Rainfall drives atmospheric ice-nucleating particles in the coastal climate of southern Norway

Ice-nucleating particles (INPs) active at modest supercooling (e.g. −8 C; INP−8) can transform clouds from liquid to mixed phase, even at very small number concentrations (< 10 m−3). Over the course of 15 months, we found very similar patterns in weekly concentrations of INP−8 in PM10 (median= 1.7 m−3, maximum= 10.1 m−3) and weekly amounts of rainfall (median= 28 mm, maximum= 153 mm) at Birkenes, southern Norway. Most INP−8 were probably aerosolised locally by the impact of raindrops on plant, litter and soil surfaces. Major snowfall and heavy rain onto snow-covered ground were not mirrored by enhanced numbers of INP−8. Further, transport model calculations for large (> 4 m−3) and small (< 4 m−3) numbers of INP−8 revealed that potential source regions likely to provide precipitation to southern Norway were associated with large numbers of INP−8. The proportion of land cover and land use type in potential source regions was similar for large and small numbers of INP−8. In PM2.5 we found consistently about half as many INP−8 as in PM10. From mid-May to mid-September, INP−8 correlated positively with the fungal spore markers arabitol and mannitol, suggesting that some fraction of INP−8 during that period may consist of fungal spores. In the future, warmer winters with more rain instead of snow may enhance airborne concentrations of INP−8 during the cold season in southern Norway and in other regions with a similar climate.


Introduction
Most precipitation in the midlatitude and polar regions is linked to ice formation in clouds (Field and Heymsfield, 2015).Biological particles catalyse the freezing of supercooled cloud droplets at temperatures between −1 and −15 • C, whereas other particles (e.g.mineral dust, soot) are active at colder temperatures (Després et al., 2012;Murray et al., 2012).The ice particles formed grow to snowflakes through vapour deposition and, more rapidly, through riming and aggregation.A few initial ice particles (< 10 m −3 ) catalysed at −8 • C near the cloud top can rapidly glaciate a shallow supercooled cumulus (Mason, 1996;Crawford et al., 2012).This is possible through explosive ice multiplication by riming and splintering graupel pellets between −3 and −8 • C (Hallett and Mossop, 1974).Therefore, icenucleating particles (INPs) active at −8 • C or at warmer temperatures (from here on collectively denominated as INP −8 ) could strongly influence precipitation development, despite their usually small number concentration in the atmosphere, for example compared to mineral dust or soot particles.The few atmospheric data on INP −8 suggest that vegetated lands are stronger sources than deserts (Conen et al., 2015), and that rainfall triggers the aerosolisation of such INPs from forest (Hara et al., 2016).Aerosolisation is probably due to the mechanical impact of raindrops on surfaces hosting organisms that, as a whole or in parts, can serve as INPs.Increased concentration during and after rainfall has also been demonstrated for INPs active at colder temperatures (−15 • C) by Bigg and Miles (1964, examined in more detail in Bigg et al., 2015) at 24 sites in Australia, and by Huffman et al. (2013), Tobo et al. (2013) and Prenni et al. (2013) at a site Published by Copernicus Publications on behalf of the European Geosciences Union.
F. Conen et al.: Rainfall drives atmospheric ice-nucleating particles in Colorado, USA.Investigations at the latter site were accompanied by detailed characterisation of aerosolised particles regarding their size distribution, fluorescence, morphology and biological origin.Huffman et al. (2013) suggested that "follow-up studies in other environments shall elucidate whether the observed rain-related bioaerosol increase is a common feature of terrestrial ecosystems or specific for the investigated semi-arid environment".Thus, we took this suggestion as a starting point to investigate rainfall effects in the coastal climate of southern Norway.

Material and methods
Unlike previous studies, we continuously sampled aerosol particles over the course of 15 months, from October 2013 to December 2014, on a filter replaced once a week.The low time resolution of 1 week provided a large sampled volume, which enabled the reliable detection of INP −8 in all samples.

