Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 We have inferred the most probable height distribution in a set of eleven areas of desert dust aerosol plumes over the eastern tropical and subtropical Atlantic Ocean using multispectral outgoing reflected radiance data collected during the Mediterranean Israeli Dust Experiment (MEIDEX), conducted on board the STS-107 space shuttle mission, from 16 January to 1 February 2003. It is shown that one can remotely infer the average height distribution of desert aerosol plumes from space in a specified atmospheric volume, if one has available calibrated, simultaneous, and co located radiances in the UV, the visible, and the NIR.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Desert dust is the most massive of the atmosphere's aerosols and it is found as aeolian deposits far away from its sources virtually all over the world, land surfaces and sea bottom. It is strongly linked to the climate system in several ways. It affects the Earth's radiation balance as well as the hydrological cycle and therefore the energy balance of the Earth-atmosphere system (EAS). It is the main supply of iron of the world ocean's fauna and flora and to the Amazon jungle. It strongly and negatively affects coral reefs when its concentration is high. The large dust plumes traversing the atmosphere from continent to continent transport spores, microbes and maybe even viruses. Finally, it impacts vision through the atmosphere and may clog air conditioners [Karyampudi et al., 1999; Kaufman et al., 2005].
 The source regions of desert aerosols are minute compared to the area of the world's deserts [Koren et al., 2003]. They are mostly dried out playas or deltas of wadis, which contain silt, brought down from neighboring mountains or imported by winds from remote sources. Moreover, the sources are only activated when both soil and meteorological conditions are just right. The main features of the desert aerosol are therefore its sporadic nature, spatial variability and the relatively large size of its particles, larger than 0.5 micron in diameter. The desert aerosol traverses the atmosphere over distances of thousands of kilometers in the form of giant plumes but each of the latter only exists for up to weeks at a time and may be composed of many subplumes at varying heights and possibly of different composition [Israelevich et al., 2002; Karyampudi et al., 1999].
 The regional and global radiative effects of the desert aerosol on the surface and the atmosphere vary from heating to cooling depending not only on the aerosol's optical properties but also on the albedo of the underlying surface [e.g., Twomey, 1977].
 Therefore, if one wants to ascertain the integrated effect of desert aerosol on both global and regional weather and climate, the variation with time of its three-dimensional distribution in the atmosphere must be studied from space [Kaufman et al., 2002]. This is currently done by a series of satellites, monitoring the Earth in spectral regions varying from the UV, through the visible and NIR to the TIR.
 The vertically integrated total aerosol spectral optical depth is the basic quantity inferred from most remote sensing experiments. However, the height distribution of the desert aerosol (hence to forth, DA) is also a needed parameter. First, the DA may absorb as well as scatter both incoming and outgoing solar radiation in the UV, visible and NIR. The height of the layer in which the DA is located is then important for several reasons.
 In the UV this is because of the strong Rayleigh scattering over a dark surface over both land and sea. If an absorbing aerosol, like desert dust, is embedded inside the Rayleigh atmosphere, the latter is essentially divided into two parts. The lower part under the desert aerosol may receive very little solar radiation and thus reflect little. The part of the Rayleigh atmosphere above the dust is then the main gaseous scattering agent. The effect of an absorbing aerosol on the outgoing TOA radiance is then strongly dependent not only on the total optical depth of the aerosol but also on its vertical distribution. This is not the case for purely conservative scatterers, e.g., nonabsorbing aerosol. These will just increase the amount of backscattered radiance [e.g., Herman et al., 1997; Torres et al., 2002]. The desert aerosol optical depths in the UV as inferred from TOA radiance data will therefore depend strongly on both the measured TOA radiances as well as on the distribution with height of the aerosol. The latter is usually not available at the time of the measurement and therefore climatological information has been mostly used up to recently to model the height dependence [Sinyuk et al., 2003; Torres et al., 2002].
