Observations and modelling of the solar and terrestrial radiative effects of Saharan dust: a radiative closure case-study over oceans during the GERBILS campaign

Authors

  • J.M. Haywood,

    Corresponding author
    1. Observation Based Research, Met Office, Exeter, UK
    2. College of Engineering, Maths, and Physical Sciences, University of Exeter, UK
    • Met Office, OBR, FitzRoy Road, Exeter, EX1 3PB, UK.
    Search for more papers by this author
    • The contributions of these authors were written in the course of their employment at the Met Office, UK and are published with the permission of the Controller of HMSO and the Queen's Printer for Scotland.

  • B.T. Johnson,

    1. Observation Based Research, Met Office, Exeter, UK
    Search for more papers by this author
    • The contributions of these authors were written in the course of their employment at the Met Office, UK and are published with the permission of the Controller of HMSO and the Queen's Printer for Scotland.

  • S.R. Osborne,

    1. Observation Based Research, Met Office, Exeter, UK
    Search for more papers by this author
    • The contributions of these authors were written in the course of their employment at the Met Office, UK and are published with the permission of the Controller of HMSO and the Queen's Printer for Scotland.

  • J. Mulcahy,

    1. Global Model Evaluation, Met Office, Exeter, UK
    Search for more papers by this author
    • The contributions of these authors were written in the course of their employment at the Met Office, UK and are published with the permission of the Controller of HMSO and the Queen's Printer for Scotland.

  • M.E. Brooks,

    1. Global Model Evaluation, Met Office, Exeter, UK
    Search for more papers by this author
    • The contributions of these authors were written in the course of their employment at the Met Office, UK and are published with the permission of the Controller of HMSO and the Queen's Printer for Scotland.

  • M.A.J. Harrison,

    1. Atmospheric Dispersion, Met Office, Exeter, UK
    Search for more papers by this author
    • The contributions of these authors were written in the course of their employment at the Met Office, UK and are published with the permission of the Controller of HMSO and the Queen's Printer for Scotland.

  • S.F. Milton,

    1. Global Model Evaluation, Met Office, Exeter, UK
    Search for more papers by this author
    • The contributions of these authors were written in the course of their employment at the Met Office, UK and are published with the permission of the Controller of HMSO and the Queen's Printer for Scotland.

  • H.E. Brindley

    1. Space and Atmospheric Physics, Imperial College, London, UK
    Search for more papers by this author

Abstract

Mineral dust events exert a significant perturbation to the Earth's radiation balance via scattering and absorption in both solar and thermal infrared wavelengths. This study documents aircraft-based measurements of the solar and terrestrial radiative effects of a mineral dust outbreak off the west coast of Africa while the FAAM BAe-146 was in transit to the GERBILS observational measurement campaign. By comparing model and measurements of upwelling irradiance, an instantaneous top-of-atmosphere broadband solar direct radiative effect (DRE) of −33 ± 6 W m−2 is determined for an aerosol optical depth at 0.55 μm (τ0.55) of around 0.26 ± 0.04 with the variability in τ0.55 being estimated from the uncertainty in the aerosol models used in the radiative transfer calculations. Measurements of the spectral dependence of the solar radiative effect indicate that this is well modelled both above and below the dust layer whether using spherical, spheroid or irregular particle models. The terrestrial radiative impact at the top of the atmosphere is estimated to be +9 ± 3 W m−2 or around 25–30% of the solar radiative effect in this particular case, although the relative magnitude will of course be dependent on the underlying surface albedo, surface skin temperature, and details of the vertical profile of dust, temperature and humidity. The DRE retrieved from the Spinning Enhanced Visible and InfraRed Imager (SEVIRI) is in reasonable agreement with the aircraft measurements. Mineral dust aerosol optical depths derived from the Met Office global numerical weather prediction model show a reasonable spatial distribution but a general underestimation when compared against SEVIRI. Additionally, as expected from a relatively low-resolution global-scale model, the high τ0.55 values associated with mesoscale convective events are not well reproduced. Copyright © 2011 Royal Meteorological Society and British Crown Copyright, the Met Office

1. Introduction

Atmospheric aerosols are important components in climate, as they scatter and absorb short-wave and long-wave radiation, thereby exerting a significant direct effect. Aerosols also act as cloud condensation nuclei in liquid and ice clouds and can therefore influence the microphysical and optical properties of clouds and perturb cloud lifetime (e.g. Forster et al., 2007).

Atmospheric mineral dust particles are of sufficiently large size that they exert a significant direct radiative effect in both the short-wave and long-wave regions of the spectrum (e.g. Haywood et al., 2003, 2005; Highwood et al., 2003). The net radiative effect at the top of the atmosphere is the residual obtained from summing the (generally) negative short-wave radiative effect and the (generally) positive terrestrial long-wave radiative effect. However, in clear skies the short-wave radiative effect will be strongly dependent on the aerosol optical properties, the underlying surface albedo and the solar zenith angle (e.g. Boucher et al., 1998), and the long-wave radiative effect will be strongly dependent on the temperature structure and opacity of the atmosphere and the surface skin temperature (e.g. Haywood et al., 2005). Thus, the local and global net radiative effect remains difficult to quantify.

A number of measurement studies have been performed over land areas of the Sahara Desert. The Saharan Mineral Dust Experiment (SAMUM) based in southern Morocco performed a number of measurement and modelling studies of the chemical, physical and optical properties of Saharan dust and their associated impact on radiation (Heintzenberg, 2008; Otto et al., 2008). The radiative impact of mineral dust was also investigated during the African Monsoon Mulitidisciplinary Analysis (AMMA: Lebel et al., 2010), and in particular during the Dust and Biomass-burning Experiment (DABEX: Haywood et al., 2008). Haywood et al. (2008) measured a combined influence of both mineral dust and biomass burning on the surface radiation budget that reached 250 W m−2 when the combined mineral dust and biomass-burning optical depth at 0.55 μm reached around 1. Slingo et al. (2006) document a reduction in the surface radiation budget from mineral dust alone of up to 250 W m−2 when the mineral dust optical depth at 0.55 μm reached around 3–4. The surface radiative forcing efficiency (i.e. the change in surface flux per unit optical depth) is therefore 3–4 times greater in the presence of biomass-burning aerosol, showing the considerable influence of atmospheric absorption. Mineral dust itself is significantly absorbing, but questions remain as to the best representation of the absorption properties of Saharan dust. Forster et al. (2007, and references therein) present some considerable evidence that transported mineral dust is less absorbing than the model proposed by WMO (1986). One way to establish conclusively the absorption properties of mineral dust is to perform a so-called ‘radiative closure’ study. However, full radiative closure studies are rare and logistically very challenging owing to the necessity of measuring the aerosol physical and optical properties and the irradiances above and below the mineral dust concurrently. Over land, derivation of the top-of-the-atmosphere solar direct radiative impact is hampered by the lack of spectral contrast between the aerosol and the relatively bright surface (e.g. Osborne et al., 2008, 2011; Johnson et al., 2009).

The high spectral contrast between overlying bright dust and underlying dark ocean means that measurements over oceans are more amenable to radiative closure studies. Previous measurement campaigns of relevance over ocean include the SaHAran Dust Experiment (SHADE) off the coast of West Africa during 2000 (Tanré et al., 2003), where the solar and terrestrial direct radiative forcings were detected, quantified, and compared to satellite retrievals (Haywood et al., 2003a; Highwood et al., 2003). Measurements over ocean were also performed during the Dust Outflow and Deposition over the Ocean (DODO: McConnell et al., 2008), although significant dust events were few and far between over the ocean during DODO. The US National Aeronautics and Space Administration (NASA) AMMA (NAMMA) campaign (Hansell et al., 2010) investigated the surface terrestrial direct radiative impact of mineral dust aerosols over the Cape Verde Islands concluding that it offset a significant fraction (∼42%) of the reduction in the diurnally averaged short-wave flux.

