Six different feldspar minerals, sourced from two industrial quarries, were investigated for their IN activity. Two K-rich feldspars were sourced directly from the French industrial minerals company Imerys, collected from different extraction fronts. These samples are referred to as ImerysK1 (IK1) and ImerysK2 (IK2). Although the company did not disclose specific mine locations, the samples are known to originate from active industrial operations. Four additional samples were provided by the LLANSÀ, S.A. company, a feldspar quarry located in Llançà, Catalonia, Spain. These included two samples with predominantly Na-rich feldspars and two with a roughly balanced K / Na composition, depending on the extraction front. Photographs of the extraction fronts are shown in Fig. S1 of the Supplement. The samples were labeled LlançaNa1 (LNa1), LlançaNa2 (LNa2), LlançaNaK1 (LNaK1), and LlançaNaK2 (LNaK2); see Fig. 1b and Table 1 for the full list and classification of the feldspar minerals used. The bulk rocks were processed in two steps. First, portions of each were crushed and homogenized via ball milling to produce fine powders, primarily for ice nucleation experiments and also for granulometric, chemical, and mineralogical analysis. Second, unprocessed fragments were cut and polished to prepare thin sections and mounts for petrographic and microstructural characterization. This dual approach enabled correlation of ice nucleation behavior with crystallographic structure and surface features of the parent minerals. Ultrapure water (resistivity 18 MΩ cm, pH 6.2), dispensed from a Milli-Q^® Integral 3 water purification system (Merck Chemicals GmbH, Darmstadt, Germany), autoclaved at 121 °C for 15 min and filtered through a 0.1 µm bottle top filtration unit (VWR International GmbH, Darmstadt, Germany), was used in all immersion freezing experiments.

The bulk chemical composition of the feldspar samples was determined using a Philips PW 2400 X-ray fluorescence (XRF) sequential spectrometer located at the Centres Científics i Tecnològics de la Universitat de Barcelona (CCiT-UB). Major oxides were quantified as weight percentages using the fused bead method, while trace elements (in ppm) were measured using pressed powder pellets (see Table 1 and Tables S1 and S2 in the Supplement). Quantification was calibrated against certified geological standards to ensure analytical accuracy. In addition, thermal treatments were conducted to assess Loss on Ignition [ASTM D7348-21].

Mineralogical characterization was conducted through a combination of X-ray diffraction (XRD), Optical Microscope (OM), and Scanning Electron Microscope (SEM). XRD measurements were performed on a PANalytical X'Pert System diffractometer, configured in Bragg-Brentano geometry with Cu Kα radiation (λ=1.54061 Å). Phase identification and semi-quantitative analysis were conducted with the XPert Graphics Identify software (Philips). For microstructural and petrographic examination, double-polished thin sections were observed using a Nikon ACT-1 optical microscope. The same sections were examined using a Hitachi TM-1000 tabletop SEM, coupled with energy-dispersive X-ray spectroscopy (EDS) to analyze microstructural features and verify elemental composition.

Particle-size distribution (PSD) of the milled feldspar powders were measured by laser diffraction using a Mastersizer 2000 analyzer and its Hydro 2000SM accessory. Powders were dispersed in deionized water, and the suspensions were sonicated for at least 5 min prior to measurement to avoid particle aggregation. The specific surface area (Sa) was determined by N₂ adsorption/desorption at 77 K using an ASAP 2020 (Micromeritics Inc.). Prior to measurements, feldspar powders were degassed at 393 K for 24 h. Sa was calculated by applying the Brunauer–Emmett–Teller (BET) method (Brunauer et al., 1938).

Ice-nucleating behavior was assessed using the immersion-freezing method, which evaluates the temperature at which aqueous droplets containing suspended mineral particles freeze. Feldspar suspensions were serially diluted in ultrapure water using an automated liquid handling station (epMotion ep5073, Eppendorf, Hamburg, Germany) to obtain concentrations of 10 wt %, 5 wt %, 2 wt %, 1 wt %, 0.5 wt %, 0.1 wt %, 0.05 wt %, 0.01 wt % and 0.001 wt %. However, not all concentrations were tested for each feldspar. The 10 wt % and 5 wt % concentrations yielded very similar results and were therefore not both tested for all samples. Additionally, very low concentrations were, in some cases, indistinguishable from pure water experiments and were thus excluded. The tested concentrations are shown in Fig. S4 of the Supplement. For each concentration, 96 droplets of 3 µL were dispensed into two 384-well plates, which were subsequently subjected to IN experiments using the high-throughput Twin-plate Ice Nucleation Assay (TINA), which has been described in detail elsewhere (Kunert et al., 2018). All suspensions were freshly prepared within 2 h prior to the experiments to minimize aging effects, as prolonged water exposure has been shown to modify feldspar IN efficiency (Harrison et al., 2016).

