The aerosol chamber is a 0.729 m³ aluminium cube with an inlet and multiple outlet ports, allowing different instruments and sampling inlet systems to be fitted. Before any experiments took place, the chamber was flushed with filtered dry air overnight to ensure a low background aerosol concentration. After flushing, the sample material was aerosolised into the chamber as described below. Aerosolised material entered the chamber through an inlet on the side of the chamber. A small internal mixing fan positioned directly below this inlet ensured that the aerosol was well distributed throughout the chamber. To minimise sampling biases, outlet ports for additional instruments were within 7 cm of the centre of the chamber base. During operation, the aerosol chamber was maintained at ambient pressure. A schematic diagram of the aerosol sampling system can be seen in Fig. .

Due to differing physical properties, each material required an appropriate dispersion method. Dry, particulate samples were dispersed using the dry-dispersion approach. For this method, the sample material was loaded into a 15 mL centrifuge tube, which served as the dust aerosoliser. The aerosoliser was mounted at the top of the dust tower (Grimm Aerosol Technik GmbH, Germany), a cylindrical chamber 97 cm in height, 26 cm diameter, used to mix and dilute the aerosol before entering the main chamber. A HEPA-filtered air line pressurised to 2 bar was connected to a valve adjacent to the aerosoliser. When triggered, this valve briefly opened, directing a pulse of pressurised air across the sample surface. This rapid airflow caused the particles to become aerosolised before entering the dust tower. Within the tower, the aerosol was diluted with filtered air introduced at the base, regulated by a Sierra SmartTrak^® 100 (Sierra Instruments, USA) mass flow controller (MFC) at 30 L min⁻¹ or, if PINE was active, 34 L min⁻¹. The dust tower operated at a slightly elevated internal pressure, ensuring a continuous and stable flow of aerosol from the dust tower to the main sampling chamber.

For liquid sample suspensions and filtrates, a medical nebuliser (Omron MicroAir U22, Japan) was used. The nebuliser had a mass median aerodynamic diameter of 4.2 µm. The nebuliser was placed inside a clean, sealable polyethylene bag with a filtered airflow. The airflow was set to 34 L min⁻¹ during experiments with PINE and 30 L min⁻¹ without PINE. To prevent the bag from over-pressurising and exploding, the filtered air was split, with only half passing through the bag and the other half through a bypass line. Before entering the chamber, the two air lines reconnected. When the nebulised sample entered the aerosol chamber, the water evaporated due to sub-saturated conditions, leaving dry aerosol particles.

For all experiments in this study, particle size distributions were measured using a TSI Aerodynamic Particle Sizer (APS) 3321 spectrometer and a TSI Scanning Mobility Particle Sizer (SMPS) 3938 spectrometer (TSI Inc., Aachen, Germany). The APS sampled 5 L min⁻¹ from the chamber and quantified particles from 0.523 to 19.81 µm in aerodynamic diameter. The SMPS sampled at 3.3 L min⁻¹ and measured particle concentrations between 14.6 and 685.4 nm in electrical mobility diameter. The size distributions from both instruments were merged to obtain a complete aerosol size spectrum from 20 nm to 14 µm, which was then used to calculate the total aerosol surface concentration for subsequent analysis. The APS and SMPS were merged following the procedure described in , and more details can be found in Sect. S4.

For INP filter measurements, we used two Mesa Laboratories Inc. (Butler, NJ, USA) BGI PQ100 filter samplers. A 50 cm aluminium sample tube with an internal diameter of 2.3 cm was connected to the PQ100 filter holders and inserted directly into the chamber. Each sample tube was placed 25 cm away from the chamber centre (50 cm apart). Then, 47 mm diameter filters were inserted into the filter holder, which was 6 cm below the end of the sample tube. In parallel samplers, we used 0.4 µm pore diameter polycarbonate track-etched membrane filters (Whatman^® Nuclepore 10417112, Cytiva Life Sciences, USA) and 1.2 µm equivalent pore diameter polytetrafluoroethylene (PTFE) membranes (Sartorius^® 11803–47——N, Germany). Despite having different equivalent pore sizes, both filter types are expected to have high aerosol collection efficiencies (> 90 %; 0.01–0.4 µm), with reported efficiencies exceeding 95 % for particles larger than 0.15 µm . Collection efficiency depends on filter characteristics and sampling conditions, but the variability due to collection efficiency is expected to be smaller than potential differences between the wash-off and drop-on techniques. To assess potential sampling-location bias, both samplers were operated simultaneously using the same filter type while sampling chamber aerosol. The resulting INP activities agree within 1 °C, indicating no systematic difference between sampling positions (Sect. S2 in the Supplement). The filter type assigned to each sampler was alternated throughout the campaign to minimise any residual position-related bias. Each sampler was controlled by a Sierra SmartTrak^® 100 mass flow controller (MFC) connected to a vacuum pump (S55JXTJN8535, Gast, USA). Both MFCs were set to 10 L min⁻¹. The vacuum pump pulled air from the aerosol chamber through the filter membranes, which collected any particles in the sampled air.

