIPSO lidar data used for these studies are the version 2.01 lidar level 1 attenuated backscatter returns at the 532 nm perpendicular and parallel channels and 1064 nm total attenuated backscatter. The volume depolarization ratio is determined from the perpendicular and parallel channels and used to identify dust aerosols [Liu et al., 2009; Omar et al., 2009]. For NAMMA underflights of CALIPSO and near spatial coincidences where both missions observed dust layers of optical depths greater than about 0.3, we compare the extinction profiles from in situ measurements to CALIPSO profiles. In such cases, we calculate the extinction using Sa that was determined using the transmittance method or an Sa ratio constrained by the in situ extinction profiles. In both cases, we use the two-color methods to retrieve the 1064 nm Sa, after determining the 532 nm Sa. These two methods, transmittance and two-color, are discussed below.

2.1. Transmittance Methods

[7] The transmittance method uses the following equation describing the relationship between optical depth and integrated attenuated backscatter, as in the work of Platt [1973],

equation image

[8] Here γ′ is the integrated (from layer base to top) attenuated backscatter,

equation image

[9] τ is optical depth, η is a multiple scattering parameter, T2 = exp(−2ητ) is the layer-effective two-way transmittance, and Sa = σa/βa, where βa is the aerosol backscatter coefficient and σa is the aerosol extinction coefficient. This ratio is assumed constant throughout a feature. Note that the quantities Sa, γ′, and τ describe characteristics of an aerosol layer, i.e., they are associated with the backscatter and extinction of aerosol particles only. If we define an effective Sa ratio, S* = ηSa, we can rewrite equation (1) as follows,

equation image

[10] The effective two-way transmittance is typically obtained by fitting the returns both above and below a feature to a reference clear air scattering profile obtained from local rawinsonde measurements or meteorological model data [Young, 1995]. In this study, the transmittance method is used to determine S* from the 532 nm CALIPSO measurements whenever clear air scattering signals are available both above and below an aerosol layer. However, the same method is not applicable to the 1064 nm CALIPSO measurements, because a reliable measurement of the clear air scattering at 1064 nm, which is about 16 times smaller than that at 532 nm, is not available. To determine S* at 1064 nm, the two-color method described in section 2.2 is used.

2.2. The Two-Color Method

[11] The two-color or two-wavelength method was first proposed by Sasano and Browell [1989] and adapted to spaceborne lidar measurements using an optimization technique by Vaughan et al. [2004]. The method requires a priori knowledge of Sa at 532 nm and a suitable profile of 532 nm attenuated backscatter amenable to the calculation of 532 nm aerosol backscatter coefficient profiles. For the NAMMA cases described below, these preconditions were satisfied. Whenever a suitable region of clear air was identified both above and below an aerosol layer, the Sa at 532 nm was determined using the transmittance method described above. A clear air layer is defined as a region of low attenuated scattering ratios with a mean value equal to or less than 1 and a slope with respect to altitude of approximately zero. This is further confirmed by low volume depolarization ratios in a small region (∼1/2 km) below the aerosol layer. In cases where coincident NAMMA measurements are available, the Sa ratio at 532 nm is the value that provides the best fit between the retrieved CALIPSO extinction profiles and the NAMMA in situ extinction profiles obtained by summing the total scattering and absorption measured by a nephelometer and a Particle Soot/Absorption Photometer (PSAP), respectively, aboard the DC-8.

[12] Once Sa ratio is determined at 532 nm, the value at 1064 nm can be calculated using the two-color method. Note that this technique can be used to derive Sa at 532 nm if the value at 1064 nm is known. Given a solution of the particulate backscatter at 532 nm (β532,p) the two-color method uses a least squares method to minimize the difference between the measured attenuated total backscatter at 1064 nm (B1064) and the attenuated backscatter at 1064 nm (right-hand side of equation (4)) reconstructed from the extinction and backscatter coefficients at 532 nm,

equation image

[13] The only unknowns (underlined) in equation (4) are the Sa ratio at 1064 nm (S1064) and the backscatter color ratio χ (defined as β1064,p/β532,p). These are both intensive aerosol properties defined by the layer composition, size distribution, and shape of its constituent particles. Assuming these characteristics do not vary substantially in a given aerosol layer, we can infer that S1064 and χ are constant within the layer. The algorithm details and optimization techniques are discussed at length by Vaughan et al. [2004].

3. Numerical Calculation Based on NAMMA In situ Measurements

3.1. Aerosol Microphysical Properties

[14] We use measurements of the aerosol size distributions based on number concentrations from the Aerodynamic Particle Sizer (APS, TSI, Inc., Shoreview, MN) and the Ultra-High Sensitivity Aerosol Spectrometer (UHSAS, Droplet Measuring Systems, Boulder, CO). The UHSAS measures the fine mode aerosol size distributions with particle diameters from 0.06 to 0.98 μm, and the APS measures the coarse mode size distributions from 0.6 to 5.5 μm in diameter. The two instruments drew samples from a window-mounted, shrouded inlet, which was maintained isokinetic by continually adjusting downstream flow in response to changes in aircraft speed and ambient air density. The inlet was characterized during a series of tower flybys [McNaughton et al., 2007] and found to efficiently transmit particles smaller than 4 μm in diameter under flight conditions similar to those encountered during NAMMA. Our calculations suggest that the short sample lines (<1 m) leading from the inlet to the instruments had a negligible effect on measured size distributions. Possible errors associated with the UHSAS and APS measurements are discussed by Chen et al. (submitted manuscript, 2010). We use the size distributions to identify the presence of aerosol dust layers. In many cases, these intense dust layers were visually identified by the instrument operators and in some cases specifically targeted by the DC-8 operators for sampling. Additional information about the composition of these layers is available from the NAMMA data archives ( For each size distribution sampled during a 5 s interval, we fit the discrete measurements to the best continuous bimodal lognormal size distribution, as shown in Figure 1. The geometric mean radius and standard deviation of a fine and coarse mode derived from the in situ measurements are used in the numerical calculations.

Figure 1.

(a) A Two-dimensional probability plot of NAMMA in situ size distributions measured in a dust layer over a 22 min period (∼260 size distributions) during Flight 4 on 19 August 2006. The mean measured size distribution is denoted by the yellow line. (b) The best approximate bimodal lognormal particle size distribution (blue squares) with fine (coarse) mean radius of 0.0648 (0.5627) μm and a fine (coarse) geometric standard deviation of 1.696 (1.572) fitted to the average measured size distribution (red squares).