The airborne activities associated with the Geostationary Earth Radiation Budget Intercomparison of Long-wave and Short-wave radiation (GERBILS) measurement campaign took place over the western Sahara and Sahel during June 2007 (Haywood et al., 2011). The UK Facility for Airborne Atmospheric Measurements (FAAM) BAe-146 aircraft flew a transit flight from Agadir, Morocco to Nouakchott, Mauritania on 18 June 2007 to make measurements of the radiative impact of Saharan dust. In order to maximise the utility of this transit flight, aircraft instrumentation was fully functional allowing remote sensing and in situ measurements to be made of the broadband and spectrally resolved radiative effects of mineral dust in both the solar (0.3–3.0 μm) and terrestrial (3–20 μm) spectral regions. In situ aerosol characterisation was also performed.

Section 2 describes the instrumentation onboard the BAe-146, section 3 the meteorological and radiative transfer models, and section 4 describes the flight patterns that were flown to attempt to obtain ‘radiative closure’. Section 5 presents the bulk of the radiative closure results and sections 6 and 7 provide comparisons of the Spinning Enhanced Visible and InfraRed Imager (SEVIRI) satellite retrievals and modelled dust fields both against each other and against the aircraft measurements. A conclusion is presented in section 8.

2. Instrumentation

In situ measurements available on the BAe-146 aircraft are very similar to those available during DABEX and have been discussed elsewhere (e.g. Haywood et al., 2008). Those directly relevant to the measurements, and corrections made, are briefly summarised here.

2.1. In situ sampling instrumentation

Aerosol scattering was determined at three wavelengths (0.45, 0.55, 0.70 μm) with a TSI 3563 nephelometer via a Rosemount inlet using the corrections of Anderson and Ogren (1998) to correct for instrument truncation and deficiencies in the illumination source. When the TSI instrument was situated on the C-130 aircraft during the SHADE measurement campaign in 2000 (Tanré et al., 2003), a significant correction had to be made to account for the loss of super-micron aerosol particles in the inlet/pipe-work (Haywood et al., 2003a). However, comparison of aerosol optical depths derived from the nephelometer when the TSI instrument was situated on the BAe-146 aircraft against Aerosol Robotic Network (AERONET) sun-photometers during DABEX (Johnson et al., 2008) suggests that the majority of super-micron dust particles are sampled. Thus, in this study, no correction for super-micron particle losses was made. Particulate absorption of radiation of wavelength 0.567 μm was measured with a Radiance Research Particle Soot Absorption Photometer (PSAP) using the corrections for instrumental artefacts documented by Bond et al. (1999). Aerosol chemistry was determined from isokinetic sampling onto filter substrates as described by Formenti et al. (2008) and Klaver et al. (pers. comm., 2011). Aerosol size distributions with radii between 0.05 and 1.5 μm were determined throughout the campaign with a Particle Measuring System Passive Cavity Aerosol Spectrometer Probe 100X (PCASP). Comparison of the sub-micron aerosol particles size distribution with those derived from AERONET radiance inversion techniques have been shown to be in reasonable agreement (Haywood et al., 2003b). The Small Ice Detector (SID: Hirst et al., 2001) that was originally designed for measuring ice particle size and shape of ice crystals has previously been used to detect and size large dust particles (Haywood et al., 2003a). The instrument used has been further developed into SID-2 (Cotton et al., 2010), which has a greater detector sensitivity. During this flight on 18 June, the PCASP was unserviceable owing to a mechanical failure of the motor powering the pump; we have to rely on the campaign-mean size distribution derived from the PCASP and SID-2 (Johnson and Osborne, 2011) once the motor had been replaced. While certainly not an ideal situation, we shall show that the campaign-mean size distribution and the associated optical properties are able to produce the observed broadband irradiances and spectral variability in the irradiances that are in remarkable agreement. While this agreement may be fortuitous, our assumption of using a campaign-mean aerosol size distribution does appear to compromise the results presented here.

Gas phase chemistry measurements of ozone and carbon dioxide were performed using a standard TECO 49 ultraviolet (UV) photometric instrument and a UV fluorescence Aero-Laser AL5002 (Gerbig et al., 1999) instrument respectively. Dropsondes measuring the temperature, humidity and wind fields were also deployed.

2.2. Radiation instruments

The radiation instruments include upward- and downward-looking clear-domed and red-domed broadband Eppley radiometers covering the spectral ranges 0.3–3.0 μm and 0.7–3.0 μm, instrumentation that has previously been used to detect the radiative effect of mineral dust off the coast of West Africa (e.g. Haywood et al., 2001, 2003a). The broadband radiometers (BBRs) are installed at a nominal 3° pitched forward angle to the airframe to account for the pitch of the aircraft when under standard operating conditions. The Short-wave Hemispherical Integrating Measurement System (SHIMS) uses two temperature-controlled Carl Zeiss spectrometer modules operating in the spectral range 0.30–0.95 μm and 0.95–1.70 μm. The pixel separation is approximately 0.0033 μm in the 0.30–0.95 μm module and 0.006 μm in the 0.95–1.70 μm module, giving approximate spectral resolutions of 0.010 μm and 0.018 μm with an in-house designed integrating head (see Osborne et al. (2011) for more information). The SHIMS instrument provides counts per millisecond and, during this campaign, the data was not fully calibrated so we use model radiative transfer data to calibrate the modules. The upper BBRs and the SHIMS instruments are subject to standard pitch and roll offset corrections that correct for both the attitude of the aircraft and for any slight misalignments of the instrument with the aircraft fuselage (see section 5.2). The Airborne Research Interferometer Evaluation System (ARIES) has previously been used to measure the spectrally resolved impact of mineral dust in the long-wave region of the spectrum at a spectral resolution of approximately 0.5 cm−1 (Highwood et al., 2003). A similar methodology is used here with zenith and nadir radiances converted to brightness temperatures.

3. Meteorological, dust and radiative transfer models

3.1. Met Office Numerical Atmospheric-dispersion Modelling Environment (NAME) model

NAME is a Lagrangian particle model (Ryall and Maryon, 1998) in which emissions are represented by parcels released into a model atmosphere driven by the meteorological fields from the Met Office's numerical weather prediction model. In this study the driving meteorology is from the Crisis Area Mesoscale (CAM) model (Greed et al., 2008). Each parcel carries mass of one or more atmospheric species (e.g. mineral dust); this mass can change due to various physical and chemical processes during the parcel life span. Although originally designed as an emergency-response nuclear accident model, subsequent development has greatly enhanced the capability of NAME so that it is now used in a wide range of applications. Amongst these capabilities is a back-attribution facility which enables the identification of locations that contribute to observed concentrations over a chosen time period. In this instance, a number of inert tracers were tracked backwards in time to the surface, which indicates the origin of the air mass observed during the flight.

3.2. Dust modelling in the Met Office global model

The production, transport and deposition of mineral dust is modelled using the scheme initially developed by Woodward (2001) for use in the Hadley Centre global climate model, but adapted for high-resolution mesoscale modelling and global numerical weather-prediction modelling as described by Greed et al. (2008) and Milton et al. (2008) respectively. Here we examine the performance of the global model (see Milton et al. (2008) for a more complete description). The horizontal resolution is 0.5625° longitude by 0.375° latitude which equates to a resolution of 63 km at the Equator, and fifty levels are used in the vertical. Further modifications and tuning have been made to the model as described in Johnson et al. (2011). Dust emissions are dependent upon clay, silt and sand fractions, the vegetative fraction, soil moisture, a threshold friction velocity and the surface-layer friction velocity. The threshold friction velocity is a function of the soil moisture. Dust is transported in six size bins spanning 0.06–60 μm in diameter. The radiative properties of dust are calculated for each of the size bins for both short-wave and long-wave regions of the spectrum using the refractive indices of Balkanski etal. (2007) using Mie scattering theory and assuming spherical particles. Optical depth diagnostics (aerosol optical depths, AODs) at several wavelengths have been developed for each bin size and for the sum of the six bins. This study adopts AOD as representing the aerosol optical depth at a wavelength of 0.55 μm.

