Mineral dust aerosol plays an important role in the Earth's radiative budget on both regional and global scales. To better understand the impact of this component of the Earth's atmosphere, the extinction spectra for several key components of mineral dust aerosol have been measured in an environmental aerosol reaction chamber. The extinction spectra are measured over a broad wavelength range, which includes both IR (650 to 5000 cm−1) and UV-Vis (12,500 to 40,000 cm−1) spectral regions. Experimental data are compared with Mie theory simulations derived from available literature optical constants. In a few cases, we have needed to modify the published optical constant data sets to ensure Kramers-Kronig consistency. In general, the Mie-based simulations are in excellent agreement with experimental data over the full IR-UV spectral range, except in the immediate neighborhood of the IR resonance absorption lines where particle shape effects on the resonance line profiles can be significant.
 Mineral dust aerosol is of great interest because of the effect it can have on global and regional climate [Tegen et al., 1996; Haywood and Boucher, 2000; Satheesh and Moorthy, 2005]. Annual mineral dust aerosol emission fluxes into the atmosphere range between 800–1500 Tg, as estimated by recent modeling studies [Bauer et al., 2004]. Mineral dust can affect the global radiative budget by both direct and indirect means. Mineral dust aerosol can have a cooling (negative forcing) effect on the atmosphere by scattering incoming solar radiation at short wavelengths, and a warming effect (positive forcing) by absorbing outgoing terrestrial radiation at long wavelengths. Thus mineral dust aerosol impacts temperature profiles, photolysis rates and atmospheric dynamics. Estimating the effects of mineral dust on climate forcing requires accurate modeling of dust optical properties across the entire spectral range from infrared (IR) to the ultraviolet (UV), including both nonresonant scattering and resonance absorption regions [Sokolik and Toon, 1996; Sokolik and Toon, 1999; Quijano et al., 2000; Conant et al., 2003]. Unfortunately, the radiative forcing of mineral dust aerosol in the troposphere is not well understood; even the sign of the direct radiative forcing contribution due to mineral dust aerosol is uncertain [Haywood and Boucher, 2000; Myhre and Stordahl, 2001]. Indeed, the impact of aerosols remains one of the most uncertain forcing components in climate models [Forster et al., 2007].
 Mie theory is commonly used to model key optical properties of mineral dust, such as optical thickness, single scattering albedo and asymmetry factor [Hess et al., 1998; Ramanathan et al., 2001]. Mie theory requires knowledge of the dust particle concentration, size distribution, and index of refraction (the optical constants) [Sokolik and Toon, 1999]. Optical constants have been measured and tabulated for many of the most abundant components of mineral dust aerosol, but these are usually measured under bulk conditions and the results have generally not been validated in experiments on aerosols. In addition, optical constants in different spectral regions (such as in the IR and in the visible) are usually measured in different laboratories using different experimental methods. In some cases, the data sets are in disagreement where they overlap and this can present problems when attempting to model optical properties across a broad spectral range.
 Even in cases where the optical constants are well established, Mie theory is strictly valid only for homogeneous spheres. It is known from both experimental and theoretical studies that the use of Mie theory can lead to serious errors in some cases [Kalashnikova and Sokolik, 2002, 2004], particularly for angle resolved studies of light scattering [Mishchenko et al., 1997; Volten et al., 2001; Veihelmann et al., 2004; Kahnert et al., 2006]. T-matrix based calculations on distributions of randomly oriented spheroids suggest that the errors associated with using Mie theory to calculate angle-integrated properties such as total extinction may be small [Mishchenko et al., 1997]. However, these conclusions have not been widely tested against laboratory measurements of extinction for well-characterized particles across a broad spectral range that includes both nonresonant scattering and resonant absorption regions.
 Rayleigh-Gans theory offers an alternate approach that might be used to model the effects of larger irregularly shaped particles and particle aggregates [Bohern and Huffmann, 1983; Wang and Sorenson, 2002; Garcia-Lopez et al., 2006]. This method requires a model for particle shape (the form function), which is not always available for atmospheric mineral dust aerosols. While the Rayleigh-Gans method may have advantages over Mie theory for large, irregular particles, it has not yet been widely implemented in atmospheric radiative transfer calculations.