 A second reason is that the atmosphere will be thermally stabilized under the aerosol and destabilized above. The DA is composed mostly of particles with diameters larger than 0.5 μm therefore they may exhibit greenhouse gas-like behavior due to the absorption in the bands of the silicates in the thermal IR main atmospheric “window” [e.g., Pierangelo et al., 2004]. Moreover, the DA layer itself is often neutrally stable, with an adiabatic lapse rate, due to the atmosphere's properties at its source. This is important by itself as it affects the thermohydrodynamics of the region [Joseph, 1984; Alpert et al., 1998] and also relevant as a constraint in inverting a remotely sensed TOA thermal IR radiance spectrum for the temperature profile.
 Last, the exact location of the aerosol layer together with the albedo of the underlying surface are important in determining whether it heats or cools the atmospheric column. The height interval in which the DA is located also plays a significant role in determination of its efficiency to instigate rain-producing clouds and in its propensity to be coated with sulfate aerosols. The latter would greatly enhance the probability and amount of rain from a given cloud [Levin et al., 2005; Teller and Levin, 2006].
 Therefore, the height distribution has always been the target of many investigations [e.g., Torres et al., 2002, and references therein], including MEIDEX. The preceding acronym means Mediterranean Israeli Dust Experiment, the experiment carried out from the shuttle flight STS-107. It has been found that the desert dust over the eastern tropical Atlantic is transported thousands of kilometers in the so-called SAL layer [e.g., Karyampudi et al., 1999]. The transport in summer takes place mainly above the Marine Boundary Layer and reaches altitudes of over 5 km. However, low-altitude plumes may be transported within the boundary layer from sources near to the ocean [Chiapello et al., 1995; Di Sarra et al., 2001]. In winter, however, the dust transport is done primarily by the low-altitude trade winds, then blowing out of Africa, and is usually below the 2 km level [Chiapello et al., 1995; Di Sarra et al., 2001; Torres et al., 2002; Alpert et al., 2004; Pierangelo et al., 2004].
 In the present study, we have available forecast dust plume location as well as its height distribution made by the MEIDEX Dust Forecast Model, specifically developed for the MEIDEX Project. The model has been extensively studied [Alpert et al., 2002, 2004; Kishcha et al., 2005] and supplies 3D distributions of desert aerosol (DA) in a grid of 0.5° resolution, covering the eastern Atlantic, North Africa and the Mediterranean Sea. The dust forecast model is based on the one-particle-size SKIRON system, and after modification, was put into operation and has been used for short-term dust predictions at Tel Aviv University from November 2000 until the end of 2005 [Kishcha et al., 2005; Alpert et al., 2002]. Several modifications were made to the model including development of a new dust initialization procedure, determination of the dust sources employing Ginoux et al.'s  method, and expansion of the forecast area to include the Atlantic Ocean. The model domain was 0–50_N, 50_W–50_E. The model had a horizontal resolution of 0.5° and 32 vertical levels (Table 1). Dust forecasts were initialized with the aid of the Total Ozone Mapping Spectrometer aerosol index (TOMS AI) measurements [Alpert et al., 2002]. The initial dust vertical distribution over each grid point, within the model domain, was determined according to the value of TOMS indices with four categories of model calculated averaged dust profiles over the Mediterranean and among four other profiles over North Africa.
Table 1. Center Wavelengths and Widths of the Filter Transmissions
 From the onset of the MEIDEX project it was clear to us, in view of what is currently available from several satellites, many special campaigns as well as from the AERONET and similar networks, that it is clearly not very useful and not too original to supply the scientific community with a small additional data set on desert aerosol optical depth over the Atlantic. On the other hand, the distribution with height of desert aerosol is a parameter which is important but not yet been monitored continuously until very recently by the A-train satellites. Large amounts of historical experimental satellite data are available that may be mined for the height distribution. We decided therefore to investigate the feasibility of constructing methods to remotely determine the height distribution on a continuous basis.
 We selected a combination of six wavelength bands for our camera that spanned 113 the UV, the visible and the NIR (see Table 1). In our study, carried out over the northeastern tropical Atlantic, the spectral variation of the complex refractive index of the aerosol in the spectral range from the UV to the NIR as well as its size distribution are taken as in the Shettle Model for Background Desert Aerosol which is part of the latest version of the so-called 6S Radiative Transfer Model [Vermote et al., 1997, and references therein; Torres et al., 2002; Sinyuk et al., 2003]. Computed optical depths are compared with colocated and simultaneous of MODIS measurements.
