CALIPSO lidar observations of the optical properties of Saharan dust: A case study of long-range transport

Authors


Abstract

[1] An extensive dust storm originating on 17 August 2006 in North Africa was observed and tracked by the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) lidar. Over the next several days, the dust layer moved westward across the Atlantic Ocean and into the Gulf of Mexico. The initial stages of the event were examined using a sequence of CALIPSO measurements. The first of these was acquired very near the source on 18 August. Successive measurements were made over the Atlantic Ocean on 19 and 20 August, at respective locations approximately ∼1300 km and ∼2400 km from the source region. The later stages of the event were assessed using measurements acquired by the NASA Langley Research Center airborne HSRL over the Gulf of Mexico on 28 August. Within the free troposphere, the intrinsic optical properties of the dust remain relatively unchanged for the first 3 d of transport over the Atlantic Ocean. This is consistent with previous in situ measurements that have shown that there is little change in the size distribution of dust as it crosses the Atlantic. After the 10 d journey to the Gulf of Mexico, some changes are seen in the lidar ratios, the backscatter color ratio, and the optical depth ratio. The linear depolarization ratio appears to remain essentially constant (∼0.32) at all four locations mentioned above, demonstrating a notable consistency in the dust particle nonsphericity. The measured 532 nm lidar ratios are 41 ± 3, 41 ± 4, 41 ± 6 and 45.8 ± 0.8 sr, respectively, at locations near the source, over the Atlantic Ocean, and in the Gulf of Mexico. The corresponding 1064 nm lidar ratios are 52 ± 5, 55 ± 5, 54 ± 13 and 44 ± 8.3 sr. The 532 nm lidar ratios are consistent with previous measurements and with CALIPSO's prelaunch models. The lidar ratios retrieved at 1064 nm are somewhat larger than would be expected on the basis of existing modeling studies. The backscatter color ratios are 0.74 ± 0.07, 0.75 ± 0.08, 0.72 ± 0.04 and 0.62 ± 0.01, and the optical depth ratios are 0.97 ± 0.02, 1.01 ± 0.05, 0.93 ± 0.17 and 0.62 ± 0.13, respectively.

1. Introduction

[2] The Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) mission was launched successfully in April 2006 [Winker et al., 2006, 2007]. The primary payload on board the CALIPSO satellite is a two-wavelength, polarization-sensitive backscatter lidar known as CALIOP (i.e., the Cloud-Aerosol Lidar with Orthogonal Polarization). Among the many instruments aboard the A-Train constellation of satellites, CALIOP is unique in its ability to measure high-resolution vertical profiles of both clouds and aerosols within the Earth's atmosphere [Winker et al., 2004]. Since launch, CALIPSO has been accumulating a near-continuous global data set of range-resolved backscatter measurements.

[3] Among the key scientific objectives of the CALIPSO mission are the acquisition of a global set of measurements that will greatly improve the accuracy of our estimates of the direct and indirect climate forcings due to aerosols. Validating the CALIPSO aerosol models and aerosol classification algorithms is thus an important metric for gauging mission success. Desert dust is a common and important aerosol type that can exert significant influences on the atmosphere. Because its effects are felt both locally, at the source region, and at distances far from the source region, accurate characterization of the physical and optical properties of dust is a key element in establishing the success of the CALIPSO aerosol retrievals of optical properties.

[4] North Africa is the largest dust source in the world. Dust from North Africa is frequently transported over the Atlantic Ocean and the Mediterranean and Caribbean Seas [Prospero, 1996; Prospero and Carlson, 1972; Karyampudi et al., 1999; Ganor and Mamane, 1982; Moulin et al., 1998; Colarco et al., 2003]. Transport of Saharan dust across the tropical North Atlantic Ocean and into the Caribbean region reaches a maximum during the summer. The mechanisms for dust transport into the Caribbean have been studied for some time [e.g., Carlson and Prospero, 1972]. Cool, moist northeasterly air crossing the Mediterranean into Africa experiences intense heating over the arid continent. The air advects westward in the predominantly easterly flow, developing into a dust-laden, well-mixed layer extending from the surface to an altitude of several kilometers. As this hot, dry air emerges from the west coast of Africa, the base of the air mass rises quickly because it is undercut by the relatively cool and moist trade winds. The dust layer is usually confined by two inversions, one above the layer and one below. This elevated layer of warm, dry, dusty air is frequently referred to as the Saharan Air Layer (SAL). Transport of the SAL across the Atlantic typically requires 5 to 6 d (the westward speed of the Saharan air mass is usually about 8 m/s). During this same time, the SAL drops vertically by an average of 1–2 mm/s between Africa and the Caribbean [Carlson and Prospero, 1972].

[5] Previous studies have analyzed the transport of Saharan dust by combining space-based lidar measurements from the Lidar In-space Technology Experiment (LITE) with additional satellite data acquired by passive sensors [Karyampudi et al., 1999]. CALIOP represents a substantial improvement over LITE. The addition of polarization-sensitive optics in the CALIOP receiver permit the unambiguous recognition of dust layers using depolarization ratio measurements. Furthermore, CALIOP's much larger dynamic range minimizes the potential for misclassifying especially dense dust layers as clouds. CALIPSO measurements thus provide a greatly enhanced data set that is ideal for the study of dust vertical structure and transport. In this paper, we exploit the improved measurement capabilities provided by CALIOP to analyze the changes in spatial and optical properties that occur along the transport track of an extensive Saharan dust event.

[6] For the dust storm that occurred over the Sahara desert on 17 August 2006, dust particles were lofted to a maximum altitude of ∼6.6 km and transported over Atlantic Ocean to the Gulf of Mexico. The resulting dust plume was observed by CALIOP for more than 10 d after its generation. The plume was also measured by the NASA Langley Research Center airborne high-spectral-resolution lidar (HSRL) [Hair et al., 2006] off the coast of Texas during a CALIPSO validation flight conducted on 28 August. Extended data segments acquired by CALIOP along the transport track of the dust plume near the source, and again at locations ∼1300 km and ∼2400 km away from the source, have been analyzed. This paper presents the results of these analyses and the dust optical properties derived from collocated HSRL measurements acquired off the coast of Texas.

2. Long-Range Transport of a Saharan Dust Event

[7] Figure 1 presents CALIOP observations of the time evolution and transport of the Saharan dust event. The red lines in the image, which originate in the Gulf of Mexico and extend eastward to the west coast of Africa, represent back trajectories produced by the NASA Langley Trajectory Model (LTM) [Pierce and Fairlie, 1993; Pierce et al., 1994]. The 10-d back trajectories are initialized on 28 August at the HSRL flight segment marked with a white line in Figure 1. The vertical images superimposed on the map show the 532 nm attenuated backscatter measured by CALIOP when passing over the dust transport track. The data from 17 through 23 August were acquired during nighttime. The 25 and 28 August data were measured during daytime. The dust layers in the CALIOP images, which are marked with the letter “D”, have been identified from their volume depolarization ratios (i.e., the ratios of perpendicular to parallel polarization components of the measured total signal including aerosol and molecular scattering). For irregular particles like dust, the lidar return signal is depolarized and hence the depolarization ratio is frequently used to discriminate nonspherical particles from spherical particles [Sassen, 2000; Murayama et al., 2001]. The altitudes of the trajectories are seen to be lower than that of the dust layer observed by CALIPSO before the dust plume reaches the Caribbean Sea. This may be because the trajectories are calculated using wind fields resolved in a 2° × 2° grid, and do not account for other processes that might have influenced the vertical displacement of the layer, such as gravitational settling, wet removal or subgrid-scale processes.

