In this study, we used five coal fly ash (CFA) samples that were collected from the electrostatic precipitators (EPs) of five different power plants – four in the USA and one in the UK. The four CFA samples from the USA were supplied by the Fly Ash Direct Ltd.©, USA. The CFA samples were sourced from the following power plants: Clifty Creek Power Plant in Madison, Indiana (hereafter labelled as, CFA_Cy); Miami Fort Generating Station in Miami Township, Ohio (hereafter labelled as, CFA_Mi); Joppa Generating Station in Joppa, Illinois (hereafter labelled as, CFA_Ja); and J. Robert Welsh Power Plant in Titus County, Texas (hereafter labelled as, CFA_Wh). This is the same set of samples also studied and reported in Garimella (2016). Garimella (2016) grouped CFA_Ja and CFA_Wh fly ash samples as class C type, whereas CFA_Cy and CFA_Mi were grouped as class F, which is broadly based on the calcium oxide (CaO) composition. A typical mass fraction of CaO in class F CFA particles is ∼1–12 wt%, whereas class C has higher CaO contents, sometimes up to 40 wt% (Ahmaruzzaman, 2010). A new CFA standard classification system suggests that CFA samples can be sialic (S), calsialic (CS), ferrisialic (FS), and ferricalsialic (FCS) (Vassilev and Vassileva, 2007). However, no further information on chemical composition was provided by Garimella (2016) for a more quantitative classification of the USA CFA samples.

The UK coal fly ash sample was obtained from one of the major power plants in the UK and is referred to as CFA_UK throughout this report. The operator of the UK power plant prefers anonymity; hence, no specific name is mentioned here. The CFA particles collected from EPs are the same particles that could have been directly released into the atmosphere in situations where EPs malfunction or are inefficient. Also, the CFA particles which are emitted indirectly into the atmosphere by road transportation, application in agricultural fields, industrial sites, road construction, and other sources are the same CFA particles as collected from the EPs (Buhre et al., 2005). First, all raw CFA samples were sieved with a FRITSCH sieve set-up (ANALYSETTE 3, 03.7020/06209, Germany) to obtain 0–20 µm diameter size fractions, which were later used for the experiments.

All investigations were carried out in the Aerosol Interactions and Dynamics in the Atmosphere (AIDA) aerosol and cloud simulation chamber. This is an 84 m3 aluminium vessel sitting in a temperature-controlled housing, where the pressure, temperature, and relative humidity are well-controlled depending on the experimental requirements. In addition, a suite of instruments is connected to the chamber for direct in situ measurements or extractive measurements after sampling air from the chamber. A detailed description of the AIDA chamber and its instrumentation has been previously reported in various works (including but not limited to Möhler et al., 2003; Steinke et al., 2011; Wagner et al., 2009). Here, a brief overview of the devices which were employed in our study is highlighted.

A combination of an aerodynamic particle sizer (APS, TSI GmbH, USA), and a scanning mobility particle sizer (SMPS, TSI GmbH, USA) was used to measure the size distribution of the CFA aerosol particles in the AIDA chamber. The SMPS instrument measures in the size range from 13.3 to 835.4 nm, whereas the APS has a larger detection size range (0.5–20 µm). Both instruments were operated at the same time to obtain the full-size distribution spectrum of the particles. A condensation particle counter (CPC3010, TSI, USA) was used to measure the number of aerosol particles in the chamber per volume. We also deployed two optical particle counters (OPCs, WELAS 2000, PALAS GmbH, Germany), which were connected to the base of the chamber to sample and count aerosol particles, cloud droplets, and ice crystals and also measure their respective optical sizes. Each of the OPCs had a different detection range (0.7–46 and 5–240 µm, respectively). The data obtained from the WELAS systems were later used to calculate the ice particle number concentration in the chamber during expansion cooling experiments with an uncertainty of ±20 %. The water vapour concentration in AIDA at every stage of the experiment was measured with tunable diode laser (TDL) spectrometers, from which the relative humidities with respect to water (RHw) and ice (RHice) were calculated with ±5 % uncertainty (Fahey et al., 2014). The spatial and temporal homogeneity of the temperatures in the AIDA chamber is better than ±0.3 K. In this report, the mean gas temperatures will be given throughout the paper.

