Simultaneous Fourier transform infrared (FTIR) extinction spectra and aerosol size distributions have been measured for some components of mineral dust aerosol including feldspars (albite, oligoclase) and diatomaceous earth, as well as more complex authentic dust samples that include Iowa loess and Saharan sand. Spectral simulations for single-component samples, derived from Rayleigh-theory models for characteristic particle shapes, better reproduce the experimental spectra including the peak position and band shape compared to Mie theory. The mineralogy of the authentic dust samples was inferred using analysis of FTIR spectra. This approach allows for analysis of the mineralogy of complex multicomponent dust samples. Extinction spectra for the authentic dust samples were simulated from the derived sample mineralogy using published optical constant data for the individual mineral constituents and assuming an external mixture. Nonspherical particle shape effects were also included in the simulations and were shown to have a significant effect on the results. The results show that the position of the peak and the shape of the band of the IR characteristic features in the 800 to 1400 cm−1 spectral range are not well simulated by Mie theory. The resonance peaks are consistently shifted by more than +40 cm−1 relative to the experimental spectrum in the Mie simulation. Rayleigh model solutions for different particle shapes better predict the peak position and band shape of experimental spectra, even though the Rayleigh condition may not be strictly obeyed in these experiments.
 Scattering and absorption from atmospheric dust also affects the interpretation of remote sensing data. Key atmospheric properties e.g., temperature profiles of the atmosphere are determined based on measurements in the IR spectral region [Sokolik, 2002]. A wide range of atmospheric and oceanic properties can be determined by remote sensing, using measurements from narrow band IR sensor channels of such instruments as the Moderate Resolution Imaging Spectroradiometer (MODIS), the Advanced Very High Resolution Radiometer (AVHRR), the Atmospheric Infrared Sounder (AIRS), the High-Resolution Infrared Radiation Sounder (HIRS/2), and Geostationary Operational Environmental Satellites (GOES-8) [Ackerman, 1997; Pierangelo et al., 2004; Sokolik, 2002]. It is important to take the radiative effect of atmospheric dust into account when processing remote sensing data, which requires an accurate determination of dust optical properties [DeSouza-Machado et al., 2006; Thomas and Gautier, 2009]. In fact, errors may occur in the analysis of climate trends when atmospheric dust is not properly modeled in satellite retrievals [Ackerman, 1997].
 For particles with diameters considerably smaller than the wavelength of light (D ≪ λ) (Rayleigh regime), simple analytic solutions have been derived to calculate absorption cross sections for particles with simple characteristic shapes, such as disk or needle shapes, or for a “continuous distribution of ellipsoids” (CDE) model [Bohren and Huffman, 1983]. A CDE model is widely used to simulate absorption for cosmic dust particles [Fabian et al., 2001]. The effect of particle shape on IR extinction spectra for some prevalent components of atmospheric mineral dust aerosol was investigated by Hudson et al. [2008a, 2008b]. Those studies focused on dust aerosol with particle diameters <2 μm (accumulation mode). The comparison of experimental IR spectra with Mie simulations and with models derived in the Rayleigh regime, show that the analytic solutions based on Rayleigh theory better reproduce the resonance peak positions and band shapes than Mie theory. The poor results from Mie theory are caused by a breakdown of the spherical particle assumption. Specifically, the prominent Si—O stretch resonance for silicate clay minerals (illite, kaolinite, and montmorillonite) are better fit by assuming a disk shape [Hudson et al., 2008a]. For other mineral dust aerosol components (quartz, dolomite, and calcite), it was found that a CDE model gives better results [Hudson et al., 2008b]. These previous studies have focused on single component mineral dust aerosol samples. Little, however, is known about the effect of varying composition and particle shape on IR absorption profiles for more complex authentic dust mixtures.
