Simultaneous size distributions and Fourier transform infrared extinction spectra have been measured for several clay components of mineral dust aerosol (illite, kaolinite, and montmorillonite) in the fine particle–sized mode (D = 0.1–1 μm). Published optical constants have been used in combination with the measured size distributions in spectral simulations for comparison to the measured extinction spectrum. In general, the Mie theory simulation does not accurately reproduce the peak position or band shape for the prominent Si—O stretch resonance near 9.5 μm (ca. 1050 cm−1) for any of the clays. The resonance peak in the Mie simulation is consistently blue shifted (27–44 cm−1) relative to the experimental spectrum. Additionally, the integrated absorbance in the resonance band is underpredicted for all three clay compounds. Spectral simulations based on various distributions of ellipsoid-shaped particles better reproduce the experimental spectrum including the peak position, band shape, and integrated absorbance for these fine clay aerosols.
 Mineral dust aerosol, associated with windblown soil, plays an important role in determining the physical and chemical equilibrium of the atmosphere. Aerosols can affect the global radiation balance through the direct scattering and absorption of light; atmospheric dust can exert a negative climate forcing effect by scattering incoming short-wavelength solar radiation, but can also have a positive forcing effect by absorbing outgoing infrared terrestrial radiation. Recent studies have shown that the outgoing IR radiances observed by satellites at the top of the atmosphere are noticeably affected by dust [Ackerman, 1997]. However, the direct radiative forcing effect of atmospheric aerosol, particularly mineral dust aerosol, represents one of the largest uncertainties in our current understanding of climate change; even the sign of the net direct forcing contribution of mineral dust aerosol is not certain [Penner et al., 2001].
 Mineral dust particles in the atmosphere can also affect the global radiation balance through indirect means, by acting as nucleation sites for cloud condensation [DeMott et al., 2003; Rudich et al., 2002], or catalytic surfaces for heterogeneous chemistry that can alter the chemical balance for important atmospheric species such as sulfur and nitrogen oxides [Sullivan et al., 2007; Usher et al., 2003]. Mineral dust aerosol can thus impact temperature profiles, photolysis rates, atmospheric dynamics, chemical cycles, and transport on regional and global scales.
 Modeling the impact of mineral dust on the physical and chemical balance of the atmosphere requires knowledge of the dust loading, as well as information about the aerosol composition, shape, and size distributions, and how these may change as dust is transported and aged. Aerosol optical depth can be measured through remote sensing, including high-resolution or narrow band infrared spectral measurements from satellite instruments such as the advanced very high resolution radiometer (AVHRR), the Moderate Resolution Imaging Spectrometer (MODIS), and the Atmospheric Infrared Sounder (AIRS) [Ackerman, 1997; Pierangelo et al., 2004]. For example, Ackerman  has suggested that various difference measurements between specific narrow band IR sensor channels of MODIS (the BT8, BT11, and BT12 channels centered near 8, 11, and 12 μm, respectively) could be used to indicate the presence of dust. Sokolik  has considered this question in a recent modeling study, and found that dust has a unique spectral signature characterized as a “negative slope” in the 820–920 cm−1 (10.87–12.2 μm) range of the brightness temperature (BT) spectrum, that could be used to determine dust loading. Sokolik  has also suggested that high-resolution studies in the spectral region from 1099 to 1200 cm−1 may be particularly useful for determining dust composition because many of the abundant mineralogical constituents of dust aerosol (including the silicate clays and quartz) have characteristic vibrational resonance features in this region.
 The IR region is also of particular importance because satellite measurements determine key atmospheric and oceanic properties such as the atmospheric temperature profile, water vapor and trace gas concentrations, and sea surface temperature through IR spectral studies of the Earth's radiative emission [Sokolik, 2002]. Under moderate or high–dust loading conditions, the radiative effect of atmospheric dust must be accounted for in the spectral measurements to accurately determine these properties [DeSouza-Machado et al., 2006].