[15] On 25 August 2008, the DC-8 flew through a dense elevated dust layer measuring nearly 1 km in thickness at a mean altitude of about 2.3 km in an hour-long mostly straight and level flight. The APS and UHSAS measured coarse and fine size distributions, respectively, through the dust layer at intervals of 5 s. Figure 2 shows the probability distributions of the geometric mean fine and coarse radii of the dust layer. Figure 2 is generated by taking the discrete 5 s size distribution measurements and fitting these to a bimodal lognormal distribution as described above. For this dust layer, the microphysical properties of mean, median, and standard deviations of the fine (coarse) radius distributions, as shown in Figures 2a and 2b, are 0.059, 0.061, and 0.0064 μm (0.54, 0.57, and 0.083 μm), respectively. The mean, median, and standard deviations shown in Figures 2c and 2d for the fine (coarse) geometric standard deviation distributions are 1.613, 1.630, and 0.101 (1.495, 1.545, and 0.151), respectively.

Figure 2.

The distributions of the (a) fine and (b) coarse geometric mean radii and the (c) fine and (d) coarse geometric standard deviations for a dense elevated dust layer measured during a flight on 25 August 2006.

[16] The distributions in Figure 2 show that Saharan dust properties after lofting of these layers remain relatively unchanged within the plume. Other studies [Liu et al., 2008; Maring et al., 2003; Prospero and Carlson, 1971, 1972] have shown the same consistency in properties after long range transport of dust. In each case, the means and medians of the size distribution descriptors are close, i.e., the size descriptors are nearly normally distributed. The standard deviation is a small fraction (<0.16) of the means, i.e., the variance of the data is small and thus the layer is quite homogenous with respect to size across the 1 km vertical extent of the dust plume.

3.2. Scattering Models

[17] Mie scattering calculations [Mie, 1908], when applied to dust, are adequate for total scattering, albedo, and other flux-related quantities, but result in large errors when used to retrieve optical depth from satellite reflectance measurements. In particular and central to the theme of this paper, are Sa ratios calculated from measured size distributions. Mie calculations underestimate Sa by up to a factor of 2.0 leading to substantial errors in the lidar-derived aerosol optical depths [Kalashnikova and Sokolik, 2002]. This has been known experimentally for quite some time: laboratory measurements by Perry et al. [1978] showed nonspherical particles, when compared to spherical particles having the same equivalent volume, enhance side scattering and suppress backscattering. To account for nonsphericity of dust particles, we use T-matrix calculations with the assumption that the dust shapes can be modeled by randomly oriented prolate spheroids. T-Matrix is a matrix formulation of electromagnetic scattering first proposed by Waterman [1971] and subsequently improved and extended to much larger sizes and aspect ratios by Mishchenko et al. in a series of papers [Mishchenko, 1991, 1993; Mishchenko and Travis, 1994; Mishchenko et al., 1996a, 1996b; Mishchenko and Travis, 1998]. The T-Matrix code used in these calculations is described in detail by Mishchenko and Travis [1998]. This method is particularly suitable for light scattering calculations of nonspherical, polydisperse, randomly oriented particles of identical axially symmetric shape with size parameter, x (x = πdp/λ, dp is particle diameter and λ is the wavelength), smaller than 30.

[18] It is challenging to determine representative statistics of the mean shape for dust particles because of their complexity and variety in shape. These particles are not only confined to desert regions but are ubiquitous in continental areas where they contribute quite significantly to the extinction budget [Omar et al., 1999]. Fortunately, however, randomly oriented prolate and oblate spheroids can adequately represent the scattering properties of nonspherical particles of the same aspect ratio [cf. Mishchenko et al., 1995; Mishchenko and Travis, 1994]. The aspect ratio is the ratio of the largest to the smallest particle dimension. A prolate (oblate) spheroid is a rotationally symmetric ellipsoid with a polar diameter greater (smaller) than the equatorial diameter.

[19] There have been several studies of aspect ratio distributions representative of dust aerosols. From an analysis of scanning electron microscope images of yellow sand particles, Nakajima et al. [1989] found that the distribution of the minor to major particle radius ratio peaked around 0.6, equivalent to an aspect ratio of 1.67. An investigation of mineral dust particle shapes using electron microscopy by Okada et al. [1987] found a mean aspect ratio of 1.4 ranging from 1.0 to 2.3. Hill et al. [1984] compared the measured scattering properties of 312 samples of soil dust with the simulated scattering properties of randomly oriented prolate spheroids using T-matrix. They found the distributions of the aspect ratio of prolate spheroids centered near 2.3 most closely reproduced the measured scattering properties. For this study, we use a mean aspect ratio of 2.0 based on the above studies and investigate the sensitivity of the Sa ratio to aspect ratios ranging from 1.7 to 2.3 partly to account for the dependence of the aspect ratio on size as discussed by Kalashnikova and Sokolik [2004].

[20] There is quite a wide range of estimated and measured mineral dust complex refractive indices (mik). For wavelengths of 550 and 1000 nm, d'Almeida et al. [1991] estimate the real part (m) of 1.53 and 1.52, respectively, and a spectrally invariant imaginary part (k) of 0.008 for dust-like aerosols, and 1.53 – i0.0055 and 1.53 – i0.001, respectively, for mineral dust. Ackerman [1998] used values of 1.53 – i0.0043 and 1.53 – i0.0063 to calculate dust Sa ratios of 19–23 and 17–18 sr at 532 and 1064 nm, respectively. These values are much lower than more recent 532 nm dust Sa ratios of 40–60 sr [Cattrall et al., 2005; Di Iorio et al., 2003; Muller et al., 2007; Murayama et al., 2003; Voss et al., 2001] because the Ackerman study assumed spherical particles.

[21] Retrievals from radiances measured by ground-based Sun-sky scanning radiometers of the Aerosol Robotic Network (AERONET) over a 2 year period yielded dust complex refractive index values of 1.55 ± 0.03 – i0.0014 ± 0.001 at 670 nm and 1.55 ± 0.03 – i0.001 ± 0.001 at 1020 nm at Bahrain (Persian Gulf) and 1.56 ± 0.03 – i0.0013 ± 0.001 at 670 nm and 1.56 ± 0.03 – i0.001 ± 0.001 at 1020 nm at Solar Village in Saudi Arabia [Dubovik et al., 2002]. Using vertically resolved aerosol size distributions in a scattering model constrained by lidar measurements of aerosol backscattering coefficient at 532 nm, Di Iorio et al. [2003] estimated dust refractive indices of 1.52–1.58 (real part) and 0.005–0.007 (imaginary part). Kalashnikova and Sokolik [2004] calculated effective refractive indices from component mixtures and found values of 1.61 – i0.0213 and 1.59 – i0.0032 for Saharan dust at wavelengths of 550 and 860 nm, respectively. The values for Asian dust, from the same monograph, are 1.51 – i0.0021 and 1.51 – i0.0007 at 550 and 860 nm, respectively. Using a twin angle optical counter, Eidhammer et al. [2008] estimated the indices of refraction to be in the range 1.60–1.67 for the real part and 0.009–0.0104 for the imaginary part.

