Surface shortwave aerosol radiative forcing during the Atmospheric Radiation Measurement Mobile Facility deployment in Niamey, Niger



[1] The Atmospheric Radiation Measurement (ARM) Program's Mobile Facility (AMF) was deployed to Niamey, Niger, during 2006. Niamey, which is located in sub-Saharan Africa, is affected by both dust and biomass burning emissions. Column aerosol optical properties were derived from multifilter rotating shadowband radiometer, measurements and the vertical distribution of aerosol extinction was derived from a micropulse lidar during the two observed dry seasons (January–April and October–December). Mean aerosol optical depth (AOD) and single scattering albedo (SSA) at 500 nm during January–April were 0.53 ± 0.4 and 0.94 ± 0.05, while during October–December mean AOD and SSA were 0.33 ± 0.25 and 0.99 ± 0.01. Aerosol extinction profiles peaked near 500 m during the January–April period and near 100 m during the October–December period. Broadband shortwave surface fluxes and heating rate profiles were calculated using retrieved aerosol properties. Comparisons for noncloudy periods indicated that the remote sensing retrievals provided a reasonable estimation of the aerosol optical properties, with mean differences between calculated and observed fluxes of less than 5 W m−2 and RMS differences less than 25 W m−2. Sensitivity tests showed that the observed fluxes could be matched with variations of <10% in the inputs to the radiative transfer model. The calculated 24-h averaged SW instantaneous surface aerosol radiative forcing (ARF) was −21.1 ± 14.3 W m−2 and was estimated to account for 80% of the total radiative forcing at the surface. The ARF was larger during January–April (−28.5 ± 13.5 W m−2) than October–December (−11.9 ± 8.9 W m−2).

1. Introduction

[2] Understanding the impact of anthropogenic aerosol radiative forcing is important in predicting future climate change. It is difficult to accurately model aerosol radiative forcing in global climate models because the radiative forcing of aerosol varies considerably with the composition and optical properties of the aerosol, and is therefore strongly regionally dependent. Recent studies have indicated the importance of mineral dust aerosol to climate, due both to direct radiative forcing [Haywood et al., 2003, 2005; Highwood et al., 2003] and to potential influences on tropical cyclogenesis [Dunion and Velden, 2004]. The Sahara is an important source region for mineral dust aerosol; mineral dust plumes originating in Africa frequently travel over the Atlantic Ocean, some reaching as far as the eastern United States [Prospero and Lamb, 2003]. Anthropogenic factors, such as overgrazing in the Sahel region (the southern border of the Sahara), may lead to changes in the amount of mineral dust aerosol, making it an important factor to consider in estimates of anthropogenic climate change [Tegen et al., 1996; Mahowald et al., 2004].

[3] Estimates of the effect of mineral dust on the top of atmosphere (TOA) radiation budget have been made using satellite [Li et al., 2004] and aircraft observations [Haywood et al., 2003; Highwood et al., 2003], but limited observations of aerosol radiative effects in the Sahel have been made at the surface. In 2006, the Atmospheric Radiation Measurement (ARM) program's Mobile Facility (AMF) was deployed to Niamey, Niger for the yearlong Radiative Atmospheric Divergence Using the AMF, GERB Data, and AMMA Stations (RADAGAST) experiment [Miller and Slingo, 2007]. This deployment of the AMF, which is a mobile suite of ground-based remote sensing instruments similar to those found at a fixed ARM site, provides the first extended time series of aerosol and collocated broadband radiation measurements at the ground in the Sahel region.

[4] The combination of the AMF measurements with the Geostationary Earth Radiation Budget (GERB) [Harries et al., 2005] broadband radiometer and the multichannel Spinning Enhanced Visible and Infrared Imager (SEVIRI) [Schmetz et al., 2002] onboard the Meteosat-8 geostationary satellite will allow examination of the radiative divergence of the atmospheric column. Climate models may be tuned to produce agreement with observed globally averaged radiative fluxes at the top of the atmosphere (TOA) from satellite measurements, but they often have trouble matching both the TOA and surface radiation budgets simultaneously [Wild et al., 2006]. Combining the surface and satellite measurements to estimate the radiation budget of the entire column will provide an important constraint for climate model evaluation. As a first step, both the surface and TOA measurements need to be evaluated independently. In this study, we focus on examining the aerosol properties and surface radiative effects calculated from the AMF measurements.

[5] Slingo et al. [2008a] provide an overview of the meteorological and thermodynamic conditions observed at Niamey during the RADAGAST experiment. Niamey has a monsoonal climate with a high-rainfall period occurring in May through September and very little rainfall at other times of the year. During the AMF deployment, the onset of the wet season, which is marked by the northward passage of the Inter-Tropical Front (ITF) through Niamey, occurred on 5 May 2006. The end of the wet season, which is marked by a sudden drop in dew point temperature as the ITF passes southward through Niamey, occurred later than usual, near the end of October [Slingo et al., 2008a]. However, although there was increased moisture in the atmosphere associated with the monsoon season, no precipitation was recorded at the AMF site during the month of October. During the wet season, clouds dominate the radiative forcing at both the surface and TOA, but during the dry season the effects of dust and biomass burning aerosol dominate the surface radiative forcing [Slingo et al., 2008b].

[6] In this study, we focus on quantifying the effect of aerosol on the shortwave radiative fluxes at the surface during the dry season, from 9 January to 31 April and 6 October to 31 December 2006. We assess our ability to reproduce the measured shortwave radiative fluxes at the surface for the range of aerosol conditions observed during the period using aerosol optical properties measured by the AMF and a state of the art radiative transfer model. This type of radiative closure study is important to identify possible shortcomings in aerosol property retrieval algorithms or the treatment of aerosol in current radiative transfer models [Michalsky et al., 2006]. Additionally, we present initial estimates of the vertical profile of shortwave radiative heating in the atmospheric column over Niamey.

2. Data and Methodology

[7] The deployment of the AMF to Niamey is described in detail by Miller and Slingo [2007]. In this section we describe only the particular instruments and retrievals used in our study. The AMF was deployed at the Niamey airport (13.48°N, 2.18°E) from January through December 2006 with the official start of operations occurring 9 January 2006. An ancillary site, containing only radiation and surface meteorological instruments was deployed at Banizoumbou, Niger (13.52°N, 2.63°E). Data from the AMF instruments are freely available from the ARM archive (

[8] The AMF deployment to Niamey was timed to coincide with the special observing periods of the African Multidisciplinary Monsoon Analysis (AMMA) [Redelsperger et al., 2006], an extensive project to study the West African monsoon. Measurements taken as part of AMMA are useful for placing the AMF observations in context. In particular, in situ measurements taken during aircraft flights out of Niamey in January 2006 as part of the Dust and Biomass Experiment (DABEX) [Johnson et al., 2008a, 2008b; Osborne et al., 2008] provide information about the vertical structure of the aerosol composition over Niamey.

2.1. Meteorological Measurements

[9] Vertical profiles of atmospheric temperature and water vapor at 15-min resolution are calculated following the method of Mather et al. [2007] in which vertical profile information from radiosondes (which are launched 4 times/d at AMF site) are combined with more frequent measurements of precipitable water vapor (PWV) from a 2-channel microwave radiometer (MWR) and surface air temperature (from a surface meteorological station). Above 20 km, radiosonde profiles are not valid and information from a standard atmospheric profile is used. The surface skin temperature is obtained from a downward looking infrared thermometer (IRT).

2.2. Broadband Radiation Measurements

[10] Downwelling shortwave (SW) broadband radiation measurements at 1-min resolution were provided by a suite of radiometers. Diffuse broadband SW radiation was measured by an Eppley Model 8–48 black and white pyranometers and direct SW radiation by a normal incidence pyrheliometer (NIP). Total SW downwelling flux is calculated using the component sum method, which consists of combining the measured diffuse irradiance and the direct normal irradiance from the NIP multiplied by the cosine of the solar zenith angle. Upwelling broadband SW irradiance was measured by a downward looking pyranometer, mounted at approximately 2 m above the surface. The ratio of the upwelling to downwelling flux is used to estimate the local broadband albedo at the site.