 In this study, IR and UV-Vis extinction spectra have been measured in an environmental simulation chamber for several key components of mineral dust aerosol, including silicate clay (illite, kaolinite, and montmorillonite), oxide (quartz and hematite), carbonate (calcite), and sulfate (anhydrite) constituents. Aerosol optical properties can then be investigated over the spectral range from the IR to the UV. Experimental spectra are compared with Mie theory-based simulations that use optical constants derived from published literature data sets and a best fit lognormal size distribution function. This comparison allows us to test, both the reliability of Mie theory across the full spectral range from the IR to the UV (including both resonant absorption and nonresonant scattering regions), and the database of published optical constants for these important mineral constituents.
2. Experimental Methods
2.1. Environmental Aerosol Chamber
 An environmental aerosol reaction chamber is used to measure the extinction spectra for selected components of mineral dust aerosol. A schematic diagram of the chamber is shown in Figure 1. It is an FEP-Teflon coated steel chamber with 151 L volume and a surface to volume ratio of 10.7 m−1. It is sealed by large flanges at the top and bottom, and is fitted with eight sidearms to access the interior of the chamber. A mechanical pump is connected through an opening near the bottom of the chamber. The pressure in the chamber is measured with capacitance manometers. All the experiments in this study are carried out at room temperature, ca. 296 k.
 The mineral dust components investigated in this study were obtained from several sources. The manufacturer information and specifications, along with the literature references for the optical constants used for the different mineral dust aerosol samples, are given in Table 1. Note that only well-characterized single component mineral samples are investigated in this work. The samples have each been characterized by X-ray diffraction to verify the mineralogy. Results for more authentic mineral dust samples that are more complex internal or external mixtures of minerals will be reported in future work.
Table 1. Sources and Literature References for the Optical Constants of the Components of Mineral Dust Aerosol Used in This Study
 An Ocean Optics UV-Vis spectrometer (Ocean Optics, SD 2000) is used to measure the extinction in the range 12,500 to 40,000 cm−1. The UV-Vis spectrometer beam is coupled through quartz windows to the chamber using optical fibers with matched collimating lenses. An adjustable aperture of diameter ∼8 mm is placed in the beam path, immediately before the final focusing lens, to limit the field of view and discriminate against near forward scattered light. Even with the aperture in place, we cannot entirely rule out some error in the UV-Vis extinction signal at short wavelengths resulting from the collection of very near forward scattered light from large particles. The window-to-window path length for the UV-Vis beam is 66.5 ± 0.5 cm.
 A Mattson Infinity 60 AR FT-IR spectrometer is used to measure the infrared extinction spectra in the range of 650–5000 cm−1. The IR beam is directed through Germanium windows mounted on the chamber in a single-pass configuration, using a series of gold-coated mirrors and an off-axis parabolic mirror to focus onto an external liquid-nitrogen cooled, MCT detector. The path length for the IR beam is 58.5 ± 0.5 cm.
 The IR and UV-Vis spectroscopic probe beams are perpendicular to each other, and lie in the same horizontal plane. The data integration times for the instruments are also adjusted so that they are matched. Since the IR and UV-Vis spectrometers are collecting data over the same time intervals, and the probe beams lie in the same horizontal plane in the chamber, the two instruments sample the same particle distribution. Note that the instruments used in these studies leave a gap in the spectral coverage between ∼5000–12,500 cm−1. While it is possible to modify the FTIR source and detector to better cover this near IR gap, we would sacrifice the IR spectral range <2000 cm−1 that encompasses the resonance absorption lines.
2.3. Experimental Protocol
 The chamber is evacuated to its base pressure of ∼10 mTorr, at the beginning of the experiment, using a mechanical pump. The chamber is then filled to atmospheric pressure with purge air, generated in the laboratory. A background reference spectrum is taken with both spectrometers before the aerosol is introduced into the chamber. The mineral dust sample under investigation is typically held under vacuum for two hours prior to the experiment to remove water. Montmorillonite is a swellable clay and retains water; in this particular case the sample holder was heated and pumped overnight in order to better remove the residual water.
