Mineral dust aerosol net direct radiative effect during GERBILS field campaign period derived from SEVIRI and GERB



Colocated Spinning Enhanced Visible and Infrared Imager (SEVIRI) retrieved dust optical depths at 0.55 microns, τ055, and Geostationary Earth Radiation Budget (GERB) fluxes at the top of atmosphere are used to provide, for the first time, an observationally based estimate of the cloud-free net direct radiative effect (DRE) of mineral dust aerosol from geostationary satellite observations, providing new insights into the influence of time of day on the magnitude and sign of the shortwave, longwave, and overall net effect during sunlit hours. Focusing on the Geostationary Earth Radiation Budget Intercomparison of Longwave and Shortwave radiation (GERBILS) campaign over North Africa during June 2007, the presence of mineral dust aerosol reduces the outgoing longwave radiation at all times of day with the peak reduction clearly following the diurnal cycle of surface temperature. The instantaneous shortwave DRE shows strong dependencies on pristine sky albedo and solar zenith angle such that the same dust loading can induce a positive or negative value dependent on time of day. However, the area mean net DRE over the GERBILS period is dominated by the longwave component at all sampled times of day, with mineral dust inducing a reduction in outgoing net flux of the order of 10W m−2. Hence, in the mean sense, Saharan dust is found to warm the Earth-atmosphere system over northern Africa and the Middle East.

1 Introduction

The recognition that mineral dust aerosol is a significant component in the climate system coupled with recent developments in remote sensing satellite products has resulted in a significant increase in dedicated research over the last 10–15 years into its physical and optical properties [e.g., Sokolik et al., 2001; Balkanski et al., 2007; Shao et al., 2011; Formenti et al., 2011]. Mineral dust is an efficient scatterer of solar SW radiation with a single scattering albedo at 0.55μm typically quoted in excess of 0.9 [Tegen and Lacis, 1996]. However, due to the relatively large micron size of dust particles, it is also an efficient absorber of longwave (LW) terrestrial radiation capable of exerting a significant greenhouse effect depending on the atmospheric burden and vertical distribution [e.g., Myhre and Stordal, 2001; Haywood and Boucher, 2000]. This impact on both SW and LW radiation can lead to changes in atmospheric circulation through the modification of the surface energy budget and also by directly changing atmospheric temperatures. Dust may also change boundary layer mixing and therefore modify its subsequent advection patterns and emission [Grini et al., 2006]. Significant efforts are now being made to assimilate the effects of dust within forecasting Numerical Weather Prediction models [Greed et al., 2008; Morcrette et al., 2009].

Considered as a whole, the Sahara desert is the largest source of mineral dust to the atmosphere in the world [Washington et al., 2003]. There have been a number of in situ field campaigns on the border of and within the Sahara (e.g., Saharan Dust Experiment) [Tanré et al., 2003], African Monsoon Multidisciplinary Analyses [Redelsperger et al., 2006], Bodélé Dust Experiment [Todd et al., 2007], Dust and Biomass-burning Experiment [Haywood et al., 2008], Dust Outflow and Deposition to the Ocean [McConnell et al., 2008], Saharan Mineral Dust Experiments (SAMUM) [Ansmann et al., 2011], Geostationary Earth Radiation Budget Intercomparison of Longwave and Shortwave radiation (GERBILS) [Haywood et al., 2011], and FENNEC [Washington et al., 2012]. One aim has been to address deficiencies in our current knowledge of dust composition, size distribution, and optical properties and to achieve radiative closure by coordinating these measurements with radiative flux observations [e.g., Osborne et al., 2011], thus helping reliable empirical relations between the radiative impact of dust and its properties to be established. Similarly, recent advances in sensors and innovative retrieval techniques have enabled quantitative estimates of aerosol—primarily dust—optical depth to be made over bright desert regions [Carboni et al., 2012]. Combining these estimates with top of atmosphere (TOA) SW and LW fluxes derived from instruments measuring broadband radiation has allowed the direct cloud-free TOA radiative effect (DRE) of aerosol to be obtained empirically [e.g., Hsu et al., 2000; Zhang and Christopher, 2003; Patadia et al., 2008; Chen et al., 2009]. However, all of these studies suffer from limited temporal sampling since they employ observations from satellites in low Earth orbit.

The Spinning Enhanced Visible and Infrared Imager (SEVIRI) and Geostationary Earth Radiation Budget (GERB) instrument on board the Meteosat-9 geostationary satellite provide observations which can in theory be exploited to investigate how the impact of mineral dust aerosol on the TOA radiation budget varies with time of day. Here we use a previously developed and evaluated dust retrieval algorithm designed for SEVIRI (see section 2.2) in combination with the radiative fluxes available from GERB to estimate, for the first time, the impact of dust aerosol on the net TOA radiation budget on a subdiurnal timescale. For the purposes of this study, we focus on the north African and Arabian region (0°–40°N, 25°W–65°E) over the period 20–29 June 2007 which coincides with the GERBILS observational field experiment.

