Ship-based measurements of ice nuclei concentrations over the Arctic, Atlantic, Pacific and Southern Ocean

Ambient concentrations of ice-forming particles measured during ship expeditions are collected and summarized with the aim of establishing a spatial distribution and variability of ice nuclei in oceanic regions. The presented data from literature and previously unpublished data from over 23 months of ship-based measurements stretch from the Arctic to the Southern Ocean and include a circumnavigation of Antarctica. In comparison to continental observations, ship-based measurements of 5 ambient ice nuclei show one to two order of magnitude lower mean concentrations. To quantify the geographical variability in oceanic areas, the concentration range of potential ice nuclei in different climate zones is analysed by meridionally dividing the expedition tracks into: tropical, temperate and polar climate zones. We find that concentrations of ice nuclei in these meridional zones follow temperature spectra with similar slope, but vary in absolute concentration. Typically, the frequency with which specific concentrations of ice nuclei are observed at a certain temperature follows a log-normal distribution. A consequence of 10 the log-normal distribution is that the mean concentration is higher than the most frequently measured concentration. Finally, the potential contribution of ship exhaust to the measured ice nuclei concentration on board of research-vessels is analysed as function of temperature. We find a sharp onset of the influence at approximately -36 ◦C, but none at warmer temperatures that could bias ship-based measurements. 1 https://doi.org/10.5194/acp-2020-466 Preprint. Discussion started: 16 June 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
A small fraction of atmospheric aerosol particles possesses properties that induce the nucleation of ice. Herein the term ice nuclei (IN) is used to describe these, without specificity about their nature, equivalent to the proposed term ice nucleating particles (Vali et al., 2015). The concentration of IN active at a certain temperature affects the extent of ice crystal formation in clouds. Thereby, IN have an effect on precipitation formation, cloud optical properties and cloud persistence (e.g. DeMott 5 et al., 2010). Stratiform clouds are more affected by the IN concentration than convective clouds because of differences in cloud water content and dynamics (Zeng et al., 2009). Since the ratio between stratiform and convective cloud amount changes meridionally, with more stratiform clouds towards higher latitudes, the effect of IN concentration on cloud properties increases from the tropical to polar regions (Zeng et al., 2009). The most important ice nucleation mechanism for mixed-phase clouds is immersion freezing. The temperature range of immersion freezing, in which IN are first immersed in supercooled droplets 10 before triggering ice formation, is limited at the lower end by the onset temperature of homogeneous ice nucleation (-38 • C) below which droplets freeze without the necessity for catalysing IN. At the high end, exceptionally efficient, biological IN are known to nucleate ice at temperatures as high as -1.3 • C (Schnell and Vali, 1972) whereas, for example, sub-micron desert dust IN typically show high activity only below -25 • C (Boose et al., 2016). After the onset of immersion freezing, the glaciation of mixed-phase clouds can be enhanced by secondary ice formation, a cascade process where supercooled droplets freeze and 15 shatter on contact with ice crystals (Mason and Maybank, 1960) or by ejection of ice splinters during the formation of rime on graupel (Hallett and Mossop, 1974).
To describe the atmospheric IN population comprehensively, IN concentrations in the entire temperature range of immersion freezing (0 • C to -38 • C), their temporal fluctuation and geographical variation should be considered. The first summary and parametrization of the change in IN concentration as a function of temperature was given by Fletcher (1962), based on a 20 collection of ambient measurements available at the time. Even before Fletcher's parametrization, efforts had been made to quantify the concentration of atmospheric IN dependent upon air mass origin and weather conditions. For example, Findeisen and Schulz (1944) determined an IN temperature spectrum for Central Europe on the basis of 8 years of aircraft and ground based observations. The spatially most comprehensive measurements so far have been achieved by coordinated, ground-based measurements (e.g. Bigg and Stevenson, 1970), aircraft campaigns (e.g. DeMott et al., 2010) and ship-based observations in 25 the Southern Ocean between Australia and Antarctica (Bigg, 1973). As 71% of earth's surface is covered by oceans, ship-based observations offer an effective way to cover large areas. The Bigg (1973) data set, averaging three years of observations (shown in Fig. 1), has been used as a reference on which modelled IN concentrations have been validated (e.g. Vergara-Temprado et al., 2017).
