The diurnal and seasonal variability of ice nucleating particles at the High Altitude Station Jungfraujoch (3580 m a.s.l.), Switzerland

Cloud radiative properties, cloud lifetime, and precipitation initiation are strongly influenced by the cloud phase. Between ∼ 235 and 273 K, ice nucleating particles (INPs) are responsible for the initial phase transition from the liquid to the ice phase in cloud hydrometeors. This study analyzes immersion-mode INP concentrations measured at 243 K at the High Altitude Research Station Jungfraujoch (3580 m a.s.l.) between February 2020 and January 2021, thereby presenting the longest continuous, high-resolution (20 min) data set of online INP measurements to date. The high time resolution and 5 continuity allow to study the seasonal and the diurnal variability of INPs. After exclusion of special events, like Saharan dust events (SDEs), we found a seasonal cycle of INPs, highest in April (median in spring 3.1 INP std L−1), followed by summer (median: 1.6 INP std L−1) and lowest in fall and winter (median: 0.5 INP std L−1 and 0.7 INP std L−1, respectively). Pollen or subpollen particles were deemed unlikely to be responsible for elevated INP concentrations in spring and summer, as periods with high pollen loads from nearby measurement stations do not coincide with the periods of high INP concentrations. 10 Furthermore, for days when the site was purely in the free troposphere (FT), no diurnal cycle in INP concentrations was observed, while days with boundary layer intrusions (BLI) showed a diurnal cycle. The seasonal and diurnal variability of INPs during periods excluding SDEs is with a factor of 7 and 3.3, respectively, significantly lower than the overall variability observed in INP concentration including SDEs of more than three orders of magnitude, when peak values result from SDEs. The median INP concentration over the analyzed 12 months was 1.2 INP std L−1 for FT periods excluding SDEs, and 1.4 INP 15 std L−1 for both FT and BLI, and incl. SDEs, reflecting that despite SDEs showing strong but comparatively brief INP signals, they have a minor impact on the observed annual median INP concentration.


Introduction
The ratio of ice crystals to liquid droplets in a cloud strongly determines its radiative properties, lifetime, and precipitation initiation (e.g., Lau and Wu, 2003;Lohmann and Feichter, 2005;Hoose and Möhler, 2012;Murray et al., 2012;Mülmenstädt (nephelometer, Airphoton IN101) as well as meteorological standard parameters (e.g., ambient temperature, relative humidity, atmospheric pressure, wind speed and direction). Meteorological standard parameters for the JFJ, precipitation rates, and pollen concentrations for other stations were obtained from the IDAWEB interface of MeteoSwiss (https://gate.meteoswiss.ch/idaweb, last accessed April 27, 2021).