4. Flight patterns

The flight pattern flown by the BAe-146 aircraft is shown in Figure 1.

Figure 1.

Track of the BAe-146 aircraft on flight B294 from Agadir, Morocco, to Nouakchott, Mauritania. The locations of the dropsondes launched from the aircraft and the intensive measurements are shown by the open circles.

Essentially the aircraft followed the coastline but was 50–100 km offset to the west which means that the impact of the bright land-mass upon the upwelling solar radiation is negligible in the hemispherically integrating BBR and SHIMS instrumentation. Four dropsondes were launched and the approximate positions of the dropsonde launches are shown on Figure 1. Also marked on Figure 1 is the intensive flight pattern area at approximately 20°N–21°N. A schematic diagram of this intensive flight pattern is presented in Figure 2.

Figure 2.

Schematic diagram showing the intensive measurement flight pattern performed by the FAAM BAe-146 during flight B294.

Figure 2 shows that the flight consisted of an approximate hexagonal pattern in the vertical with the following specific manoeuvres:

  • Run R3.1: A long approximate straight and level run (SLR) at a high altitude of FL250 (25 000 ft or 7.5 km) covering latitudes 29°N–21°N over the period 1503 UTC to 1645 UTC. Three dropsondes were launched on this run as indicated in Figure 2. Visual observations suggest that there was significant broken stratocumulus north of 26°N, but south of 26°N no cloud was apparent below the aircraft.

  • Profile P3: A profile descent from FL250 at ∼1000 ft/min (300 m/min) above 5000 ft and 500 ft/min from 5000 ft to 100 ft. The profile was broken midway through the descent and a reciprocal turn at constant altitude performed before the descent to 100 ft.

  • Run R4.1: A straight and level run of 6 minutes duration at 100 ft above sea level (ASL).

  • Profile P4: A profile ascent at 500 ft/min from 100 ft to 5000 ft then ∼1000 ft/min (300 m/min) above 5000 ft. The profile was broken midway through the ascent and a reciprocal turn at constant altitude performed before continuing the profile ascent to FL200.

  • Run R5.1: A straight and level run of 33 minutes duration at FL200 (20 000 ft or 6 km).

5. Results

5.1. Analysis of the campaign-mean aerosol size distribution and the associated optical properties

The derivation of the size distribution and associated optical properties are discussed in detail in Johnson and Osborne (2011), and the methodology for deriving these properties is not discussed in detail here. The mean mineral dust size distribution is represented by a series of four log-normal size distributions with mode radii and geometric standard deviations of 0.12, 0.32, 1.32 and 2.70 μm and 1.30, 1.68, 1.40 and 1.85 respectively. The refractive indices are assumed to be identical to those of Balkanski et al. (2007) assuming a haematite mass content of 1.5% yielding a refractive index of 1.520–0.00147i at 0.55 μm. Three different scattering solutions are derived: Mie scattering theory for spherical particles, non-spherical particles represented by prolate and oblate spheroids (Dubovik etal., 2006), and irregularly shaped non-spherical particles (Kokhanovsky, 2003).

The spheroid package of Dubovik et al. (2006) was used to calculate the single-scattering properties and scattering phase functions for the size distribution using 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. The distribution of aspect ratios is specified at 25 logarithmically spaced points and is a minimum at an aspect ratio of 1 (Dubovik et al., 2006). This model was found to lead to the closest agreement with measured spectral radiances derived from aircraft orbits (Osborne et al., 2011). This model will henceforth be referred to as the spheroid model.

The irregular particles were assumed to consist of hexagonal prisms of aspect ratio unity (the aspect ratio is defined as the column length-to-diameter) between 0.12 and 1.0 μm radius and polyhedral particles between 1.2 and 20 μm radius, with the aspect ratio of the polyhedral particles remaining invariant with respect to size. The polyhedral model is based on Macke et al. (1996) which represents particle irregularity by randomising a second-generation triadic Koch fractal, commonly referred to as the ‘polycrystal’. Although, the polycrystal model was originally applied to study the scattering properties of cirrus, it has been applied to study the scattering properties of large mineral dust aerosols by Kokhanovsky (2003) who showed that the polycrystal could replicate to high accuracy laboratory-derived scattering phase matrices of mineral dust. Because the derivation of the full scattering solutions from these irregular particles is computationally expensive, the calculations are performed at eight wavelengths and interpolated between these wavelengths to provide spectral data across the solar spectrum. This model will henceforth be referred to as the irregular model.

At a wavelength of 0.55 μm, the specific extinction (ke) is 0.48, 0.44 and 0.30 m2 g−1, the asymmetry factor (g) is 0.73, 0.73 and 0.62, and single scattering albedo (ωo) is 0.95, 0.96 and 0.97 for spheres, spheroids and irregular particles (Johnson and Osborne, 2011).

5.2. Analysis of the vertical profiles of temperature, humidity, ozone and aerosol from dropsondes and aircraft measurements

The temperature and humidity structure from the four dropsondes are shown by the tephigrams in Figure 3.

Figure 3.

Dropsonde profiles of temperature (solid lines) and humidity (dashed lines) on standard tephigrams. The latitudes and longitudes of the sondes (launch positions) are (a) 26.6°N, 14.9°W, (b) 23.4°N, 16.7°W, (c) 20.9°N, 17.8°W, (d) 19.3°N, 17.5°W.

Saharan air layers (SAL) are characterised by isothermal or slightly stable layers with vertically well-mixed specific humidity structures (e.g. Thorncroft et al., 2003). The northernmost dropsonde (Figure 3(a)) does not indicate any evidence of an SAL, while the southernmost dropsonde (Figure 3(d)) shows a well-established SAL with specific humidities of 2–3 g kg−1. The intermediate profiles shown in Figure 3(b) and (c) show progressively more influence of the SAL. Figure 3(b)–(d) shows a very strong inversion of up to 10 K (Figure 3(d)) capping the shallow marine boundary layer. Extrapolation of the temperature profile within the SAL to the surface provides an estimate of the surface temperature from the Sahara of around 318 K while the temperature near to the sea surface is 293–295 K. Figure 3(d) shows strikingly similar features to those shown during the SHADE which was based in the Cape Verde Islands during September 2000 (Haywood et al., 2003).

Temperature and humidity profiles from the aircraft measurements from profiles P3 and P4 (Figure 2) are very similar to those in Figure 3(d) and are not therefore shown. However, it is worth assessing the vertical profiles of aerosols and trace gases from the aircraft instrumentation. Figure 4(a) shows the profile of aerosol scattering measured by the nephelometer and Figure 4(b) shows corresponding measurements of ozone and carbon monoxide.

Figure 4.

Profile measurements showing (a) the nephelometer green, red and blue scattering as a function of the pressure, and (b) the measured ozone (thick line), climatological ozone (thin line) and carbon monoxide (dots). Measurements are from profile P3. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Mineral dust is present in a thick layer between 500 hPa and 800 hPa with a maximum dust scattering at around 750 hPa. A second rather small dust layer is evident at around 900 hPa. Throughout the main dust layer, the blue, red and green scatterings are very similar, which is typical for large-particle mineral dust with a mean Ångström exponent calculated between 0.45 μm and 0.70 μm, Å0.45–0.70, and 400–800 hPa of −0.24. Note that the increase in scattering in a thin layer near the surface (1010–980 hPa) is likely to be due to the presence of sub-micron marine aerosol because Å0.45–0.70 is significantly different, at 1.14.