Site description
The Birkenes Observatory (58 • 23 N, 8 • 15 E, 219 m a.s.l.) is situated approximately 20 km from the Skagerrak coast in southern Norway (Fig. 1) and is located on a minor hilltop in an undulating terrain, which allows for efficient ventilation and air mass mixing.Kristiansand (61 000 inhabitants, 2016) is the nearest city, located 25 km south/south-west of the station.The observatory is located in the Boreonemoral zone with mixed coniferous and deciduous trees, accounting for 65 % of the land use near the site.Meadows and lowintensity agricultural areas account for 10 % each, whereas 15 % are freshwater lakes.Birch is the most common deciduous tree in the area around the observatory, for which budding starts in late April and the shedding of leaves is over by mid-October.
By its proximity to the coast and low altitude, the Birkenes Observatory experiences a coastal climate, with relatively mild winters (T mean Jan-Feb 2014 = −0.3• C) and moderately warm summers (T mean Jun-Aug 2014 = 15.6 • C).The annual amount of precipitation around Birkenes was 2077 mm in 2014, about 1.5 times the normal  amount.Of this, 12 % precipitated as snow.The mean temperature in 2014 exceeded the norm by 2-3 • C (MET, 2015).The prevailing wind direction is from the west and the south-west, occasionally with quite high wind speeds.Hence, the observatory is situated downwind of major emission regions in continental Europe.

Aerosol sampling
Ambient aerosol filter samples were collected as part of the Norwegian national monitoring programme (Aas et al., 2015), using two Kleinfiltergerät low-volume samplers with a PM 10 and a PM 2.5 inlet to collect aerosols on prefired (850 • C; 3 h) single-quartz fibre filters (Whatman QM-A; 47 mm in diameter).Both samplers operated at a flow rate of 38 L min −1 , corresponding to a filter face velocity of 47.3 cm s −1 .Filters were conditioned at 20 ± 1 • C and at 50 ± 5 % RH (relative humidity) for 48 h before and after exposure.Filters were kept in petri slides for transportation and storage.Post conditioning, filters were stored at 4 • C for approximately 1 month for subsequent analysis of OC / EC, then at −18 • C prior to analysis of arabitol and mannitol.

Analysis of ice-nucleating particles
Number concentrations of INP −8 on PM 10 and PM 2.5 filter samples were determined with 108 punches (1.0 mm diameter) from each filter.Each punch was immersed in Milli-Q water (0.1 mL) in a tube (1.5 mL, Eppendorf Safe-Lock), cooled from −4 to −12 • C (0.3 • C min −1 ) in a cold bath (Lauda,model RC6).The number of frozen tubes were counted every 1 • C temperature step to calculate the number concentration of INPs in sampled air (Conen et al., 2012).The punches from 10 filters in each size fraction were tested a second time after they had been immersed for 10 min in a water bath at 90 • C. We also tested 24 field blanks the same way.Only one had a small positive signal (0.08 INP −8 m −3 ) and 23 were negative at −8 • C, thus we did not do a blank correction.As we have analysed punches of each filter in two lots of 54, we can estimate the uncertainty of our procedure.The mean deviation of INP −8 derived from single lots of 54 punches, from that of both lots taken together (108 punches), was 22, 19, 13 and 12 % for INP −8 < 1, 1 to 2, 2 to 4, and > 4 m −3 , respectively.

Analysis of arabitol and mannitol
Arabitol and mannitol have previously been identified as amenable tracers of fungal spores (Bauer et al., 2008).Concentrations of arabitol and mannitol in PM 10 filter samples were determined using Waters Acquity ultra-performance liquid chromatography (UPLC) in combination with Waters Premier XE high-resolution time-of-flight mass spectrometry (HR-TOF-MS) operated in the negative electrospray ionisation (ESI) mode: resolution > 10 000 FWHM (full width at half maximum).The analytical methodology is based on that described by Dye and Yttri (2005) for monosaccharide anhydrides, deviating from the original one only by choice of the column (2.1 × 150 mm HSS T3, 1.8 µm, Waters Inc.).Arabitol and mannitol were identified on the basis of retention time and mass spectra of authentic standards (ICN Biomedicals).Response factors for arabitol and mannitol were calculated from external standards.Isotope-labelled standards of levoglucosan ( 13 C-levoglucosan, 98 %, Cambridge Isotopic Laboratories) were used as the internal recovery standard.