 In this study, we have investigated IR extinction spectra of single components of mineral dust, feldspars (albite and oligoclase) and diatomaceous earth, as well as multicomponent authentic dust samples (Iowa loess and Saharan sand). Feldspars are known to be common constituents of atmospheric aerosol from many sources [Menéndez et al., 2007; Shen et al., 2009; Abed et al., 2009; Klaver et al., 2011]. Diatomaceous earth is a major component of dust transported from central Africa [Warren et al., 2007; Schwanghart and Schütt, 2008]. We are primarily interested in accumulation mode particles, typical for long-range transported aerosol. It has been estimated that mineral dust may comprise approximately to 30% of submicron particles in dust storms [Arimoto et al., 2006].
 The paper is organized as follows. Section 2 describes the experimental and simulation methods used in this study. Section 3.1 presents the results of model simulations for single component minerals. Section 3.2 includes description of mineralogy estimation for complex authentic samples, and spectral simulations using the results from section 3.1 and previously published data [Hudson et al., 2008a, 2008b]. Section 4 highlights the importance of the results of this work on remote sensing analysis. Section 5 summarizes these findings and offers future directions.
2. Experimental Methods
2.1. Sources of Materials and Sample Preparation
 Albite (Item 49 V 5851, Evje, Norway) and oligoclase (Item 46 V 5803, Madawaska, Ontario, Canada) were purchased from WARD'S Natural Science Establishment, Inc. and ground into finer powders using a mortar and pestle. Diatomaceous earth was purchased from Alfa Aesar (Item 89381). Authentic materials were obtained from the Saharan desert and the Loess Hills of western Iowa. The exact location of the source of Saharan sand sample was presented by Krueger et al. . Iowa loess and diatomaceous earth were used as received, while Saharan sand was passed through a 100 um sieve before using.
 Aerosol particles were formed by atomizing (TSI Inc., Model 3076) a suspension of the compound in Optima water (Fisher Scientific, W7–4, Lot 111718). Upon exiting the atomizer aerosol with a flow rate of 1.5 lpm is passed through diffusion dryers (TSI Inc., Model 3062) to dry particles to ∼5% relative humidity. The total residence time of aerosol through dryers is ∼0.4 min. The atomizer limits dust particle size in the flow to the accumulation mode range with diameter D ∼ 0.01–2.5 μm.
2.2. Characterization of Materials
 Bruker D-5000 q - q diffractometer (CuKα1line) equipped with a Kevex energy-sensitive detector was used for powder X-ray diffraction measurements (XRD). The scanning electron microscope (SEM) images were acquired with a Hitachi S-4800 instrument. The elemental composition of mineral dust used in this study was determined using a Hitachi S-3400N SEM in combination with energy dispersive X-ray (EDX) spectrometer. Particles were collected on carbon stubs directly from the stream of aerosol. The details of the analytical methods used were described earlier byCwiertny et al. .
2.3. Size Distribution and FTIR Extinction Spectra
 Aerosol size distributions and IR extinction spectra were measured simultaneously using a Multi-Analysis Aerosol Reactor System (MAARS). The experimental set up of MAARS has been described in detail [Gibson et al., 2006]. Briefly, the full size distribution of aerosol is measured using a scanning mobility particle sizer (SMPS, TSI, Inc., Model 3936) and an aerodynamic particle sizer (APS, TSI, Inc., Model 3321), operating in tandem. An algorithm is then applied that combines the aerodynamic and mobility size data into an effective volume equivalent size distribution, as described by Khlystov et al.  and Hinds . The method also allows a determination of χ, the effective aerodynamic shape factor for the aerosol. Important equations and relations that were used are summarized in auxiliary material Table S1 in Text S1.
 An FTIR spectrometer (Thermo Nicolet, Nexus Model 670) was used to measure the IR extinction spectra in the spectral range from 800 to 4000 cm−1 using 8 cm−1 resolution. In order to relate measurement of the size distribution with the corresponding IR spectrum, the SMPS, APS, and IR scan times (210 s) were synchronized. In this work we focused our investigation on the prominent IR resonance features in the 800 to 1400 cm−1spectral range. In order to compare experimental spectra with the model simulations, a sloping baseline, which is related to a fall-off in the lamp intensity toward the low energy end of the spectral range, was corrected in the experimental IR spectrum using a linear interpolation over the region 800 to 1200 cm−1.