[11] The radiometers at the AMF were cleaned daily. The best estimates of the accuracy of the downwelling radiation measurements from years of operation during the ARM program are 13.6 W m−2 (direct), 9.0 W m−2 (diffuse) and 9.0 W m−2 (total), as reported by Slingo et al. [2006]. Uncertainties are likely to be larger during unusual circumstances, such as very high turbidity during dust storms.

[12] Detailed spectral information about the surface albedo at Niamey is required for the aerosol optical property retrievals (section 4) and the radiative transfer calculations (section 5). However, this information is not available. We use a spectral albedo of dry sand [Tanre et al., 1986] and scale the profile so that the broadband albedo matches the daily average broadband albedo that is calculated from the ratio of the average upwelling and downwelling fluxes measured by the pyranometers between 1000 and 1600 LST each day. The mean scaling factor used during the study period was 1.00, with a standard deviation of 0.02. During the AMF deployment, a white cooling box was located within the field of view of the downward-looking pyranometers. The effect of this bright white surface within the field of view is expected to increase the broadband albedo by several percent over the actual value. The broadband surface albedo at Niamey varies over the deployment period from an average of 0.25 ± 0.01 during the first observed dry season (January–April) to 0.20 ± 0.02 during the wet season (May–September) and 0.23 ± 0.01 during the second observed dry season (October–December) as the surface vegetation and soil moisture change. The error bars indicate the standard deviation of the albedo measurements. The higher variability in the albedo during the wet season is likely due to the impact of precipitation on the surface albedo: wet and dry soils can have significantly different SW albedos [Idso et al., 1975].

[13] Examination of satellite surface albedo retrievals indicates that the region around Niamey has a very heterogeneous surface albedo and the local-scale albedo derived from the upward- and downward-looking pyranometers may not be representative of the larger area. The local broadband albedo at the Banizoumbou site, which is located approximately 60 km away and is more vegetated than the Niamey site, shows the same temporal variation as the Niamey site during the first part of the year, but is generally 0.05 higher. During the second observed dry season, the Banizoumbou broadband surface albedo increases more rapidly than the Niamey site, perhaps owing to the differing vegetative properties at Banizoumbou, but is on average only 0.03 higher than the Niamey albedo.

2.3. Aerosol Remote Sensing Measurements

[14] Column-integrated values of aerosol optical depth (AOD), Angstrom exponent, and aerosol optical properties are calculated from the MultiFilter Rotating Shadowband Radiometer (MFRSR) data at the Niamey and Banizoumbou sites. The MFRSR measures total and diffuse solar irradiance at six wavelengths (415, 500, 615, 673, 870, 940 nm). Direct solar irradiance is obtained by differencing the total and diffuse measurements, and spectral values of aerosol optical depth are obtained via a Langley regression [Harrison and Michalsky, 1994]. The Angstrom exponent is calculated from the optical depths at 415 and 870 nm. For a well-calibrated MFRSR, uncertainty in retrieved AOD is estimated to be 0.01 [Michalsky et al., 2001; Alexandrov et al., 2007]. A detailed examination of possible sources of error in MFRSR AOD retrievals is given by Alexandrov et al. [2007]. The MFRSR uses a shadow band to block the sun for the diffuse sky measurement, and estimates the solar aureole contribution to the blocked measurement by taking two sideband measurements. For most aerosol conditions, errors in AOD due to the underestimate of the solar aureole contribution are negligible; however, they become more significant for aerosol effective radius >1 μm owing to the larger forward scattering contributions [Alexandrov et al., 2007].

[15] Column-averaged values of aerosol single scattering albedo (SSA) and asymmetry parameter (AP) at MFRSR wavelengths below 940 nm are retrieved from MFRSR observations using the retrieval technique described by Kassianov et al. [2007]. In this retrieval, the aerosol size distribution is assumed to be well modeled by a bimodal lognormal distribution of spherical particles. The widths (variance) of the fine and coarse modes and real part of the refractive index are assumed known. The retrieval is performed in two steps. First, the parameters of the two lognormal distributions (volume median particle radius and volume concentration of each mode) are iterated to match the observed spectral dependence of the AOD. Since the contribution of the coarse mode to the spectral variability of AOD in the MFRSR spectral range (415–870 nm) is relatively small, the retrieval is not sensitive to the coarse mode. Second, the value of the imaginary refractive index is iterated until the calculated spectral values of the ratio of the diffuse to direct irradiances matches the observed values. The second step involves an additional assumption that the spectral values of surface albedo (at MFRSR wavelengths) are known. Kassianov et al. [2007] illustrated that uncertainties in the surface albedo only weakly affect the MFRSR-retrieved SSA and AP values when the surface albedo is relatively small (∼0.1). For Niamey, such small values are observed for two wavelengths (415 and 500 nm). As a result, we expect that surface albedo variations do not substantially change the MFRSR-retrieved SSA and AP values for these two wavelengths. Uncertainties in MFRSR-retrieved SSA and AP values associated with uncertainties in AOD values and size distribution assumptions vary, on average, in 0.03–0.04 range for typical values of AOD (0.1–0.3 at 500 nm) [Kassianov et al., 2007]. However, these uncertainties can be substantially larger for cases with large aerosol loading (e.g., during dust storms).

[16] An additional factor which can contribute to uncertainties in MFRSR-retrieved optical properties of aerosol is nonsphericity of aerosol particles. Since dust particles have complex geometrical shapes [e.g., Kalashnikova and Sokolik, 2004], the influence of particle nonsphericity on aerosol optical properties has been addressed by numerous studies [e.g., Yang et al., 2007, and references therein]. In particular, the modeling results demonstrated that particle nonsphericity can substantially change the scattering phase function, but has little impact on the total optical cross section, SSA and AP [e.g., Mishchenko et al., 1997; Yang et al., 2007]. As a result, the influence of particle shape is important in aerosol retrievals based on multiangular satellite [Kahn et al., 1997] or ground-based observations [Dubovik et al., 2006]. The MFRSR retrieval discussed here involves diffuse irradiance from hemispherical observations. This irradiance is a function of the AOD, SSA and AP, and therefore it should not be very sensitive to details of scattering phase function associated with the particle shapes. The same should be true for the MFRSR-derived aerosol optical properties. However, accounting for the effects of nonspherical aerosol particles in the MFRSR retrieval and estimation of associated uncertainties in retrieved optical properties requires further investigation.

[17] AOD and Angstrom exponent retrievals are performed between 0800 and 1800 LST at 20-s resolution for the entire deployment. The aerosol optical properties (SSA and AP) are retrieved only during the dry periods (January–April and October–December) for times between 1000 and 1600 LST at 1-min resolution. We apply a cloud-screening methodology based on the variability of the retrieved AOD and remove all AOD and optical property retrievals for times identified as cloudy. We interpolate the AOD and optical property measurements over cloudy periods and missing (including nighttime) data to develop a continuous time series.

[18] Information on aerosol optical properties in the longwave spectrum can be retrieved from the Atmospheric Emitted Radiance Interferometer, which was also deployed in Niamey as part of the AMF [Turner, 2008]. In this study we focus only the shortwave optical properties and radiative effects at the surface. In future work we plan to combine the shortwave and longwave measurements to get a consistent picture of the aerosol properties and radiative effects.

