1. INTRODUCTION

Minerogenic particulate matter is an optically dominating constituent in coastal waters, inland lakes, and rivers. It also plays an important role in the ocean, where, e.g., deposition of atmospheric dust is seen as a major source for micronutrients [e.g., iron (Fe)] in oligotrophic waters. Therefore, the optical properties of minerogenic material suspended in seawater have recently been a focus of research [1]. Indeed, pioneering work on properties of pure minerals and other mineral-rich matter in suspension conducted by Babin, Stramski, and co-workers [2–6] determined mass-specific optical properties for light absorption, identified possible major absorbing compounds (iron oxides), and investigated the role of particle size for these properties [5]. For most materials, the absorption is high in the ultraviolet (UV: 200–400 nm) and very low in the near-infrared radiation (NIR) regions (NIR: 780–3000 nm). The extent of the absorption coefficient at near-infrared wavelengths for minerogenic material and that of natural particle assemblages in coastal waters is still uncertain [1,2,7] and it needs more accurate information about the related optical properties for a better understanding of NIR absorption by natural suspended particles [5]. This study is intended to achieve a better understanding of NIR absorption by mineral particles.

Many materials investigated by the work mentioned above [2–6] had mass-specific absorption coefficient in the NIR region below the detection limit of the optical methodology used (1 cm cuvette inside an integrating sphere). Other materials, however, showed significant values of the coefficient of on average ${0.02}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 800 nm [4]. On the other hand, natural particle assemblages are reported to have rather high absorption and specific absorption coefficients, with values of ${0.02}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ [1], or up to ${2}\;{{\rm m}^{- 1}}$ for the absorption coefficient at 750 nm found in the River Elbe [7]. In [7], negligible NIR absorption was rarely found and only in cases where biogenic material (phytoplankton) dominated the particle assemblage. It should be noted that even when phytoplankton, which possess very weak NIR absorption, is dominating, significant NIR absorption is often observed. Fortunately, mass-specific values of either the natural particulates or the strongest in the NIR absorbing mineral-rich material are quite similar, as the high absorption coefficients in rivers are accompanied by high particle mass concentration in these rather turbid environments.

In terrestrial environments, spectral remote sensing or, more generally, spectral analysis for the discrimination of different minerals, is based on spectral differences in optical properties, here mainly in light absorption properties, and this is normally based on received reflectance spectra of the material alone. In aquatic environments, where minerogenic particles occur suspended in water, the absolute mass concentration and spectral discrimination are equally interesting and can only be determined in optically complex waters with very accurate optical models that account for the optical properties the different materials. As pointed out by Stramski and Mobley 1997 [8], a mechanistic understanding of the optics in water demands information about each individual optical compound in the water.

For discrimination of minerals and rocks, the full NIR spectral region is typically used in addition to the visible radiation region (VIS: 400–780 nm) [9]. Accurate optical properties of suspended minerogenic material in natural water might help to identify specific minerals and to improve optical modeling in very turbid mineral-dominated environments. Accurate optical information is therefore necessary for the NIR spectral region, while light absorption here will always be much lower than in the ultraviolet and visible regions, except for rather black material, for which, however, important examples exist that might play a role in natural waters (black carbon, magnetite). Reflectance measurements have shown that many minerals possess specific, broader or narrower, absorption bands, associated either with absorption maxima of water and water structures (as well as O-H groups) in the crystals or with specific metal elements in the crystal matrix, e.g., from iron (see [9]). The absorption coefficient of suspended particles is rarely determined for the NIR spectral region but would allow optical modeling of such suspensions. At the same time, a spectral analysis to optically identify specific minerals in natural particle suspensions could be used, keeping in mind that the strong light absorption by water in the long-wavelength NIR region (${\gt}{1000}\;{\rm nm}$) will here never allow any in situ measurements. The objectives of the work presented here are (1) to determine mass-specific absorption coefficients of common minerals and mineral-rich material in the full NIR spectral region using a much higher accuracy, thereby achieving a much lower detection limit when combining an accurate methodology with a method where particles are concentrated on filters to increase sensitivity; (2) to get full spectral, UV to NIR, mass-specific absorption coefficient; and (3) to confirm upper limits of these mass-specific coefficients for common minerals in the short-wavelength NIR (780–1000 nm), a spectral region that is still relevant for optical remote sensing approaches in turbid coastal waters. It should be noted that particle size strongly influences mass-specific absorption coefficients. Thus, size is reported here and measurements of the dependence on size will be reported later.

2. MATERIALS AND METHODS

A. Mineral Materials

Different natural and artificial pure mineral materials are examined. The materials chosen are either important for natural waters (like laterite earths [LAT] [10]), were used in earlier work (source clays [3,5,10]), or have physical properties that can be assumed to be constant in terms of shape, density, refractive index, or mineral composition (like tourmaline or malachite). These materials included a red quartz (QTZ), a colorless quartz (DOR), feldspar (FES), mica (MIC), olivine (OLI), a steatite (STE), several laterite earths, a black tourmaline called “Schörl” (TUR), malachite (MAL), magnetite (MAG), artificial silicon carbide (SIC), and several source clays. Most of the materials (QTZ, MIC, OLI, LAT, MAG) were chosen as they can likely end up in natural waters as minerogenic suspended detritus via erosion and fluvial transport and are, in specific regions, often part of natural suspended material. The laterite earths are, e.g., important for coastal waters around New Caledonia, as they are transported via rivers after heavy rainfall events on the island [10]. TUR and MAL are chosen, as this material is rather homogeneous with very small crystal sizes allowing size fractionation without changing the material composition for the different size fractions. Table 1 summarizes all mineral materials used, their color appearance, origin, and particle size distribution.