CFA aerosol particles were injected into the AIDA chamber with a rotating brush generator (RBG, RBG1000, PALAS GmbH, Germany) connected to the chamber with cleaned Teflon and stainless-steel tubing. We coupled the RBG to two cyclones placed in series to eliminate particles larger than 3 µm diameter. Cyclone 2 (D50 cut-off = 3.7 µm) was placed before cyclone 3 (D50 cut-off = 2.3 µm) in the set-up. The overall aim was to obtain smaller sized particles (<∼2.5 µm), which are more atmospherically relevant, especially for long-range transportation in the atmosphere (Prospero, 1999).

Samples of CFA particles were collected on a Nuclepore filter (25 mm diameter, 0.02 µm pore size, Whatman^®, USA) from the AIDA chamber. The sampling was carried out with a mass flow controller (MFC, Tylan^®, UK) running at 2 L min-1 for 30 min. The loaded filters were sputter-coated with 1 nm platinum to improve the conductivity, and the images were taken with an environmental scanning electron microscope (ESEM, FEI Quanta 650 FEG). Coating of the filters did not affect the morphology of our samples because the coating thickness was 1 nm and thus below the SEM resolution. A different model of ESEM (ThermoFisher Scientific Quattro S) was used for the USA CFA samples. With this new ESEM model, we were able to obtain images of the CFA particles under grazing viewing angles similar to 3-D images (see Fig. 1).

We adopted the Brunauer–Emmett–Teller (BET) method (Brunauer et al., 1938) to measure and analyse the specific surface areas (SSAs) of the five CFA samples. The CFA samples were degassed at ∼368 K for 24 h before measuring the molecular adsorption on the particles (a five-point BET model was used). During the degassing process <8.5 % mass loss was recorded for all the CFA samples. Specifically, we used argon gas (87.3 K) as the adsorbent instead of the standard nitrogen gas, hence, we tagged it BETAr. Argon gas provides better adsorption for the estimation of SSA because of its monatomicity and non-localization of the adsorbent during adsorption (Rouquerol et al., 2014; Thommes et al., 2015). This measurement was performed with an Autosorb 1-MP instrument (Quantachrome, Germany). The pore size volumes were calculated with models based on DFT/Monte Carlo methods assuming a mixture of spherical and cylindrical pores on an oxygen-based substrate (Landers et al., 2013; Thommes et al., 2006). The SSA (m2 g-1) from the BETAr measurements and the calculated pore volumes for all the CFA samples are presented in Table 1. All adsorption and desorption isotherms of the different CFA samples are available in the Supplement (Fig. S1).

CFA aerosol particles were first injected into the chamber filled with synthetic air at a particular temperature – hereafter referred to as start temperature (Tstart) – and mixed with the aid of a big fan installed at the lower level of the chamber. After the injection into the AIDA chamber, the CFA particles were probed in two different ways. In the first type of experiments, the particles' inherent ice nucleation ability was tested at temperatures between 261 and 228 K by means of an expansion cooling cycle. For this purpose, the pressure of the chamber was reduced with the aid of a vacuum pump (Möhler et al., 2005). Cooling and the concomitant increase of the relative humidity triggered the droplet activation of the particles, and a subset of the CFA particles nucleated ice via immersion freezing during continued pumping. Generally, pumping was stopped when the maximum RHice was reached.

In the second type of experiments, a temperature-cycling and freezing (TCF) protocol was adopted. Previously, this method had been used for similar experiments with other aerosol types in the AIDA chamber (Wagner et al., 2012, 2016). In the TCF procedure, the CFA particles were injected into the AIDA chamber (∼1300–1600 particles per cm3) at ∼253 K and cooled to ∼228 K. During the cooling process, a rate of 5 K h-1 was achieved. The CFA aerosol particles were then warmed to 253 K (or the desired Tstart) at 2.5 K h-1, as described by Wagner et al. (2016). During the entire cooling and warming process (Fig. 2), the relative humidity prevalent in the AIDA chamber was slightly below ice saturation, as controlled by an ice layer on the inner chamber walls. The slight sub-saturation of the chamber air with respect to ice may be attributed to some internal heat sources which increased the gas temperature by a few tenths of a Kelvin compared with the wall temperature (Wagner et al., 2016). After warming, the particles' ice nucleation ability was probed in an expansion cooling run as described above. Details of the various experiments that we conducted and the outcomes are shown in Table 2.