The GERBILS field campaign had four major aims: to characterize the geographic distribution and physical and optical properties of mineral dust, to assess the impacts of mineral dust on the radiative budget, to assess satellite retrievals of mineral dust associated radiative impacts, and to use the aircraft and satellite remote sensing techniques to validate and improve the performance of dust modeling within numerical models [Haywood et al., 2011]. A key motivation for the campaign was provided by results suggesting that significant discrepancies existed between numerical model simulations of outgoing longwave radiation and those observed from satellite over the north-west African region [Haywood et al., 2005]. This study suggested that the lack of a specific treatment of mineral dust in the model was likely to be at least partly responsible for the differences seen.

For ease of notation, from now on in this study we use the term DRE to refer to the TOA clear-sky DRE. The instrumentation and methods used to obtain dust optical depth and DRE are introduced in section 2, including a detailed explanation of the method used for deriving the pristine sky planetary albedo, crucial for estimating the TOA SW DRE. Dust optical depths and the LW and SW DRE for June 2007 are presented in section 3, followed by specific discussions on the impact of solar zenith angle on the pristine sky planetary albedo and instantaneous SW DRE. The net effect of mineral dust aerosol over the GERBILS period is then derived. Finally, in section 4 we draw conclusions from this study and outline areas for future work.

2 Data and Methodology

2.1 Meteosat-9 Instrumentation

Meteosat-9 is the second of the Meteosat Second Generation satellites launched in December 2005 and becoming operational in May 2007. It is situated in a geostationary orbit above the equator at 0°W and carries two Earth observation instruments, SEVIRI and GERB-1. The SEVIRI imager measures in 11 narrowband spectral channels, each with a spatial sampling resolution of 3×3 km2at nadir, and in one high-resolution broadband channel with a resolution of 1×1 km2. The instrument has a temporal resolution of 15 min [Schmetz et al., 2002]. The GERB broadband radiometer is the only instrument to measure broadband-emitted thermal and reflected solar fluxes at the Earth's top of atmosphere from a geostationary orbit. Observations are made in two broadband channels: a shortwave channel covering the reflected solar spectrum (0.32–4 μm) and a total channel covering a wider range (0.32–100 μm). A longwave observation (4–100 μm) is obtained via subtraction. Harries [2005] details the GERB instrument design, data processing routines, and available products. The coverage of the two instruments extends from approximately 60°N–60°S and 60°W–60°E.

Here we make use of GERB High Resolution (HR) level 2 shortwave and longwave fluxes contained within the GERB HR level 2 product (hereafter GHR2). The GHR2 products are derived as follows. Initially, a narrow-broadband conversion is performed on the observed SEVIRI radiances within a native GERB footprint (∼50×50 km2at nadir), producing GERB-like HR radiances, with a resolution of ∼10×10 km2 at nadir. The ratio between the observed native GERB radiance and the GERB footprint mean GERB-like HR radiance provides a correction factor which is constrained to vary smoothly at the subfootprint (HR) scale. Multiplying the GERB-like HR radiances by this factor produces the equivalent GERB HR values [Dewitte et al., 2008]. Shortwave GERB HR fluxes are determined from these radiances by the application of the appropriate Clouds and Earth's Radiant Energy Tropical Rainfall Measuring Mission angular distribution models chosen according to the GERB scene ID which consists of a cloud detection algorithm and a fixed land surface map. In the longwave, the radiance to flux conversion is performed using a regression based on the viewing zenith angle and the measured radiances in the SEVIRI 6.2, 10.8, 12.0, and 13.4 μm channels [Clerbaux et al., 2003].

For the purposes of this initial study we make use of SEVIRI and GERB HR observations taken at 0900, 1200, and 1500 UTC, although the methodology outlined here could be easily extended to higher time resolution observations.

2.2 Dust Optical Depth Retrieval

Daytime dust loading is quantified from SEVIRI infrared channels using the method of Brindley and Russell [2009] which essentially exploits the expected thermal contrast between lofted dust layers and the underlying surface. The magnitude of this contrast, which is observed as a depression in the 10.8 μm channel brightness temperature (TB108), can be related to dust loading provided that a dust-free or reference TB108 can be identified for each time slot and pixel where a retrieval is performed.