Table 1. Mineral Materials and Their Particle Sizes

View Table | View all tables in this article

B. Size Fractionation

Data presented here shall provide general absorption data of different mineral materials. As the mass-specific absorption coefficient of particulate matter is particle size dependent, the specific size distribution of each material shall be known. When possible, the absorption data are presented for a specific size fraction only, generally of a size of around 30–40 µm, as this is the smallest fraction that can easily be produced by sieving. For each pure mineral material (MAL, MIC, QTZ, MAG, OLI, and TUR) an optical homogeneous block or an optical homogenous sample of small crystals was first crushed using a clean metal hammer and a titanium plate to produce a fine granulate/sand (${\lt}{500}\;{\unicode{x00B5}{\rm m}}$). The mineral material was covered by a robust tissue to avoid contamination by metal material from the hammer or the plate. The crushed material was milled for 1–2 h using a simple ball mill with ceramic balls to produce even finer particles (all material used for this milling step is not very hard and contamination by abrasion from the ceramic balls is not expected). The LAT material was already in the form of sand and could be easily broken up by hand at which point it was smaller than the pore size of the smallest sieve. The milled material was fractionated for particle size by wet sieving using a vibratory sieve tower (Retsch Vibro) equipped with a set of different sized metal sieves (Retsch sieves DIN 4188, pore sizes 20–350 µm) to produce 16 different size fractions according to the sieve sizes for a different experiment. For the experiment here, the material was collected from the smallest sieves (20 or 25 µm). If not enough mass was available in the smallest fraction (20–25 µm sieve sizes), the next larger fraction (25–30 µm sieve sizes) was used. The expected particle size was 22.5 to 32 µm (or 27–40 µm), but the distribution lastly depends on the shape of the particles. For MIC enough material was available for a further fractionation that was done using micro-gazes of 5 µm and 10 µm mesh sizes, and the fraction between 5 µm and 10 µm is used. The sieve with the retained material was washed with purified water and placed in a drying cabinet at 60°C for several hours. The dried particles were removed from the sieve and stored in small plastic vials until analysis. The particles were left in the plastic vials for several days to allow reactions with atmospheric oxygen to produce particles, as they would occur in natural suspension. Note that corrosion/degradation effects were observable for some minerals and might constitute a significant error source in the final determination of the absorption coefficient. However, the most significant reaction was found for the QTZ and OLI samples that consisted of several different crystals; milling did specifically destroy some black crystals and reduced their relative content. Nevertheless, we reported the results for QTZ and OLI for a general comparison. Drying at 60°C does not remove the water (crystalline or free) in the minerals, whereas drying at 105°C is normally used to remove free water associated with the crystals.

For the fine materials, such as the source clays, Arizona Test Dust sample (ATD), DOR, FES, and the laterite earths, the powders were used without a further size fractionation, as the particle sizes are already smaller than the available smallest pore size of a sieve. Only STE was dry fractionated using the 20 µm and 25 µm sieves. While further fractionation to just a few micrometers (see [3]) would better approximate particles in natural suspensions, the effort required to achieve such a size fractionation is significant, as masses of about 1 g are needed for the NIR absorption measurements. Furthermore, the optical methods employed here do not require particles to be in suspension for several minutes, making measurements with the easily available size fractions of 20–25 µm or 25–30 µm more practical. Another advantage of using larger particle sizes is that they can be dried on the sieves to produce enough material that can easily be used for an accurate mass determination when preparing the suspension.

C. Mass and Size Determination

When preparing suspensions from these particle samples, subsamples (1–500 mg) were weighed using a microbalance (Sartorius BP210D) that is calibrated shortly before measurements using standards masses (10 mg and 100 mg; VWR International) with a precision of $\pm 0.01\;{\rm mg}$.

Determination of the particle size distribution (PSD) of each final particle sample was performed using a laser diffraction meter (CILAS 1180). Suspensions of the different particle samples were prepared by suspending 100 mg material in 500 ml of purified water contained in clean glass bottles. The suspension was placed in an ultrasonic bath for 5 min immediately prior to the measurements to destroy particle aggregations. All measurements were made in triplicate and the sample cuvette in the instrument was flushed with purified water after each sample to ensure a clean system and to avoid retention of particles from the previous sample. The final PSDs were calculated by averaging the three replicate measurements for each sample. Exact sizes for the fraction of MAG (25–30 µm) and OLI (20–25 µm) were not determined due to the low amount of material available.