 Mie theory is commonly used for modeling the optical properties of aerosols in both radiative forcing calculations and satellite data retrieval algorithms [Conant et al., 2003; DeSouza-Machado et al., 2006; Moffet and Prather, 2005; Wang et al., 2002]. However, Mie theory is strictly valid only for spheres. Atmospheric aerosol particles are generally highly irregular in shape which can have a significant effect on the optical properties. It is known that Mie theory does a poor job in predicting the resonance line positions and band shapes for nonspherical particles, even for small particles that fall in the Rayleigh regime, D ≪ λ, where D is the particle diameter and λ the wavelength of incident light [Bohren and Huffman, 1983]. This could lead to errors in aerosol retrievals depending on the spectral overlap between the actual and calculated line profiles, and the specific narrow band IR sensor channels used to determine aerosol optical depth and dust composition.
 Simple analytic relations have been derived in the small particle (Rayleigh) limit to model resonance absorption cross sections for particles with characteristic shapes such as ellipsoids, disks, and needles [Bohren and Huffman, 1983]. The analytic solutions for the absorption due to a continuous distribution of ellipsoidal (CDE), disk-shaped, or needle-shaped particles are given in equations (1) through (3) [Bohren and Huffman, 1983]:
where ɛ = ɛ′ + iɛ″, is the complex dielectric constant, k = (2π/λ), and v is the volume of the particle, related to the diameter of an equivalent volume spherical particle by v = (πD3/6). Note that these analytic solutions are derived under the assumption that X = (πD/λ) ≪ 1 and ∣m∣X ≪ 1, where ∣m∣ is the magnitude of the complex index of refraction [Bohren and Huffman, 1983]. Results for the “continuous distribution of ellipsoids (CDE)” model for particle shape are commonly used in the astronomy literature to simulate absorption spectra for cosmic dust particles, which are thought to be small [Fabian et al., 2001]. This raises several interesting questions. Is such an approach useful for atmospheric mineral dust particles? What simple particle shapes might best characterize the optical properties of atmospheric mineral dust? In what size range does the small particle limit break down?
 To quantitatively investigate, on a fundamental level, the appropriateness of various models of light extinction by atmospheric aerosols, an instrument capable of simultaneously measuring size distributions and infrared extinction spectra of aerosols has been designed and implemented. A scanning mobility particle sizer (SMPS) and an aerodynamic particle sizer (APS) are used to measure the particle size distribution. Mie theory is then used to generate a simulated spectrum from the measured size distribution and available literature optical constants. This allows for an absolute comparison between the simulated and measured IR extinction spectra, with no adjustable parameters, thereby providing a quantitative evaluation of the errors associated with using Mie theory.
 Although T matrix methods have been derived for the analysis of shaped particles as well [Dubovik et al., 2006; Mishchenko, 1991; Mishchenko and Travis, 1998; Mishchenko et al., 1997], here we specifically investigate the potential role of particle shape effects using the “small particle” analytic model results given in equations (1)–(3) [Bohren and Huffman, 1983] for different characteristic particle shapes including distributions of spheres, disks, needles, and ellipsoids. As with the Mie simulations, these calculations are performed using the measured particle concentration, size distribution, and available literature optical constants.
 Testing and validation of the experimental method has been described previously [Hudson et al., 2007]. Here we apply this approach to study infrared extinction spectra for three silicate clays that are important components of mineral dust aerosol: illite, kaolinite, and montmorillonite. Our focus here will be on clay particles that fall in the accumulation (or clay-sized) mode. Field studies have shown that, in general, the mass median size for mineral dust aerosol over the oceans is typically in the 2–3 μm diameter range [Prospero, 1999]. Further, recent studies have estimated that up to 30% of submicron aerosol may be associated with mineral dust in a dust event [Arimoto et al., 2006]. It is expected that the clay minerals illite and kaolinite will be particularly abundant in these small dust particles [Prospero, 1999].
2.1. Source of Clays and Sample Preparation
 Illite (IMt-1), kaolinite (KGa-1, low defect) and montmorillonite (Cheto-SAz-1) were all purchased from the Source Clays Repository. Illite was coarsely ground using a mortar and pestle before being more finely ground using a Wig-L-Bug. Kaolinite and montmorillonite were used as received. All aerosols were formed from atomizing a suspension of the clay compound and Optima water (Fisher Scientific, W7-4).