In this study, we used the ice-active fractions to compare the data from the various experiments performed. The fraction of ice frozen (i.e. the ice-activated fraction, fice) was calculated as the number of ice particles detected divided by the total number of seed aerosol particles present in the chamber (Vali, 1971). The uncertainty associated with our fice calculations is ∼±20 % (Möhler et al., 2006). The fice data in each experiment are plotted in Figs. 3, 4, 6, and S2–S5. For each experiment, the maximum ice-activated fraction (fice,max) values are presented in Figs. 5, 7, and 8.

We start our discussion with the CFA_UK particles. When probed in an expansion cooling run at Tstart=261 K, the ice-active fraction was generally below the detection limit of 0.02 % (Fig. 3a). However, at Tstart=253 K, about 0.19 % of the particles had nucleated ice via the immersion freezing mode in the course of the expansion cooling run until the minimum temperature of 244 K was reached (Fig. 3b). The ice-active fraction encountered during the expansion cooling run at Tstart=245 K was a factor of 10 higher than the run started at 253 K. At t∼300 s, the homogeneous freezing mode kicked in (see the illustration in Fig. 3c). In our analyses, ice particles detected just before, during, and after such events were omitted from the ice particle counts. In summary, the CFA_UK particles were observed to be active in the immersion freezing mode at temperatures below 253 K; however, the ice-activated fractions were rather low and exceeded 1 % only at temperatures very close to the homogeneous freezing threshold of pure water. The homogeneous freezing threshold temperature observed in our experiments (237.0 K) agreed with previous reports (Benz et al., 2005; Schmitt, 2014). In contrast, for the experiment at Tstart=228 K (Fig. 3d), more than 64 % (T=220 K) of the aerosol particles nucleated ice directly from the CFA_UK particles at very low supersaturations. This means that within a change of only 9 K from the homogeneous freezing temperature of pure water (237 K) to the expansion run started at 228 K, the ice-active fraction of the CFA_UK particles increased by almost 2 orders of magnitude. A similar increase in the heterogeneous ice nucleation ability has been previously observed for zeolite and illite particles (Wagner et al., 2016), and temperature-cycling experiments based on these particles have substantiated that the PCF mechanism is the most likely explanation for the sudden increase of the particles' ice nucleation behaviour below the homogeneous freezing temperature of supercooled water. Following the experiment at Tstart=228 K, we hypothesized that PCF may also be the dominant nucleation pathway for the CFA particles. To verify this hypothesis, we adopted the TCF approach as discussed in Sect. 3.2.

Other CFA samples studied here – CFA_Cy, CFA_Mi, CFA_Ja, and CFA_Wh – were also tested for their inherent ice-nucleating properties in the immersion freezing mode at Tstart=251, 250, 251, and 248 K, respectively (Figs. 6, S3, S4, S5). The onset temperatures (Tonset) are reported in Table 2. Here, we defined our Tonset in each experiment as the temperature where the fice is >0.1 %. In order to compare the inherent ice nucleation behaviour of the five CFA samples investigated, we have tabulated the fice,max (%) for experiments with a similar starting temperature of about 250 K (Table 2, experiment numbers 3 and 5–8). The results reveal a significant spread in the ice-activated fractions, with CFA_Wh (∼26 %) > CFA_Ja (∼17 %) >> CFA_Cy (∼1.5 %) = CFA_Mi (∼1.5 %) > CFA_UK (∼0.17 %). This huge variation in the particles' inherent ice nucleation activity is probably related to differences in morphology, elemental composition, and/or surface functionalization. The observed differences in their inherent ice-nucleating abilities may also be due to variabilities in their chemical and mineralogical compositions. Garimella (2016) reported that the four CFA samples from the USA belonged to different classes of fly ash and these groupings are based on the chemical compositions (Garimella, 2016). Further analyses on the distribution of the ice nucleation active sites densities of these CFA particles are outside the scope of the current report and will be presented in a separate communication.

