3.1 Sample preparation

Basic raw materials are single feldspar crystals (adularia, amazonite), one ground sample for use in pottery (labradorite AU) and rocks (all the others). All raw samples are ground for 5 min in a tungsten carbide disc mill (Retsch, RS1, 1400 rpm). The ground sample is aerosolized in a fluidized bed aerosol generator (TSI Model 3400A). To investigate particle size effects special attention is given to select monodisperse particles. The aerosol is passed through two cyclones (d₅₀=3 and 1 µm) and one impactor (d₅₀=0.4 µm) before size selection by a differential mobility analyser (DMA; TSI Model 3081). As the size selection with a DMA is not directly based on the particle size but on electrostatic mobility, multiply charged particles of larger sizes pass the DMA and are selected as well. Depending on the initial particle size distribution, the fraction of multiply charged particles can contribute a substantial fraction to a certain mobility diameter. In particular, small particle sizes originating from the upslope of the initial size distribution contain non-negligible amounts of multiply charged, larger particles. To reduce distortion to the size selection, a centrifugal particle mass analyser (CPMA; Cambustion), able to isolate a specific particle mass, is used in sequence with the DMA for particle sizes of 100 and 200 nm. Spherical particle shape and bulk density (2.6 g cm⁻³) was assumed to select singly charged particles and exclude larger particles by mass. The use of the CPMA for the selection of larger particles (400, 800 nm) was not necessary as the fraction of larger particles is sufficiently reduced by the cyclones and the impactor upstream of the DMA (see Fig. 2), which confines the size distribution of the particles entering the DMA to below 1 µm (aerodynamic diameter). For the 400 nm size selection, doubly and triply charged particles (700 and 994 nm) amounted to 16 % and 4 %, respectively, whereas for the 800 nm particles, the doubly charged particles (1481 nm) made up 4 %. The CPMA was not used for the selection of 50 nm particles due to the very low number concentration of this particle size in the output size distribution of the fluidized bed generator. A substantial fraction of larger, multiply charged particles (73 nm (∼ 30 %) and 91 nm (∼ 7 %) for 2 and 3 charges, respectively) among the 50 nm (∼ 63 %) particles is expected. The contributions of singly and multiply charged particles to the frozen fractions is approximately proportional to their abundance and does not impact any of the conclusions made in this work. As all feldspar samples have similar hardness and were ground in the same way, the fraction of multiply charged particles at a selected size is expected to be comparable among samples. The contribution of multiply charged particles to the 50 nm particle frozen fraction is discussed in Appendix C.

3.2 Immersion freezing experiment

A flow of 1 L min⁻¹ of sample air containing the dry feldspar particles layered between twice 2.5 L min⁻¹ sheath air is introduced into the immersion mode cooling chamber (IMCA) developed by Lüönd et al. (2010). The IMCA walls are 0.6 cm apart and layered with continuously wetted filter paper. A horizontal and vertical temperature gradient is maintained to generate supersaturated conditions to form and subsequently precondition water droplets for the experiment in the Zurich Ice Nucleation Chamber (ZINC; Stetzer et al., 2008). In the top part of IMCA particles are exposed to relative humidity with respect to water of ∼ 120 % at 313 K. The high relative humidity ensures all particles transitioning through IMCA activate into cloud droplets, independent of their properties. In the lower part of IMCA, droplets containing the immersed feldspar particles are cooled to the experimental temperatures prevailing in ZINC. ZINC is a parallel-plate continuous-flow diffusion chamber (1 cm gap between the wall plates) layered with a thin ice layer as source of water vapour at temperatures below 273 K. The temperature of the parallel wall plates of the chamber is controlled independently by two cryostats (Lauda134 RP890). At the transition from IMCA into ZINC, additional sheath air (2 L min⁻¹ on either side) is added in order to maintain a width of 1 mm for the sample lamina and to prevent turbulence. Constraining the sample width to 1 mm allows us to control and constrain the temperature and relative humidity that the particles are exposed to. ZINC is operated such that droplets transitioning from IMCA can be exposed to temperatures from 238 to 258 K at water saturation conditions to prevent evaporation of the droplets. After a residence time of ∼ 10 s, water droplets and nucleated ice crystals are detected in-line (see Fig. 2) by the ice optical depolarization detector (IODE) described in Nicolet et al. (2010) and Lüönd et al. (2010). The water droplet and ice crystal counts at a specific temperature are used to derive the fraction of droplets that freeze due to ice nucleation on the immersed feldspar particle or by homogeneous freezing. The reported frozen fraction is defined as