2.2. Measurements of Particle Size Distributions, Fourier Transform Infrared Extinction Spectra and SEM
 The experimental method and the MultiAnalysis Aerosol Reactor System (MAARS) apparatus have been described in detail [Hudson et al., 2007]. Briefly, the Fourier transform infrared extinction spectrum (Thermo Nicolet, Nexus Model 670) of the atomized clay aerosols (TSI, Inc. Model 3076) is simultaneously acquired with size distributions from a scanning mobility particle sizer (SMPS, TSI, Inc., Model 3936) and an aerodynamic particle sizer (APS, TSI, Inc. Model 3321). The IR spectra were measured from 800 to 4000 cm−1 at 8 cm−1 resolution by co-adding 256 scans. As described previously, an aerodynamic shape factor, χ, can be determined from the independent measurements of mobility and aerodynamic diameter.
 For additional particle shape information, particle samples were collected on a piece of mica affixed to a scanning electron microscope (SEM) stub. The SEM stub was exposed to particles from the aerosol stream upon exiting the diffusion dryer for 30 to 45 min depending on the concentration of the sample. The SEM images were acquired with a Hitachi S-4000 instrument.
 A theoretical simulation of the extinction spectrum using the measured particle size distribution (from 20 nm to 20 μm) and optical constants drawn from the published literature can then be compared to the measured IR extinction spectrum. Because the particle concentration, size distribution, and optical constants are all known, the simulation offers an absolute comparison with the experimental data, with no adjustable parameters. This allows for a quantitative evaluation of the errors associated with using Mie theory to model the IR resonance absorption of mineral dust aerosol.
3.1. Size Distributions
Figure 1 shows (left) partial and (right) full size distributions for the three clay compounds studied, (a) illite, (b) kaolinite, and (c) montmorillonite. The process of overlapping the SMPS and APS data has been described in detail previously [Hudson et al., 2007]. Briefly, similar to the method of Khlystov et al.  the SMPS and APS systems are overlapped in a diameter region where both instruments have good sampling efficiencies and where the counting results for the two instruments are in good agreement. For the clays discussed in this work, the agreement between the two instruments is best from 700 to 850 nm mobility diameter. In this region the SMPS and APS data sets are overlapped and the shape factor, χ, is determined according to equation (4) [DeCarlo et al., 2004; Hinds, 1999]:
where Dm and Da are the measured mobility and aerodynamic diameters, respectively, ρo is the reference density (1 g cm−3), ρp is the density of the particle, and Cs(Dm), Cs(Da), Cs(Dve) are the Cunningham slip factors for the mobility, aerodynamic, and volume equivalent diameters, respectively.
Figure 1 (left) shows a partial region of the measured SMPS mobility, measured APS aerodynamic and calculated APS mobility size distributions. The points represent the median diameter the size bin with that number concentration (dN/dlogD). It can be seen that the SMPS and APS mobility diameters are in good agreement with the exception of the first few size bins of the APS where the sampling efficiency decreases, as has been previously reported [Armendariz and Leith, 2002; Hudson et al., 2007; Khlystov et al., 2004]. The experimentally determined shape factors, also shown in Table 1, are χ = 1.3, 1.1 and 1.1 (ρp = 2.8, 2.6 and 2.35 g cm−3) for illite, kaolinite and montmorillonite, respectively. χ is a measure of the deviation of particle shape from that of a sphere where χ = 1. It is therefore not surprising that the shape factors are greater than unity as clay particles are highly irregular and sheet-like in appearance [Nadeau, 1987], as can be seen in the SEM images shown in Figure 2.
Table 1. Median Diameter and Width Values Determined From Lognormal Fits to Size Distributions, Mass-Weighted Mean Diameter Calculated From Full Size Distributions, and Shape Factor χ Determined From Overlap of SMPS and APS Data
Statistical error in χ determined from the average of the experimentally determined χ for 8–12 size distributions of each compound.
153.6 ± 2.0
0.88 ± 0.01
418.9 ± 7.3
1.30 ± 0.02
409.6 ± 55.3
0.59 ± 0.02
673.9 ± 52.6
1.05 ± 0.04
208.8 ± 12.4
1.26 ± 0.09
578.3 ± 46.4
1.11 ± 0.03
Figure 1 (right) shows the full combined size distributions with respect to mobility and volume equivalent diameter. The volume equivalent diameter is calculated according to DeCarlo et al.  as:
The dashed box highlights the expanded region shown in Figure 1 (left). Because the analytic shape solutions are derived according to volume equivalent diameter, the volume equivalent diameter is used as the size distribution input into the Mie and analytic shape solution simulations.