 This is a reasonable assumption since the desert aerosol used in the 6S in the region of the tropical and subtropical Atlantic has been extensively investigated and verified [e.g., Vermote et al., 1997; Reid et al., 2003].
 In summary, we chose as our research goal in this project to find the height distribution of the desert aerosol by comparing the simulated TOA radiances in the UV with the measured ones, knowing the optical depth and assuming the size distribution and the refractive index. The remotely inferred height profiles of the desert aerosol were compared with the ones forecast by the MEIDEX Dust Forecast model, available simultaneously and at the same location. This choice prescribed the need to construct an instrument that would observe the same atmospheric volume simultaneously in the UV, visible and NIR, a battery of instruments in an airplane for ground truth data as well as a model that would forecast aerosol spatial distribution.
 The absolutely calibrated spectral CCD camera developed for and used during the MEIDEX campaign on the space shuttle Columbia during STS-107 was unique among the various space instruments in orbit at the time, in that it covered simultaneously the UV as well as the visible and NIR spectral regions. The wavelength bands chosen were 340 and 380 nm in the UV, similar to one of the TOMS instruments, and 440, 550, 660 and 875 nm similar to those in MODIS. In section 2, the instrument and the amount, availability, quality, calibration and other properties of the data are described. In section 3, we describe the complete analysis process and apply it to the MEIDEX data. In this section, we apply our desert aerosol discrimination and inversion techniques to the measured TOA radiances and derive optical depths in the full spectral region of the MEIDEX instrument, UV to NIR.
 Finally, we derive and analyze the most probable height distribution of the desert aerosol in each selected case and compare with the independently forecast height profile by the MEIDEX Dust Forecast model [Alpert et al., 2004]. Section 4 is the summary and conclusions part of this paper.
2. Columbia STS-107 Mission and the MEIDEX Instrument
2.1. General Description
 The instrument was planned at Tel Aviv University (TAU) and constructed and integrated into the space shuttle systems in the U.S. at NASA GSFC by Orbital Systems Inc. and Omitron, Inc. It was positioned in the aft section of the cargo bay on the space shuttle Columbia and was part of the Mediterranean Israeli Dust Experiment (MEIDEX) conducted between 16 January and 1 February 2003. The shuttle was launched to an orbit with a 39° inclination and a height of 278 km, enabling us to cover tropical and subtropical latitudes.
2.2. MEIDEX Payload
 The main science instrument in MEIDEX was a Xybion radiometric camera Model IMC-201, equipped with a rotating filter wheel with six narrowband filters. The wavelengths of these filters were chosen so as to simulate those used by TOMS and MODIS satellite instruments for aerosol observations. The central wavelengths were 340 nm, 380 nm, 470 nm, 555 nm, 665 nm and 860 nm. The respective FWHM of these filters were 4 nm, 4 nm, 30 nm, 30 nm, 50 nm and 40 nm, respectively. The camera was equipped with a 50 mm Hamamatsu UV lens, adjusted with a special baffle to mitigate stray light from entering the optics. The FOV of the camera was rectangular, 10.76° vertical and 14.04° horizontal (diagonal 17.86°). The CCD had 486 over 704 pixels, where each pixel corresponds to 1.365·10–7 sr. This geometry led to a nominal ground resolution of about 70 m per pixel which increased to 200 m because of the point spread function of the instrument. The video format of the IMC-201 camera was NTSC which means that it produced its video output at a 30 Hz (33.3 ms/frame). The camera was operated in a “running mode,” where the filters are sequentially changed at the frame rate (33.3 ms/filter) so that that the complete six-filter sequence took 200 ms. The exposure time with each filter was automatically increased from 50 ns to 4 ms (in 50 ns increments) until the maximal light level in each frame reached a predetermined level. Boresighted with the Xybion was a second, low-light level color video camera with a wide FOV of ±60°. This camera served as a viewfinder for the crew and assisted in real-time observations. Both cameras were mounted on a single-axis gimbal which had a ±22° cross-track scanning ability. The data from both cameras was routed to digital recording devices within the payload and the crew cabin, and also downlinked in real time to the ground for quick analysis and storage. The entire payload was housed inside a 5 cubic feet canister, filled with dry nitrogen and sealed at the top with a 16” coated quartz window (see Figure 1). The canister was mounted on a cross-bay structure in the aft part of shuttle cargo bay as part of an array of other experiments (http://spaceflight.nasa.gov/shuttle/archives/sts-107/cargo/).
