Physical and optical properties of mineral dust aerosol measured by aircraft during the GERBILS campaign

Authors


Abstract

This paper presents aircraft measurements of the physical and optical properties of mineral dust from the GERBILS campaign. The campaign involved ten flights of the UK Facility for Airborne Atmospheric Measurements (FAAM) BAe-146 aircraft over the western region of the Sahara desert. Vertical profiles showed dust layers at varying altitudes extending as high as 6.5 km. Dust layers were typically associated with a deep well-mixed boundary layer or a residual boundary layer above (the Saharan air layer). Aerosol optical depths (AODs), measured by integrating vertical profiles of extinction coefficient, ranged from 0.3 to 2.4 (at 0.55 μm). Aircraft AODs were generally within 20% of AERONET and Microtops sun-photometer measurements. Single-scattering albedos at 0.55 μm were measured in the range 0.92–0.99 with a campaign mean of 0.97. The in situ size distribution compared well with AERONET retrievals made at Banizoumbou (Niger) and Dakar (Senegal). The proportion of aerosol volume associated with particles of radii >1.5 μm was highly variable and also more difficult to measure. Models of dust as spheres, spheroids and more complex irregular-shaped particles were used to calculate single-scattering optical properties. The single-scattering albedo showed a low sensitivity to particle shape. The asymmetry parameter and specific extinction coefficient showed greater sensitivity to particle shape. Copyright © 2011 British Crown copyright, the Met Office. Published by John Wiley & Sons Ltd.

1. Introduction

The role of mineral dust aerosol in the climate system has received considerable interest over recent years, owing to its interaction with solar and terrestrial radiation (e.g. Sokolik et al., 2001). These interactions are governed by the microphysical, optical (Balkanski et al., 2007) and chemical (Chou et al., 2008) properties of mineral dust as well as its spatial and temporal distribution (Colarco et al., 2003). These properties vary greatly with the chemical composition and particle size distribution that depend on source region (e.g. McConnell et al., 2008) and various size-dependent emission and deposition processes. Therefore, estimates of mineral dust loading and radiative effects remain uncertain and scientific understanding of the governing processes is still considered to be low (Forster et al., 2007). Much of the uncertainty stems from the large range of particle sizes that contribute significantly to the interaction with radiation (e.g. 0.1–10 μm) and the difficultly in making consistent and reliable measurements across this size range, particularly from aircraft (Wendisch et al., 2004). Additional uncertainty arises from the mixing of dust with black carbon, organic carbon, and inorganic pollutants such as sulphates in some regions (e.g. Capes et al., 2008; Kandler et al., 2009). Thus there is a continuing need to catalogue dust properties from different regions and to test our current understanding of dust by validating existing models with new data.

Much recent research on mineral dust aerosol has been focussed on the North African region since the Sahara Desert is the largest global source of dust aerosol. Recent field campaigns focussing on the properties of Saharan dust include the Saharan Dust Experiment (SHADE: Tanré et al., 2003), the Puerto Rico Dust Experiment (PRIDE: Reid et al., 2003), the Dust Outflow and Deposition to the Ocean experiment (DODO: McConnell et al., 2008), the Dust and Biomass-burning Experiment (DABEX: Haywood et al., 2008), the Bodélé Experiment (BODEX: Washington et al., 2006), and the Saharan Mineral Dust Experiment (SAMUM: Heintzenberg, 2009). These experiments have demonstrated the success of using multiple measurement platforms (ground-based, aircraft, satellite) and interdisciplinary methods (observations and modelling on varying time and length scales) to further the understanding of mineral dust aerosol. The involvement of aircraft in such experiments enables in situ measurements over a large range of altitudes through the atmosphere and allows targeted measurements over remote land and ocean areas.

The key motivation for conducting the Geostationary Earth Radiation Budget Intercomparisons of Long-wave and Short-wave radiation (GERBILS) experiment was to investigate errors in the radiation budget of the western Sahara within the Met Office Global Numerical Weather Prediction model (Haywood et al., 2005). A positive bias of up to 50 W m−2 in outgoing long-wave radiation was identified within the Met Office model compared to satellite data. This was attributed mainly to the lack of inclusion of mineral dust aerosol in that model (Haywood et al., 2005). The main objective of the GERBILS campaign was therefore to characterize the physical, chemical and optical properties of mineral dust in the western Sahara and evaluate its effects on the short-wave and long-wave radiation budget. A further objective was to measure the characteristics of the desert surface such as temperature, emissivity and spectral albedo (Haywood et al., 2011; Osborne et al., 2011), as these are also key parameters in the short-wave and long-wave radiation budgets. Another objective was to investigate consistency between aircraft in situ measurements, aircraft remote sensing measurements, surface-based sun-photometers and satellite retrievals (Christopher et al., 2009). In addition, the observational data has been used to assess the performance of dust forecasts generated by the Met Office Crisis Area Model (Johnson et al., 2011).

The purpose of this paper is twofold. Firstly it describes the flight plans and measurements made by the Facility for Airborne Atmospheric Measurement (FAAM) aircraft during GERBILS (sections 2 and 3). Secondly it presents a summary of the in situ measurements (sections 4–8) including aerosol size distributions, vertical profiles of extinction and number concentration, and optical properties such as aerosol optical depth (AOD), single-scattering albedo (SSA) and Ångström exponent.

2. Flight plans and flight objectives

The GERBILS campaign comprised ten flights of the United Kingdom community Facility for Airborne Atmospheric Measurements (FAAM) BAe-146 research aircraft over North Africa and the Atlantic coast from 18 to 29 June 2007. As shown in Figure 1, the flights covered a broad area between Niamey (Niger) and the coastal regions of Senegal and Mauritania. Five of these flights took a standard route between Niamey and Nouakchott (Mauritania) involving a long segment along the line of 18°N. Two flights involved work over the ocean (B294 and B297) and one flight (B296) followed a more southerly route between Niamey and Nouakchott to encounter a large dust-storm that was over southern Mali and Senegal during 21 June. Table I lists each flight with the date, departure and arrival airports and a summary of the flight objectives. The main objective of the standard route between Niamey and Nouakchott was to cross the region where large discrepancies had been identified between the Met Office global model outgoing long-wave radiation and that measured by the GERB satellite (Haywood et al., 2005; Allan et al., 2009). Preliminary studies during the planning of GERBILS showed the region of model error to be centred on ∼18°N and 5–10°W, halfway along the standard route. The standard route between Niamey and Nouakchott took about 5 hours.