D. Spectrophotometric Measurements

As the determination of the particulate absorption coefficient with the quantitative filter technique (using wet and dry filters) is influenced by variation in the path length amplification factor and an often inhomogeneous particle distribution on the sample filters, accurate light absorption coefficients, ${a_p}(\lambda)$, of particle suspensions were determined in the visible spectral region (400–700 nm) using a point-source integrating-cavity absorption meter (PSICAM) [11,12]. Particle suspensions were measured after a preparation of the suspension directly in the cavity of the PSICAM. Potential sinking of particles would lead to a rather inhomogeneous distribution of the particles inside the PSICAM cavity, basically leading to particle self-shading at higher concentration when particles lay on the bottom of the cavity in several layers. To avoid sinking of particles, a white magnetic stir bar was placed at the bottom of the cavity, the PSICAM was placed on a magnetic stirrer plate, and the content of the cavity was stirred as strong as possible during measurements. Calibration of the PSICAM was done with the spinning stirrer bar in the cavity to adjust for the different reflectivity of the bar compared to the wall. For preparation of a suspension, the cavity was filled to 90% with purified water. Then 1.5 ml of purified water was added to the prepared, weighed subsample in a vial (1–500 mg), the vial was shaken and then placed in an ultrasonic bath for half a minute to destroy agglomeration of particles and to ensure a surface moistening of all particles (avoid particles sticking to the water surface due to surface tension). The treated sample was then added to the cavity while stirring, and the material was washed out of the vial with purified water several times. Lastly, the cavity was filled completely with purified water and closed with the regular stopper. The measurements were done at least in triplicate to determine the related uncertainty using purified water as the reference. Temperatures of the sample suspension and the reference fluid were near room temperature, but recorded for each sample and reference, and afterwards a temperature correction was made [11].

The mass-specific absorption coefficient, $a_p^*(\lambda)$ [${{\rm m}^{2}}\,{{\rm g}^{- 1}}$], of each sample was calculated as

Thereby the mass on the filter was exactly known and the volume used not important for the later calculation. Measurements with wet filters can be performed up to 900 nm. Due to the increasing absorption by water with wavelength, measurements at longer wavelengths are only possible after the free water in the filter was removed by drying the filters at 60° for several hours. All filter samples were measured using a Lambda 950 UV/VIS/NIR spectrophotometer (Perkin-Elmer) equipped with a 150 mm integrating sphere that had a photomultiplier and an indium-gallium-arsenide (InGaAs) detector, according to the method described in [13]. Test measurements were performed to achieve an optimal signal-to-noise ratio, a good spectral resolution (optimized by the slit width), and optical densities (OD) in the range of 0.01–0.15 (by adjusting the material mass on the filter). The scans typically start at the longest wavelength and the detector changes at 860 nm from the InGaAs detector to the photomultiplier. The light source changes at 320 nm from tungsten to deuterium. The photomultiplier’s sensitivity adjusts automatically, and the slit width was set to a spectral resolution of 2 nm at a wavelength ${\lt}{860}\;{\rm nm}$. The slit width above 860 nm is automatically regulated by the photometer in the so-called “servo mode” in response to the setting for integration time and detector gain. Using the “program mode” of the spectrophotometer, the gain here is set individually for different wavelength regions to keep the slit width between 2 and 4 nm. At ${\gt}{1900}\;{\rm nm}$ it exceeds 4 nm and sometimes was up to 8 nm. To achieve a good signal-to-noise ratio and a high spectral resolution, several preliminary tests were made to run a program that individually optimized the measurements for each NIR wavelength region and sample. Nevertheless, the detector change at 860 nm induces a significant artifact around these wavelengths that cannot be easily avoided and are still visible in some spectra. Visible absorption features around 860 nm need, thus, be interpreted with care.

To receive optimal optical densities for the wavelength range between 200 nm and 2500 nm, three different filters, with different loads of material, were prepared. The lowest filter load was used for measurements in the UV/VIS (200–500 nm), the medium in the VIS/NIR (350–1000 nm), and the highest load in the NIR (860–2500 nm) region. All filter loads were individually adjusted for each material to receive a sufficient optical density with values not exceeding 0.15 to avoid particle shelf-shading and, thus, to ensure the linearity of the path length amplification factor [13,14]. Due to this low OD, no filter was needed in the reference beam, as automatic adjustment of the photomultiplier gain would not lead to a significant change in signal to noise. Placing a reference filter was not possible in this integrating sphere arrangement. Instead the instrument was set to “autozero” with a dry filter in the sample beam (inside the integrating sphere) and “reference/blank” filters (either wet or dry) were measured as samples regularly and subtracted from the sample filter measurements. This is rather important in the short UV, where effects of small changes in the water content of the filter on the OD of the filter are significant. Here, filter-to-filter differences can be significant and absolute precision in the short UV is much less than in the other parts of the spectrum. For the UV region quartz fiber filters (MK360, Munktell) were used instead of glass fiber filter (GF/F, Whatman) as for the other spectral regions.

Similar filter-to-filter problems arise at ${\gt}{900}\;{\rm nm}$ with wet filters due to the strong absorption by water. As water absorption is very high in the longer-wavelength NIR, some tests were made to control the effects of air humidity onto filter taken out of the drying chamber for a longer time (1 h). No significant water absorption features were observed when such filters were scanned several times for the next 60 min. A full NIR scan (of a dry filter) typically required about 15 min.