FLEXPART
FLEXPART is a Lagrangian particle dispersion model (Stohl et al., 1998(Stohl et al., , 2005) ) and is used to investigate the origin of air masses and their potential for emission uptake.The model is driven by 1 × 1 • operational meteorological data from the European Centre for Medium Range Weather Forecast (ECMWF) with 3 h temporal resolution and 137 vertical levels.The model calculates the trajectories of tracer particles using the interpolated mean winds plus random motions representing turbulence and moist convection.The particles are subject to dry and wet deposition, the latter of which is described in detail in Grythe et al. (2017).
For this study, FLEXPART was run for 20 days backward in time with a black carbon tracer, which experiences wet and dry deposition.Black carbon was used as a proxy for INPs, as both are susceptible to dry and wet scavenging.For the exact time interval in which each filter sample was collected, 400 000 particles were released.The FLEXPART output represents potential emission sensitivity in units of seconds.It quantifies the impact of potential emissions on the aerosol concentration at the measurement site.If multiplied with known emission fluxes, the aerosol concentration at the receptor is obtained.Since emissions of INPs are not known, we use arbitrary constant emission densities for the different land use types, based on the International Geosphere-Biosphere Programme (IGBP) data (Belward, 1999) at 1×1 • resolution.The land use categories used are urban, agriculture, range land, deciduous forest, mixed forest, coniferous forest, water, desert, wetland, agriculture/range land, rocky areas, snow and rainforest.Our procedure thus quantifies the relative impact that INP emissions in these land use types would have on the INP concentration in a measurement sample.

Time series of INP −8 and precipitation
Our observations revealed a general seasonal pattern.Values of INP −8 in PM 10 were mostly < 2 m −3 during spring and summer, and > 4 m −3 over the course of several weeks during autumn.Values decreased with snowfall in winter and were on average 1.2 m −3 as long as the ground was covered by snow (Fig. 2).Elevated INP −8 levels of short duration in spring and summer were associated with rainfall exceeding 20 mm per week.Similar rates of snowfall (January 2013) were not accompanied by additional INP −8 .Deposition velocity of snow is in the order of 1 m s −1 (Garrett et al., 2012), while that of even a small raindrop (1 mm diameter) is already four times as large (Gunn and Kinzer, 1949).Hence, for the same mass, the kinetic energy of rain (proportional to the velocity squared) is at least an order of magnitude larger compared to precipitation in the form of snow, and so the energy is available for dispersion and aerosolisation of particles.Hara et al. (2016) saw enhanced concentrations of INP −7 during rainfall, but not during snowfall at Kanazawa, a coastal city in Japan.Huffman et al. (2013) suggested that rainfall may trigger the release of INPs from vegetation through its mechanical impact.Such impact happens on leaves, but also on other surfaces, such as twigs, stems, soil and on leaf litter covering the forest floor.If leaves on trees were the only relevant source of INPs, we should have seen a marked decrease in atmospheric concentrations once leaves were shed by mid-October.No change of this kind was discernible in the autumn of 2013 or 2014.Numbers of INPs on shed leaves increase substantially upon decay (Vali et al., 1976).At first sight, it seems unlikely that INPs produced on the forest floor are transported above the canopy.However, once trees are defoliated, raindrops hit the ground with less interception by canopies, and turbulent momentum penetrates more easily into the subcanopy layer (Staebler and Fitzjarrald, 2005).Continued occurrence of large numbers of airborne INP −8 during the defoliated period suggests that the reduced likelihood of aerosolisation from the forest floor is largely compensated by the increased source strength of INPs due to the decay of the shed leaves.Throughout the year, the increase in INP −8 with weekly rainfall occurs despite the fact that scavenging of such particles is also enhanced by rain.
An increase or decrease in the amount of rainfall from one week to the next was followed in 76 % of all cases by .53 for all weeks with neither snowfall nor snow on the ground (n = 55).Both correlations are statistically significant (p < 0.01).There are at least two reasons for the deviations: first, depending on rain intensity and raindrop size, the same amount of weekly rainfall can have a very different impact on aerosolising microorganisms from vegetation (Paul et al., 2004).Second, the presence and density of IN-active (ice nucleation active) material on surfaces hit by raindrops may change with season or on shorter timescales.Considering both issues, the time course of INP −8 mirrors that of rainfall surprisingly well (Fig. 2).