Figure 1.

Summary of the BAe-146 flight tracks during GERBILS. The x-axis is degrees of longitude (positive is east), y-axis is degrees of latitude (positive is north). The position of the Dakar and Banizoumbou AERONET sites are annotated.

Table I. Summary of the flights of the BAe-146 aircraft during GERBILS in June 2007.
FlightDateDeparture airportArrival airportFlight description
B29418 JuneAgadir (Morocco)NouakchottDust over ocean. Mainly radiation measurements including orbits.
B29519 JuneNouakchottNiameyStandard route. in situ sampling and radiation measurements, including orbits.
B29621 JuneNiameyNouakchottNon-standard route chasing a big dust storm over southern Mali and Senegal. Mainly in situ measurements.
B29722 JuneNouakchottDakarDust over ocean still following big dust storm. Radiation and in situ measurements including orbits.
B29822 JuneDakarNiameyTransit flight, late in the evening.
    Some in situ sampling during profiles.
B29924 JuneNiameyNouakchottStandard route. in situ sampling and high-altitude radiation measurements.
B30025 JuneNouakchottNiameyStandard route. Mainly radiation measurements.
B30127 JuneNiameyNouakchottStandard route. Long low-altitude run along 18°N for in situ sampling and surface characterization. Some localized dust events
B30228 JuneNouakchottNiameyStandard route. Long low-altitude run along 18°N for in situ sampling and surface characterization.
B30329 JuneNiameyMarrakech (Morocco)Transit back across Sahara. Mainly high-altitude radiation measurements.

Flights typically consisted of a deep profile ascent out of the departure airport, taking 20–25 min, followed by a series of straight and level runs (SLR) varying in length from 5–10 minutes to 2 hours. Runs at high altitudes above the dust were used to measure upwelling radiation, low-altitude runs were used to measure surface characteristics and the impact of dust on surface radiation, and mid-altitude runs (e.g. 2–5 km) were used for in situ sampling. Typically, profiles were also performed during the middle part of the flight to investigate the vertical distribution of aerosol in the target dusty region. A deep profile was then performed on descent into the arrival airport. Banked orbits were also performed at low altitude during flights B294, B295 and B297 allowing short-wave spectrometer measurements of sky radiances over a range of scattering angles (Osborne et al., 2011). Stacked SLRs (i.e. high- and low-altitude run over the same ground location, connected by a profile), were performed on flights B294, B297, B299 and B300 for the purpose of radiative closure studies (Haywood et al., 2011; Osborne et al., 2011). On other flights it was not possible to make such series of reciprocal runs due to the time/fuel constraints.

Dust was encountered on all GERBILS flights and on most (including B295, B296, B297, B299, B300, B301, B302), dust was observed visually, either below or at the altitude of the aircraft along at least half of the flight path. An overview of dust conditions during each flight is presented in the GERBILS overview article (Haywood et al., 2011). Based on the satellite data, dust forecasts and in situ evidence, the aircraft did not intercept any active emission events during the GERBILS flights but rather encountered widespread large-scale dust layers (e.g. 100 km or more in the horizontal) that may have been from disparate sources. The nephelometer (see section 3) provided the most straightforward means of identifying when the aircraft was within a dust layer. A low Ångström exponent, and a lack of chemical signatures such as CO, SO2 and NOx was generally taken as a confirmation that the aerosol layer was dominated by mineral dust.

3. Instrumentation

The instrumentation on board the BAe-146 aircraft has been described previously by Highwood et al. (2007) and Haywood et al. (2008). However, the aerosol in situ instruments are described in detail below with comments on errors involved when sampling mineral dust.

Aerosol size distributions were measured in situ with a Passive Cavity Aerosol Spectrometer Probe (PCASP-100X) for the accumulation mode (nominally 0.05–1.5 μm radius) and a Small Ice Detector (SID-2) for coarse mode particles (nominally 1–30 μm radius). These probes were mounted on a pylon beneath the wing of the aircraft. The PCASP instrument and errors involved in its application to dust have been described previously by Osborne et al. (2008). The PCASP data in this paper have been adjusted to account for the effect of refractive index on the scattering signal measured by the PCASP. The usual method for calibrating the PCASP is to use latex spheres of known sizes and predict the scattering signal using Mie theory. Here we have corrected the PCASP radii by repeating Mie calculations using a refractive index more representative of dust (assumed to be 1.53 + 0.0015i for all of the 15 size bins, based on Haywood et al. (2003b) and Balkanski etal. (2007), as compared to 1.58 + 0i for latex). The correction factors are generally small, between 1.0 and 1.1, i.e. the PCASP tends to under-size dust particles by a small amount if the standard calibration is assumed. The PCASP was serviceable only on flights B299–B303 and the smallest two bins of the PCASP suffered from excessive noise, so data from these bins was not included in the analysis. Hence the effective minimum detection size was close to 0.07 μm.

The SID-2 instrument was originally designed to distinguish between small spherical (liquid) and non-spherical (ice) cloud particles. Its predecessor SID-1 is described in Hirst et al. (2001) and SID-2 is described in Cotton etal. (2010). The SID-2 provided the concentration and equivalent-spherical radius of particles in the size range ∼1–30 μm. The standard method for interpreting the SID-2 data is to relate the scattered laser intensity to particle size using Mie theory for water spheres. The application of these data to dust has been investigated using the T-matrix code of Mishchenko (1991). We assume the dust to consist of a mixture of randomly-orientated oblate and prolate spheroids having an aspect ratio of 1.7. The phase functions for water spheres and dust spheroids were integrated over scattering angles 9–20°, i.e. the detection range of SID-2. The scattered light intensity over 9–20° was lower for equivalent-area spheroids relative to spheres by a factor of about 1.4 for particle radii up to 5 μm. This implies that SID-2 will under-size dust particle radii by a factor of about 1.2 due to the combined effects of particle shape and refractive index. The SID-2 bin radii were therefore multiplied by this correction factor; hence, after correction and ignoring the first two bins due to detection inefficiency, the bins effectively covered the size ranges 2.4–3.6, 3.6–4.8 μm and so on at intervals of 1.2 μm up to 30 μm.