E. Mass-Specific Absorption and Merging of Spectra

The mass-specific absorption from spectrophotometric measurements was first calculated as

Due to scaling to the PSICAM results, any wavelength-independent uncertainty of the spectrophotometric filter measurements is of minor importance, as long as our assumption of a constant path length amplification is valid. Therefore, the filter measurements were not done in triplicates; filter scans were simply repeated once to check for any errors during scanning or during sample filter preparation. The precision of the PSICAM absorption measurements was below $\pm 0.01\;{{\rm m}^{- 1}}$, its accuracy is estimated to be ${\lt}\pm 3\%$; uncertainties for the mass determination are also ${\lt}{1}\%$. Larger errors are expected for the final concentration due to the necessary sample handling after weighing, e.g., by small losses of material in the preparation steps; however, PSICAM measurements have an overall uncertainty for the $a_p^*$ of only $\pm 2\%$. Relative errors are larger for smaller values, if the absorption coefficient varies more than 3 orders of magnitude over the wavelength region for a single filter sample, i.e., the absorption coefficient approaches the detection limit in one wavelength region (typically in the infrared) while in another (typically in the UV/blue) it approaches the upper OD limit for the filter measurements. This detection limit for $a_p^*$ is not constant, as it depends on the mass used for each filter. One prerequisite for the approach used here is not to overload the filters, i.e., avoiding buildup of a significant layer of low-absorbing particles that would change the scattering of the sample filter compared to an empty filter. An upper limit of 500 mg was experimentally found for most low-absorbing materials. For this upper limit we can estimate a detection limit for the $a_p^*$ determination: variations in OD between filters and between sample and reference filters, even for dry filters, were observed to be in the range of $\pm 0.0015$; the standard deviation of the spectrophotometric measurements varied between $\pm 0.0003$ and $\pm 0.0006$ (for the here used photometer settings for slit width and scan speed). Using the higher value and a factor $k = 3$, the estimated detecting limit for the OD determination would be $\pm 0.0018$ ($k \times \sigma$; ${3} \times 0.0006$,99% confidence interval), quite similar to the value of observed variations. A theoretical detection limit of ${2} \times {{10}^{- 6}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ was determined by using a mass of 500 mg and an amplification correction factor of 1/4.5 (from [13]). Due to the above-mentioned spectral behavior of the material or for material having strong absorption maxima in the NIR or in the case when absorption was generally very high, a mass as low as 10 mg had to be used, and the detection limit was then only ${1} \times {{10}^{- 4}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$. In only one case (STE) the $a_p^*$ values were below the detection limit in a narrow spectral region (here 1500–2000 nm).

3. RESULTS AND DISCUSSION

The highest accuracy in the determination of $a_p^*$ was achieved with the measurements using the PSICAM in the region of 400–700 nm. The filter pad measurements were scaled to these results, by using individual path length amplification factors determined by a direct comparison of filter pad and PSICAM results and, in principal, using this factor for the other wavelengths and filter results of the same material equally. However, several error sources exist, and their effects have to be compensated. These are real differences in the amplification factor for different filters used for different spectral regions. Among these are the effect of drying the filter for the NIR region, the change due to an inhomogeneous distribution of particles on the filter, and small but still significant offset errors for each filter. An inhomogeneous particle distribution will alter the OD and this effect cannot be distinguished from an alteration of the path length amplification when the filter is measured a second time after manipulation. The long scanning time needed did not allow measuring all filters several times by using different parts of the filter to check for inhomogeneity. Additionally, it should be noted that minerals typically have high real parts of the refractive index (${\gt}{1.4}$). The absorption efficiency will change for a specific particle size when the refractive index of the surrounding medium changes, as here from about 1.33 for water to about 1.00 for air. To elucidate whether two spectra differ due to a measurement offset or due to different path length amplification, it was ensured that all three filter-derived spectra do spectrally overlap significantly (100–200 nm). The overlapping region was used to adjust all spectra to the PSICAM results by adjusting an offset value and amplification factor for each spectrum. The necessary assumptions are that both the offset and the amplification correction are not spectrally dependent for a single filter.

This spectral dependence was first examined for the difference induced by drying the sample filter for the VIS to short-wavelength NIR spectral range, as this region can be measured with a wet and dry filter. This comparison focuses on OD between 0.01 and 0.1. Lower values are not used due to significant relative errors for those values and higher values are not used due to the known non-linearity of the path length amplification with OD at those high values [13]. Except for three filters, all filters examined (${n} = 18$) showed a rather constant relative change of the OD after drying, i.e., a constant change of the path length amplification, with linear regression coefficients (wet versus dry filter results) of ${{r}^2}\gt {0.998}$. The extent of this effect was variable in the sense that the OD of a dry filter was higher, rather equal, or lower for single filters. The factors varied between 0.6 and 1.3 over all filters without a visible correlation to material type, particle size, or filter load. It should not be forgotten that a homogeneous distribution of the rather large particles on the filter is rarely reached. Taking this into account, drying the filter did not change the OD in a predictable way, such that when other uncertainties are accounted for, it is also valid to say that drying did not have a systematic effect on the spectral distribution of the OD, i.e., the effect filter drying had on the path length amplification in the filter was not wavelength dependent.