Potential source regions
We summarised source receptor sensitivity (SRS) fields for situations with > 4 INP −8 m −3 and when INP −8 were < 4 m −3 in PM 10 (Fig. 3).Higher concentrations of INP −8 were not associated with additional source areas to those seen when INP −8 were < 4 m −3 .Hence, the higher concentrations of INP −8 were not transported to Birkenes from specific strong sources afar.The main difference when INP −8 were > 4 m −3 was a weaker influence from the north-east (Fennoscandia, Norwegian Sea, Barents Sea, north-western Russia, Siberia).Air masses from the north-east mainly ar-rive at Birkenes with high-pressure systems and bring little or no rain.Large amounts of rain arrive with cyclones of the North Atlantic storm track.Although high INP −8 concentrations are also associated with enhanced transport from northern Africa, this transport signal is due only to a few cases and unlikely to cause the higher INP −8 levels.Furthermore, the Saharan Air Layer sampled off the western coast of northern Africa on Tenerife over the course of a year, and analysed by the same method as we used here, was previously found to contain no more than 1 INP −8 m −3 , even during dust storms (Conen et al., 2015).Overall, there was only a minor difference in the proportion of influence from land surfaces between situations with higher or lower values than 4 INP −8 m −3 (48 and 51 %, respectively) and even smaller differences in land cover type.The small difference in the SRS fields supports the idea that it is rain and its impact on aerosolising INP −8 locally that drives temporal variation in INP −8 concentrations at Birkenes.

Size of INP −8
At seven other sites across the Northern Hemisphere, Mason et al. (2016) found 38 to 90 % of INP −15 collected with a cascade impactor to be larger than 2.5 µm (mean = 62 %, s.d.= 20 %).In the present study, INP −8 were equally distributed amongst the fine (PM 2.5 ) and the coarse fraction (PM 10−2.5 ) of PM 10 (Fig. 4).Such a 50-50 % distribution can be expected, if there is a unimodal, normal distributed peak of INP −8 , which peaks around 2-3 µm aerodynamic diameter.Some findings support this presumption.Schumacher et al. (2013) continuously measured the size distribution of fluorescent biological aerosol particles (FBAP) over 18 months at a site in southern Finland, 960 km to the east  of Birkenes, while also observing a marked decrease in atmospheric concentrations at times when snow covered the ground.Their results showed a dominant, often extremely narrow mode between 2.5 and 3.0 µm.Similar observations at a forest site in Colorado revealed rain-induced increases in FBAP with a peak at 2.0 µm (Schumacher et al., 2013).Huffman et al. (2013) have demonstrated a close link between FBAP and INPs for the site in Colorado.They saw rainfall that immediately increased the number of airborne biological particles with a size of 2-3 µm, while also increasing INP −15 .In addition, larger particles (4-6 µm), comprising INPs, appeared several hours later and lasted for up to 12 h after rainfall had ceased.Huffman et al. (2013) were able to identify several IN-active species from aerosol sam-ples collected during rainfall, including two fungi previously unknown to be IN-active that produced spores between 1 and 4 µm in geometric size.Strong correlations between FBAP and fungal spore markers, such as arabitol, strongly supported the notion that most FBAP aerosolised during rain were of fungal origin (Gosselin et al., 2016).
Another important source of FBAP in PM 10 is pollen (Manninen et al., 2014).Pollen from birch, the most abundant deciduous tree around the Birkenes Observatory, has an aerodynamic diameter of 20 µm (Efstathiou et al., 2011).Smaller fragments of pollen are generated by osmotic rupture when pollen grains get wet.These fragments are as INactive as intact grains (Pummer et al., 2012).On rainy days their abundance increases in the fine fraction (Rathnayake et al., 2017).However, we can exclude a major contribution of pollen-derived INPs, because exposure to 90 • C deactivated all INP −8 in 9 of the 10 samples treated that way and 92 % of INP −8 in the remaining sample (26 March-2 April 2014).Still, a few heat-resistant INP −8 might have escaped observation at concentrations below the limit of detection in our approach (0.08 m −3 ).If so, they would constitute, on average, 7 % or less of INP −8 found prior to treatment.The INPs of bacteria and most fungi are proteins and denatured at this temperature (Pummer et al., 2015).INPs from pollen or fractured pollen are most likely polysaccharides and would have remained active after heating to 90 • C (Pummer et al., 2012(Pummer et al., , 2015)).
Recently, Wang et al. (2016) described a mechanism that generates airborne soil organic particles (ASOPs) of submicron size by air bubbles bursting at the water-air interface of impacted raindrops.Soil organic matter can harbour large numbers of INP −8 (Schnell and Vali, 1972;Conen et al., 2011;O'Sullivan et al., 2015;Hill et al., 2016).Therefore,