Figure 1.

A dust event that originated in the Sahara desert on 17 August 2007 and was transported to the Gulf of Mexico. Red lines represent back trajectories indicating the transport track of the dust event. Vertical images are 532 nm attenuated backscatter coefficients measured by CALIOP when passing over the dust transport track. The letter “D” designates the dust layer, and “S” represents smoke layers from biomass burning in Africa (17–19 August) and South America (22 August). The track of the HSRL measurement is indicated by the white line superimposed on the 28 August CALIPSO image. The HSRL track is coincident with the track of the 28 August CALIPSO measurement off the coast of Texas between 28.75°N and 29.08°N. Plot courtesy of Kurt Severance at NASA Langley Research Center.

[8] In addition to dust, other aerosol types are also evident in Figure 1. The letter “S” is used to indicate smoke plumes that originate from biomass burning events in Africa (17–19 August) and South America (22 August). These smoke layers are easily distinguished from the dust by their very low depolarization ratios and by the strong wavelength dependence of the signal attenuation within the layer. The attenuation of smoke aerosols is generally much higher at 532 nm than 1064 nm.

[9] From Figure 1, the images from CALIPSO observations along with the back trajectory analysis indicate that the dust storm originated over the Sahara on 17 August 2006 (or perhaps somewhat earlier), and that the dust plume was subsequently transported to the Gulf of Mexico. A second set of back trajectories, produced with the NOAA Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model [Draxler and Rolph, 2003] and shown in Figure 2, confirm the transport of the dust plume from the Sahara desert areas. These trajectories show clearly that the air masses over the Gulf of Mexico where we had the HSRL measurement on 28 August are mostly from North Africa.

Figure 2.

Back trajectories for three heights started at 1900 UTC on 28 August 2006 at a location of 28.84°N and 94.78°W along the ground track of the HSRL measurement for three altitudes of 500 m (red curves), 1500 m (blue curves), and 2500 m (green curves), produced with HYSPLIT (http://www.arl.noaa.gov/ready/). (top) Horizontal components and (bottom) vertical components of the trajectories.

[10] A closer look at the vertical distribution of the dust is provided by the averaged profiles of the volume depolarization ratio and attenuated backscatter signal that are presented in Figure 3. These profiles have been averaged for 500 single-shot profiles for nighttime measurements (17–23 August) and 1000 for daytime measurements (25 and 28 August). Profiles with cloud signals present in 0–8 km altitudes have been removed. Standard CALIPSO data processing provides cloud and aerosol discrimination on a layer basis using layer averaged attenuated backscatter and total color ratio [Liu et al., 2004]. Here we simply set a threshold based on the measured attenuated backscatter. Those individual profiles that contain samples exceeding the threshold value are not used in constructing the average. Error bars indicate the standard deviation due to detection noise, and are computed using the technique based on the so-called noise scale factor (NSF) developed by Liu et al. [2006]. The NSF represents the proportionality between the variance and the mean of random samples, and can be used easily to estimate uncertainties for each sample in the lidar data. This factor is measured and provided for each profile in the standard CALIPSO lidar level 1 data products. The additional solar background present in daytime measurements cause the 25 and 28 August profiles to be noisier, despite a greater amount of data being used to generate the averages.

Figure 3.

(top) Depolarization ratio and (bottom) attenuated backscatter average profiles. The average number of laser shots is 500 for 17–23 August and 1000 for 25 and 28 August, corresponding to the ranges represented in Figure 3 (bottom). Vertical resolution is 30 m. Cloudy profiles have been removed using a threshold screening method. Error bars indicate ±1σ errors estimated using the noise scale factor [Liu et al., 2006]. Because the 25 and 28 August data were acquired during daytime, these profiles are nosier than nighttime measurements (17–23 August). Numbers in Figure 3 (top) indicate the top height of the dust layer measured during night (17–23 August) where the depolarization ratio is larger than 0.04. Because of the additional solar background noise present in daytime depolarization ratio measurements, the top height of the dust layer for 25 and 28 August could not be determined in the same way as was done for nighttime data. However, the top heights for 25 and 28 August appear to be between 3 and 4 km.

[11] Examination of the profiles in Figure 3 show that the dust vertical structure changes during the transport. The lofted dust particles are seen to attain a maximum altitude of ∼6.6 km at the source region. As it progresses across the Atlantic Ocean, the top of the layer slowly descends, so that upon reaching the Gulf of Mexico the maximum height of the layer is ∼3.3 km.

[12] From the previous model studies [e.g., Colarco et al., 2003], the possible processes that influence the dust vertical distribution during the transport across the North Atlantic Ocean could be sedimentation, the general descent of air through downward vertical winds, and wet removal. However, these processes may have played different roles at different stages of the transport. Sedimentation is an important mechanism in the removal of large dust particles over the source region. However, in situ measurements conducted over the Canary Islands near the west coast of Africa and over Puerto Rico have shown that the vertical descending speed of dust particles is smaller than the Stokes gravitational settling speed by ∼0.33 cm/s [e.g., Maring et al., 2003a]. As a result, dust particles smaller than 7.3 μm are not preferentially removed during transport over the Atlantic Ocean, and the size distribution remains largely unchanged for smaller dust particles. In addition, the dust size distributions measured in Puerto Rico show only a small height dependence [Maring et al., 2003b; Reid et al., 2003]. In this paper, the intrinsic optical properties retrieved for dust in free troposphere are seen to remain essentially constant throughout the transport across the Atlantic Ocean. Furthermore, the vertical homogeneity of these properties (see section 4) suggests a nearly constant size distribution of dust particles, consistent with previous in situ measurements by Maring et al. [2003a], Maring et al. [2003b] and Reid et al. [2003].

[13] The mechanisms driving the transport of the dust layer are beyond the scope of this paper and will not be discussed further. However, we note several phenomena that are immediately obvious from the CALIPSO measurements. The transport of this dust event appears to have followed a “typical” pathway: the SAL generated over North Africa continent, then emerged from the west coast of Africa, advected across the tropical North Atlantic, and finally entered the Caribbean region and was trapped in the area for several days. The transport from the west coast of Africa to the Caribbean took 5–6 d. The mean descending motion of the top of the dust layer is ∼6 mm/s (based on the depolarization profiles of 17 and 23 August). After entering the Caribbean and Gulf of Mexico, the dust layer top altitude does not appear to change dramatically. In the source (17 August), dust particles are distributed from the surface to ∼6.6 km. Over the Atlantic Ocean (18–23 August), the majority of the dust particles are confined to the free troposphere, showing some classic SAL features [Prospero and Carlson, 1972]. However, the depolarization ratio measurement indicates the presence of some amount of dust in the marine boundary layer (MBL). Colarco et al. [2003] has suggested regional subsidence may bring mineral dust from aloft into MBL over the Atlantic Ocean. After entering the Caribbean region (25 and 28 August), it is evident from the depolarization measurement that more dust particles are mixed into MBL and there is no clear boundary between the dust layer in the free troposphere and MBL. This vertical distribution pattern was also observed by in situ measurements in this area [Reid et al., 2002].

[14] The same dust transport event, from North Africa to the Gulf of Mexico, was also observed by MODIS, and is plainly evident in the aerosol optical depth (AOD) measurements provided by MODIS-Aqua. Figure 4 presents the Aqua AODs measured on 17–28 August, along with the CALIPSO ground tracks for the images shown in Figure 1. We note the time difference between the Aqua track and the corresponding overlaid CALIPSO track. The Aqua data were acquired during day and the CALIOP data were acquired during following night for 17–23 August and during day for 25 and 28 August, so that the Aqua measurements occur about 11 h earlier than the corresponding CALIPSO nighttime track and about 1 min earlier than the CALIPSO daytime track. The Aqua data clearly shows the outflow of a dense aerosol layer off the west coast of North Africa that originated inland on 17 August.