The path length amplification observed in the direct comparison of PSICAM and filter pad measurements was also in most cases constant over all wavelengths (and OD), and non-linearity was rarely observed. A non-linearity can be explained by small offsets in the filter pad measurements. In some isolated cases, the non-linearity was stronger and was then attributed to errors in the PSICAM measurements, when absorption coefficients were rather high and possibly outside the linear range of the PSICAM, i.e., higher than ${5}\;{{\rm m}^{- 1}}$ at the shortest wavelengths; results at these wavelengths were ignored. As observed earlier [13], the amplification factor was highly variable and varied between 2.5 and 5.8; again, one need to consider that inhomogeneity of the particle distribution on the filter might have induced part of this variability. Nevertheless, this factor corrects for both path length amplification and errors due to inhomogeneous particle distributions, and, hence, these errors are here less relevant for the final determination of $a_p^*$. Most important is the variability of the factor with wavelength or OD, which was minor in most cases (${\lt}{10}\%$). Fortunately, in cases where these variations over wavelength were significant, it was obvious from comparison of the four different spectra (UV/VIS, VIS/NIR, NIR, PSICAM) that artifacts were observed in just a single spectrum and could be compensated by taking the other spectra carefully into account. Procedural repetition of measurements would have helped in this aspect, but the long full scanning time (${\gt}{1}\;{\rm h}$ for the three filters) did not permit repetitions for all filter samples without a much larger effort, as in most cases it was not necessary. Figure 1 shows an example of the results before and after corrections for the path length amplification effects and small offset adjustments to the spectrum resulting from PSICAM measurements.

 

Fig. 1. One example of the construction of a final mass-specific absorption coefficient spectrum ($a_p^*$) by combining PSICAM and filter-pad measurements. Top: original results for PSICAM and different filter pad measurements (QFT) without applying correction for the path length amplification in the filters. Bottom: final spectral $a_p^*$ after path length correction applied as described in the methods. Note: both axes in logarithmic scale.

Download Full Size | PDF

A. Particle Size

As particle size has a significant effect on $a_p^*$, it is reported here (Table 1). In addition, when possible, the measurements are done using a narrow size fraction for the material (see method). The mean particle size (${D_{v50}}$) of the particle volume distribution varied from 1.7 µm for the ATD to 42.3 µm for the MAL sample (sieve fraction 25–30 µm). For the sieved material, the observed sizes and size distribution correspond well to the sizes expected from the different sieve pore sizes used, taking the non-regular shapes of the particles into account as well as general methodological constraints of the laser diffraction method used here for determining the size distribution. The mean particle sizes were 7.0 µm for MIC (sieve fraction 5–10 µm), 23.7 µm for STE, and 32.6 µm for QTZ (all sieve fractions 20–25 µm), and 41.2 and 42.3 µm for TUR and MAL, respectively (both sieve fractions 25–30 µm). Based on these results, the mean sizes for the two materials for which the size was not determined (due to a too low mass available) are estimated to 30 µm for OLI (sieve fraction 20–25 µm) and 40 µm for MAG (sieve fraction 25–30 µm). The material for DOR, FES, and SIC was purchased as fine grained material; the mean sizes were 9.1, 34.9, and 4.8 µm, respectively. All these materials had a relatively narrow size distribution as can be seen from the ${D_{v10}}$ and ${D_{v90}}$ values in Table 1.

The other materials (source clays, dust, and laterite earths) were not fractionated. Their sizes were rather small and ranged from 2.1 µm (SWy-2) to 14.7 µm (SHCa-1) for the source clays, and 6.8 µm (LAT sapro) to 11.5 µm (LAT rouge) for the laterite earths samples. Their size distributions were, as expected, generally larger (on a relative basis) than for the fractionated samples.

B. Mass-Specific Absorption

Four of the mineral types (MAG, TUR, SIC, ATD) examined here were not colored but have a rather dark appearance: MAG and TUR are black, SIC and ATD are gray but also have smaller particle sizes. As expected from this visual impression, these four mineral samples had generally very high $a_p^*$ values and rather spectrally flat absorption coefficient in the VIS, which for SIC and ATD clearly were higher toward the UV (Fig. 2). The following discussion will mainly focus on the NIR region and will only take UV and VIS features into account for the visual appearance. Together with MAL (which due to its green color is discussed later) they all showed the highest mass-specific absorption of all examined mineral types over the full NIR spectral range. In the VIS to NIR region SIC had the highest values over all that ranged from ${0.025}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 700 nm to ${0.040}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 2500 nm, slightly increasing with wavelength, while MAG had values of ${0.025}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 700 nm slightly decreasing to ${0.022}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 2500 nm (Fig. 2). Both mineral samples did not have any narrow spectral features in this spectral range. The differences in absolute values of these two samples need to take the differences in particle size into account (SIC: 4.8 µm versus MAG: 40 µm), hence we can expect that $a_p^*$ of MAG samples with smaller particle size might be significantly higher, while SIC of larger size might be significantly lower. The ATD sample (1.7 µm in size) had much lower values of only ${3.9} \times {{10}^{- 3}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 700 nm, decreasing to ${2.2} \times {{10}^{- 3}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 2400 nm.