Coal fly ash particles from other sources have been reported to nucleate ice inherently at much higher temperatures. Previously studied CFA particles were suspended in deionized water before ice nucleation properties were investigated on a cold stage set-up. For example, a particular sample from one of the UK power plants was reported to already nucleate ice in the immersion freezing mode starting at 257 K (Umo et al., 2015). This sample also showed a steep curve in the fice, indicating the presence of unique ice-active sites which may be similar to what we observed in CFA_Ja and CFA_Wh. Grawe et al. (2018) reported even higher freezing temperatures (at 265 K) for CFA particles obtained from a power plant in Germany. This was again attributed to the unique composition of CFA samples. However, both studies were performed with drop freezing assay techniques and with much larger particles than reported here. Moreover, in a drop freezing assay method, a droplet can contain many particles, whereas each cloud droplet activated in the AIDA chamber only contains a single particle. Hence, the probability of observing freezing events in drop freezing assay at much higher temperatures was higher than in the AIDA experiments where smaller particle sizes were explored. A combination of both techniques in future studies could ultimately yield a parameterization of the heterogeneous ice nucleation activity of the CFA particles over the entire range of temperatures in the mixed-phase cloud regime. In another study, particles in a plume from a coal-fired power plant were not considered ice active at temperatures above 253 K (Schnell et al., 1976). However, when similar experiments were conducted at a higher supersaturation, the particles' ice nucleation ability increased, indicating that CFA particles could act as good INPs even at temperatures as high as 263 K (Parungo et al., 1978). However, in these experiments, not many details on the exact experimental conditions are available for a direct comparison with our experiments. Also, the particles in the plume were not well characterized; hence, the plume may have contained other ambient aerosol particles.

Generally, for investigations with a measurement set-up that requires a dry generation method, much lower temperatures are reported as inherent ice-nucleating temperatures of CFA as INPs. A study of CFA samples from Germany in a laminar flow tube in Leipzig called Leipzig Aerosol Cloud Interaction Simulator (LACIS) showed ice nucleation from ∼247 to 236 K (Grawe et al., 2016, 2018). Although this freezing temperature range is comparable to what we observed with our samples (Table 2), it should be noted here that the particle size of the CFA samples used in Grawe et al. (2016, 2018) is different from the size range used in our study. First, the average median particle diameter of our CFA samples is 0.58 µm, whereas Grawe et al. (2016) reported an average diameter of 0.3 µm. This can also have an impact on the behaviour of INPs (Garimella, 2016). Second, we should state here that these particles are from different sources – hence, they might have different mineral (or chemical) compositions as well as surface properties. Aerosol compositions and surface properties have been clearly established to influence the ice nucleation behaviour of INPs (Fitzner et al., 2015; Harrison et al., 2016; Isono and Ikebe, 1960; Lupi et al., 2014; Mason and Maybank, 1958). Third, the different measurement techniques applied in each study can also introduce some differences (Grawe et al., 2018). In comparison with other aerosol types, the ice nucleation activities of CFA particles in the immersion freezing mode are considerably higher than e.g. soot particles (Mahrt et al., 2018), but less active compared with some biological materials (Suski et al., 2018). Generally, the ice-nucleating abilities of CFA samples are similar to the ice-nucleating potential of some mineral components of desert or agricultural soil dusts (Grawe et al., 2018; Umo et al., 2015).

In the previous section, we reported the inherent ice nucleation activity of CFA particles. Here, we show the results for CFA particles that were temporarily exposed to a lower temperature (228 K) before the expansion cooling experiments were conducted. Freezing data after the temperature-cycling and freezing (TCF) procedure are presented in Figs. 4a–c, 6b, and panels b and c of Figs. S2–5.

