The SOA particles were produced in the CLOUD (Cosmics Leaving Outdoor Droplets) chamber at CERN during the CLOUD9 campaign between 25 October and 3 November 2014. SOA particles were produced from ozonolysis of α-pinene at low relative humidity with respect to water (RHw) (5–15 %) at four different temperatures, -10, -20, -30 and -38 ∘C, which are relevant for the free troposphere. α-pinene and ozone were introduced to the chamber at the rate of 10 and 1000 mLmin-1 (respectively) over a time period of a few minutes. UV lights (Philips TUV 130 W XPT lamp) were on during this time, resulting in the photolysis of O3 and OH formation. High number concentrations of particles (>105 cm-3) quickly formed a monomodal size distribution which grew to approximately 100 nm in diameter. During the particle growth, α-pinene and ozone were continually introduced into the chamber. The particles were grown to approximately 600–800 nm in diameter in order to follow the size-dependence of their optical properties. The depolarization ratio was measured, and based on the determined depolarization of the backscattered light, it was determined whether the particles were aspherical and thus (highly) viscous (detectable depolarization). Then the RHw was slowly increased for the optical detection of the phase transition of the particles from higher to lower viscosity or liquid phase, i.e. the point where the depolarization ratio decreased significantly to a level of spherical particles. The transition RHw increased with the decreasing temperature, from 35 % at -10 ∘C to 80 % at -38 ∘C. Further experimental details are explained in . The ice nucleation ability of the SOA particles was measured by sampling them from the CLOUD chamber at different growth stages in order to examine particles with different sizes. In Table  the particle sizes and number concentrations for the different experiments are included together with information about the chamber temperature and humidity conditions. The particle number size distributions were measured with several Scanning Mobility Particle Sizer (SMPS) systems (20–800 nm), and optically with an Ultra High Sensitivity Aerosol Spectrometer (UHSAS, Droplet Measurement Technologies, Inc.) in the range from 60 nm to 1 µm. In addition, an Aerodyne high-resolution time-of-flight aerosol mass spectrometer (HR-ToF AMS) was used to determine the chemical composition of the SOA particles.

We used the Spectrometer for Ice Nuclei (SPIN) to measure the ice nucleation efficiency of the SOA particles generated in the CLOUD chamber. SPIN is a new, commercially available portable INP counter manufactured by Droplet Measurement Technologies, Inc. The final version and the performance of SPIN are described in more detail by . The main difference between the version of SPIN used in this study and the final version is related to better temperature control of the final version.

SPIN is a continuous flow diffusion chamber with parallel plate geometry adapted from the design of the Portable Ice Nucleation Chamber PINC and the Zurich Ice Nucleation Chamber ZINC . The aerosol sample flow is set to 1 Lmin-1 surrounded by a sheath flow of 10 Lmin-1 flowing through a chamber where a supersaturation of water vapour with respect to ice is obtained by keeping two ice covered walls at different temperatures below 0 ∘C. The residence time in the upper part of the chamber where ice nucleation may take place is approximately 10 s. The aerosol sample and sheath air flows are then exposed to an isothermal, separately temperature-controlled evaporation section where the unfrozen droplets evaporate while the ice particles are retained, prior to particle detection with a linear polarization optical particle counter (OPC). The OPC laser is polarized, and the intensity of the backscattered light perpendicular to the incident polarization and the intensity of the backscattered polarized light are measured on a particle by particle basis. The polarization-equivalent ratio between the two intensities provides information about the phase state of the particles. The size measurements of the SPIN OPC were calibrated using two sizes of polystyrene latex spheres (PSL), 0.9 and 2 µm, and glass beads of 5 and 8 µm. A power law fit was applied to the calibration data, and the full size distribution was extrapolated from the fit. The detection efficiency of the OPC was investigated with different sizes of monodisperse PSL spheres in the range from 300 nm to 1.0 µm. The lower detection limit of the SPIN OPC was approximately 400 nm, with a detection efficiency of ∼50 %. For particles larger than 700 nm, the detection efficiency was close to 100 %. For particle sizes between 550 and 600 nm, the detection efficiency was approximately 80 %. This allows the SPIN OPC to detect also the seed aerosol particles if they are large enough. Typical backgrounds during the experiments with viscous SOA particles were of the order of 10–20 particles per litre. The frozen fraction lower limit of detection was typically of the order of 1×10-4.