2.4. Aerosol Extinction Profiles

[19] Vertical profiles of aerosol extinction were calculated using a micropulse lidar (MPL) and the column AOD retrieved from the MFRSR. The MPL is a 523-nm eye-safe autonomous lidar system [Campbell et al., 2002]. The measured backscatter profile is corrected for dead time, after-pulse, background signal and overlap using standard methods [Campbell et al., 2002]. Large changes in temperature can affect the operation of the MPL by leading to changes in the telescope focus which affect the overlap correction (E. J. Welton, personal communication, 2006). Owing to the environmental conditions expected at Niamey (where the diurnal change in temperature can be as large as 19K), additional temperature control measures were added to the MPL container as part of the AMF deployment. During the Niamey deployment, the mean diurnal variation in laser temperature was <3K, which should not result in significant variations in the overlap correction.

[20] Vertical profiles of aerosol extinction are retrieved from the lidar backscatter measurements following the method of Welton et al. [2000]. MPL backscatter profiles are averaged over 15 min to improve the signal-to-noise ratio. A cloud detection algorithm based on the method of Pal et al. [1992] is implemented to identify clouds in the MPL data. For the noncloudy daytime periods, the value of the extinction-backscatter ratio (also known as the lidar ratio, S) is iterated in the lidar inversion equation [Fernald, 1984] until the AOD calculated from the lidar extinction profile matches the AOD value derived from the MFRSR measurements. The retrieved lidar ratio is interpolated across nighttime and cloudy periods.

[21] Visual inspection of the MPL images and aircraft profiles over Niamey from DABEX in January 2006 indicate that the aerosol observed at Niamey is generally confined to the lowest 7 km of the atmosphere. For nighttime periods or time periods where the MPL detects cirrus cloud above 8 km and there are no AOD retrievals, we calculate the extinction profile by using the interpolated lidar ratio in the Fernald [1984] forward equation. For times where the MPL detects cloud below 8 km, the MPL profile at this time is not used, and the extinction is interpolated in time. The MPL shutter is closed for a half-hour period near local solar noon each day to protect the optics from the nearly overhead sun. The calculated extinction profile is interpolated over these periods.

2.5. Radiative Transfer Calculations

[22] We calculate shortwave broadband fluxes and radiative heating rate profiles at the AMF site in Niamey using the 1D version of the SHDOM radiative transfer model [Evans, 1998] combined with the RRTM correlated k-distribution [Clough et al., 2005; McFarlane and Evans, 2004]. To best match the characteristics of the Eppley radiometers, we use 13 bands in the shortwave, from 0.2 to 3.86 μm. This set of radiative transfer models was chosen because of its high accuracy and the flexibility of the input parameters; the extinction, single-scattering albedo, and Legendre coefficients of the scattering phase function are specified for each wavelength band at each input grid point. For this study, we assume the aerosol phase function can be modeled by a Henyey-Greenstein phase function and we use 8 streams in the radiative transfer calculations.

[23] Since complete information about the spectral variability of the aerosol optical properties is not known, the aerosol optical properties are extrapolated from the 5 MFRSR wavelengths to the center wavelength of each wavelength band used in the radiative transfer model. The AOD is extrapolated using the Angstrom expression τλ = AOD0equation image, where λ is the desired wavelength, λ0 is 415 nm, AOD0 is the AOD at 415 nm, and α is the Angstrom exponent, calculated using the 415 and 870 nm wavelengths. Use of the Angstrom expression to extrapolate AOD beyond measured wavelengths has been shown to give the best fit to observed fluxes [Michalsky et al., 2006]. The extrapolation of single scattering albedo (SSA) and asymmetry parameter (AP) is less well-constrained since the ultraviolet (UV) and near-infrared (NIR) aerosol properties are not known. Many previous studies have assumed that the AP and the SSA or imaginary index of refraction are constant with wavelength [e.g., Kim et al., 2005; Halthore and Schwartz, 2000] (also some models used by Michalsky et al. [2006]). In this study, we extrapolate the SSA and AP to the center wavelengths of the radiative transfer model bands using exponential fits based on the observed wavelength dependence. For the SSA, the exponential increases with increasing wavelength; however the SSA at wavelengths greater than 870 nm is fixed to the MFRSR derived value at 870 nm. For the AP, the exponential fit decreases with increasing wavelength. Given the uncertainties inherent in the extrapolation and the fairly narrow wavelength bands in the model, using the central wavelength rather than a weighted average across the wavelength bands does not greatly affect the calculated optical properties. For cloudy periods, the aerosol properties obtained during adjacent clear-sky periods are interpolated. The vertical distribution of the aerosol is determined by the extinction profile calculated from the MPL as described above.

[24] We perform radiative transfer calculations at 15-min resolution for the periods January–April and October–December 2006 at the Niamey airport site. Two sets of calculations are performed using the same set of temperature and humidity profiles; one with aerosol (AER) and one without aerosol (CLEAR). No clouds are included in the radiative transfer calculations. A sensitivity analysis of the radiative transfer calculations is performed in section 4.2.

3. Aerosol Results

3.1. Aerosol Optical Properties

[25] In this section, we examine the statistics of the time series of aerosol measurements and compare the aerosol properties to those reported in the literature. The time series of daily averaged, cloud-screened AOD (Figure 1) indicates the large variability in aerosol optical depths seen in Niamey, with daily averaged AOD at 500 nm ranging from a low of 0.08 during November to a high of 2.5 during the peak of a dust storm which occurred in March [Slingo et al., 2006]. Peaks in AOD throughout the year, including the broad peak in June, are associated primarily with other dust transport events. Although the data at the Banizoumbou site are more limited, the good correspondence between the time series of retrieved AOD at the two sites, which are located roughly 60 km apart, indicates that much of the daily aerosol variability is controlled by large-scale rather than local influences.

Figure 1.

Time series of daily averaged cloud-screened AOD at 500 nm retrieved from the MFRSR at the Niamey airport and Banizoumbou.

[26] Large variability in the retrieved Angstrom exponent and column-integrated aerosol optical properties is seen throughout the year (Figure 2 and Tables 1 and 2), which is likely due to variability in the relative amounts of dust and biomass burning aerosol in the column. Biomass burning in Africa occurs on a seasonal cycle, related to changes in precipitation. Satellite observations of fire counts and smoke-related aerosols indicate that burning in the Sahel region generally begins in October–November, peaks in January, and then declines throughout the spring, with little biomass burning observed in May through September [Duncan et al., 2003]. Aircraft flights out of the Niamey airport during the DABEX experiment indicated the frequent occurrence of biomass burning aerosol layers existing above dust layers [Johnson et al., 2008a; Osborne et al., 2008]. Back-trajectory analysis [Slingo et al., 2008a] indicated that air parcels observed at 500 m and 1000 m over Niamey during the dry season were primarily from the north and east. Many of these trajectories passed over the Bodele depression in Chad, a major dust source region. During January–April, a large percentage of the air parcels at 4000 m originated from south and east of Niamey (i.e., over Nigeria, a source of biomass burning aerosol) while the trajectories were more variable during October–December. The Angstrom exponent is much higher (indicating smaller particles) during January (Figure 2), although the average values of the Angstrom exponents are similar over the entire January–April and October–December periods (average values of 0.48 and 0.47, respectively). During the January–April part of our study period, when biomass burning was more prevalent near Niamey, the SSA is lower (indicating smaller or more absorbing aerosol) (Table 1). The SSA, AP, and Angstrom parameter show larger variability with wavelength in January–April time period than in the October–December period (Table 2). Note that this variability represents observations with 1-min resolution. Corresponding daily averaged variability is smaller (Figure 2).

Figure 2.

Time series of daily averaged cloud-screened (top) Angstrom exponent, (middle) asymmetry parameter, and (bottom) single scattering albedo retrieved from the MFRSR at the Niamey site. Angstrom exponent retrieved at the Banizoumbou site is also shown in the top panel.