 The mineral dust sample is rapidly introduced into the chamber by pressurizing the sample holder up to 100 psi with an inert gas, and activating a pulsed solenoid valve between the chamber and sample holder. An impactor plate is placed in the flow path, directly behind the nozzle, to ensure efficient deagglomeration of the sample. The mixing time of the chamber following sample introduction is less than ∼1 minute. Extinction is measured with both the spectrometers after the dust is well mixed in the chamber, typically after about 10 minutes. While there may be some variation in the particle size distribution and concentration as a function of height in the chamber due to sedimentation, persistent local density gradients in a horizontal plane are unlikely to be important after this mixing time.
 Extinction is measured simultaneously in both the IR and UV-Vis. Both spectrometers are set to collect data in the same horizontal plane and over the same 53 s time period. For the FTIR this corresponds to an average of 256 scans at 8 cm−1 resolution. For the UV-Vis, the integrated signal corresponds to an average of ∼3300 individual scans over the period. The effects of any short-term density fluctuations on the spectra are averaged out during the scan averaging. These 53-s data collection periods could then be repeated many times as the dust settled. To account for the differences in path length between the two beams, the extinction signals measured by the different spectrometers are scaled and normalized to a constant 100 cm path length. Reproducibility in the results was ensured by repeated measurements of the extinction spectra over successive days. Spectra taken at different times or on different days show no significant spectral variations when normalized to the same peak intensity.
 We have measured extinction spectra for several of the abundant components of typical mineral dust aerosol including the silicate clays, illite, kaolinite, montmorillonite, as well as quartz, hematite, calcite, and anhydrite, over the spectral range covering the IR (650–5000 cm−1) and UV-Vis (12,500–40,000 cm−1) regions [Claquin et al., 1999]. Optical constants derived from the published literature and assumed lognormal size distribution functions are used to calculate a Mie theory extinction spectrum for comparison.
 The Mie extinction for aerosol particles in air is given by
where an and bn are the scattering coefficients given by Bohern and Huffmann  in terms of Ricatti-Bessel functions, k = (2π/λ), where λ is the wavelength, m = n + iκ is the complex index of refraction, and X is the dimensionless size parameter (X = πD/λ, where D is the particle diameter).
 In the Mie simulations, refractive index data tabulated in the published literature are used as a basis for the calculations. The literature references for the optical constant data sets used are given in Table 1. Optical constants from these reference sources are often used to model the radiative properties of mineral aerosols [Sokolik and Toon, 1999]. Because these measurements cover such a broad wavelength range it is generally necessary to combine optical constants from multiple sources, measured in different spectral regions. These tabulated optical constants have been measured on different samples using different methods. Perhaps not surprisingly, in some cases the optical constants are in some disagreement in the spectral regions where the data sets overlap (e.g., kaolinite and calcite). In such cases the mismatch in optical constants can lead to an unphysical discontinuity in the simulated spectrum (vide infra). In addition, the Kramers-Kronig relations are clearly not satisfied. This situation may also be encountered in efforts to model the effects of real mineral dust aerosol in radiative forcing calculations. For such cases, we have made an attempt to provide a smoothed and Kramers-Kronig consistent set of optical constants for the simulation. This procedure is outlined below.
 In a typical case, a table of optical constants that have been measured in the IR, up to ∼4000 cm−1, must be connected to a different set of optical constants measured across the UV, visible, and near IR, down to ∼4000 cm−1. Thus E ∼ 4000 cm−1 is the typical “overlap region” for these data sets. In most cases, there is very low absorption in this overlap region, and the imaginary parts of the refractive index values for the two different data sets are small and not too dissimilar. However, in a few cases there is an appreciable mismatch in the real part of the refractive index at this point. In such a case, the imaginary part of the refractive index is assumed correct (averaging the small values from the two data sets near 4000 cm−1, as needed). The imaginary component of the refractive index and the subtractive Kramers-Kronig relations are then used to generate a new real part for the refractive index. The subtractive Kramers-Kronig dispersion principle relates real and imaginary part of index by the following equation [Clapp et al., 1995]:
where ν is frequency, n(ν0) is the real index at a wavelength at some “anchor point” inside the integration region, and P indicates the Cauchy principle value.