The onboard TSI 3563 integrating nephelometer was used to estimate the scattering and backscattering coefficients at 0.45, 0.55 and 0.70 μm via a Rosemount inlet. In practice the nephelometer integrates scattered light over angles of 7–170° and an angular-truncation correction is needed to convert this to total scattering (i.e. 0–180°). We used the super-micron corrections provided by Anderson and Ogren (1998), and these gave correction factors of 1.31–1.37 for the 0.55 μm channel (values are similar across each of the three wavelengths) for the typical range of Ångström exponents we observed in dust-dominated aerosol layers (−0.2 to 0.2, see section 5.2). Mie calculations using the GERBILS campaign-mean size distribution from the PCASP and SID-2 gave an angular truncation correction factor of 1.40 for the 0.55 μm scattering coefficient. Further scattering calculations gave a similar correction factor for spheroids but a much lower correction factor of 1.16 for an irregular-shaped particle model (Osborne et al., 2011). This shows that the nephelometer scattering coefficient could be overestimated by 11–15% of the corrected value if aerosol scattering phase functions are assumed to follow the irregular-shape model. Quirantes et al. (2008) also shows that the correction associated with super-micron particles can be ambiguous when the Ångström exponent approaches zero. Using a new formulation for the correction they found 30% less truncation than the Anderson and Ogren (1998) method (this would imply our scattering coefficients could be overestimated by 8% of the corrected value). A further issue is the potential loss of super-micron particles in the Rosemount inlet and sample line. This would lead to an underestimation error (opposing the angular truncation error) although particle losses are not well known. An additional source of uncertainty is the role of water, although this would only have had a small influence on a few of the profiles in and out of Niamey. Elsewhere there was little evidence of anthropogenic or other emissions linked to secondary (i.e. hygroscopic) aerosol, and ambient relative humidity was generally low (<50%) or very low (<25%). We therefore neglected the role of hygroscopic growth in this analysis. Due to the combination of issues above, we attach an overall error of +/−20% to the corrected scattering coefficients. A comparison of nephelometer-derived AODs against sun-photometer AODs in section 8 shows that this level of uncertainty seems reasonable.

The Radiance Research Particle Soot Absorption Photometer (PSAP) was used to measure aerosol absorption coefficient at 0.567 μm and this was extrapolated to 0.55 μm using Mie calculations. The sample flow rate was maintained at ∼3 L min−1. The standard correction procedures of Bond et al. (1999) have been adopted here that includes unit-specific corrections to the aerosol spot size and sample flow rate. Data sampling was carried out using a 30 s running average and only data from straight and level flight was used, as recommended by other researchers (e.g. Petzold et al., 2009). The PSAP corrections limit the accuracy of the measurement of absorption coefficient to at least 20–30%. An addition source of uncertainty is the potential loss of large particles due to impaction in the Rosemount inlet, sample lines and PSAP instrument. Since the absorption efficiency also increases with particle size, such sampling losses may lead to significant underestimation of aerosol absorption. However, the overall losses from the sampling system are not well known. As an illustrative example, a sampling efficiency that reduced from 100% to 0% between radii of 1 and 3 μm would lead to an underestimation of aerosol absorption coefficient by a factor of 2 for typical dust size distributions measured in this campaign (assuming refractive index to be uniform with particle size). Whilst this sampling inefficiency error is somewhat speculative it would dominate other sources of error in PSAP and therefore we consider it to be a reasonable upper bound on the overall uncertainty in aerosol absorption measurement. Since the nephelometer AOD estimates in this campaign were in good agreement with Aerosol Robotic Network (AERONET) observations (section 8, Figure 10), without any correction for sampling inefficiency, we do not believe the nephelometer to have been strongly affected by losses in the Rosemount inlet. The speculated factor-of-2 uncertainty in PSAP therefore translates approximately to a factor-of-2 uncertainty in the single-scattering co-albedo (1 − SSA), as discussed in section 5.1.

4. Size distributions

Figure 2 shows size distribution data (normalized by the total concentration) from the 14 SLRs during flights B299–B303 where the PCASP and SID-2 probes were both functioning. Each line gives the average over an SLR at a particular altitude. These averages themselves hide much variation, particularly within runs B301 R3.1 and B302 R2.1, which were long (∼2 h) low-altitude runs along the latitude line of 18°N. The SID-2 was not sensitive to particles below about 2.4 μm radius, which left a gap in the size distribution between the upper detection limit of the PCASP (maximum corrected radius ∼1.6 μm) and the SID-2 (minimum corrected radius ∼2.4 μm). This gap unfortunately corresponded with the peak in the volume and extinction distributions. The gap can be bridged by fitting the available data with log-normals (see Table II and Figure 3). The use of log-normal modes rather than raw data is a prudent method for the interpretation of in situ size distribution measurements, as demonstrated by Haywood et al. (2003a) and Osborne and Haywood (2005), and approximating the data in this way should not introduce significant errors. However, the interpolation of the log-normals across the large gap between the PCASP and SID-2 data inevitably introduces a level of uncertainty that is difficult to quantify. Flight testing of the next generation of SID probes, SID-2h and SID-3, is now underway. Improvements to the photodetector arrays and electronics should make these instruments more sensitive to the smaller aerosol particles, thus giving some overlap with the PCASP. The SID-2 data become noisy in size bins above 10 μm radius due to poor sampling statistics, given the low number concentration of such particles and the limited sample volume of the instrument (∼80 cm3 s−1).

Figure 2.

PCASP (solid lines and crosses) and SID-2 (dashed lines and circles) size distributions from the straight and level runs during GERBILS. The x-axis shows the equivalent area spherical radius. The key shows the flight number, run number and the mean run height in metres.

Figure 3.

The GERBILS mean size distribution based on all run data together with four log-normal fits and their summation. Also included are the average size distributions for runs where nephelometer scattering at 0.55 μm exceeded 300 Mm−1 and was less than 75 Mm−1.