Retrievals are only performed on pixels which meet the following conditions:

  1. A pixel must be identified either as cloud free (diagnosed either via the scheme of Derrien and Le Gléau [2005] or via the standard GERB detection scheme [Ipe et al., 2004]) or dust flagged [Banks and Brindley, 2013]. The test for dust is applied subsequent to the cloud masking such that dusty pixels that may orignally have been misidentified as cloud are reinstated.

  2. The solar (θz) and viewing (θv) zenith angles must be less than 70°.

  3. The pixel must be classified as land by the GERB surface-type mask.

In this study, the retrieval algorithm stores the previous 16 days of cloud-screened TB108, and the most pristine (lowest τ055) observation at each time step for each pixel is identified. An atmospheric correction is applied to the most pristine TB108 observation to account for changes in the skin temperature and total column water vapor between this observation and the day of the actual retrieval. The atmospheric fields are taken from ERA-Interim data. The mean correction term is found to be 33% of the final ΔTB108 when the aerosol optical depth at 550 nm (hereafter τ055) exceeds 0.5 but increases to 68% for τ055 below 0.5. Thus, the retrieved τ055 at low dust loadings will be more sensitive to errors in the atmospheric fields. The corrected observation is the dust-free brightness temperature, TB108df, such that the residual difference (ΔTB108) between it and TB108 on the retrieval day is attributable to the presence of dust alone

display math(1)

The ratio of ΔTB108/ΔTB134 is then used to determine τ055 where ΔTB134 is derived in a similar manner from observations in the 13.4 μm channel and is used to minimize the impact of variations in dust layer altitude on the brightness temperature signals. The method is discussed in more detail in Brindley [2007] and Brindley and Russell [2009].

Retrievals using this method have been carefully evaluated in Banks and Brindley [2013] and Banks et al. [2013]. During boreal summer over the study region here, the retrieval performs reasonably well, showing good correlation with available colocated Aerosol Robotic Network (AERONET) measurements and a similar spatial pattern to retrievals from instruments in low Earth orbit. There is a suggestion in these studies that in more humid conditions, τ055 may be overestimated, but comparisons with colocated AERONET and aircraft-derived optical depths over the GERBILS period show encouraging levels of agreement given the uncertainties associated with each data source (Figures 1 and 2). In situ τ055 observations from GERBILS have a measurement error of ±20% due to instrumental uncertainties [Johnson and Osborne, 2011]. The corresponding SEVIRI τ055 was determined by colocating the latitude and longitude of the aircraft with the nearest SEVIRI pixel at 1 s time steps for the duration of the aircraft τ055 profile. The mean and standard deviation of all the colocated observations provides the matched τ055 and associated σ. Similarly, the mean SEVIRI τ055 of 5 × 5 pixels surrounding the AERONET locations is compared with AERONET observations averaged over ±30 min centered on the SEVIRI retrieval time step. Thus, the observation standard deviations shown include both spatial and temporal variation. Note that the uncertainties indicate that for some points, the relative uncertainty is of the order 100%.

Figure 1.

Comparisons of colocated SEVIRI retrieved τ055 with GERBILS flight τ055. Vertical bars indicate ±σ of SEVIRI data, and horizontal bars correspond to ±20% of GERBILS τ055. Dashed line represents 1:1, and solid line represents best fit.

Figure 2.

Colocated AERONET τ055 versus τ055 derived from SEVIRI observations during GERBILS observation period. Vertical and horizontal error bars indicate spatial and temporal ±σ of τ055, dashed line represents 1:1, and red line represents best fit.

2.3 DRESW Methodology

The mineral dust aerosol DRESW is defined as the difference between the GERB planetary albedo observed for a given τ055 and the corresponding pristine sky planetary albedo (αprist). To determine αprist for each time step and location, linear regression is performed on pairs of colocated GERB albedo and SEVIRI τ055 retrievals from the 10 day study period using a method similar to Chen et al. [2009], who used Multiangle Imaging Spectroradiometer (MISR) data to determine the TOA albedo change induced by aerosols over different land surface types. To provide enough valid data pairs over the short observing period while also minimizing the effect of spatial inhomogeneities in the surface albedo over North Africa, the regression is performed over 0.25°× 0.25° latitude-longitude boxes.

Regressions are only performed on boxes containing 10 or more data pairs with τ055< 0.5 provided that the τ055 dynamic range exceeds 0.15. This confines the regression to the linear regime between τ055 and TOA albedo. To be considered successful, the root-mean-square (RMS) error of the regression must either be smaller than 0.025 or the absolute correlation coefficient must exceed 0.5. If these conditions are met, the y intercept of the regression line is taken as αprist. Missing boxes where the regression is unsuccessful or there are insufficient points to perform the regression due to persistent cloud cover are filled via interpolation of αprist to the appropriate θz from boxes that show the same mean climatological surface reflectance, Rs [Derrien and Le Gléau, 2005]. This results in 99.5% of boxes being filled at all times.