[22] The Saharan Mineral Dust Experiment (SAMUM) [Heintzenberg, 2009; Rodhe, 2009] based in Morocco in 2006 produced several independent estimates of the complex refractive indices of Saharan dust. Kandler et al. [2009] determined Saharan dust aerosol complex refractive index from chemical/mineralogical composition of 1.55 – i0.0028 and 1.57 – i0.0037 at 530 nm for small (diameter < 500 nm) and large particles (diameter > 500 nm), respectively. Schladitz et al. [2009] derived mean refractive indices of 1.53 – i0.0041 at 537 nm and 1.53 – i0.0031 at 637 nm from measurements of scattering and absorption coefficients, and particle size distributions. Using similar methods during SAMUM, Petzold et al. [2009] found real parts of the refractive indices of Saharan dust ranging from 1.55 to 1.56 and imaginary parts ranging from 0.0003 to 0.0052. Some of the estimates of refractive indices reported in the literature are summarized in Table 1.

Table 1. Summary of Complex Dust Refractive Indices From Previous Studies
Wavelength (nm)Real PartImaginary PartSource
5001.500.0045Volz [1973]
10001.530.008d'Almeida et al. [1991], dust-like
5501.530.0055d'Almeida et al. [1991], mineral dust
10001.530.001d'Almeida et al. [1991], mineral dust
5321.530.0043Ackermann [1998], dust
10641.530.0063Ackermann [1998], dust
6701.55 ± 0.030.0014 ± 0.001Dubovik et al. [2002], Bahrain dust
10201.55 ± 0.030.003 ± 0.001Dubovik et al. [2002], Bahrain dust
6701.56 ± 0.030.0013 ± 0.001Dubovik et al. [2002], Solar Village dust
10201.56 ± 0.030.001 ± 0.001Dubovik et al. [2002], Solar Village dust
5321.52–1.580.005–0.007Di Iorio et al. [2003], Saharan dust
5001.420.003Israelevich et al. [2003], Sede Boker dust
8601.510.0032Kalashnikova and Sokolik [2004], Saharan dust
5501.510.002Kalashnikova and Sokolik [2004], Asian dust
8601.510.0007Kalashnikova and Sokolik [2004], Asian dust
6701.450.0036Omar et al. [2005], dust cluster
5501.530.0015Cattrall et al. [2005], mineral dust
10201.530.0005Cattrall et al. [2005], mineral dust
300–7001.60–1.670.009–0.0104Eidhammer et al. [2008], Wyoming dust
3551.52 ± 0.0040.008 ± 0.001Raut and Chazette [2008], Saharan dust
5371.530.0041Schladitz et al. [2009], Saharan dust
6371.530.0031Schladitz et al. [2009], Saharan dust
4501.55–1.560.0003–0.0052Petzold et al. [2009], Saharan dust
7001.55–1.560.0003–0.0025Petzold et al. [2009], Saharan dust
5301.550.0028–0.0037Kandler et al. [2009], Saharan dust

[23] To perform the sensitivity study described in section 6, we use values of the real part of the refractive index ranging from 1.45 to 1.55 (normally distributed) and the imaginary part ranging from 0.00067 to 0.006 (log normally distributed) with a central value of 1.50 – i0.002. For the scattering calculations using NAMMA size distributions, we use the central values of the refractive indices along with the nearly instantaneous (5 s interval) size distribution measurements to generate profiles of the aerosol properties. Figure 3 is a plot of the fine mode, coarse mode, and total phase function of the dust plume encountered on 19 August 2006. The fine mode and coarse mode phase functions are computed from the mean of the instantaneous size distributions, and the total phase function is the area-weighted composite of the fine and coarse mode phase functions. The phase functions are driven largely by the coarse mode, especially at 532 nm, and exhibit a more pronounced peak in the backscattering direction at 532 nm than 1064 nm.

Figure 3.

The (a) 532 nm and (b) 1064 nm Saharan dust phase functions calculated from the size distributions measured aboard NASA DC-8 during Flight 4 on 19 August 2006.

4. Data Analyses of CALIPSO NAMMA Coincident Measurements

[24] For this study, we analyzed coincident measurements of the CALIOP 532 nm extinction profiles and in situ extinction profiles measured at wavelengths near the CALIOP green channel. The in situ extinction coefficient is obtained by summing the scattering (550 nm) and absorption (532 nm) coefficients measured by a nephelometer and a PSAP, respectively. Data from the nephelometer and PSAP have been corrected for errors associated with the limited detector viewing angle [Anderson and Ogren, 1998] and scattering from the filter media [Virkkula et al., 2005], respectively. We make the assumption that the scattering properties are invariant over the 532–550 nm range for these large dust particle sizes. We chose 3 days on which there were near collocated CALIPSO and NAMMA measurements of nearly the same air mass.

4.1. 19 August 2006 NAMMA Flight 4

[25] 19 August was one of the days the DC-8 performed an under flight of CALIPSO. Figure 4 shows the time-altitude flight track of the DC-8. A nearly coincident in situ profile was obtained during the second ascent leg shown in Figure 4. In this segment, the DC-8 climbed from 300 m at 50,460 s to 10 km at 52,575 s after UTC. The climb flight time was 2115 s (∼35 min) at a rate slightly lower than the DC-8's nominal rate of climb of 4.8 m s−1. Atmospheric context for this flight is given by Figures 5a and 5b, in which both the DC-8 and CALIPSO flight tracks are superimposed on images of measurements made by the Moderate Resolution Imaging Spectroradiometer (MODIS) and Measurements of Pollution in the Troposphere (MOPITT), respectively. The DC-8 flight tracks are the black irregular octagons, and the CALIPSO orbit tracks are the straight lines in both images. The underflights in Figures 5a and 5b are segments where the DC-8 flies under the CALIPSO satellite and the two are momentarily both temporally and spatially collocated. Unfortunately, the direct underflight of CALIPSO by the DC-8 was a nearly level flight with no in situ profile information. In fact the DC-8 was at a high altitude near Flight Level 330 (∼10 km) throughout the underflight and therefore did not encounter any significant aerosol layer at this altitude to sample. We used the ascending leg of the DC-8 flight, which corresponds to the flight segment during ascent to the CALIPSO underflight portion denoted by blue dots in Figure 5a. The in situ profile segment is shown in Figure 4.

Figure 4.

The DC-8 altitudinal flight path showing the segment during which the in situ measurements were made. The direct CALIPSO underflights are the five segments (a, b, c, d, and e) at the highest altitudes. The time is in seconds after midnight UTC.

Figure 5.

CALIPSO flight tracks superimposed on (a) the MODIS optical depth and (b) the MOPPITT CO concentrations. The DC-8 flight track for 19 August 2006 is shown by the black lines in the middle of both figures. The blue dots in Figure 5a denote the portion of the DC-8 flight segment during which the in situ measurements were taken. Daytime CALIPSO tracks run from the southeast to the northwest while the nighttime tracks run from northeast to the southwest.