2.3. Camera Calibration
 The radiometric calibration of the Xybion IMC-201 camera is described in detail by Yair et al. . It was performed in the Laboratory for Atmospheres at Goddard Space Flight Center, MD. The basic procedure was to get an absolute radiometric calibration against a calibrated source. This was done by measuring the constant spectral radiance N [Watt/ster/cm2/nm] of an aperture of an integrating sphere with different exposure times, t, measured in milliseconds, for all the six filters that were mounted on the filter wheel. The product (N·t) has an almost linear correlation with the video signal of the aperture expressed in gray-level units [GL0]. The polynomial fit of third degree, N·t = f3(GL0), shows that such fit has a residual less than 1% over most of the dynamic range of the camera. This was shown for all six filters. By normalizing this third-order polynomial dependence for all filters one can show that the radiometric response of the camera is the same for all filters.
 Prolonged activation during the thermal qualification tests of the payload has shown that the CCD temperature changes with time. In addition, the orbital attitudes of the space shuttle changed continually throughout the mission, and put external thermal constraints on the payload. This necessitated a calibration of the temperature effect on the absolute calibration. The response of the camera was measured and enabled us to obtain a correction factor as a function of the temperature for the typical operation range, 20 to 40°C. The correction slope is small and varies between 0.053%/°C for filter 6 to 0.436%/°C for filter 1. The correction factors for the other four filters fall in between. In practice, no temperature corrections were necessary during a single measurement.
 Last, a flat field calibration was performed in order to derive the pixel-to-pixel nonuniformity correction. This was achieved by the use of an integrating sphere that has a rather large aperture with a constant spectral radiance N [Watt/ster/cm2/nm] all over its aperture. The images of this aperture are corrected first to a polynomial surface that is filter-dependent. This correction reduces the nonuniformity from a distribution with ∼15% FWHM to a distribution with ∼5% FWHM.
 The residual nonuniformity is caused by pixel-to-pixel nonuniformity that is constant for all filters. By removing this nonuniformity the distribution of pixel response to uniform radiance is reduced to distribution with ∼1% FWHM. These calibrations were completed at GSFC and were repeated once more when the payload was integrated to the shuttle prior to launch. This ensured that the Xybion IMC-201 could conduct measurement of radiance levels that can vary by 5 orders of magnitude with absolute accuracy of 1%. The noise level of individual images was found to be quite high, especially in the UV and of the order there of 30%. It was therefore decided to form data mosaics, using running averages of 36 images at a time. This reduced the noise level of the mosaics to 5%.
 The inferred optical depth then may have an error of 0.05 in the visible and NIR and 0.1 to 0.2 at 340 nm. Two regions of interest (ROIs) were defined for the MEIDEX experiment: the eastern tropical and subtropical Atlantic Ocean off the coast of Africa and the Mediterranean Sea, south of 39°N. The shuttle crew detected dust or biomass burning aerosol plumes, the instrument usually being pointed manually, on the basis of the daily MEIDEX numerical forecasts for the presence of dust supplemented by visual observation from the shuttle during the measurement. The cameras and measurement sequences were usually run by the astronauts, though limited and less efficient ground control was possible and used when the crew was unavailable.