2.4. Model Simulations
 Mie theory and Rayleigh theory were used to simulate the extinction spectra. Mie theory is used to model the optical properties of spherical particles and, in the Rayleigh limit (D ≪ λ), analytic solutions have been derived to predict absorption cross sections for particles of different shapes such as disks, needles, and a CDE model [Bohren and Huffman, 1983]. In the spectral region investigated, the magnitude of the scattering cross section is negligible compared to the extinction cross section, so absorption cross sections are approximately equal to extinction cross sections. The simulation code used here (adapted from Hung and Martin ) calculates an extinction spectrum for a given measured particle size distribution and set of optical constants. The code essentially computes the extinction for a given size bin, and then sums over the size bins weighted by the particle number density in each bin to obtain the final spectrum. The procedures, equations, and calculations have been described in detail [Hudson et al., 2011] and are summarized in Table S1. To simulate the extinction spectra for complex samples a weighted linear combination of the modeled IR spectra of the minerals composing particular dust sample was performed.
3. Results and Discussion
3.1. Single Component Minerals
 SMPS and APS data was combined to get a full particle size distribution for diatomaceous earth, albite and oligoclase. The full size distributions are shown in Figure S1 as a function of both mobility (Dm) and volume equivalent (Dve) diameter. Volume equivalent diameters are used in the simulations. To measure the average particle size the mass-weighted mean diameter (Dw) is usually used. Details of determining mass-weighted mean diameter calculations from full size distribution have been described previously [Hudson et al., 2008a]. For the particle distributions presented in Figure S1the mass-weighted mean diameters are Dw = 586 ± 9, 491 ± 22, and 530 ± 21 nm for diatomaceous earth, albite and oligoclase, respectively. Within the IR resonance region studied here these diameters correspond to the dimensionless size parameter x = πD/λ ∼ 0.2.
 The shape factors, experimentally determined for these samples from SMPS and APS data overlap, are χ = 1.34 ± 0.02 for diatomaceous earth (density ρp = 2.42 g cm−3), 1.08 ± 0.02 for albite (density ρp = 2.62 g cm−3), and 1.05 ± 0.03 for oligoclase (density ρp = 2.65 g cm−3). The density of diatomaceous earth is specified by the manufacturer (Alfa Aesar - A Johnson Matthey Companyhttp://www.alfa.com). The densities of albite and oligoclase were reported by Anthony et al. . The errors in the shape factor values indicate the standard deviation of a series of experiments (∼10–15) for each compound. The value of the shape factor determined for the diatomaceous earth sample is quite large, which is usually indicative of particles with extreme shapes or aggregates with voids [DeCarlo et al., 2004]. Indeed, SEM images of the diatomaceous earth sample show it to consist of aggregates of thin wafers and rod-like structures, with many voids, as can be seen inFigure S2.
Figures 1 and 2 show experimental and simulated resonance extinction spectra for diatomaceous earth (Figure 1) and feldspars (Figure 2). Aerosol samples were dried to a relative humidity ∼5% by passing the aerosol through two diffusion dryers before measuring the IR spectrum; remaining weak signals from gas-phase water have been subtracted from the spectra. The size distributions depicted inFigure S1 were combined with optical constants obtained from the literature (vide infra) and used as inputs into the spectral simulations.
 Feldspars are a group of minerals which have a silicate framework. Albite and oligoclase are members of the plagioclase feldspar series with different amounts of sodium and calcium. As can be seen in Table S2, the chemical compositions of albite and oligoclase are very similar, so the set of IR optical constants for albite, published by Mutschke et al. , were used for both feldspar samples. The experimental line shapes for albite and oligoclase show broadly similar envelopes, but different unique features of the band shapes resulting from their differing mineralogies. Moreover, the distinctive band shapes for both albite and oligoclase are sample dependent and we should not necessarily expect to reproduce those details in the spectrum.