Figure 3 indicates that αprist at a given location derived in this manner increases with θz in agreement with the study of Wang et al. [2005]. This behavior is shown more clearly for a specific surface reflectance in Figure 4, where αprist is found to increase by ∼ 0.1 as θz increases from 0° to 70°, which is also typical for other surface reflectance values. The change in planetary albedo due to the presence of dust is calculated for each grid box by taking the difference αpristαGERB where αGERB is the instantaneous measured GERB SW albedo averaged over the box from all pixels where τ055 is available. This value is multiplied by the incident TOA solar flux averaged over the box to quantify the DRESW in W m−2.

Figure 3.

(top–bottom) Pristine sky albedo at 0900, 1200, and 1500 UTC derived using linear regression method using pairs of τ055 and GERB SW albedo observations as described in text. Concentric circles denote solar zenith angle contours at 10° intervals.

Figure 4.

Mean pristine sky albedo as solar zenith angle increases from 0° to 70° for grid boxes with surface reflectance = 32%. Error bars indicate ±σ for the bin. Δαp indicates the difference between max and min pristine sky albedo.

2.4 DRELW Methodology

The method used to isolate DRELW due to mineral dust aerosol is described in Brindley [2007], and it follows a similar approach to the τ055 retrieval outlined in section 2.2. For each pixel and time step, GERB longwave flux observations corresponding to the most pristine TB108 selected within the 16 day running window are denoted as the dust-free LW flux, FLWdf. An atmospheric correction, ΔFLWp, is then applied to account for changes in the skin temperature and also the upper and lower tropospheric humidities between the retrieval and pristine day as determined from ERA-Interim data.

Typically, when τ055 exceeds 0.5, the mean atmospheric correction applied to the LW flux signal is 33% of the final DRELW. However, when τ055 is below 0.5, this contribution rises to 64%, making the results more sensitive to errors in the representation of the temporal behavior of the atmospheric state [see Brindley, 2007].

DRELW is then estimated as the difference between the corrected pristine LW flux and the measured LW flux, FLW.

display math(2)

DRELW is initially calculated for each pixel and then gridded to the same 0.25°×0.25° latitude-longitude resolution to match the DRESW resolution.

3 Results and Discussion

3.1 TOA LW Flux Sensitivity to Dust Loading Over Land

The main factors which would be expected to affect DRELW for a given dust loading are dust layer altitude, underlying surface temperature (Ts), dust size distribution, and dust composition. The atmospheric conditions will also significantly modulate the outgoing longwave radiation (OLR), especially in instances of low dust loading. The spatial pattern of mean DRELW (Figure 5) shows a reasonably strong agreement with the mean τ055 (Figure 6). Figure 6 also suggests a modulation of the mean DRELW by Ts with regions of high DRELW moving westward at later UTC times despite a similar spatial distribution of mean τ055. The magnitude of the mean LW TOA warming effect is greatest at 1200 UTC (Table 1) corresponding to local noon over West Africa when surface temperatures there are at a maximum.

Figure 5.

(top–bottom) Mean LW DRE for 20–29 June 2007 at 0900, 1200, and 1500 UTC. Color bar denotes LW DRE in W m−2 where negative values indicate cooling at the TOA, and positive indicate warming.

Figure 6.

(top–bottom) Mean τ055 for 20–29 June 2007 at 0900, 1200, and 1500 UTC. Color bar denotes τ055.

Table 1. Mean DRESW, DRELW, and DRENet for 20–29 June 2007 at 0900, 1200, and 1500UTC in W m−2 Across Whole Observing Region and the Corresponding Mean AOD at 550 nm
AOD (550nm)0.450.450.42
DRESW (W m−2)0.873.290.43
DRELW (W m−2)9.4910.77.21
DRENet (W m−2)10.314.18.14

The relationship between DRELW and τ055 is more clearly exemplified by Figure 7 (top) which shows binned DRELW as a function of τ055 derived for each time slot from all points that provided a valid τ055 across the study region over the 10 day period. The lines indicate quasi-linear behavior at low τ055 (between 0.0 and 1) before flattening off at higher values in agreement with modeled results shown by Hansell et al. [2012]. Here the impact of τ055 on OLR saturates as the emission is either fully coming from the dust layer itself or, in opaque spectral regions, from the atmosphere above the dust. The shading shows the standard deviation in DRELW seen within each τ055 bin at 0900 UTC. Given that the plot is a composite for the whole region, the scatter is perhaps not surprising, implicitly containing the effects of changes in dust composition, size, and height. The increased scatter at higher τ055 is a reflection both of the greater sensitivity to these factors as atmospheric dust loading increases and the reduced sample size (Figure 7, bottom).