Figure 4.

MODIS Aqua aerosol optical depth (AOD) between 1300 UTC and 2100 UTC measured on 17–28 August 2006. The map boundaries are 120–0°W and 10°S–40°N. Solid lines represent corresponding CALIPSO ground tracks for the images in Figure 1. The Aqua data were acquired during day, and the CALIPSO data were measured during both night (17–23 August) and day (25 and 28 August). A time difference between Aqua and CALIPSO footprints in the dust region is about 11 h for the CALIPSO nighttime orbit and about 1 min for the CALIPSO daytime orbit.

[15] The Navy Aerosol Analysis and Prediction System (NAAPS, available http://www.nrlmry.navy.mil/aerosol/#aerosolobservations) model results for the same time period are presented in Figure 5. The NAAPS model predicts the generation and subsequent transport of a dust plume, corresponding almost exactly to the one observed by CALIOP and Aqua. According to the NAAPS simulation, the dust that is transported to the Gulf of Mexico remains largely unmixed with smoke aerosol. However, the LTM results (Figure 1) indicate a possibility that after entering the Caribbean region the dust layer may have become mixed with a very small amount of smoke aerosol transported from Africa.

Figure 5.

Navy Aerosol Analysis and Prediction System (NAAPS) simulated optical depths at 550 nm for the time period of 17–28 August. The map boundaries are 10–80°W and 0–40°N for 17–22 August and 50–100°W and 10°S–30°N for 23–28 August.

3. Lidar Retrieval Methods

[16] The CALIPSO data segments identified for the dust model validation studies occurred on 18 August near the source, and on 19 and 20 August approximately 1300 and 2400 km away from the source. Analyses were conducted to retrieve a complete set of optical properties for both layers, including lidar ratio, particulate depolarization ratio, 1064 nm to 532 nm particulate backscatter ratio (or color ratio), and dust optical depth ratio (also 1064/532). To assess the changes that occur after the long-range transport, these derived optical properties are then compared with the optical properties measured by the HSRL over the Gulf of Mexico.

[17] Two methods are used for the CALIOP lidar signal analyses in this study. Each is described in the sections below.

3.1. Optical Depth Constraint Method

[18] It has long been established that for a given value of the particulate lidar ratio, Sp, profiles of particulate backscatter coefficients, βp(r), can be retrieved using [Fernald, 1984]

equation image

In this expression, X(r) is the range-corrected lidar backscatter signal. For CALIOP measurements, additional corrections are applied to X(r) at 532 nm to account for absorption by stratospheric ozone. Sm is the lidar ratio for molecular scattering and is readily derived from theory [e.g., Bucholtz, 1995]. βm(r) is the molecular backscatter coefficient, which is computed from available meteorological data (e.g., for CALIPSO, from model data provided by NASA's Global Modeling and Assimilation Office [Bloom et al., 2005]). CT2 is the boundary condition and can be determined, for example, by fitting the available clear air signal to the model molecular signal immediately above the layer top.

[19] To obtain aerosol backscatter and extinction profiles, the CALIOP Level 2 production code selects lidar ratios via a measurement-based assessment of aerosol type [Omar et al., 2005, 2006]. On those occasions when an independent measurement of layer optical depth is available, the initial, model-provided value of Sp can be iteratively adjusted until the calculated optical depth obtained from (1) matches the measured optical depth. If sufficient regions of clear air exist both above and below a layer [Sassen and Cho, 1992; Young, 1995], reliable estimates of optical depth at 532 nm can be obtained directly from the CALIOP data itself. This clear-air technique can be used to determine the optical depth of cirrus clouds and some lofted aerosol layers. However, the dust layers measured by CALIOP typically extend down to the surface, and thus the required high-resolution optical depth measurements cannot generally be obtained in this manner.

[20] Recently, a new technique has been described for obtaining estimates of layer optical depth for those transparent cloud and/or aerosol layers that lie above opaque water clouds [Hu et al., 2007b]. For a single, opaque cloud layer in an otherwise clear atmosphere, the integrated attenuated backscatter can be expressed as [Platt et al., 1999]

equation image

Here βc(r) is the range-resolved cloud attenuated backscatter coefficient, Sc is the cloud lidar ratio, and η is the layer effective multiple scattering factor [Platt et al., 1999]. If, however, there are additional transparent cloud or aerosol layers overhead, then the integrated attenuated backscatter of the opaque layer is reduced by the two-way transmittance of the transparent overlying layer [Hu et al., 2007b]; that is

equation image

For opaque water clouds measured using a polarization-sensitive lidar like CALIOP, equation (3) represents a single equation in one unknown: τp. γopaque is obtained by integrating the backscatter signal between layer top and (apparent) base. Sopaque is the lidar ratio for water clouds, and is known to have a uniform value very near 19 sr [Hu et al., 2007b; Pinnick et al., 1983; O'Connor et al., 2004]. The final piece of the puzzle is the water cloud multiple scattering factor, ηopaque, which can be computed from water cloud layer-integrated volume depolarization ratio, δv,int [Hu et al., 2007a]:

equation image

The optical depth of the overlying layer is thus derived using

equation image

The value of τp thus obtained can in turn be used to constrain equation (1), so that the layer lidar ratio as well as backscatter and extinction coefficients can be retrieved via iterative improvement [Sassen and Cho, 1992; Young, 1995]. The optical depth constraint method is applied to the 20 August data wherever opaque water clouds occur beneath the dust layer. We note that because the depolarization measurement is not available at 1064 nm for CALIOP, the depolarization ratio measured at 532 nm is used in the 1064 nm analysis, which may introduce additional error. Monte Carlo simulations [Hu et al., 2006] show that the difference in depolarization ratios at 532 nm and 1064 nm wavelengths in water clouds arose from the multiple scattering is dependent on cloud types but smaller than 5%.

[21] Having an extinction solution allows the subsequent retrieval of the particulate depolarization ratio profile, δp. The volume depolarization ratio, δv, includes contributions from both aerosol and molecular scattering components (i.e., δv is the ratio of the perpendicular and parallel components of the attenuated backscatter signal). The particulate depolarization ratio is the ratio of the perpendicular to parallel components of the particulate scattering signal only, and is determined using the following equation:

equation image

Here R is the backscatter ratio, such that R(r) = (βp(r) + βm(r))/βm(r), and δm is the molecular depolarization ratio. Because CALIOP has a very narrow bandwidth filter (∼37 pm), only the central Cabannes line of the Rayleigh scattering signal is received, and thus δm ≈ 0.0036 at 532 nm.

3.2. Two-Wavelength Method

[22] The two-wavelength method derives layer-effective optical properties from simultaneous measurements acquired at two different wavelengths (e.g., for CALIOP, at 532 nm and 1064 nm) [Sasano and Browell, 1989; Liu et al., 2000; Vaughan, 2004]. For the technique utilized in this work, the layers being analyzed are assumed to be homogeneous with respect to particle composition and size distribution, so that (1) the backscatter color ratio, χ, defined as the ratio of particulate backscatter coefficients (i.e., χ(r) = βp,1064(r)/βp,532(r)) is a constant throughout the layer and (2) the lidar ratios at both wavelengths are likewise constant throughout the layer [Vaughan, 2004]. Given a solution to the lidar equation at one wavelength, the two-wavelength method then determines the required optical parameters necessary to generate a solution at the second wavelength that is guaranteed to be consistent with the governing assumptions.