 Diatomaceous earth is an amorphous silica material consisting of the remains of dead diatoms in marine sediments, and XRD confirmed the lack of crystallinity of the material and the amorphous character of diatomaceous earth. Therefore optical constants for silica [Steyer, 1974] were used to simulate the optical properties of diatomaceous earth.
 The first panels of Figures 1 and 2 show the experimental spectra and Mie simulations for the prominent silicate stretch resonance spectral region from 800 to 1400 cm−1 for diatomaceous earth, and, albite and oligoclase, respectively. The study of this spectral range is crucial since it can be used to estimate loading of mineral dust using measurements from narrow band IR sensor channels of satellites [Ackerman, 1997; DeSouza-Machado et al., 2006; Sokolik, 2002].
 Two parameters were used to quantitatively compare the experimental spectra with the simulations. First, the position of the silicate peak in experimental and simulated spectrum was compared. Second, the χ2 error between the experimental spectrum and simulation was determined for the 800–1400 cm−1 spectral range. The χ2 error is the square of the difference between the simulation (Esim) and the experimental spectrum (Eexp) at a particular wave number, summed over a chosen range of wave numbers (ν1 − ν2) and divided by the number of data points (N), and can be calculated according to formula (1):
For all three samples there are significant differences between the simulations based on Mie theory and the experimental results in the position and intensity of the Si—O stretch resonance peak, and in the overall line shape. The Mie-based simulation for diatomaceous earth shows a resonance peak that is blue shifted by 41 cm−1 relative to the experimental spectrum.
 Results for CDE, disk, and needle-shaped models are also shown inFigures 1 and 2 for diatomaceous earth, and, albite and oligoclase. The commonly used CDE model offers a significant improvement in simulated band shape and position over the Mie results for diatomaceous earth (Figure 1). The Si—O peak position is significantly better predicted by the CDE model then by Mie theory; the peak is shifted by 11 cm−1 relative to the experiment (versus 41 cm−1 peak shift in Mie theory) and χ2 error is smaller for CDE model (1.84 × 10−6) than for Mie theory (4.56 × 10−6). The observation that the CDE model distribution gives a good qualitative fit to the experimental line shape is reasonable, since the SEM images of diatomaceous earth presented on Figure S2 definitely show a broad range of particle shapes ranging from thin rods to thin wafers.
 Analytic solutions derived from a Rayleigh theory for shaped particles also show some improvement over the Mie theory simulation in the overall bandwidth of the resonance profile of feldspars. In addition, as can be seen in Figure S2, SEM images of both albite and oligoclase show the variety of ellipsoids rather than spheres. However, unlike the case of diatomaceous earth noted above, or for the mineral dust results reported by Hudson et al., [2008a, 2008b] none of the limiting forms for the Rayleigh simulation here do a very good job of simulating the band shape for the feldspars. The smallest χ2error is found for needle-shaped particles (1.78 × 10−7 for albite and 2.18 × 10−7 for oligoclase) but the CDE model gives a very similar result (1.82 × 10−7 for albite and 2.54 × 10−7for oligoclase). However, both Mie theory and Rayleigh-based models fail to reproduce the observed structure in the IR band. As was discussed (see above), the structure of the feldspar extinction spectrum is sample dependent and clearly changes with feldspar mineralogy, which means that the optical constants are also likely to be sample dependent. This may partly explain the observation that none of the models reproduce the peak structure in detail. It must also be noted that the limiting forms for the Rayleigh model solutions can only give a crude approximation to the actual shape distributions for mineral dust. Furthermore, mixed shape models (i.e., mixtures of needles and disks) also do not significantly improve the fit.