Time series of arabitol and mannitol
Arabitol and mannitol serve as carbohydrate stores in fungal spores.Their ambient air concentration has been found to correlate well with number concentrations of airborne fungal spores (Bauer et al., 2008), but not necessarily with ergosterol (Burshtein et al., 2011), a dominant sterol in most fungi (Weete et al., 2010).It seems that arabitol and mannitol are specifically associated with spores released under moist conditions, as occur in forests during night-time (Zhu et al., 2016).At Birkenes, arabitol and mannitol had very similar temporal patterns throughout the year (Fig. 5).Their concentrations were low from January to mid-April, then increased and remained enhanced throughout summer.During the warmer part of the year, from mid-May to mid-September, they correlated significantly with INP −8 (mannitol: r = 0.72, p < 0.01; arabitol: r = 0.48, p < 0.05).These correlations (Fig. 6) support the above-mentioned notion (Sect.3.3) that FBAP and INP −8 may to some extent be of fungal origin, at least during the warmer part of the year.Another hint in this direction comes from measurements about 300 km north-east of Birkenes.There, in summer, arabitol and mannitol had a unimodal size distribution peaking between 2 and 4 µm aerodynamic diameter (Yttri et al., 2007), which coincides with our interpretation of the 50-50 % distribution of INP −8 amongst PM 2.5 and PM 10−2.5 .Concentrations of the fungal spore markers increased towards the end of 3 rather dry weeks in September and reached their annual maxima with intensive rainfall at the beginning of October, followed 1 week later by a 7-fold increase in INP −8 .Changes Assuming 1.2 pg arabitol and 1.7 pg mannitol per fungal spore (Bauer et al., 2008), we estimate for the period from mid-May to mid-September average spore concentrations of 4.6 and 5.8 × 10 3 m −3 .The average concentration of INP −8 during the same period was 1.6 m −3 .If all INP −8 were spores, there would have been 2.8 or 3.5 × 10 −4 INP −8 per spore, which is in the upper range of values reported by Morris et al. (2013, Fig. 1) for urediniospores of rusts.However, spores found in the atmosphere are not only produced by rusts.For example, Cladosporium species contribute a large proportion of airborne spores (Maninnen et al., 2014) and their onset of freezing is well below −25 • C (Iannone et al., 2011).At the same time, some fungi release INP −8 from their mycelium in the form of macromolecules (Fröhlich-Nowoisky et al., 2015), which are unlikely to contain storage products, such as arabitol or mannitol.
For the last 3 months of the year, the average INP −8 to spore ratio was about twice as large as compared to the preceding 4 months.Since the time course of INP −8 was no longer directly related to that of fungal spore markers during this period, it might be that the fungal composition had changed, or that bacteria had become a more important source of INP −8 .The expression of ice nucleation activity in bacteria is favoured by cold temperatures and nutrient limitation (Nemecek-Marshall et al., 1993).Hence, when similar numbers of bacteria are aerosolised by the same amount of rainfall, they likely contribute larger numbers of INP −8 during the cold season than during the warm months.Overall, the relative contribution of INP −8 from any type of microorganism might have changed by the end of September, as a result of leaves starting to be shed by deciduous trees.Decaying leaves provide the substrate for a highly dynamical succession of interacting fungal and bacterial populations (Purahong et al., 2016).Rain-released INP −8 are equally distributed amongst the fine (PM 2.5 ) and the coarse fraction (PM 10−2.5 ) of PM 10 .Sensitivity to heat treatment (90 • C) suggests bacterial and fungal sources and not pollen.The assumption of relevant fungal sources is supported during the warmer part of the year by some similarities in the temporal pattern of INP −8 and the fungal spore markers, arabitol and mannitol.However, major shifts in microbial community composition occur when leaves are shed in autumn and start to feed a highly dynamical succession of interacting fungal and bacterial populations.These dynamics likely also affect the strength and composition of the various sources of INP −8 .
In general terms, we expect similar relations between rainfall and warm-temperature INPs in other coastal regions with a comparable climate and ecosystem, such as the Pacific coasts of Canada and Chile, Japan, and New Zealand.Global warming may lead to shorter periods of snow cover on the ground and a greater proportion of precipitation falling as rain instead of snow.These changes would probably result in larger airborne concentrations of INP −8 during the cold season.Whether they have an effect on cloud development in these regions remains an interesting question for further studies.