 For the clay samples studied, the size distribution can be well fit by a lognormal distribution. This is generally the case when employing at atomizer source to generate aerosols of soluble species and appears to hold for the atomization of insoluble compounds as well. The median and width values for the lognormal fits are given in Table 1. Since the absorption for small particles scales as particle volume (or mass), the mass-weighted mean diameters (Dw) for the distributions are also given in Table 1. The mass-weighted mean diameters are larger than the median volume equivalent diameter and range Dw = 419–674 nm. Within the IR resonance region of interest these diameters correspond to an effective size parameter range of 0.12–0.25. The implications of using the analytic shape solutions derived for particles where X ≪ 1 with X values that are arguably not much less than 1 will be discussed later.
3.2. Mie Simulation Results
Figures 3, 4, and 5show experimental and simulated IR spectra for illite, kaolinite and montmorillonite, respectively. Water has been reduced in the aerosol samples from passing the aerosol through two diffusion dryers prior to measuring the IR spectrum, but weak adsorbed water bands are still apparent in the spectra (vide infra). Infrared absorptions due to gas-phase water present in the spectrum from slight changes in the conditioning tube have been subtracted out. Gas-phase carbon dioxide (CO2) is also kept to a minimum using a commercial dry air generator (Parker Balston, Model 75-62) but can be observed in the spectrum as a doublet centered at 2348 cm−1. The volume equivalent size distributions shown in Figure 1 are used as inputs in the Mie simulation in combination with optical constants drawn from the published literature [Querry, 1987].
Figures 3a, 4a, and 5a show the experimental spectrum and Mie simulation from 800 to 4000 cm−1 for illite, kaolinite, and montmorillonite, respectively. An expanded view of the prominent Si—O stretch resonance spectral region is shown in Figures 3b, 4b, and 5b. All three clay samples show the characteristic Si—O stretch (1036–1048 cm−1) as well as structural O—H stretches near 3620 cm−1.
 Clear differences between the Mie theory simulation and the experimentally observed resonance line profiles are apparent for the three clay samples. Discrepancies are readily seen in the strong Si—O stretch resonance region where there are significant differences in band position, band shape, and peak intensity. This spectral range is particularly important because the silicate stretch region can be used to characterize mineral dust loading from narrow band and high-resolution satellite spectral data [Ackerman, 1997; DeSouza-Machado et al., 2006; Sokolik, 2002]. The experimental Si—O resonance peaks for illite, kaolinite and montmorillonite are red shifted ∼27–44 cm−1 relative to the Mie simulation. Peak assignments and mode descriptions for the experimental and Mie simulated spectra are given in Tables 2a–2c with comparison to literature values.
Table 2a. Vibrational Assignments for Illite (IMt-1)
 In addition to the discrepancy in the Si—O resonance region, the baseline slopes in the 2500–4000 cm−1 range also differ somewhat. The poor simulation of the baseline slope may be due to residual adsorbed water in the clay samples and is discussed in detail below. The adsorbed water is most evident in the montmorillonite sample shown in Figures 5a and 5b. In addition to the Si—O stretch observed near 1050 cm−1 and the inner O—H stretch at 3616 cm−1, the O—H stretching of adsorbed water (3000–3400 cm−1) and H2O bending (1633 cm−1) modes are present in the experimental spectrum. These modes are due to the adsorbed water in the inner layers of montmorillonite as it is a swellable clay. The structure consists of 2 tetrahedral sheets sandwiching a central octahedral sheet (2:1 Si:Al ratio) that allows for water to penetrate into the interlayer molecular spaces. Because of their structures, illite and kaolinite are “less swellable” and the water signatures are correspondingly weaker. There might also be some slight variations in resonance line shapes associated with water uptake in these clay samples. Recent work by the Tolbert group has shown that the presence of water does not lead to a major change in the silicate Si—O resonance peak structure [Frinak et al., 2005]. For example, even in montmorillonite, the clay with the most water association, the shift in the Si—O resonance peak position under dry and wet conditions is less than 10 cm−1 [Frinak et al., 2005]. Other studies in our lab involving attenuated total reflectance measurements on bulk clay powders have shown that there can be some slight differences in line position and band shape among different mineral samples from different sources, though these variations are smaller than the line shifts observed here [Schuttlefield et al., 2007]. The Mie simulations should not capture the adsorbed water features.