After sampling, the filters were immediately analysed using the Microlitre Nucleation by Immersed Particle Instrument (µL-NIPI). This technique is an offline droplet freezing assay to quantify the immersion mode freezing INP spectra . The µL-NIPI methodology depends on the filter material. The polycarbonate filters were processed using the wash-off µL-NIPI method , while the PTFE filters were processed using the drop-on NIPI method . The different techniques are illustrated in Fig. . In all experiments, Agar Scientific rectangular glass cover slips (48 mm × 64 mm) were washed in ultrapure water (18.2 MΩ water filtered by Arium Pro UV, Sartorius, Germany) and isopropanol before being immersed in ClearVue^® rain repellent (Turtle Wax, USA) for 30 min. The slide was then rinsed again with water and dried using zero-grade nitrogen. The ClearVue created a hydrophobic coating on the glass slide to increase the contact angle between the glass slide and the water droplets.

During the wash-off technique, after the polycarbonate filter had been removed from the sampler filter holder, it was placed into a 50 mL centrifuge tube with 5 mL of ultrapure water. The centrifuge tube was mixed on a rotary mixer at 30 rpm for 30 min. Following mixing, approximately 140 droplets of 2 µL volume of the sample suspension were pipetted (Picus^® Electronic Pipette, 0.2–10 µL, Sartorius, Germany) directly onto the treated glass slide. Prior to each set of repeated experiments, the suspension was remixed on a rotary mixer for 10 min. The pipette was capable of dispensing five droplets before refilling was required. To minimise settling of larger particles, the suspension was manually agitated before each refill. During the drop-on technique, the PTFE filter containing the sampled particles was placed directly onto a treated glass slide (as described above), and approximately 100 × 2 µL droplets of ultrapure water were pipetted directly onto the filter surface.

After pipetting, the treated glass slides with droplets were placed on the µL-NIPI cold stage. The cold stage was programmed to cool down at 1 °C min⁻¹. To prevent frost formation on the cold stage during analysis, we used a low flow (0.4 L min⁻¹) of zero-grade nitrogen over the droplets during the experiment. A camera placed over the cold stage recorded a video that detected droplet freezing. A Python software was used to automatically detect nucleation events and determine the temperatures associated with each frozen droplet .

Throughout the study, filter handling blanks were taken before every new aerosol sample material was introduced. These handling blanks sampled the same volume as the samples, from a clean aerosol chamber with filtered inlet air. These were subtracted from the sample data to remove any background contamination . All the data from the study (handling blanks and samples) were binned into 0.5 °C temperature bins. The differential freezing nucleus spectrum, k(T) (cm⁻³ °C⁻¹), was then calculated for each bin using Eq. () : 1kassay(T)=-1Vdrop×ΔT×ln⁡1-ΔNN(T), where Vdrop is the volume of the pipetted droplet (2 µL), ΔT is the temperature bin width (0.5 °C), ΔN is the number of freezing events in the temperature bin (between T-0.25 °C and T+0.25 °C), and N(T) is number of unfrozen droplets in the temperature bin (freezing colder than T-0.5 °C).

Background k(T) curves were created for the PTFE and the polycarbonate filter blanks (Sect. S1). The average of all the blanks was used as the background k(T), and the standard deviation of the background runs represented the uncertainty in the background measurement. For sample data, if multiple runs were carried out on one wash-off suspension, the k(T) value was the mean of the separate runs, and the uncertainty was the standard deviation. For experiments with no repeats, the error was determined by the Poisson counting statistics and propagated to k(T) . To determine the k(T) value, the background value, kbg(T), was subtracted from the sample value, ksample(T) (Eq. ). If the sample k(T) value happened to fall within the background, this resulted in a sample k(T) value of 0 cm⁻³ °C⁻¹ at that given temperature. The background subtraction method used during this study is further described in and . 2k(T)=ksample(T)-kbg(T)

We then calculated the cumulative freezing nucleus concentration, K(T), by summing the background-subtracted k(T) values for temperatures warmer than T. The cumulative freezing nucleus concentration can be calculated using Eq. (). 3K(T)=∑0Tk(T)×ΔT K(T) does not account for the volume of sampled air or the dilution by the amount of water used. To directly compare samples of different sampling volumes, the INP number concentration (NINP, L⁻¹) was used. The INP concentration gives the number of INPs in a given volume of sampled air. We use Eq. () to calculate the INP concentration for the wash-off technique and Eq. () for the drop-on technique: 4NINP(T)=K(T)×VsuspVair5NINP(T)=K(T)×A×VdropVair×α, where Vsusp is the volume of the wash-off suspension, Vair is the volume of sampled air, A is the area of the filter exposed to the airstream in the sampler, Vdrop is the volume of a pipetted droplet, and α is the footprint area of the drop in contact with the filter. For this study, Vsusp = 5 mL and Vdrop = 2 µL. Additionally, A = 11 cm² and α = 1.357 mm² .

