Physical and Chemical Properties of Cloud Droplet Residuals and Aerosol Particles During the Arctic Ocean 2018 Expedition

Abstract Detailed knowledge of the physical and chemical properties and sources of particles that form clouds is especially important in pristine areas like the Arctic, where particle concentrations are often low and observations are sparse. Here, we present in situ cloud and aerosol measurements from the central Arctic Ocean in August–September 2018 combined with air parcel source analysis. We provide direct experimental evidence that Aitken mode particles (particles with diameters ≲70 nm) significantly contribute to cloud condensation nuclei (CCN) or cloud droplet residuals, especially after the freeze‐up of the sea ice in the transition toward fall. These Aitken mode particles were associated with air that spent more time over the pack ice, while size distributions dominated by accumulation mode particles (particles with diameters ≳70 nm) showed a stronger contribution of oceanic air and slightly different source regions. This was accompanied by changes in the average chemical composition of the accumulation mode aerosol with an increased relative contribution of organic material toward fall. Addition of aerosol mass due to aqueous‐phase chemistry during in‐cloud processing was probably small over the pack ice given the fact that we observed very similar particle size distributions in both the whole‐air and cloud droplet residual data. These aerosol–cloud interaction observations provide valuable insight into the origin and physical and chemical properties of CCN over the pristine central Arctic Ocean.


GCVI sampling efficiency
The GCVI does not sample the cloud particle distributions with an efficiency of unity (Shingler et al., 2012;Karlsson et al., 2021). In order to retrieve concentrations of cloud residuals at representative ambient concentrations, we therefore need to determine the effective sampling efficiency. With our set-up, this can be achieved by comparing the cloud residual number concentrations to: a) the ambient cloud particle concentration, determined by integrating the FSSP cloud particle size distribution, b) the difference between the particle concentrations measured by the whole-air inlet and the interstitial inlet, and/or c) the whole-air particle size distribution for warm clouds, assuming larger particles activate first (i.e. comparing the accumulation mode concentration of cloud residuals to that of the whole air particles). a) Cloud particle concentration comparison: Cloud particle number size distributions were measured by a forward scattering spectrometer probe (FSSP; Particle Metrics Inc., USA, Model FSSP-100), measuring particles from 0.5 to 47 um diameter. Unfortunately, we were not able to use the measured cloud particle size distributions from the FSSP to infer the GCVI sampling efficiency. Figure S6b shows a density scatter plot of the liquid water content (LWC) as measured by a PVM versus the LWC calculated from the FSSP size distributions by assuming spherical droplets with the density of water. The calculated LWC is only about 30 % of the measured LWC, which suggests that the cloud particle concentrations measured by the FSSP were too low. Figure S6c shows a density scatter plot of the visibility measured by the sensor next to the GCVI versus April 7, 2022, 11:35am : X -5 visibility calculated from the FSSP size distributions assuming spherical particles and using the Koschmeider formula (Seinfeld & Pandis, 2016) and the PyMieScatt Python package (Sumlin et al., 2018). The calculated visibility is, on average, several times higher than the measured one, which, again, suggests that the FSSP was undercounting cloud particles. The FSSP laser failed on Sep 6, so it is possible that there were technical issues leading up to the failure. b) Total minus interstitial concentration comparison: The total minus interstitial particle concentration should, in theory, be on the same order of magnitude as the total cloud particle concentration, since we are subtracting the non-activated (interstitial) particles from the total particles which include activated particles (i.e. cloud particles). Therefore, comparing this difference to the cloud residual concentrations could give an indication of the transmission efficiency of the GCVI inlet. Figure S7a shows a density scatter plot of the difference between total and interstitial particle concentrations measured by the CPCs versus the concentration of cloud residuals.
The orthogonal distance linear regression resulted in a slope of about 6 (i.e. a sampling efficiency of about 17 %); however, the correlation between the two concentrations is quite weak (ρ = 0.3) and there are data points around the 1:1 line which would indicate that the cloud residual number concentration is too high, or that the total minus interstitial concentration is too low or does not accurately represent the cloud particle concentration.
CPC intercomparisons performed weekly during the expedition showed that the interstitial CPC on average measured 10% higher concentrations than the whole-air CPC (within standard measurement uncertainty: Wiedensohler et al., 2012), which can be part of the explanation why the total minus interstitial concentration sometimes appears too low. X -6 : c) Accumulation mode concentration comparison: The third method compares accumulation mode particle number concentrations of the cloud residuals to that of the whole air particles. This comparison can only be done when the clouds are mostly liquid, since it is based on the assumption of liquid droplet activation in which accumulation mode particles activate first. The comparison was made for clouds at temperatures above -2 • C for a range of different boundaries, D cut , between the Aitken and accumulation mode.
The number size distributions were integrated between D cut and 921 nm (i.e. the last size bin) to get the accumulation mode concentrations for the comparison. The temperature boundary was not put at 0 • C because that would only have included the cloud event during the first MIZ station which might not be representative for all the data. Ice nucleation was not observed above -5 • C during AO18 or other Arctic studies (Porter et al., 2021, and references therein), so the boundary of -2 • C should be sufficient to eliminate the presence of the vast majority of ice particles.
To determine the ideal D cut , the slope was chosen for the case with the highest coefficient of determination of R 2 = 0.92 ( Fig. S7b and c), which resulted in a sampling efficiency of about 6 % (slope of 16.93) at a correlation coefficient of ρ = 0.56. This is likely the most trustworthy of the estimated GCVI sampling efficiencies, due to the weak correlation in the total minus interstitial method and the fact that the FSSP cloud particle size distribution is somewhat questionable and could not be used. Karlsson et al. (2021) evaluated these approaches for a similar GCVI, and the accumulation mode comparison in their study agreed very well with the sampling efficiency as determined by using the ambient cloud particle size distributions. As such, we have corrected all cloud residual size distributions by a factor of 17 (rounded value of 16.93). :