2.4. MEIDEX Data
 Dust plumes were observed off the coast of West Africa on 27, 28 and 29 January 2003. Figure 2 and Table 2 show the sections of the orbits over which data was acquired by the astronauts within the Atlantic ROI. Average chlorophyll content for January 2003 in the relevant regions, based on NASA/SEAWIFS data (http://reason.gsfc.nasa.gov/OPS/Giovanni/mpcomp.ocean.shtml), was used in the calculation of the sea surface albedo in the 6S radiative transfer program. The total amount of dust data was 6.38 Gb from which eleven sequences of cloud-free spectral image strips suitable for analysis were chosen on 27 and 29 January (see Table 2). The data was prepared for analysis in the following stages. Copies were made of recordings of individual orbits from the original digital tapes to CDs. Calibration and temperature corrections were applied to these data and the final data were rerecorded. In order to facilitate analysis, low-resolution browse movies of the data were made in spectral B/W as well as in RGB formats.
Table 2. List of Chosen Image Strips in January 2003
Date in January 2003
Latitude of Strip Start
Longitude of Strip Start
Latitude of Strip End
Longitude of Strip End
Number of Pixels
smoke not used
smoke not used
smoke not used
 The small amount of data collected on desert dust during the MEIDEX mission made it feasible to use minimalist, mostly nonautomated, methods in all phases of the analysis. The successive stages in the analysis described in section 3 are (1) detection of aerosol plumes, using the RGB browse movies and then the spectral digital images; (2) choice of suitable sets of images containing significant amounts of dust by visual inspection (see Table 2); (3) detection and filtering of cloudy areas by visual inspection; (4) discrimination between aerosol types on the cloud-free images by use of spectral contrast; (5) determination of histogram of spectral TOA radiance of each band of the sample strips; (6) simulation of dependence of spectral radiance on optical depth at viewing conditions and geometry of each strip by use of the most recent 6S radiation code [Vermote et al., 1997; Kotchenova et al., 2006]; (7) determination of histogram of optical depth in each band of a strip by use of a TOA radiance versus optical depth diagram (the TOA radiance histogram was projected onto the axis of optical depth via the simulated curve of the dependence of TOA radiance on optical depth, calculated with 6S, using the known observation conditions); (8) creation of curve of spectral variation of optical depth; (9) validation of spectral optical depths in the NIR, visible and 380 nm by comparison with coincident Aqua and TOMS optical depths (the former were obtained from the NASA database (I. Koren, private communication, 2006), and the TOMS optical depths were calculated from the measured TOMS Aerosol Index using a statistical method [Ginoux and Torres, 2003]); and (10) estimation of the desert aerosol layer most probable height distribution by use of the projection of a two-dimensional histogram in the known optical depths and the measured TOA radiance at 340 nm onto a height axis.
3.2. Detection and Filtering of Cloudiness: Discrimination Cloud/Other
 One of the most serious problems in the analysis of satellite data of atmospheric aerosol is the correct identification and removal of clouds. There does not seem to be a really foolproof numerical method of removing clouds from an image available for analysis.
 In addition, it was shown in a recent study that the area that is considered cloud-free atmosphere in the vicinity of clouds (within a cloud field) has unique optical properties due to contributions of forming and evaporating cloud fragments and hydrated aerosols. The gradual transition from cloudy to dry atmosphere is proportional to the aerosol loading, suggesting an additional aerosol effect on the composition and radiation fluxes of the atmosphere as well as the great difficulty of defining cloud fraction [Koren et al., 2007].
 All methods that were tried in this study, whether based on spatial or on temporal properties of clouds and cloud fields, leave strips of clouds that are clearly identifiable by eye and which strongly affect the accuracy of the analysis. It was decided then to determine the cloud filter threshold by eye for each mosaic strip separately, since it was found that in this way the discrimination could be readily, reliably and repeatably determined. Three experienced observers independently picked out cloudy areas by texture using low-resolution browse movies and then full resolution spectral images. The findings of the observers were then intercompared and the final cloudy areas chosen by consensus. The fully cloudy areas thus found were then removed from the data strips and from the analysis. The application of this method for an exemplary strip is shown in Figures 3a and 3b, showing the unfiltered and filtered strips, accordingly.