To compare the ice-nucleating activity derived from different experiments with different aerosol loadings and particle size distributions, we use the active site density (ns(T), cm⁻²). The active site density is the number of ice-nucleating active sites per unit of surface area of ice-nucleating material . The active site density is defined in Eq. (): 6ns(T)=NINP(T)s, where s is the total surface area of the particles per unit volume of sampled air and has units of µm2cm-3. The total surface area was calculated by finding the area under the curve from the dSdlog⁡Dp vs. particle diameter (Dp) curve, mathematically shown in Eq. (): 7s=∫dSdlog⁡Dpdlog⁡Dp.

The uncertainty associated with the background-subtracted k(T) values was propagated through the calculation of K(T) and subsequently through to NINP(T) for both the drop-on and wash-off techniques. The uncertainty in ns(T) was calculated by combining the NINP(T) uncertainty with that of the total particle surface area, s, using standard error propagation in quadrature. While NINP(T) shows how numerous INPs are in a sample, ns(T) shows how active the INPs are, making ns(T) useful for this study, where the aerosol loading and aerosol type vary. For each individual experiment, ns(T), plotted with its uncertainty, can be found in Sect. S3.

In some experiments, the Portable Ice Nucleation Experiment version 1B (PINE-1B, Bilfinger GmbH, Germany) was connected to the aerosol chamber. PINE is an expansion chamber with pressure-drop-induced cooling to activate INPs via immersion and deposition mode freezing . The PINE-1B is described in . All PINE experiments in this study used the PINE-1B (PINE hereafter).

The PINE operated in three modes: flush, expansion, and refill. During the flush mode, sample air was drawn through the system at 4 L min⁻¹ for 5 min. The air entered through the inlet, was dried by two counterflow Nafion dryers in parallel, and then flowed into the top of the PINE chamber. It exited through the bottom of the chamber and passed through the WELAS 2500 (Palas GmbH, Germany) optical particle counter (OPC) connected to a Promo 2000 control unit (Palas GmbH). After 5 min of flushing, the instrument switched to expansion mode. The inlet valve at the top of the chamber closed, and air was pumped out at 4 L min⁻¹. As the chamber pressure decreased, the relative vapour pressure over ice and water increased, while the temperature decreased quasi-adiabatically, triggering the formation of a mixed-phase cloud through activation of INPs and CCN. The resulting cloud droplets and ice crystals passed through the WELAS OPC, which measured particle diameter and number. During the expansion phase, the incoming sample air was bypassed around the chamber directly to the outlet to maintain the upstream flow rates. Once the chamber pressure reached 800 hPa, the PINE switched to refill mode. The refill valve above the chamber opened, and the chamber outlet was closed. This allowed sample air to slowly refill (approximately 1 L min⁻¹) and repressurise the chamber. When the chamber pressure reached 950 mbar, the PINE returned to the flush mode, allowing the chamber to equilibrate with ambient pressure. The entire cycle lasted approximately 8 min and was then repeated throughout the experimental period.

The phase of sampled particles was determined by whether the detected particle was larger or smaller than the ice threshold, which was manually determined based on an individual expansion's particle size distribution, which was typically bimodal. The ice threshold was defined as the size representing the cut-off between the smaller-diameter water droplet mode and the larger-diameter ice particle mode. Each expansion was then divided into five equal-sized pressure bins, and any data below 850 hPa were removed (Sect. S5). The total number of ice particles detected in each pressure bin was Nmeas. We can then determine the INP concentration (NINP(T)) using Eq. (). As the PINE was an expansion chamber and cools the air parcel via pressure-controlled adiabatic cooling, the adiabatic temperature was used for measurements . The initial, pre-expansion, temperature was measured by thermocouples in the chamber. The temperature during the expansion was then calculated assuming the expansion was adiabatic, using the dry adiabatic lapse rate until the parcel was saturated before switching to the saturated adiabatic lapse rate, as used in . The adiabatic assumption was supported by the match between measured droplet onset relative humidity and water saturation reported by . 8NINP(T)=Nmeas×τtbin×Vd, where Nmeas is the ice crystal number concentration detected by the OPC in each pressure bin, τ is the mean transit time of particles through the detector, tbin is the pressure bin duration, and Vd is the detection volume of the OPC. Since we use pressure bins, the duration is the duration of each pressure bin (usually between 3 and 8 s), and the volume is the volume of air passed through the OPC over the course of the pressure bin. The temperature used for each pressure bin was the coldest temperature of the detected ice particles. The WELAS used in the PINE-1B did not sample the entire flow from the chamber (only approximately 10 %; ), so to account for this, the transit time and detection volume must be included in the calculation. To calculate the active site density, we use Eq. ().