After the TCF process, experiments were conducted with the processed CFA_UK particles following the schematic in Fig. 2. Specifically, we conducted two independent series of experiments, each with a fresh load of aerosol particles, following the sequences Tstart∼250 K → 254 K → 264 K (series I, experiments 9, 10, and 11, data shown in Fig. 4) and Tstart∼251 K → 254 K → 263 K (series II, experiments 12, 13, and 14, data shown in Fig. S2). As the results from both series are very similar, we focus our discussion on the experiments conducted during series I. At Tstart=250 K, we clearly observed an increase in the fice,max of the CFA_UK particles (up to 11 % at T∼246 K) compared with the unprocessed CFA_UK particles that only showed fice,max of 1.6 % at T∼238 K, which was even at a lower start temperature (Tstart=245 K). The processed CFA_UK particles nucleated ice at water-subsaturated conditions with a nucleation threshold in terms of RHice of only about 101 %. In contrast, the unprocessed CFA_UK particles nucleated ice in the immersion freezing mode after exceeding water saturation during the expansion run (corresponding to RHice∼130 %). This means that there was a change in the ice nucleation mode in comparison with the unprocessed CFA_UK particles in the same Tstart range. For the processed CFA_UK particles, there was no droplet activation before the emergence of ice, i.e. ice formation cannot be ascribed to “classical” immersion freezing (Fig. 3a–c). Rather, the ice particles observed were formed directly on the pre-activated CFA_UK particles. Following the history of these particles, we suggest that the ice particles may have been formed by the depositional growth on the ice germs formed in the pores of the particles during temperature cycling.

After the first expansion at Tstart=250 K, we warmed the chamber to 254 K and performed another expansion cooling run. The ice-activated fraction decreased by a factor of 2 compared with the run at Tstart=250 K (fice,max∼3 %), but was still significantly higher than what was observed for the unprocessed CFA_UK particles at a similar temperature (fice,max=0.19 % at T∼245 K). Ice formation by the processed CFA_UK particles again occurred by the depositional growth mode at low ice supersaturation (RHice,max∼109 %), whereas the much smaller ice-activated fraction of the unprocessed particles was due to immersion freezing at water-saturated conditions corresponding to RHice=124 % at Tstart=253 K (Fig. 3).

Afterwards, the same processed CFA_UK aerosol particles were warmed to Tstart=264 K for another expansion cooling run (Fig. 4c). At this start temperature, the ice nucleation ability of the unprocessed CFA_UK particles was below our detection limit of 0.02 % for fice. For the processed CFA_UK particles, however, a maximum ice-activated fraction of 1.3 % was observed at T∼251 K. In contrast to the runs conducted at Tstart=250 and 254 K, the ice cloud was not formed at low supersaturation values with respect to ice, but appeared just at the instant of droplet activation (RHice∼107 %, Fig. 4c). Given the absence of any ice formation for the unprocessed particles, it is highly probable that the nucleation mode of the processed CFA_UK particles, although similar to a classical immersion freezing mode, is in fact related to ice growth from an existing ice germ formed during temperature cycling. This implies that at least 1.3 % of the processed CFA_UK particles still contained ice-filled pores even after warming to 264 K. Such ice formation modes have already been observed for other particle types in similar scenarios (e.g. Mahrt et al., 2018; Wagner et al., 2016), and have been ascribed to the condensational growth of the ice germs formed in the pores or crevices of these particles. Figure 5 shows the summary of the ice nucleation enhancement of CFA_UK particles described above with fice,max and their corresponding temperatures as well as the respective start temperatures of each experiment. It is clear that for the processed CFA_UK particles, the fice,max values are significantly higher than those for the unprocessed particles at a similar Tstart.