Figure 7.

(top) Mean DRELW versus τ055 for bin size Δτ055= 0.05. The grey shaded area marks 1σ in each bin up to τ055 = 3 at 0900UTC; 1200 and 1500 UTC show similar σ. (bottom) Number of observations in each τ055 bin at 0900 UTC (other times show similar distribution).

Our results are broadly similar to previous studies that have attempted to quantify DRELW over north Africa. We note here that differing methodologies are employed in these studies and that they cover different time periods and regions, so an exact match should not be expected. Hsu et al. [2000] use a combination of data from the Tropospheric Ozone Monitoring Spectrometer (TOMS) and the Earth's Radiation Budget Experiment instrument to estimate a monthly averaged instantaneous LW effect of dust ranging from 0 to 33 W m−2 over the region 10–35°N, −20–40°E for July 1985. Zhang and Christopher [2003] use aerosol optical depth (AOD) retrievals from MISR coupled with fluxes derived from the Clouds and the Earth's Radiant Energy System (CERES) instrument to determine an instantaneous monthly mean DRELW of 7 W m−2 for September 2000 over the area 15–35°N, −18–40°E. Over smaller subregions the monthly mean value varies from −1 to 15 W m−2. Finally, Yang et al. [2009] estimated an area mean instantaneous DRELW of 11.4 W m−2 over the region 15–30°N, −10–30°E using 2 years of TOMS, MISR, and CERES data covering the months of June–September. None of these studies explicitly attempts to account for the effects of variability in surface temperature and atmospheric moisture although Hsu et al. do limit observations to those which show a precipitable water of 2 cm or below, and Yang et al. attempt to base their regression between AOD and DRELW on typical Saharan summertime conditions.

Perhaps surprisingly, Figure 6 indicates regions of cooling at all time steps, mainly occurring along the northern coasts of Algeria, Libya, and Egypt where the mean τ055 is less than 0.2. This low dust loading would be expected to exert a relatively small influence on the OLR, and thus, the atmospheric conditions and therefore the atmospheric correction to the flux signal are likely to be dominating the estimated change in LW flux. The negative sign suggests that this correction is consistently too large in this region, although the magnitude of DRELW is smaller than the expected grid box uncertainty (see section 4). We also note that SEVIRI τ055 estimates are themselves more uncertain at low dust loading [Banks and Brindley, 2013]. Studies by both Haywood et al. [2005] and Allan et al. [2011] also find negative OLR differences along the north of Africa when comparing cloud-screened model output with observations. Since this was not the focus of either of these studies, mechanisms for this negative difference were not proposed, but a possible candidate is the representation of the water vapor profile within the model simulations.

A strong positive warming is also observed in the western Sahara at 1200 UTC and 1500 UTC despite only a small mean τ055. Significant differences in regional dust composition across northern Africa have been noted by Schuster et al. [2012] who found an illite concentration of 80% in western Saharan dust and only 10% in eastern Saharan dust resulting in different real parts (1.45 to 1.55) of the complex refractive index. Variations in the refractive index will alter the optical properties of dust, and it is possible that this could affect both the broadband LW flux and the quality of the retrieved τ055. While an assessment of the sensitivity of the retrieval algorithm to the choice of refractive index has been performed [Ansell, 2013] and shows relatively little sensitivity, this was in the context of the GERBILS and AERONET observations shown in Figures 1 and 2, which do not sample the eastern Sahara. A more complete evaluation study is currently in progress which should help to address this question more comprehensively.

Finally, the vertical distribution of dust, particularly in relation to the water vapor profile, and the potential variation of dust properties with height will both influence the DRELW and as such is an area which merits further investigation. This would require very well constrained observations, for example, similar to those available from SAMUM [e.g., Köhler et al., 2011]. GERBILS did not have dedicated ground sites providing a similar level of detailed information, and although some insight may be gained from colocated Cloud-Aerosol Lidar with Orthogonal Polarization observations, the 10 day period considered in this study did not allow enough overpasses that coincided with valid SEVIRI retrievals to make it possible to investigate this further here.

3.2 TOA SW Flux Sensitivity to Dust Loading Over Land

The area averages of all the grid box DRESW values during the GERBILS observing period (Table 1) suggest a negligible overall SW impact on the earth-atmosphere system due to dust aerosol at 0900 and 1500 UTC and only a small positive DRESW at 1200 UTC corresponding to peak insolation over West Africa. This region is where we expect to see maximum dust loading at this time of year [e.g., Ashpole and Washington, 2012].