[23] The method proceeds in two stages. First, to initiate a two-wavelength solution, we specify a value for Sp,532, and apply (1) to generate a solution at 532 nm. We then invoke the assumptions cited earlier, and use the solution derived at 532 nm to formulate a nonlinear least squares (NLLS) problem that expresses the 1064 nm particulate backscatter measurements in terms of the 532 nm particulate backscatter solution and the unknown parameters Sp,1064 and χ [Vaughan, 2004]:

equation image

In this expression, γp,532 is the layer integrated 532 nm particulate backscatter, defined by

equation image

and B1064(rk) is the 1064 nm particulate-attenuated total backscatter coefficient at the kth range bin. Standard numerical techniques [e.g., Dennis and Schnabel, 1996] can be used to solve equation (7), thereby generating optimal estimates (in a least squares sense) of Sp,1064 and χ. The value of Sp,1064 can then be used in equation (1) to derive a profile of particulate backscatter coefficients from the 1064 nm measurement.

[24] When the layer optical depth at 532 nm can be derived using either the clear-air technique or opaque water cloud technique, the initially selected value of Sp,532 can be iteratively adjusted until the calculated optical depth obtained from equation (1) matches the measured optical depth. In such cases, the value of Sp,1064 subsequently derived by solving the NLLS problem will yield the optimal estimate of layer optical depth at 1064 nm [Vaughan, 2004]. For the HSRL measurement, the 532 nm lidar ratio and corresponding particulate backscatter profile are measured directly. The two-wavelength technique can then be used to determine the lidar ratio at 1064 nm. This technique has been applied to the HSRL dust measurement on 28 August to derive 1064 nm optical properties.

[25] For this study, a second approach has been developed to handle those cases when the lidar ratios and optical depths at both 532 nm and 1064 nm are all unknown, and cannot be estimated using existing techniques. Similar to the procedure used in previous studies [Sasano and Browell, 1989; Liu et al., 2000], we specify a range of 532 nm lidar ratios from 1 sr to 100 sr, in increments of 1 sr. We note that both the range and increment are adjustable. The range should be selected large enough so that the true value of Sp,532 is within the range. The range from 1 sr to 100 sr covers lidar ratios for most aerosol types; in particular, it comfortably spans the full range of lidar ratios previously reported for Saharan dust [Mishchenko et al., 1997; Barnaba and Gobbi, 2001; Liu et al., 2002; Shimizu et al., 2004; Sakai et al., 2002; Muller et al., 2003, 2007; Mattis et al., 2002; Dubovik et al., 2006; Tomasi et al., 2003; Cattrall et al., 2005; Berthier et al., 2006; Mona et al., 2006]. For each value of S532, an extinction solution is attempted at 532 nm. For successful attempts, the 532 nm lidar ratio and the derived extinction solution then serve as the inputs for the solution of the NLLS problem. Repeating this procedure for all specified values of Sp,532 generates a sequence of triplets, {Sp,532, Sp,1064, χ}, where each triplet represents a set of optical parameters guaranteed to satisfy our assumptions regarding layer composition homogeneity. The triplet of values that best satisfies these assumptions is defined as that triplet for which the value of F(χ, Sp,532, Sp,1064) computed by equation (7) is minimized. We therefore record the value of F(χ, Sp,532, Sp,1064) obtained for each of the trial Sp,532 values. Upon completion, the resulting series is searched to ascertain the global minimum. The set of {Sp,532, Sp,1064, χ} values corresponding to this global minimum is subsequently used to retrieve the remaining optical parameters from the layer. These include the particulate backscatter and extinction profiles, optical depths, and particulate depolarization ratio profiles.

[26] To ensure the numerical stability of the solutions, this second approach is only applied to relatively dense and homogeneous layers that have high signal-to-noise ratios (SNR). The real-world performance of the two-wavelength technique depends on the sensitivity of the extinction/backscatter retrieval (i.e., as in equation (1)) to the lidar ratio; the more sensitive the retrieval, the better stability and accuracy of the lidar ratio determination. Weakly scattering layers are largely insensitive to lidar ratio specification, as solutions can be generated for a wide range of lidar ratios while producing little change in the derived optical depths. Conversely, obtaining successful solutions for strongly scattering layers with relatively large optical depths depends critically on the accurate selection of layer lidar ratio. For these layers, physically meaningful solutions can be derived only within a narrow range of lidar ratios. The dust layers measured by CALIPSO on 18 and 19 August are both physically and optically thick, with a very robust backscatter signatures, and thus are excellent candidates for successful application of the second two-layer approach.

3.3. Calibration Procedures

[27] Accurate retrievals of layer optical depths and lidar ratios require an accurate calibration of the backscatter signals. The CALIOP 532 nm profiles are calibrated by normalizing the measured data in an altitude range of 30–34 km to the appropriate molecular model [Hostetler et al., 2006]. The 532 nm data calibration is thought to have an uncertainty of ∼3%, due to measurement noise, error in determining molecular number density, and possible presence of a small amount of aerosols in the normalization region. The 1064 nm data is calibrated relative to the 532 nm channel, using backscatter signals from well characterized scattering targets to compute a scale factor. In CALIOP's automated processing system [Vaughan et al., 2004], this scale factor is determined from the ratio of signals at 1064 and 532 nm measured in strongly scattering cirrus clouds, with an assumption of known backscatter color ratio of these cirrus clouds. A constant cirrus color ratio of unity is used in the current data release. For each daytime and nighttime pass, a mean 1064 calibration constant is computed by averaging all individual retrievals accumulated throughout the duration of the orbit segment. This single value of the 1064 calibration constant is applied to all profiles within the segment. Calibrating the 1064 nm data in this fashion is estimated to have an uncertainty of ∼10%, in part due to uncertainties in the correct value of the cirrus cloud backscatter color ratio [Reagan et al., 2002; Hostetler et al., 2006]. Because the molecular scattering cross section at 1064 nm is ∼16 times smaller than that at 532 nm, 1064 nm signal levels are too weak, and the signal is too noisy, to calibrate using the same high-altitude normalization scheme that is employed for the 532 nm data calibration. Furthermore, the presence of aerosol may result in significant error in the 1064 nm calibration even though it is insignificant at 532 nm. For example, an aerosol with a backscatter ratio of 1.01 at 532 nm could have a backscatter ratio of 1.03–1.16 depending on the aerosol type (the aerosol backscatter color ratio can range from 0.2 to 1.0).

[28] Because the estimated calibration errors are relatively large at 1064 nm, additional analyses were undertaken to improve the value of the 1064 nm calibration constant used in this study. To do this, 1064 nm profiles suitable for use with the molecular normalization technique were obtained by averaging extensive regions of clear air. Similarly averaged 532 nm profiles were searched to identify those regions of exceptionally clear air located in the midtropopause (6–14 km) where the SNR at 1064 nm was adequately high. As a result of these calculations, the 1064 nm calibration constants used in this study were increased by a factor of ∼10% above the daily mean 1064 nm calibration constants reported in the CALIPSO data products. This adjustment is consistent with other recent findings. A just-concluded, 1 year ground-based study of cirrus clouds using a two wavelength lidar [Tao et al., 2008] measured a distribution of cirrus cloud backscatter color ratios peaked at about 0.88, with a full width at half maximum of 0.12. A similar distribution of cirrus cloud backscatter color ratios has also been observed by the airborne Cloud Physics Lidar (CPL) during the CALIPSO-CloudSat validation campaign, suggesting that the CALIOP 1064 nm calibration coefficient may have been underestimated in the current data release.