In contrast to the CFA_UK particles, the CFA particles from the USA power plants showed less modification of their ice nucleation ability after the temperature-cycling process. A distinct depositional ice growth mode, such as shown in Figs. 4a and b for the CFA_UK particles, was not observed for any of the USA particle types. However, some particle types revealed an improved ice nucleation ability due to the condensational ice growth mode, as exemplified in Fig. 6 for the CFA_Cy particles. While the ice-activated fraction of the unprocessed CFA_Cy particles remained below 0.5 % for temperatures above 244 K (Fig. 6a), the particles subjected to temperature cycling showed ice formation with fice>0.5 % already observed at 249 K (Fig. 6b). Similar to the experiment with CFA_UK at Tstart=264 K (Fig. 4c), this ice mode was instantaneously formed upon droplet activation, i.e. it is most likely related to a condensational ice growth mode. The CFA_Cy particles also showed a tiny depositional growth mode indicated by a few ice particles detected before the droplet activation (Fig. 6b). To better illustrate the generally small differences in the ice nucleation ability of the CFA particles from the USA with and without temperature cycling, we summarize the ice-activated fractions as a function of temperature for both the expansion cooling runs with processed and unprocessed particles in Fig. 7. For the corresponding data of the CFA_Cy particles as discussed above, there is a clear shift of the ice nucleation spectrum towards higher temperatures after temperature-cycling. The difference is much less pronounced for other CFA particles from the USA.

Pre-activated CFA_Ja particles did not show any significant improvement of their ice nucleation ability after the temperature-cycling experiment for expansion cooling experiments started at around 250 K (Fig. 7). Obviously, pre-activation cannot compete with the already very high inherent heterogeneous ice nucleation ability of the CFA_Ja particles at this temperature, meaning that there is no further detectable increase in the ice-activated fraction after the TCF cycle. However, the pre-activation phenomenon becomes visible when further warming the pre-activated CFA_Ja particles to a higher starting temperature (256 K, Fig. S4c). Here, the processed CFA_Ja particles showed a small nucleation mode with fice∼1 % at 252 K just when exceeding water saturation during the expansion run. Given that the threshold temperature for exceeding an ice-activated fraction of 1 % for the unprocessed CFA_Ja particles was as low as 246 K, the observed ice nucleation mode for the processed CFA_Ja particles at 252 K can most likely be ascribed to the condensational growth of pre-existing ice, generated in the pores of the particles during the TCF cycle.

Similar to the CFA_Ja particles, the CFA_Wh particles also did not significantly change their ice nucleation ability after the TCF cycle when probing them at starting temperatures of 248–249 K (Fig. 7), i.e. in a temperature range where the particles' inherent heterogeneous ice nucleation ability is already very high. However, the smaller nucleation mode with fice∼2 % that was observed after further warming the processed CFA_Wh particles to 256 K (Fig. S5c) is likely again due to the condensational ice growth mode. The CFA_Mi particles showed the smallest variation with respect to their ice nucleation ability after the TCF cycle. In addition to the comparable ice nucleation behaviour before and after temperature cycling at a starting temperature around 250 K (Fig. 7; Fig. S3a, b), the processed CFA_Mi particles also revealed only a tiny condensational ice growth mode at a higher starting temperature of 255 K with fice,max∼0.1 % (Fig. S3c).

The degree of ice nucleation enhancement by CFA particles differs from sample to sample. The enhancement capability of the CFA samples studied here adheres to the following order: CFA_UK >>>> CFA_Cy > CFA_Wh > CFA_Ja > CFA_Mi. The ranking is based on the start temperature, fice,max, and the relative humidity as summarized in Figs. 5 and 7. Morphology, chemical composition, surface area, and pore volume are important parameters influencing the efficiency of the PCF mechanism. In the following, we discuss whether differences in these properties can account for the different behaviour of the CFA particles after temperature cycling.

The morphology of the five samples is shown in Fig. 1 for selected typical particles. The SEM images showed that the CFA particles have some degree of roughness, coatings, layers, and mesh-like structures on their surface. Although the overall particle habit is spherical, as many electron micrographs of CFA have shown (Blissett and Rowson, 2012; Fisher et al., 1978), they have no smooth surface. Of the five CFA samples, CFA_UK had the highest degree of deformity on the surface as indicated in Fig. 1a–f. We attempted to focus into the surface (up to ∼50–100 nm resolution) to identify the potential pores and crevices, but it was difficult to establish a clear view of the pores (Fig. 1c, f). Classical nucleation theory (CNT) and empirical calculations have shown that pore diameters of about 5–8 nm (mesopores) contribute to a particle's pre-activation ability at ice sub-saturated conditions (Marcolli, 2014; Wagner et al., 2016). A more recent study using CNT and molecular dynamics has shown that it is not enough to have pore diameters of the above size but that a network of closely spaced pores is necessary to overcome the free energy required for a macroscopic ice-crystal growth from narrow cylindrical pores (David et al., 2019).