Sample suspensions were cooled from 0 to -30 °C at a constant rate of 1 °C min⁻¹. Droplet freezing was monitored using two infrared cameras (Seek Thermal Compact XR, Seek Thermal Inc., USA), providing a temperature resolution of ±0.2 °C. From the resulting data, the fraction of frozen droplets, fice(T), was calculated as a function of temperature. Cumulative freezing spectra were derived from fice(T) using the approach proposed by Vali (1971, 2014, 2019). Each experiment was performed in triplicate on independent sample batches. Control measurements using pure water showed a background freezing point with an average temperature of -24.3 °C. At concentrations below 0.001 wt %, freezing behavior converged with that of pure water and did not provide additional information on feldspar-induced ice nucleation. Consequently, these data were excluded from the quantitative analysis, which was restricted to temperatures above -20 °C.

The mineralogical composition of the feldspar samples was analyzed using powder XRD, enabling a semi-quantitative assessment of the mineral phases present. The main minerals determined are Na-feldspar, K-feldspar, and quartz (Qz), and a secondary phase is muscovite (KAl(Si₃Al)O₁₀(OH,F)₂), a mica-type mineral. XRD data indicate that all feldspar samples from the LLANSÀ S.A. mine (LNa1, LNa2, LNaK1, and LNaK2) are richer in Na-feldspar (Ab) than in K-feldspar (Mc or Or). The LNaK samples exhibit a more balanced distribution between the two phases. Quartz and minor muscovite (Ms) are also detected in these samples. In contrast, the Imerys samples (IK1 and IK2) are rich in K-feldspar, with only minor Na-feldspar presence. Neither Qz nor Ms is identified in these latter samples. Semi-quantitative abundances of these mineral phases, expressed as weight percentages, are shown in Fig. 1b.

XRF analysis (Table 1) reveals that IK1 and IK2 contain more than 10 wt % K₂O and exhibit K₂O / Na₂O ratios between 3 and 5, consistent with a predominance of K-rich feldspar with respect to plagioclase. In contrast, LNa1 and LNa2 are Na-rich, with Na₂O contents exceeding 5 wt % and K₂O / Na₂O ratios well below 1. LNaK1 and LNaK2 show intermediate alkali compositions, with K₂O / Na₂O ratios close to unity. Minor oxides (< 0.5 wt %) are reported in Table S1 in the Supplement, together with full oxide compositions as in Table 1. Loss on ignition (LOI) values are also reported to account for adsorbed water and volatiles, which are found to be negligible. Notably, the Si / Al ratio in the Llançà samples is ∼ 4.5, while in the Imerys samples it is closer to 3.0, suggesting the presence of quartz in the former. Although classification based solely on K₂O / Na₂O ratios may be insufficient, the ice-nucleating behavior of the samples correlates closely with this ratio. Furthermore, samples collected from the same extraction front, for example, LNa1 and LNaK1 (see Fig. S1 in the Supplement), exhibit markedly different K / Na ratios and correspondingly distinct ice-nucleating behaviors. Therefore, we maintain this classification to facilitate a comprehensive analysis of the results.