 The Mie simulation is quantitatively compared to the experimental spectra in two ways. First, the integrated area of the Si—O resonance peaks can be compared. Specifically for kaolinite and montmorillonite, the experimental peak amplitudes are roughly a factor or two larger. In earlier work we presented the Mie simulation results for the illite extinction spectrum and explored possible sources of error for the observed spectral differences [Hudson et al., 2007]. The largest systematic uncertainties are in the optical path length, and in the possibility that there may be some particle transmission losses between the extinction cell and the particle counting and sizing instrumentation. We estimated the total systematic error from these sources to be ∼12% in the resonance peak amplitude, with minimal error (<2 cm−1) in the resonance peak position [Hudson et al., 2007]. Second, the χ2 error is determined for the Mie simulation relative to the experimental spectrum over the spectral range from 900 to 1300 cm−1. The χ2 error is defined as the sum of the square of the difference between the simulation and the experimental spectrum. The χ2 error between the experimental spectrum and Mie simulation, and the integrated area, normalized to the area of the experimental spectrum, are shown in Table 3. The χ2 error will be discussed with respect to the analytic shape simulations below.
Table 3. Quantitative χ2 Error and Relative Integrated Area of Resonance Region (900–1300 cm−1)a
Percent Area Relative to Experimental Spectrum
Statistical error from an average of 8–12 spectra for each compound.
23.2 ± 8.9
6.2 ± 3.0
86 ± 9
102 ± 11
112 ± 86.1
77.3 ± 52.8
47 ± 7
55 ± 9
36.3 ± 17.6
9.2 ± 5.7
59 ± 7
79 ± 9
3.3. Shape Simulation Results
 Analytic shape simulations were also calculated according to Bohren and Huffman  for the sphere, CDE, disk and needle using the volume equivalent size distributions and literature optical constants for comparison to the experimental spectra. These results are shown in Figure 6 for illite. The analytic results for spheres essentially reproduce the Mie theory results as expected since our particle size distribution lies largely in the small particle range (D ≪ λ). However, the calculated resonance line profiles for disks, needles, or for the “continuous distribution of ellipsoids” are all markedly different. The χ2 error was used to determine the best fit to the data. Although the more commonly used CDE model offers an improvement over the Mie results, the disk simulation gives the best fit quantitatively to the experimental data. As a result, the disk simulation has also been added to Figures 3b, 4b, and 5b. It can be seen that, with respect to peak position and band shape of the Si—O resonance peak, the disk simulation results in a better match to the experimental data than the Mie simulation. As compared to the 27–44 cm−1 blue shift of the Si—O peak in the Mie simulation, this peak is now red shifted to a much smaller magnitude (6–13 cm−1). That small clay particles might be better fit by a disk shape model is not at all surprising; Nadeau  explored particle size and shape effects in the infrared spectrum of kaolinite and suggested that clay-sized particles (D < 2 μm) of kaolinite typically have a mean thickness of 0.03–0.15 μm.
 Note that these analytic solutions are derived under the assumption that X = (πD/λ) ≪ 1 and ∣m∣X ≪ 1, where ∣m∣ is the magnitude of the complex index of refraction [Bohren and Huffman, 1983]. The value for ∣m∣ varies dramatically across the resonance region, but peaks at ∼2.5 in each case. Thus the peak values of ∣m∣X range up to near 0.5. As in the case of X, the ∣m∣X values also cannot be considered to be ≪1, yet the analytic shape theory results give better overall agreement than Mie theory for modeling line positions, shapes, and integrated intensities. A subtraction of the Mie and disk simulation from the experimental results is shown as an inset in Figures 3b, 4b, and 5b as well. This result will be discussed in detail below.
 Systematic errors in the integrated absorbance for the major silicate resonance bands may affect radiative forcing calculations; in particular, a systematic underestimate in the integrated absorbance could lead to an underestimate of the positive forcing contributions of mineral dust aerosol due to outgoing terrestrial infrared absorbance. However, in this context it is important to note that the aerosols studied here (D = 0.1–1 μm) may give a relatively small contribution to overall aerosol optical depth in the IR owing to their small volume. It does, however, raise the question whether similar errors in Mie theory simulations may also be present in estimating dust absorption for larger mineral aerosol particles, with D > 1 μm.