Table 1. Average Retrieved Aerosol Optical Properties During the Two Dry Periods
Wavelength (nm)January−AprilOctober−December
Table 2. Average, Standard Deviation, Minimum, and Maximum of Aerosol Optical Properties at 500 nm During the Two Dry Periods
Angstrom parameter0.480.34−

[27] Estimates of optical properties of mineral dust in the literature show a wide range of values. Previous studies have indicated that optical properties of mineral dust are strongly tied to the mineral composition of the dust and may vary significantly with dust source region [Sokolik and Toon, 1999; Koven and Fung, 2006]. The red color of the dust from visual observations and the strong absorption in the MFRSR blue wavelength (415 nm) relative to the other wavelengths indicates that the dust at Niamey probably contains iron oxides, such as hematite. Alfaro et al. [2004] performed laboratory studies to examine the absorption of desert dust aerosols as a function of iron oxide content. They found that mass absorption efficiencies at 325 nm were approximately 6 times larger than those at 660 nm for samples containing iron oxide and that absorption efficiency increased linearly with iron oxide content. Their soil sample from Niger had the largest iron oxide content (6.5% by mass) of the three desert samples studies. Turner [2008] retrieved dust mineral composition from AERI measurements during the AMF deployment by using the differences in longwave absorption features of several minerals to distinguish between them. However, iron oxides have no spectral absorption features in the 8–13 um band so cannot be identified by the AERI measurements. Turner [2008] found that gypsum and kaolinite were the most common components contributing to the longwave absorption and that the mineral composition of the dust varied with both the stage of the monsoon and the prevailing wind direction.

[28] Carlson and Benjamin [1980] performed an early study of radiative heating of Saharan dust and summarized available measurements of dust refractive indices. They show an order of magnitude range of variability in the imaginary component of the refractive index (which is related to absorption), especially in the 500–700 nm region [Carlson and Benjamin, 1980, Figure 3]. Similarly, Sokolik and Toon [1999] reported visible SSA values from the literature ranging from 0.6 to 0.95 and asymmetry values from 0.65 to 0.94. Fouquart et al. [1987] derived SSA of dust from in situ broadband radiation measurements near Niamey in October–December 1980. They found a mean SSA of 0.95, with a range from 0.93 to 0.99. More recent studies of Saharan dust have reported values in general agreement with the values found by Fouquart et al. [1987]. Kaufman et al. [2001] derived a SSA of 0.97 at 0.64 μm from satellite measurements of Saharan dust over Senegal although slightly smaller SSA (around 0.92) were found from ground-based remote sensing measurements. During the Saharan Dust Experiment (SHADE), which examined aerosol advected off the coast of West Africa, Haywood et al. [2003] measured values of SSA at 550 nm ranging from 0.95 to 0.99 with in situ aircraft instruments. These more recent studies suggest that lower values of SSA found for mineral dust in previous experiments were due to mixing of mineral dust with more absorbing aerosol, such as that produced by biomass burning or pollution.

[29] The range of SSA retrieved from the MFRSR at 500 nm during the January–April period is 0.73 to 1.00 (Table 2), which is consistent with the range of values reported in the literature and is also consistent with values of SSA measured by the aircraft instruments during DABEX [Johnson et al., 2008a; Osborne et al., 2008]. Osborne et al. [2008] reported SSA (at 550 nm) as a function of height from all level aircraft runs within the vicinity of Niamey or Banizoumbou [Osborne et al., 2008, Figure 11]. Values of SSA measured by the in situ aircraft range from 0.73 to 1.00 and show a dependence on height, with lower SSA values (<0.85) associated with biomass burning aerosol or mixed aerosol conditions above 2000 m, intermediate values of SSA between 1000 and 2000 m, and higher SSA values (>0.96) associated with dust below 1000 m. During a flight northeast of Niamey, the aircraft also sampled a “pure” dust layer, which had SSA of 0.99. Since the MFRSR retrieves column-integrated values, we cannot separate out the contributions of dust and biomass burning aerosol to the retrieved optical properties.

[30] During the second observed dry period, when less biomass burning aerosol is expected near Niamey, the average SSA and AP retrieved from the MFRSR are higher than in the January–April period. However, there is still a relatively large range in SSA at 500 nm (from 0.82 to 1.00), which does not agree with the DABEX low-altitude dust measurements. This range in retrieved optical properties from the MFRSR may indicate differences in mineral composition of the dust or the influence of local industrial or cookfire emissions from the Niamey area on the AMF measurements.

3.2. Aerosol Vertical Profiles

[31] The MFRSR only measures column-integrated aerosol properties. However, information on the vertical distribution of aerosol throughout the AMF deployment is available from the MPL. These measurements, along with the other lidar deployments during AMMA [e.g., Heese and Wiegner, 2008], represent the first set of ground-based lidar measurements in the Sahel region. The MPL derived extinction profiles (described in section 2.4) were compared to observations from the in situ aircraft probes for profiles taken with 100 km of the Niamey airport during January–February 2006 [Johnson et al., 2008b]. In these comparisons, the MPL and aircraft extinction profiles generally compared well in the upper layers (above 2000 m) while the aircraft overestimated the extinction relative to the MPL in the layers dominated by dust aerosol. This disagreement may be partly caused by the assumption of a constant lidar ratio in the MPL retrieval algorithm, which may cause the MPL to overestimate the extinction due to biomass burning aerosol (which has a lower lidar ratio than dust) and underestimate the extinction due to dust layers. There is also uncertainty in the aircraft profiles, due to correction factors required for the in situ aircraft probes, especially for the larger particles associated with dust conditions. The overall uncertainty in the aircraft extinction coefficients was estimated as 10% for biomass burning aerosols and 25% for dust aerosol [Johnson et al., 2008b].

[32] Uncertainty in the MPL derived extinction profiles is due both to uncertainty in the MPL signal itself and to uncertainty in the aerosol extinction retrieval algorithm. Uncertainty in the derived MPL signal is described in detail by Welton and Campbell [2002]. For the Niamey deployment, where boundary layer aerosol is of interest, the largest source of uncertainty is the lidar overlap correction. The overlap function is a multiplicative factor that corrects for loss of signal due to poor optical efficiency of the telescope in the near range of the lidar. For the MPL, the overlap correction is significant (>3%) to a vertical distance of about 3.5 km from the lidar. The exact uncertainty in the overlap function used for the Niamey MPL is not available; error analysis by Welton and Campbell [2002] indicates that uncertainty in the overlap correction is typically on the order of 5–10% of the overlap function itself. Sensitivity tests on the Niamey retrievals indicate that an uncertainty of 10% in the overlap correction can correspond to errors of up to ±0.045 km−1 in the derived extinction in the thick aerosol layers below 5 km.

[33] Uncertainty in the MPL extinction profiles associated with the retrieval of extinction from the MPL backscatter signal is due primarily to uncertainty in the AOD used to scale the derived profiles (15%; see section 4.2) and the assumption of a constant lidar ratio with height. For cloudy and nighttime periods, there is also uncertainty associated with the interpolation of the lidar ratio over these periods. Given the average AOD of 0.43 during the study period and the tendency for aerosol to be concentrated in the lowest 5 km, the error in extinction due to the uncertainty in derived AOD is estimated to be ±0.013 km−1.

[34] To estimate the uncertainty in the MPL extinction profiles associated with the assumption of a constant lidar ratio, a simple sensitivity test was performed. An idealized aerosol extinction profile was constructed, consisting of a dust aerosol from 0 to 2 km and biomass burning aerosol from 2 to 6 km. The total AOD of the profile was 0.61, with dust aerosol contributing 72% of the extinction and biomass burning aerosol contributing 28% of the extinction. These values are based on the DABEX campaign-averaged profile [Johnson et al., 2008b]. Given the idealized aerosol extinction profile, a standard molecular scattering profile, and the average backscatter-extinction ratios for dust (55 Sr) and biomass burning aerosol (75 Sr) calculated by Heese and Wiegner [2008] from Raman lidar measurements during the DABEX campaign, a theoretical backscattering profile was calculated. This profile was input to the lidar inversion algorithm, and the extinction profile was retrieved assuming a constant lidar ratio. The resulting retrieved column lidar ratio was 58.8 Sr, close to the assumed value of 55 Sr for the dust aerosol layer, due to the fact that the dust aerosol dominated the total extinction. At the peak of the dust layer (extinction values of 0.58 km−1) the extinction was overestimated by 0.035 km−1, or 6%; while at the peak of the biomass burning aerosol layer (extinction values of 0.09 km−1) the extinction was underestimated by 0.016 km−1, or 18%. This sensitivity test illustrates that the vertical extinction is sensitive to the use of a constant lidar ratio, but the uncertainty is larger for the biomass burning layer since the dust tends to dominate the extinction in the column. Given the above analysis, and assuming independent errors, the total uncertainty in the derived extinction profiles is estimated to be 0.093 km−1 in the dust layers and slightly less in the biomass burning layers.