The K / Na distribution in selected samples was further investigated through optical microscopy and SEM analysis of metallographically polished feldspar surfaces. SEM imaging was performed on three samples exhibiting a wide range of K / Na ratios – IK1, LNaK2, and LNa1 – as shown in Fig. 2a, c, and e, respectively. The mineralogy identified via optical microscopy and SEM corroborates previous results obtained from XRD. All examined samples exhibit characteristic exsolution textures between the two alkali feldspar phases. SEM observations of Imerys' samples: IK1 (Fig. 2a) and IK2 reveal a typical perthitic texture, where K-feldspar – identified as Mc (brighter regions due to its higher average atomic number, Z≈18) – hosts elongated intergrowths of Ab-type sodic plagioclase (darker regions, Z≈11). Feldspars from Llançà display two distinct characteristics, depending on the exploitation level where they were collected. The LNaK samples contain both Ab and Or (see Fig. S2 in the Supplement), whilst LNa feldspars, which are more albitic, contain mainly Ab with smaller amounts of Or in the form of anti-perthites. The XRD (Fig. 2) peaks associated with Or exhibit lower intensity and considerable broadening, particularly in the LNa powders. These observations suggest that the dominant K-feldspar phase in the Llançà samples is more disordered, corresponding to Or rather than the well-ordered Mc. In the LNaK2, small apatite crystals [Ca₅(PO₄)₃(OH, Cl, F)] are also observed as solid inclusions (see inlet Fig. 2c). In contrast, the LNa1 sample (Fig. 2e) shows an anti-perthitic texture, where Ab forms the dominant matrix with fine exsolutions of Or. Apatite inclusions are again present. All samples exhibit dark-contrast features associated with fluid inclusions, as observed in SEM images and corroborated by optical microscopy, which enables the visualization of internal features. Moreover, optical observations show aligned porosity across all samples (see Fig. S2). The coexistence of aligned porosity and fluid inclusions suggests a shared history of hydrothermal alteration or mineral-fluid interactions. These features are predominantly developed within the K-feldspar phases.

Figure 2b, d and f show the XRD spectra for each of the studied samples in the 2θ range of 28–32°, where characteristic peaks associated with Mc, Or, and Ab structures can be clearly identified (the full-range spectra are also provided in Fig. S3 of the Supplement). Sharp peaks corresponding to Mc and Ab are clearly visible in the spectra of the IK samples from Imerys (Fig. 2b). In contrast, Na-rich powders from Llançà (Fig. 2d) display well-defined Ab peaks, but significantly less pronounced Mc peaks. These peaks exhibit lower intensity and considerable broadening, particularly in the LNa samples (Fig. 2f), where Mc reflections nearly disappear. These observations suggest that the dominant K-feldspar phase in the Llançà samples are more disordered, likely tending toward Or rather than the well-ordered Mc.

Particle size characterization was performed using laser diffraction, and the key metrics – D10, D50, D90, and average particle size – are presented in Table 2. D10, D50, and D90 values represent the diameters below which 10 %, 50 %, and 90 % of the particles fall, respectively. Complementary surface area measurements were conducted using the BET method. These analyses revealed surface area variations of up to a factor of four across the samples, ranging from 1.35 m² g⁻¹ for LNaK1 (the lowest) to 5.29 m² g⁻¹ for LNaK2 (the highest). For the LNa2 sample, two distinct powders were provided, each produced from the same bulk material but subjected to different milling durations, resulting in varying particle size distributions.

Figure 3a shows the fraction of frozen droplets, fice(T), for suspensions of the Na-rich feldspar LNa2 over a wide range of concentrations. At the highest concentration (5 wt %), nearly 90 % of the droplets froze once the temperature reached approximately -4 °C. As the concentration decreased, freezing shifted progressively to lower temperatures; at 0.01 wt %, 90 % of the droplets remained unfrozen until the temperature dropped below -18 °C. At the lowest concentrations (≤ 0.001 wt %), the freezing behavior became indistinguishable from that of pure water and was therefore excluded from the analysis shown in Fig. 3b and c, resulting in a maximum of seven concentrations being plotted.

Each fice(T) curve in Fig. 3a represents data from 192 to 288 droplets (corresponding to 2–3 replicates of 96-droplet arrays). Across the full range of concentrations tested, between 1500 and 2500 freezing events were recorded per feldspar powder, providing a large dataset for characterizing the ice-nucleating activity of the mineral. The complete set of fice(T) curves for all samples is shown in Fig. S4 of the Supplement. Additional tests were performed to assess the robustness of the measurements. When fice(T) was recorded over three successive freeze–thaw cycles using the same droplet arrays, no significant differences were observed, confirming the stability of the nucleation behavior (see Fig. S5 in the Supplement). Although the low LOI values indicate the absence of substantial organic content in the feldspar samples, selected samples were also tested after heating at 90 and 110 °C for 30 min (see Fig. S6 in the Supplement). No changes in freezing behavior were detected, further suggesting that biological contamination does not contribute to ice nucleation in these samples