Sokolik  has suggested that high-resolution studies in the silicate stretch spectral region, 1099–1200 cm−1 may be particularly useful for determining dust composition because many of the abundant mineralogical constituents of dust aerosol (including the silicate clays and quartz) have characteristic vibrational resonance features in this region. However, errors in simulated peak position and line shape could adversely affect determinations of mineral composition based on high-resolution IR satellite data. For example, we have found that the clay Si—O stretch resonance peak is red shifted by ca. 35 cm−1 from the Mie theory prediction. Indeed, the observed Si—O stretch mode in illite, kaolinite, and montmorillonite all overlie the strong O3 resonance absorption line near 1042 cm−1 which may make it problematic to quantify the clay silicate features in the spectrum. On the other hand, lab measurements of quartz dust spectra find that the Si—O stretch in quartz aerosol is also red shifted from the Mie simulation and in fact lies near the silicate resonance band position for the clays predicted by Mie theory [Bohren and Huffman, 1983]. Thus due care must be taken when attempting to use Mie theory to derive compositional information from high-resolution IR spectral data in the Si—O stretch region. Spectral simulations based on distributions of ellipsoid models might be preferable in modeling high-resolution spectra of mixed mineral dust aerosol obtained from field or satellite measurements.
 In this regard a recent modeling study of atmospheric dust loading based on AIRS data is noteworthy. DeSouza-Machado et al.  show a series of clear-sky and dust-sky biases for retrieved optical depths from high-resolution spectral data from AIRS. The data shows a characteristic “V” shape signature that is commonly associated with silicate-based absorbers. DeSouza-Machado et al.  have attempted to remove the effects of dust extinction from the biases to obtain more accurate temperature and water vapor measurements. This analysis uses a Mie theory–based radiative transfer model to simulate the effects of Saharan dust, which is largely silicate clay [Volz, 1973]. The clear-sky bias for retrieved optical depth shows evidence for a residual absorption band in the 1020–1060 cm−1 range that is attributed to errors in the ozone concentration. However, an alternate explanation may be that the residual dip in the BT spectrum is associated with the silicate clay absorption resonance centered near 1040 cm−1, which was misplaced in the Mie simulation. The insets of Figures 3b, 4b, and 5b show a dispersion shape profile when the Mie simulation is subtracted from the experimental absorption spectrum that is similar to the retrieved clear sky bias data by DeSouza-Machado et al.  in the 1020–1120 cm−1 range. A similar effect is observed in the disk simulation subtraction but to a lesser degree.
 Although, the overall agreement between the simulations and the experimental spectra are poor, the way that the simulations are utilized in satellite data retrievals may be more forgiving. Ackerman  has suggested that comparing difference measurements between specific narrow band IR sensor channels of MODIS (BT8–BT11vs. BT11–BT12) could be used to indicate the presence of dust. However, the simulated BT differences can also be affected by errors in peak placement and band shape. Sokolik  has previously reported modeling results describing the effect that light and moderate dust loadings (where dust composition is varied as a function of a mixture of quartz and clays) has on brightness temperature retrievals for four narrowband sensors, the High-Resolution Infrared Radiation Sounder (HIRS/2), AVHRR, MODIS and Geostationary Operational Environmental Satellites (GOES 8) [Sokolik, 2002]. Table 4 shows the narrowband channels for these satellites used for integration. For comparison, a top hat integration over the narrowband range for each channel has been carried out for the experimental and simulated spectra presented here. The ratio of integrated peak areas for the Mie and the analytic disk simulation, relative to the experimental extinction spectrum are shown in Figure 7 for illite, kaolinite and montmorillonite. Figures 7a–7d present the bands for each respective satellite, whereas Figure 7e shows a side-by-side comparison of all satellites divided according to the wavelength of integration. Note that Figure 7 does not present results from simulated BT measurements, but the extinction results presented here are closely related to the BT spectra and give useful insight. Only slight differences are observed between satellite sensors for common wavelength integrations. For example, in the brightness temperature band centered at 8 μm (BT8) using the HIRS 8.01–8.42 μm (1188–1248 cm−1) and MODIS 8.4–8.7 μm (1149–1190 cm−1) bands, the Mie and disk simulated integrated areas are larger than the experimental from 15 to 162% respectively for kaolinite and montmorillonite. However, in the HIRS and MODIS BT8 bands, there is only a small difference between the Mie and disk simulations for all three clays. With the exception of the BT8 bands, the disk simulation more closely approximates the integrated area relative to the experimental spectrum. There is no strong bias from satellite to satellite as the integrating range for each is not significantly different. The error bars represent the statistical error calculated from the standard deviation in the average of multiple measurements of experimental spectra with size distributions used in the Mie and disk simulations.