3.3. Identification of Aerosols
 In the cloud-free areas, the same methods were used to identify the presence of aerosols. The identification was done on the basis of the texture that is visually different from that of clouds as well as that of the sea surface. The low-resolution browse color movies, made from the imagery, were again scanned by eye by three independently working and experienced observers to select aerosol in images, using texture and color as selection criteria to discriminate the possibly remaining clouds from aerosols. The selected images were then scrutinized by eye again, using full-resolution dusty images to double check the identification. Finally, aerosol-containing areas in images were selected for analysis by a consensus of the 3 independent experts. As stated already, eleven selections were performed and they represent a consensus as to the presence of aerosol and the nonpresence of clouds.
3.4. Discrimination Between Desert, Biomass Burning, and Marine Aerosols
 It was found by trial and error analysis of the theoretical spectral contrasts of the radiances between all six bands that, in our case, the ratio of the radiances in the 550 and 875 bands is the best, most sensitive, choice to distinguish between the possible types of aerosols. The distinction between desert aerosol (DA)and biomass burning aerosol (BBA) was therefore made by use of experimental and simulated contrast at the conditions of observation between band 4 (550 nm) and band 6 (875 nm). In Figure 4 we show eight examples of such discrimination. The relevant locations, dates, times, viewing angles as well as number of data points are summarized in Table 2. The marine aerosol was not distinguishable from the desert aerosol in this method. The full line represents the desert aerosol model, and the dotted line represents biomass burning aerosol. The band radiances, I, for DA and BBA are calculated using 6S, as a function of optical depth at the particular set of viewing conditions and time, and are plotted one against the other, namely, I(550) as a function of I(875). The clouds of dots show the distribution of the many thousand measured pixel radiances in each case. It is clear that in all selected cases except case C and case E, the aerosol is of type “desert aerosol.” Case C is ambiguous, case E represents a mixture of biomass burning aerosol and desert dust at low latitudes. The measured data always cluster along either the DA or the BBA curve, unless the optical depth is significantly below about 0.1, as in case C. Thus, the discrimination of DA from BBA, for AOT larger than 0.1, is immediate and obvious. Case C and case E were neglected in the further analysis. Figure 4 shows that our discrimination is reasonably exact and unique.
 A method to find the net contribution of dust to the total aerosol optical depth (AOD) by using the MODIS AOD and fine-fraction retrievals is described by Kaufman et al. . Our method is very different from that developed by Kaufman et al., even though in both cases one differentiates between different types of aerosols. The latter is a statistical method, suitable for application to large volumes of data, to find the net contribution of dust to the total aerosol optical depth (AOD) by using the MODIS AOD and fine-fraction retrievals is described by Kaufman et al. .
3.5. Determination of Optical Depths of the Selected Aerosol Imagery
 Because of the relatively small amount of data on aerosols obtained during the MEIDEX, the actual analysis was done case by case by using 6S simulation of TOA spectral radiances at the actual times and viewing conditions for each strip. The 6S model used was the latest one published, which includes the height distribution, is vectorized and is available on the Internet [Kotchenova et al., 2006]. For each selected spectral image strip containing dust (DA) or biomass burning aerosol (BBA), we went through the following procedure:
 1. We constructed a histogram of measured TOA radiances in the region of interest (ROI) at each of the six spectral bands, applied in each case at actual times and viewing conditions.
 2. We simulated the TOA radiance as a function of optical depth using the 6S for the same times and viewing conditions. Histograms of spectral optical depths were then constructed by projection of the TOA radiance histogram on the optical depth axis, using the theoretical 6S simulation curve of optical depth as function of TOA radiance as the transfer curve. The process is shown in Figure 5 for case A, 27 January 2003, 1424 UT. On the LHS of Figure 5, we show the histograms of the measured spectral radiances. In the middle, we show the method of projection of the radiance histograms onto the optical depth axis of the simulation. Finally on the RHS we show the histograms of the spectral optical depths. We studied the effect of ±5% bias and random errors in TOA radiances on inferred AOT and found the effect to be minor. The scales on the x axis of the optical depth curves vary from band to band in order to show the shape of the histograms.