[26] The MODIS optical depth near the coincident flight segment is about 0.5. The MOPPITT image shows moderate CO concentrations in the vicinity of the coincident flight track and therefore indicates that most of the aerosol is dust and not continental pollution or biomass burning. This can also be confirmed by the CALIPSO depolarization measurements. The image also shows high CO concentrations (>2.5 × 1018 molecules/cm2) to the south of the DC-8 flight tracks most likely due to biomass burning and identified as such by the CALIPSO aerosol subtyping scheme illustrated later in this section.

[27] The CALIPSO browse images (e.g., Figure 6) are plots of the attenuated backscatter color coded by intensity varying from blue (weak) to white (very strong). A horizontal line near the 0 km mark denotes the surface. The CALIPSO data used for comparison with the in situ data come from the 80 profiles (∼27 km horizontal average) between the white lines in the browse image (Figure 6). To determine the optimal 532 nm Sa ratio for these data, we iteratively adjust Sa until the difference, in a least squares sense, between the retrieved CALIPSO profile, σa,532 nm, and the measured NAMMA profile, σa,550 nm, is minimized. This is analogous to the CALIOP validation efforts of Chazette et al. [2010] using ground based and airborne measurements.

Figure 6.

Browse image of the CALIPSO flight track on 19 August 2006 shown in blue in the inset map and the location of the near coincidence of the CALIPSO flight track and the DC-8 flight 4 during NAMMA is denoted by the two white lines near 15°N, 19°W. In the image, colors above the surface are coded as follows: blue indicates clear sky, white indicates clouds, and warm colors (green, yellow, orange, and red) indicate aerosols. The white parallel lines bound the CALIPSO profiles closest to the in situ measurements.

[28] Figure 7 is an image of the aerosol scattering ratios measured by the Lidar Atmospheric Sensing Experiment (LASE) on board the DC-8 during the ascent leg of Flight 4 on 19 August 2006 as indicated in Figure 4. The measurements were made at 815 nm, and the extinction calculation was performed using an Sa ratio of 36 sr [Ismail et al., 2010]. The DC-8 flight track is shown by the solid line in Figure 7 during which the in situ measurements shown in Figure 8 were made. Figure 7 also shows the region nearest to the CALIPSO overpass where measurements of the CALIPSO profiles shown in Figure 8 were made. LASE observed a dust layer extending to an altitude of 6 km at the coincident point. This dust layer is optically and geometrically thick and is lofted over a layer of lower optical depth (aerosol scattering ratio ∼1) in the marine boundary layer. Some differences in altitude and extent of the layer between the CALIPSO and LASE measurements can be attributed to temporal and spatial mismatch.

Figure 7.

Aerosol scattering ratio (βa/βm at 815 nm) image measured by the Lidar Atmospheric Sensing Experiment (LASE) aboard DC-8 during the ascending segment passing through a Sahara Air Layer (SAL) and the underflight of CALIPSO as shown in Figure 4. The DC-8 point of nearest approach to the CALIPSO flight track is near 15°N–19°W.

Figure 8.

A comparison of the profiles of the in situ extinction coefficient derived during the DC-8 ascending segment as shown in Figure 4 and the extinction coefficient retrieved from the CALIPSO measurement averaged over the region indicated by the two white lines in Figure 6. The Sa ratios were determined using the in situ extinction as an additional constraint.

[29] Figure 8 is a plot of the extinction profiles retrieved from CALIPSO's 532 nm backscatter profiles and the profiles of the sum of scattering (550 nm) and absorption (532 nm) measured by the nephelometer and the PSAP aboard the NASA DC-8, respectively. These profiles were taken during the ascent leg shown in Figure 4 corresponding to the flight tracks shown in Figure 5a. As in the LASE observation in Figure 7, the aerosol is stratified into two layers. The boundary between the two layers is determined by the increase in the extinction coefficient near 2 km. The two Sa ratios are values that provide the best fit in a least squares sense of the CALIPSO data to the NAMMA in situ measurements. The 532 nm Sa ratios are consistent with a dust plume (Sa = 35.7 sr) above sea salt in the marine boundary layer (Sa = 25 sr). The root mean square (rms) of the differences between the CALIPSO and the NAMMA in situ extinction coefficient profiles are 20.2 and 47.4 Mm−1 for the marine and dust layers, respectively. The rms of the differences in the extinction coefficients in the clear region (4–8.5 km) above the dust layer in Figure 6 is 12.3 Mm−1. Some differences between the two extinction profiles in Figure 8 are likely due to the temporal and spatial mismatches of 30 min and 160 km, respectively, between DC-8 and CALIPSO.

[30] Figure 9 is a plot of the results of (Figure 9a) the cloud-aerosol discrimination and (Figure 9b) aerosol classification algorithms applied to the data shown in the browse image including the NAMMA underflight (Flight 4 of 19 August 2006). The CALIPSO level 2 algorithms first discriminate between aerosol and clouds [Liu et al., 2009] and then classify the aerosol layers into aerosol subtypes [Omar et al., 2009]. Figure 9a shows that some of the optically thick aerosol near 15°N was misclassified as clouds and thus was not examined by the aerosol subtyping algorithm. The presence of biomass burning smoke and polluted dust (mixture of dust and smoke) during the first part of the flight depicted in Figure 9b is borne out by the high CO concentrations in the MOPPITT data to the south of the DC-8 flight tracks shown in Figure 5b. Although the small lump of aerosol at the surface near 15°N is classified as pure dust, it is more likely a mixture of dust and marine aerosol. The white line in the Figure is the midpoint of the 80 profiles averaged for the retrieval of the extinction profile discussed above.

Figure 9.

Products of the CALIPSO (a) cloud-aerosol discrimination and (b) aerosol classification schemes for the data in the browse image shown in Figure 6. The cloud aerosol discrimination shows the extensive aerosol layer at 3 km and the aerosol classification image shows that this aerosol is a Saharan dust layer to the north and a polluted dust and smoke layer to the south.

[31] CALIPSO also measured a dense Saharan dust layer to the southwest of the coincident measurements during a nighttime orbit on the same day (19 August 2006). The browse images of attenuated backscatter at 532 nm for this measurement are shown in Figure 10. The inset map in Figure 10 shows the CALIPSO ground track in blue. This dust layer appears to be a more robust part of the same dust plume observed during the coincident measurement.

Figure 10.

(a) Browse image of the orbit segment denoted by the blue track in the inset map and (b) an expansion of the dust layer region of the plot marked by yellow dots in Figure 10a. In both cases, the color code is as in Figure 6. The red dashed oval shows the elevated (layer top ∼ 5km) dust layer. Aerosol optical depths at 532 nm for all five segments are shown and are near 0.3. The range corrected mean total attenuated backscatter (red lines) and molecular backscatter (blue lines) profiles are shown in Figure 10c for the five segments of the dust layer analyzed in Figure 10b.