Table II. Summary of the log-normal fitting parameters to the mean GERBILS number size distribution as measured by PCASP and SID-2.
Moders.d.WnWextWv
  1. r = radius (μm), s.d. = standard deviation, Wn = fractional contribution to the total number, Wext = fractional contribution to extinction at 0.55 μm, Wv = fractional contribution to volume.

10.121.300.72500.04030.0084
20.321.680.22890.33250.1239
31.321.400.04580.59860.7086
42.701.850.00030.02860.1592

Figure 3 shows the campaign-mean aerosol number size distribution from the PCASP and SID-2 probes. As this is an average over all SLR data there is less noise at the large-particle end of the distribution. Some of the small-scale features in the PCASP and SID-2 distributions are probably due to uncertainties in bin widths rather than real features of the ambient dust distribution. The fitted curve, composed of four log-normal modes, provides a more realistic interpretation of the data that smoothes over noisy features in the raw data. The third mode of the fitted curve has the highest fractional contribution to the aerosol volume and extinction at 0.55 μm (see Wv and Wext values in Table II). The fitted curve does not capture the number concentration well at radii <0.1 μm, but scattering calculations with the raw size distribution data showed these particles contribute less than 1% to the extinction. Figure 3 includes two other averaged size distributions. These are averages over all available SLR data with nephelometer scattering coefficients of less than 75 Mm−1 or greater than 300 Mm−1. These show that the proportion of coarse particles is higher when the scattering coefficient is higher (see section 5.2 for discussion). Because these size distributions were averaged over shorter sections of data, the SID-2 had to be truncated at 20 μm.

The mean volume size distributions from AERONET retrievals at Banizoumbou and Dakar (see Figure 1 for their locations) are compared with the FAAM data and the fitted curve in Figure 4. The AERONET data from each site were averaged over all available retrievals during the period 18–29 June 2007. The aerosol volume curve fitted to the aircraft measurements gives a peak at slightly smaller sizes than the AERONET retrievals. The fitted curve also suggests a narrower peak than AERONET. These differences are due to a large drop in volume concentration between the largest bin of the PCASP (equation imagem) and the smallest bin of the SID-2 data (equation imagem). This may be an indication that SID-2 underestimated the concentration and/or size of the non-spherical dust particles, at least in the smaller size bins (size range 2.4–6 μm). However, the overall agreement is very encouraging given that we are comparing a retrieved column mean against an in situ mean with large differences in temporal and spatial coverage.

Figure 4.

Campaign-mean volume distribution from the BAe-146 aircraft (PCASP and SID-2, plus log-normal fit) and from AERONET retrievals made at Banizoumbou and Dakar.

Figure 5 compares the GERBILS log-normal fit with log-normal fits from other Saharan dust campaigns using aircraft measurements. These are plotted as volume distributions to emphasize the differing proportions of coarse particles. The DODO curves (McConnell et al., 2008) have been curtailed at 1.5 μm as there were no measurements above this size during that campaign. The lack of coarse-mode data for DODO and the normalization of the curves explain the apparent high proportion of fine particles. Coarse-mode measurements were also lacking in the SHADE campaign but Haywood etal. (2003b) used AERONET retrievals to judge the size and shape of the coarse mode at sizes above the PCASP range (i.e. r > 1.5 μm). In the DABEX campaign the coarse mode was measured by a PCASP-X, which extended the measured size range, although only to 5 μm (Osborne et al., 2008). During SAMUM, aircraft measurements from the German DLR Falcon of dust size distribution were made using a PCASP for the small–medium size range (0.05 < r < 1.5 μm) and two Forward Scattering Spectrometer Probes (FSSPs) for the medium–coarse size range (1.5 < r < 50 μm) (Weinzierl etal., 2009). The DABEX, GERBILS and SHADE log-normal fits all give similar distributions with the volume mode between 1.5 and 3 μm, but SAMUM shows a much larger volume mode at 10 μm. It is reassuring that the SHADE, DABEX and GERBILS give similar size distributions despite the difficulties encountered in measuring coarse dust on each of those campaigns. Perhaps this agreement is influenced to some degree by a reliance on AERONET retrievals, as these were used in each case, either as a way of verifying instrument performance or in the case of SHADE as filler for missing coarse-mode data.

Figure 5.

Comparison of log-normal fits to normalized mean volume size distributions from various campaigns listed in Table III.

The discrepancy between SAMUM and the other campaigns raises an important question: is this difference real or an indication of measurement errors? Some real differences are likely, given the difference in geographic region (Morocco versus south-western parts of the Sahara) and the spatial/temporal variability of dust. However, it is also probable that the big difference between SAMUM and the other campaigns points to fundamental difficulties in the sizing of coarse mineral dust. There are many sources of uncertainty in both optical particle sizing and sun-photometer retrievals of mineral dust aerosol (e.g. Reid et al., 2003). The scattering phase function of non-spherical dust particles is difficult to model or constrain observationally, making the interpretation of PCASP, SID or FSSP data uncertain. The efficiency of sampling inlets is also difficult to define accurately and one cannot rule out the possibility of unknown problems with flow distortion or instrument performance. Radiometric constraints on the size distribution are therefore important. The agreement between modelled and measured short-wave sky radiances and nadir terrestrial radiances during GERBILS (Haywood et al., 2011; Osborne et al., 2011) suggests that the observed size distribution is reasonable. Sensitivity tests showed that the SAMUM size distribution was less suitable for the interpretation of GERBILS aircraft radiation measurements (Osborne et al., 2011).

Differences in size distributions can be conveniently summarised by comparing integrated properties such as the mass specific extinctions (kext) predicted by Mie theory, whilst holding density and refractive index constant. Values of kext are shown in Table III, for all the campaigns compared in Figure 5. In the case of GERBILS and SAMUM, uncertainty estimates are also given based on information regarding the variability of size distributions. The absence of coarse-mode data (sizes above the PCASP range) during DODO leads to high kext values. For the other campaigns, where the size distributions do extend into the coarse range, there is a surprisingly large range of kext values; they differ by a factor of ∼6 between DABEX and SAMUM. Such differences are attributed to differences in the abundance of very large particles (e.g. r > 5 μm) that contribute greatly to mass but proportionately little to the short-wave extinction.