 This calculation requires an anchor point, νo, where the refractive index is known. Without another independent measurement of the optical constants, there is some ambiguity about how to select this point. To choose the anchor point, in each individual case a judgment is made about which data set is deemed more reliable, and an anchor point is selected from that data set. The choice is inevitably somewhat arbitrary and open to debate. As a result, we cannot claim that the “new” derived optical constants are “correct”, only that they are (at least) consistent with the Kramers-Kronig relations. The choice of a different anchor point would lead to a different (shifted) real index. This would change some of the fine details of the fittings, but the essential conclusions drawn from this work would not be altered (vide infra).
 In addition, several of the mineral samples studied here (quartz, hematite, and calcite) are anisotropic (birefringent) materials with o-ray and e-ray optical constants for light transmission along different crystal axes. The accepted method for calculating the optical properties for a distribution of randomly oriented dust particles is to calculate individual spectra for both o-ray and e-ray indices, and then to take a weighted average of the spectra in the form [(2/3) o-ray + (1/3) e-ray], since there are 2 equivalent axes for the o-ray and 1 for the e-ray. [Bohern and Huffmann, 1983]. We refer to this as the spectral averaging (SA) method. Alternatively, while not rigorously justified, one could first average the o-ray and e-ray optical constants with a (2/3):(1/3) weighting, and then calculate the spectra from the averaged optical constants, which we refer to as the optical constant averaging (OCA) method. It is interesting to ask what errors may result in a practical case from using the numerically simpler, but less justified OCA method. For the quartz, hematite, and calcite samples here, Mie extinction is calculated using both the SA and OCA methods for comparison to experiment.
 In addition to the optical constants, Mie theory simulation requires knowledge of the particle size distribution. For this analysis, a lognormal size distribution function is used,
with three adjustable parameters, the particle number density N, mode diameter Dp, and distribution width σ. The tabulated optical constants (modified as described above if necessary) and an initial guess for the size distribution function are used to generate an initial Mie theory extinction spectrum for comparison with the experimentally observed spectrum. The root-mean square (RMS) difference between the simulated and experimental results is then calculated. The lognormal size distribution parameters are then iteratively adjusted in a nonlinear least squares fitting procedure to minimize the RMS difference. Because the IR resonances are not well modeled by Mie theory, the size distribution is fit by matching the spectra only over the nonresonant scattering range, ∼1500 cm−1 – 40,000 cm−1. For the purpose of adjusting the size distribution function, a simple linear interpolation is used to cover the gap region between the IR and visible data in the experimental spectrum (5000 cm−1 – 12,500 cm−1). The result is the ‘best fit’ lognormal size distribution function. This and the corresponding final best fit simulated extinction spectrum are given for comparison with the experimental spectrum for each dust sample. The best fit size parameters (N, Dp, σ) are summarized in Table 2. Note that the extinction due to scattering scales roughly like the projected surface area of the particle, and as a result larger particles in the tail of the lognormal distribution dominate the extinction. A more appropriate measure of the effective particle size in these experiments is the commonly used effective radius (projected surface area weighted radius), Reff, which is also given in Table 2 for reference.
Table 2. Lognormal Size Distribution Parameters Determined by the Least Squares Fitting Algorithm Discussed in the Modeling Sectiona
Mean Diameter (Dp), nm
Number Density (N), cm−3
Effective Radius, Reff, μm
For kaolinite, parameters are given for both the modified and unmodified literature (in parentheses) optical constants. For montmorillonite, parameters are given for both the combined Roush and Egan data sets and for the combined Querry and Egan data sets (in parentheses). See text for details.
4.34 × 106
2.39 × 106 (4.10 × 106)
1.42 × 105 (6.58 × 105)
1.31 × 107
6.10 × 106
5.47 × 106
4.87 × 107
5.70 × 104
 We note that individual optical constant data sets used in the fitting process may be in error. In addition, there may be source dependent variability in the silicate clay samples. These uncertainties in the optical constants will impact the accuracy of our derived size distributions; varying the optical constants would lead to somewhat different best fit size parameters. However, the size distribution is determined by fitting only over the nonresonant scattering part of the spectrum where (with the exception of hematite) the imaginary index is small and the scattering is determined primarily by the real index value. We have carried out sensitivity tests to determine the effect on the size distribution fits associated with variations in the real index value. This is most easily done using the SKK relation and changing the index value at the anchor point. In fact, while the best fit lognormal size distribution mode parameters (N, Dp, σ) do vary appreciably with changes in the real index value there is relatively little variation in the large particle tail of the distribution that is primarily important in determining the extinction spectrum, i.e., the derived size distribution is robust for diameters >∼700 nm. As a result, Reff is quite insensitive to changes in the optical constants. (See, for example, the results for kaolinite and montmorillonite in the next section.) In other words, while our experiment does not constrain the small particle part of the distribution well, the large particle tail of the distribution and the effective radius are well constrained. This result is consistent with idea that the extinction due to scattering for large particles is dominated by diffraction, which depends primarily on the particle cross-sectional area and is relatively independent of particle shape and refractive index [Bohern and Huffmann, 1983].