[32] As shown (by the red dotted oval) in Figure 10a, the layer exceeds 1000 km in horizontal extent (from 18.3°N to 10.3°N). The 532 nm aerosol optical depth (AOD) is greater than about 0.3 across most of the 1000 km orbital segment shown in Figure 10a. Figure 10b is a magnified illustration of the region in Figure 9a subtended by the yellow dotted line. We divided the layer into five segments and applied the transmittance method of section 2.1 to the mean backscatter of 100 profiles of each segment (about 34 km) to calculate a 532 nm Sa and the two-color method of section 2.2 to calculate a 1064 nm Sa ratio. These values are shown in green (532 nm) and red (1064 nm) in Figure 10b. Clear air regions above and below each dust layer were identified manually by inspection of the profiles. Note that for this mesoscale layer the 532 nm Sa ratios range from 38 to 41 sr with an average of 40.1 sr and the 1064 nm Sa ratios range from 45.8 to 54.2 with an average of 50.9 sr.

[33] Figure 10 shows that the Sahara dust layers once elevated are consistent both geometrically (the layer is confined to 3–5 km altitude band) and optically (Sa variation at both wavelengths is small and the layer optical depth is near 0.3). The Sa ratio is an intensive aerosol property that depends on the composition, size distributions, and particle shape of the aerosol and its consistency is an indication that these layers stay intact over very long distances. Other studies have shown the transport of relatively unmixed Saharan mineral dust to the south American rainforest [Ansmann et al., 2009; Graham et al., 2003] and western Atlantic Ocean [Formenti et al., 2003; Kaufman et al., 2005], including the U.S. eastern seaboard [cf. Liu et al., 2008].

4.2. 26 August 2006

[34] The 26 August DC-8 flight included an underflight of CALIPSO. As is the case with the coincident CALIPSO-NAMMA measurements on 19 August, the collocated measurements are at one level and lack in situ profile measurements. In situ profiles of the size distributions were estimated from the DC-8 data obtained during the descent flight segment shown by a red dashed tilted oval in Figure 11a and flown about 2 h earlier than the CALIPSO underflight. The MOPITT CO levels (Figure 11b) for this period are slightly elevated. The extinction comparison shows that the layer observed by CALIPSO is more elevated than the one encountered by the DC-8. Figure 12 is a browse image of the CALIPSO 532 nm attenuated backscatter coefficients measured during the orbital segment corresponding to Flight 8 of the DC-8. The CALIPSO data used for comparison with the in situ extinction profiles were extracted from the region between the two white lines shown in Figure 12.

Figure 11.

CALIPSO flight tracks superimposed on (a) the MODIS optical depth and (b) the MOPPITT CO concentrations. The CALIPSO flight tracks are colored cyan and gray in the MODIS and MOPPITT images, respectively. The DC-8 flight track for 26 August 2006 is shown by the heart-shaped drawing in the middle of both figures. The CALIPSO underflight segment is the portion of the flight track in alignment with the CALIPSO flight track in cyan or gray in Figures 11a and 11b, respectively. The DC-8 in situ data for this obtained during the descent flight segment are shown by a red dashed tilted oval in Figure 11a.

Figure 12.

Browse image of the CALIPSO flight track shown in green in the inset map and the location of the near coincidence (in space) of the CALIPSO flight track and the DC-8 flight 8 on 26 August 2006. The color code is as in Figure 6. The white lines denote the CALIPSO data used for comparison with the DC-8 measurements.

[35] Figure 13 compares the CALIPSO extinction profile with a profile of the extinction derived from in situ measurements (i.e., the sum of the absorption and scattering coefficients). The maximum extinction coefficients of 160 and 140 Mm−1 by CALIPSO and the DC-8, respectively, are comparable, showing that the layer is intact after 2 h. The altitudes of the maximum layer extinctions for the CALIPSO and NAMMA measurements are offset by at least 1 km. The in situ measurements and CALIPSO observations are far removed from each other in this case (2 h and 1250 km). The data shown in Figure 12 are north of the NAMMA flight segment. These differences are shown by the mismatch in layer heights shown in Figure 13. Because of the relatively large mismatch, the constraint method, which requires good coincidence, is not applicable to this case to derive Sa. However, the transmittance technique can be applied using the CALIPSO measurement averaged over the region bounded by two white lines in Figure 12. Note that the dust layer overlies a streak of marine stratus just above the marine boundary layer. At 532 nm, the dust layer has an optical thickness of 0.41 and an Sa ratio of 38.2 sr. Although the Saharan layers can have horizontal extents of thousands of kilometers, it is possible that the layer observed by CALIPSO is not the same as the one measured by the in situ instruments aboard the DC-8. Notwithstanding this possibility, the properties measured by the two methods provide independent characterization of the Saharan dust and polluted dust layers.

Figure 13.

A comparison of the NAMMA in situ extinction profile and the extinction profile calculated from CALIPSO's attenuated backscatter measurement using 532 nm Sa ratio of 38.2 sr obtained using the transmittance method. The two measurements both taken on 26 August 2006 are offset by 2 h and ∼1250 km.

4.3. 1 September 2006

[36] Flight 10 of the DC-8 on 1 September 2006 neither directly underflew nor intercepted the CALIPSO orbit tracks. However, this flight sampled some of the same aerosol layers measured during a CALIPSO orbit track, as shown in the MODIS aerosol optical depth image in Figure 14a. The region of interest is denoted by the red circle. The DC-8 flight segment most relevant and of closest proximity is the descending leg located in this region. The MOPPITT CO concentrations are low (1.5–2 × 1018 molecules/cm2) in the comparison region. It is therefore likely that most of the aerosol mass is Saharan dust. CALIPSO preceded the DC-8 by 14 h. As in the previous flight of 26 August, the observed layers by CALIPSO and the in situ measurements may be different.

Figure 14.

CALIPSO flight tracks superimposed on (a) the MODIS optical depth and (b) the MOPPITT CO concentrations. The CALIPSO flight tracks are colored cyan and gray in the MODIS and MOPPITT images. The DC-8 flight track for 1 September 2006 is shown by the black lines in the middle of both figures.

[37] The CALIPSO browse image (Figure 15) shows the layers of interest marked by two white lines and comprising of 80 profiles. Note that though there are low clouds immediately above the marine boundary layer, there exists a small region above the clouds of relatively clean air. The transmittance and two-color methods yielded Sa ratios of 39.8 and 56 sr at 532 and 1064 nm, respectively. The optical depth determined from the inversion of the CALIOP attenuated backscatter using this lidar ratio (39.8 sr at 532 nm) is 0.55. In Figure 14a, the MODIS AOD is about 0.5 in this region.

Figure 15.