The temperature of the aerosol sample flow and the supersaturation with respect to ice are modelled based on the continuous measurement of the wall temperatures. Due to a temperature gradient between the walls, there is a buoyancy effect to the air mass inside the chamber, pushing the sample flow closer to the colder wall. This effect has been taken into account in the 1-D modelling and calculations of the sample temperature and humidity which are done according to . A condensation particle counter (CPC-3720, TSI) was run parallel to SPIN during half of the runs, when the sample particle size was smaller than 500 nm. SPIN was operated by keeping the colder wall at constant temperature at its lower limit and ramping up the temperature of the warm wall, thus raising the sample temperature from approximately -43 to -35 ∘C and ice saturation ratio Sice from 1 to approximately 1.45–1.5 inside the instrument. In order to minimize fluctuations of the wall temperatures, the regime of operation for the current study was limited to a minimum temperature of -43 ∘C for the colder wall.

SPIN sampled the SOA particles from the CLOUD chamber through a 1.5 m long, quarter inch stainless steel tubing, in which the residence time was approximately 2 s. As the concentrations of the SOA particles in the CLOUD chamber typically exceeded 10 000 cm-3, a dilution system partly filtering away aerosol particles was used to dilute the sample so that the particle number concentrations entering SPIN would stay under 1000 cm-3 in order to avoid saturating the optical detector. The stainless steel tubing together with the dilution system was insulated, but the temperature of the sample was not continually measured at the SPIN inlet, and it is possible that the temperature in the sampling tube was higher than in the CLOUD chamber.

Homogeneous freezing of highly diluted ammonium sulphate droplets was used to evaluate the performance of the SPIN chamber and the optical detector. Ammonium sulphate particles were generated from a 1.0 mass % solution with a medical nebulizer, dried in a diffusion dryer and then size-selected by a Differential Mobility Analyser (DMA). Mean mobility diameters of 200 and 500 nm were chosen, and a CPC (CPC-3010, TSI) was run in parallel to measure the total particle concentration.

The freezing experiments were designed in the following way: first a supersaturation with respect to ice and water was created inside SPIN by diverging the wall temperatures in order to obtain the necessary temperature gradient, while keeping the sample temperature below -37 ∘C, which is the upper threshold temperature for homogeneous freezing in the atmosphere . After water (super) saturation was achieved and ice formation observed, both walls were heated at the same rate in order to increase the sample temperature but keep the water supersaturation, leading to the disappearance of ice. The temperature of the evaporation section walls was kept constant at -35 ∘C.

An example of such an experiment is displayed in Fig. . In the upper left panel, the SPIN wall and sample (lamina) temperatures, as well as the relative humidity with respect to liquid water are plotted as a function of time. The upper right panel shows the particle number concentrations in different size bins from the same experiment. At subsaturated conditions with respect to liquid water only the 500 nm seed aerosol particles are observed. When water saturation is reached shortly after 14.3 h, most of the seeds activate to droplets which then freeze, forming a distinct mode with sizes around 5 µm. As the sample temperature is increased, the ice mode vanishes at approximately 14.4 h and a mode consisting of liquid droplets with sizes of ∼2–3 µm remains.

In the lower three panels of Fig. , the size distributions measured by the SPIN OPC from the same experiment are shown. Panel (a) shows the dry 500 nm seed aerosol particles that have not yet activated. Panel (b) illustrates the situation when most of the ammonium sulphate particles have activated into droplets and frozen. Panel (c) depicts the seed aerosol and liquid, highly diluted ammonium sulphate droplets at water supersaturation at temperatures above the homogeneous freezing point. All the data are normalized with respect to total particle counts from the OPC. The polarization-equivalent ratios (S/P ratios) from the OPC are also different for ice crystals and liquid droplets: for ice, S/P ratios vary between 0.5 and 0.6, whereas for liquid droplets the ratios are significantly smaller, between 0.1 and 0.2. This further confirms that the distinct modes in the size distributions correspond to different phase states, and the size distributions can also be used to define ice nucleation onset points.