4. Results

4.1. The 18 and 19 August Cases

[29] The 18 August data lying between the two white lines shown in Figure 6 are analyzed using the second of the two-wavelength methods described in section 3.2. The data are initially averaged over 5 km (15 shots) and then further smoothed over 20 averaged profiles. Because the lower part of the dust layer may include additional aerosol contributions from the MBL, our analysis was limited to the altitude region between 1 km and 6 km (i.e., the free troposphere). The results are presented in Figure 7. The optical depth of the layer (1–6 km) studied ranges from 0.6 to 1.2, with a mean value of 0.87. The layer-averaged depolarization ratio is 0.31 ± 0.01 (0.004); the backscatter color ratio is 0.74 ± 0.07 (0.014); the optical depth ratio is 0.97 ± 0.02 (0.02); and the lidar ratios are 41 ± 3 (1) and 52 ± 5 (2) sr, at 532 nm and 1064 nm, respectively. The determination of uncertainties is described in the follow paragraph. The numbers given in parentheses are the standard deviations for each quantity computed along track (66 averaged profiles). They represent random variations due to noise and intrinsic variation of the corresponding dust optical properties.

Figure 6.

Attenuated backscatter at 532 nm acquired by CALIOP on 18 August 2006 very close to the dust source. The data segment between two white lines from (22.0°S, 19.2°W) to (18.6°S, 20.0°W) is analyzed.

Figure 7.

(a) Optical depth at 532 nm; (b) 1064 nm and 532 nm lidar ratio; and (c) optical depth ratio, color ratio, and depolarization ratio, retrieved for the 18 August case using the two-wavelength method. Error bars in Figures 7a and 7b represent uncertainties determined by adjusting the 532-nm and 1064-nm calibration coefficients by ±3% and ±10%, respectively. Error bars in Figure 7c are standard deviations computed from vertical bins within 1–6 km altitude range of each averaged profile.

[30] The error sources include (1) measurement noise, (2) uncertainties in the calibration constants, and (3) the range-invariant assumptions for lidar ratio and color ratio. Uncertainties due to the measurement noise can be estimated using the NSF technique [Liu et al., 2006]. However, a direct derivation of the uncertainty components for all the error sources is challenging because of the nonlinear process involved in the analysis. Uncertainties were therefore determined in a way that converts estimated errors into uncertainties in the normalization coefficient. Accurate retrievals using the two-wavelength technique require that both the lidar ratios and the backscatter color ratio of the layer be range invariant. Fortunately, because the size distributions are nearly vertically constant in the SAL [Maring et al., 2003b; Reid et al., 2003], it is expected that both lidar ratio and color ratio remain invariant in the SAL if it is unmixed with other type aerosols. The analyses in this paper also show evidence for the vertical homogeneity of dust intrinsic properties. Figure 8 presents an example of retrieved dust backscatters at 532 nm and 1064 nm (Figure 8a), depolarization ratio (Figure 8b), and 1064/532 backscatter ratio (Figure 8c). It is seen from this example that the dust depolarization ratio and the backscatter color ratio remain quite constant vertically, with standard deviations of ∼17% for the depolarization ratio and ∼10% for the color ratio. We note that these values include contributions from measurement noise, as indicated by the error bars in Figure 8 computed using the NSF technique [Liu et al., 2006]. The intrinsic variation of dust depolarization ratio and color ratio appears to be much smaller than these numbers. For 532 nm analyses of uncertainties, we estimated all errors using an uncertainty of 3% in the 532 nm calibration constant, C532. We then repeated the analyses, using values of 1.03 × C532 and 0.97 × C532 in the retrieval. The uncertainty for each parameter was then determined from the difference of the retrieved values when using 1.03 × C532 and 0.97 × C532. For the 1064 nm analyses, all the uncertainties have been estimated in the same way, using an equivalent error of ±10% in the 1064 nm calibration constant. All retrievals are summarized in Table 1. Table 1 also includes retrieved values for the 19 and 20 August cases, described in the following section, and for the HSRL measurements acquired on 28 August in section 4.3.

Figure 8.

An example of retrieved dust (a) backscatter coefficients at 532 nm and 1064 nm, (b) depolarization ratio, and (c) 1064/532 backscatter ratio at a location of (20.33°S, 19.61°W) from CALIOP measurement on 18 August 2006. The data were averaged for 15 single-shot profiles and further smoothed for 20. Vertical resolution is 30 m. Error bars in all the panels indicate random errors due to detection noise [Liu et al., 2006]. Standard deviations (Stddev) in Figures 8b and 8c were computed from vertical bins within 1–6 km altitude range, representing variations in the retrieved particulate depolarization ratio and backscatter ratio due to detection noise and natural variability of the dust layer. Both dust depolarization ratio (Figure 8b) and backscatter ratio (Figure 8c) show a good vertical homogeneity.

Table 1. Dust Layer Optical Properties Derived From the CALIOP and HSRL Measurements
DateCALIOPHSRL: 28 Aug
18 Aug19 Aug20 Aug
Locationnorthwest coast of AfricaAtlantic Ocean ∼1300 km from the coastAtlantic Ocean ∼2400 km from the coastGulf of Mexico ∼7600 km away from the source
Vertical extent1–6 km1.5–5 km2–5 km1.5–3.3 km
Optical depth at 532 nm0.6–1.20.3–0.450.29 ± 0.030.08–0.09
S53241 ± 3 sr41 ± 4 sr41 ± 6 sr45.8 ± 0.8 sr
S106452 ± 5 sr55 ± 5 sr54 ± 13 sr44 ± 8 sr
Backscatter color ratio0.74 ± 0.070.75 ± 0.080.72 ± 0.040.62 ± 0.01
Optical depth ratio0.97 ± 0.021.01 ± 0.050.93 ± 0.170.62 ± 0.13
Depolarization ratio0.31 ± 0.010.31 ± 0.010.32 ± 0.010.32 ± 0.01
Analysis methodsecond two wavelengthsecond two wavelengthopaque water cloudHSRL at 532 nm, first two-wavelength at 1064 nm

[31] In general, the retrieval at 532 nm is more accurate than at 1064 nm, because the backscatter intensity and SNR at 532 nm are both larger than the corresponding quantities at 1064 nm. The uncertainty for the depolarization ratio measurement is estimated conservatively to be ∼0.01, including errors due to the cross talk between two polarization channels. Both CALIOP and HSRL have a small cross talk (typically smaller than 1%). Retrievals of backscatter color ratio and lidar ratio are sensitive to the uncertainty in the normalization coefficient.

[32] The second two-wavelength method was also applied to the 19 August case, for which the dust layer has now been transported ∼1300 km out over the Atlantic Ocean. A data segment between 12.97°N, 32.09°W and 15.03°N, 31.64°W as indicated by the two white lines in Figure 9 has been selected. The results are presented in Figure 10. For the 19 August case, the intrinsic properties of the dust layer are essentially identical to the properties measured on 18 August. The layer-averaged depolarization ratio is 0.31 ± 0.01; the backscatter color ratio is 0.75 ± 0.08; the optical depth ratio is 1.01 ± 0.05; and the lidar ratios are 41 ± 4 and 55 ± 5 sr, at 532 nm and 1064 nm, respectively. Some change in extrinsic properties is evident, as the optical depth now ranges from 0.3 to 0.45.

Figure 9.