Our hypothesis is that INMs are adsorbed to the polycarbonate filter and do not come off in the washing process. To test this, we analysed punches of the washed filters for INPs using the InfraRed-Nucleation by Immersed Particle Instrument (IR-NIPI) (Fig. ) . The IR-NIPI is an immersion mode freezing instrument that uses an infrared camera to detect ice nucleation events in aqueous aliquots contained in the wells of 96-well plates, into which we placed 5 mm punches of filters. When we used the IR-NIPI, the wash-off suspension had been in the freezer for about a year. The centrifuge tubes containing the frozen wash-off suspension with the immersed polycarbonate filter were allowed to thaw at room temperature for 30 min. They were then placed on a rotary mixer (Clifton™ RM2, UK) for 30 min. Afterwards, the filter was removed from the suspension, and 15 × 5 mm diameter punches were taken using a 5 mm diameter stainless steel tissue biopsy punch (WellTech Rapid Core 5.0, Taiwan). The punches were placed into wells in half of a 96-well plate and submerged in 100 µL of pure water each. The remaining filter was put into a clean centrifuge tube with clean water and was washed vigorously by shaking and spraying with ultrapure water to remove any further particles. An additional 15 punches were placed in wells in the other half of the 96-well plate with 100 µL of water each. The punches and water were put in every other well to avoid being influenced by the release of latent heat of freezing from adjacent wells. To assess the background of the analysis method, we performed four experiments with punches of blank filters.

The assay tray was then placed onto the IR-NIPI's cold stage and cooled down at a rate of 1.3 °C min⁻¹, corresponding to the tray actually cooling at approximately 1 °C min⁻¹ . An infrared camera recorded the temperature of the water's surface in each well. When the water froze at subzero temperatures, the temperature in the well jumped to 0 °C due to the release of latent heat and then slowly cooled back down to the plate temperature. The nucleation temperature was assumed to be the coldest temperature recorded for each well before the jump to 0 °C. The freezing temperatures can then be used to calculate a fraction frozen, giving the fraction of frozen wells at a given temperature (Eq. ). 9f(T)=nice(T)ntot, where nice(T) is the number of frozen wells at a given temperature, and ntot is the total number of wells with water and filter punches. Background measurements of new, clean filters were plotted next to the data from washed filters to provide context that the samples were above or in the background. To determine whether the washed-filter-punch analysis was different from the background, we used an integrated squared difference (ISD) test . We only draw attention to this (ISD) test if the washed filter analysis was found not to be statistically significant compared to the background. There is a temperature gradient between the measured surface film temperature (measured by IR) and the base of the wells that is poorly characterised . Because the filter punch samples could not be placed and held at a consistent position within the wells, individual samples likely experienced different thermal conditions (e.g. variations in vertical position or contact with the well walls). This introduced an unknown and variable temperature uncertainty, which we estimate to be within approximately 5 °C. Hence, we do not derive active site density (ns(T)) from IR-NIPI data and instead present fraction frozen curves that provide a measure of the ice nucleation activity of material retained on the washed filters relative to clean filters.

We dry-dispersed three samples containing biogenic INPs: Alaskan Copper River dust, UK agricultural soil dust, and a lyophilised bacterial material (Snomax). Information about each material can be found in Table , and size distributions are in Fig. S4 in the Supplement.

We wet-dispersed four samples that contain aqueous filtrates of free biogenic INPs: agricultural soil, lichen, pollen, and Snomax. In addition, we wet-dispersed a sea surface microlayer sample. Information about each material can be found in Table , and size distributions are in Fig. S8.

To address our hypothesis that INMs stick to polycarbonate filters, we have studied Snomax material in both a dry-dispersed form, where coarse-mode material is present, and also in a wet-dispersed filtrate form, where particles larger than 0.2 µm have been removed, leaving behind free proteins. The results are shown in Fig. . The drop-on and wash-off techniques for the dry-dispersed samples agree to within about 3 °C (with substantial run-to-run variability), whereas for the wet-dispersed filtrate, the wash-off method produces freezing curves that are about 9 °C colder than the drop-on technique. The PINE results are consistent with the drop-on results and inconsistent with the wash-off results for the wet-dispersed filtrate experiments. Analysis of the washed polycarbonate filter punches from both the dry Snomax and Snomax filtrate reveals that highly active ice-nucleating material remains on the filter, even after a second, more vigorous wash (Fig. d). These experiments support our hypothesis that the free INpro are adsorbed to the polycarbonate filter, resulting in a low bias in the wash-off results.