X -7
The factor 17 is significantly higher than the average factor of around 2.2 that was observed in Karlsson et al. (2021). Part of this difference can be explained by the fact that the GCVI transmission efficiency is size dependent (Shingler et al., 2012) and the fact that the cloud droplet size distributions differ between AO18 and Zeppelin Observatory.
Because the FSSP data were questionable and the concentrations appeared too low, we did not use them to estimate the overall transmission efficiency. If we nevertheless assume that the shape of the cloud droplet size distribution measured by the FSSP is reasonably accurate, we can compare the theoretical fraction of sampled droplets between AO18 and Zeppelin Observatory based on the different cut sizes of the GCVI inlets used and on the extrapolated size-dependent transmission efficiency from Shingler et al. (2012) used in Karlsson et al. (2021). Figure S8 below shows that we theoretically could sample about 14% of the cloud droplets during AO18, as opposed to 46% at Zeppelin Observatory (from Karlsson et al., 2021). This is still around a factor of 2 off from the 6% obtained by the accumulation mode comparison and we cannot explain the last part of the difference except to emphasize that there are many uncertainties involved in the comparison presented here.
We are aware that using a constant correction factor is not ideal, and the concentrations obtained this way should be considered approximate. For looking at average cloud residual size distributions, a constant factor should be sufficient, but for individual size distributions the concentrations could be either under-or overestimated. However, the correction factor 17 leads to cloud residual number concentrations that are in a similar range as CCN concentrations reported from previous expeditions (cf. Bigg & Leck, 2001;Mauritsen et al., 2011;Leck & Svensson, 2015). The limit of detection (LOD) for each compound was calculated from the mean plus one standard deviation of the signal during HEPA filter background measurements. The mean background and standard deviations for each compound at 1 min time resolution are listed in Tab. S1 below.
Detection limits for the relevant averaging periods were calculated according to Eq. 1.
The tables that follow show the measured concentrations for the AMS data from the figures in the main manuscript. Values where the signal is weaker than the LOD are marked in red. Figure S12 shows a modified version of Fig. 5 in the main manuscript, in which data points with values that lie below the LOD have been marked in gray (all LODs were calculated according to Eq. 1 based on the exact number of data points included in the average). Note that the top row in Fig. S12 shows that the correlations discussed in the main manuscript still remain (although weaker for sulfate and stronger for organics) even if only above-LOD data are considered.
April 7, 2022, 11:35am : X -9 4. Additional information on data treatment 4.1. Loss correction All DMPS and SMPS particle number size distribution data were corrected for diffusion, impaction and sedimentation losses using the Particle Loss Calculator (von der Weiden et al., 2009), assuming spherical particles with a density of 1.5 g cm −3 which accounts for the presence of organic substances with low density (e.g. 1.3 g cm −3 , Siegel et al., 2021) and inorganic salts such as ammonium sulfate (1.77 g cm −3 ) or sea salt (e.g. 2.0 g cm −3 , Zieger et al., 2017). Overall, the SMPS and both DMPS systems agreed well in terms of the modal diameters of the aerosol size distributions, but the SMPS often showed higher concentrations than both DMPSs in both aerosol size modes (but particularly the Aitken mode, see Fig. S14) which could be due to the different inlet lines and/or remaining differences in transmission efficiencies (e.g. within the instruments, since only line losses were accounted for). This limits us to studying the total minus interstitial size distributions as averages over long periods, when the SMPS can be compared to itself since it was alternating between inlets every hour (see Fig. S13).