Attenuated backscatter at 532 nm acquired by CALIOP on 19 August 2006 at a site ∼1300 km away from the source.

Figure 10.

(a) Optical depth at 532 nm; (b) 1064 nm and 532 nm lidar ratio; and (c) optical depth ratio, color ratio, and depolarization ratio, retrieved for the 19 August case using the two-wavelength method. Error bars in Figures 10a and 10b represent uncertainties determined by adjusting the 532-nm and 1064-nm calibration coefficients by ±3% and ±10%, respectively. Error bars in Figure 10c are standard deviations computed from vertical bins within 2–5 km altitude range of each averaged profile.

4.2. The 20 August Case

[33] The optical-depth-constrained method is applied to the CALIOP 20 August measurement where opaque water clouds beneath the dust layer are available. Figure 11 presents attenuated backscatter at 532 nm. By this point in time, the dust layer has separated into two distinct regions. The upper, pure dust portion extends from 2 km to 4.5 km. The lower portion contains broken clouds, with tops at ∼2 km. Within the region between the two white lines shown in Figure 11, a search was conducted to identify those profiles in which the dust layer lies over a completely opaque cloud. A total of 90 single-shot profiles were found, and these were averaged together to improve the profile SNR. Figure 12a presents the averaged attenuated backscatter profiles at 532 nm and 1064 nm. Error bars in Figures 12a and 12b indicate random errors due to measurement noise computed using the NSF. The optical depths of 0.29 ± 0.03 and 0.27 ± 0.05 were computed using equations (4)(6), at 532 nm and 1064 nm, respectively for the dust layer. An equivalent uncertainty in the normalization coefficient is approximately 5% at 532 nm and 10% at 1064 nm. Uncertainties for all the retrieved parameters were estimated by changing the normalization coefficient by ±5% and ±10% at 532 nm and 1064 nm, respectively, and repeating the retrievals.

Figure 11.

Attenuated backscatter at 532 nm acquired by CALIOP on 20 August 2006 at a site ∼2500 km away from the source. Profiles with underlying opaque water clouds between two white lines, from (16.9°S, 42.0°W) to (14.2°S, 42.6°W), have been averaged and analyzed using the optical depth constrained method.

Figure 12.

(a) Attenuated backscatter profiles at 532 nm and 1064 nm averaged for 90 single-shot profiles where there are underlying water clouds and (b) retrieved backscatter profiles using the optical-depth-constrained method. Error bars represent random errors due to measurement noise computed using the NSF. Vertical resolution is 30 m.

[34] Error sources for these calculations include measurement noise, uncertainty in the normalization coefficient, and variations in the water cloud lidar ratio (19.1 sr are used for both 532 nm and 1064 nm). The uncertainty at 1064 nm is larger than 532 nm because of the lower SNR and the larger variation in lidar ratio for water clouds at 1064 nm. In addition, because the 532 nm depolarization ratio was used in the 1064 nm analysis, some additional error (<5%) may be introduced into the retrieved 1064 nm parameters.

[35] Particulate backscatter profiles at 532 nm and 1064 nm, retrieved iteratively using equation (3) with the optical depth as a constraint, are presented in Figure 10b. The rest of the parameters determined are listed in Table 1. Once again, the intrinsic properties of the layer show little change. For the 20 August measurements, the lidar ratio is 41 ± 6 sr at 532 nm and 54 ± 13 sr at 1064 nm; the color ratio is 0.72 ± 0.04; the optical depth ratio is 0.93 ± 0.17; and the depolarization ratio is 0.32 ± 0.01.

4.3. The 28 August Case

[36] Figures 13a and 13b show, respectively, the aerosol backscatter ratios and the particulate depolarization ratios (i.e., the ratio of perpendicular to parallel polarization components of aerosol scattering) measured over the Gulf of Mexico by the HSRL on 28 August. Similar to the CALIOP 28 August measurement (see Figure 3), the aerosol layer is bifurcated into an upper part (∼1.5 to 3 km) and a lower part (surface to ∼1.5 km). These two regions are clearly delineated by abrupt changes in depolarization ratio (Figure 13b). Unlike CALIOP, the 532 nm channel of the HSRL is equipped with an iodine absorption filter. The laser wavelength is tuned to the center of an iodine absorption line. The lidar return signal is transmitted through the iodine filter, where the aerosol and cloud scattering component are absorbed. Since the molecular scattering component has a wide spectral distribution due to its higher kinetic energy and fast molecular thermal motion, it is partially transmitted through the filter. Thus, the two scattering components in lidar return signals are separated spectrally. As a consequence, in addition to its ability to measure both backscatter and depolarization at 532 nm and 1064 nm, the HSRL can also make direct measurements of aerosol extinction and the lidar ratio at 532 nm and provide profiles for these parameters.

Figure 13.

HSRL data acquired on 28 August 2006 off the coast of Texas: (a) 532 nm particulate backscatter ratio; (b) 532 nm particulate depolarization ratio; (c) layer-averaged 532 nm lidar ratio and 532 nm particulate depolarization ratio for the aerosol layers from 0.2 to 1.2 km and from 1.5 to 3.3 km; and (d) 1064 nm lidar ratio, 1064/532 backscatter color ratio, and optical depth ratio retrieved for the dust layer between 1.5 and 3.3 km. Data in Figures 13a and 13b have a horizontal resolution of ∼1 km and a vertical resolution of 60 m. Data in Figures 13c and 13d have been further averaged for 20 profiles.

[37] Figure 13c shows the values measured directly by the HSRL for both the 532 nm lidar ratio and the 532 nm particulate depolarization ratio averaged over both the upper part of the layer (1.5–3.3 km) and the lower part (0.2–1.2 km). The first two-wavelength technique [Vaughan, 2004] described in section 3.2 has been applied to the HSRL data to derive 1064 nm lidar ratio, backscatter color ratio, and optical depth ratio for the dust (upper) portion of the layer. The result is presented in Figure 13d. Error bars indicate standard deviations computed from vertical samples in the corresponding altitude ranges of each averaged profile, with an exception for the lidar ratio at 1064 nm. Only one value is retrieved for the lidar ratio at 1064 nm for each averaged profile using the two-wavelength technique and hence a vertically computed standard deviation is not available for this parameter. Instead, we display in Figure 13d the standard deviation computed horizontally over the averaged profiles. These error bars represent the variations due to measurement noise and natural variability of the corresponding parameter. For the upper layer (1.5–3.3 km), the measured depolarization ratio, color ratio and lidar ratio at 532 nm appear to be relatively invariant along the track (relative variations are 1%, 1.9% and 1.8%, respectively). The depolarization ratio and color ratio also show some vertical homogeneity (relative variations are 5% and 10%, partly because of noise). The measured lidar ratio at 1064 nm and optical depth ratio show some variations both vertically and along the track.