Table III. Summary of mean specific extinction (kext) values at a wavelength of 0.55 μm.
Campaignkext m2 g−1No. of modesDateReferenceComment
  1. Based on size distribution data from various aircraft campaigns of note that took place over northern Africa. All calculations have been calculated for a refractive index of 1.52 − 0.00147i and for a particle density of 2.65 g m−3.

GERBILS0.48 ± 0.104Jun 2007this paper 
DABEX0.865Jan 2006Osborne et al. (2008) 
DODO-10.654Feb 2006McConnell et al. (2008)No coarse mode
DODO-20.974Aug 2006McConnell et al. (2008)No coarse mode
SAMUM0.13 ± 0.084May 2006Weinzierl et al. (2009) 
SHADE0.385Sep 2000Haywood et al. (2003b)AERONET coarse mode

5. Measured aerosol optical properties

5.1. Single-scattering albedo (SSA)

The aerosol SSA was estimated at 0.55 μm using the 0.55 μm scattering coefficient from the nephelometer and the absorption coefficient of the PSAP. In sufficiently thick aerosol layers, i.e. scattering coefficient at 0.55 μm, σsca > 10 Mm−1, the SSA can be estimated by averaging over 5-minute sections of SLRs. This averaging period and scattering coefficient criteria ensured that uncertainties due to instrumental noise and/or sampling statistics were minimal. Figure 6(a) shows the histogram of SSA estimates from all suitable sections of data from GERBILS. The distribution is highly skewed towards the top end, with the majority of observations in the range 0.97–0.99, about a third in the range 0.95–0.97 and 10% in the range 0.92–0.95. The mean SSA was 0.97 with a nominal instrumental uncertainty of +/−0.02 (due to standard nephelometer and PSAP corrections, see section 3) and a standard deviation of 0.013. The mean value was not sensitive to the duration of the averaging period (3, 5 or 10-minute integration periods were tested) or the way the data was weighted (e.g. unweighted or weighted by extinction coefficient); all lead to a mean of 0.97 (to the nearest 0.005). The median value was also 0.97 to the nearest 0.005.

Figure 6.

Histograms of single-scattering albedo at 0.55 μm observed from (a) the BAe-146 aircraft, and retrieved from AERONET sun-photometers at (b) Banizoumbou and (c) Dakar.

If major sampling losses were to have occurred in the sampling inlet and PSAP system then the SSA values could be overestimated to a greater degree than suggested by the nominal instrument uncertainty. Although these losses are not well known it is conceivable that aerosol absorption coefficient from PSAP could have been underestimated by a factor of 2. This degree of underestimation was considered plausible following the analysis of similar airborne measurements of mineral dust during DABEX (Osborne et al., 2008). A factor-of-2 underestimation would reduce the mean and median values of SSA to 0.94, which would bring closer consistency with the AERONET retrievals from Dakar (Figure 6). However, the possibility of much more substantial absorption (e.g. SSA > 0.90) was not supported by radiative closure studies, i.e. the comparison of radiative transfer calculations with airborne radiometric observations during GERBILS (Haywood et al., 2011; Osborne et al., 2011).

The overall range of SSA values is similar to the range observed during past aircraft campaigns investigating Saharan dust aerosol with the same instrumental set-up. For example, Haywood et al. (2003b) report values of 0.95–0.99 from SHADE and an average value of 0.97. McConnell et al. (2008) report a range of 0.95–0.99 and averages of 0.98 and 0.99 for DODO-1 and DODO-2 respectively. SSA estimates for pure dust aerosol from DABEX were all consistently high with values in the range 0.98–0.99 (Osborne etal., 2008). A comparison with the ground data of Schladitz et al. (2009) during SAMUM in Morocco is also favourable. Here the authors used similar instrumentation again (i.e. TSI 3563 nephelometer and PSAP) and found the mean SSA within significant dust events to be 0.96 ± 0.02 at a wavelength of 0.537 μm. For dust measured in recent years in the Asian outflow, Anderson etal. (2003) found from aircraft measurements using a TSI nephelometer and PSAP that the dust SSA including the coarse mode was 0.96 + /−0.01 at 0.55 μm. Saharan dust transported across the Atlantic to Florida was measured with a ground-scanning radiometer and the column SSA was determined (at 0.55 μm) as 0.97 + /−0.02 (Cattrall et al., 2003). But with such long-range transport much of the coarse mode is presumed to have sedimented out. From long-term AERONET sun-photometer measurements shown in Dubovik etal. (2002) the column SSA of Arabian desert aerosol was found to be lower (0.93 + /−0.03) than that for Saharan dust advected over the Cape Verde islands (0.96 + /−0.01).

During GERBILS the aircraft covered a large geographic area and we might have expected to see some relationship between the observed SSA and the region of measurement. However, there was no obvious relationship between the observed SSA and the geographic location of the measurement, nor other factors such as the altitude of the measurement, the Ångström exponent, or aspects of the size distribution. It may be that mineral composition, and other aspects such as particle shape, confounded any relationship with size distribution. We conclude that the relative magnitude of short-wave absorption compared to scattering (expressed by 1 − SSA) is highly variable in Saharan mineral dust and difficult to summarise based on region or other observable properties.

Figure 6(b) and (c) show histograms of SSA retrievals (version 2, level 2.0) made by AERONET sites at Banizoumbou and Dakar during the GERBILS period. These have been interpolated to 0.55 μm from the 0.44 and 0.675 μm retrievals assuming a linear relationship with wavelength. These give lower mean values: 0.91 for Banizoumbou and 0.94 for Dakar, and show less-skewed distributions. The cause of this discrepancy is not clear but the small number of AERONET retrievals available during the GERBILS period (18–29 June 2007) may affect the results. However, extending the time period to the whole of June did not significantly change the results. Another possibility is that the aircraft values of SSA are erroneously high due to particle losses, as discussed above.