To further evaluate the IN behavior of feldspar powders, the cumulative freezing spectrum of each mineral was derived based on Vali's methodology (Vali, 1971). This approach assumes that each individual INP has a characteristic freezing temperature that is independent of its thermal history. Within each droplet, the INP with the highest nucleation temperature governs the freezing event. Given an initial number of droplets, N0, and the number of frozen droplets, NF(T), the cumulative number of active INPs Nm(T) can be calculated as in Eq. (1) (Vali, 1971, 2019; de Almeida Ribeiro et al., 2023) 1Nm(T)=-1Xln⁡NL(T)N0=-1Xln⁡1-fice(T) where NL(T)=N0-NF(T) is the number of unfrozen droplets at temperature T, and X is a normalization factor (e.g., per unit mass, volume, or surface area of the INPs). For mineral dust, X is often taken as the ice-nucleating surface area available per droplet (X=Sa), which yields the cumulative active-site density per unit area, denoted as Ns(T). In Fig. 3b, the Ns(T) spectrum is shown, obtained from Eq. (1), and normalized to the specific surface area of the powders, which was determined using the BET method. Notably, the contributions derived from different suspension concentrations overlap well within the temperature range where each concentration exhibits significant freezing activity, indicating that the relative distribution of ice nucleation sites is largely independent of dilution. These spectra therefore provide a robust quantitative basis for comparing IN efficiency among the different feldspar samples.

To further illustrate the concentration dependence of the IN activity, Fig. 3c presents Ns(T) for each suspension concentration of one representative feldspar powder. Equivalent behavior is observed for the other feldspars (see Fig. S7 in the Supplement), with the Ns(T) curves offset along the y-axis to facilitate comparison of the temperature ranges and functional forms associated with each concentration. Although our experiments span a wide range of concentrations, including values exceeding those typically expected for atmospherically suspended mineral dust (e.g., 10 wt %, 5 wt %, and 2 wt %), the data clearly show that high-temperature IN activity is observed not only in highly concentrated suspensions but also at lower concentrations (1 wt %, 0.5 wt %, and 0.1 wt %). These observations indicate that the high-temperature IN activity is unlikely to be explained solely by particle aggregation at elevated concentrations, although such effects cannot be entirely excluded. Instead, the results point to intrinsically active ice nucleation sites associated with the feldspar particles. At the highest suspension concentrations, particle aggregation and sedimentation may occur, potentially modifying the effective ice-nucleating surface and limiting the direct atmospheric relevance of these measurements. Nevertheless, the observation of high-temperature IN activity at lower concentrations (≤ 1 wt %) indicates that such active sites remain operative under more dilute concentrations (see shaded region in Fig. 3c). Moreover, mineral particles active at relatively high temperatures have been reported in several previous studies (Harrison et al., 2016; Peckhaus et al., 2016; Kanji et al., 2017; Whale et al., 2017; Keinert et al., 2022). In particular, the Ns(T) spectra derived from suspensions of a microcline perthitic feldspar (labelled FS06) reported by Keinert et al. (2022) at concentrations of 1 wt %, 0.1 wt %, and 0.01 wt % exhibit temperature ranges and curve shapes that are highly comparable to the Ns(T) results presented here.

Figure 4 compares the cumulative active-site density spectra, Ns(T), of feldspars grouped by composition. Figure 4a shows the K-rich samples (IK1, IK2), Fig. 4b presents the Na-rich samples (LNa1, LNa2), and Fig. 4c compares feldspars with balanced K / Na compositions (LNaK1, LNaK2). Figure 4d contrasts representative spectra from each compositional group: IK1 (highest K content), LNa1 (highest Na content), and LNaK1 (lacking a clear dominance of either K or Na), allowing a direct comparison of IN efficiencies. Figure 4a and b show that, for all powders, the surface density of ice nucleation sites increases sharply as temperature decreases to approximately -5 °C. Between -5 and -15 °C, however, distinct temperature dependencies emerge. The IK samples display a continuous increase in site density on a logarithmic scale, whereas the LNa samples exhibit a pronounced plateau, indicating limited activation of additional nucleation sites within this temperature interval. At temperatures below -15 °C, both feldspar groups again show a marked increase in Ns(T). The LNaK powders exhibit site densities and temperature dependencies that lie between those observed for the IK and LNa samples. This behavior is further highlighted in Fig. 4d, where IK and LNa samples show comparable increases in Ns(T) above -5 °C, while LNaK samples exhibit substantially lower site densities. Below -15 °C, the site densities of the LNa and LNaK samples converge, whereas those of the IK samples remain systematically higher. Collectively, these results indicate that, although the Na / K ratio cannot be used as a sole criterion to determine the ice-nucleating activity of feldspars, it plays an important role for the perthitic feldspars studied here. Samples with similar alkali compositions exhibit nearly overlapping freezing spectra, thereby supporting the classification presented in Table 1.