4.1 Temperature dependence

The shape of the temperature-dependent frozen fraction provides information on the heterogeneity of ice nucleation properties among particles. Comparing frozen fraction shown in Fig. 3 reveals different slopes of the temperature dependence for different samples as well as particle sizes. As an example, the increase in frozen fraction from 0.1 to 1 occurs over a 7 K range for 800 nm microcline particles; however, this range broadens with decreasing particle size to 10 K for 100 nm particles. To illustrate this further, in Fig. 4 we show the range of freezing temperatures for the frozen fraction 0.1–1 for all 800 nm particle samples. We find that the freezing temperature range for orthoclase spreads over 10 K already for the 800 nm particles. The broader range of activation temperatures suggests that the ice-nucleation-active sites in the 800 nm orthoclase are more heterogeneous than the ones in the 800 nm microcline sample (see Fig. 4). Additionally, we observe that the microcline and orthoclase samples initiate freezing at higher temperatures compared to the Na∕Ca-feldspars suggesting that K-feldspars are in general more ice active in accordance with, e.g., Atkinson et al. (2013). The microcline amazonite sample is the most effective sample for immersion freezing. Upon close observation of the freezing range depicted in Fig. 4, however, it can be seen that the narrow freezing range is neither unique to microcline nor the K-feldspar polymorphs as a freezing range of 5 K also exhibited by the labradorite CH2 sample. Even though labradorite CH2 is much less ice active than orthoclase, the narrow spread in freezing temperatures suggests that the active sites of this sample are less heterogeneous than those of orthoclase.

Figure 3Frozen fractions of size-selected feldspar polymorphs. The frozen fraction of water droplets due to homogeneous freezing in the set-up is given by the grey shaded area. Smaller particles (50 and 100 nm) were only tested for the more ice active samples. Horizontal error bars represent the variation in the sample temperature across the 1 mm aerosol layer in ZINC, and vertical error bars represent uncertainty in the measured frozen fraction due to overlap in depolarization ratio of droplets and ice crystals in the IODE signal.

Download

Figure 4Spread in freezing temperatures corresponding to frozen fraction of 0.1 to 1 and T₅₀ for 800 nm feldspar particles. Hatched region indicates homogeneous freezing as the dominant mechanism.

Download

Freezing curves for the Na∕Ca-feldspars extend to temperatures where homogeneous freezing sets in, suggesting that a fraction of these particles, up to 25 % (considering the 800 nm particles in Fig. 3), cannot nucleate ice heterogeneously. In contrast, freezing on the K-feldspars (except sanidine) occurs above the homogeneous freezing regime. Potential reasons for why sanidine is an exception to the rest of the K-feldspars are discussed in Sect. 5 below. For all K-feldspar samples, the effect of temperature on the frozen fraction has a distinct size dependence where larger particle sizes exhibit a stronger temperature dependence; i.e. the change in frozen fraction with temperature is larger for large particles and smaller for small particles (see Fig. 3). This can be explained by the increased probability of large particles to host at least one exceptionally efficient active site compared to smaller particles with fewer actives sites. Most plagioclase feldspar particles are ineffective heterogeneous INPs or initiate freezing only when temperatures approach homogeneous nucleation temperatures, also observed by Harrison et al. (2016).

A comprehensive analysis of how to obtain good immersion freezing parameterizations, including the fitting of the microcline (amazonite) data set shown in this study, can be found in Ickes et al. (2017). Parameterizations for other samples can be found in Niedermeier et al. (2015) and Peckhaus et al. (2016). Compared to the temperature-dependent increase in frozen fraction predicted by a single contact angle representation within classical nucleation theory, the observed temperature dependence is less steep. This feature can be parameterized by, e.g., a contact angle distribution (Ickes et al., 2017). In terms of nucleation sites per surface area (ice nucleation active site (INAS) densities), the number of sites increases by 1 order of magnitude with every additional 2 K of supercooling for most samples (see Fig. C1). Sanidine and pericline show a weaker increase in frozen fraction with decreasing temperature.