5.2. Ångström exponent and relationship to size distribution

The Ångström exponent was estimated from the nephelometer 0.45 and 0.70 μm scattering coefficients and was averaged over all 5-minute sections of flight data with aerosol scattering greater than 10 Mm−1 at 0.55 μm. Figure 7(a) shows the Ångström exponent as a function of the number-mean radius, estimated from the PCASP and SID-2 size distribution during flights B299–B303. Most Ångström exponents were in the range −0.2 to 0.2, which is consistent with the nephelometer measurements from DABEX (Osborne et al., 2008). Some higher values of Ångström exponent, up to 0.6, were recorded near Niamey and are possibly influenced by local conurbation pollution. Tesche etal. (2009) also indicate some high Ångström exponents (≥0.6) from lidar measurements during SAMUM and attribute the high values to urban influence. The Ångström exponent has a strong relationship with the number-mean radius, which reinforces confidence in the ability of these instruments to observe the natural variability of dust size distributions. This is encouraging given that the PCASP and SID-2 are mounted under the aircraft wing and the nephelometer is mounted inside the cabin. It demonstrates that the nephelometer is responsive to changes in the true size distribution. The relationship between the nephelometer Ångström exponent and the area-mean radius (i.e. effective radius) and volume-mean radius (not shown) was not clear due to the low sampling statistics of large particles in the upper bins of the SID-2.

Figure 7.

(a) Number-mean radius from the PCASP and SID-2 versus nephelometer Ångström exponent (0.45–0.7 μm) for the same 5-minute sections of data. (b) Nephelometer scattering (σsca) at 0.55 μm versus nephelometer Ångström exponent for 5-minute segments of all flight data within aerosol layers (σsca > 10 Mm−1) during flights B299–B303. The horizontal grey lines show the 75 Mm−1 and 300 Mm−1 boundaries in σsca.

Figure 7(b) shows the Ångström exponent as a function of the 0.55 μm scattering coefficient. The Ångström exponent also tended to decrease as the scattering coefficient increased. The observations with low extinction and high Ångström exponent may be associated with a background of fine dust and some anthropogenic or biogenic fine particles. The observations of high extinction and low Ångström exponent are probably related to strong or recent localized dust emission where one would expect a higher proportion of coarse particles in the atmosphere. These may drop out with time whilst the plume disperses and travels downwind. Figure 3 shows the mean size distribution associated with the ‘strong’ aerosol layers (σsca > 300 Mm−1) and the ‘weaker’ aerosol layers (σsca < 75 Mm−1). The relative drop in number concentration is greater at the larger sizes. This may reflect the drop-out of large particles as dust plumes age and disperse. It could also be that the ‘weaker’ plumes were simply more dilute, and fine non-dust aerosols from the background air mass have shaped the size distribution, particularly at sizes smaller than 0.2 μm. Surprisingly there was no obvious relationship between the altitude of the aerosol and the Ångström exponent or size distribution.

6. Derived aerosol optical properties

Here we present single-scattering optical properties of the dust aerosol based on theoretical calculations for a variety of shape assumptions. Firstly, Mie calculations were performed to estimate optical properties given the spherical shape assumption. Secondly, the spheroid package of Dubovik et al. (2006) was used to calculate the single-scattering properties and scattering phase functions for the full range of particle sizes detected by the PCASP and SID-2. This package uses look-up tables constructed from T-matrix calculations for size parameters up to 30–40 and a geometric optics method for larger size parameters. The spheroids were assumed to be randomly-orientated prolate and oblate particles having a non-equal distribution of aspect ratios between 0.3 and 3.0, in a similar manner to Dubovik et al. (2006). As a proof of numerical and theoretical consistency the T-matrix method was also tested with spheres and gave identical results to those from the Mie code. An ‘irregular’-shaped particle model was also employed, as described in detail in Osborne et al. (2011), providing a third independent method of estimating the optical properties.

In each model the refractive index was based on the Balkanski et al. (2007) dataset assuming a haematite mass content of 1.5%. This gave a refractive index of 1.520 − 0.00147i at 0.55 μm and provided the best agreement between derived SSA and the campaign-mean SSA from the nephelometer and PSAP measurements (see section 5.1). The Balkanski database uses computed refractive indices that are based on mineralogical data and compares well with AERONET data from dust-prone sites.

The single-scattering optical properties of asymmetry parameter, SSA and kext were averaged over the size distribution function given by the log-normal fit from section 4 (Table II). Results are given in Tables III and IV for the wavelength of 0.55 μm including a comparison with the measured SSA based on nephelometer and PSAP data. Further details of the phase functions and their application to radiative transfer can be found in Osborne etal. (2011).

Table IV. Summary of mineral dust optical properties at a wavelength of 0.55 μm derived from scattering computations of spheres, spheroids and ‘irregular’ shapes, together with aircraft data of SSA.
Particle typegSSAkext(m2 g−1)
spheres0.730.950.48
spheroids0.730.960.44
irregular0.620.970.30
measured0.97 ± 0.02

The shape assumption has a significant influence on single-scattering properties, especially those of kext and g. The irregular-shaped model is somewhat the outlier whereas the spherical and spheroid models give fairly similar results. The significantly lower value of asymmetry parameter for the irregular particles is due to increased side scattering from sharp edges on the particles (see Osborne et al. (2011) for plots of the scattering phase functions). All SSA estimates lie within the measurement uncertainty, although this simply shows that the refractive index was specified to a suitable value and cannot be used to infer which shape assumption is more realistic. The value of kext for spheres matches that shown in Table III. The other two shape assumptions lead to lower values of kext, showing that these are optically less efficient shapes, although the spheroid shape is only slight less so.

7. Vertical profiles

The vertical profile of aerosol mass has a strong influence on the interaction with long-wave radiation (Osborne et al., 2011) and in some situations a significant impact on the short-wave interaction too (Meloni et al., 2005; Johnson et al., 2008). The vertical extent of dust also affects its residence time and long-range transport, and therefore it is an important aspect to characterize. During GERBILS a total of 34 profiles were made from the surface (or near-surface) to 6 or 7 km to build a database of the vertical profile of dust and how this related to the thermodynamic structure of the atmosphere.

Figure 8 shows two examples of vertical structure that were observed, although there was strong variability from flight to flight and within flights. Figure 8(a)–(c) shows an early afternoon profile (∼1415 UTC) made on the approach to Nouakchott during flight B299. As the profile finished at the coast, the influence of a relatively cool, moist marine boundary layer exists near the surface with moderately low dust amounts. This is capped by a temperature inversion and hydrolapse between 0.3 and 1 km, and a peak in aerosol extinction and aerosol volume lies at the top of this inversion. Above that, a weakly stratified layer lies from 1 to 5.5 km containing several distinct aerosol layers at altitudes between 3 and 5 km. There is little variation of extinction coefficient with wavelength (i.e. Ångström exponent close to zero) indicating a dominance of coarse particles on the interaction with solar radiation.