The ice nucleation onsets in these homogeneous freezing experiments were obtained in the following way. In Fig. , we use the transition from regime B to regime C, not from regime A to regime B, to obtain the ice nucleation onset temperature and ice saturation ratio. Due to the risk of cold pockets forming during active wall cooling, we scan the aerosol sample temperature upwards until the observed ice crystal mode vanishes and only liquid droplets remain. This temperature, at which a certain fraction (e.g. 10 %) of ice remains, is considered the ice nucleation onset. Although the ice nucleation onset here is actually ice offset, the conditions correspond to the ice nucleation onset conditions.

The homogeneous freezing temperatures for a frozen fraction of 10 % are shown in Fig. . The homogeneous freezing temperatures are all found in the range from -37.9 to -36.6 ∘C, with an average value of -37.3 ∘C. The variation in Sice results from using slightly different temperature gradients to introduce a supersaturation with respect to water inside SPIN. Based on classical nucleation theory and the parameterization of homogeneous nucleation rates of water presented by , a frozen fraction of 10 % can be expected for a temperature in the range from about -38.2 to -37.6 ∘C. Uncertainties and variations in the residence time and the sizes of droplets formed inside SPIN result in the uncertainty in the expected homogeneous freezing temperature. Hence, the homogeneous freezing temperatures reported in Fig.  are within the range or slightly higher than what could be expected from theory. However, with a temperature difference between the SPIN chamber walls close to 20 ∘C, according to model simulations, the aerosol sample temperature range is approximately ±0.4 ∘C with respect to the average temperatures presented in Fig. . Roughly half of the droplets formed inside SPIN will thus be exposed to temperatures down to ∼0.4 ∘C below the average aerosol sample temperature. It would be expected that > 10% of the droplets formed inside SPIN are exposed to temperatures > 0.3 ∘C below the reported average aerosol sample temperatures, which can explain most of the gap between the reported average aerosol temperatures and the theoretical homogeneous freezing temperature.

The freezing point depression caused by ammonium sulphate in the water droplets was calculated using the following formula: ΔT=iM×1.86, where i is the van't Hoff factor of the solute, M is the molar concentration of the solution, and 1.86 KM-1 is the freezing point depression for an ideal solution . For ammonium sulphate, i=2.04 . The droplet diameter here was estimated to be 3 µm. For 200 nm ammonium sulphate particles, this solute effect is negligible (ΔT=0.015∘C); for 500 nm particles the freezing point depression is 0.24 ∘C, which is still within the uncertainty range of the aerosol sample temperature inside SPIN. Thus, we can conclude that the droplets were dilute enough so that the presence of ammonium sulphate did not significantly influence the freezing temperature. With all the uncertainties considered, the correspondence between theory and experimentally obtained homogeneous freezing temperatures with SPIN is good, with a systematic tendency of the experimental average aerosol temperatures being slightly too high (≲ 0.5 ∘C). This off-set is likely to be due to occasional locally slightly colder chamber wall sections in between thermocouples relative to the temperature set-points. Based on the results presented in Fig.  it can be concluded that the aerosol sample conditions inside SPIN can be reproduced with good accuracy between different experiments, and the inferred aerosol sample conditions correspond well to what can be expected from theory. The standard deviation of the experimentally determined homogeneous freezing temperatures for 10 % frozen fractions is 0.50 ∘C. This variation can be considered to reflect the experimental random errors on the average aerosol sample temperature for the instrument for these operation conditions.