[38] The mean depolarization ratio in the upper part of the layer is ∼0.32, which is typical for dust [e.g., Murayama et al., 2001; Liu et al., 2002; Barnaba and Gobbi, 2001; Gobbi et al., 2000]. The mean depolarization ratio in the lower part of the layer is ∼0.15, suggesting that here the layer is a mixture of dust and maritime aerosol and possibly some other aerosol type or types. This is consistent with the CALIPSO measurement, and in situ measurements in this area [Reid et al., 2002]. Maritime aerosols generally have a lower lidar ratio at 532 nm and hence would decrease the composite lidar ratio when mixed with dust particles. Humidity can change the optical properties of maritime aerosol. Theoretical computations showed that increasing relative humidity can decrease lidar ratio of maritime aerosols slightly at 527 nm and substantially at 1053 nm, and that for polluted environments, the lidar ratios at both wavelengths increase [Liu et al., 2000]. The measured lidar ratio at 532 nm in southern hemispheric marine conditions is 23 ± 5 sr [Franke et al., 2003]. Over the tropical Indian Ocean, the measured lidar ratio for the MBL increases to 49 ± 19 sr in the presence of soot-like particles [Franke et al., 2003]. Furthermore, maritime aerosols are generally spherical and therefore have lower depolarization ratios. The measured mean depolarization ratio of 0.15 is consistent with a mixture of dust and maritime aerosol. The 532 nm lidar ratios show a similar disparity, with the mean value of S532 in the upper portion of the layer hovering near 46 sr (a value that once again is consistent with lidar ratios measured for dusts [e.g., Liu et al., 2002; Tomasi et al., 2003; Mona et al., 2006; Muller et al., 2007]) and dropping to ∼33 sr in the lower portion of the layer.

5. Discussion

[39] On the basis of the optical properties derived from CALIOP and HSRL measurements (Table 1), the SAL in the free troposphere does not appear to have undergone a substantial transformation or size distribution change during the first 3 d of transport over the Atlantic Ocean. This indication of invariant size distribution is consistent with the in situ measurements of dust size distributions by Maring et al. [2003a]. Because the SAL is mostly transported in the free troposphere over the Atlantic Ocean [Prospero and Carlson, 1972], it is expected that the dust particles have remained largely unmixed with other aerosol types. We note that the SAL is normally quite dry, and thus humidity only slightly impacts the dust optical properties. After a 10 d transport period, some changes are evident in the color ratio, the optical depth ratio, and the lidar ratios measured by the HSRL over the Gulf of Mexico. Variations in these parameters suggest that the aerosol layer is undergoing some changes in its composition and/or size distributions. Changes in compositions [Sokolik and Toon, 1999] and size distributions [Ferrare et al., 2001; Dubovik et al., 2006] of the dust layer can alter dust optical properties. Because of the proximity of the layer to the highly polluted US eastern seaboard, it is possible that the dust may have become intermixed with polluted air arising from local sources. There is also possibility of mixing of the dust layer with small amount of smoke aerosol transported from African after entering the Caribbean region as suggested by the LTM results (Figure 1). However, the depolarization ratios measured in this area are essentially identical to the values measured over the Atlantic Ocean, which suggests that any possible contamination from a continental source or transported smoke has a rather small effect on the composition of the dust layer. Ship-borne in situ measurements show a surface wind flowing toward land for several days prior to 28 August, and these measurements are consistent with the NAAPS model results as shown in Figure 5. The significant change in the measured optical properties in the lower part of the dust layer over the Gulf of Mexico is most likely the consequence of mixing of dust particles with the MBL [Reid et al., 2002]. The most notable change is the gradual decrease in the SAL optical depth over the transport track, which can be attributed to a decrease in the number density of the dust particles.

[40] It is also useful to compare the results given in this work to those previously reported for other lidar measurements of dust. The depolarization ratios measured in this event are ∼0.32, and show almost no change over the course of the transport from the Sahara to the Gulf of Mexico. For desert dust, the particulate depolarization ratio is usually high because of the nonsphericity of dust particles, whereas the single-scattering depolarization ratio for spherical particles is zero. Depolarization ratios reported for dust typically range from 0.1 to 0.4 [Murayama et al., 2001; Liu et al., 2002; Gobbi et al., 2000; Shimizu et al., 2004; Sakai et al., 2002; Muller et al., 2003; Barnaba and Gobbi, 2001, Mattis et al., 2002].

[41] It is now well understood that the lidar ratio for dust aerosols is higher than the lidar ratio for spherical particles with equivalent size distributions. This difference is due to the nonsphericity of the dust particles, which introduces a significant reduction in the 180° scattering efficiency [Mishchenko et al., 1997; Liu et al., 2002; Mattis et al., 2002; Tomasi et al., 2003; Dubovik et al., 2006]. Very recently, Dubovik et al. [2006] further pointed out that on the basis of theoretical computations using dust models, the lidar ratio can also increase with increasing sizes of large spheroid particles. Dust lidar ratios measured at 355 nm and 532 nm can and do vary, depending on the source from which the dust is generated, the location of the observations, the pathway of transport, and the length of time that the dust particles have remained aloft. For example, dusts generated in different sources can have largely different mineral compositions [Sokolik and Toon, 1999; Chiapello et al., 1997]. Passing over highly industrialized area, the dust layer can become contaminated with polluted aerosols [e.g., Muller et al., 2007]. For these reasons, measured values of the dust lidar ratio vary widely, ranging from 30 sr to 80 sr [Mishchenko et al., 1997; Barnaba and Gobbi, 2001; Liu et al., 2002; Shimizu et al., 2004; Sakai et al., 2002; Muller et al., 2003, 2007; Mattis et al., 2002; Dubovik et al., 2006; Tomasi et al., 2003; Cattrall et al., 2005; Berthier et al., 2006; Mona et al., 2006]. The lidar ratio at 532 nm derived in this study is 41 ± 3 sr, 41 ± 4 sr, 41 ± 6 sr and 45.8 ± 0.8 sr, respectively, near the source, over the Atlantic Ocean, and in the Gulf of Mexico. The measured lidar ratios at 532 nm over the Atlantic Ocean are very close to the modeled value for CALIOP (40 sr) [Omar et al., 2006].

[42] We note too that the lidar ratios retrieved from the CALIOP data used in this study are not the true values, but instead are effective values, whose magnitudes are reduced to some degree by the presence of multiple scattering. However, the differences are considered to be small: the lidar ratios reported in this study should represent an underestimate by only a few percent because of the multiple scattering in the CALIOP measurement [Winker, 2003]. Pure and fresh dust plumes were measured using Raman lidar at Ouarzazate in South Morocco, a site rather close to the source, in the spring 2006 during the Saharan Mineral Dust Experiment (SAMUM). The measured lidar ratios range mostly between 50 and 60 sr at 355 nm and 532 nm [Muller et al., 2007]. The 532 nm lidar ratio was 40 sr for a dust storm measured in January 2005 in Beijing, about 70 km away from desert areas [Muller et al., 2007]. In the beginning and at the end of the dust storm, the measured lidar ratio varied between 30 and 35 sr.

[43] The lidar ratio at 1064 nm is 52 ± 5 sr near the source, 55 ± 5 sr and 54 ± 13 sr over the Atlantic Ocean, and 44 ± 8.3 sr at the Gulf of Mexico. The 1064 nm optical properties for the HSRL measurement were retrieved using the two-wavelength method described by Vaughan [2004]. Again, no significant change is seen over the ocean though the uncertainties in the retrieval are relatively large. In the Gulf of Mexico, the 1064 nm lidar ratio decreases by ∼15%. Unlike the values obtained in analyses of AERONET measurements [Cattrall et al., 2005], the lidar ratios derived in this work are larger at 1064 nm than 532 nm over the ocean. We note that the ground-based AERONET measurements do not directly measure the backscattered signal, and that subsequent computations of the lidar ratio are based on an assumed particle shape model (e.g., spheriods). Larger values of the effective lidar ratio at 1064 nm (S1064 = 35 ± 18.3 sr versus S532 = 26 ± 4.8 sr) were previously retrieved from the LITE measurements near the Sahara desert [Vaughan, 2004]. The lidar ratio of fresh Saharan dusts measured at the SAMUM site showed no wavelength dependence at 532 nm and 355 nm. However, the lidar ratios reported for aged Saharan dusts transported to Central Europe are higher at 355 nm (by 10–20%) than at 532 nm [Muller et al., 2007]. Similarly, the lidar ratios measured in this study show a wavelength dependence at 532 nm and 1064 nm for dust lying over the Atlantic Ocean. Our analyses of other Saharan dust cases measured by CALIOP over the North Africa and northern Atlantic Ocean using the transmittance constraint method [Young, 1995], the opaque water cloud method [Hu et al., 2007], and the two-wavelength methods [Liu et al., 2000; Vaughan et al., 2004] show that the lidar ratio at 1064 nm is larger than 532 nm for most cases. It is not definitively clear why the lidar ratios are larger at 1064 nm than 532 nm. It is possible that the difference can be attributed to differences in the absorption at the two wavelengths. For example, illite which is a major component in desert dusts [Sokolik and Toon, 1999] has stronger absorption at 1064 nm than 532 nm, which would tend to cause larger lidar ratios at 1064 nm than 532 nm. The lidar ratios measured in the Gulf of Mexico for the dust aged over 10 d in the free troposphere are effectively identical at 532 nm and 1064 nm, though the 1064 nm retrieval has larger uncertainties.