INP measurements
INP concentrations are measured using an automated continuous flow diffusion chamber, the automated Horizontal Ice Nucle-95 ation Chamber (HINC-Auto, . HINC-Auto sampled ambient air and measured the INP concentration at a center lamina temperature of T = 243.15 K and at a supersaturated saturation ratio with respect to water of S w = 1.04 between February 7, 2020 and January 31, 2021, in unites of INP std L −1 (per standard liter, normalized to T = 273.15 K and an atmospheric pressure of p = 1013.25 hPa). Ambient air was sampled via a heated (T = 293.15 K) total aerosol inlet (Weingartner et al., 1999), which feeds also the other aerosol measurements at the JFJ. Before entering HINC-Auto, the sampled air 100 is dried using a diffusion dryer (S w ≤0.008 at 20 • C, filled with 4 Å-molecular sieve). The sampling volume rate was set to 0.283 std L min −1 . The sampling interval was 20 min and consisted of sampling ambient air (15 min) and particle-free ambient air (5 min) between each measurement. In HINC-Auto, false-positive counts can arise due to frost breaking off the chamber walls. They are present during the measurement of ambient, as well as particle-free air. During the 5 min before and after an ambient air measurement, the number of false-positive frost particles are counted and time-proportionally subtracted from the ambient air measurement to yield the background-corrected ambient INP concentrations. Because of Poisson statistics it is likely that during the time-normalized particle-free measurements more or fewer false-positive counts arise than during the time-normalized ambient air measurement. This leads to a variable systematic bias of the background-corrected ambient INP concentrations for each sampling interval even if the true atmospheric INP concentration was to remain constant. The standard deviation of the resulting probability density function corresponds to the stated counting uncertainty of ± 1 σ, provided after 110 each stated INP concentration in this work, and is on median ± 1.37 INP std L −1 . The bias in background correction can lead to negative reported values of background-corrected ambient INP concentrations whenever the true INP concentration is close to or below the chamber background. These negative readings are retained in the data set. In addition, applying a moving average over 3 measurements greatly reduces the number of reported negative INP concentrations. For HINC-Auto, the counting uncertainty is identical to the limit of detection (LOD). See  for a more detailed description of 115 HINC-Auto and the derivation of the INP concentrations.
The measured INP concentrations are here statistically described by the median and 25 th to 75 th percentiles (Q 25% -Q 75% ).
We assume atmospheric INP concentrations away from sources to be log-normally distributed, as proposed generally for atmospheric pollutants by Ott (1990) and supported by other studies (e.g., Welti et al., 2018;Schrod et al., 2020). Therefore, an adequate statistical description would be the log-mean and log-standard-deviation. However, measurements reporting a 120 negative INP concentration don't allow to include all data when calculating the logarithm. For log-normally distributed data without any skewness, the log-mean is identical to the median. Hence, we chose to report the median. https://doi.org/10.5194/acp-2021-710 Preprint. Discussion started: 12 October 2021 c Author(s) 2021. CC BY 4.0 License.
Frequent construction work at the JFJ during the observation period caused intermittent interference from local pollution (Bukowiecki et al., 2021). Therefore, the INP measurements had to be filtered (75.5% remain after filtering). High frequency fluctuations in the total particle concentrations were observed during periods with pollution from the construction site. These 125 fluctuations in the CPC and the ≥ 0.3 µm -optical particle counter channel of HINC-Auto were used to obtain unpolluted INP measurements (N = 15843). A more detailed description of the data filtering process can be found in Brunner et al. (2021, in review The distinction between undisturbed FT air masses and FT affected by boundary layer intrusions, hereafter simply referred to as BLI conditions, is made according to Brunner et al. (2021, in review). It is based on a combined criterion considering both the 222-Radon concentration ( 222 Rn, Griffiths et al., 2014, first factor in equation (1)) and the total number concentration of particles with diameters larger than d ≥ 90 nm (N 90 , Herrmann et al., 2015, second factor in equation (1)).Subsequently, the probability (P FT ) of the sampled air to represent undisturbed FT air mass is obtained using: where PDF FT and PDF BLI are the probability density functions with FT or BLI log-normal fit parameters, respectively, k is the slope factor to capture the seasonality as discussed later, here k = 0.1, and N 90,TH is the N 90 threshold midpoint, here N 90,TH = 120 cm −3 . PDF FT and PDF BLI are inferred from from long-term radon measurements at the JFJ between January 1, 2009 and December 31, 2020, where the frequency distribution of the logarithm of the 222 Rn concentrations, has a bimodal shape 140 with two discernible but overlapping modes. The two normal distributions were fitted to this frequency distribution, where the modes at lower and higher concentrations are assumed to represent undisturbed FT and BLI conditions, respectively. The resulting probability density functions for 222 Rn concentrations of FT and BLI air masses are according to: where the log-normal fit parameters are µ FT = -0.139 Bq/std m 3 , σ FT = 0.239 Bq/std m 3 for PDF FT , and µ BLI = 0.403 Bq/std 145 m 3 , σ BLI = 0.238 Bq/std m 3 for PDF BLI . Herrmann et al. (2015) showed, depending on the seasons at the JFJ, for values of N 90 ≥ 100 -150 cm −3 to be only present during BLI conditions. N 90 values below this threshold do not exclude BLI conditions. Therefore, the second factor of equation (1) forces P FT to low probabilities whenever N 90 exceeds the defined threshold, while below the threshold P FT becomes identical to using the first factor of equation (1) ( 222 Rn term) only. A more detailed description can be found in Brunner et al. (2021, in review). 222 Rn is measured at the JFJ according to Griffiths et al. 150 (2014) and N 90 retrieved from SMPS measurements. The temporal resolution of the 222 Rn measurements is one every 30 min., and one every 6 min. for the N 90 measurements. In the total period analyzed, 31 SDEs were recorded with a total duration of 55 days and 20 hours. Periods without a SDE 165 signal are labeled as background periods (BG = total -SDE). BG periods are further divided into FT and BLI periods (FT BG and BLI BG , correspondingly). The conditions for a positive SDE-signal and further information can be found in Brunner et al. (2021, in review).