Figure 8.

Vertical profiles from two flights: (a), (d) nephelometer-based extinction coefficient at three wavelengths; (b), (e) total aerosol volume from the PCASP and SID-2 and fine-mode aerosol volume (radii <0.5 μm) multiplied by 10; (c), (f) temperature and dew-point temperature.

The second example, in Figure 8(d)–(f), is a mid-afternoon (∼1515 UTC) descent into Niamey. This shows moderate levels of aerosol extinction layer from the surface to 1 km. This is presumed to be the local daytime boundary layer since the temperature profile is fairly close to adiabatic and aerosol extinction seems fairly constant with height. The ratio of fine aerosol volume to total aerosol volume is higher here than elsewhere, which could be an indication of anthropogenic aerosols from Niamey, although dust still dominates the scattering as there is still very little difference between the scattering coefficients at the different wavelengths. Above the local boundary layer is a weakly stratified zone extending from 1 to 6 km. Within this region is a thick layer of aerosol between 1 and 4 km and a weaker layer of aerosol between 4 and 6 km.

The temperature and humidity profiles in these two examples are fairly typical of the conditions observed during GERBILS and show a weakly stratified ‘Saharan air layer’, capped by a temperature inversion and hydrolapse. These show similar thermodynamic characteristics to profiles measured during JET-2000 (aircraft observations of the African Easterly Jet system) (Thorncroft et al., 2003). During GERBILS the capping inversion was typically at altitudes of 5–6 km but occasionally as high as 6.5 km. Dust was observed at all altitudes within this Saharan air layer but there was great variety in the altitude ranges and vertical depths of dust layers amongst the sample of 34 profiles that were measured. The examples in Figure 8 reflect this variability but are not to be viewed as typical cases. The only consistent pattern was a general decrease in aerosol concentration and scattering towards the top of the Saharan air layer. The dust concentration tended to decrease smoothly at the top of the highest layer rather than falling abruptly to zero.

In all profiles there was a strong correlation between the aerosol extinction and aerosol volume, including total volume and fine aerosol volume (radii of 0.05–0.5 μm from PCASP bins 1–8). This is exemplified by Figure 8 (a)–(b) and (d)–(e). However, the ratio of fine aerosol volume to total aerosol volume was generally lower in layers with high aerosol extinction (e.g. at 0.7 km in the B299 P9 case and at 2.7 km in the B302 P6 case). This is consistent with results from section 5.2 where high aerosol extinction was related to lower Ångström exponents and a larger mean radius (see section 5.2 for discussion of results).

8. Aerosol optical depth

The AOD at 0.55 μm was estimated by vertically integrating the aerosol extinction coefficient profile from the surface to the top of the aerosol layer. The AOD was largely determined by the nephelometer signal, since scattering accounted for 92–99% of the extinction (as shown by the SSA in Figure 6). The absorption coefficient during profiles was estimated using an appropriate SSA because the PSAP instrument is difficult to regulate during pressure changes. The appropriate SSA was taken either from adjacent SLRs, or the campaign-mean value of 0.97 if there were no adjacent runs. The AOD estimates from all 34 profiles are shown in Figure 9 both by a map showing the location of the profiles (Figure 9(a)) and as a scatter plot versus longitude (Figure 9(b)). This shows AODs ranging from 0.3 to 2.4 with no particular east–west trend. The average AOD from these estimates was 0.85. The FAAM AOD estimates have an assumed uncertainty of +/−20% as the overall error in extinction is dominated by the 20% uncertainty in nephelometer scattering coefficient (see section 3).

Figure 9.

Aerosol optical depth at 0.55 μm from FAAM aircraft profiles versus (a) geographic location and also (b) plotted against longitude.

AODs were also measured by a Microtops hand-held sun-photometer at Nouakchott and AERONET sun-photometers at Banizoumbou (13.5°N, 2.7°E) and Dakar (14.4°N, 17.0°W) (Holben et al., 1998). Figure 10 compares the FAAM AODs with sun-photometer measurements for profiles made within 200 km of a sun-photometer site. Despite this rather loose spatial constraint on the data, the FAAM AODs are in fairly good agreement with the sun-photometers. Most of the data points agree within the 20% error bars on the FAAM AODs estimates. The uncertainties in the sun-photometer AODs at 0.55 μm are almost negligible compared to the uncertainties of the FAAM estimates; typically around +/−0.01 for AERONET (Eck et al., 1999) and +/−0.02 for Microtops (based on a 3-year intercomparison with AERONET data: Ichoku et al., 2002).

Figure 10.

(a) FAAM AOD derived from nephelometer and PSAP profiles, and (b) column-mean Ångström exponent derived from nephelometer profiles versus sun-photometer data from AERONET (Banizoumbou and Dakar) and a Microtops hand-held sun-photometer operated at Nouakchott. The dashed line is the 1:1:line. The solid line in (a) is the linear regression best-fit line. The dotted lines in (b) are the 1:1:line +/−0.1.

The best-fit regression line shows no significant bias and the correlation coefficient is 0.885. Previous comparisons of FAAM aircraft derived AODs have been less successful. Results from DABEX showed a 20% overestimate of AOD (Osborne et al., 2008). Results from the Aerosol Direct Radiative Impact Experiment (ADRIEX), within industrial pollution, showed underestimation of AOD by up to a factor of 2 (Osborne et al., 2007), possibly due to evaporation of water and nitrates during the sampling process. It is unclear why there was better agreement during the GERBILS experiment. It could either indicate that the sampling efficiency of the Rosemount is better than previously thought (Haywood et al., 2003b) or that the sampling efficiency error in this case cancels against errors associated with the angular truncation correction (see section 3).