LNa2 powder was provided by the mine in two different particle-size fractions, referred to as “small-particles” and “large-particles,” with specific surface areas of 5.43 and 1.37 m² g⁻¹, respectively (see Table 2). The only difference in their preparation was the grinding time. Figure 5 presents the corresponding cumulative freezing spectra normalized by mass (Fig. 5a) and by specific surface area (Fig. 5b). When normalized by mass, the finer particles exhibit higher Nm(T), reflecting their larger surface area per unit mass rather than an intrinsically higher IN activity. Upon normalization to surface area, however, the spectra converge to nearly identical Ns(T) curves, indicating that IN efficiency is governed by the mineral surface itself and is independent of particle size within the size range investigated here.

The onset temperatures and spectral trends observed in this study are consistent with previously reported cumulative active-site density freezing spectra, Ns(T), for perthitic feldspars. To illustrate this, Fig. 6 compares representative Ns(T) spectra from each feldspar type analysed here with literature data for perthitic feldspars (Whale et al., 2017), together with two feldspars exhibiting exceptionally high IN activity, nearly pure microcline and albite, reported by Harrison et al. (2016).

A key observation, both in this study and in previous work (Whale et al., 2017; Harrison et al., 2016), is that freezing initiates within a narrow temperature interval, approximately between -2 and -4 °C. This range is clearly resolved in our measurements, most likely because the relatively high feldspar concentrations employed (2 wt %, 5 wt %, and 10 wt %) allow for a precise determination of initial freezing events. Notably, onset temperatures are remarkably similar across all samples investigated. The average temperature of the first ten freezing events, T(i,10), falls within this narrow interval for all feldspars: T(i,10)(IK1)=-2.4 °C, T(i,10)(IK2)=-2.9 °C, T(i,10)(LNaK1)=-3.9 °C, T(i,10)(LNaK2)=-2.4 °C, T(i,10)(LNa1)=-2.7 °C, T(i,10)(LNa2)=-2.3 °C (see Table S3).

In contrast, Fig. 6 shows that onset temperatures reported in the literature (Whale et al., 2017; Harrison et al., 2016) are generally slightly lower than those measured here, with the exception of the Dark Shap feldspar. Specifically, calculated values of T(i,10) from reported data are: -3.9 °C (KB14), -5.5 °C (Perthite), -6.3 °C (Keystone), -2.5 °C (Dark Shap), -4.5 °C (Light Shap), -4.3 °C (LD4). It is important to note that these studies used substantially lower feldspar concentrations (0.1 wt % and 1 wt %) than those employed in the present work. At lower concentrations, the detection of IN sites that exists in a very low density on the surface, becomes more challenging, which may bias the apparent onset temperature toward lower values. For both microcline (TUD#3) and albite (Amelia albite) from Harrison et al. (2016), and microcline in particular, the freezing spectra exhibit a very steep (nearly vertical) increase at higher temperatures, indicating high IN efficiency. This behavior suggests that increasing particle concentration would not substantially shift the onset temperature to warmer values. The narrow and consistent onset temperature range observed across alkali feldspars therefore points to a common type of highly efficient nucleation site, likely shared among perthitic feldspars.

In contrast, other feldspar groups, such as plagioclase, exhibit significantly lower onset temperatures (around -10 °C; Harrison et al., 2016; Welti et al., 2019). Although immersion freezing experiments for plagioclase were not conducted in this study, it is unlikely that increasing particle concentration would shift their initial freezing temperatures into the range observed for alkali feldspars. Consequently, the differences in the cumulative freezing spectra (Fig. 5) are best interpreted as reflecting variations in the surface density of these highly active sites, expressed as Ns(T) (sites per cm⁻²), as well as the presence and relative abundance of additional, less active site populations that contribute to IN at lower temperatures.