[44] Dust is a complicated mixture of various minerals whose optical constants vary widely from mineral to mineral [Sokolik and Toon, 1999]. The shape of dust particles is also complex, and has a wide distribution of aspect ratios [Okada et al., 1987, 2001]. All these features together make the theoretical study of dust optical properties complicated and challenging [Kalashnikova and Sokolik, 2002]. For the purposes of lidar ratio determination, there are currently no direct methods readily applicable to dust at the 1064 nm wavelength. Theoretical simulations of 1064 nm lidar ratios are saddled with unquantifiable errors arising from both dust shape and composition uncertainties.

[45] The retrieved backscatter color ratios range between 0.61 and 0.74, and appear to have decreased slightly (∼15%) over the course of transport. The retrieved optical depth ratio is 0.97 ± 0.02, 1.01 ± 0.05 and 0.93 ± 0.17 for three CALIOP measurements and 0.62 ± 0.13 for the HSRL measurement. The retrieved color ratio from the LITE dust measurement was 0.86 ± 0.01, which corresponds to an optical depth ratio of ∼1.1 determined using the measured lidar ratios [Vaughan, 2004]. We note that the large uncertainty in the optical depth ratio could be a consequence of the large uncertainty in the lidar ratio at 1064 nm. Since the size distributions of dust layers are typically dominated by coarse mode particles, the optical depth ratios are expected to have a low wavelength dependence.

6. Conclusions

[46] An extensive dust event that originated on 17 August 2006 has been observed and tracked by CALIOP. In the following days, the dust plume emerged from the west coast of Africa, advected across the Atlantic Ocean, and finally moved into the Gulf of Mexico. The transport from North Africa to the Caribbean region took approximately 6 d. Over the source, the dust particles attained a maximum altitude of ∼6.6 km. During the course of the transport between the source and the Caribbean Sea, the top of the dust plume dropped by ∼3.3 km. After entering the Caribbean region, the dust transport motion slowed, and the top of the dust layer showed no further significant change. The observed transport pathway is quite typical in summer [e.g., Carlson and Prospero, 1972; Karyampudi et al., 1999]. After 10 d journey, the remnant of the dust event was still observed by both CALIOP and the NASA Langley Research Center airborne HSRL off the coast of Texas on 28 August. Over the Atlantic Ocean, the dust plume showed some classic SAL features as described by Prospero and Carlson [1972], with the dust being transported mostly in the free troposphere. Throughout the transport track, depolarization measurements showed the presence of some amount of dust particles in the MBL. In the Caribbean and the Gulf of Mexico, dust particles have become well mixed into the MBL, as observed by previous in situ measurements [Reid et al., 2002]. This same dust event was observed by the MODIS instrument aboard the Aqua satellite. Both the dust event and the subsequent transport path were predicted by the NAAPS model.

[47] Optical properties derived from measurements made by CALIOP and the NASA Langley Research Center HSRL were used to investigate the evolution of dust particles throughout the transport event. The study uses extrinsic properties (optical depth) and intrinsic properties (layer-averaged depolarization ratios, color ratios, and lidar ratios) to track changes in the amount and composition of dust in the free troposphere. The properties near the source and over the Atlantic Ocean were observed by CALIOP, while the measurements at the Gulf of Mexico were made by the HSRL. The depolarization ratios observed near the source, over the Atlantic Ocean, and at the Gulf of Mexico range between 0.31 and 0.32, showing notable consistency in the nonsphericity of the dust particles. The measured dust intrinsic optical properties in the free troposphere over the Atlantic Ocean showed homogeneity both along the track and vertically in the layer, supporting previous findings by in situ measurements [Maring et al., 2003a, 2003b; Reid et al., 2003]. However, we do not know if this homogeneity persists equally over the other dust transport regions such as the northern Pacific Ocean and the Mediterranean Sea. The SAL optical depth decreases significantly with time, suggesting a decrease in the number density of dust particles as a consequence of the general descent of air, sedimentation, and wet removal of the dust particles, though these processes may have played different roles in different stages of the transport. Changes in the color ratio, lidar ratio and optical depth ratio at the Gulf of Mexico suggest that either the composition or size distribution or both have changed. The observed color ratios are 0.74 ± 0.07, 0.75 ± 0.08, 0.72 ± 0.04 and 0.62 ± 0.01, and the measured optical depth ratios are 0.97 ± 0.02, 1.01 ± 0.05, 0.93 ± 0.17 and 0.62 ± 0.13, at the source, over the Atlantic Ocean, and at the Gulf of Mexico, respectively. The observed lidar ratios at 532 nm and 1064 nm are, respectively, 41 ± 3 sr and 52 ± 5 sr near the source, 41 ± 4 sr and 55 ± 5 sr, 41 ± 6 sr and 54 ± 13 sr over the Atlantic Ocean, and 45.8 ± 0.8 sr and 44 ± 8.3 sr in the Gulf of Mexico.

[48] We report a significant departure from previously modeled estimates of 1064 nm lidar ratios. The 1064 nm lidar ratios determined in this study (52 ± 5 sr near the source, 55 ± 5 sr and 54 ± 13 sr over the Atlantic Ocean) are higher than the observed 532 nm lidar ratios and higher than the modeled 1064 nm values hitherto reported in the literature. That the lidar ratios increase with wavelength is not unprecedented. Such behavior has been observed previously for aerosols where large particles made up a significant part of the size distribution [Vaughan, 2004; Mattis et al., 2002]. The discrepancy between our measurements and the model predictions reflects the complexity in the shapes, size distribution, and composition of dust particles, which makes the accurate modeling of dust optical properties challenging. The discrepancy also emphasizes the importance of acquiring validation measurements for modeling studies. The discrepancy in the measured value and the CALIOP modeled value for the 1064 nm lidar ratio could induce an underestimate of dust aerosol extinction coefficient by as large as 40% in the CALIOP lidar data processing (though as of this writing the extinction product is still being validated and has not been currently released for public use). A value of 30 sr of the 1064 nm lidar ratio is modeled for the CALIOP dust aerosol used in the prelaunch algorithms [Liu et al., 2005; Omar et al., 2006]. Further theoretical and observational investigations are needed. The CALIOP aerosol models will be updated in the data processing for future releases following studies based on actual observations by CALIOP.

Acknowledgments

[49] This work was supported by the Radiation Sciences Program of NASA.

Ancillary