Table 4. Narrowband Satellite Sensors Operating in the Infrared Region
 Note that Mie theory simulations could introduce an error in data retrievals from the BT8–BT11 difference signals for dust clouds that are clay-rich, as Mie theory appears to systematically overestimate the extinction signal in the BT8 range and underestimate the extinction in the BT11 and BT12 bands. The brightness temperature spectrum is directly related to the atmospheric transmission and, as such, appears as the inverse of the absorption spectrum. Thus an overestimate in the BT8–BT11 difference signal in the extinction (absorption) spectrum, corresponds to an underestimate in the BT8–BT11 difference signal in the BT spectrum. As a result, the BT8–BT11 brightness temperature difference signals will tend to be underpredicted in the Mie simulation for clay particles in this size range. The extinction signals in the BT11 and, especially, the BT12 band regions are weak and it is difficult to draw a firm conclusion about the BT11–BT12 difference signal. However, it appears that the BT11–BT12 differences calculated by Mie theory, as compared to the disk simulation, may be somewhat closer to the experimental result. That is, even though Mie theory may underestimate the total extinction for fine clay particles in the 11–12 μm spectral range, the BT11–BT12 band differences, which are related to the slope of the extinction spectrum rather than its absolute value, may be more consistent with the Mie theory simulation. This latter point is important because Sokolik  has suggested using the slope in the 820–920 cm−1 (10.9–12.2 μm) spectral region as a determinant for dust. Thus, while the main resonance peak structure may not be well modeled in Mie theory, the measured extinction slope (and, by inference, the BT slope) in the 820–920 cm−1 range may be consistent with the Mie theory simulations for the fine silicate clay particles (D = 0.1–1 μm) studied here.
5. Conclusions and Atmospheric Implications
 We have measured simultaneous size distributions and Fourier transform infrared extinction spectra for several clay components of mineral dust aerosol, illite, kaolinite and montmorillonite. The measured size distributions are used in combination with published optical constants for comparison to the measured extinction spectrum. Within the fine particle size mode (D = 0.1–1 μm), Mie theory does a relatively poor job in predicting the resonance peak position, intensity and line shape for the prominent Si—O stretch ca. 1050 cm−1. The experimental Si—O resonance bands are red shifted by ca. 27–44 cm−1 relative to the Mie theory simulation. Simulations based on distributions of ellipsoid shaped particles, the continuous distribution of ellipsoids (CDE), disk and needle shaped particles, are used to investigate particle shape effects on the extinction spectrum. Interestingly, the analytic theory results for disk-shaped particles appear to give the best overall agreement with the experimental line shape data. The more commonly used CDE model also offers an improvement over Mie theory.
 The results from this study have further been examined in the context of remote sensing measurements. Satellite-based high-resolution IR spectral measurements, or differences in brightness temperature measurements from narrowband IR sensor channels could be useful for determining atmospheric dust loading and composition. Our studies show that care must be taken when using Mie theory to model the spectral data in satellite retrieval algorithms since Mie theory does a poor job in predicting the band positions and line shapes for the main silicate absorption features in the 8–12 μm spectral range. For example, BT8–BT11 brightness temperature difference measurements are likely to be significantly underpredicted by Mie theory for dust clouds that are clay-rich. The situation in the BT11–BT12 band region is less clear; Mie theory may be somewhat more reliable over that spectral range, which corresponds to the far red wing of the silicate absorption band. Simple analytic models for shaped particles, derived in the small particle limit, may offer a better fit to the major band features for interpreting high-resolution spectra.
 In future work we will extend these measurements to include several important nonclay mineral dust constituents such as quartz, calcite, and gypsum. Additionally, we plan to extend our analyses to include other theoretical approaches (e.g., T matrix theory) for calculating the extinction properties of mineral dust aerosols.
 This material is based upon work supported by the National Science Foundation under grant ATM-0425989. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. The authors would like to thank Courtney Usher for her assistance in obtaining SEM images and Jennifer D. Schuttlefield for discussions of clays and water adsorption.