 3. The next step was the construction of spectra of AOT from the NIR to the UV for the individual cases. These AOT spectra were validated in the NIR and visible and UV from 875 to 380 nm by direct comparison with colocated and almost contemporaneous Aqua MODIS and TOMS data. These spectra of optical depth are validated for the first time simultaneously in both the UV and the VISIBLE/NIR regions. In order to calculate validation optical depths in the UV at 380 nm from TOMS data, we used a statistical method developed at NASA, connecting the TOMS Aerosol Index to AERONET optical depths [Ginoux and Torres, 2003]. In this way continuous spectra of the aerosol optical depths from 380 to 860 nm were determined, spanning the UV, visible and NIR spectral regions. In Figure 6, we show the comparison of our results with those of Aqua MODIS and the TOMS at the same times and locations. In all cases, the comparison is good to very good.
 4. The inferred AOT at 550 nm is used to define the aerosol amount in 6S. This latter parameter and the complex refractive index at 340 as defined in 6S were used to calculate the optical depth at 340 nm.
3.6. Determination of Most Probable Height of DA
 The primary aim of the desert dust part of the MEIDEX was to study the dependence of the measured TOA radiances in the UV on the height distribution of the dust. The TOA radiance in the 340 nm band is a function of the aerosol's refractive index, total amount, size distribution and its distribution in the atmosphere. In order to find the latter, using the TOA radiance, one must therefore have independent information on the other properties of the aerosol. The amount and size distribution are determined using the measured spectral data in the visible and NIR. The refractive index at 340 nm is taken from 6S.
 The dependence of the radiance at 340 nm with a given optical depth on the heights is then calculated using the latest version of 6S. The atmosphere is divided into ten 500 m thick layers containing DA and air from the surface up and an eleventh layer at the top, containing only clear air. The implicit assumption here is that in winter, the desert aerosol is transported at relatively low altitudes in the boundary layer by the trade winds coming out of Africa. Consequently, the dust plume over the Atlantic is then usually not more than one km thick and is confined to low altitudes.
 The modeled UV optical depths at 340 nm were assumed to have a Gaussian distribution in height, with a width of 1 km, distributed in ten layers each of thickness of 500 m from zero to five km. The maximum of the distribution was put in turn in each of the ten layers. The aerosol of each optical depth in the plume at a particular set of viewing conditions, determined by the location and time of day, was then distributed according to each of the above 10 Gaussian profiles in turn and the resulting outgoing radiance was determined using the 6S. A 3D surface was then constructed of the heights versus the measured optical depth and radiance. Using the measured optical depths and outgoing radiances, a histogram of heights was constructed by a linear 2D interpolation, using MATLAB routine “grid data.” This defines the most probable height distribution of the desert aerosol layer.
 The process is summarized schematically in Figure 7 for one data point. We see the colored surface, calculated using 6S, defined by the outgoing radiance, the optical depth and the height of the maximum. The height of the maximum was taken as the dependent variable, making the surface stand up at a slant, somewhat like a sail. The measured data are shown as a 2D cloud of many thousands of black points in the plane of the optical depth and the outgoing radiance.
 In order to analyze the shape of the data cloud, two-dimensional histograms in the radiance and the optical depth were constructed. The following constraints were then put on the data. Optical depths below 0.1 and their radiances were filtered out to reduce the noise. At the high radiance side of the data cloud the peak of outliers extends somewhat beyond the boundaries of the model. There the shape of the peak is influenced by the wide range of the radiances due to their natural variability and to a 5% noise level in the data. Finally, another source of discrepancies may be differences between the model assumptions and the actual state of affairs.
 The two ends of the data “cigar” were truncated to a 5% noise level in order to check the effect on the final result for the height and the centroid of the data cloud was determined and found to have been unchanged. Therefore, all data in the cigar were used to determine the histogram of the height distribution.