In previous studies, it has been shown that the specific surface area and pore volume of fly ash particles generated from pulverized coal combustion are very likely dependent on the particle size (Schure et al., 1985; Seames, 2003). To better understand the nature of the CFA surfaces, we measured the specific surface area (SSA) of the sieved bulk samples (0–20 µm) using the BET method but with argon gas rather than nitrogen (Gregg et al., 1967; Thommes et al., 2015). We obtained five-point BETAr surface areas as tabulated in Table 1. The BETAr of CFA_UK had the highest SSA of 14 m2 g-1, which was a factor of 3 higher than those of the other CFA particles: CFA_Cy (5 m2 g-1), CFA_Mi (4 m2 g-1), CFA_Ja (4 m2 g-1), and CFA_Wh (3 m2 g-1). The high SSA of CFA_UK is indicative of the presence of crevices in the form of pores or grooves and could therefore account for the ice nucleation enhancement exhibited by the pre-activated CFA_UK particles compared with the other CFA particle types in this study. Note that this does not necessarily mean that all particles with high SSA such as soot particles will show pre-activation and ice nucleation enhancement. For example, pre-activation was not observed for water-processed soot particles (Wagner et al., 2016), although other soot types have been suspected of showing considerable ice activity via the PCF mechanism (Mahrt et al., 2018; Wagner et al., 2016).

We also report the pore volume (PV) of the investigated particles (Table 1). The PV was calculated with a DFT/Monte Carlo model assuming that the pore diameters are not greater than 100 nm. In our results, CFA_UK had the highest PV (0.05 cm3 g-1), about 4 to 5 times higher than the other CFA samples. Amongst these other CFA samples, there was no clear correlation between PV and corresponding ice nucleation enhancement. For example, CFA_Ja and CFA_Wh had very similar PV (0.009 and 0.010 cm3 g-1, respectively) but CFA_Wh showed a higher ice susceptibility to pre-activation than the former. Another example is CFA_Cy (0.012 cm3 g-1), which has a PV similar to the CFA_Mi sample (0.013 cm3 g-1), but only the processed CFA_Cy particles showed a clear pre-activation ability due to the PCF mechanism. Specific surface areas correlate with the PV (Sigmund et al., 2017), however, it is difficult to ascertain the geometries of the pores or crevices contributing to the surface area. CFA particles are very unique particles in that some of them can be cenospheres (hollow particles with a tiny opening). They can also be plerospheres, i.e. a case whereby smaller particles fill the larger cenospheres (Alegbe et al., 2018; Fisher et al., 1978; Goodarzi, 2006; Goodarzi and Sanei, 2009). The cenospheres and plerospheres present in the CFA samples could increase the pore volume of these particles, thereby leading to a higher uncertainty in estimating the pore size. Currently, it is highly difficult to estimate the pore sizes of the CFA particles based on the PV alone, except in the case of a well-defined pore model and morphology. We suggest that knowing the possible geometries of defects on the surface of INPs may help to predict their pre-activation behaviour.

In a previous study, Wagner et al. (2016) investigated the pre-activation behaviour of INPs by the PCF mechanism in the AIDA cloud chamber with a similar measurement routine as described in Sect. 2.6. In this study, a wide range of INPs was tested including illite NX, diatomaceous earth, zeolites, dust samples from the Canary Islands, the Sahara, and Israel, graphite spark generator soot (GSG soot), and volcanic ash (Wagner et al., 2016). It was reported that illite NX, diatomaceous earth, and mesoporous zeolite CBV 400 showed a significant ice nucleation enhancement in the depositional ice growth mode, with ice-active fractions of 5.9 %, 3.8 %, and 3.7 %, respectively, at a starting temperature of ∼250 K (Fig. 8). At higher starting temperatures, the ice-activated fractions in the condensational ice growth mode were typically around 1 %. Another group of INPs including CBV 100 (untreated microporous zeolites), Canary Island dust, and GSG soot showed much smaller depositional ice growth modes with ice-activated fractions below 1 %. Finally, volcanic ash, water-processed GSG soot, and Saharan and Israeli dust particles did not show any enhancement after the pre-activation process in the depositional nor the condensational ice growth mode.