4.2 Size dependence

The minimum particle size able to trigger heterogeneous ice nucleation above temperatures where droplets freeze homogeneously is investigated within the limits of the experiment (see Sect. 3.1.). The range of homogeneous nucleation temperatures for water droplet sizes of 10–20 µm generated in IMCA was found to be 235–237 K (see Fig. 3). The lower particle threshold size for heterogeneous ice nucleation determines how effectively small inclusions, internally or externally mixed with mineral dust aerosol, will impact ice formation.

Experiments on single immersed, size-selected particles show a strong particle size dependence of ice nucleation ability. The specific slope of the median freezing temperature (T₅₀) as a function of geometric surface area varies between samples. Examples are the almost linear scaling of frozen fractions with particle size (surface area) observed for 100–200 nm microcline, 200–400 nm orthoclase and 400–800 nm adularia (see Fig. 5). The two orthoclase (adularia) varieties, which show a low ice nucleation activity for particle sizes smaller than 200 (400) nm are observed to have a stronger increase in activity with surface area compared to the highly active samples (e.g. amazonite). This behaviour could suggest that for larger surface areas than accessible with these experiments, differences among K-feldspar species may become smaller as the T₅₀ converge for the K-feldspars (Fig. 5). The minimum size triggering immersion freezing is found to be 50 nm for microcline particles. In particular, the 50 mm amazonite particles exhibit a T₅₀ of 243 K, which is higher than previously reported values of a variety of dust and clay samples (Lüönd et al., 2010; Welti et al., 2012; Kanji et al., 2013; Augustin-Bauditz et al., 2014; Hartmann et al., 2016). For most of the feldspar samples, including orthoclase and sanidine, particles of 400 or 800 nm are required to observe effective immersion freezing, which contradicts the hypothesis of Atkinson et al. (2013) that generally tiny K-feldspar inclusions in dust particles will contribute to ice nucleation. An exception is the microcline (amazonite) sample showing that even particles as small as 50 nm are able to act as INPs, suggesting that the hypothesis could be true for this K-feldspar polymorph. We note that despite the multiply charged particles expected in the 50 nm particle population (see Sect. 3.1), particles of this size substantially contribute to the frozen fractions observed, as demonstrated in Fig. B1.

Figure 5Median freezing temperature (T₅₀) where 50 % of particles are observed to initiate freezing as a function of geometric particle surface area. Parameterizations for desert dust from Niemand et al. (2012) and K-feldspar from Atkinson et al. (2013) are included for comparison. Dashed lines indicate extrapolation of the parameterization to lower temperatures. Only T₅₀ above homogeneous freezing temperatures is shown.

Download

5.1 Crystal structure and surface texture

None of the K-feldspar samples exhibits a close lattice match (see Table 2) to hexagonal ice Ih (a=4.52 Å, c=7.35 Å) or cubic ice Ic (a=6.35 Å), typically found at lower temperatures. There are two crystal structures in the feldspar samples tested: monoclinic and triclinic. While Na∕Ca plagioclases are all triclinic, the K-feldspar sanidine and orthoclase have a monoclinic crystal lattice. Microcline is a triclinic K-feldspar. Triclinic microcline forms by the transformation of monoclinic K-feldspar upon cooling (see, e.g., Waldron et al., 1993) or, as suggested by Collins and Collins (1998), by K replacement of Na∕Ca-feldspar. The lattice angles of monoclinic and triclinic K-feldspar polymorphs are summarized in Table 2. It can be seen that structural differences among feldspars are small even when comparing triclinic to monoclinic lattice angles. Mason and Maybank (1958) reported experiments on triclinic microcline and monoclinic orthoclase. While microcline was found to be active at −9.5 ^∘C, their orthoclase sample was inactive above −18 ^∘C. The data shown in Fig. 3 support a difference in ice nucleation temperature of these K-feldspars, with microcline forming ice at higher temperatures than the monoclinic orthoclase, although the absolute temperatures differ from Mason and Maybank (1958) because of the single-particle nature of our study. In addition, the least efficient K-feldspar sample sanidine is monoclinic. The hypothesis that within the K-feldspar polymorphs, a triclinic crystal structure is a more active ice nucleation property was put forward by Augustin-Bauditz et al. (2014), based on their indirect observation that mixed dust samples containing orthoclase were less ice active than a microcline sample. However, given that orthoclase is much more ice active than sanidine (Fig. 3) and both have a monoclinic structure, their difference in ice activity cannot be explained by the crystal lattice alone. Another difference between monoclinic and triclinic K-feldspars is the order or disorder of Al³⁺ and Si⁴⁺ in the tetrahedral sites building the crystal structure. Monoclinic sanidine retains a disordered crystal structure (although often forms defect-free crystals), while microcline transforms to an ordered crystal. Orthoclase has intermediate disorder in this regard. The polymorphs of K-feldspar are distinguished by the randomness of this order distribution, which depends on the temperature and cooling rate during rock formation. The sequence of feldspar melting temperatures given in Lutgens et al. (2016) is Ca-feldspar > Na∕Ca-feldspar > Na-feldspar > K-feldspar (sanidine > orthoclase > microcline). While Ca-feldspar is completely ordered, Na-feldspar can have varying degrees of order and Na∕Ca-feldspars are disordered. The melting points of K-feldspar correspond to the ascending order of the observed T₅₀ and the degree of order–disorder, indicating that ordered feldspar crystals are more active INPs than disordered crystals. However, since the crystal structures do not match that of ice, a direct correlation is not plausible.