 The spectral fitting is based on the assumption that the particle size distribution function is lognormal, which may not be true in practice. However, the previous discussion also suggests that this assumption is not critical. Another distribution with a different shape for small particle diameters, but with a large particle tail that falls off similarly to the lognormal would lead to a similar fit to the extinction data.
 The experimental extinction spectra measured over the IR and UV-Vis spectral ranges are shown in Figures 2–4, together with the best fit Mie simulations. The Mie simulations are based on optical constants drawn from the published literature and the best fit lognormal size distribution (shown as an inset in each panel). The figures also show an expanded view of the IR region for each mineral aerosol sample. The absolute extinction signal levels vary significantly for the different dust samples. For clarity each individual spectrum is scaled independently. The scales are given along the vertical axes.
 For illite the optical constants in the IR [Querry, 1987] and in the near IR, visible, and near UV [Egan and Hilgeman, 1979] are in good agreement in the spectral region where they overlap (3850–4000 cm−1). These optical constants, together with an assumed lognormal size distribution, are used to simulate the extinction spectrum. The least squares fitting procedure described above yields the best fit lognormal size distribution parameters: mode diameter Dp = 107 nm, width σ = 2.57, and number density N = 4.34 × 106, yielding an effective particle radius, Reff = 0.50 μm. The size distribution is shown in the inset to Figure 2a, and the fit parameters are summarized in Table 2. The resulting best fit simulated extinction profile is shown in Figure 2a with an expanded view of the IR shown in the right panel.
 For kaolinite (and for montmorillonite below), the NIR - UV optical constants are also taken from Egan and Hilgeman , but there are two different IR data sets that can be evaluated [Querry, 1987; Roush et al., 1991]. For kaolinite the Roush and the Querry data sets are in good agreement in the major Si-O stretch resonance absorption region near 1050 cm−1, leading to very similar fits to the resonance profile. However, the Roush data extends only to 2000 cm−1, leaving a significant interpolation gap to the onset of the Egan data near 4000 cm−1. In addition, there is a weak kaolinite absorption band near 3650 cm−1 assigned to an O-H stretch mode that is then missed in the interpolation between the Roush and Egan data. For these reasons only the comparison using the Querry and Egan optical constant data sets for kaolinite [Querry, 1987; Egan and Hilgeman, 1979] are presented.
 Unfortunately, the Querry and Egan data sets for kaolinite do not match well in the overlap region near 4000 cm−1. Using a table of optical constants that simply splices these two data sets together yields a discontinuity in the real part of the index of refraction at the connection point. Using these combined optical constants and following the least squares fitting procedure described above, leads to a best fit simulation to the extinction data shown in Figure 2b for kaolinite. The best fit lognormal size distribution parameters are summarized in Table 2 (in parentheses). Note the step near 4000 cm−1 in the expanded IR view of the simulated spectrum (right panel of Figure 2b), which results from the mismatch in the optical constant data sets.
 To improve the simulation we have used the subtractive Kramers-Kronig relations as previously described to derive a smooth and Kramers-Kronig consistent set of optical constants. To select the anchor point, we note the reasonably good agreement between the Roush et al.  and Querry  IR data sets in the region where they overlap. On the basis of this agreement, the Querry data set is assumed to be reliable and an anchor point is selected from this data at 3000 cm−1 (m = 1.376 + i 0.049). With this choice, an improved set of optical constants that are Kramers-Kronig consistent is determined. (These modified optical constants are available upon request). The best fit size distribution parameters are then reevaluated using the modified optical constants and the iterative procedure described above, and the results are also given in Table 2. The resulting extinction spectrum is displayed in Figure 2b, and best fit size distributions are shown in the inset.