[35] A curtain plot of cloud-screened MPL-derived aerosol extinction during the dry season is shown in Figure 3. During the first observed dry period there is significantly more variability above 2 km, which is associated with the frequent presence of biomass burning aerosol layers. The few points with very high aerosol extinction above 4 km are likely thin clouds that were not caught by the cloud-screening algorithm. The background AOD level is higher in the January–April period than in the October–December period, although both periods show frequent spikes in AOD associated with dust events. The MPL extinction profiles show clear seasonal differences in the vertical distribution of aerosol (Figure 4). Large dust storms occurred in both dry periods, giving March, April, and December the largest average extinction values. During the early dry period, the extinction profiles peak near 500 m in all months while in the later period the aerosol extinction profiles peak much closer to the surface, near 100 m. Analysis of the radiosondes from the AMF [Slingo et al., 2008a] indicates that the mixed layer is often deeper in the beginning of the year which may result in the aerosol mixing higher in the boundary layer.

Figure 3.

Curtain plot of cloud-screened aerosol extinction derived from MPL at Niamey. In periods of very optically thick aerosol (such as day 65), lidar is attenuated and cannot see the entire column.

Figure 4.

Average vertical profile of cloud-screened aerosol extinction derived from MPL for the entire study period (thick black line) and for each month separately.

4. Aerosol Surface Radiative Effects

4.1. Case Study

[36] To illustrate the effects of aerosol at Niamey on the SW surface fluxes and heating rates, we examine the results for a case study on 21 January 2006. During the dry season at Niamey, the typical atmospheric structure consisted of multiple atmospheric layers in the midtroposphere, often with large humidity differences between layers [Slingo et al., 2008a]. The strong solar heating near the surface often created a deep, well-mixed layer during the daytime. The 21 January case exemplifies this structure. Aircraft profiles during the DABEX experiment indicated that the total aerosol extinction in the atmospheric column over Niamey on 21 January was dominated by dust aerosol although biomass burning aerosol existed above 2 km (Ben Johnson, personal communication, 2007). The corrected backscatter and extinction profiles derived from the MPL for the daylight hours of 21 January show that the majority of the aerosol loading and extinction is below 1 km, although significant amounts of aerosol exist up to 5 km (Figure 5). Considerable vertical structure is seen in the aerosol layers throughout the day. The aerosol optical properties were fairly constant over the day. The AOD at 500 nm ranged from 0.45 to 0.53 between 0630 and 1715 UTC. The average values of the single-scattering albedo and asymmetry parameter at 500 nm were 0.89 and 0.68, respectively, and both of these parameters had standard deviations less than 0.01 over the daylight period.

Figure 5.

(a) MPL corrected backscatter (relative units) and (b) MPL-derived extinction profile for 21 January case study. Missing backscatter data near 1200 UTC is due to closure of the MPL shutter to protect the optics; extinction profiles are interpolated over this period.

[37] Good agreement is seen between the observed (OBS) fluxes and the calculated AER fluxes for this case study (Figure 6 and Table 3). Average differences, given as AER-OBS for the 0630–1715 UTC period are 4.1 W m−2 for the direct flux (1.4% of the average observed flux), 11.6 W m−2 for the diffuse (6.0% of observed), and 15.6 W m−2 total flux (3.1% of observed). The good agreement between the calculated and observed fluxes indicates that the input parameters to the radiative transfer model are a reasonable approximation of the atmospheric state and aerosol properties.

Figure 6.

Observed and calculated (a) shortwave total, (b) diffuse, and (c) direct fluxes at the surface for the 21 January case study. AER calculations include aerosol but no clouds. CLEAR calculations use the same atmospheric state inputs but have no aerosol.

Table 3. Statistics of Calculated and Observed Fluxes for 21 January Case Studya
 Total SW Flux (W m−2)Diffuse SW Flux (W m−2)Direct SW Flux (W m−2)
  • a

    All averages are taken over daylight times (solar zenith angle > 0°).

Mean observed497.0193.4303.6
Mean AER512.6205.0307.7
Mean CLEAR588.549.5539.0
Mean difference (AER-OBS)15.611.64.1
Maximum difference ∣AER-OBS∣28.730.013.7

[38] These flux differences are consistent with values seen in previous aerosol radiative flux closure studies [Halthore and Schwartz, 2000; Henzing et al., 2004], despite the considerably larger aerosol optical depths seen at Niamey. A more recent study, performed during a field experiment at the ARM Southern Great Plains (SGP) site in Oklahoma found smaller flux differences, with instantaneous biases in direct flux less than 1% and diffuse flux less than 2% [Michalsky et al., 2006]. However, in the SGP study the radiometers were well-calibrated before the study and the aerosol properties were well characterized by in situ absorption and scattering measurements.

[39] The calculated SW heating rate for the 21 January case study is shown in Figure 7a. The SW heating is concentrated in several layers in the troposphere, corresponding to moister layers identified from the sounding data (Figure 8). The SW heating is strongest in layers between 3 and 4 km and below 1 km, corresponding to the layers of strong aerosol extinction (Figure 5b). Above 5 km, SW heating is due primarily to water vapor absorption which begins to drop off above 8 km as the water vapor mixing ratio and effectiveness of water vapor absorption decrease. To further examine the effect of aerosol on the calculated heating rate, we look at the difference between the AER and CLEAR calculations (Figure 7b). As expected, the strongest aerosol heating is associated with the layers of high aerosol extinction. The AER calculation assumed a constant lidar ratio for the entire aerosol column. In reality, we expect the biomass burning aerosol to be more absorbing than the dust aerosol, which would result in more absorption in the 3–4 km layer and reduce the absorption in the 0–1 km layer. At high solar zenith angles (before 0900 LST and after 1700 LST), the aerosol decreases the heating below the top of the aerosol layer due to the reflection of SW radiation from the top of the aerosol layer and the associated decrease in water vapor absorption in the lowest 2 km. The aerosol also slightly increases the absorption in the clear sky region above the aerosol layer because of reflection from the top of the aerosol layer and the resulting increased water vapor absorption. Because of the bright surface and dry conditions at Niamey in January (PWV ∼ 2.0 cm) this increase in absorption above the aerosol layer due to reflection from the aerosol layer is very small (<0.05 K/d per 100 m layer) and not clearly seen in Figure 7b.

Figure 7.

(a) Calculated shortwave heating rate profile for 21 January case study. (b) Percentage difference in aerosol and clear-sky shortwave heating rate profile, illustrating effect of aerosol on the shortwave heating. Difference is defined as 100 × (AER-CLEAR)/CLEAR.

Figure 8.

Average (a) temperature and (b) water vapor mixing ratio profiles derived from radiosonde data from 21 January 2006.

4.2. Uncertainty in Radiative Transfer Calculations

[40] In this section we examine sources of uncertainty in the radiative transfer modeling, which include uncertainty in the input parameters (surface albedo, AOD, AP, SSA, PWV, ozone) and use of the Henyey-Greenstein (HG) phase function in the flux calculations. Uncertainties in the aerosol optical properties come both from possible errors in the MFRSR-retrieved values and from extrapolation of the retrieved values to the wavelengths used in the RRTM model. Given fixed column AOD and average optical properties, uncertainties in the vertical distribution of extinction (from the MPL profiles) will affect the heating rate calculations, but do not significantly affect the calculated SW fluxes at the surface, although they would have larger impact on TOA fluxes [Johnson et al., 2008b].