Table 2Lattice parameters for K-feldspar polymorphs from Hovis (1986) and Ostrooumov and Banerjee (2005).

Download Print Version | Download XLSX

Figure 6Representative scanning electron microscope images of (a) orthoclase, (b) adularia, (c) sanidine, (d) microcline, (e) amazonite, (f) labradorite AU, (g) pericline, (h) labradorite CH2 and (i) labradorite CH1. For reference, the 400 nm scales are indicated in the figures. The lamellae (perthitic) structure can be seen for the K-feldspars (a–e) and to a lesser extent for the plagioclase feldspars (f–i) but is almost absent for pericline (g) and labradorite CH2 (h). Images taken using a LEO 1530 Gemini SEM. Dust samples were applied to adhesive, carbon conductive tapes and sputter-coated by a thin platinum layer. High-resolution secondary electron (in-lens detector) images were acquired at 2 kV.

Download

The surface texture of the nine samples was investigated by scanning electron microscopy (SEM). Representative images are shown in Fig. 6. Steps, lamellae and clustering are prominent features. Smaller particle sizes appear smoother and rounder. Due to the triclinic structure of Na∕Ca-feldspar, twinning (i.e. symmetrical intergrowth, phase exosolution or Na–K exchange between feldspar phases) with triclinic microcline is common (David et al., 1995; Parsons et al., 2015; Whale et al., 2017). However, given that the α and γ crystallographic angles are very close to 90^∘ (see Table 2), it is difficult to detect twinned structures in SEM images (Lee et al., 1998, and references therein). The surface sites where transitioning between microcline and plagioclase occurs has been proposed to be a topological feature critical for ice nucleation (Whale et al., 2017). All K-feldspars exhibit perthitic structures (lamellae), whereas on the plagioclase feldspars this feature is less prominent or missing (Whale et al., 2017). In Fig. 6, the layering from the lamellae is clearly visible for panels (a)–(e) (K-feldspars), whereas for panels (f)–(i) the layering is less visible and even absent for panels (g) and (h) (pericline and labradorite CH2). It has been reported that K-feldspar contains macropores of 0.1–1 µm diameter (David et al., 1995; Lee et al., 1995). Hodson (1998) found that the contribution of internal surface structures such as pores to the total particle surface increases with decreasing particle size for some K-feldspars (e.g. sanidine), whereas no particle size relation to pore surface was found for other samples such as microcline. For the current study the internal surface area is not of interest since in comparison to the outer surface in contact with the water droplet, ice formation is depressed in narrow confinements (Marcolli, 2014). Therefore, using the particle surface area available to nitrogen adsorption (BET surface; Brunauer et al., 1938) as a measure of surface area might overestimate the available surface for immersion freezing nucleation. Geometric surface area assuming a spherical shape is used here instead (Fig. 5). In none of the SEM images, intact pores are visible. This is possibly because of the Pt coating prior to imaging. Also, the grinding process applied to the samples may have eliminated this feature, as cavities and macropores of this size present favoured sites of fracture. Lamellae could stem from split pores. Before milling, the K-feldspar samples used here were translucent to opaque, indicating that they reacted pervasively with a fluid causing exsolution, and thus the perthitic structure is expected (Parsons et al., 2015). All samples were subject to the same preparation (grinding) procedure. The assumption that physical properties (e.g. hardness) are comparable among the tested feldspar species implies that the same degree of artificial surface features are introduced to all samples. The vast difference in ice nucleation efficiency between the feldspars suggests that any grinding artefacts (such as fractures or exposing high-energy surfaces) did not create ice-nucleating properties, as otherwise all samples would have more comparable ice nucleation activities. Zolles et al. (2015) tested the effect of milling on a sample of microcline, albite and andesine and found none or only a minor increase of 1 K on the median freezing temperature. We note that this is not the case for all minerals as a large influence of milling has been noted for haematite (Hiranuma et al., 2014).