Similar to previous studies [Patadia et al., 2008], the spatial pattern of DRESW is fairly noisy, and although there is evidence of a relationship between dust loading and DRESW, it is not as strong as that between τ055 and DRELW. This is due to the impact of both surface albedo and θz on the SW radiative effect. Despite the negligible mean values over the whole of the study region, particularly strong mean SW warming of order 10–20 W m−2 is seen in the region around the Bodélé Depression (17°N, 18°E) which is the world's most prolific source of mineral dust to the atmosphere [Knippertz and Todd, 2012]. The mean τ055 around the Bodélé Depression is in excess of 0.6 (Figure 6), and crucially, it has a high pristine sky albedo ∼ 0.4 (Figure 3) such that at these loadings the semitransparent elevated dust aerosol appears “less bright” than the underlying surface. Similarly, distinct SW cooling with DRESW<−20 W m−2 is seen in the region below 18°N and west of 0° which has a relatively low pristine sky albedo of less than 0.25.

Besides the regions identified above, it is difficult to distinguish robust relations between the pristine sky albedo, τ055, and DRESW from the spatial patterns. Therefore, data from the entire observing period (all grid boxes at all time steps) were binned according to τ055 and αprist in an attempt to identify whether a stronger relationship could be seen.

3.2.1 Pristine Sky Planetary Albedo Dependence

The pristine sky albedo is well recognized as strongly modulating the SW effect of mineral dust aerosol. Essentially if the underlying surface albedo appears “darker” than the dust, SW cooling occurs, whereas over “brighter” surfaces the darker dust will result in warming. The critical surface albedo represents the boundary at which the DRESW changes sign for a given dust layer from heating to cooling [Haywood and Shine, 1995].

Figure 8 shows the mean instantaneous dust SW effect expressed in terms of the change in planetary albedo as a function of αprist and τ055. A critical pristine sky planetary albedo (αcrit) is clearly evident at ∼ 0.35 where the addition of extra dust does not result in a marked change in the planetary albedo. Below this value, increasing τ055 leads to greater SW cooling with heating occurring when αprist exceeds 0.35. Due to the presence of the dust layer, αcrit can be considered as the αprist at which the reduction in the surface contribution to upwelling SW flux is balanced by the additional upwelling reflected SW flux. This is in agreement with Yang et al. [2009] who found that while DRESW appeared insignificant across the Sahara desert as a whole, the values implied a slight cooling at surface albedo below 0.32 and a near-zero effect or slight warming at surface albedo above 0.36.

Figure 8.

(top) Change in planetary albedo due to dust presence (αpristαGERB) with τ055. Bin intervals of 0.1 and 0.01 are used for τ055 and αprist respectively, and αprist is binned with an interval of 0.01. The mean of all data from the entire observing period is used. (bottom) Standard deviation of the data in each bin.

The relationship between DRESW and τ055 clearly shows that the amount of dust necessary to cause a significant change in the planetary albedo increases as αprist converges to αcrit either from above or below. For darker (lower) αprist, the addition of only a small amount of dust has an immediate cooling effect as the reduction in SW radiation reflected by the surface and transmitted to the TOA is small relative to the additional contribution from dust layer itself. As αprist increases, the initial reduction in surface contribution increases relative to the additional upwelling contribution from the dust layer, and thus, a larger τ055 is required for the dust layer contribution to dominate the change in planetary albedo and result in SW cooling.

Figures 3 and 4 show that αprist can increase by ∼0.1 at a given location as θz ranges from 0° to 70°. This suggests that the same dust loading could induce heating or cooling over the same region depending on the time of day, as αprist may exceed αcrit at later local times even if this condition is not met near the Sun's zenith. Specific instances of this effect were seen in the instantaneous DRESW for June 2007 (not shown).

3.2.2 Solar Zenith Angle Dependence

The standard deviation of the data in each τ055 and αprist bin does increase markedly with τ055 (see Figure 8) as the dust layer begins to dominate the SW emission. This is likely due to the data not being separated according to θz despite the dust-SW interaction being known to have an angular dependency [Boucher, 1998]. In Figure 9 the instantaneous DRESW-binned data are shown for 10°<θz<20°, and in Figure 10 for 40°<θz<50°; limited data prevent smaller ranges in θz being considered. The bin standard deviations for the limited θz cases decrease markedly at higher τ055 compared to that seen in the entire data set. This indirectly indicates the importance of considering radiative effects of mineral dust aerosol at different local times of day, as opposed to quoting a single SW radiative efficiency.