3.2. Complex Mineral Mixtures
3.2.1. Mineralogy of Complex Samples
 Modeling mineral dust aerosol as an external mixture of individual mineral components often gives good results for determining mineral dust optical properties [Köhler et al., 2011]. Thus, the extinction spectra for our authentic dust samples were mimicked computationally using a linear combination of the IR spectra of the proposed individual constituents of the dust. IR extinction spectra were measured for quartz, amorphous silica, kaolinite, montmorillonite, illite, calcite, dolomite, albite, oligoclase, and for the authentic samples, Iowa loess and Saharan sand. For comparison, IR spectra were normalized to the unit aerosol volume. To determine the mineralogy of the complex authentic dust samples (Iowa loess and Saharan sand), IR extinction spectra of these were compared with experimental IR extinction spectra of the individual mineral components that might be expected to be present in the authentic samples (see below). Spectra of the components were added together in different proportions and the best overall fit to spectra of the authentic mineral dust sample was determined by a least squares minimization procedure. Note that this is a purely empirical fit involving only experimental line profiles.
 The composition of the mixture which yielded the best overlap with the experimental spectrum was taken as the composition of the sample. The resulting mineral compositions are shown in Figure 3. Those results indicate that that the clay content is approximately 88% and 81% of the total for Iowa loess and Saharan sand, respectively. This is in a good agreement with the fact that clays are one of the most abundant constituents of the fine fraction of mineral dust aerosol [Prospero, 1999]. The results for Saharan sand are also in a good agreement with Linke et al. , who has reported that the major components of Saharan dust are clays, silicates, feldspars with small addition of carbonaceous materials (dolomite and calcium), gypsum and iron oxides (hematite and goethite). The simulated spectra that result for an external mixture of mineral components are shown in Figure 3.
 Experimental measurement of the elemental compositions of the authentic dust samples using EDX analysis was then compared with the composition derived from spectral fitting of the IR data. The results are summarized in Table 1. As can be seen, there are some discrepancies between the analyses, particularly with the iron content. A possible reason for such disagreement may be due to the fact that common iron oxides and hydroxides (hematite, goethite) do not have strong extinction features in the IR spectral region from 800 to 1400 cm−1. Thus, our IR measurements are insensitive to the presence of common iron oxides such as hematite. In addition, the Fe content of the clays and feldspars can be highly variable.
Table 1. Elemental Composition of Authentic Dust Samples and Corresponding External Mixtures Mimicked Using a Linear Combination of the IR Spectra of the Components
External Mixture (Wt %)
EDX Experiment (Wt %)
External Mixture (Wt %)
EDX Experiment (Wt %)
 There are also notable differences in the concentrations of other trace metals. For example, there are some discrepancies in the sodium and potassium content and in the magnesium and calcium content. These differences are easy to understand. Most clay and feldspar minerals have very similar spectral signatures, which make it challenging to distinguish between different clays and feldspars in the extinction spectrum. On the other hand, different clays and feldspars can have quite different trace metal concentrations. Nevertheless, the comparison of the silicon to aluminum ratio is the most important and Si:Al ratios are in fairly good agreement for both Iowa Loess (55:22 from IR fit and 48:19 from the EDX analysis) and Saharan sand (48:23 from IR fit and 53:19 from the EDX analysis).
 The results suggest that spectral fitting approach can yield a quantitative estimate of the relative amounts of clays, feldspars, carbonates, and silicon dioxide in mineral dust aerosol, directly from high resolution IR spectral data.
3.2.2. Model Simulation Results
Figure S3 shows the full particle size distributions for Iowa loess and Saharan sand, obtained from the SMPS and APS data overlap. For the size distributions shown in Figure S3the mass-weighted mean diameters are Dw = 473 ± 27, and 423 ± 44 nm for Iowa loess and Saharan sand, respectively. For the investigated IR region it corresponds to dimensionless size parameters x in the range of ∼0.1.