 Between the two sets of results for kaolinite (with the original and the modified optical constants) there are significant differences in the real index value in the near IR and visible. These differences in the optical constants lead to quite different best fit size distribution mode parameters. However, the tails of the distributions overlap well and the effective radius values are nearly identical.
 For montmorillonite, we similarly find that the Querry and Egan optical constant data sets do not match well in the overlap region near 4000 cm−1 [Querry, 1987; Egan and Hilgeman, 1979]. In this case the “best” spectral fit is very poor over the IR. The results are shown in Figure 2c and the best fit size parameters are given in Table 2 (in parentheses). We have attempted to use the subtractive Kramers-Kroenig procedure outlined above to generate an improved set of optical constants for montmorillonite based on the Querry and Egan data sets, with no real improvement in the quality of the spectral fit through the IR. (The “kink” in the data at 4000 cm−1 can be eliminated but the overall fit through the IR is still very poor.) This could indicate that the Querry optical constants may be in error. However, it may also simply reflect differences in the montmorillonite samples that have been used in these experiments. Montmorillonite does not define a unique chemical structure but refers to a class of smectite clays that can show significant variability.
 The Roush optical constants for montmorillonite, however, lead to a much improved spectral fit. In this case we simply combined the Roush data set in the IR with the Egan data set in the NIR -UV, with a linear interpolation through the gap, 2000–4000 cm−1. This leads to the results also shown in Figure 2c, with the size distribution parameters as shown in Table 2. Again note that despite the significant differences in mode parameters for the two size distributions used in this analysis, the effective radius parameters are not markedly different. Also, note this montmorillonite size distribution is shifted toward much larger particles than the kaolinite or illite clay samples.
4.2. Oxides (Quartz, Hematite)
 Quartz optical constants (both e-ray and o-ray) are available in the literature over the IR and near IR, visible and UV [Longtin et al., 1988] spectral ranges. The least squares fitting procedure discussed above is used to find the best fit lognormal size distribution parameters using the OCA method, since it is numerically simpler and more readily applied in the nonlinear least squares fitting procedure. The best fit size lognormal size distribution parameters are N = 1.31 × 107 cm−3, Dp = 22 nm, and σ = 3.89 for the number density, mode diameter, and distribution width, respectively. These values are summarized in Table 2, and the distribution is shown in the inset to Figure 3a. In this case the mode diameter seems anomalously small. However, recall that our experimental method really only constrains the large particle part of the distribution well, and it is the large particles that dominate the extinction spectrum. In this case the effective radius for the distribution is a very reasonable Reff = 1.1 μm. The Mie extinction for quartz is then calculated by both the averaging methods for the optimized size distribution. The comparisons between Mie extinction (using both the SA and OCA methods) and the experimentally measured extinction spectra are also shown in Figure 3a.
 A similar study has been carried out for hematite. The optical constants (e-ray and o-ray) for hematite have been published by Longtin et al. . As hematite is a birefringent material, the extinction is calculated using both averaging methods, and the comparisons are also shown in Figure 3b. The comparison between the measured and best fit theoretical size distributions using the OCA method is shown in the inset of Figure 3b.
 Note that the calculated spectrum for hematite shows clear evidence for interference fringes across much of the visible portion of the spectrum. The interference fringes result from the ‘spherical particle’ assumption inherent in Mie theory. Interference fringes are common and expected in calculations with a narrow distribution of particle sizes. Here they are clearly evident even for the broad distribution of particle sizes shown in Figure 3b. It is interesting that the Mie results for the other samples under study do not show such interference fringes, which are washed out in the average over particle size in the highly dispersed samples used. This difference can be rationalized on the basis that, for hematite, the index of refraction is very different in both the real and imaginary parts from the other materials studied; in particular, the imaginary part of the index begins to rise rapidly in the visible as a result of the onset of a strong visible absorption band. The experimental spectrum shows no evidence for interference fringes because the experiment averages over a distribution of actual particle shapes.
 Because hematite is a strong absorber in the visible and near UV, it is particularly important in the atmospheric radiative balance. In our apparatus we cannot determine experimentally the ratio of scattering to absorption (the scattering albedo), but we can extract this information from the Mie model results. The model results for the absorption and scattering contributions to the extinction are shown in Figure 5.