We present a summary of ship-based measurements from the literature and extend this data set with unpublished data dating 30 back as far as 1996 and measurements from several recent ship expeditions. Definition of expedition acronyms and a description of expedition objectives can be found in Appendix A. The recent data-sets were obtained during expeditions in the Arctic (PS106), the Southern Ocean (ACE) and during Atlantic transects (ACE, PS95 and PS79). Previously unpublished -15 • C data include observations from the Pacific (SHIPPO, ACAPEX), Southern Ocean (Tan1502) and central Arctic Ocean (AOE-96, AOE-2001).

Sampling and measurement
Ship-based measurements of IN concentration record the abundance of potential IN in the marine boundary layer, some meters above sea level. Much of available literature data are limited to IN concentrations at -15 • C or -20 • C, because of instrumental 5 limitations and a focus on the influence of IN abundance on precipitation formation. During recent expeditions, emphasis was put on collecting data covering a broad temperature range (0 • C to -38 • C), because of a shift in focus to the cloud radiative effect and the realization that already low IN concentrations can have a large effect by triggering secondary ice formation (e.g. Sullivan et al., 2018). Additionally to the importance for cloud microphysics, observations over a broader temperature range allow the construction of IN temperature spectra, providing information on the abundance of potential IN sources from their 10 slope (e.g. Welti et al., 2018).
Two measurement techniques are used for recent ship-based measurements. First, continuous flow diffusion chambers (CFDCs), e.g., CSU CFDC measuring during CAPRICORN (McCluskey et al., 2018a) or SPIN (SPectrometer for Ice Nuclei, DMT, see Garimella et al., 2016, for technical details) measuring during PS95 and PS106. CFDCs cover the low temperature range from approximately -38 • C to -24 • C. Secondly, drop freezing techniques, using aerosol collected on filters during expeditions. Drop 15 freezing techniques typically cover the temperature range from -26 • C to 0 • C.
For the new data sets in the Arctic, Atlantic and Southern Ocean (marked as circles in Fig. 1), sampling was conducted on board the research vessels Polarstern (expeditions indicated with PS) and Academic Tryoshnikov (during ACE). For a description of the ACE and PS106 (PASCAL) expeditions we refer to Schmale et al. (2019) and Wendisch et al. (2018), respectively. On board both vessels we used the upper deck (monkey bridge), about 30 m above the ocean surface, to install sampling devices. 20 SPIN sampled from within a container through a whole-air inlet system with subsequent drying of the aerosol. The SPIN measurement strategy was to sample half hourly intervals at a constant temperature (selected between -38 • C and -24 • C) and constant relative humidity (85%, 95% or 103% with respect to water). Each sampling condition was repeated three times within 24 h. This allowed measurements of IN concentrations at different temperatures and relative humidity with a high spatial coverage. Measurements above water saturation, representing the immersion freezing mode, are reported here. Filter samples were 25 collected in 8 h intervals using a Low Volume Aerosol Sampler (DPA14, Digitel) with PM10 inlet during ACE and PS106.