The anti-correlation between the dust layers and tropospheric ozone that has been reported in previous studies (e.g. Bonasoni et al., 2004; Marsham et al., 2008) is evident in Figure 4(b). Given that the carbon monoxide concentrations show very little structure, this feature suggests that dust/ozone heterogeneous chemistry may result in the destruction of ozone. To assess whether this feature could be of climatic significance, we perform radiative transfer calculations assessing the radiative impact of the depletion of tropospheric ozone by mineral dust. We assume that the vertical profile of ozone is perturbed from that of McClatchey et al. (1972). The mass mixing ratio of ozone from McClatchey et al. (1972) is also shown on Figure 4 and indicates that the ozone profiles approximately match those of McClatchey et al. (1972) when there is little dust present (i.e. above ∼500 hPa and 920 hPa–950 hPa). We assess only the long-wave radiative impact as this dominates over the short-wave impact (e.g. Haywood et al., 1998). Calculations are performed for a sea-surface temperature of 292.7 K (see section 5.5), and an additional calculation is performed for a surface skin temperature of 330 K which would correspond to an approximate maximum over desert regions (e.g. Haywood et al., 2005). The vertical profiles of temperature and humidity are from the measurements from dropsonde no. 3, and other trace gases are included as in section 5.5. The results suggest that the instantaneous radiative forcing at the top of the atmosphere is approximately −0.23 W m−2 at a surface skin temperature of 292.7 K, and −0.38 W m−2 for 330 K. Thus, the climatic impact of mineral dust via heterogeneous removal of tropospheric ozone appears relatively minor, although a more comprehensive global modelling study would be needed to assess the global impact more thoroughly.

To assess the origin of the particles contributing to the scattering shown in Figure 4, the Met Office NAME model was run using the CAM meteorological fields to investigate the origin of the mineral dust by initiating a release of particles at 750 hPa and running the model backwards to determine where the particles are incident with the surface (see section 3.1). An additional calculation was performed for 1000 hPa. The results are shown in Figure 5.

Figure 5.

Back-maps showing the origin of particles detected in the aircraft profile shown in Figure 4 at (a) 750 hPa and (b) 1000 hPa. The units are arbitrary. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Figure 5(a) shows that the main area of dust that is observed at 750 hPa is likely to originate from Mali and Algeria although there are some less extensive sources in Mauritania that could also contribute to the observations. The Mali/Algeria area is highlighted by the satellite observations and analysis of Washington et al. (2003) as being the second most significant area for dust production in Africa north of the Equator after the Bodélé depression. Figure 5(b) confirms that the particles close to the surface are likely to be of marine origin, which ties in with the analysis of Å0.45–0.70.

In deriving the aerosol optical depth, τ, we follow the work of Johnson etal. (2008), who compared the aerosol optical depths derived from nephelometer measurements on the BAe-146 aircraft during the DABEX campaign with co-located AERONET sun-photometers and found that the Rosemount inlet and nephelometer system reasonably efficient at sampling both sub- and super-micron particles. Thus, unlike the system installed on the C-130 aircraft during SHADE (Haywood et al., 2003a), no correction is deemed necessary to account for loss of large super-micron particles. Integrating the scattering from the nephelometer and assuming a campaign-mean single-scattering albedo of 0.97 (Johnson and Osborne, 2011) leads to an aerosol optical depth of around 0.51 at 0.55 μm with an uncertainty of around 20% derived from uncertainties in corrections applied to the nephelometer (Anderson and Ogren, 1998), and thus we quote a τ0.55 of 0.51 ± 0.10.

5.3. Analysis of short-wave broadband radiometer data

To determine the direct radiative effect of dust from the high-level SLR at an altitude of 25 000 ft (375 hPa, R3.1 Figure 2), the methodology from Haywood et al. (2001, 2003a) is used. Essentially, the short-wave upwelling radiation with aerosol present (SWaer↑) is measured and compared to model radiative transfer calculations without aerosol present (SWno_aer↑). The direct radiative effect upon upwelling short-wave radiation (DRESW↑) is:

equation image

The total short-wave DRESW↑ can be estimated using the broadband radiometers (BBRs). Model short-wave calculations are performed using the flexible Edwards and Slingo (1996) radiation code with 220 bands and 300 bands in the short-wave and long-wave regions of the spectrum respectively. The two-stream radiative transfer calculations account for the scattering and absorption by gaseous components. Profiles of ozone and water vapour are taken from the aircraft descent profile and from dropsonde no. 3 (see Figure 4), while concentrations of carbon dioxide, oxygen, nitrous oxide and methane are set to climatological values. Above the level of the aircraft, the climatological concentrations from McClatchey etal. (1972) are assumed. The spectrally dependent sea-surface reflectance is assumed to be that of Cox and Munk (1954) using a 10 m wind speed of 4.8 m s−1 as measured by dropsonde no. 3. The solar constant is set to 1321.9 W m−2 and variations in the solar zenith angle are accounted for.

SWaer↑ is shown as a function of latitude in Figure 6. SWaer↑ is seen to range from approximately 60 W m−2 at 26°N and decrease to a minimum of approximately 51 W m−2 around 23°N before increasing rapidly to around 83 W m−2 around 21°N. The reduction in SWaer↑ between 26°N and 23°N is due to the change in the solar zenith angle as a function of time. There are a few abrupt spikes in the data between 21.5°N and 21°N caused by the presence of high-altitude cirrus streaks that passed quickly away from the area and did not unduly influence the results presented here. Model radiative transfer calculations excluding aerosol are also shown by the dashed line (with the fainter dashed line representing an estimate of the error of the model calculations) which shows that the model and the measurements are indistinguishable from one another northwards of approximately 22.5°N. This agreement between the model and the measurements suggest that very little aerosol is present in the atmospheric column below the aircraft and that the reduction in SWaer↑ between 26°N and 23°N is entirely due to the change in the solar zenith angle and is well modelled. Southwards of 22.5°N, the upwelling irradiance shows a sharp increase from around 50 W m−2 to around 80–85 W m−2; this indicates that the dust is reflecting additional solar radiation back to space and the DRESW↑ is approximately −30 to −35 W m−2. Use of water vapour data from dropsonde no. 2 rather than dropsonde no. 3 (positions marked on Figure 6) changes the upward fluxes by less than 0.2 W m−2 and is not significant. The error in the BBR fluxes is estimated as ±5 W m−2 (Haywood et al., 2003a) and the error in the modelling is estimated as ±3 W m−2 and we therefore quote an SWaer↑ of 83 ± 5 W m−2 and a DRESW↑ of −33 ± 6 W m−2.

Figure 6.

The upwelling irradiance as a function of latitude measured by the clear-dome (0.3–3.0 μm) instrument on the BAe-146 aircraft while performing SLR R3.1 (see Figure 2). The launch positions of sonde no. 2 (S2) and sonde no. 3 (S3) are shown. The most northerly data are from around 1540 UTC, while the most southerly data are from around 1645 UTC. The dashed lines show the model-calculated irradiances accounting for the variation in the solar zenith angle.

We estimate the aerosol optical depth from the BBR measurements by performing radiative transfer calculations including mineral dust within the radiative transfer code using the vertical profile from the nephelometer and scaling the mass concentration until the measured and modelled fluxes are in agreement (Haywood et al., 2001, 2003a). The aerosol optical depth at 0.55 μm, τ0.55, is 0.27 for spherical particles, 0.30 for the spheroid model and 0.22 assuming the irregular particle model. These calculations show a large sensitivity to the aerosol model used, mainly due to the variation in the scattering phase function of the aerosols and asymmetry parameter. The results differ significantly from those determined by integrating the nephelometer scattering (section 5.2: 0.51 ± 0.1). Reasons for these differences will be discussed in section 8.