[41] First, we extend the case-study analysis to compare the calculated and observed fluxes during the entire study period. All of the noncloudy 15-min periods are identified using the MPL and MFRSR cloud-screening methodologies. The calculated and observed downwelling surface fluxes show good agreement during noncloudy periods, with mean differences of 2% or less, indicating confidence in the retrieved aerosol properties and radiative transfer calculations (Table 4).

Table 4. Statistics of Observed and Calculated Downwelling SW Fluxes at the Surface for Noncloudy 15-min Periodsa
 All (975)January−April (324)October−December (651)
  • a

    Number of 15-min periods included in each set of statistics is given in parentheses. All fluxes and flux differences are given in W m−2. Mean flux difference is given as AER-OBS.

Mean observed flux689.8450.2239.6713.8383.0330.8677.9483.6194.2
Mean flux difference−2.11.0−3.2−0.9−7.97.0−2.75.5−8.2
RMS flux difference9.614.013.811.915.

[42] In the flux calculations, a Henyey-Greenstein (HG) phase function (which depends only on the asymmetry parameter) is used. The HG phase function generally underestimates the forward scattering peak relative to a full Mie calculation and can affect the calculated aerosol radiative forcing [Boucher, 1998]. To test the effect of using the HG phase function, a sensitivity test was performed. For two different cases, representative size distributions and imaginary values of the refractive index were saved from the MFRSR retrievals. The refractive index was extrapolated to the RRTM wavelengths and Mie calculations were performed using the retrieved size distribution to calculate the AP, SSA, and full phase function. Then flux calculations were performed using the calculated SSA and AP and either the full Mie phase function or the HG approximation. For each of the two periods, which had effective radius of 0.68 μm and 0.77 μm, respectively, the use of the full phase function increased the diffuse flux at the surface by less than 1.5 W m−2 compared to the calculations with the HG phase function. Although larger differences between the full phase function and HG phase function are expected for larger particles that are observed during fresh dust events, this analysis indicates that use of the HG phase function is not a large source of uncertainty for this study.

[43] A complication in interpreting both the MFRSR retrievals and the comparisons of calculated and observed fluxes is due to the geometry of the MFRSR and broadband radiometers and the partitioning of the direct and diffuse fluxes. Theoretically, the SW direct flux is that part of the incoming solar radiation that has not been scattered while the diffuse component consists of light that has been scattered at least one. In radiative transfer model calculations, these contributions can be easily separated. However, in observations it is not as straightforward to estimate or remove contributions from the solar aureole, or forward scattering into the instrument field of view. The MFRSR uses a shadow band to block the sun for the diffuse sky measurement, and takes two sideband measurements to estimate the solar aureole contribution to the blocked measurement. As aerosol effective radius increases, and forward scattering contributions become more important, the underestimate of the solar aureole contribution increases [Alexandrov et al., 2007]. The MFRSR essentially measures an “effective” AOD that is smaller than the true AOD because of the contribution of forward scattered light. The SSA may also be underestimated owing to a larger measured direct/diffuse ratio.

[44] Russell et al. [2004] estimate the contribution of forward scattering by dust aerosols to Sun photometer and pyrheliometer measurements as a function of field of view. For instruments with half-angle η = 1.85°, AOD correction factors at 354 nm can be as large as 10%, and for η = 2.8° they can be up to 16%. Comparison of the MFRSR retrieved AODs to values from a narrow field of view (η < 1°) Sun photometer that was collocated at the Niamey site during the latter half of the year, indicate that the MFRSR underestimates AOD by 10–15% compared to the Sun photometer owing to the forward scattering effect. A correction for the effect of large particle scattering on the MFRSR retrievals at Niamey, taking into account the MFRSR shadowband geometry and the aerosol Angstrom coefficient, is under development but is not yet available (C. Flynn, personal communication, 2008).

[45] However, the other radiometric instruments at the AMF are also impacted by the forward scattering of the large aerosol particles. The NIP instrument also has a relatively large field of view, η = 2.8°, and the shading disk for the shaded pyranometer was designed to have the same geometry. Owing to the similar effective fields of view of the NIP, shaded PSP, and MFRSR, the underestimate of AOD from the MFRSR is partially compensated for by an overestimate of direct flux from the NIP and underestimate of diffuse flux from the PSP, which explains the good agreement in the radiative transfer calculations. In the future, if the “effective” AODs measured by the MFRSR are corrected for forward scattering, then the NIP and shaded PSP must also be corrected to examine flux closure of the direct and diffuse components. The issue of correcting the observed direct and diffuse broadband fluxes has been addressed in some cloudy-sky flux closure studies [McFarlane and Evans, 2004], however it has neglected in many aerosol closure studies because nondust aerosol tend to be small enough that the aureole scattering only requires a correction on the order of 0.1% [Halthore et al., 1997].

[46] Finally, we perform a sensitivity analysis to determine the sensitivity of the calculated fluxes to uncertainty in the input parameters, so that uncertainty in the derived radiative forcing values can be estimated. Calculations for the 21 January cases were performed in which each of the above model input parameters was varied by 10%, which is a reasonable estimate of the uncertainty in the retrieved parameters. Results of this sensitivity study are shown in Table 5. The direct flux is affected only by changes in AOD, PWV, and ozone, with AOD having the largest impact. A 10% change in AOD resulted in roughly a 15 W m−2 change in direct flux, indicating that the mean direct flux could be matched by about a 3% reduction in AOD. Changes in SSA had the largest effect on the modeled diffuse flux, with a 10% reduction in SSA causing a 26.8 W m−2 (or 5.2%) reduction in modeled diffuse flux. Changes in asymmetry parameter and AOD also had significant impacts on the diffuse flux. Changes in surface albedo, ozone, and PWV had little effect on the diffuse flux, with changes of less than 1.5 W m−2 for each parameter. The sensitivity tests show that the diffuse flux could be easily matched by changes of about 5% in the retrieved SSA (Table 6).

Table 5. Mean and Standard Deviation of Calculated Daily 24-h Aerosol Radiative Forcing, Aerosol Radiative Forcing Efficiency, and Total Radiative Forcinga
 Study PeriodJanuary−AprilOctober−December
  • a

    Aerosol radiative forcing, ARF; aerosol radiative forcing efficiency, ARFE; and total radiative forcing, TRF.

ARF (W m−2)−21.1 ± 14.3−28.5 ± 13.5−11.9 ± 8.9
ARFE (W m−2τ−1)−48.0 ± 23.4−56.3 ± 10.1−37.6 ± 30.3
TRF (W m−2)−26.3 ± 19.4−34.9 ± 19.2−15.5 ± 13.4
Table 6. Average Fluxes Over Daylight Period for 21 January Sensitivity Studya
 ParameterChange in ParameterSW Total (W m−2)SW Diffuse (W m−2)SW Direct (W m−2)
  • a

    Calculated fluxes based on changing a single parameter by the stated amount are given. Difference in sensitivity test fluxes minus original AER calculated fluxes is given in parentheses for each sensitivity test.