5.2 Chemical composition

The individual samples have been analysed by XRF spectroscopy for their bulk chemical composition and trace elements (see Table 3). Using actively ground samples from a homogeneous base material should preclude large variations in chemical composition with particle size for the samples investigated here (see INAS densities in Appendix A). A statistical classification of the XRF results using random forest analysis (Liaw and Wiener, 2002) to identify the important compositional predictors of T₅₀ revealed Pb, Rb and Sr to be most relevant. Pb and Rb are highly positively correlated (r=0.91), whereas the concentration of Sr in the samples shows a negative correlation to Pb ($r=-\mathrm{0.66}$) and Rb ($r=-\mathrm{0.51}$). The best linear correlation between a single compound and the ice nucleation activity characterized by T₅₀ was found for the Pb content in the feldspar samples (Fig. 7). However, there is an exception with using Pb as the single compound predictor since the microcline sample has a lower lead content than the orthoclase but a higher T₅₀. Cziczo et al. (2009) have shown that the presence of Pb could increase the ice nucleation activity of dust (clay) particles. However, high Pb content alone is insufficient to explain the ice nucleation activity of a mineral dust sample in general. As an example, the ice nucleation activity of a weathering product of feldspars (kaolinite) does not appear to depend on its Pb content. XRF analysis of the kaolinite samples used in the experiments reported in Welti et al. (2012), which had a lower ice nucleation activity than microcline, revealed a ∼ 10-fold higher Pb content (2888.4 ppm) in comparison to, e.g., the amazonite sample (247.0 ppm). The high Pb content measured in kaolinite could be a result of its high adsorption capacity for Pb (Jiang et al., 2010) and can be introduced at any point after the weathering from feldspar to kaolinite. Based on these arguments and that the correlation of Pb content to T₅₀ is not monotonic (e.g. lower Pb content in microcline than orthoclase; Fig. 7) we focus on other predictors suggested by the random forest analysis.

Figure 7Particle-size-dependent median freezing temperature (T₅₀) as a function of lead (Pb) content. Particle size is indicated by the coloured legend.

Download

Table 3Chemical analysis from XRF of the different feldspar samples used in this study. N/D stands for not detectable but could be above zero. A zero implies reliable absence of a compound. ^* LOI stands for “loss on ignition” and indicate the amount of volatile compounds (e.g. H₂O, CO₂, F, Cl, S) in a sample that are removed by heating during sample preparation.