In this context, the ice nucleation enhancement observed for the CFA_UK particles at a starting temperature of 250 K in the depositional growth mode with fice,max∼11 % (Fig. 4a) is by far the highest value for any particle type investigated so far (Fig. 8). In contrast, the pre-activation efficiency of the CFA particles from the US power plants is comparable in magnitude to the above-mentioned group of CBV100, Canary Island dust, and GSG soot particles with much lower ice-activated fractions. The mean diameters of the particles investigated by Wagner et al. (2016) ranged from 0.21 to 0.43 µm, and were thus smaller than the mean diameters of our CFA particles except for CFA_Mi (0.42 µm). The different pore sizes, morphology, and chemical composition of these INPs may control their susceptibility to the PCF pre-activation mechanism. More studies are required to investigate the role that each of these parameters plays.

Ice nucleation by CFA particles pre-activated via the PCF mechanism could be important for different cloud types. When CFA particles are lofted into the atmosphere, these particles can act as INPs or CCN as well as sinks for other atmospheric species (Dlugi and Güsten, 1983; Havlíček et al., 1993; Herndon, 2016; Korfmacher et al., 1980; Muduli et al., 2014). During their residence time in the atmosphere, the CFA particles can be transported through different relative humidity and temperature regimes. If the particles were temporarily exposed to temperatures below 237 K at high ambient relative humidity, their ice nucleation ability might improve by the formation of ice-filled pores. There is a high potential that the pre-activated CFA particles can be re-circulated as INPs via a sedimentation process into the lower atmosphere to contribute to ice formation in mixed-phase clouds as illustrated in Fig. 9. Some of the atmospheric processes that could aid the re-circulation of the pre-activated INPs are radiative cooling, deep convective flows, sedimentation, and feeder–seeder mechanisms (Carruthers and Choularton, 1983; Highwood and Hoskins, 1998; Hong et al., 2004; Salathé and Hartmann, 1997). By convective atmospheric dynamics, these pre-activated particles could then be released to lower altitudes and trigger ice formation at higher temperatures than expected from their inherent ice nucleation ability. In addition, some CFA particles that initiated cloud glaciation can also be released via cloud evaporation or the sublimation of the ice particles releasing the CFA ice residues back into the atmosphere. These pre-activated CFA INPs can then re-initiate cloud formation at higher temperatures than inherently expected for the same CFA INPs. This process is not peculiar to CFA particles, but is also relevant for other natural and anthropogenic INPs with unique properties such as illite NX, zeolite, and GSG soot that exhibit the PCF mechanism and can have a wider atmospheric implication in cloud formation. Despite the dearth of information on the number concentration of CFA particles in the atmosphere at higher altitudes, there is some evidence that CFA particles are found in ice residues of cirrus and mixed-phase clouds (DeMott et al., 2003; Liu et al., 2018).

We suggest that future modelling work should focus on the impact that pre-activated INPs or INPs with ice-filled pores can have on cloud formation processes. Some observations show that more ice particles are observed at higher temperatures than the amount expected by the available INPs (Hobbs and Rangno, 1985). Aside from secondary ice multiplication processes (Hallett and Mossop, 1974; Phillips et al., 2018), it could be possible that pre-activated INPs also contribute to the higher concentration of ice crystals than are observed in some cases. There are other open questions in these areas such as understanding the timescale and frequency (often or episodic) with which this phenomenon occurs in clouds, the impact of this process in mixed-phase and cirrus cloud formation, and the occurrence at regional and global levels. The PCF mechanism could be potentially important for cirrus cloud systems because CFA particles entrained into the upper troposphere at lower temperatures could already have their pores filled with ice. For instance, our experiment with CFA_UK particles at T∼220 K showed over 60 % ice activation (Fig. 3d).