24 h filter samples were collected during PS79 and PS95 using a High Volume Aerosol Sampler (DHA-80, Digitel) with PM1 inlet. The latter used 150 mm quartz fiber filters, from each of which 103 sub-samples were extracted. IN concentration was examined in a drop freezing array setup, similar to the technique used by Conen et al. (2012) and previously described in Welti et al. (2018). With the Low Volume Sampler, 8 h samples were collected on 47 mm-diameter Nuclepore polycarbonate 30 membranes (Whatman) with a pore size of 0.2 µm. Field blanks, which had undergone the same handling except for aerosol collection, were collected regularly to determine the filter background. Filters were stored in a -20 • C cold room on board and kept cool during transport home before storage at -20 • C until analysis. To determine IN concentrations, membrane filters are immersed in 10 m ultra pure water to extract collected particles and 96 aliquots (each 50 µ ) of the sample solution were subsequently examined in a drop freezing assay. IN concentrations are estimated from the number of frozen aliquots at a specific temperature according to Vali (1971). -15 • C data from AOE, TAN1502, ACAPEX, SHIPPO were also obtained from filter samples. TAN1502, ACAPEX and SHIPPO filters were analysed using the Colorado State University Ice Spectrometer (IS) (Hill et al., 2016). Collection (open-faced, 0.2 µm filters) and processing protocols for the IS are as described in McCluskey 5 et al. (2018a, b). During these expeditions IN concentrations have been measured over a broader temperature range, and results will be reported elsewhere. A discussion of the membrane filter technique used to determine IN concentrations during AOE can be found in Bigg (1990). Tab. A1 provides an overview on the location, on-board equipment, sampling time intervals and sampled air volume during expeditions from which data are included. An analysis showing that contamination from ship exhaust had no influence on ship-based IN concentration measurements during PS106 and ACE, is presented in Sec. 6 and Appx. C.   Bigg, 1973;Schnell, 1977;Borys, 1983;Rosinski et al., 1986Rosinski et al., , 1988Rosinski et al., , 1995

Geographic variability
Regional variations of IN concentrations on the continents coincide with strong local sources (e.g. Isono et al., 1959;DeMott 20 et al., 2003) or IN generating events (e.g. McCluskey et al., 2014;Suski et al., 2018). In remote maritime environments the sources of the most common IN type by concentration can be long range transport of mineral dust (e.g. Bigg, 1973) or oceans themselves. Oceans have been identified as a source of IN (e.g. by Brier and Kline, 1959;Schnell and Vali, 1975;Ickes et al., 2020)   Due to the different conditions (wind speed, temperature) and location (e.g. proximity to desert) of different parts of the world's  Bigg (1973); Schnell (1977); Borys (1983); Rosinski et al. (1986Rosinski et al. ( , 1988Rosinski et al. ( , 1995 seas and oceans, meridional variations in the IN temperature spectrum (Sec. 5) could be expected. Effects of atmospheric transport and local sources include, for example, the westward transport of large amounts of dust from the Sahara desert, higher biological productivity along coasts and in higher latitudes, and furious storm tracks or sea ice cover that promote or prevent wave-derived aerosol generation.
A comparison of continental and maritime IN concentrations observed at -15 • C, from the surface-based measurements in the 5 study of Bigg and Stevenson (1970) and meridional averages of ship-based data from Fig. 1, is shown in Fig. 2. Averaging zones of Fig. 2(b) are indicated in Fig. 1 and defined in Tab. B1. Concentrations over the Ocean are lower than the reported average concentrations over continents and the variation is higher. The continental data set consists of measurements taken in the same year and season with one type of sampling device and analysis method, while ship-based measurements are a compendium of data obtained with different methods over decades, which could add to the variability in the ship-based data  Bigg and Stevenson, 1970). However, some contemporary data (e.g. Mason et al., 2016;DeMott et al., 2017) suggest that the values from Bigg and Stevenson (1970) may underestimate the present magnitude and range 5 typifying the boundary layer over continental North America and Europe by a factor of 3 or more, while other data, e.g., in the Eastern Mediterranean region remain consistent with the prior assessment (Gong et al., 2019).
In general, the highest concentrations of an ice active aerosol are found at its source and concentrations decrease with increasing distance away from the source due to random dilution with IN-free air during transport (Anderson et al., 2003). An increase in dilution time can have the same effect as transport. Random dilution causes concentration to vary even at a fixed position 10 relative to the source (Ott, 1990). The higher concentration and smaller variation range of continental IN concentrations could indicate vicinity to the sources (minimal dilution), while the IN population on the ocean is more diffuse, suggesting extended, random dilution during transport or low, fluctuating source strength.