Browse image of the CALIPSO flight track shown in cyan in the inset map and the location of the near coincidence (in space) of the CALIPSO and the DC-8 flight tracks on 1 September during NAMMA. The color code is as in Figure 6. The white lines subtend the region averaged for the extinction comparison.

[38] It is a general transport pattern that Saharan dust aerosol goes through a phase of rising motion near the source then relative horizontal transport culminating in descent and deposition near the Americas [Ansmann et al., 2009; Formenti et al., 2003; Graham et al., 2003; Kaufman et al., 2005; Okin et al., 2004]. If the layer observed on 1 September 2006 by CALIPSO is the same as the one observed by the in situ measurements, then it was in the rising phase at a rate of approximately 0.1 km/h. The rate of ascent is based on the time difference between the CALIPSO overpass (3:30 am Local Time) and DC-8 flight track (6:30 pm Local Time) of nearest approach. The extinction profiles' comparison (Figure 16) between the DC-8 in situ measurements and the CALIPSO measurements shows an offset of 1.5 km in the altitude of maximum extinction coefficient. The layer shown at the same location has been elevated by 1.5 km since CALIPSO sampled it. During this rising phase, there is no evidence of significant deposition, since the maximum extinction coefficient, a property of the aerosol loading, does not decay appreciably.

Figure 16.

The NAMMA/CALIPSO extinction comparison for Flight 10 on 1 September 2006. Note that the centroid of the layer has shifted upward by about 1.5 km in between the CALIPSO measurement and NAMMA sampling. The CALIPSO measurement is ∼15 h earlier and nearly 750 km away.

5. Extinction-to-Backscatter (Sa) Ratios Calculation Based on NAMMA In situ Measurements

[39] To determine profiles of Sa ratios and validate the retrieved values, we perform numerical calculations for Saharan dust based on NAMMA in situ size distribution measurements. We use the DC-8 APS and UHSAS measurements (Chen et al., submitted manuscript, 2010) to determine coarse and fine size distributions and then calculate coarse and fine mode phase functions, as in Figure 3, using a T-Matrix scheme. The Sa ratio of the aerosol is derived from an area-weighted integral of the two modes. Figure 17 is a plot of the profile of Sa ratios of the 2 km dust layer encountered by NAMMA Flight 4 on 19 August 2006. Figure 17 shows a profile of the coarse number concentration, which marks the bottom and top of the layer at 2.5 and 4.6 km, respectively. This is very similar to the coincident CALIPSO extinction profile shown in Figure 8. The 532 and 1064 nm Sa ratios calculated by this method are 34.3 ± 2.0 and 50.2 ± 5.7 sr, respectively. These are in good agreement with Sa ratios of 38–41 sr at 532 nm independently determined from CALIPSO measurements using the transmittance technique (Figure 10) on the same day, albeit further downfield. The values are also in good agreement with 45.8–54.2 sr at 1064 nm found using the two-color method shown in Figure 10.

Figure 17.

A profile of the 1064 and 532 nm Sa ratios and integrated number distribution of coarse particles (0.583–6.264 μm) taken from a 20 min segment of the APS measurements during Flight 4 on 19 August 2006. The dust base and top are shown by the lines at altitudes of 2.5 and 4.6 km, respectively. The mean 532 and 1064 nm Sa ratios for the layer is 34.3 ± 2.0 and 50.2 ± 5.7 sr, respectively.

[40] Part of the flight on 25 August 2006 was dedicated to an intercomparison of the in situ measurements on the NASA DC-8 and the British BAe146. The DC-8 made a nearly straight and level flight through a dust cloud near 2 km. Figures 18a and 18b show the DC-8 altitudinal flight tracks and the calculated Sa ratios (and the coarse number concentration) for this flight, respectively. The dust layer is quite tenuous with maximum coarse number concentrations ∼20 cm−3. Scattering coefficients varied from 50 to 75 Mm−1 on intercomparison legs near 19°N latitude. The condensation nuclei (CN) concentration in the dust layers on this flight was fairly low, around 300 cm−3, while CO mixing ratio was 85 parts per billion by volume (ppbv) and the relative humidity was ∼50%–60%. Both the CN and CO concentrations infer the predominance of dust in the aerosol layer. The profiles of the 532 and 1064 nm Sa ratios shown in Figure 18 for this dust layer are quite consistent with means of 38.0 ± 2.5 and 48.7 ± 3.2 sr, respectively. The small standard deviations in both Sa ratios indicate that the layer remains very uniform with respect to composition, size distribution, and shape.

Figure 18.

(a) The flight track of the DC-8 on 25 August 2006. The straight and nearly level portion is through a dust layer at ∼2 km. (b) The calculated 532 and 1064 nm Sa ratios and coarse number particle concentration for the straight and level portion of this flight.

[41] Size distribution measurements were made during the NAMMA DC-8 Flight 8 on 26 August 2006 of a low-density dust layer between 0.6 and 1.6 km during the descent phase of the flight. This phase includes one stair step with the straight and level segment near 1.6 km. Profiles of Sa ratios calculated from these measurements are shown in Figure 19. The calculated values are 39.0 ± 1.5 and 45.9 ± 2.2 sr at 532 and 1064 nm, respectively.

Figure 19.

A profile of the 1064 and 532 nm Sa ratios and integrated number distribution of coarse particles during Flight 8 on 26 August 2006. The mean 532 and 1064 nm Sa ratios for the layer is 39.0 ± 1.5 and 45.9 ± 2.2 sr, respectively.

[42] This layer is optically thinner than the dust layer encountered on 19 August 2006. Nevertheless, the Sa ratios determined by T-Matrix calculation for this layer and the direct measurement for the denser layer on 19 August 2006 (40.1 and 50.9 sr at 532 and 1064 nm, respectively) are quite close. The calculated Sa ratio at 532 nm is also consistent with the value (38.2 sr) retrieved from the CALIPSO measurements on the same day.

[43] During the return flight on 26 August 2006, the DC-8 performed a stair-step descent flight consisting of two level sections and two descent sections. The first straight and level section (Leg 1 in Figure 20a.) was flown at an altitude of ∼2.25 km in a dense dust plume with coarse number concentrations near 35 particles cm−3. The mean Sa ratios are 42.4 ± 1.3 and 53.3 ± 2.0 sr at 532 and 1064 nm, respectively. The plume is fairly consistent as shown by the small standard deviations in the Sa ratios at both wavelengths. Figure 20b are profile plots of the two descent legs (legs 2 and 4 shown in Figure 20a the flight path). The break in ordinate demarcates the straight and level section (leg 3) of the flight. The 532 and 1064 nm Sa ratios for leg 2 are 46.6 ± 1.3 and 52.6 ± 1.7 sr, respectively. In leg 4, the aircraft has begun its descent into the marine boundary layer, and there is a sharp decline in the coarse number concentration. The Sa ratio at 532 nm has also dropped considerably signifying a change in the aerosol composition. The calculated Sa ratios for leg 4 are 32.8 ± 1.5 and 51.4 ± 10.8 sr at 532 and 1064 nm, respectively. Note that for leg 4, the 532 nm Sa ratios are fairly consistent while the 1064 nm values are quite noisy and decrease at lower altitudes. From 1.84 to 1.52 km, the decreasing trend in lidar ratios at 1064 nm with altitude corresponds to a decrease in the aerosol coarse mode number concentration. Since the aerosol coarse mode has a disproportionate impact on the 1064 nm Sa and the coarse mode number concentrations below 1.52 km are so low (<2/cm3), the calculated 1064 nm Sa is quite noisy compared to the 532 nm values.