Figure  shows results of a typical heterogeneous ice nucleation experiment with viscous SOA particles; this example is from 26 October 2014. Analogous to Fig. , the middle three panels show two particle number size distributions observed during subsaturated conditions with respect to liquid water, and a size distribution from an experiment, when the RH with respect to liquid water was greater than 100 %. The leftmost panel (a) shows the 550 nm seed aerosol and the middle panel (b) depicts the seeds and ice (a size mode around 5 µm) at water subsaturated conditions. A prominent droplet mode can be seen in the third panel (c), when the RHw was greater than 100 %, and it is possible that the SOA particles liquefied and froze homogeneously, as we have observed homogeneous freezing of diluted ammonium sulphate droplets at similar conditions. The different possible freezing mechanisms will be discussed in Sect. 3.4. The droplet mode in panel (c) likely consists of frozen droplets that have not grown very much due to potential activation close to the evaporation section of SPIN. This could also explain the slightly higher S/P ratios (0.3) compared to the S/P ratios of 0.1–0.2 for the liquid droplets in Fig. c.

The fraction of particles activated as ice crystals was defined as the number of particles larger than the minimum between the seed aerosol and ice crystal modes, or between the liquid droplet and ice modes, divided by the total number of particles. In the case of SOA particles, this threshold size was 3 µm. It cannot be ruled out, however, that there is some size overlap between ice crystals and liquid droplets, since due to the design of SPIN, not all particles nucleate ice at the same time and not all the droplets necessarily evaporate in the evaporation section.

We investigated different SOA particle sizes from 120 nm to approximately 800 nm (see Table  for details). Figure  shows exemplary frozen fractions of 330, 550 and 800 nm viscous SOA particles as a function of the ice saturation ratio. The 330 nm particles were produced in the CLOUD chamber at -38 ∘C, and at the time of ice formation the temperature inside SPIN was -38.5±0.5∘C. The 550 nm particles were produced at -20 ∘C, and nucleated ice at -37.2±0.5∘C; the 800 nm particles produced at -10 ∘C nucleated ice at -38.2±0.5∘C. The maximum frozen fractions observed were ∼6–20 % but they did not show any clear dependency on the seed particle size; nor did the 1 or 10 % ice activation onset values, although there is some variation among the different particle sizes. This can also be seen from Table S1 which details all the ice nucleation onset conditions from each experiment day.

SOA particle properties relevant for ice nucleation could depend on the chemical composition and morphology, which could depend on the chamber conditions. The mean atomic oxygen to carbon ratio (O : C) of the SOA particles inferred from AMS measurements was 0.25 throughout the different stages of the different experiments : this would suggest possibly similar chemical composition of the particles throughout the experiments. The apparent similarity in the chemical composition for different SOA particle sizes does not explain why there is no significant change in the frozen fraction or freezing onsets when the seed particles grow larger. Surface area dependency has previously been shown for deposition ice nucleation of different mineral dusts by , but there is no indication of such a dependency for the viscous SOA particles studied here.

The ice nucleation onset conditions were systematically investigated for frozen fractions of 1, 5 and 10 %. These data are listed in Table S1 in the Supplement. In general, the conditions for the observations of 1, 5 and 10 % frozen fractions were very similar. In Fig.  the conditions for 10 % frozen fractions are depicted. In contrast to Fig. , the data in Fig.  were obtained from actual onset of ice formation as depicted in Fig. , not from ice offset measurements. The random instrumental error on the average aerosol sample temperature is expected to be similar to the variations observed above for homogeneous freezing with a standard deviation of 0.5 ∘C. The depicted range of the saturation ratio with respect to ice (Sice) is the modelled equilibrium maximum range the aerosol sample possibly is exposed to – in the vertical and horizontal dimensions – based on the pairwise temperature readings of the chamber walls at four locations. This is illustrated in Fig. S1 in the Supplement. The modelling is done by taking the warm and cold wall temperature pairs at the four thermocouple locations where they are monitored and then calculating the aerosol lamina temperatures and saturation ratios with respect water and ice at those locations according to . The reported ice nucleation onset temperatures in Fig.  are the mean values of the four aerosol lamina temperatures, and the reported Sice values are the mean values of the four calculated lamina Sice values. All the maximum saturation ratios for the observed 10 % frozen fractions are below the depicted lines in Fig.  indicating where homogeneous freezing occurs. In terms of the saturation ratio with respect to water, the maximum modelled values are found in the range 0.90–0.98 and in the range 0.92–0.98 for frozen fractions of 1 and 10 %, respectively. For the water saturation to reach 1 for all of these freezing conditions, a systematic wall temperature deviation in between thermocouples of > 2 ∘C would be required. Such a systematic wall temperature deviation is highly unlikely considering the reasonable and reproducible homogeneous freezing results presented above. Hence, it is highly unlikely that the observed freezing occurring at subsaturated conditions with respect to water is homogeneous freezing.