As shown in Fig. 4, powders originating from the same mine and exhibiting similar mineralogy and K / Na ratios display nearly identical cumulative freezing spectra within the statistical uncertainty of the measurements. For example, IK1 and IK2 show comparable freezing behavior and active-site densities (Fig. 4a), although IK1 exhibits a slightly higher site density down to approximately -15 °C, below which deviations become more pronounced. It should be noted that, although IK1 and IK2 were collected from different extraction fronts within the same mine, the absence of precise spatial coordinates limits a more detailed assessment of site-specific geological variability.

In contrast, feldspars from the Llançà mine were collected with well-defined location data, enabling a more detailed evaluation of the influence of geological variability on IN properties. Samples were obtained from three distinct extraction fronts separated by more than 500 m, each characterized by different alkali compositions: one dominated by Na-rich feldspar, another by K-rich feldspar, and a third exhibiting more heterogeneous Na / K compositions (see Fig. S1 in the Supplement). Feldspars LNa1 and LNa2, sourced from the Na-rich front and the compositionally variable front, respectively, display nearly identical freezing spectra (Fig. 4b), consistent with their similar K / Na ratios and mineralogical characteristics (Table 1). A comparable trend is observed for LNaK1 and LNaK2, collected from the compositionally variable and K-rich fronts, respectively. Although LNaK1 exhibits a slightly higher K / Na ratio, it shows a lower density of active nucleation sites down to -15 °C (Fig. 4c). The close similarity in ice-nucleating behavior between these samples suggests that they may expose analogous surface features governing IN activity, potentially arising from similar geological histories or alteration processes. However, since similar surface motifs can form through different mechanisms (Kiselev et al., 2021), this interpretation should be considered indicative rather than definitive. The robustness of these observations is further supported by the finding that, within the particle size range investigated, powder preparation from bulk rock has a negligible impact on intrinsic IN efficiency. Powders of the same mineral ground to different particle sizes were compared (Fig. 5). At equal mass concentrations, finer powders induce a higher number of freezing events per temperature interval, reflecting increased Nm(T) due to greater exposed surface area. However, after normalization by the BET-derived specific surface area (Sa), the freezing spectra converge across all particle sizes, indicating that the observed particle-size dependence arises solely from differences in surface area exposure rather than from changes in intrinsic surface properties.

A previous study by Welti et al. (2019) reported a correlation between particle size and IN activity; however, this effect was restricted to particles smaller than 1 µm, which are significantly finer than those examined in the present study. The same work also proposed the Rb / Sr ratio as a potential predictor of freezing efficiency, with higher ratios associated with enhanced IN activity. Our results are consistent with this observation: IK1 and IK2 exhibit Rb / Sr ratios nearly an order of magnitude higher than those of the other samples (Table S1) and correspondingly show higher active-site densities between -5 and -20 °C (Fig. 4a).

To further explore differences in freezing spectra across feldspars, the distribution of heterogeneous nucleation temperatures was analyzed using the Heterogeneous Underlying-Based (HUB) method (de Almeida Ribeiro et al., 2023; Renzer et al., 2024). This approach employs a stochastic optimization algorithm to fit the experimental cumulative spectra with a linear combination of Gaussian subpopulations, each representing a distinct class of IN sites. The resulting differential spectra resolve the distribution of nucleation temperatures, thereby enabling characterization of the underlying subpopulations.

Figure 7a shows the cumulative freezing for LNa1, with the optimized HUB fit (in red), assuming the differential spectrum consists of a linear combination of three Gaussian subpopulations (Fig. 7b). Figure 7c displays the corresponding normalized distribution function, ns(T), representing the differential freezing spectrum. A mean squared error (MSE) below or approximately 0.01 was achieved when a linear combination of three distinct subpopulations was included in the fitting. Fitting the spectra with two subpopulations, corresponding to two distinct IN classes, failed to reproduce the initial freezing behavior of feldspar powders, yielding an MSE of 0.0652. A more accurate solution is achieved by considering three subpopulations, which lowers MSE to 0.0041 (see Fig. S8 and Table S4 in the Supplement).