Results
The observed seasonal and diurnal variability of INP concentrations are discussed in the following sections. First, the total 170 investigated period is brought into context with previous seasons with regard to standard meteorological parameters and the type of air mass present at the site. Then, the seasonal signature of different air masses is analyzed. Finally, the observed diurnal variability is presented for different air masses.

The seasonal INP variability at the JFJ
During February 2020 -January 2021, the mean temperature at the JFJ was -5.9°C, representing the warmest period on record 175 since 1933 and 2.0°C warmer than February 1933 -January 1971. On February 10, the storm "Sabine" led to wind speeds of up to 54 m/s, which corresponds to the 99.6 th percentile of all daily maximum wind measurements at the JFJ. Yet, this is still well below the record wind velocity of 74 m/s, measured on January 6, 1998. The site was 40% less in clouds (where a relative humidity (RH) greater than 96% is considered as in-cloud) compared to previous years (2020-2021: 16.1% vs. 1970-2019: 26.6%). Note that the prevalence of in-cloud conditions based on RH is underestimating the true prevalence of in-cloud 180 conditions, but relative changes are well captured (e.g., see Herrmann et al., 2015). There were two exceptionally dry period intrusions often go along with ozone concentrations >70 ppb (Stohl et al., 2000)). During the late March/early April period, 185 air masses were largely advected from high northern latitudes. With 38%, the JFJ experienced 5% more FT periods in relative terms than in the 2008-2019 period (36%). Figure 1 shows a more detailed representation of the air masses present during the analyzed period compared to historic observations. The fractions of days purely in FT, of days with a mix of FT and BLI, and of days only in BLI air masses for February 2020 -Jan 2021 were overall representative for the fractions observed during previous years, as they rarely exceed the climatological Q 25% -Q 75% range. May and August 2020 had more days purely in 190 the FT at the expense of days only with BLI for May. February and November 2020 showed a significantly higher fraction of mixed air masses while there were fewer days purely with BLI.  and January 2021. Overall Saharan dust contributed to 74.7 ± 0.2% of the total INPs observed at the sampling conditions at the JFJ (see Brunner et al., 2021, in review). BG concentrations in June were lower compared to spring and July/August. Dividing BG periods into FT BG and BLI BG shows similar distributions in both cases for most of the investigated period. Median and Q 75% concentrations in April, July, and -to a lesser extent -in August were higher during BLI BG than in FT BG air masses.

205
Independent of the air mass, April showed the highest BG INP concentrations. In contrast to observations of INP active at warmer temperatures Schneider et al., 2020), no correlation between the ambient temperature and the INP concentrations is evident in our data (Spearman's rank correlation coefficient = 0.149, R 2 = 0.012). We hypothesize that because the investigated ice-activation temperature in our study (243 K) is distinctively colder than the ambient temperature were four times higher than before this period, but remained at the same level or increased further afterwards, suggesting for the high INP concentrations in April are not connected to the mentioned dry period. Multiple species of pollen and subpollen particles were found to be ice-active in previous laboratory studies, specifically birch, juniper, pine, orchard grass and redtop grass (e.g., Gute and Abbatt, 2020, and references in Table 1). Note, juniper belongs to the cypress family. Given their seasonality, could pollen or subpollen particles be responsible for the distinctively  Also for artificial measurements with higher prescribed INP concentrations, the median and mean agree well with the set prescribed INP concentration, however, the shape of the distribution follows a normal and not a log-normal distribution. The seasonal and overall frequency distributions, in contrast, show a log-normal distribution, as expected by theory (Ott, 1990) and discussed more in Brunner et al. (2021, in review). Comparing the artificial signals with the measurements, the total (SDE+BG) true median INP concentration in winter was likely between 0 and 1 INP std L −1 . This is supported by the median    with studies at different locations (e.g., Schrod et al., 2020). Comparing the frequency distribution of the measurements from this work to Lacher et al. (2018) emphasises the difference in observed concentrations (see Fig. A2), and proves that the nine single field campaigns in the mentioned earlier work, all targeted to sample SDEs between mid-January and the beginning of March and between May and August in the years 2014-2017, were successful in probing an over-proportional fraction of SDEs. This is further supported by the INP frequency distributions, which for Lacher et al. (2018) are not log-normal as expected 310 compared to those of the present work. Overall, the seasonality has a minor impact on the observed INP concentrations, which is consistent with other work (e.g., Tobo et al., 2020;Schrod et al., 2020). This statement holds for FT conditions, was well as within the PBL. The seasonal INP number concentrations vary by a factor of up to 7 for identical statistical metrics, e.g., when comparing the median concentrations, thus, well below the variation observed within all measurements or compared to SDEs.