There is substantial experiment-to-experiment variability in the dry Snomax results. For example, experiment 72 produced higher ns(T) values for both the wash-off and drop-on techniques (Fig. S5). This difference across experiments is likely due to the inconsistencies in our dry-dispersion technique for dry Snomax. When comparing dry Snomax size distributions (Fig. S5a and Table S1 in the Supplement), experiment 72 had a smaller peak diameter than experiments 70 and 71. We noted that dry Snomax readily accumulated electrostatic charge during handling, which may have contributed to variability in both the size distribution and ice-nucleating activity. In contrast, reproducibility was excellent for the wet-dispersed Snomax experiments, where the aerosol generation process did not appear to be affected by static charging. The use of an aerosol neutraliser prior to sampling could potentially reduce variability associated with particle charging and improve the reproducibility of the generated aerosol in future experiments. Despite the experiment-to-experiment variability in dry-dispersed Snomax, within each experiment, the wash-off ns(T) results are systematically around a factor of 10 lower than the drop-on ns(T) results (Fig. S5). This coincides with substantial ice nucleation activity on the filter from the analysis of the washed filter. These results with dry-dispersed Snomax are consistent with our hypothesis that, while larger particles can be recovered into the wash-off volume, a substantial proportion of the INpro remains adsorbed to the filter and is therefore lost from the analysis.

Our dry Snomax results have a similarly shaped ns(T) curve to that reported in the literature , with a steeply increasing ns(T) above ∼ -10 °C and a plateau at lower temperatures. This parametrisation is derived from a combination of offline droplet freezing and online dry-dispersed techniques. There are no ns values for wet-dispersed Snomax filtrate, but nm curves in the literature qualitatively show a similar trend to our data, with the nm values plateauing around -10 °C.

Like bacterial ice-nucleating material, the ice-nucleating ability of lichens is also thought to be related to ice nucleating proteins ; hence we might expect similar interactions between the INpro and polycarbonate filters as for Snomax. The results presented in Fig. a for lichen filtrate confirm this, with the wash-off technique detecting lower INP activity than the drop-on technique. PINE data are unfortunately only available for temperatures below -30 °C, making a comparison between PINE and the filter techniques difficult. We also note that there is greater uncertainty in individual wash-off experiments due to the proximity to the backgrounds, which introduces uncertainty when performing the background-subtraction procedure (see Fig. S10 for uncertainties associated with this data). Analysis of the washed polycarbonate filter reveals that the freezing temperatures are above the background runs by about 1 °C; however, by using the ISD test described in , the difference between the most active background and the washed filter punches is not statistically significant. This means that we cannot say with certainty that we detected a signal above the background on the washed filters.

We studied pollen filtrates because the ice-nucleating entities are thought to be polysaccharides and may have different interactions with polycarbonate filter surfaces compared to INpro. The pollen filtrate results show good agreement between techniques, with both the wash-off and drop-on measurements agreeing within an order of magnitude below -14 °C. The pollen filtrate spectrum is very steep, with all measured activity occurring between -13 and -17.5 °C, consistent with the previous literature . Despite the strong agreement between the two filter measurement techniques, a substantial amount of ice-nucleating material remained on the polycarbonate filter after washing (Fig. b). Although we expect a few degrees of temperature bias in the IR-NIPI due to the larger well volume, the reported data are about 3.5 °C warmer than the warmest point in either the wash-off or drop-on results. It has been suggested that the activity of birch pollen washing water can actually increase after freeze drying or numerous freeze–thaw cycles . The polycarbonate filter used for IR-NIPI analysis was stored in the freezer for about 1 year, possibly leading to the increase in activity seen on the washed filter. Polysaccharides have also been observed to adhere to surfaces, being a key cause of biofouling in membrane bioreactors used for water treatment . Different polysaccharides interact differently with surfaces, suggesting there are different mechanisms for polysaccharides to adsorb to surfaces . However, adsorption due to polysaccharides is thought to be more reversible than adsorption of proteins . Our work suggests that a significant proportion of ice-nucleating polysaccharides in the pollen are recovered into suspension; hence a low bias in the wash-off is not observed.