At the surface, the direct radiative effect upon downwelling solar radiation (DRESW↓) is:

equation image

where SWaer↓ and SWno_aer↓ represent the downwelling irradiance including aerosols and excluding aerosols respectively, and are again taken from the measurements and the modelling calculations respectively, when the aircraft is below the mineral dust aerosol (R4.1, Figure 2). Whilst considerable technological efforts have been made to produce levelling systems for broadband radiometers (Wendisch et al., 2001; Bucholtz et al., 2008), this methodology has proved too costly for installation on the pressurised BAe-146 aircraft. Hence, the BBRs are subject to standard pitch and roll corrections that account for the attitude of the aircraft and any potential misalignment of the BBR when mounted on the aircraft frame. Traditionally this misalignment of the BBRs with the aircraft frame has been accounted for by performing a ‘box-pattern’ where the aircraft is flown in four close-to-orthogonal into-, down- and cross-sun runs. Optimal pitch and roll corrections are then applied to minimise the difference in the measured irradiance (taking into account the variation in the solar zenith angle). We adopt a new procedure: a ‘pirouette’ was performed while the aircraft was on the ground in Nouakchott on 19 June in cloud-free conditions. We then apply pitch and roll offsets to the aircraft data to minimise the standard deviation in the derived fluxes. Figure 7 shows the results from this procedure.

Figure 7.

The downwelling irradiance measured by the clear-domed Eppley pyranometer as a function of the relative heading for the ‘pirouette’ manoeuvre while on the runway. The diamonds represent the uncorrected data and the crosses represent the corrected data using a pitch correction of +0.4° and a roll correction of −0.4°.

Figure 7 shows that when no pitch or roll correction is applied, the BBR measures an irradiance approximately 5% larger when pointing towards the Sun (relative heading 0°) compared to away from the Sun (relative heading 180°). This is due in the most part to the installation angle being set as 3° forward to the aircraft frame. When a correction of +0.4° (−3.0° + 0.4° = 2.6° total correction) is applied to the pitch and −0.4° is applied to the roll, the standard deviation in the irradiances is minimised at 2.7 W m−2, leading to an error estimate of around 5 W m−2 owing to levelling corrections.

The pitch- and roll-corrected measured and modelled SWaer↓ and modelled SWno_aer↓ are shown in Figure 8.

Figure 8.

The measured (bold solid line with fainter solid lines representing the measurement uncertainty of ±9 W m−2) and modelled (dashed lines) SWaer↓ for the sphere, spheroid, and irregular dust particles. The modelled SWno_aer↓ is also shown by the dashed line. The measurements and modelling are for the SLR R4.1 (see Figure 2).

The slope of the measurements and the modelling (even without aerosol included) are in good agreement, suggesting that the primary cause of the variability in the measurements is the change in the solar zenith angle rather than any significant change in the aerosol optical depth. Comparison of SW from the high-level run R3.1 suggests that the BBRs are accurate to around 1–2%; we adopt 2% as an upper limit: this error exceeds that introduced from the levelling correction. The mean SWaer↓ and SWno_aer↓ and associated estimated errors are 449 ± 9 W m−2 and 496 ± 3 W m−2 yielding an estimate of DRESW↓ of −47 ± 9 W m−2.

Radiative transfer calculations are performed to model SWaer↓ including the three mineral dust models described in section 5.1 and the associated τ0.55 derived by matching SWaer↑. The resulting SWaer↓ and DRESW↓ are summarised in Figure 8 and Table I. Figure 8 shows that the sphere, spheroid and irregular models are all able to represent the measured SWaer↓ within the uncertainty of the measurements.

Table I. The modelled and measured SW irradiances and DRE at the surface.
 SWno_aer↓SWaer↓DRESW↓
  1. Derived from measurements and modelling as described in the text. τ, ωo and g correspond to the values at 0.55 μm. τ represents the aerosol optical depth that is necessary to fit SWaer↑. The estimated error in modelled values of SWaer↑ represents estimated errors in modelling the atmospheric radiative transfer for a given set of aerosol optical parameters.

Measured (BBRs)449 ± 9−47 ± 9
Modelled Spherical (τ = 0.27, ωo = 0.95, g = 0.73)496 ± 3452 ± 5−44 ± 6
Modelled Spheroid (τ = 0.30, ωo = 0.96, g = 0.73)496 ± 3451 ± 5−45 ± 6
Modelled Irregular (τ = 0.22, ωo = 0.97, g = 0.62)496 ± 3445 ± 5−51 ± 6
Modelled Spherical WMO (τ = 0.52, ωo = 0.82, g = 0.73)496 ± 3382 ± 5−114 ± 6

Table I shows SWaer↓ and DRESW↓ using a mean solar zenith angle of 59.3°. However, both the spherical model and spheroid model yield values of τ0.55 that differ considerably from the value of τ0.55 = 0.51 ± 0.10 that was derived from the nephelometer (section 5.2). A further calculation was performed using the spherical model combined with the refractive indices of WMO (1986) for dust-like particles, yielding a more absorbing aerosol with ωo0.55 = 0.82 (Table I). These calculations suggest that while the derived τ0.55 value of 0.52 necessary to match SWaer↑ is in good agreement with the nephelometer, the SWaer↓ of 382 ± 5 W m−2 is in significant disagreement with the measured value of 449 ± 9 W m−2. This evidence supports the use of a more modest τ0.55 (0.22–0.30 depending on the model used) and a significantly less absorbing ωo (0.95–0.97).

5.4. Analysis of the short-wave spectral irradiance

The spectral variation of the DRE (DRESW_λ) can be investigated using the SHIMS instrument. First we analyse the impact on DRESW_λ↑. Rather than estimate the temporal evolution, we chose two time periods from the SLR corresponding to a 10-minute section of the aerosol-free part of the run (near sonde no. 2 on Figure 6) and a 5-minute section of the run with significant aerosol (near sonde no. 3 on Figure 6). As the SHIMS instrument is essentially uncalibrated, we perform spectrally resolved radiative transfer calculations representative of aerosol-free conditions to calibrate the data using data from sonde no. 2. This calibration procedure provides spectrally resolved data calibration coefficients (units W m−2/μm/count) that are assumed to be temporally invariant at constant temperature. The results from these measurements are shown in Figure 9.

Figure 9.

The measured and modelled upwelling spectral irradiances, with (SWaerλ↑) and without (SWno_aerλ↑) mineral dust aerosol. SWaerλ↑ is shown for three different models. The measurements and modelling are for a short section of SLR 3.1 (see Figure 2) as described in the text. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Figure 9 shows a significant increase in the upwelling spectrally resolved irradiances when going from no aerosol (SWno_aerλ↑) to aerosol conditions (SWaerλ↑). Integrating the area under the curves suggests ΣλSWno_aerλ↑ = 53 W m−2 (in forced agreement with the modelling shown in Figure 6 owing to the calibration procedure), while ΣλSWaerλ↑ = 78 W m−2, which is in reasonable agreement with the 83 ± 5 W m−2 derived from the BBRs. These measurements suggest a DRESW_λ↑ of around −28 W m−2 when the temporal variation in the solar zenith angle is accounted for (SWno_aer↑ = 50 W m−2, see Figure 6), which is within the measurement uncertainty of the DRESW↑ of approximately −33 ± 6 W m−2 derived from the BBRs.

The modelled spectral irradiances are also shown on Figure 9. All of the models (sphere, spheroid, and irregular particles) show practically identical spectral responses. The agreement between the model and observations is within 5% between 0.4 and 0.8 μm. At longer wavelengths the modelled spectral irradiances tend to overestimate those from the measurements; this is likely due to differences in actual and assumed aerosol size distribution, particularly in the coarser particles, and also spectral variations in the refractive index, which may not match those of Balkanski et al. (2007). It may also be due to differences in the surface reflectance under clear-sky and diffuse light conditions that cannot be accounted for using this calibration procedure. At wavelengths shorter than 0.4 μm there are some additional discrepancies between the model and the measurements. The measured SWaerλ↑ is less than SWno_aerλ↑ in this spectral region which suggests that the calibration procedure may not work sufficiently accurately here or that the absorption by ozone or mineral dust is not well modelled. These wavelengths are subject to absorption by ozone, and the calibration procedure essentially assumes that ozone is invariant; the observations in Figure 4 show that this assumption is unlikely to be valid. Additionally, the absorption properties of mineral dust at these wavelengths may not be accurately modelled, which is certainly possible given the uncertainty in the imaginary part of the refractive index at these wavelengths. Nevertheless, the general agreement between the model and measured spectral irradiances suggests that the optical properties of the aerosol are reasonably well modelled.