 

Fig. 2. Top: mass-specific particulate absorption coefficient, $a_p^*$, as a function of light wavelength in the ultraviolet (UV) to near-infrared radiation (NIR) spectral range, 200–2500 nm, for particles (of discrete size) of different dark and colored pure minerals suspended in water (the mean particle size of each mineral type is given, see legend); logarithmic scale for the absorption axis. Middle and bottom: $a_p^*$ on linear scale for the spectral range of 900–2500 nm to highlight narrow spectral features. See Data File 1 for underlying values.

Download Full Size | PDF

There is only a faint, but spectrally very narrow, feature at 1912 nm indicating a very low contribution of crystal water. The TUR sample, despite being rather black, had lower values than MAG, of about ${0.012}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 700 nm, and showed a much stronger decrease with wavelength, reaching values as low as ${7} \times {{10}^{- 4}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$. TUR showed some broad spectral features with maxima at 730 and 1100 nm, a local minimum at 900 nm, and narrower features at 2204, 2246, 2302, and 2370 nm. The black color and the broad spectral features arise from the high amount of Fe in the crystal structure of this specifically black Fe-Tourmaline (Schörl). The finer structures at ${\gt}{2000}\;{\rm nm}$ indicate significant absorption by O-H groups that are known to be part of the crystal structure of tourmalines. The very broad absorption features of MAG (Magnetite: ${\rm Fe}_3{\rm O}_4$) arise from a specific ${{\rm Fe}^{2 +}}/{{\rm Fe}^{3 +}}$-complex [15], while SIC (SiC) absorption comes from the carbon in the chemical structure. ATD is a mixture of quartz and clay components, but also includes ${\rm Fe}_2{\rm O}_3$ that could produce the spectrally broad absorption by this material.

Two mineral samples examined had a clear green color: the dark green MAL and the greenish OLI. Hence, the absorption shows clear minima in the green spectral band (500–550 nm) and higher absorption in the UV/blue and red part of the light spectrum (Fig. 2). MAL was most obvious, with a distinct minimum at 530 nm. The generally dark color of MAL is indicated by the generally higher $a_p^*$ values, which are among the highest measured together with the black mineral samples. MAL consists mainly of malachite, which is a copper carbonate hydroxide (${\rm Cu}_2{\rm CO}_3{({\rm OH})_2}$) mineral, where the O-H groups lead to the absorption structure at ${\gt}{2200}\;{\rm nm}$ and the ${{\rm Cu}^{2 +}}$ is responsible for the green color by having absorption maxima in the blue and red/NIR (600–1000 nm) regions with a clear gab in the green. The broad absorption in the NIR (1100–1500 nm) must come from some other minerals in the samples. The OLI sample is a mixture of several minerals with olivine (${({{\rm Mg}^{2 +}},{{\rm Fe}^{2 +}})_2}{\rm SiO}_4$) as a major compound. It possesses a broad, specific absorption maximum (around 1050 nm, ${1.1} \times {{10}^{- 3}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$) due to the octahedral structure of a specific ${{\rm Fe}^{2 +}}$ complex in the olivine crystal (see [9]). All these UV-to-NIR features and the additional narrow absorption maxima at 1920 nm and 2320 nm of the OLI sample reproduce rather exactly absorption features visible in the reflectance spectrum for olivine shown in [9] (see Fig. 2 in [9]).

 

Fig. 3. Same as in Fig. 2 for the laterite earths and some source clays (SWy-2, SAz-1, STx-1b). Obvious structures around 860 nm have to be interpreted with care (see methods). See Data File 2 for underlying values.

Download Full Size | PDF

 

Fig. 4. Same as in Fig. 2 for DOR, FES, QTZ, MIC, and STE. Here, obvious structures around 860 nm have to be interpreted with care (see methods). See Data File 3 for underlying values.

Download Full Size | PDF

The two samples MIC and FES contain silicate minerals, mica and feldspar, respectively, and are white in appearance. Like all the rather white or red mineral samples, they possess their highest specific absorption in the UV (Figs. 3 and 4). The absorption of MIC and FES in the VIS–NIR is about 1 to 2 orders of magnitude lower than in the UV (Fig. 3). MIC shows $a_p^*$ of maximally ${3.2} \times {{10}^{- 3}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 400 nm and minimally ${1.9} \times {{10}^{- 4}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 1850 nm, with broad absorption maxima around 700 nm and 1100 nm and narrower maxima at 1414 nm, 1930 nm, 2206 nm, and 2350 nm. Micas form a wide group of minerals that can contain a variety of typical elements in their crystal structure. The two broad NIR maxima seem very similar to the feature observed with the TUR sample and would then indicate the occurrence of a similar Fe structure in MIC. The very fine structures at 1414 nm and 2206 nm indicate the occurrence of specific O-H bonds in the sample and occur at exactly the same wavelength and in the same composition as in the mica mineral muscovite (see Fig. 8 in [9]). The slightly broader features at 1930 nm and at ${\gt}{2200}\;{\rm nm}$ with their spectral shapes and positions come from free water in the crystal structure and would be accompanied by a structure around 1416 nm, that here is probably masked by the stronger O-H absorption feature at 1412 nm. The FES sample consists of a rather pure feldspar mineral that has neither O-H structures in the crystal matrix nor any Fe; it shows one of the lowest $a_p^*$ values in the NIR (${9.1} \times {{10}^{- 6}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$) and no broad absorption features. Some wider spectral features can be observed at 1416 nm, 1924 nm, 2212 nm, and 2316 nm (Fig. 3, Table 2), all of which indicate the presence of water in a specific crystal configuration.

