Ice nucleation active particles are efficiently removed by precipitating clouds

Ice nucleation in cold clouds is a decisive step in the formation of rain and snow. Observations and modelling suggest that variations in the concentrations of ice nucleating particles (INPs) affect timing, location and amount of precipitation. A quantitative description of the abundance and variability of INPs is crucial to assess and predict their influence on precipitation. Here we used the hydrological indicator δ18O to derive the fraction of water vapour lost from precipitating clouds and correlated it with the abundance of INPs in freshly fallen snow. Results show that the number of INPs active at temperatures ≥ −10 °C (INPs−10) halves for every 10% of vapour lost through precipitation. Particles of similar size (>0.5 μm) halve in number for only every 20% of vapour lost, suggesting effective microphysical processing of INPs during precipitation. We show that INPs active at moderate supercooling are rapidly depleted by precipitating clouds, limiting their impact on subsequent rainfall development in time and space.


Ice nucleation in cold clouds is a decisive step in the formation of rain and snow. Observations and modelling suggest that variations in the concentrations of ice nucleating particles (INPs) affect timing, location and amount of precipitation. A quantitative description of the abundance and variability of INPs is crucial to assess and predict their influence on precipitation.
Here we used the hydrological indicator δ 18  Ice formation in clouds contributes to the development of precipitation at mid-latitutdes [1][2][3][4] . Ice nucleating particles (INPs) of biological origin can be effective in promoting ice nucleation at temperatures around − 10 °C or warmer [5][6][7] , whereas at colder temperatures inorganic substances are likely to be responsible for an increasing fraction of ice particles formed in the atmosphere 8 . Here we focus on the cumulative number of INPs active at temperatures warmer than − 10 °C (INPs −10 ), the range where the activity of INPs of biological origin seems to be dominant. Such INPs include certain bacteria, fungal spores and pollen, but a large fraction of INPs from biological sources in the atmosphere may also be composed of ice nucleation active macromolecules associated with mineral and soil particles 9,10 . Because of usually very small number concentrations in the atmosphere, the potential role of such particles in conditioning precipitation is still contentious 11,12 .
Elevated concentrations of INPs associated with dust from desert storms on other continents and with far away and regionally emitted INPs were recently found to contribute to precipitation over the Western USA 4 and the Amazon basin 13 respectively. Overall, it is likely that there is a coincidence in time and space of the concentration of INPs and the intensity of precipitation events 14 , raising the general question of where and when cloud glaciation and subsequent precipitation are limited or facilitated by INPs. To address this question, it is crucial to understand the major factors driving the variation of atmospheric concentrations of INPs, which have been observed to range over several orders of magnitude 3,15 .
Feedbacks between human activities and climate modifications could be, or become, partly influenced by INPs. In fact, intensifying land use and climatic change are likely to increase future emissions of INPs associated with wind-blown soil dust 16 . Changes in vegetation cover, crop type and management may also affect emissions of ice nucleating particles from vegetation 12 . In this study we intend to quantify the relation between the fraction of water lost from air masses and the residual concentrations of INPs −10 .
INPs get rapidly lost from a precipitating cloud. Over a 10-month period (December 2012 to September 2013) we sampled snow within precipitating air masses that had lost between 22 and 95% of their initial water content before arriving at the observatory (Fig. 2). The decision to initiate a sampling campaign depended on weather forecasts that predicted snowfall for at least two full days in a row, to assure that we could collect multiple samples within the same campaign. A total of 304 mm were collected, reaching approximately 20% of the total amount of precipitation fallen in the same period of 10 The predominant factor with a similarly marked variability that correlates with the abundance of INPs −10 observed in precipitation, is the fraction of residual water vapour in clouds (Fig. 2).
The abundance of INPs −10 in snow declines exponentially with increasing proportions of water lost before arrival of an air mass at Jungfraujoch (Fig. 2b), and is best described by equation (1): = − . , < . , = ) s R 052; Spearman' r 061 p 0 001 n 91 2 We can derive from this empirical equation, fitted to our data, an estimate for the largest number of INPs −10 to expect in 1 ml of precipitation at Jungfraujoch. If the very first precipitation from an air mass is just about triggered at the observatory (1-f V = 0) we would expect it to contain around 10 3 INPs −10 · ml −1 . This number is probably a consequence of the strength of the sources of INPs influencing Jungfraujoch. Nevertheless, we can presume the exponential relationship to hold also in other places because of a generally geometric behaviour observed in precipitating particles 22 . Physical processes during the course of precipitation define the factor − 7.57 in the exponent, which might have a similar value also at other locations where the same physical processes are at work. It suggests that the concentration of INPs −10 halves with about every 10% of moisture lost from a precipitating air mass (e.g. moving 1-f V = 0.5 to 1-f V = 0.6 results in: e (−7.57 ⋅ 0.6) /e (−7.57 ⋅ 0.5) = 0.47).
Selective removal of INPs. The question remains whether there is experimental evidence for an INP being more likely to be deposited from an air mass than a particle of similar size that is not ice nucleation active. INPs should, in principle, be the starting point for snowflakes precipitating from a cloud. The average activation diameter for cloud condensation nuclei at Jungfraujoch is about 0.1 μ m and most INPs are probably larger than 0.5 μ m 23,24 . Relating concentrations of INPs −10 in precipitating snow to all particles larger than 0.5 μ m (N > 0.5 ) in the same air volume reveals a significant negative trend in the ratio of INPs −10 to N > 0.5 with an increasing fraction of vapour lost, despite the large scatter of data (Fig. 3).
If the ratio of INPs −10 to N > 0.5 were independent from the fraction of water vapour lost, we could say that both kinds of particles are removed with equal efficiency from a precipitating cloud. This is clearly not the case. The function fitted to our data (Fig. 3) suggests that the ratio of INPs −10 to particles of similar size N > 0.5 is reduced to 0.69 times of what it was before with every 10% of initial water vapour lost from a precipitating cloud (e.  25 . This selective loss of INPs is highly significant, but explains only about one sixth of the total variation in the ratio of INPs −10 to N >0.5 . The remaining variation might be due to source-related factors and could reflect temporal and spatial differences in INPs −10 , N >0.5 and in the proportion of INPs among particles N >0.5 emitted to the atmosphere before precipitation. Part of the scattering of INPs −10 and N >0.5 data may be also due to differences in the dimensions of INPs and total particles in each sample. In fact, not only nucleation but also impaction scavenging of aerosols can contribute to the simultaneous removal of particles, with an efficiency largely depending on the size of aerosols and precipitation intensity [26][27][28] .