In a similar manner, the upper SHIMS instrument can also be used to determine SWaerλ↓ and radiative transfer modelling may be performed to determine SWno_aerλ↓ and hence derive DRESW_λ↓. As is the case for the broadband radiometers, the measured irradiance will be a strong function of the pitch and roll of the aircraft; pitch and roll corrections are therefore applied. We apply the identical aerosol optical depths given in Table I so that the modelled irradiances are consistent with the estimates of DRESW_λ↑. The results are shown in Figure 10.

Figure 10.

The measured and modelled downwelling spectral irradiances, with (SWaerλ↓) and without (SWno_aerλ↓) mineral dust aerosol. SWaerλ↓ is shown for three different models. The standard pitch and roll corrections are applied to the aircraft data. The measurements and modelling are for SLR 4.1 (see Figure 2). This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Once again, the modelled spectral irradiance is in good agreement with the measurements independently of the aerosol model utilised despite the difference in the assumed aerosol optical depth in the model simulations. Between 0.45 μm and 1.6 μm the difference between the modelled and measured fluxes does not exceed 5% except in the strong water vapour absorption bands. The integrated flux obtained from the measurements with SHIMS is approximately 452 W m−2, in good agreement with the measurements from the BBRs which suggest 449 ± 9 W m−2 (Table I).

5.5. Determination of the long-wave spectral radiative effect

Mineral dust also exerts a significant influence on terrestrial radiation. To determine this influence we examine the ARIES radiances in the 8–12 μm atmospheric window from around 23.5°N (no dust below, see Figure 5) and 21°N (dust below, see Figure 5). The radiances are then converted into equivalent brightness temperatures. Corrections are made for variations in the sea-surface temperature between the dusty and no-dust cases by removing 1.2 K from the brightness temperature for the no-dust case. This difference corresponds to the difference between the near-surface temperatures determined from sonde no. 2 and sonde no. 3. Figure 11(a) shows the brightness temperature obtained from the dusty (black line) and the no-dust (red line) cases, while Figure 11(b) shows the difference between the two sets of measurements.

Figure 11.

ARIES spectra across the 700–1300 cm−1 wave-number range during sections of SLR R3.1 (see Figure 2). (a) The brightness temperature derived from nadir views of the sea surface for the no-dust (1610–1620 UTC, 23.0°–23.8°N, black line) and dusty (1640–1645 UTC, 20.9°–21.3°N concurrent with sonde no. 3, grey line) cases. The no-dust data are corrected for variations in SST between the two views by removing 1.2 K. (b) The difference (no-dust minus dust) in the brightness temperature with the 1.2 K offset. The ARIES spectrum is shown in grey, with a 50-point running mean shown in the thin black line. Calculations of the difference in brightness temperature are also shown for Fouquart (thick black line), Volz (thick dotted line), WMO (thick dashed line), and Balkanski (red line) refractive indices. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

It is immediately apparent from Figure 11(a) that dust reduces the upwelling brightness temperature by absorbing and re-emitting terrestrial radiation in the atmospheric window by a maximum of approximately 2 K. This is approximately half of the 4 K brightness temperature difference observed with the same instrument during a Saharan dust outbreak with τaerλ=0.55 of 0.55 during the SHADE campaign (Highwood et al., 2003). Thus if we assume in this study that τaerλ=0.55 derived from the solar radiation measurements in Table I (0.22–0.30) are of higher accuracy than that those derived from the nephelometer, we appear to have reasonable consistency between the impact on long-wave (LW) brightness temperatures in the atmospheric window between the two studies. The reduction in water vapour in the dusty case compared to the non-dusty case that is apparent in the tephigrams shown in Figure 3 is also apparent in Figure 11, as evidenced by the reduction in the strength of the sharp water-vapour absorption lines.

We perform a modelling estimate of the spectral impact of Saharan dust in the LW spectral region by assuming an optical depth at 0.55 μm of 0.30 corresponding to the AOD for spherical particles (see Table I), and performing radiative transfer calculations in the LW region of the spectrum with and without dust using the Edwards and Slingo (1996) radiation code. Both radiance and irradiance calculations are performed excluding aerosol and including aerosol using the vertical profile of scattering determined from the nephelometer, but scaled to a total τ0.55 of 0.30. The surface temperature is set to 292.7 K which corresponds to that derived from ARIES for the low level run (R4.1, Figure 2). Four different refractive indices are used in the infrared region of the spectrum: those of Volz (1973), Fouquart et al. (1987), WMO (1986), and Balkanski et al. (2007). Figure 11(b) shows that the Balkanski et al. (2007) refractive indices appear to best represent the radiative impact of Saharan dust in this particular case.

The irradiance calculations suggest that the top-of-atmosphere (TOA) outgoing LW radiation (OLR) is around 300.1 W m−2 with dust aerosol present, but 309.2 W m−2 in the absence of aerosols, suggesting a radiative impact of mineral dust at the top of the atmosphere of around 9 W m−2 when using the Balkanski et al. (2007) refractive indices. The uncertainty in this estimate is bolstered by the relatively good agreement between the measured and modelled spectral radiances, and we estimate a TOA LWaer↑ of +9 ± 3 W m−2.

6. Comparison of aircraft measurements of SW, DRESW↑ and τ0.55 with those from SEVIRI

Previous measurements of the radiative impact of Saharan dust over the oceans have shown that the Clouds and Earth Radiant Energy System (CERES) is capable of retrieving the SW in the presence of mineral dust and hence DRESW↑ may be obtained to a reasonable accuracy (Haywood et al., 2003a). We perform a similar analysis, but utilise retrievals from the SEVIRI (Spinning Enhanced Visible and InfraRed Imager) instrument.

Broadband short-wave radiances and fluxes derived from SEVIRI are obtained under the auspices of the Geostationary Earth Radiation Budget (GERB) project. In this project, these products are referred to as ‘GERB-like’ High Resolution, with a spatial resolution of 3 × 3 SEVIRI pixels (∼10 × 10 km at nadir) and a temporal resolution of 15 minutes. They are essentially the result of a narrow- to broad-band conversion on the SEVIRI solar band channels at 0.6, 0.8 and 1.6 microns, coupled with a radiance-to-flux conversion via the application of CERES Tropical Rainfall Measuring Mission (TRMM) angular distribution models (ADMs: Loeb et al., 2003; Dewitte et al., 2008). This latter conversion requires a scene identification which is derived from SEVIRI, but at present the available GERB-like fluxes are obtained without accounting for the impact of aerosol on the anisotropy of the scene. Brindley and Russell (2008) investigated the impact of this omission over ocean and found that it could result in short-wave flux errors of up to ∼50 W m−2 depending on the solar and viewing geometries. In all cases presented here the original ADM applied to the GERB-like radiances was representative of clear-sky conditions.