Temperature spectra
The ship-based data are used to construct zonal temperature spectra of the IN concentration and to investigate the frequency 15 with which concentrations are measured (shown in Fig. 3). The ambient temperature spectrum at any location is a superposition of IN contributions from several sources activating at specific temperatures. As different sources can contribute IN active at different temperatures, a change of nearby sources and source strength, can generate a change in the temperature spectrum shape (cf. Bigg (1961); Welti et al. (2018) for a discussion of temperature spectrum features). In the ship-based data, examples of non-monotonic temperature spectra with step-like features were observed close to continents, e.g. elevated IN concentration at -15 • C in the vicinity of harbors in Fig. 1. An increased slope at about -10 • C in the meridional averaged temperature spectra is found in the tropics (measurements 350 km -680 km off the coast of West Africa, see Fig. 1), suggesting a local source that concentrations from all additional data sets (cf. Tab. A1) are also shown to compare the range in which ambient concentrations vary. The normalized frequency with which concentrations occur at each temperature in the temperature spectra is shown in the right column of Fig. 3. The symmetry of contours in Fig. 3 and the offset of mean concentrations from the centers of the ranges in Fig. 2(b) suggest that concentrations follow normal distributions when plotted on a log-scale, indicating an underlying log-normal distribution. It has previously been noted that the frequency with which specific IN concentrations are observed 15 follows such a distribution (described in Isaac and Douglas, 1971;Welti et al., 2018). Two consequences of the log-normal frequency distribution can be seen in Fig. 3. While the complete data set (left column of Fig. 3) shows a spread over 3 to 4 orders of magnitude, frequently measured concentration (≥ 20%) are confined to less than one order of magnitude fluctuation (reddish colours in right column of Fig. 3). Note that at temperatures where data is sparse, high occurrence frequency is biased towards the available data points, which may compromise its reliability. Within the range of the most frequent concentrations 20 (yellow and reddish colours) the temperature dependent change in IN concentration, i.e., the slope of the IN concentration versus temperature, is similar for all five meridional zones and the absolute concentration changes by less than two orders of magnitude between zones. The skewed (log-normal) frequency distribution causes the mode (most frequently measured IN concentration) to be less than the arithmetic mean. Therefore, individual concentration measurements are often below average concentrations. 25 The temperature above which it becomes most probable to measure no IN in a sample of approximately 10 m 3 (highest frequency on zero-line in the right column of Fig. 3) is at -14 • C in N-and S-Polar and the S-Temperate zone. By contrast, the temperature threshold in the Tropical and N-Temperate climate zone is found between -10 • C to -9 • C.

Contamination from ship exhaust
To avoid contamination during ship-based measurements, the aerosol inlet and filter sampling devices are placed in front of the ship's chimney towards the bow of the ship. In situations with wind speed from the rear exceeding the speed of the ship, i.e. relative wind direction from the chimney towards the sampling location, high number concentrations of ship exhaust aerosol particles are encountered. Moderate contamination by turbulent mixing is also observed at relative wind speed close to zero. 5 The potential contribution of ship exhaust particles is determined by comparing IN concentration measured with SPIN to soot particle concentration measured with a single particle soot photometer (SP2, DMT). Fig. 4 shows the concentration of soot versus IN concentration, both binned to common 10 s intervals. Moderate correlation is found at -36 • C, whereas soot concentration is higher than the IN concentrations at higher temperatures and lower at lower temperatures. Comparing the temperature spectra in Fig. 3, where a distinct increase in IN concentration at -36 • C can be seen, confirms that ship exhaust 10 potentially distorts the ambient temperature spectrum at and below -36 • C, but not above. Increased concentrations at -38 • C could be caused by the onset of homogeneous freezing. This gives confidence that the effect of contamination from ship exhaust on the immersion freezing experiments with filter samples is negligible. Fig. C1 in Appendix C confirms no effect of sampling air coming from the direction of the chimney of the ship on filter measurements at T ≥ -20 • C. It can be argued that at temperatures above -36 • C, the effect of soot if any should be a suppression in activity of other particles by covering their 15 ice active sites (Mossop and Thorndike, 1966). Note that the present analysis is valid for the RV Polarstern exhaust. Other type of fuel or ship engine might yield a different result (Thomson et al., 2018). However, Schnell (1977) also tested a possible contamination from the exhaust of the USNS Hayes by purposely exposing membrane filters to the exhaust plumes from galley and engines and found no IN from either source active at -15 • C. Also McCluskey et al. (2018a) found no statistical significant difference due to varying black carbon levels from exhaust contamination at temperatures between -16 • C and -26 • C on board 20 the RV Investigator during the CAPRICORN campaign.

Discussion
The maritime IN concentration data presented here can be used to predict the temperature spectra in a range of oceanic regions.