Discussion
Land use and climate change alter the distribution, the quality and the size of soil and vegetation cover in a landscape, and with it the strength and distribution of sources of different INPs 12,16,29 . As we illustrate here, the ratio of stable water isotopes in precipitation can be used in novel way to characterize the history of air masses in terms of residual abundance of INPs. Despite simplifying assumptions, our approach explains more than 50% of the large variation of INPs −10 observed in snow both within short sampling campaigns and over the year. Much of the unexplained variation is probably due to variations in the initial concentration of INPs before precipitation, which depends both on the source strength of INPs and on the degree of the mixing of air masses with different initial concentrations of INPs (e.g. from different altitudes or regions). Source strength in the lowland north of Jungfraujoch is scattered over two orders of magnitude during most of the year, but does not seem to change with season 30 . In the same study the seasonal variation observed on Jungfraujoch seemed to be driven by microphysical processing of INPs through activation and deposition from approaching air masses. The same process may explain much of the observed temporal variation of INPs in this study (Fig. 2). It is in fact supported by the finding that INPs −10 are deposited more efficiently than other particles of similar size (Fig. 3).
Carrying out similar measurements on stable water isotopes and on INPs also at other stations will firstly lead to the constant improvement and refinement of our calculations and, secondly, it will shed new light on the evolution of concentrations of INPs before and during precipitation events over the trajectories of air masses.
This will provide an important contribution for mapping the probabilities of the abundance of INPs and their exchanges across regions, in particular the estimation of how far from a source and along a LGR and all results presented here were related to the standard VSMOW. The local meteoric water line obtained from the whole set of yearly data fits well with the equation associated to the global meteoric water line (δ 2 H = 7.7 δ 18 O + 10.6; R 2 = 0.98). This indicates the absence of significant disequilibrium conditions at Jungfraujoch compared to the global behaviour of precipitations. The remaining water vapour fraction (f V ) was calculated from δ 18 O‰ values measured in snow (δ L ) following the method described in Rowley 2001 19 .
The evolution of δ 18 O in vapour (δ V ) can be described by a Rayleigh-type fractionation model 21,32 : In our calculations, the fractionation factor from liquid to vapour α L/V along the trajectory of the cloud was assumed constant during the entire path of a precipitating cloud and proportional to the average value between the temperature of the air at Jungfraujoch and the estimated temperature at the sea surface from where the air mass originated. The dependence of α L/V from absolute temperature (T) was calculated according to Majoube 1971 33 : The isotopic ratio of the vapour at Jungfraujoch (δ V ) was calculated from the isotopic ratio of snow (δ L ) and the fractionation factor liquid-vapour at the temperature of the air recorded at the station: Seawater was considered the principal and constant source of moisture in calculating the isotopic ratio of the initial water vapour (δ V,0 ): with the isotopic δ 18 O ratio of seawater (δ L,0 ) homogeneously equal to 0‰, since it coincides with the standard reference for water stable isotopes measurements and the fractionation factor between seawater and vapour α V/L , equal to 1/α L/V . Over the year, the station is affected by intermittent influence of the boundary layer 34 , with air masses coming from different geographical regions and its location in a saddle allows air systems to be channelled along two main directions, mainly North-West and South-East 35  Atmospheric Administration NOAA database (http://www.nodc.noaa.gov/OC5/indprod.html), grouped per season, and used to calculate α V/L. A constant relative humidity factor (h) of 0.8 was used for the North Atlantic and for Mediterranean Sea, a value reasonably analogous to those recently reported in Pfahl 2014 36 since the local meteoric water line shows a deuterium excess comparable to the average precipitations on Earth. Consequently, α V/L values were corrected for disequilibrium processes occurring during evaporation from the sea, which tend to increase isotopic fractionation, according to the relationship 37 : Obtained values for δ V,0 ranged from − 13.76‰ (North Atlantic, winter) to − 12.16‰ (Mediterranean Sea, summer), comparable to what is reported in IAEA 2001 18,37 . Total number of particles N >0.5 . The total number of particles with a diameter larger 0.5 μ m (N >0.5 ) was measured with an optical particle counter (Grimm TM , Dust Monitor 1.108). Particles up to 40 μ m size are aspired through a heated sample inlet, dried and detected, even when activated as cloud condensation nuclei and part of hydrometeors or ice 38,39  Statistics. Statistical analyses presented here were done with PAST software version 2.17 44 and refined with the use of R software version 3.0.1. 45 . Parametric regression was done on logarithmic values of INPs as correction for normality to understand how much of the total variability was covered by our tests and R 2 values have been reported. These results are accompanied by non-parametric Spearman's correlation results (r coefficient and p values expressing the probability that variables are not correlated), as a more robust test for the significance of the relationships found. For the comparison among months for the values of INPs a Kruskal-Wallis test was done.