Figure 12 shows the TOA GERB-like SW as a function of latitude along the aircraft track (red crosses). Also shown is SW measured by the aircraft corrected to the TOA. A clear offset is apparent between the GERB-like estimate and the aircraft observations, particularly at the lowest and highest latitudes, associated with the highest and lowest aerosol loadings respectively. We apply two methods to try to account for the effect that the dust aerosol will have on the scene anisotropy. The first utilizes the theoretical dust ADM developed by Brindley and Russell (2008), using retrieved SEVIRI optical depths and the appropriate solar/viewing geometry to derive a representative anisotropic factor which can then be applied to the GERB-like radiances (referred to as ‘retrieved’). The second applies the approach of Loeb et al. (2003) to correct the anisotropic factor taken from the CERES ADM based on the observed radiance, the applied CERES ADM radiance, and a theoretical aerosol–radiance look-up table (referred to as ‘corrected’). In both cases the new fluxes show a flattening in SW as a function of τ0.55, which is consistent with the expectation of a reduction in scene anisotropy as an increasingly thick aerosol layer covers an ocean surface. Results from Brindley and Russell (2008) indicate that for the aerosol loadings and solar viewing zenith angles sampled by SEVIRI along the aircraft track, applying the Loeb et al. (2003) correction will provide the most reliable SW estimates. This would seem to be corroborated by the improved agreement with the aircraft corrected values for this case.

Figure 12.

The top-of-atmosphere outgoing short-wave radiation from the observations derived from SLR 3.1 (adjusted from the level of the aircraft as shown in Figure 6 by adding 15 ± 2 W m−2) and from three different retrieval methods derived from SEVIRI radiances as described in the text. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

For all three cases it is possible to estimate DRESW↑ by comparing the SW at the maximum aerosol loading (at 20.8°N) with those with minimal mineral dust loading. Figure 13 shows the DRESW↑ derived by subtracting the mean fluxes at 24°–25°N from the retrievals. DRESW↑ values obtained in this way over the range of latitudes 21°–21.5°N (coincident with the −33 ± 6 W m−2 derived from the aircraft) covered by the aircraft observations are −49 W m−2, −43 W m−2 and −38 W m−2 for the GERB-like, retrieved and corrected cases respectively. The estimated potential error in DRESW↑ includes the impact of changes in aerosol other than dust for 24°–25°N and 21°–21.5°N, differences in the vertical profile of water vapour, and changes in the surface reflectance as a function of solar zenith angle, and may reach 10 W m−2. Thus the observed DRESW↑ is in reasonable agreement, particularly for the ‘retrieved’ and ‘corrected’ methodologies.

Figure 13.

The Direct Radiative Effect DRESW↑ from the aircraft measurements during SLR R3.1 (Figure 2) and the SEVIRI retrievals. Also shown are the model DRESW↑ derived from the GLOB-H and GLOB-LN simulations. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

τ0.55 derived from SEVIRI radiances is shown in Figure 14 for 1500 UTC. We chose 1500 UTC because this is close to the measurements shown for the aircraft and is concurrent with the model output which is available every 3 hours. Two different approaches are used in the retrieval process. Over land the technique of Brindley and Russell (2009) is employed, utilizing dust-induced brightness temperature deviations in the SEVIRI 10.8 and 13.4 μm channels. Comparisons with co-located AERONET and aircraft measurements (Brindley and Russell, 2009; Christopher et al., 2011) indicate a maximum uncertainty of ∼0.3. Indeed, the comparison with the concurrent mean daily data from AERONET which is also shown on Figure 14 shows reasonable agreement in all but the Cape Verde site (which can be considered oceanic in practice for the retrievals documented here). Over land regions Figure 14 shows a broad area of τ0.55 > 0.6 stretching from southern Algeria through central Mali into southern Mauritania and Senegal. Maximum τ0.55 exceed 2.5 on the Algeria/Niger border and in a small area of central Mali. The very significant dust event that was recorded over Tamanrasset (τ0.55 = 2.19) appears well represented in the SEVIRI retrieval.

Figure 14.

τ0.55 determined from the SEVIRI sensor on Meteosat Second Generation. Areas associated with cloud are coloured white. Sun-glint angles (SGA) of 20° and 40° are shown by the ellipses. τ0.55 from five AERONET sites are shown by the coloured circles. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Over ocean, reflectances in the SEVIRI short-wave bands are used as input to appropriate look-up tables as described in Brindley and Ignatov (2006). De Paepe etal. (2008) show that retrievals using this method exhibit root-mean-square (RMS) differences with co-located MODerate-resolution Imaging Spectroradiometer (MODIS) optical depths which are typically less than 0.1. However, in this case there is a problem: sun-glint affects the retrievals, making them unreliable, which can readily be seen by the poor comparison of the retrieval and the concurrent AERONET τ0.55 in Cape Verde. In the standard operational algorithm, retrievals are considered unreliable if the sun-glint angle (SGA) is less than 40°; this angle is marked on Figure 14 by the outer circle. No retrievals of SW, DRE or τ0.55 would be possible in almost the entire domain if this strict criterion were applied. We relax the criterion to an SGA of 20° (inner circle) which then allows some comparison against the aircraft measurements (shown by the red line). This relaxation threshold is based on the fact that when retrievals are analysed in more favourable SGA conditions, the increases in τ0.55 in the geographical area north of the SGA = 20° line on the aircraft track are still evident. Also, the lack of a discontinuity when comparing τ0.55 determined in pixels over land to adjacent pixels over ocean close to the aircraft track north of the SGA = 20° line suggests that features seen here are real and not an artefact of sun-glint. However, we acknowledge that SW, DRESW↑ and τ0.55 over ocean may be less accurately determined than would be the case in more favourable SGA conditions.

7. Modelling the Saharan dust outbreak

Previous operational modelling of Saharan dust in the Met Office suite of models (MetUM) has included use of mesoscale models (Greed et al., 2008) and global models (Milton et al., 2008). While Greed et al. (2008) validated the mesoscale model using aircraft and satellite data (Christopher et al., 2008), here we focus our attention on validating the dust forecast from the global version of the MetUM. Two specific dust uplift schemes are investigated: the scheme developed for the HadGEM climate model (GLOB-H) dust scheme based on Woodward (2001), and the Dust Emission and Deposition (DEAD) scheme based on Zender et al. (2003) which imposes dust emissions into log-normal size distributions (GLOB-LN). Further details of the modifications to the mineral dust scheme are detailed in Johnson etal. (2011).

The model radiative diagnostics are output every three hours. We analyse data from 1500 UTC, which most closely corresponds to the data from the aircraft, which was obtained between 1540 UTC and 1645 UTC. Figure 13 shows the DRESW↑ obtained from both the GLOB-H and GLOB-LN simulations. While the magnitude appears reasonable, the GLOB-H and GLOB-LN simulations indicate that the area of significant DRESW↑ is displaced to the south by 1–2 degrees. Such spatial discrepancies are frequently observed in coarse-resolution models, and can be due to timing errors, errors in meteorological fields, and errors in the model fields of clay fraction, vegetation and moisture (e.g. Milton et al., 2008). To present a more comprehensive investigation of the performance of the dust model, we examine the spatial coherence of the model fields of τ0.55.