Observed flux  497.0193.4303.6
Original AER calculated flux  512.6205.0307.7
Sensitivity testsaPWV+10%509.9 (−2.7)204.1 (−0.9)305.8 (−1.9)
  −10%515.5 (2.9)205.8 (0.8)309.7 (2.0)
 Ozone+10%511.7 (−0.9)204.4 (−0.6)307.2 (−0.5)
  −10%513.6 (1.0)205.5 (0.5)308.1 (0.4)
 Surface albedo+10%514.1 (1.5)206.5 (1.5)307.7 (0)
  −10%511.1 (−1.5)203.5 (−1.5)307.7 (0)
 AOD+10%505.9 (−6.7)213.8 (8.8)292.1 (−15.6)
  −10%519.5 (6.9)195.3 (−9.7)324.2 (16.5)
 AP+10%519.4 (6.8)211.8 (6.8)307.7 (0)
  −10%506.2 (−6.4)198.5 (−6.5)307.7 (0)
 SSA+10%535.4 (22.8)227.8 (22.8)307.7 (0)
  −10485.8 (−26.8)178.1 (−26.8)307.7 (0)
  −5498.9 (−13.7)191.2 (−13.7)307.7 (0)

[47] The flux closure results (Table 4) illustrate that the error in the SW fluxes for noncloudy time periods are <4 W m−2 on average, although the RMS flux difference of 14.0 W m−2 indicates that individual time periods can have larger errors. The results of the sensitivity tests (Table 6), which used realistic estimates of uncertainty in the model input parameters, indicate that maximum errors in the diurnal (24-h) averaged fluxes are on the order of ±12.5 W m−2. These uncertainty estimates should be considered when interpreting the estimated effects of aerosol on the surface radiative fluxes in section 4.3.

4.3. Aerosol Surface Radiative Forcing

[48] In this section, we examine the effects of aerosol on the surface radiative fluxes at Niamey. We first examine the change in transmitted surface fluxes due to the aerosol by subtracting the calculated CLEAR downwelling flux from the observed downwelling flux at the surface during noncloudy periods and normalizing by the incoming SW flux at the TOA (Figure 9). Aerosol reduces the direct SW flux at the surface relative to the CLEAR calculation because of the increased extinction. However, the increased scattering due to the aerosol layer increases the SW diffuse flux, which partially compensates for the reduced direct flux. In general, the direct effect is larger than the diffuse effect and the total SW radiative effect at the surface due to aerosol is negative. Positive total SW radiative effects are seen for some cases of small AOD (Figure 9c). These instances may be due to improper specification of the atmospheric state properties in the radiative transfer models or to errors in the observed fluxes. More cases with calculated positive aerosol radiative effects are observed during the October–December period, which may indicate issues with the instrumentation near the end of the field experiment. The change in transmittance is correlated strongly with the observed AOD, as expected. For the largest AOD values, aerosol reduces the total transmission at the surface by 30% relative to the CLEAR calculation.

Figure 9.

Change in surface transmittance due to aerosols during noncloudy periods for (a) direct SW, (b) diffuse SW, and (c) total SW. Also shown is (d) aerosol radiative forcing (ARF) for noncloudy periods. Linear fits are calculated for the January–April (black) and October–December (blue) periods, with coefficients given in the plots.

[49] For comparison to other studies, we also calculate the instantaneous surface aerosol radiative forcing (ARF), defined as the difference between the net flux (downwelling minus upwelling) at the surface and the same quantity with no aerosol.

[50] ARF = (Fsfc,aerFsfc,aer) − (Fsfc,bgFsfc,bg), where F and F are the downwelling and upwelling fluxes at the surface, respectively, “aer” denotes aerosol conditons and “bg” denotes background conditions with no aerosol [Christopher et al., 2003]. We calculate the ARF by subtracting the calculated CLEAR net flux at the surface from the observed net flux (Figure 9d). The maximum ARF, which occurred during the March dust storm, was −270.9 W m−2. The average ARF during the noncloudy periods is −38.5 W m−2, with a standard deviation of 38.4 W m−2, indicating the large variability in radiative forcing.

[51] The frequent presence of cirrus clouds during the dry season makes it difficult to examine the daily averaged effects of aerosols using the observed fluxes. Therefore we estimate the daily averaged surface ARF (difference in net fluxes at the surface) by subtracting the calculated CLEAR net fluxes from the calculated AER net fluxes (in which the aerosol extinction and optical properties have been interpolated over cloudy periods) instead of from the observed fluxes. This calculation assumes that the aerosol does not vary over the cloudy periods. Since most of the cloudy periods during the dry season are due to high cirrus (above 8 km) we believe this is a reasonable assumption. We also estimate the daily averaged (24 h) total radiative forcing (TRF) (aerosol + clouds) by subtracting the CLEAR calculation from the observed net flux. The average values of the daily surface ARF and TRF are given in Table 5.

[52] A time series of the calculated daily averaged surface ARF is shown in Figure 10. The impact of clouds on the surface radiative forcing can be seen by comparing the black and blue lines in Figure 10. The average calculated surface ARF is −21.1 ± 14.3 W m−2 d−1 while the average calculated total radiative forcing (including clouds and aerosol) is −26.3. ± 19.4 W m−2 d−1. Therefore, during the dry season at Niamey, we estimate that aerosol accounts for about 80% of the SW radiative forcing at the surface on average. The aerosol radiative forcing at the surface is much stronger in the January–April period (−28.5 ± 13.5 W m−2) than in the October–December period (−11.9 ± 8.9 W m−2), owing to the higher AOD and lower SSA in the January–April period. The estimated uncertainties in the calculated daily averaged radiative fluxes (section 4.2) should be considered when interpreting the significance of the ARF estimates.

Figure 10.

Time series of (a) daily-averaged surface SW aerosol radiative forcing (ARF) and aerosol + cloud total SW radiative forcing (TRF). The time series of (b) daily-averaged AOD is included again to illustrate how the radiative forcing is tied to the aerosol loading. Note the discontinuity in the time axis.

[53] Regional estimates of surface SW radiative forcing over North Africa from climate models range from roughly 8 W m−2 for yearly means [Tegen et al., 1996; Woodward, 2001] to −25 to −30 W m−2 for December–February averages [Miller et al., 2004], which are comparable to the values estimated here. This strong surface radiative forcing of the aerosol has implications for the surface energy balance and boundary layer meteorology [Miller et al., 2004, also Seasonal contrasts in the surface energy balance of the Sahel, submitted to Journal of Geophysical Research, 2008]. Reduction of shortwave radiation absorbed at the surface can reduce the surface temperature and decrease the strength of boundary layer mixing. Inability to correctly model the effects of dust on the surface radiation budget in global models can lead to temperature biases and poor surface energy balance, with implications for the hydrological cycle.

[54] Owing to the limited availability of ground-based observations, few constraints on model estimates of surface radiative effects due to mineral dust are available in this region. Zhu et al. [2007] combined satellite estimates of AOD with a radiative transfer model to calculate the seasonal average shortwave ARF of dust over 3 regions. Off of the Saharan coast, they found a regional mean value of −23.0 W m−2 at the surface averaged over the June–August period, which is similar to our value of 21.1 W m−2. Christopher et al. [2003] calculated the surface ARF from ground-based observations of Saharan dust aerosol transported over the Atlantic to Puerto Rico. Their calculated daytime monthly mean shortwave ARF at the surface was −18 W m−2, which is significantly less than the value of −40 W m−2 that we find if we calculate a daytime rather than 24-h average radiative forcing. The difference is due primarily to the reduced AOD (average of 0.26 in their study compared to 0.41 in our study) after transport across the Atlantic, but may also be due to differences in aerosol properties after transport and aging.