 The shape factors experimentally determined for these samples are χ = 1.05 ± 0.02 and 1.31 ± 0.05 using densities of ρp = 1.43 and 2.30 g cm−3 for Iowa loess and Saharan sand, respectively. The experimentally measured densities are consistent with previously published values for loess (1.45 g cm−3) [Bettis et al., 2003] and for Saharan sand (2.5 g cm−3) [Linke et al., 2006]. The results for Saharan dust are in a good agreement with Davies  who reported that a dynamic shape factor of particle mixtures of different composition ranges from 1.3 to 1.6.
Figure 4 shows the comparison of the extinction spectra for Iowa loess and Saharan sand samples with the results from different model simulations using the mineralogy determined from the fit shown in Figure 3. Optical constants for components of the authentic samples were obtained from the literature. (Data published by Querry  was used for illite, montmorillonite, kaolinite and dolomite; data from Mutschke et al.  was used for albite, data from Steyer  for silica and data from Long et al. for calcite.) Aerosol samples were dried and gas-phase water absorptions have been subtracted from the spectra as described earlier.
 Spectral simulations were carried out for each component using the corresponding optical constants and measured size distributions for the authentic samples. The simulations are based either on Mie theory or on Rayleigh theory using the characteristic particles shapes for each component that have been previously determined. Specifically, the clays illite, kaolinite, and montmorillonite have been modeled as disks based on the prior work of Hudson et al. [2008a]. All the remaining non-clay minerals (amorphous silica, calcite, dolomite, albite, oligoclase) have been modeled by the CDE approximation based on the discussion above and the results ofHudson et al. [2008b]. Figure 4 (top) shows the comparison between the experimental spectra of authentic samples and Mie simulations. Figure 4 (bottom) gives the comparison between the experimental data and the Rayleigh based model simulations for Iowa loess and Saharan sand. Only the resonance region from 800 to 1500 cm−1 is shown, which includes both the silicate stretch region and the carbonate region.
 It is clear from the results in Figure 4 that the analytic solutions based on Rayleigh theory that include nonspherical particle shape effects show significantly improved overlap with the experiment spectra in comparison to Mie theory. For Iowa Loess χ2 error is more than 5.5 times higher for Mie theory (17.6 × 10−7) compared to Rayleigh theory (3.11 × 10−7); for Saharan sand χ2value is also significantly lower for Rayleigh-based model (3.51 × 10−8) than for Mie theory (12.7 × 10−8).
 The peak in the simulation based on Mie theory is blue shifted by ∼40 cm−1relative to the experiment. The simulations based on Rayleigh-based analytic solutions provide better results. The peak maximum is now red shifted by less than 3 cm−1. In addition, the overall band shape is much better fit by the Rayleigh model simulations.
4. Atmospheric Implications
 As mentioned previously, the presence of dust could be denoted using narrow band IR sensor channels of satellites [Ackerman, 1997]. The use of Mie theory could introduce unfavorable errors in satellite retrievals. Brightness temperature (BT) retrievals of HIRS/2, AVHRR, MODIS and GOES-8 narrowband sensor channels are affected by dust and the effect of different dust loadings was published previously [Sokolik, 2002].
Figure 5 shows a plot of the ratio of the integrated area of simulated spectra to the experimental spectra over narrowband channels BT8, BT11 and BT12 for four satellites mentioned above. The measure of the quality of the simulation is the proximity of this ratio to one. As can be seen in Figure 5, the Rayleigh simulations for nonspherical particles, in general, give better results in the BT11 and BT12 bands. In the band centered at 8 μm (BT8) using the HIRS/2 8.01–8.42 μm (1188–1248 cm−1) and MODIS 8.4–8.7 μm (1149–1190 cm−1) bands, there is no significant difference between the integrated areas derived from Mie theory and shape simulations. In the BT11 and B12 bands, there are more noticeable distinction between the experiment and the simulations based on Mie and Rayleigh theory. The Rayleigh model simulations better approximate the integrated area to the experimental data over channels BT11 and BT12 and Mie theory underestimates the extinction in the BT11 and BT12 bands.