From our measurements, we have a strong indication that the studied α-pinene SOA induced ice nucleation heterogeneously. Despite instrumental limitations, the results were reproducible and the uncertainty for the ice nucleation onset temperatures and supersaturations could be inferred.

The results presented in the section above clearly indicate heterogeneous freezing of SOA particles below saturation with respect to water vapour. Various freezing mechanisms could potentially be in play. It can be speculated that we observe (i) deposition nucleation occurring directly onto highly viscous SOA particles; (ii) immersion freezing of partly deliquesced SOA particles, where the core of the particle is still (highly) viscous; (iii) hygroscopic growth of the particles leading to freezing of droplets due to suspensions of large organic molecules. The relevance of the first two potential freezing processes are related to the relative timescales of the viscosity transition vs the freezing of the SOA particles for increasing humidities as discussed e.g. by , and . conclude that heterogeneous freezing of biogenic SOA particles would be highly unlikely at temperatures higher than 220 K in the atmosphere since according to their modelling, the timescales of equilibration would be very short. On the other hand, the modelling results presented by indicate that α-pinene SOA particles are likely to exhibit viscous core-liquified shell morphologies on timescales long enough to facilitate ice nucleation via the suggested mechanism (ii) in our study.

In this context, it is worth mentioning that the maximum ice nucleation time in SPIN is of the order of 10 s. However, nucleation taking place on much shorter timescales can be observed if the nucleation rates are high enough to yield detectable numbers of ice crystals. In other words, the observed number of ice crystals corresponds to the time integral over the nucleation rate distribution. This implies that from our measurements, no further conclusions concerning nucleation times and rates can be drawn.

The (iii) potential freezing mechanism has been reported for ice nucleating macromolecules (INM) originating from pollen . It is not likely that the molecules formed in the current study grow to masses comparable to the several kDa reported for the pollen macromolecules , but it does not necessarily rule out that large enough molecules or agglomerates to facilitate freezing may have been produced during the conducted experiments, even though it did not seem to be the case in previous comparable studies . Based on the current study, it is not possible to conclude which heterogeneous freezing mechanism(s) may be dominating.

Figure  shows our results together with selected literature data. The ice saturation ratios we have observed for the ice nucleation onset temperatures for viscous α-pinene SOA are qualitatively comparable with ice nucleation data from other SOA or SOA proxies, but the lack of data points at lower temperatures makes quantitative comparison challenging. There is a notable difference between our results and those reported by and , who found α-pinene SOA to be a poor INP, nucleating ice homogeneously. However, also found that pre-cooling of the particles made them slightly better INPs, and interpret it as a result of a transition to a more viscous state, although the phase state of the particles was not studied experimentally.

When comparing the frozen fractions of this study to earlier studies with SOA proxies, such as the substances studied by Wilson et al. (2012) and Murray et al. (2010), it is worth noting that much higher frozen fractions are achieved in this study. Most likely the difference lies in the experimental methods used: in Wilson et al. (2012) and Murray et al. (2010), the freezing was studied in an expansion chamber, not a continuous flow diffusion chamber such as SPIN. Thus, we would also not necessarily expect similar frozen fractions

When compared to other INPs, such as different mineral dusts, viscous α-pinene SOA requires higher ice saturation ratios for 10 % activated fraction than e.g. kaolinite and illite at -35 and -40 ∘C . The maximum ice fractions, on the other hand, are of the same order of magnitude. The viscous SOA particles seem to be more efficient INPs than volcanic ash at -35 and -40 ∘C measured by , with similar or lower ice saturation ratios needed for 10 % activation and higher maximum ice fractions.