Pollution flag
The lab containers and inlets were set up such that they were upstream of the ship exhaust as long as Oden faced into the wind. The bow of the ship was kept facing into the wind as much as possible to avoid pollution from the ship stack contaminating the measurements. In addition, activities on the front deck were restricted and all smaller exhausts (e.g. ventilation from containers) were piped to the back of the ship.
A pollution flag was developed based on 1 s resolution number concentration data from the MCPC behind the whole-air inlet. When the MCPC concentration range in a 30 s X -10 : period exceeded the mean absolute concentration over a 6 h period centered on the same point, and the range exceeded 50 cm −3 , then the data point was flagged as contaminated.
The minimum requirement on the range was added to limit the flagging of uncontaminated data points when the total number concentrations were low. In addition to this, all points where the concentration exceeded 10 4 cm −3 were flagged (this does not eliminate new particle formation events as concentrations above the MCPC cut size remained below this threshold during the events defined in Baccarini et al. (2020)), as well as any periods noted in the logbook as potentially influenced by pollution. Finally, the number concentration time series was visually inspected to flag any remaining outliers and suspicious data points.
Lower-resolution data were cleaned based on the 1 s flag, such that if suspected contamination occurred at any point during the measurement interval, it was flagged as contaminated.

Definition of cloudy periods
As previously stated, the GCVI was run in manual mode during the AO18 expedition, so whether the air we sampled qualified as "cloudy" or not needed to be decided in postprocessing. To do this, we used a combination of the visibility data recorded by the GCVI and the LWC measured by the PVM. To achieve an optimal data coverage, we chose a maximum visibility of 2 km in combination with an LWC of more than 0.01 g m −3 to be considered as cloudy conditions. Events where such conditions persisted for at least 27 min (i.e. a minimum of 3 DMPS scans) were classified as cloud events and are included in the data presented in this paper. All in all, there were 25 cloud events with a total of 327 measured cloud residual size distributions, or 48.8 hours of in-cloud measurements. Note that this does not correspond to the total cloud occurrence during the expedition, since : X -11 the GCVI was not continuously sampling. During the included cloud events, the average LWC ranged from approximately 0.01 to 0.2 g m −3 , while the effective cloud particle radius (measured by the PVM) ranged between 7.0 and 9.2 µm (25 th and 75 th percentiles). This is in good agreement with the FSSP data which had an effective radius between 7.7 and 12.2 (main droplet number mode around 8 µm diameter). The CVI cut-off only allows us to sample a subset of the cloud particle distribution and we can only assume that there is no partitioning of physical and chemical cloud residual properties between the smaller and larger cloud particles. This assumption, however, would need more sophisticated experimental work to be proved or disproved. It is also important to note that cloud residuals do not necessarily directly correspond to CCN or INP -the particles can be modified by other atmospheric processes like scavenging (e.g. Baumgardner et al., 2008) or secondary ice processes (e.g. Field et al., 2016).

Cloud parcel modelling
A pseudo-adiabatic cloud parcel model (G.-J. Roelofs & Jongen, 2004) further developed to allow for adiabatic ascent of the air parcel (Partridge et al., 2011(Partridge et al., , 2012, and a description of surface-active organics (Lowe et al., 2019) was used to estimate the smallest activation (dry) diameters and fraction of aerosol particles that activate into cloud droplets using different chemical compositions and updraft velocities. Aerosol activation and condensation/evaporation of water are calculated according to the Köhler equation and parameterized according to Hänel (1987); G. Roelofs (1992). For the simulations performed, the model is initialised with a relative humidity (90%), temperature (270 K), and pressure (980 hPa). The initial dry aerosol size distribution is described using the lognormal distribution fit parameters for the aerosol number size distribution modes present (Table 1 whole air in 32-79 nm size range, see main manuscript), the aerosol chemical composition and updraft velocity. Particles with wet radii larger than the critical radii as predicted by Köhler theory were counted as cloud droplets. In the simulations shown here, both Aitken and Accumulation mode had the same composition. We performed three different modelling scenarios under a range of updraft velocities which spans the approx. range of the observed distribution in Arctic stratus (Shupe et al., 2013) (from 0.1 ms −1 to 1 ms −1 ): • Assuming the aerosol particles comprised 70% organics (Org) and 30% ammonium sulfate (AS). The organics were assumed to be soluble and the particles well mixed. These simulations are therefore referred to as the bulk Köhler, BK (Org+AS) simulations.
April 7, 2022, 11:35am : X -13 • Assuming all the particles comprised sulfuric acid (SA) to probe the impact of extremely hygroscopic Aitken mode on the minimum activation diameter size. These simulations are referred to as the BK (SA) simulations.
• Assuming the same composition as for the BK (Org + AS), but only 50 % of the organics were treated as soluble. The insoluble organics were instead treated as surface active forming a minimum 0.2 nm thick layer onto the particle surface. This organic surface layer had a surface tension set to 40 mNm −1 . For more information on how the surface-active organics are described we direct the readers to Lowe et al. (2019). These simulations are referred to as the compressed film, CF (Org+AS) simulations. Figure S15 shows the vertical evolution of the parcel model size bins for the BK (SA) case under 0.5 ms −1 updraft. At maximum supersaturation the activated size bins have clearly diverged from the unactivated bins.