 The extinction data are inversely related to the brightness temperature data. Accordingly, if the extinction in the BT11 and BT12 bands is underestimated, the BT11–BT12 brightness temperature will be overestimated. Our results suggest that spectral simulations based on analytic solutions derived from Rayleigh theory are more reliable than Mie theory for modeling the extinction spectra of atmospheric aerosol consisting of a mixture of various minerals. Additionally, the use of these experimentally measured extinction spectra can be used to improve modeling and retrieval data [Klüser et al., 2012].
 Fourier transform infrared extinction spectra and full size distributions have been measured simultaneously for feldspars (albite, oligoclase), diatomaceous earth and complex authentic samples, Iowa loess and Saharan sand. Particle shape effects on the IR resonance features were investigated using simulations assuming different specific particle shapes based on Rayleigh theory. Rayleigh-theory based simulations for a needle-like or CDE particle shape model appear to better overlap with the experimental line shape data for feldspars than Mie theory, although the fits are still not especially good. This may be due in part to uncertainties in the optical constants for feldspars. The Rayleigh solution assuming a CDE gives the best overlap with the extinction spectra of diatomaceous earth. Diatomaceous earth consists of amorphous silica particles so our result is in agreement with previously published results where it was shown that the CDE model gave a good fit for non-clay silicates.
 The mineralogy of authentic mineral dust was inferred from IR extinction spectra and the comparison of calculated and experimentally measured elemental analysis confirms the quality of the derived mineralogy. The measured size distributions were combined with optical constants from the literature to generate model simulations, which are compared to the measured IR spectrum. Mie theory and Rayleigh theory, with appropriate particle shapes for each component, were used to simulate IR spectra. It was found, for the samples studied here, that the silicate resonance peak is blue shifted by more than 40 cm−1relative to the Mie theory prediction. In previous studies, it was shown that for several clay (kaolinite, montmorillonite, and illite) and for several representative non-clay (quartz, calcite, and dolomite) components of atmospheric aerosol the silicate stretch peaks in Mie simulations are also blue shifted compared to the experiment. Spectral simulations of authentic dust using the derived mineralogy and a simple Rayleigh-based analytic theory for appropriate shape models for each component better reproduce the experimental spectra.
 It is somewhat surprising that analytic solutions based on Rayleigh theory give such good results for mineral dust aerosol particles even though the Rayleigh condition (D ≪ λ) is not quite fulfilled in these experiments. However, both theories are approximations. Mie theory is really only valid for homogeneous spheres; Rayleigh theory is only valid for very small particles. It is obvious that for particles in this size range the application of Mie theory to nonspherical particles are considerably worse than the use of the Rayleigh theory in a regime where the value of dimensionless size parameters x is not much smaller than 1.
 Errors in the peak position and line shape associated with Mie theory simulations could unfavorably affect determination of composition of mineral dust using the high-resolution IR data from satellites. Radiative forcing calculations may also be adversely affected by the errors in the integrated absorbance in the prominent Si–O resonance region. It was shown that Mie theory over predicts BT11–BT12difference data. Simple analytic models, derived in the Rayleigh regime, offer a better fit in the prominent resonance region to interpret high-resolution IR data. Based on the results of this study it can be suggested that the accuracy of modeling atmospheric dust radiative transfer effects in the IR can be improved by using Rayleigh theory for particles in the accumulation mode size range. We expect that more sophisticated calculations of the extinction properties of mineral dust aerosols using T-matrix theory, the discrete dipole approximation method or the assumption of nonsymmetric hexahedral shape might offer a more reliable results and might also be extended to a larger size distribution [Draine and Flatau, 1994; Mishchenko et al., 2002; Bi et al., 2010]. We plan to perform calculations based on these methods in future studies.
 The authors would like to thank Jennifer Alexander for helpful discussions and technical assistance. This research was supported by the National Science Foundation under grant AGU-096824. 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.