4.3. Sulfates and Carbonates (Anhydrite, Calcite)
 Anhydrite is a common sulfate component of mineral dust aerosol. The optical constants in the IR (3–4000 cm−1) have been determined by Long et al. , and in UV/Vis (4000–50,000 cm−1) by Ivlev and Popova . In the overlap region near 4000 cm−1 the data sets are in reasonably good agreement, and so we simply use the combined data sets without any modification. The simulated extinction spectrum, calculated from Mie theory and using the best fit size distribution, is compared to the measured extinction spectrum in Figure 4a. The best fit size distribution parameters for anhydrite are given in Table 2.
 In our study on calcite, we have measured the extinction for two different calcite particle samples, with very different particle size distributions, a “small particle” sample (OMYA Products) and a “large particle” sample (EM Science). The extinction spectra for the two samples, shown in Figures 4b and 4c are markedly different. The large particle sample has a smaller extinction over the entire spectral range for the same mass loading in the chamber when compared to that of the small particle sample. The large particle sample also shows a much earlier onset and steeper rise in the short wavelength scattering part of the spectrum.
 Calcite is birefringent. The e-ray and o-ray optical constants have been measured in the IR by [Lane, 1999] and in the near IR, visible, and near UV by [Ivlev and Popova, 1973]. In this case the two data sets do not match well in the overlap region near 2000 cm−1. As discussed in previous section for the cases of kaolinite and montmorillonite, this mismatch in the optical constant data set leads to an unphysical kink in the extinction spectrum in the overlap region. In order to improve the fit we have used the subtractive Kramers-Kronig relations to generate a modified and Kramers-Kronig consistent data set for both the e-ray and o-ray optical constants. For calcite an anchor point for the index in visible spectral region at 18,182 cm−1 is chosen since calcite optics are commonly used in UV-Vis spectroscopy and the optical properties are well determined in this region.
 For both the large and small calcite particle distributions, the modified optical constants are used in a Mie simulation to optimize the size distribution for comparison to the experimental extinction spectra in Figure 4. Once the size distributions have been optimized (using the OCA method to handle the birefringence), we show the comparison between the measured and simulated extinction spectra for both SA and OCA methods in Figures 4b and 4c. For the small particle calcite sample a best fit lognormal size distribution (shown in the inset to Figure 4b) with fit parameters N = 4.9 × 107 cm−3, Dp = 52 nm, and σ = 2.68 is determined, resulting in an effective radius of Reff = 0.30 μm. For the large particle calcite sample the best fit parameters are N = 5.7 × 104 cm−3, Dp = 271 nm, and σ = 3.06, resulting in an effective radius Reff = 3.1 μm, roughly a factor of 10 larger. These very different size distributions lead to the distinctly different spectra observed in Figures 4b and 4c.
 As seen in Figures 2–4, the measured extinction spectrum for each mineral sample matches well with the Mie simulation over the entire spectral range from ∼1500–40,000 cm−1, corresponding essentially to the nonresonant scattering part of the extinction spectrum. Note that the optical constants are fixed in these calculations and the only adjustable parameters are N, σ, and Dp, which characterize the lognormal size distribution function, making this a 3 parameter fit to the data over the full spectrum in each case.
 It is also worth noting that we have an additional experimental check on the best fit size distributions that we derive. Recall that the size distributions are determined by fitting the spectra over the nonresonant scattering regions. However, the resonance absorption line strengths calculated using the best fit size distributions are also in reasonable agreement with the experimental line strengths. Since the resonance absorption depends on total aerosol mass, this result essentially verifies that the total aerosol mass in the best fit size distribution is roughly correct in each case.
 We now turn to a general discussion of the extinction results for these samples across the full IR-UV spectral range. In every case, the Mie simulation is in good agreement with the measured extinction throughout the non-resonant scattering regions of the spectrum from the IR to the UV. Although the particles are not spherical, Mie theory predicts the nonresonant spectral regions very well suggesting that it can be used with confidence to calculate the extinction for nonspherical particles in the spectral regions away from any sharp absorption features. This is in agreement with T-matrix calculations by Mishchenko et al.  that suggest that the errors in using Mie theory to describe angle-integrated properties such as total extinction are small in the nonresonant scattering regions. Interestingly, even for hematite, which has a strong visible and near UV absorption band, the extinction fit is quite good if one averages through the nonphysical interference fringes in the Mie theory simulation. For the birefringent materials (quartz, hematite, and calcite), the extinction spectra calculated by both averaging methods match well with each other and with the experimental result, suggesting that the averaging method has relatively little effect on the extinction in the non-resonance scattering regions.