Download Print Version | Download XLSX

The strongest chemical predictor for T₅₀ was found to be a high Rb∕Sr ratio. Contrary to the Pb content, a higher Rb∕Sr monotonically corresponds to a higher T₅₀ for all samples and particle sizes, shown in Fig. 8. Initially each feldspar has a specific Rb∕Sr partitioning influenced by its K content, the melt temperature from which it crystallized and its melt composition. If it is subsequently changed, the Rb∕Sr ratio can be an indication of mineralization processes (Plimer and Elliott, 1979). Due to their ionic radii, electronegativity and ionization potentials, Rb⁺ replaces K⁺ while Sr²⁺ is incorporated in the feldspar crystal instead of Ca²⁺. The Rb∕Sr ratio is indicative of the position of feldspar on the ternary phase diagram (Fig. 1) and the absolute concentrations of trace elements depends on the cooling or crystallization process and abundance of trace elements in particular magmatic fluid (Parsons et al., 2009). The replacement of K+ by Rb+ results in a lower microcline polymorph and a higher Rb∕Sr ratio, correlating to higher ice nucleation activity. This suggests that the observed correlation reflects a possible relationship between ice nucleation activity and feldspar microstructure. We note that a monotonic correlation between T₅₀ and a Pb∕Nd ratio can also be constructed, but no interpretation of the implication of such a ratio could be found in the literature and neither was Nd suggested as a predictor by the random forest analysis. Zolles et al. (2015) proposed an explanation of K-feldspar ice nucleation ability based on its surface chemistry. They argue that depending on the chaotrope (order-breaking) or kosmotrope (order-making) characteristic of feldspar surface cations which are present in the water surface surrounding the particle, they can enhance or inhibit the structuring of water and therefore the formation of the ice phase. The ions with the higher charge densities (Ca²⁺, Na⁺, Sr²⁺) tend to disturb the structuring of water (increase entropy) while those with a lower charge density (K⁺, Rb⁺) have a promoting effect (decrease entropy). Following this argument, an increased Rb⁺ concentration combined with a decreased Sr²⁺ content results in a more kosmotropic and less chaotropic surface property for the ordering of water into ice. The increased ice nucleation activity with an increasing Rb∕Sr ratio therefore corresponds to a higher ordering and thus supports the hypothesis of Zolles et al. (2015). However, it would be necessary to quantify the concentration of cations required to influence water ordering to determine how influential Rb⁺ and Sr²⁺ are for ice nucleation given they are only abundant as trace elements.

Figure 8Particle-size-dependent median freezing temperature (T₅₀) as a function of the Rb (rubidium) ∕ Sr (strontium) ratio. Particle size is indicated by the coloured legend.

Download

The observed discontinuities in freezing temperature with increasing particle size for adularia and orthoclase (see Fig. 3) indicate that chemical composition alone cannot explain the ice nucleation potential of all feldspar species. That the strong size dependence is observed for both orthoclase varieties (intermediate ordered crystal structures) but not for disordered sanidine indicates that a large enough, sufficiently ordered surface area is needed to initiate efficient ice nucleation. This interpretation corroborates the suggestion of Whale et al. (2017) that some perthitic structure (microtexture) associated with the strain between K- and Na-rich regions on the samples benefits ice nucleation, and these can range from the nanometre to the millimetre scale, varying amongst feldspar samples. The probability of such features inevitably scales with surface area and could explain the requirement of a critical particle size (surface area) in order to exhibit ice nucleation.

Natural weathering producing small feldspar particles commonly includes contact to acidic environments, causing the removal of alkalis and aluminium and the formation of a silica-rich skin. Freely exposed feldspars are reported to be populated by lichens, which can alter naturally occurring feldspar in an infinite number of ways (Smith, 1998). The surface properties (specific area, porosity, molecules on the surface) of freshly ground feldspar are therefore expected to be different from aged particles (Mangan et al., 2017). However, bulk composition and crystal structure can be assumed to be representative of natural feldspar. Additionally, a majority of airborne feldspar particles could be coming from phenocrysts (inclusions of feldspar in, e.g., a granite matrix; Lee et al., 1995), where the trace element content could be different. The ageing of microcline in concentrated solutions does not impair ice nucleation activity permanently in the case of near-neutral solutions; however, exposure to highly acidic or alkaline solutions damages the surface, inhibiting or even destroying the ice nucleation ability (Kumar et al., 2018). We note that the samples used in this study could have different ice nucleation properties from feldspar found in natural airborne dust because of the natural weather elements such as cold, low-pH conditions, resulting in a decrease in ice nucleation activity. On the other hand, high relative humidity with high-${\mathrm{NH}}_{\mathrm{3}}/{\mathrm{NH}}_{\mathrm{4}}^{+}$ conditions could enhance ice nucleation (Kumar et al., 2018). Feldspars of the same classification but from a different location could differ in immersion mode freezing properties given that feldspars can exhibit broadly similar chemistry and structure but at the atomic level will almost certainly differ in their intracrystal defects as well as in their trace element chemistry (Lee et al., 1998; Parsons et al., 2015).