Figure 1 .
Figure 1.Location of Birkenes Observatory (a) and a view of the observatory (b).

Figure 2 .
Figure 2. Time course of INP −8 in the PM 10 (black) and PM 2.5 (green) particle size fractions.Peaks in INP −8 largely mirror peaks in precipitation (blue is inverse scale to mirror precipitation values on the horizontal axis).Snow (open circles) or small amounts of rain are accompanied by very small numbers of INP-8 in the same week.Significant amounts of snow on the ground (grey bars) seem to attenuate the increase in INP −8 with intense rainfall.Total amount of precipitation in 2014 was 2077 mm (entire period shown: 2640 mm).Precipitation and snow cover data are the averages of three stations operated by the Norwegian Meteorological Institute in the municipality of Birkenes.

Figure 3 .
Figure 3. Source receptor sensitivity (SRS) fields for situations with > 4 INP −8 m −3 (a) and when INP −8 were < 4 m −3 (b) as derived from FLEXPART.The SRS unit is seconds, which would result in a mass concentration (kg m −3 ) at the receptor when multiplied with an emission flux (kg m −3 s −1 ) into the model grid cells.Since emission fluxes are not known for INP −8 , SRS values can be considered as a measure of relative impact that INP emissions from a particular area would have had on INP concentrations at Birkenes.The potential influence was strongest from areas shown in red and weakest from those in white and purple.

Figure 4 .
Figure 4. Ice-nucleating particles active at −8 • C in PM 2.5 , relative to their number in PM 10 .

Figure 5 .
Figure 5. Weekly concentration of arabitol (red) and mannitol (magenta).The time courses of INP −8 in PM 10 (black; × 5, to fit onto the same scale) and precipitation (blue) are copied from Fig. 1 to facilitate direct comparison.
in INP −8 continued to follow those of the fungal spore markers with a delay of 1 week until the beginning of December.In December, INP −8 remained elevated whereas, concentrations of arabitol and mannitol decreased markedly.The fungal spore markers and INP −8 reached their minimum with snowfall in the last week of the year.

Figure 6 .
Figure 6.Correlation between INP −8 and the fungal markers mannitol (a) and arabitol (b) for the time period from 14 May to 24 September 2014 (black circles, regression line and equation are fitted to these data).Concentrations of INP −8 before and after this period (crosses) did not correlate with either fungal spore marker.
the coastal climate of southern Norway drives the near-surface concentration of INP −8 across all seasons.Concentrations of INP −8 increase with amounts of rain.Most airborne INP −8 are probably aerosolised locally through the impact of raindrops onto surfaces hosting microorganisms that synthesise INPs.Snowfall has no such effect.When trees are defoliated between October and April, decaying leaf litter on the ground constitutes a likely INP source.During this time, snow cover on the ground strongly reduces such INP aerosolisation by rainfall.