[55] We also examine the diurnally averaged surface aerosol radiative forcing efficiency (ARFE), which represents how the aerosol radiative forcing changes with the aerosol loading. In this study, the surface ARFE is estimated as the diurnally averaged surface ARF divided by the AOD at 523 nm. The average surface ARFE is −48.0 ± 23.4 W m−2τ−1. Again, there is strong seasonal variability associated with the aerosol properties, with average values of ARFE of −56.3 ± 10.1 W m−2τ−1 in the January–April period and −37.6 ± 30.3 W m−2τ−1 in the October–December period. Li et al. [2004] also saw strong seasonal variability in the radiative forcing of Saharan aerosol over the North Atlantic, although their ARFE values are larger than those seen in the current study. Their estimates of surface ARFE from satellite measurements were −65 W m−2τ−1 for June–August (high dust period in their study) and −81 W m−2τ−1 for November–January (low dust period which was more influenced by biomass burning aerosol from equatorial Africa). Their larger ARFE values may be due to the assumptions required to derive surface ARFE from satellite measurements or to the fact that they were estimating surface ARFE for a dark surface (ocean). Zhou et al. [2005] estimate surface ARFE from ground-based aerosol measurements at several locations influenced by mineral dust around the globe. They report average values of surface ARFE of −80 W m−2τ−1 for mineral dust over dark surfaces (broadband surface albedo <0.1) and average ARFE of −48 W m−2τ−1 for sites with surface albedo between 0.3 and 0.35 (including the Banizoumbou site). East Asian dust, which is often mixed with urban aerosol from China, has stronger surface ARFE values than seen at Niamey, with average values ranging from −91 W m−2τ−1 to −106 W m−2τ−1 for dust cases at three ground sites [Kim et al., 2005].

4.4. Vertical Profiles of Aerosol Radiative Heating

[56] Over land, close to the aerosol source regions, dust is generally found in the boundary layer. However, as Saharan dust is transported off the coast of West Africa by the easterly winds, it is generally elevated as the dry, dusty air layer is undercut by cool, moist maritime air [Prospero and Carlson, 1981]. Thus vertical profiles of transported dust may be quite different from those over land, closer to the dust source regions. Optical properties may also be quite different owing to effects of transport and aging. There has been little examination of vertical profiles of aerosol and radiative heating in the Sahel, because of the lack of vertically resolved measurements. However, along with the AMF deployment, recent observations from a field campaign in southern Morocco [e.g., Ansmann et al., 2008] and from lidars deployed during the AMMA experiment [e.g., Heese and Wiegner, 2008] should significantly improve our knowledge of the vertical profiles of aerosol properties and radiative effects over land and close to dust source regions.

[57] Carlson and Benjamin [1980] calculated heating rates for idealized Saharan dust profiles over the desert and found SW heating (relative to clear sky) of 1.2 K/d with a fairly uniform heating profile from 500 to 1000 mb for an AOD of 1.0. Fouquart et al. [1987] used aircraft measurements of dust properties to calculate radiative heating profiles over Niamey. They found strong SW heating due to the dust in the lowest 2 km of the atmosphere, with a maximum value of 5 K/d (relative to clear sky) near noon on a day with AOD = 1.5 and 2 K/d near noon on a day with lower AOD (0.37).

[58] Our results show strong SW heating due to aerosol in the lowest 4 km during the January–April period, with an average SW aerosol effect at 1300 UTC of 2.5 K/d near 500 m and 1 K/d between 2 and 4 km (Figure 11). SW heating due to aerosol was significantly less in the October–December period, with maximum SW aerosol heating of 1 K/d near the surface and little aerosol heating above 2 km.

Figure 11.

Average (a) cloud-screened heating rate and (b) extinction profile at 1300 UTC for January−April and October−December dry periods.

[59] One uncertainty in the calculated heating rate profiles is the lack of knowledge of the vertical variability of the aerosol optical properties. From the MPL alone, discrimination between dust and biomass burning aerosol layers is not possible. Use of multiple lidar wavelengths (a 1064 nm ceilometer was also deployed at the AMF) might allow discrimination between dust and biomass burning aerosol and improve the representation of the vertical profile [Cattrall et al., 2005]. Additionally, combination of the MFRSR, which is more sensitive to the fine mode aerosol, with longwave sensors such as the AERI, which is more sensitive to the coarse mode aerosol, could help identify the relative contribution of dust and biomass burning aerosol to the total column extinction.

5. Summary and Conclusions

[60] The AMF deployment in Niamey, Niger provided an unprecedented opportunity to study aerosol properties and the aerosol surface radiative effects in the Sahel. We used MFRSR measurements to retrieve aerosol optical properties, including optical depth, single scattering albedo, and asymmetry parameter during the dry seasons at Niamey. The colocated MPL allowed us to estimate the vertical distribution of the aerosol extinction. Using the retrieved aerosol properties and vertical extinction profiles, we calculated the broadband fluxes and heating rates at Niamey.

[61] The retrieved aerosol optical properties were compared to values from the literature as well as values from aircraft profiles over Niamey in January–February as part of the DABEX experiment. The single scattering albedo (SSA) showed strong absorption at shorter wavelengths, which decreased with increasing wavelength, indicating the presence of iron oxides in the dust aerosol. The range in single scattering albedo from the MFRSR retrievals agreed well with the DABEX aircraft measurements, although the SSA were significantly lower than in previous observations of Saharan dust. The DABEX flights indicated that the lower SSA values were due to biomass burning aerosol layers that were prevalent during the first part of the year. The SSA values were much higher in the October–December time period, although there were still occasions when they were lower than would be expected for pure dust, which may be due to local combustion sources near Niamey or variations in mineral content. Along with the optical properties, the vertical profiles were also seen to vary between the two study periods, with the aerosol extinction concentrated in a shallower layer during October–December.

[62] Comparison of the calculated surface fluxes to observed fluxes for noncloudy periods indicated that the remote sensing retrievals provided a reasonable estimation of the optical properties. The calculations tended to underestimate the diffuse flux relative to the observations for higher aerosol optical depth (AOD) values. This difference is likely due to the fact that small uncertainties in the aerosol property retrievals are magnified as the aerosol loading increases. Sensitivity tests for the case study calculations showed that we were able to match the observed fluxes with variations of <10% in the inputs to the radiative transfer model. For example, a 5% reduction of SSA brought the calculated diffuse flux in very good agreement (∼1%) with the observed flux.

[63] We calculated the instantaneous direct surface aerosol radiative forcing (ARF) by subtracting the modeled CLEAR fluxes (radiative transfer calculations assuming no aerosol) from the observed net fluxes for each of the 15-min noncloudy periods. The average ARF was −38.5 ± 38.4 W m−2, indicating both the strong effect of the aerosol on the surface radiative budget and the large variability in the radiative effect associated with the large variability in dust loading. The daily average ARF was calculated by subtracting the calculated CLEAR net fluxes from the calculated AER net fluxes (in the AER calculations, aerosol properties were interpolated over cloudy periods). The average daily ARF was −21.1 ± 14.3 W m−2, which is comparable to values estimated from satellite data and to estimates for Northern Africa from climate models with sophisticated dust parameterizations. As with the aerosol optical properties, strong variability was seen in the surface aerosol radiative forcing between the two study periods. The daily average surface aerosol radiative forcing efficiency (ARFE), defined as the ARF divided by the AOD, was −48.0 W m−2τ−1 and also showed strong seasonal variability.


[64] The AMF was deployed to Niamey as part of the RADAGAST project through funding from the Atmospheric Radiation Measurement (ARM) Program, Office of Biological and Environment Research, U.S. Department of Energy. We thank Mark Miller, the AMF lead scientist, and everyone else whose hard work contributed to the successful AMF deployment. We also thank Dave Turner for providing the MWR retrievals and Jim Mather for the merged sounding profiles used in this study. Two anonymous reviewers provided insightful comments which improved the clarity and focus of the manuscript. Funding for this project was provided by the U.S. Department of Energy, Office of Science, as part of the ARM program under contract DE-AC06-76RL01830 to PNNL. PNNL is operated by Battelle for the U.S. Department of Energy. The RADAGAST proposal, which led to the Niamey deployment of the AMF, was largely the work of Anthony Slingo. His untimely death in 2008 has robbed us of a colleague, research collaborator, and dear friend. Tony was a person of great enthusiasm, insight, and wit in everything that he did, and his scientific acumen was an important component of the deployment choices and the aircraft studies during RADAGAST. The articles in this special issue, and further articles to come, are a further tribute to Tony's insights that led to the RADAGAST deployment. We will greatly miss Tony's excitement and enthusiasm in the scientific process, his willingness to share his knowledge and understanding with all of us, and the deep friendship that we had with him.