 Note that this range of effective particle sizes studied in this work (Reff ∼ 0.3–3 μm) is primarily relevant for assessing the optical properties of mineral dust in the accumulation mode size range, typical of background aerosols and aerosols that have undergone long-range transport. Field studies during African dust events have shown that up to about of the total dust mass collected in the southeastern US and Carribean was less than ∼2 μm in aerodynamic diameter [Moulin et al., 1997; Li-Jones and Prospero, 1998]. Furthermore, studies from different geographic areas have determined that the mass median diameter of mineral dust found over the oceans falls generally in the 2–3 μm range [Prospero et al., 1989; Prospero, 1999]. For comparison, the mass weighted mean diameter for our kaolinite sample is ∼3 μm. Thus our results are primarily relevant to modeling of long-range transport aerosols.
 While the spectral fit in the nonresonant scattering regions is quite good, there are clear discrepancies in the fits to the IR resonance absorption lines apparent in the right panels to Figures 2–4. These discrepancies are due to the nonspherical nature of the mineral dust particles in our samples. It is well known that particle shape effects can be especially significant in the neighborhood of infrared resonance absorption lines, and can have a large effect on resonance line positions and line shapes [Bohern and Huffmann, 1983]. Mie theory often fails to correctly predict resonance line profiles and this can be important in modeling the effects of dust in data retrievals from high spectral resolution satellite measurements.
Hudson et al.  have recently carried out a quantitative study of particle shape effects on IR resonance absorption line profiles for a series of silicate clay mineral dust aerosol particles that fall in the accumulation mode size range (D = 0.1 − 1.0 μm). They find that Mie theory gives a poor fit to the resonance line positions, line shapes, and integrated areas. It was also pointed out that this has potentially important consequences for determining atmospheric dust loading and composition from narrow band or high-resolution satellite data. Dust retrievals based in Mie theory could be in error if the overlap between the narrow band sensor channels and the actual dust absorption resonance is different than predicted by Mie theory. Hudson et al.  also suggest that simple analytic model results, derived in the small particle limit, for absorption by particles with characteristic shapes (disks, ellipsoids, needles, etc.) may offer a better fit to the resonance line absorption profiles than Mie theory, at least for mineral dust aerosols with diameters, D < ∼1 μm. Such models are commonly used in the astronomy literature to account for particle shape effects on line profiles [Fabian et al., 2001; Min et al., 2003]. Our experimental arrangement allows us to sample larger effective particle diameters than the experiment of Hudson et al. . We will explore particle shape effects on the IR resonance line profiles, as a function of particle size and composition in a forthcoming paper.
 Extinction spectra for several important components of mineral dust aerosol including the silicate clays (illite, kaolinite, montmorillonite), as well as quartz, hematite, calcite, and anhydrite have been experimentally measured across a broad spectral range from the IR to the UV. A Mie theory based simulation of the extinction spectra, using optical constants derived from published data sets and an assumed lognormal size distribution has been carried out. For kaolinite and calcite there are discrepancies in the published data sets that lead to unphysical kinks in the simulations. In these cases the literature data sets have been modified to be internally consistent with the Kramers-Kronig relations.
 Extinction spectra simulated by Mie theory are in good agreement with the measured extinction spectra over the nonresonant scattering region of the spectra from the IR to the UV, for these components of mineral dust aerosol. Although the particles are not spherical in shape, Mie theory predicts the nonresonant spectral regions very well suggesting that it can be used with confidence to calculate the extinction for nonspherical particles in the spectral regions away from any sharp absorption resonances. As noted, this experimental result confirms the conclusions from T-matrix based theoretical calculations, suggesting that angle integrated properties such as extinction are not particularly sensitive to particle shape effects, at least in the nonresonant scattering regime.
 This material is based upon work supported by the National Science Foundation under Grants ATM-042589. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the view of the National Science Foundation. The authors would like to gratefully acknowledge helpful discussions with Dr. Paula Hudson.