While the general absorption maxima of water in crystals are determined by overtones and combinations of the major vibrations of the water molecules at a wavelength ${\gt}{2500}\;{\rm nm}$, the specific local shape and occurrence of these structures depend on the interplay of the water molecule with the crystal structure. For example, in the case of quartz the water is not interfering with the crystal structure and shows rather typical absorption features of the water molecule, whereas is many clays water interferes much more and the crystal structure includes O-H bonds, which leads to many more fine structures that can be used to identify specific mineral types. The two quartz mineral samples, QTZ and DOR, hence, show the typical structures of water in the crystals at 1412/1415 nm, 1928/1930 nm, and 2206/2208 nm (Fig. 3, Table 2). The DOR sample was rather pure and appeared white, hence the overall $a_p^*$ values in the NIR are low, ranging from ${2.3} \times {{10}^{- 5}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 1800 nm to ${8.4} \times {{10}^{- 5}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 700 nm. QTZ has a sand type color and the sample shows few additional small black crystals. The sample possesses generally higher $a_p^*$ values (minimum ${4.7} \times {{10}^{- 5}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$) than DOR, with some slightly more pronounced features in the VIS region, but very faint, but still visible, water absorption structures in the NIR. Note that of the last four samples described, MIC and DOR are of small particle size and QTZ and FES of larger, which would lead to higher values of $a_p^*$ for the former and lower for the latter material if the same particle size is considered.

Table 2. Occurrence and Position of Specific Absorption Maxima in the NIR Spectral Region That Indicate Vibrational Absorption Features Associated Mostly with Water and O-H Bonds^a

View Table | View all tables in this article

One special sample is STE, which was produced by grinding a piece of soapstone. The minerals in soapstone are Mg-silicates, like steatite (${\rm Mg}_3{\rm Si}_4{{\rm O}_{10}}{({\rm OH})_2}$). The gray/yellow stone turned into a slightly yellow, white powder during grinding. The UV/VIS absorption shows strong absorption in the UV that decreases with wavelength by nearly 3 orders of magnitude up to 620 nm, with a clear $a_p^*$ minimum of ${2.0} \times {{10}^{- 5}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ (Fig. 3).

In the NIR region STE has a broad maximum around 950 nm and a minimum of about ${3.0} \times {{10}^{- 6}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ around 1700 nm; here STE possesses the lowest $a_p^*$ values found in this study. In addition, there are two clear, very fine absorption features at 950 nm and (very pronounced) a double peak at 1392/1398 nm, and a set of additional structures between 2050 and 2500 nm (see Table 2). All these fine features are related to specific O-H bounds in the crystal structure and possibly other anionic structures with O-H bounds (for details see [9]).

The next group of samples is the laterites earth samples collected in New Caledonia. These are mineral samples from natural ultramafic soils but are of interest as the material is often washed into coastal water and there often is part of the minerogenic suspended matter of coastal waters around the island [10]. All samples examined here are red to red–yellowish colored and of medium particle size (6.8–11.5 µm, Table 1). They showed the strongest specific absorption in the UV/blue region and strong absorption throughout the VIS to NIR region (Fig. 4). Lowest $a_p^*$ values are observed for LAT (jaume) at 2150 nm with ${3.2} \times {{10}^{- 4}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$, while for LAT (sapro) $a_p^*$ was always larger than ${3.3} \times {{10}^{- 3}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$. All samples showed only small, more or less pronounced features of water absorption around 1420 nm and 1930 nm (see Table 2), and clear broad maxima at 470 nm, 670 nm, and 900 nm. These are most likely associated with the iron oxide compounds in the samples, which are already indicated by their red colors. The shape and occurrence of the water absorption features might indicate that the samples are composed of some quartz, which is a typical compound of laterite earths.