BG:
Overall: -0.9 ± 1.0 1.4 ± 1.0 4.7 ± 1.2 15.1 ± 1.6 40.4% FTBG: -1.0 ± 1.0 1.2 ± 1.0 4.2 ± 1.1 11.7 ± 1.4 100.0% BLIBG: -0.9 ± 1.5 1.5 ± 1. To study the diurnal variability of INPs at the JFJ, phase-statistics of the BG INP concentrations with a cycle period of 24 h were calculated for the total investigated period, starting at midnight of each day. These phase-statistics were divided into periods when the full day was in FT air masses (30 days), days not exclusively in FT or BLI (244 days), or days only with BLI (91 days), as shown in Figure 5 with statistics in Table 2. Phase-statistics of the total particle number concentrations during BG periods are superimposed (total particle concentrations from CPC measurements without SDE periods). For days only 320 in the FT, there is no clear diurnal cycle in INP number concentrations evident, while the median BG particle concentration   Continuous, sub-hourly measurements of the ambient INP concentration enable to statistically study the behavior of INPs during repeating meteorological events. Such long-term measurements were absent so far. In this work, continuous online INP measurements at T = 243.15 K and S w = 1.04 at the JFJ between February 7, 2020 and January 31, 2021 were analyzed 360 with regard to their seasonality and diurnal cycle. We found a seasonal cycle, highest in spring with a median of 3.1 INP std L −1 , followed by summer (median: 1.6 INP std L −1 ), and lowest in fall and winter (median: 0.5 std L −1 and 0.7 INP std L −1 , respectively), all in absence of SDEs. This is consistent with the seasonality observed in other studies at the same site Lacher et al., 2018) and at different sites (Wex et al., 2019;Tobo et al., 2020;Schneider et al., 2020). Here, INP concentrations show a larger seasonal dependency than the total particle concentrations. We hypothesize 365 this to be an effect of the different seasonality of the partitioning types of particles, e.g., mineral dust for INPs at 243 K vs.
biological and anthropogenic particles for the total aerosol concentration. A positive correlation between ambient temperature and the INP concentrations was non-existent, in contrast to earlier studies for INPs active at warmer temperatures. Based on our observations, it is unlikely that pollen or subpollen particles are responsible for the observed high background INP concentrations in April, although during peak periods their contribution cannot be ruled out up to 19.9 INP std L −1 to the 370 overall INP population at T = 243.15 K and S w = 1.04. The seasonal quartile concentrations vary by a factor of up to 7 for identical statistical metrics, which is much smaller than the observed variation due to special events, e.g., SDEs, which can cover three orders of magnitudes. No diurnal INP cycle was found for days purely in the FT, indicating that sinks and sources of INPs in the FT are either far away from the JFJ or do not follow a diurnal pattern. Atmospheric ageing, for example, which potentially makes atmospheric particles ice-active, is either a slow, (ultra violet-) light dependent process or happens only on 375 local scale. For days with a mix of FT and BLI or for days entirely with BLI, a diurnal cycle similar to the diurnal cycle in total particle concentration was found, yet more pronounced in the case of the diurnal INP variability. Limitations were faced concerning the counting uncertainty of the instrument in combination with the low ambient INP concentrations, hindering the study of INP concentrations at warmer temperatures. By using an aerosol concentrator, future measurements can be extended to warmer sampling temperatures, e.g., for INPs active at 248 K. While this study covered almost a full year, future studies 380 over multiple years to decades can help to fill knowledge gaps in spatiotemporal variability of INPs. The investigation of the interannual variability and trends could, for example, provide some insight whether the observed seasonality is linked to other parameters or how the anthropogenic land-use change and desertification (Ginoux et al., 2012) affect the INP number concentrations in the atmosphere.
Data availability. The data presented in this publication will be made available at DOI: xx.xxxx/ethz. Note by authors: data will be made For technical support and fabrication, we would like to thank Dr. Michael Rösch and Marco Vecellio, whose expertise greatly helped to improve the instrumentation.