This can help the future development of measurement techniques and sampling strategies. As an example, to avoid artefacts from the sensitivity range of measurement techniques, sample volume and probed temperature range need to be matched. In the current data set, the narrowing of variation below -20 • C in the filter measurements from the temperate climate zone (Fig. 3,   5 right column), is an artefact caused by the upper limit of detectable concentration and can be avoided by sampling smaller volumes and by using dilutions of filter suspensions. In general, the sample volume and sampling time over which concentration is averaged to one data point, affects the absolute value and the variability between measurements. Smaller sample volumes result in larger scattering of the data within experimental detection limits, including measurements of practically zero alternating with high concentrations at the upper detection limit. The smaller the sample volume the higher the limit of detection and 10 vice versa, i.e. larger, sampled air volumes allow the detection of low concentrations, while smaller sample volumes are more suitable for accessing high concentrations.
To develop parametrizations, concentrations averaged to scales required for the application should be used (Gultepe et al., 2001). Grid-box scales in numerical models are much larger than the practical sampling volume of ambient measurements. As averaging any distribution of concentrations by means of probing larger air volumes or averaging collected samples over time 15 converges to the arithmetic mean concentration, the mean IN concentration (shown in Fig. 2(b) for -15 • C in five climatic zones and in Fig. 5(a) averaged over each hemisphere) is the correct input value for large scale numerical models, even if it is not the most frequently measured (mode) concentration. This also underscores the importance to use sufficiently large data-sets for parametrization development to avoid bias.
The frequency distributions in the right column of Fig. 3 illustrate that within the sample volumes where IN concentrations 20 can be measured, low values are more frequent than high values. Averaging concentrations from a skewed (log-normal) frequency distribution results in mean concentrations higher than the most frequently measured concentration. Fig. 5 depicts the temperature spectra of a) average and b) most frequently measured IN concentration in the north and southern hemisphere.
The discrepancy between averaged and most frequently measured concentrations is most pronounced for low concentrations.
Within the sensitivity limits of measurements presented here, most frequently no IN are detected above -15 • C in 10 m 3 of 25 maritime air.
Between -25 • C and -30 • C, a jump in the N-hemisphere temperature spectra can be seen in Fig. 5. It could be speculated that the jump coincides with the temperature where mineral dust particles start to exert a strong influence (e.g. Boose et al., 2016) and the N-S difference could be due to more dust emitting land masses in the north. Another reason for the jump could be that it is an artefact from the combination of filter measurements reaching their upper detection limit with SPIN measurements 30 reaching their lower detection limit in this range of IN concentration. Further measurements and the development of measurement techniques that can bridge this gap are needed.
In Fig. 1 a clear Bigg (1973) can be seen. Bigg reported averaged concentrations measured in 5x5 degree sectors. Noting the log-normal frequency distribution of observed IN concentrations, the offset between the average and most frequently occurring concentration at -15 • C (see Fig. 5) could be one factor helping to explain the difference. A decline in IN abundance in the Southern Ocean over the last decades (suggested by Bigg, 1990) could be another reason.

Conclusions
Ship-based observations are an efficient strategy to obtain measurements with a wide spatial coverage representing IN charac- In the data presented here, the lowest IN concentrations are observed in polar regions and highest in the temperate climate 10 zones. The overall geographical variation in maritime IN concentration is surprisingly small (below two order of magnitudes).
A zonal difference was found in the temperature above which most frequently no IN are detected (-9 • C in N-Temperate, -10 • C in Tropical, but much colder at -14 • C in S-Temperate, N-, and S-Polar climate). The largest difference between northern, and southern hemisphere is found at temperatures below -28 • C, where IN concentration is higher in the northern hemisphere. This coincides with the temperature range where desert dust particles become especially active ice nuclei. Competing interests. The authors declare that they have no conflict of interest.  Bigg (1973), [4] Borys (1983, [5] Rosinski et al. (1986), [6] Rosinski et al. (1988), [7] Rosinski et al. (1995), [8] Bigg (1996 Bigg and Leck (2001) [10]   atmospheric chemistry, aerosol chemistry and physics, meteorology, oceanography and seismology (Leck et al., 2004).   Fig. 2 and zonal temperature spectra shown in Fig. 3. Differences between IN concentrations over the northern and southern hemisphere in comparison to three often used parametrizations is shown in Fig. 5.