Figure 15(a) and (b) show τ0.55 from the GLOB-H and GLOB-LN schemes respectively. Figure 15(a) and Figure 15(b) both show a τ0.55 which qualitatively indicates a similar swath of high values running from around the border of Algeria and Niger (∼20°N, 7°E) to the Dakar Peninsula, Senegal (∼15°N, 17°E). This spatial distribution is similar to the observations from SEVIRI shown in Figure 13. However, τ0.55 from the GLOB-H scheme is typically lower than that derived from the GLOB-LN scheme by around 0.2. Comparisons against the satellite observations from SEVIRI suggest that neither uplift scheme captures the intensity of the dust event. The green/yellow colours in Figure 14(a)–(b) show areas where 0.3 < τ0.55 < 0.6/0.6 < τ0.55 < 0.8 while the corresponding areas in the SEVIRI retrievals in Figure 13 show 0.6 < τ0.55 < 2.5. As the SEVIRI retrievals are estimated to exhibit an uncertainty of around 0.3 (section 6; Christopher et al., 2011), the GLOB-H and GLOB-LN models appear to produce a τ0.55 that is systematically too low for this particular event over much of West Africa, although the GLOB-LN simulations produce τ0.55 in closer agreement with the SEVIRI retrievals. Some additional geographic differences between the observations and the model are worth highlighting. The high τ0.55 in the SEVIRI retrievals south of ∼15°N in the Sahelian region and the proximity to the cloud areas (marked white) are suggestive of dust generated by gust fronts from strong convective activity (e.g. Marsham et al., 2008); the turbulent downdraughts associated with deep convective systems will not be well represented in the dust parametrization in the relatively coarse resolution MetUM because it is driven by mean wind speeds. However, even at finer horizontal resolutions, the MetUM struggles to produce enough mineral dust over Sahelian regions. Johnson et al. (2011) make a more thorough comparison of the performance of a higher-resolution mesoscale version of the MetUM model and conclude that misrepresentation of soil texture, soil moisture and vegetation, and the lack of an observational constrained or geomorphologically based preferential source term could also contribute to the underestimation of dust over the Sahel. Both GLOB-H and GLOB-LN versions produce too much dust in 20°–25°N, 0°–10°W, which may be due to errors in the soil clay fraction, surface vegetation, surface moisture or wind speed. It is difficult to conclude a discernibly better performance for the GLOB-H or GLOB-LN schemes when comparing against AERONET because of the sparseness of the AERONET observations, which is why cross-validation of satellite retrievals against AERONET are important when assessing specific case-studies (e.g. Greed et al., 2008; Brindley and Russell, 2009; Christopher et al., 2011).

Figure 15.

(a) and (b) show the geographic distribution of τ0.55 determined for the GLOB-H and GLOB-LN model respectively, while (c) and (d) show the geographic distribution of DRESW↑ determined for the GLOB-H and GLOB-LN model respectively. The locations of the Dakar and Cape Verde AERONET sun-photometers are marked by the yellow and green circles on (a) and (b). This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Figure 15(b) shows the DRESW↑ derived from the model. As expected for a partially absorbing aerosol, DRESW↑ is greater over ocean than over land areas owing to the increased efficiency of the radiative forcing over dark surfaces. The impacts of clouds can be seen in Figure 15(b) where DRESW↑ is reduced to zero or even slightly positive values. As expected from the increased τ0.55 for the GLOB-LN simulations, GLOB-LN shows a higher DRESW↑.

A final assessment of the relative performance of the GLOB-H and GLOB-LN simulations is performed by examining the correlation between AERONET and the model optical depths at 440 nm, which are reported at more sites than for 550 nm in this geographic region. Data is used over the domain 30°W–15°E and 0°–30°N for the GERBILS time period (18–30 June 2007). The model data from the first 12 hours of forecast runs is available at 3-hour resolution and compared to AERONET data within ±1 hour of the model data. The results are shown in Figure 16.

Figure 16.

A comparison of the correlation between the GLOB-H and GLOB-LN schemes and the AERONET τ0.44. The data were selected from the GERBILS period (18–30 June 2007).

Figure 16 shows that the bias is reduced in the GLOB-LN simulations and R2 is improved significantly to 0.31 from 0.16. The improved correlation suggests that in this region for this time of year the GLOB-LN simulations outperform the GLOB-H simulations, although there is a low bias in both sets of simulations.

8. Discussion and conclusions

This study documents the meteorological conditions experienced during a flight of the BAe-146 aircraft off the coast of West Africa while on transit to the GERBILS measurement campaign. The appearance of a distinct Saharan Air Layer (SAL) as the aircraft progresses southward is clearly evident in the temperature and humidity structures determined from dropsondes. Mineral dust is observed in the SAL where the vertical profile shows a significant decrease in ozone, presumably owing to heterogeneous reactions that deplete ozone on the surface of the mineral dust. Assessment of the relative importance of the radiative effect caused by the ozone depletion suggests that this is an insignificant effect compared to that from the mineral dust itself.

This study provides a relatively comprehensive measurement and modelling radiative closure study of the radiative impacts of mineral dust. While spherical, oblate and prolate sphere models can all provide reasonable agreement in the broadband and spectral upwelling and downwelling radiation, there are significant differences in the derived aerosol optical depth that is necessary to obtain radiative closure for each of the models, with τ0.55 ranging from 0.22 to 0.30 (Table I) or a factor of 1.4. This measurement study is hampered by the large disagreement between these aerosol optical depths and that derived by integrating the vertical profile from the nephelometer (Figure 4), which suggests a τ0.55 of around 0.51. The close level of agreement between the modelled and measured broadband irradiances and spectral radiances suggests that the error is likely to lie with the nephelometer, but further investigations are warranted over ground sites with high-accuracy sun-photometers (e.g. Osborne et al., 2011).

While many studies of the absorption properties of Saharan dust have suggested that mineral dust is not strongly absorbing (e.g. Haywood et al., 2003; McConnell etal., 2008; Osborne et al., 2008), these studies have relied upon In situ measurements of scattering and absorption or on retrieval algorithms where absorption is inferred from e.g. almucantar scans with AERONET sun-photometers. This measurement study again suggests that Saharan mineral dust measured is not significantly absorbing; ωo0.55 of 0.95 to 0.97 provides a reasonable match between both the measured and the modelled DRESW↓ and DRESW↑ while utilisation of the WMO (1986) refractive indices with an ωo0.55 of 0.82 leads to a τ0.55 of around 0.52 (in agreement with the nephelometer data), but a modelled value of DRESW↓ of −114 ± 6 W m−2 is in significant discrepancy (by a factor of around 2.4) with the measured value of −47 ± 9 W m−2. Put simply, using a highly absorbing aerosol model combined with a higher aerosol optical depth to get the upwelling irradiance correct leads to very large discrepancies between the measured and modelled downwelling irradiance. Some recent studies during the SAMUM campaign (e.g. Otto et al., 2008) have suggested visible ωo of as low as 0.76 to 0.79 owing to the presence of a significant number of coarse particles. However, further investigation of their results (e.g. Otto et al., 2008, their Fig. 11) suggests that the modelled downwelling irradiances are consistently significantly lower than the measured downwelling irradiances by up to 10% suggesting that particulate absorption is too high in their study. In our study the agreement between the modelled and measured irradiances is significantly better, suggesting that the absorption properties of mineral dust in this region are reasonably represented by an ωo in the range 0.95–0.97. While there is likely to be significant variability in dust absorption owing to the contribution from different source regions with different inherent mineralogy (e.g. Petzold et al., 2009), the use of WMO (1986) refractive indices for mineral dust should be discontinued for Saharan dust outbreaks.

While a single case-study such as this one cannot provide comprehensive validation of satellite retrievals or modelling estimates using GLOB-H or GLOB-LN, the synergy between the SEVIRI satellite estimates of τ0.55 and the modelling estimates shows some qualitative spatial agreement. However, neither the GLOB-H nor GLOB-LN models generate the very high τ0.55 observed in the SEVIRI retrievals or AERONET sun-photometers. The model also does not generate dust to the south of the region where dust is generated by the intense downdraughts associated with mesoscale convective systems which are poorly represented in relatively coarse-resolution models (e.g. Marsham et al., 2008; Knippertz etal., 2009). There is some evidence that the GLOB-LN model simulations show less bias in τ0.55 than the GLOB-H simulations over the GERBILS period.

Acknowledgements

Airborne data were obtained using the BAe-146-301 Atmospheric Research Aircraft (ARA) flown by Directflight Ltd and managed by the Facility for Airborne Atmospheric Measurements (FAAM), which is jointly funded by the Met Office and the Natural Environment Research Council (NERC). The staff of the Met Office, FAAM, Directflight and Avalon engineering are thanked for their dedication in making the GERBILS measurement campaign a success. Finally, JH would like to dedicate this work to Professor Tony Slingo who passed away suddenly in October 2008. Tony was an inspiration to all of those who worked with him, and he would have been fascinated by the GERBILS measurement campaign.

Ancillary