To evaluate the performance of the wash-off technique in marine regions, we used the sea surface microlayer (SML12 in ) as a proxy for sea spray aerosol, a known INP type . The results in Fig. a show substantial disagreement between the wash-off and drop-on techniques. The wash-off curves are consistently flatter above 22 °C, and the wash-off measurements exhibit over an order of magnitude variability, compared with only about a factor of 5 for the drop-on data. Individual wash-off experiments also show larger uncertainties, due to high counting uncertainty in the flatter regions and proximity to the background at lower temperatures (Fig. S12). Although PINE operated at lower temperatures than offline techniques, it aligns more closely with the higher INP activity measured by the drop-on method, but unfortunately, the overlap between the filter techniques and PINE is poor. The data from the washed-filter-punch analysis (Fig. b) reveal that we did not detect a signal above background in experiment 55, and the results from experiment 54 are borderline. To determine if there was a statistical difference, we used the ISD test. There is a 7 % probability that the washed filter from experiment 54 is within the background, but washing this filter again increases this probability to 26 %. The background signal for the washed filter analyses corresponds to the warmest freezing temperatures observed in most filter experiments, so if material is adsorbed to washed filters, the freezing signal would be weak relative to the background.

In three out of four wash-off experiments, the wash-off technique detected higher activity than the drop-on technique above -16 °C. It is possible that INPs in the wash-off suspension aggregated before analysis, meaning that some droplets contained large, highly active aggregates, while the rest of the droplets contained less active material, as discussed in . Sea spray aerosol, which is influenced by the sea surface microlayer , is extremely variable, and the INP concentration increases with phytoplankton blooms . Sea spray aerosol can contain a complex variety of different organic materials such as saccharides, amino acids, lipids, and proteins . In fact, show that different organic components can dominate freezing at different times. It is not possible to say which component was active in our sample, although suggested INP activity could be due to phytoplankton exudates, which suggest nucleate ice via polysaccharide- or protein-containing nanogels. Overall, the wash-off results are much less consistent than the drop-on results, perhaps due to a combination of macromolecule adsorption and aggregation in the wash-off suspension.

To test our hypothesis for INpro-containing soil dust samples, we carried out a similar approach to the one we used with the Snomax samples. We tested two dry-dispersed soil dusts and then a wet-dispersed filtrate sample derived from the agricultural soil dust sample. Both filter techniques generally reported good agreement for dry-dispersed Copper River dust and agricultural soil dust. However, in both samples, the average ns(T) from the drop-on experiments was about 1.5 times that of the wash-off average. The filter-based results are consistent with previous wash-off measurements of Copper River dust as well as PINE data , while the agricultural dust experiments agree with previous wash-off experiments reported by . Analysis of the washed filter punches in both samples (Fig. d and e) reported activity above background levels, implying that soil dust-derived ice-nucleating entities were adsorbed to the polycarbonate filter surface.

For the dry-dispersed soil samples, we observed that the experiment-to-experiment variability for the wash-off technique was greater than for the drop-on technique. For example, five out of the seven Copper River dust experiments (Fig. S6) report good agreement between the wash-off and drop-on results within the uncertainty. This indicates that the recovery of particles into suspension is variable. These experiments were all performed with the same methodology, i.e. on a rotary mixer for 30 min at 30 rpm. However, we have minimal control over how the filter folds and crumples during the washing process, and it is possible that areas of the filter are not as well washed in some experiments. The efficacy of the recovery of particles into suspension may also contribute to the low bias seen in the literature for ambient aerosol. The crumpling of the polycarbonate filter in the washing process has the potential to affect all wash-off samples, not just those containing free INMs. This crumpling could possibly explain why there is a greater variability in the wash-off results than there is in the drop-on results.

When the agricultural soil dust was suspended in water, then filtered to remove particles larger than 0.2 µm (Fig. c), the wash-off ns(T) is substantially lower than the drop-on above -21 °C. Once again, uncertainty in individual experiments was much larger for the wash-off data than the drop-on, due to proximity to the background (Fig. S13). The PINE measurements were only made below -28 °C, at lower temperatures than the filter measurements. IR-NIPI analysis reveals that there is ice-nucleating material that remains on the polycarbonate filter post-wash, but it is not statistically different from the background. By removing the larger particles, we have removed the warmer-temperature INPs, meaning that our filter results from the chamber experiments only start around the same temperature as the background in the IR-NIPI. Regardless, overall, the results are consistent with a proportion of the ice-nucleating material from soils sticking to the polycarbonate filters and not being recovered into the wash-off suspension.