Figure 20.

(a) Flight path and Sa ratios and the particle number concentration for the return flight on 26 August 2006. The figures are plotted by temporal profiles for the straight and level legs and vertical profiles for descent legs: (b) temporal profile of straight and level leg 1, (c) vertical profiles of descent legs 2 and 4, and (d) temporal profile of straight and level leg 3.

[44] Figure 21 shows the flight path for the in situ measurements on 20 August 2006. The time is in seconds after midnight UTC. Figure 21b shows a profile plot of the 532 and 1064 nm Sa ratios observed during the descent phase through a dust layer extending from an altitude of 1.6–4.8 km. The 532 and 1064 nm Sa ratios for this dust layer are 40.8 ± 3.2 and 51.6 ± 3.8 sr, respectively. Figure 21c shows the profiles of the 532 and 1064 nm Sa ratios observed during the ascent phase through the dust layer. The 532 and 1064 nm Sa ratios for this dust layer are 42.8 ± 3.0 and 51.8 ± 3.4 sr, respectively. The similarity of the vertical extent and the Sa ratios at each wavelength suggests that the same dust layer was sampled during both the ascent and descent legs of the flight. As noted before, these layers have spatially and temporally uniform Sa ratios and perhaps, by inference, fairly constant compositions and are geometrically quite stable.

Figure 21.

The (a) flight path of the DC-8 and Sa ratios and the particle number concentrations for the (b) descent leg and (c) the ascent leg of the flight of 20 August 2006.

[45] Figure 22 is a histogram of all the 532 and 1064 nm Sa ratios (∼1100 points) determined using the size distributions measured during NAMMA for this study. There is very little overlap of the two nearly normally distributed Sa ratios. The 532 and 1064 nm mean Sa ratios (±1 standard deviation) are 39.1 ± 3.5 and 50.0 ± 4.0 sr, respectively. The 532 nm values ranged from 30 to 53 sr and the 1064 nm values ranged from 32 to 66 sr, in both cases within the estimated ranges of 10–110 sr for all aerosol types [Ackermann, 1998; Anderson et al., 2000; Barnaba and Gobbi, 2004]. Figure 23 is a plot of the frequency distribution of the ratio of Sa ratios, i.e., Sa (1064 nm)/Sa (532 nm), a parameter used in the lidar ratio determination scheme outlined by Cattrall et al. [2005]. The plot shows that for the Sahara dust sampled during NAMMA there is very little spread in the ratio of Sa values. The mean 1064 nm Sa ratio is about 30% larger than the mean 532 nm value for this Saharan dust with a standard deviation of 10%, i.e., Sa (1064 nm)/Sa (532 nm) = 1.3 ± 0.13. Since the Sa ratio is an intensive property of the aerosol, its ratio is also an intensive property. The small spread in the ratios of Sa denotes that the Saharan dust aerosol observed during this period is quite consistent at least in its size distributions. To explore the effects of varying composition (refractive indices) and shape (aspect ratios) on the T-Matrix calculations, both of which were not directly measured for this study, we use the uncertainty analysis described in section 6.

Figure 22.

An ensemble of all the Sa ratio estimates (∼1100 records) based on the NAMMA in situ size distributions of dust layers and T-Matrix calculations. The average Sa ratios are 39.1 ± 3.5 and 50.0 ± 4 sr at 532 and 1064 nm, respectively.

Figure 23.

The ratio of the Sa ratios [Sa (1064 nm)/Sa (532 nm)] derived from the size distribution data used for this study.

[46] Table 2 summarizes the Sa ratios obtained for Saharan dust aerosols during NAMMA using the three methods. The 532 nm values are fairly consistent, while there is a somewhat wider spread in the 1064 nm values. This is particularly interesting because these measurements were made on various days and at various locations.

Table 2. Summary of Sa Ratio Measurements and Calculations
 Sa (532 nm) srSa (1064 nm) sr
Constrained Method
19 Aug 200635.7NA
Transmittance + Two-Color Methods
19 Aug 200638.953.5
19 Aug 200641.450.9
19 Aug 200638.154.2
19 Aug 200641.045.8
19 Aug 200641.050.2
26 Aug 200638.2NA
1 Sep 200639.856.0
Mean39.8 ± 1.451.8 ± 3.6
T-Matrix Using NAMMA Size Distribution Measurements
19 Aug 200634.3 ± 2.050.2 ± 5.7
25 Aug 200638.0 ± 2.548.7 ± 3.2
26 Aug 200639.0 ± 1.545.9 ± 2.2
26 Aug 200642.4 ± 1.353.3 ± 2.0
26 Aug 200646.6 ± 1.352.6 ± 1.7
26 Aug 200632.8 ± 1.551.4 ± 10.8
20 Aug 200640.8 ± 3.251.8 ± 3.4
20 Aug 200642.8 ± 3.051.8 ± 3.4
Mean39.1 ± 3.550.0 ± 4.0
Noise Laden T-Matrix Simulation
Mean39.4 ± 5.956.5 ± 16.5

6. Uncertainty Analysis of the Modeled Sa Ratios

[47] In this section, we attempt to propagate the uncertainty in the input variables to the calculated Sa ratio and determine the overall uncertainty in the calculated dust 532 and 1064 nm Sa ratios using a generalized analytical uncertainty equation. The uncertainty equation is given by the Taylor series of the deviations (yyo) of the output (y) from its nominal value (yo) and is expressed in terms of the deviations of the i inputs (xxio) from their nominal values. As in the work of Morgan and Henrion [1990], for the first three terms, the uncertainty is

equation image

[48] The subscripts Xo denote derivatives evaluated at the nominal values. Assuming that there are no covariances between the input variables, all the mixed derivatives in equation (5) would equal zero. The only terms that would not be zero are the first term and the terms with i = j and i = j = k in summations II and III, i.e., second and third derivatives of the variables, respectively. Since the covariances are not known, we cannot make the assumption that they are negligible. Given the uncertainties in the variables from which the Sa ratio is calculated, the uncertainty in Sa can be estimated without making any assumptions about covariances between inputs. To accomplish this, we use Latin Hypercube Sampling (LHS) [Iman and Conover, 1980], a statistical sampling method in which a distribution of plausible scenarios of parameter values is generated from a multidimensional distribution. Unlike classical Monte Carlo sampling methods, LHS precludes duplication by requiring that each square grid containing sample positions has only one sample in each row and each column. For this study, we generate 500 variables for each of the seven uncertain parameters used in the calculation of Sa. We then randomly combine these variables to yield 500 instances or events. Each of these events has a very high probability of yielding a unique Sa ratio. We then perform standard descriptive statistics on the Sa values.