Viscous pinene SOA particles have already been observed in the lower troposphere in field measurements in the boreal forest . The global aerosol model GLOMAP-mode (GLObal Model of Aerosol Processes) was used to investigate to what extent viscous biogenic monoterpene SOA is likely to be present in the atmosphere. The model version used is identical to that in . GLOMAP is an extension to the TOMCAT chemical transport model . It includes representations of particle formation, growth via coagulation, condensation and cloud processing, wet and dry deposition and in/below cloud scavenging. The horizontal resolution is 2.8∘×2.8∘ and there are 31 vertical sigma-pressure levels extending from ground level to 10 hPa. Formation of secondary particles in the model is based on CLOUD measurements of ternary H2SO4-organic-H2O nucleation detailed in and on a parameterization of binary H2SO4-H2O nucleation . Particles grow by irreversible condensation of monoterpene oxidation products and sulphuric acid. Monoterpene emissions in the model are taken from the database. The monoterpenes are oxidized with OH, O3 and NO3 assuming the reaction rates of α-pinene. A fixed 13 % of the oxidation products, referred to as SORG, condenses irreversibly onto aerosol particles at the kinetic limit.

Figure  shows the mean annual average concentrations of SORG in parts per trillion with respect to mass (pptm). SORG represents oxidized monoterpenes and can in this respect be considered a proxy of monoterpene SOA particles. The hatched areas mark the zones in the atmosphere where the SOA particles are likely to exist in a highly viscous or even glassy phase state according to the generic estimate of SOA glass transition temperature as a function of relative humidity given by . This parameterization agrees with the viscosity transitions of the investigated α-pinene SOA particles that were measured in the CLOUD chamber simultaneously with the ice nucleation experiments .

The modelled monoterpene SOA proxy concentrations are highest over land and especially in the tropics, but near the equator the conditions for particles containing monoterpene SOA to be highly viscous require higher altitudes (∼7 km) and colder temperatures (see Fig.  right panel). On the other hand, strong convective updrafts may play a role carrying the SOA particles even into the tropical tropopause layer (TTL) in short enough timescales for the particles to remain highly viscous , or SOA may form in convective outflow regions, after the transport of precursor gases from lower altitudes in the troposphere. Boreal forests are a significant source of monoterpenes in spring and summertime of the Northern Hemisphere, and model calculations displayed in Fig.  indicate that sufficient concentrations of viscous monoterpene SOA could exist in the upper troposphere in this region, thus being a potential source of INPs for cirrus cloud formation. This is relevant assuming that the freezing mechanism is deposition nucleation of highly viscous SOA or immersion freezing of partly deliquesced SOA with a highly viscous core. It should be noted that also particles with other types of SOA than α-pinene SOA may facilitate ice nucleation in cirrus clouds, or persist in a viscous state at higher temperatures or humidities . The model results presented here nevertheless indicate qualitatively that significant concentrations of SOA-forming vapours are likely to exist in parts of the troposphere where the particles they form would be in a sufficiently viscous state to act as INPs depositionally or via immersion freezing. On the other hand, if the freezing is initiated by organic INM suspended in droplets, high viscosity might not be a prerequisite for biogenic SOA to act as an INP.

So far, SOA has not been considered in climate models involving ice nucleation, and it is still very challenging to quantify the actual effect, but given the potentially high ice nucleation efficiency (up to 20 % frozen fractions), α-pinene and other monoterpene SOA could contribute significantly to the global INP budget. It is likely that a significant portion of biogenic SOA in the atmosphere forms mixtures with sulphates or primary particles such as mineral dusts, which will most likely also affect the ice nucleation efficiency considerably. It has already been shown that glassy aerosol containing a mixture of carboxylic acids and ammonium sulphate nucleates ice .