The results presented above indicate that variable adsorption of different types of ice-nucleating material to polycarbonate filters is a systematic feature of the wash-off technique, and the impact on the measurement of INPs varies depending on the nature of the sampled aerosol. A consistent pattern emerged across all experiments in which INpro were likely involved. Specifically, when the INP population was dominated by free INpro (i.e. the INpro were not bound to larger cell fragments or mineral particles), such as in our filtrate experiments, the wash-off technique dramatically under-reported INP activity. When INpro were internally mixed with mineral dust or cellular fragments (i.e. INpro were bound to cell fragments or mineral particles), wash-off and drop-on INP spectra were in much better agreement. However, even in the cases where the drop-on and wash-off techniques agreed within scatter between experiments, ice-nucleating material was still observed to be left behind on the polycarbonate filters, as proven by testing filter punches in the IR-NIPI. These observations highlight that the mixing state of INpro with other aerosol particle types is critical in determining the accuracy of the wash-off technique and also that externally mixed (or free) INpro are the most susceptible to adsorption onto the filter, leading directly to undercounting in wash-off measurements.

Throughout this study, we aimed to determine whether adsorption of externally mixed (“free”) INMs onto polycarbonate filters causes the wash-off technique to underestimate INP activity. For the purpose of this study, we define “free” macromolecules as those not bound to larger particles such as mineral dusts or cellular fragments. While proteins were expected to exhibit strong affinity to the polycarbonate filter, it was not clear whether other INMs, such as polysaccharides, would behave similarly. The pollen filtrate experiments show that polysaccharide-based INMs also adhere to the polycarbonate membrane, indicating that adsorption is not limited to INpro, even though the wash-off measurement was consistent with the drop-on spectra, within experimental uncertainties. This suggests that polysaccharide adsorption occurs. However, the steep freezing spectrum of pollen means that even partial recovery of ice-nucleating polysaccharides is sufficient to produce reliable wash-off spectra.

In contrast, the lower recovery of free INpro appears to be the primary driver of the discrepancies observed in proteinaceous samples, most notably in the Snomax filtrate. Free macromolecules are present in the bulk, unfiltered samples such as the dry Snomax and soil dusts, but their contribution to ice nucleation is masked by the larger, more efficient INPs associated with cell fragments or mineral particles. As a result, even when free macromolecules adsorb to the polycarbonate membrane, their loss has a minimal effect on the wash-off INP spectra. However, once the larger particles are removed, and only the small, externally mixed macromolecules remain, their adsorption onto the filter leads to the undercounting of ice-nucleating activity in the wash-off technique. This distinction between bulk and filtered samples underscores the importance of externally mixed free INpro in determining wash-off recovery.

When a sample is collected, all particles in the sampled air deposit onto the polycarbonate filter, including large aggregates, internally mixed INMs, and small free macromolecules. Once deposited, intermolecular forces between the particle surfaces and the filter membrane promote adsorption. Particles with relatively large surface areas, such as mineral dust particles in fertile soil dusts or cell fragments in dry Snomax, evidently experience sufficient hydrodynamic drag during washing to overcome these adhesive forces, allowing them to detach and enter suspension. In contrast, small macromolecules have much lower drag forces and therefore remain strongly bound to the filter membrane, meaning they are not recoverable in the wash-off suspension (Fig. ). We suggest that macromolecular adsorption is universal across different samples, although its impact is influenced by the presence of externally mixed INMs. We conclude that adsorption is a fundamental and unavoidable feature of the wash-off technique, but the degree of bias depends on sample properties.

These laboratory findings have direct implications for the interpretation of field-based INP measurements, particularly in environments where the aerosol population is dominated by free INpro. Based on previous field observations, we suggest that the presence of free INpro is variable both spatially and temporally. For example, observed good agreement in INP concentrations between techniques in the southern Great Plains in the central USA but poor agreement in the remote eastern North Atlantic. The aerosol number concentration was about an order of magnitude higher at the southern Great Plains site compared to the North Atlantic, but the total aerosol surface area was 2 orders of magnitude higher, most likely influenced by higher abundance of soil dust in the region . Additionally, the measurements in the southern Great Plains took place in the autumnal pollen period, so this site may also have been influenced by pollen . As we have seen in our experiments, soil dust and its internally mixed macromolecules (Fig. a and b) are well recovered by the wash-off technique. Additionally, although the INMs in pollen have been observed to adsorb to the filters (Fig. b), we also saw that a significant amount of material was recovered, leading to the wash-off technique agreeing with the drop-on technique within uncertainty (Fig. a). Hence, we suggest that the aerosol in this region was composed of materials that are recovered in the wash-off technique, consistent with the good agreement between the wash-off and PINE measurements. However, the eastern North Atlantic site sampled by is dominated by submicron sea spray aerosol and long-range-transported continental material . Our sea surface microlayer experiments (Fig. a) exhibited similar temperature-dependent activity to , including sporadic nucleation events at warm temperatures and a sharp increase in activity around -22 °C. Our experiments also show that detectable ice-nucleating material was present on the washed filter punches. The discrepancy reported by is thus most likely due to the presence of macromolecules that are poorly recovered by the wash-off technique.