[49] The mean and standard deviations of these values is an estimate of the nominal value and uncertainty of dust Sa. We use the nominal (or central) values in Table 3, suggested by the studies referenced in section 2, to generate the random scenarios. The fine mode radii, coarse mode radii, and imaginary refractive indices are log normally distributed. The fine and coarse geometric standard deviations (GSD), the real refractive indices, and the aspect ratios are normally distributed.

Table 3. Ranges of the Variables Used to Generate 500 Random Combinations of Inputs for T-Matrix Calculationsa
  • a

    The central values of the size distributions are based on the mean values of the dust layer observed on 19 August 2006 during the NAMMA campaign.

Geometric Fine Radius0.02160.194
Geometric Coarse Radius0.1881.69
Fine GSD1.601.80
Coarse GSD1.501.70
Real Refractive Index1.451.55
Imaginary Refractive Index0.000670.006
Aspect Ratio1.702.30

[50] The results obtained by using this method do not assume that the input variables are independent of each other or that Sa ratio is linear in the individual input variables, i.e., the second- and higher-order derivatives in equation (5) are not necessarily equal to zero. The statistics of the resulting Sa ratios provide an uncertainty envelop of the Sa ratio estimates based on the uncertainty of the inputs. Moreover, the results can be used to explore the sensitivity of the Sa ratios at each wavelength to the various aerosol properties. In Figure 24, the 532 nm Sa ratios are well constrained with a standard deviation of 15% of the mean after perturbing nominal input values shown in Table 3 and similar to the distribution shown in Figure 22. The 1064 nm Sa ratios are much more sensitive to the addition of noise as shown by the wide spread of 1064 nm Sa ratios in Figure 24. A parameter that has a significant impact on the Sa ratios is the complex refractive index. Figure 25 is a 2-D histogram of the 532 and 1064 nm Sa ratios as functions of the real and imaginary parts of the refractive indices. The range of these values is from 0.00067 to 0.006 for the imaginary part and 1.45–1.55 for the real part. The 1064 nm Sa ratios are sensitive to changes in the refractive index throughout the ranges of these variables. The 532 nm Sa ratio is sensitive to changes in the complex refractive index at the lower ranges (Figures 25a and 25c). The highest density of 532 nm Sa ratios are found in midranges of both the real and imaginary part, and these values are insensitive to changes in the complex refractive index. Note that this variation is a total derivative, i.e., all parameters are allowed to vary independently for the scenarios. The 1064 nm Sa ratios decrease with the complex index of refraction almost monotonically through the range of values (Figures 25b and 25d). Sa variations with the complex refractive indices at the two wavelengths reflect the sensitivity of these values to the size distributions used in this study. Note that for these uncertainty analyses we used the size distributions of the dust layer observed on 19 August 2006. Since these are dominated by the coarse mode, any changes in the physico-chemical properties show a greater impact on the 1064 nm Sa than the 532 nm Sa.

Figure 24.

The distribution of the Sa ratio calculated from a Latin Hypercube Sampling (LHS) of 500 events bounded by the ranges shown in Table 3. The normal means (±1 standard deviation) of the 532 and 1064 nm Sa ratios are 39.4 ± 5.9 and 56.5 ± 16.5 sr, respectively.

Figure 25.

Two-dimensional frequency plots showing the variation of the 532 and 1064 nm lidar ratios with (a, b) imaginary and (c, d) real parts of the complex refractive indices for 500 randomly generated events discussed in section 6.

7. Conclusion

[51] We have determined the Sa ratios of Saharan dust layers using three methods: transmittance constraint technique for lofted layers, in situ extinction profile constraint method, and T-Matrix calculations. We found quite robust Sa ratios at 532 nm and a slightly wider spread at 1064 nm. The three methods yielded 532 and 1064 nm Sa ratios that are quite close. Sa ratios of 39.8 ± 1.4 and 51.8 ± 3.6 sr at 532 and 1064 nm, respectively, were found using the transmittance and two-color methods applied to CALIPSO measurements of Saharan dust lofted layers. T-Matrix calculations applied to size distributions measured aboard the NASA DC-8 during NAMMA yielded Sa ratios of 39.1 ± 3.5 and 50.0 ± 4 sr at 532 and 1064 nm, respectively. The measured extinction profile obtained by the aggregate of nephelometer measurements of total scattering coefficient and the PSAP measurements of the absorption coefficient was used to constrain the inversion of the CALIPSO measurements. This technique yielded a 532 nm Sa ratio of 35.7 sr for a dust layer and 25 sr for the aerosol in the marine boundary layer. We perturbed seven microphysical and chemical properties of the dust aerosol and computed the Sa ratios by randomly combining distributions of these parameters to assess the uncertainty in the Sa calculation. This uncertainty simulation generated a mean (±uncertainty) Sa of 39.4 (±5.9) and 56.5 (±16.5) sr at 532 and 1064 nm, respectively, corresponding to percentage uncertainties of 15% and 29%. The ratio of the Sa ratios, Sa (1064 nm)/Sa (532 nm) = 1.3 ± 0.13, and is nearly normally distributed. The simulation revealed that Sa is insensitive to the middle range of refractive indices at 532 nm and nearly monotonically dependent on the refractive indices at 1064 nm. This explains the observed robust Sa ratios at 532 nm and relatively wide spread of Sa at 1064 nm.

[52] This study examined a wide range of dust loadings in various locations within ∼1250 km of Saharan Desert source regions on different days. The Sa ratios do not change significantly with dust loading, suggesting the dust microphysical and chemical properties do not vary appreciably. The profile of Sa ratios at 532 nm are very close to 40 sr using different methods and on different days. There is a wider variation in the 1064 nm Sa ratios for all methods. It is possible that the variation of Sa at 1064 nm is driven by a higher sensitivity to changes in the complex index of refraction at the 1064 nm wavelength.


[53] We thank the relevant DC-8 instrument PIs for the NAMMA data, the MODIS and MOPPITT teams for their data, and the CALIPSO team for data, guidance, and many useful discussions. We acknowledge the use of the public-domain T-matrix code by Michael Mishchenko and thank Kam Pui Lee, Susan Kooi, and Sharon Burton for the CALIPSO and LASE images. We gratefully acknowledge funding support from the NASA Radiation Sciences and Atmospheric Composition Programs, respectively, managed by Hal Maring and Bruce Doddridge.