 As an illustration of our method, the centroid of this cloud was determined, as shown in Figure 7, and serves to determine the most probable height in the entire given strip by projection on the height axis, using the 3D surface shown. In Figure 8 we show a comparison of histograms of inferred heights from the TOA radiance at 340 nm to those height profiles forecast by the MEIDEX Dust Forecast model for the whole strip. The full lines at a given height are the results of the MEIDEX analysis, the dashed curves are the forecasts of the MEIDEX Dust Forecast model at the same locations and times. The curves show the results of the analysis of six samples for band 1 at 340 nm, except for Figure 8 (case F), which is based on band 2 at 380 nm, because of high noise in the 340 nm band. The results for the average height profiles of the two models, radiative and thermohydrodynamical, are close. The peaks are within 500 m of one another, except for case G, and the maxima are at similar heights, except for in case A, where the forecast profile shows a second maximum between 2 and 3 km. Figure 9 shows the sensitivity of the dust profile to a 5% variation in the optical depth at 340 nm. The dotted lines around the height profile shows the variation induced when the optical depth at 340 nm is varied by ±5% as upper and lower boundaries. It is seen that the variation of the height profile under these circumstances is of the order of 200–300 m, which is about the resolution limit of our method.
4. Summary and Conclusions
 The MEIDEX was planned in an attempt to develop a space-based method of remote sensing to determine the height distribution of desert aerosol, using measured radiance data. The outgoing radiance from the atmosphere to space is strongly dependent on the height distribution only in the UV. The outgoing radiance in that spectral range depends on the optical properties of the aerosol, its size distribution, its total amount and on its distribution in height. Consequently, in order to derive the height distribution from the outgoing radiance in the UV, the refractive index there, the size distribution and the total amount of aerosol need to be known. The latter two parameters were derived using the optical depths derived from measurements of the outgoing radiance in the visible and NIR parts of the solar spectrum, using the recent 6S model for the desert dust. It was possible to verify the optical depths throughout the visible and the NIR parts of the solar spectrum by direct comparison with available colocated and almost contemporaneous MODIS Aqua and TOMS data. The refractive index in the UV was adopted again from 6S. Its values there are within the error bounds defined in the literature. The effective diameter of a pixel is 200 m. This number is due to the nominal resolution of the instrument (∼70 m), to the width of the point spread function as well as to use of a super pixel, a running average over thirty six spectral images. The latter was done in order to reduce experimental noise to ±5%. The measured radiances at the high end peak at values that are larger than that possible according to the 6S model. This is due to the variability of the measured radiances, to possible underestimates of cloudiness by our visual method as well as to differences between the experimental situation and our model assumptions. The derived most probable height histograms compared qualitatively well with the profiles forecast by the completely independent thermohydrodynamical MEIDEX Dust Forecast model. Even though the accuracies of both the most probable height profiles or heights of maxima, derived by the each of the two very different methods, are fairly low, the fact remains that they are quite similar.
 In summary we present a novel and simple way to estimate the height distribution of desert aerosol over the Atlantic Ocean in winter. If further validated, the present method may be used to find height distributions from combined UV and visible/NIR data available in historical databases. We plan to validate the method by the use of A-train data, throughout a season, using different height distributions, optical depths, locations and viewing conditions.
 This study is dedicated first to the memory of Yoram J. Kaufman from the NASA Goddard SFC for his valuable and continued support and advice throughout all aspects of this project. We would like to also dedicate this study to the memory of the seven astronauts of Flight STS-107 on the space shuttle Columbia. It would not have been possible without their professionalism, dedication, enthusiasm, and friendship. MEIDEX was a joint project of NASA and the Israeli Space Agency (ISA). We would like to acknowledge the support of the ISA throughout the preparations, the execution, and the analysis of the results of the MEIDEX campaign from 1998 to 2005. We would like to thank Scott Jantz and Ernie Hilsenrath of the NASA Goddard SFC for their steadfast and continued support and help in the calibration of the instrument. Zev Levin of TAU, Eli Ganor and Baruch Ziv of the Open University, and Major Meir Moallem of the IAF were also members of the nuclear MEIDEX team but were not involved in the development of this project. We would like to thank them for their valuable contributions and friendship. Finally, we are glad to acknowledge the efficient support and friendly comradeship of the FREESTAR team at NASA Goddard SFC, led by Tom Dixon of the STS-107 mission planners at NASA/Johnson SC and most of all by the astronaut team on STS-107. The work on this topic was greatly retarded because of the ill health of the lead author.