Our findings could also help provide context for the discrepancies reported by . Back trajectories illustrate variability in air mass histories, including time spent over both land and ocean and frequent transitions between the boundary layer and free troposphere. This likely results in correspondingly variable aerosol sources, such that some experiments sampled air masses containing more or fewer INpro. Consequently, the degree of agreement between the wash-off and drop-on techniques varies across experiments, which we suggest reflected the differences in the underlying aerosol populations. Samples with a higher fraction of free INpro would be susceptible to wash-off undercounting, while samples dominated by dust would be less affected. The results from and indicate that disagreement between the wash-off and other direct INP measurement techniques should not be interpreted solely as instrumental uncertainty, but rather as evidence for the presence of free INpro in the ambient aerosol population.

There are also implications for the modelling of global INP concentrations and their subsequent influence on clouds. Models rely on parametrisations of the ice-nucleating activity of various aerosol species, but also on measurements to test their accuracy . For example, a parametrisation commonly used in global models for the ice-nucleating activity of sea spray aerosol is based on wash-off measurements , which may under-report the ice-nucleating ability of sea spray. This can only be tested with new measurements. Similarly, many of the ambient INP measurements that were used in the global model study of were made with the wash-off technique, with unknown bias. This might imply that the INP concentrations in some locations are greater than currently suggested by models.

The implications of these findings extend directly to how INP sampling and measurement should be approached. While we present evidence of INMs sticking to the polycarbonate membrane, it is also possible that macromolecules adsorb onto other surfaces during the wash-off process, such as sample containers or pipette tips. These considerations highlight that sample processing steps can introduce additional opportunities for macromolecule loss. Similar losses may also occur in offline sampling approaches, as well as in filter collection that relies on collection devices, such as impingers, where macromolecules could adsorb onto the walls of the collection vessel. As a result, INPs suspended in water during sampling may be under-represented even before downstream processing begins. We should aim to minimise INP losses in our measurement techniques by using methods that require fewer processing steps and therefore fewer opportunities for macromolecule adsorption. More direct techniques, such as online or direct filter-processing methods, are not only less prone to the loss of ice-nucleating material but are also more representative of the behaviour of INPs in the atmosphere. Accordingly, these findings emphasise the need for careful method selection and testing when quantifying INPs, particularly when free macromolecules may contribute to the ice-nucleating population.

The limitations identified here do not negate the utility of the wash-off technique or the use of polycarbonate filters. Wash-off suspensions remain particularly valuable for treated-sample analysis, such as heat or chemical treatments to determine the biological contribution to INP activity . In addition, the smooth, non-fibrous surface of polycarbonate filters is advantageous for particle-resolved analyses using scanning electron microscopy with energy-dispersive spectroscopy (SEM-EDS), enabling the identification of particle types and mixing states not available through INP analysis alone. One potential avenue for improving recovery from the polycarbonate filters is the use of sonication to release sampled aerosol into suspension. Sonication has been employed in several INP measurement studies . However, sonication may also alter the structure of ice-nucleating proteins through denaturation or fragmentation , potentially reducing the measured INP activity. Sonication was not employed in this study, but its utility as a means of improving filter recovery while preserving ice-nucleating activity should be evaluated in future studies. In this way, the wash-off technique and polycarbonate filter sampling can provide important information regarding INP composition and sources, even in cases where adsorption limits the recovery of free INMs.

The findings of this study have implications for how past INP measurements obtained using the wash-off technique should be interpreted. While we can provide recommendations for future INP measurements, the extent to which historical datasets are affected is difficult to determine because the chemical composition and mixing state of the sampled INPs are rarely known for field-collected material. In the absence of this information, it is not feasible to apply a universal correction factor to account for the loss of externally mixed INMs. However, previous measurements will still serve as a reasonable lower bound of the INP activity in a region, particularly for samples containing mineral dusts or cellular fragments, for which our results suggest the wash-off recovery remains largely robust. Not all wash-off data will be impacted by adsorption, but to know whether the data are reliable, another, more direct measurement technique is required. For example, in the southern Great Plains data from , the agreement between the wash-off technique and PINE indicates that the wash-off results are accurate. In contrast, in the eastern North Atlantic Ocean, the wash-off method reports lower INP activity than PINE, suggesting that the technique is not appropriate for that environment. A similar logic applies to : agreement implies robust effective wash-off recovery, whereas divergence signals the presence of externally mixed macromolecules lost during processing. When drop-on and wash-off measurements agree, the wash-off data can be considered reliable, and reliable heat tests or chemical analysis can be performed on the aqueous suspension. Thus, simultaneous measurements using wash-off and another technique can provide not only validation of the wash-off technique, but also insight into the nature of the INPs present and the presence of free INMs.