The last group of minerals includes all the source clays (Figs. 4 and 5). These are standard materials for which many physical, chemical, and mineralogical properties are known and that have been used in other studies about particulate absorption properties [3,5,10]. When focusing on the NIR region, all obvious features observed are associated purely with water and O-H bonds in the crystal structure and are observed at the expected wavelength [9,10]. For the overall $a_p^*$ values the kaolinites (KGa-1b, KGa-2) showed the lowest values of ${1.9 - 3.4} \times {{10}^{- 5}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ around 1600 nm, whereas for the strongest absorbing clay samples (SWy-2, PFI-1) the minimum values were 1 order of magnitude higher and about ${2.6} \times {{10}^{- 4}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$. There are clear similarities in the spectral features of the same mineral type (montmorillonites and kaolinites) but significant differences between these and the other two types (PFI-1, SHCa-1). All these features exactly reproduce the known absorption features in wavelength position, local spectral shape, and extent of different clay types that are visible in reflectance spectra for montmorillonite and kaolinite [9] (Figs. 4 and 5; Table 2). For the NIR, high-resolution spectral data of all source clays are available from attenuated total reflection Fourier transform infrared (ATR-FTIR) spectroscopic measurements [10]. This data can be used to validate the measurements performed here for the wavelength accuracy. For all source clay samples, the observed absorption spectra exactly reproduce the spectra shown in [10] in terms of wavelengths position and relative peak heights. The positions vary by maximally $\pm{2}\;{\rm nm}$, a reasonable accuracy for measurements at longer NIR wavelengths (${\gt}{1000}\;{\rm nm}$) by FTIR and spectrophotometry. Overall, the spectrophotometric measurements performed here allow highly accurate determination (for 200–2500 nm) of all spectral information of mineral particles of a given narrow size distribution (in absolute terms of the absorption coefficient and wavelengths), while FTIR approaches to determine absorption (or its coefficient) of scattering material is often less accurate and cannot easily be used at ${\lt}{1000}\;{\rm nm}$, while it can even be used at ${\gt}{2500}\;{\rm nm}$, where the decreasing reflectivity of the commonly used material of the integration sphere (Spectralon) sets an upper limit.

 

Fig. 5. Same as Fig. 2 for source clays PFI-1, SHCa-1b, KGa-1b, and KGA-2. Obvious structures around 860 nm have to be interpreted with care (see methods). See Data File 4 for underlying values.

Download Full Size | PDF

For specific source clays (KGa-1b, KGa-2, SAz-1) some information is available for the UV/VIS spectral region [3,5]: when considering measurement uncertainties and small differences in mean particles size due to methodological differences between these studies and the approach used here, the overall values of $a_p^*$ do agree. All studies show rather low specific absorption coefficient for these three source clays of maximal values in the range of ${0.06 - 0.11}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ in the UV found here and values of ${0.05 - 0.2}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ found in [3,5]. Otherwise, information from [3,5] at longer wavelengths (${\gt}{500}\;{\rm nm}$) seems to be rather uncertain due to methodological constraints and the values are often below the detection limit. Here, low but significant specific absorption coefficients are observed for all source clays with values ranging from ${4.6} \times {{10}^{- 5}}$ to ${6.3} \times {{10}^{- 4}}\;{{\rm m}^2}\,{{\rm g}^{- 1}}$ at 800 nm.

 

Fig. 6. Particulate absorption coefficient, $ a_p $, spectra in the NIR spectral region of one sample from the German Bight (North Sea) taken in May 2017 and one sample from the Eastern Lagoon of New Caledonia taken in March 2014.

Download Full Size | PDF

The mass-specific absorption spectra of minerals will allow not only to identify some specific minerals (clays, quartz) in natural samples but to obtain information about relative and perhaps absolute contribution of specific minerals to the overall absorption in a particle assemblage of minerogenic material if at the same time the particle size dependence is considered. In a first attempt to follow this approach, particulate absorption of natural samples from the German Bight and from the Eastern Lagoon of New Caledonia [16] are shown in Fig. 6. Both samples originated from waters with low concentrations of minerogenic suspended matter. The samples were washed with purified water to remove all sea salts, otherwise water of hydration associated with the salt crystals after sample filter drying would dominate the absorption in the NIR. Besides a generally significant absorption coefficient in the NIR, the German Bight sample showed small but clearly visible absorption peaks in the NIR that indicate the existence of a montmorillonite, as the main peak is at about 1916 nm (see Table 2) and the peak occurrences in general fit to either quartz or clays. However, this is rather speculative, as main groups of clays, such as illites and smectites, are not included in the data presented here and illites build typically the major group of clays in North Sea sediments [17]. The sample from New Caledonia was rather red in color and the NIR spectrum is very similar to that of the laterite earths from the main island. These examples show that expected spectral signatures from minerals in the NIR occur in natural samples. In the case of the German Bight sample, knowing the spectral distribution of the specific absorption by clays and quartz, it is obvious that the relatively strong general absorption outside the peaks in this sample must come from some other material with a rather structureless absorption in the NIR. If we assume that organic material will not possess significant absorption in the NIR, some kind of black mineral material or black carbon might contribute here significantly. More detailed knowledge about specific absorption by this kind of materials is critical, as well as about their contribution to absorption in natural samples. The methods and results presented here are intended to support this.

4. CONCLUSIONS

By adapting spectrophotometric methodology for the determination of the absorption coefficient of particles in suspension to work in the full NIR spectral region (780–2500 nm), mass-specific absorption coefficients of particles from several pure minerals and natural mineral samples were determined. This was done, when possible, for a specific particle size fraction with mean sizes between 1.7 µm and maximally 42 µm. For a set of source clays and other pure minerals, these measurements nicely reproduce spectral features and signatures that are known from earlier, independent reflectance and absorption measurements of bulk minerals but will allow quantification of the mineral mass in suspension if the particle sizes distribution and the dependence of the coefficient on particle size for each mineral is known. This information and the method used when applied to samples from natural particle assemblages of aquatic/marine environments will allow identification of single absorbing components in these assemblages and quantification of their contribution to particulate absorption in these waters. More information about the particle size dependence on the mass-specific absorption coefficient is needed to obtain as much information about each single optical compound in water to accurately model in-water optics.