Figure 9.

Same as in Figure 8 except only observations where θz is between 10° and 20° are included.

Figure 10.

Same as in Figure 8 except only observations where θz is between 40° and 50° are included.

The onset of SW warming of the system occurs at a lower τ055 in Figure 9 for a given αprist compared to Figure 10. For example, at 10°<θz<20°, at αprist∼0.33 the planetary albedo change induced by a τ055∼1.2 is ∼ 0.02, whereas there is only a slight change in planetary albedo (∼0.005) under similar conditions for 40°<θz<50°. Further, strong evidence of the angular dependency of the dust-SW interaction is seen with SW cooling dominating the albedo change at higher θz but being barely evident at lower angles. These findings are in agreement with Pilinis and Li [1998] who noted that when the Sun is close to zenith the predominant forward scattering of larger particles causes the backscatter fraction to decrease, whereas it attains its maximum when the sun is at the horizon.

3.3 Net Direct Radiative Effect

The net direct radiative effect, DRENet, is an overall measure of whether mineral dust aerosol has a heating or cooling effect on the earth-atmosphere system. It is evident from the positive values of mean DRENet across the whole of the observing region and period (see Table 1) that mineral dust aerosol overall acts to warm the system at 0900, 1200, and 1500 UTC with DRELW dominating the radiative effect by reducing the outgoing flux at TOA. The contribution of the SW component to the overall mean radiative balance at the TOA over North Africa and Arabia was negligible during the GERBILS period except at 1200 UTC when it enhanced the LW warming effect. The net DRE is hence clearly strongly modulated by the surface temperature and its influence on LW fluxes as the main regions of heating progress westward with the solar zenith (Figure 11).

Figure 11.

(top–bottom) Mean Net DRE for 20–29 June 2007 at 0900, 1200, and 1500 UTC. Color bar denotes net DRE in W m−2 where negative values indicate cooling at TOA, and positive indicate warming.

While the regional mean effect of dust aerosol is dominated by longwave warming, there are locations where DRESW is the dominant effect at 0900 and 1500 UTC below 16°N in the western Sahara where there is a marked cooling (<−20 W m−2) (compare Figure 8 to Figure 3). This corresponds to a relatively high mean τ055(>0.8) over the GERBILS period, coupled with a low pristine sky albedo (<0.25) at high θz.

4 Uncertainty in DRE Estimates

It is not an easy task to precisely quantify the errors in the final 10 day area mean DRE estimates. The methods used to derive the LW and SW DRE contain a number of sources of uncertainty. In the SW these arise from uncertainty in the linear regression, intrinsic uncertainty in the retrieved SEVIRI τ055 and GERB fluxes, and, for the grid boxes whose αprist could not be generated from a regression, the spatial variability of surface reflectance with respect to the value prescribed to the grid box. For each grid box the method of Isobe et al. [1990] can be used to determine the variance in the y intercept of the regression. The RMS error in the retrieved τ055(∼0.3) is multiplied by the slope of the planetary albedo-τ055 relationship from the regression to estimate its additional uncertainty in the determination of αprist. For each grid box the total error in the instantaneous DRESW is thus due to the uncertainty on αprist and the uncertainty in the instantaneous GERB SW fluxes. Where interpolation is used to estimate αprist, the uncertainty is estimated by considering the standard deviation of the mean surface reflectance prescribed to the grid box and the standard deviation of the interpolated αprist values used to obtain the grid box mean. In the LW, error arises through the correction term used to account for atmospheric variability and the uncertainty in the instantaneous GERB LW fluxes themselves [Brindley, 2007]. Previous work has suggested instantaneous uncertainty in GERB LW and SW fluxes of the order 10 W m−2[Slingo et al., 2006]. This magnitude of uncertainty translates to errors of 15 W m−2 and 10 W m−2in individual grid box estimates of the instantaneous SW and LW DRE, respectively.

However, over longer timescales, comparisons with colocated CERES fluxes indicate overall agreement in the LW to within 2% and to within 5% in the SW on the monthly mean over northern African and the Middle East (Russell, GERB Quality Summary Version 2, available at: http://ggsps.rl.ac.uk/GERBED1_ARG_QS_v2.pdf,2011). In the SW these values tend to reduce with averaging through the diurnal cycle (local time) because of compensating effects from the applied angular distribution models used to convert radiance to flux but do not reduce significantly for a given local time as more days are averaged. For the LW, averaging through the day has less impact, but increasing the number of days included in the averaging period at a given local time does improve the match, because a greater variety of atmospheric conditions are being sampled (N. Clerbaux, personal communication, 2013). Therefore, we take 10% as a realistic estimate of the error in the instantaneous flux values for both channels. This error will, in general, be higher than the absolute estimates quoted above but should ensure that we do not overstate the significance of our final DRE values.