Further figures and tables
• Table S5 presents the list of instruments, measured parameters and their temporal resolution.
• Figure S17 shows an overview of cloud residual size distributions and corresponding total particle size distributions.
• Figure S18 shows an overview of the cloud residual number mean diameter, the cloud residual Aitken mode fraction, the the cloud residual number concentration and the whole-air particle number concentration.
• Figure S19 shows the individual fits from Fig. 4 and Table 1 in the main manuscript.      April 7, 2022, 11:35am : X -29 * Note that cloud residuals, i.e. particles measured downstream of the GCVI inlet, may not always directly correspond to activated particles. Figure S13. Diagram of the instrumental set-up. Top half: On the lab container roof. Three inlets (whole-air, GCVI, interstitial) and two cloud probes (forward scattering spectrometer probe, FSSP, and particle volume meter, PVM). Lower half: Inside the lab container(s). Instruments relevant for this study are listed, and the diagram shows which inlet they sampled from. Two valve switches were installed so that some instruments could alternate between the whole-air and GCVI inlets or between the whole-air and interstitial inlets. April 7, 2022, 11:35am : X -31 Figure S15. The vertical evolution of the wet particle size bins within the parcel model.
The divergence in growth of the activated solution droplets that occurs at maximum supersaturation shows the separation between size bins that remain unactivated, and the growth of the activated cloud droplets particles by condensation during continued ascent.
The figure visualizes every 10 th size bin (40 bins out of 400) and represents the BK (SA) case with 0.5 m/s updraft.
April 7, 2022, 11:35am X -32 : Figure S16. Cloud parcel simulation of the cloud case with high Aitken-mode particle contribution. a Smallest activation diameter vs. updraft velocity, b activated fraction vs. updraft velocity and c size distribution (see Table 1 in main manuscript) with indicated activated fraction at updraft velocity of 0.5 m/s. See text above for details on chemical composition and model assumptions.
April 7, 2022, 11:35am : X -33  Figure S17. Overview of cloud residual size distributions and corresponding total particle size distributions. Total particle (orange) and cloud droplet residual Cloud residual number concentration cm 3 c 0 5 A u g 1 0 A u g 1 8 A u g 2 2 A u g 2 4 A u g 2 6 A u g 3 1 A u g 0 3 S e p 0 4 S e p 0 5 S e p 0 6 S e p 0 8 S e p 1 0 S e p 1 1 S e p 1 2 S e p 1 6 S e p 1 7 S e p 1 9 S e p 10 1 10 2 Whole-air number concentration cm 3 d Figure S18. Variability of a cloud droplet residual number mean diameter (NMD), b cloud droplet residual Aitken mode fraction, c cloud droplet residual number concentration, and d whole-air particle number concentration during the different cloud events.
Cloud events on the same day have been grouped together. When the group includes five or more data points, the data are shown as a box plot where the whiskers extend to the minimum and maximum values. When there are fewer than five data points, the individual points are shown with blue diamond markers. In the background of each panel, the geographical periods from Fig. 3 in the main paper are shaded in: marginal ice zone stations (dark blue), icebreaking periods (medium blue), and the ice drift station (light blue).  Figure S19. Lognormal fits of the mean cloud residual (purple) and whole-air (gray) particle size distributions for each cloud residual number mean diameter (NMD) bin from April 7, 2022, 11:35am