The wavelength dependence of AOD, expressed by the Ångström exponent from 0.44 to 0.675 μm, is shown in Figure 10(b) for the same selection of profiles. For most profiles the nephelometer Ångström exponents are within 0.1 of the sun-photometer values but there are three outliers, where nephelometer Ångström exponents are more than 0.1 lower than the sun-photometer values. These three outliers correspond to profiles made to the northwest of Niamey either on the approach or departure from Niamey. On these occasions the Ångström exponent was observed to be much higher in the Niamey region than in the dust layers intersected to the northwest and at higher altitudes. Therefore the bias may be due to a horizontal variability of aerosol properties. The root-mean-square (RMS) error including all points shown in Figure 10(b) was 0.10 and there was a low bias of 0.075 when comparing the nephelometer against the sun-photometers. The RMS error reduced to 0.06 and the bias reduced to 0.05 when excluding the three outliers. The result may indicate a slight bias in the ratios of the red, green and blue nephelometer channels but it is not possible to make any firm conclusions given the poor co-location of the profiles with the sun-photometer sites.

9. Conclusions

This article presents an overview of the aircraft measurements of mineral dust microphysical and optical properties from the GERBILS campaign. The aircraft covered a large geographical region over south-western parts of the Sahara Desert and encountered a variety of dust plumes. These dust events were driven both by synoptic-scale winds (e.g. flights B296 and B297) and mesoscale ‘cold pool’ downdraughts (∼100 km) generated by convective storms (Marsham et al., 2008). Here we report on the mean properties of dust and their variation across all dust events observed.

In situ measurements of aerosol size distributions are presented from the PCASP and SID-2 optical particle instruments. Corrections were applied to account for the effect of refractive index on particle sizing. The SID-2 instrument, measuring the coarse mode, was found to perform satisfactorily in most dust events, although it suffered from poor sampling statistics in weak dust events. The campaign-mean number size distribution was approximated by four log-normal modes and compared with AERONET retrievals. These showed reasonable agreement, although the aircraft data had a slightly smaller volume-mode radius and a slightly narrower peak (Figure 4). These features were associated with a large drop in volume concentration between the PCASP and SID-2 parts of the size spectrum (between 1.5 and 3 μm). This contrasts with airborne PCASP and FSSP measurements from SAMUM where the aerosol volume continued to increase beyond the PCASP size range and peaked at radii around 10 μm. This discrepancy may be due to differences in the performance of our SID-2 and the FSSP instrument used during SAMUM. The discrepancy may also be due to differences in dust source region, dust age and biases introduced by different flight-pattern strategies. Given the consistency of our results with AERONET, it is hard to believe that the SID-2 drastically under-sized coarse mode particles, although some under-counting seems probable.

The campaign-mean SSA of 0.97 + /−0.02 is consistent with previous aircraft campaigns measuring Saharan dust with nephelometer and PSAP instruments (Haywood et al., 2003b; McConnell et al., 2008; Osborne et al., 2008) and the ground-based SAMUM measurements of Schladitz et al. (2009). The nominal error of +/−0.02 does not include potential sampling losses as these have not been quantified. The agreement with Schladitz et al. (2009) is encouraging because ground sampling is less likely to suffer from large-particle losses that are a common difficulty with aircraft studies. The results from AERONET retrievals at Banizoumbou and Dakar show significantly lower SSAs (averaging 0.91 and 0.94, respectively, during GERBILS). The discrepancy between AERONET and in situ measurements could be due to sampling losses that could lead to an underestimation of absorption from PSAP. For example, a doubling of the absorption measurement from PSAP would have led to a mean SSA of 0.94 + /−0.03 that would have been in good agreement with the Dakar AERONET retrievals.

The Ångström exponent, based on the nephelometer scattering coefficients between 0.70 and 0.45 μm, showed a strong relationship with the PCASP and SID-2 number-mean particle size. The Ångström exponent also tended to decrease as the scattering coefficient increased. High scattering coefficients were also associated with a larger mean radius in the PCASP and SID-2 data. These correlations show the combined effects of dilution and gravitational settling as dust plumes disperse and lose large particles with time. They also show that both sets of instruments responded well to changes in ambient dust microphysics despite the fact that the nephelometer aerosol sample is prone to large-particle losses via the Rosemount inlet.

Theoretical calculations showed that the single-scattering optical properties of kext and asymmetry parameter depended significantly on the assumption of particle shape. SSA was less sensitive to particle shape. The asymmetry parameter was especially sensitive to particle shape when comparing spheres or spheroids with irregular-shaped particles. Therefore the use of Mie theory (i.e. assuming spheres) to predict single-scattering optical properties should not be considered an accurate approach for modelling the interaction of dust with solar radiation. Efforts should be made to account for such errors in applications where Mie theory is retained for practical reasons.

Aircraft profiles showed considerable variability in the vertical structure of the dust plumes. No consistent pattern was found in the vertical layering of aerosol (e.g. height of the peak aerosol loading), and there was no discernible trend in the vertical distribution with geographical location. Dust always extended to at least 5 km above sea level and occasionally as high as 6.5 km. Multiple dust layers were often observed during individual profiles, e.g. Figure 8(a), possibly indicating aerosol of different origin at different heights. These were often associated with small changes in potential temperature, indicating the presence of multiple residual boundary layers.

The aircraft profiles also provided a measure of the AOD derived by integrating the measured extinction coefficient over height. AODs ranged from 0.3 to 2.4 and there was no particular correlation with latitude or longitude. These aircraft-derived AODs correlated well with sun-photometer measurements, and the agreement was generally within 20%; a better level of agreement than found previously with the same aircraft instrumentation in either dust or pollution-related aerosols (Osborne etal., 2007, 2008). This could suggest that the loss of super-micron particles during the sampling process did not introduce errors greater than 20% during the GERBILS campaign. In the future it will be possible to assess particle losses using a cyclone impactor that has recently been added to the FAAM aircraft. This impactor removes particles above a certain size (known as a function of sample flow rate) and, as there are two identical nephelometers on the aircraft, we will be able to link one directly to the inlet and one via the cyclone impactor, thus allowing discrimination of size range admitted by the inlet.

Acknowledgements

The FAAM aircraft is jointly funded by the Met Office and the Natural Environment Research Council. We thank Didier Tanré for maintaining the Banizoumbou and Dakar AERONET sites. We thank Yves Balkanski for kindly supplying the Met Office with the dataset of dust complex refractive index used in this study.

Ancillary