The next question relates to whether the errors are systematic, random, or a combination of both. In reality, the last option is likely to be correct, but diagnosing exactly the split between the two error types is not possible without substantial further study. Therefore, we consider two baseline scenarios and provide error estimates for each case. In (1) we assume that all of the flux error is random and will be reduced by averaging by math formula, where N is the appropriate number of observations. Recall that the DRE is the difference between the observed and pristine flux. Hence, instantaneous errors associated with the former quantity are reduced by a factor of math formula, where Na is the number of area grid boxes and Nt is the number of days that observations were available for a given grid box at the given time slot. In contrast, since the pristine flux does not change with time for a given time slot, associated instantaneous errors will only be reduced by math formula. In (2) we assume all of the flux error is systematic. In this case averaging has no effect, but we are taking the difference between two values with the same bias such that the final estimate will also be subject to the same percentage error. In both (1) and (2) we also include the effect of uncertainty in τ055 on the estimate of pristine sky SW flux. The final error estimates on the 10 day area means of DRESW and DRELW; using each of these methods are shown in Table 2.

Table 2. Errors on 10 Day Area Mean DRESW and DRELW at 0900, 1200, and 1500 UTC Using Methods (1) and (2) as Outlined in Text
DRESW Error (1) (W m−2)0.640.620.74
DRESW Error (2) (W m−2)0.170.410.12
DRELW Error (1) (W m−2)
DRELW Error (2) (W m−2)0.951.40.72

5 Conclusions

This study provides an observationally based estimate of the net direct radiative effect of mineral dust aerosol from geostationary satellite observations, giving new insight into the influence of time of day on the SW, LW, and net effect. Our results indicate that for the GERBILS period and time steps considered here, the net radiative effect is positive and that in the mean sense dust aerosol warms the earth-atmosphere system over North Africa and Middle East with the LW DRE dominating over a negligible SW impact. We find that the regional mean net DRE for 20–29 June 2007 over Arabia and northern Africa is of the order +10 W m−2during sunlit hours.

Decomposing the results spatially, in regions of moderate to high dust loading, DRELW is, as expected, always positive, leading to a heating of the earth-atmosphere system which is clearly modulated by the surface temperature with peak warming at local noon. Comparing the spatial distribution of the mean DRELW to maps of the fraction of dust source activations (DSAs) identified by Schepanski et al. [2009] suggests that regions of the greatest DRELW are associated with areas which have high τ055 but are also significant DSAs or near to significant DSAs. Close to source regions, it is expected that the coarse mode contribution to the dust size distribution would be at its peak, and thus, the dust would be most strongly interacting with LW radiation [Tegen et al., 2006], giving further confidence in our inferred distributions.

While the DRELW dominates the net effect in the mean sense, there are regions where DRESW has the greater magnitude. Consistent with theory, these tend to occur where the mean dust loading is high and the pristine sky albedo is low, leading to a cooling of the system. Our investigation into the DRESW also shows the importance of accounting for variations in pristine sky planetary albedo as a function of time as αprist was found to increase by ∼ 0.1 with θz through the course of the day. This, along with the solar zenith angle dependency of the dust-SW interaction, means that the same τ055 can result in DRESW changing sign over the same location depending on the local time.

There are a number of caveats associated with this study, mainly related to the assumptions contained within the dust optical depth retrieval algorithm. In particular, one fixed dust model is used to translate the signal seen at infrared wavelengths to τ055. Numerous studies have indicated that different sources across the study region would be expected to result in dust emissions having varied composite refractive indices. Variations in the dust size distribution will also change the relationship between SW and LW optical properties [e.g., Ryder et al., 2013]. However, calculations using the range of in situ size distributions sampled during GERBILS [Klaver et al., 2011] and a variety of representative refractive indices showed that no significant improvement could be seen in the agreement between the retrievals and colocated AERONET measurements compared to that obtained using the original standard model [Ansell, 2013]. Similarly, the retrieval, although designed to be as insensitive as possible to dust layer altitude, is expected to have a higher signal to noise when dust is lofted higher in the atmosphere. Variability in both dust layer height and optical properties would therefore be expected to contribute to the scatter seen in the DRE estimates reported here. Nonetheless, the underlying methodology developed and employed in this study gives results which are consistent with theoretical expectations, and further work will seek to exploit this to investigate whether the results seen over the limited time period considered here are perpetuated over the longer record now available from GERB and SEVIRI.