Regional new particle formation as modulators of cloud condensation nuclei and cloud droplet number in the eastern Mediterranean

A significant fraction of atmospheric particles that serve as cloud condensation nuclei (CCN) are thought to originate from the condensational growth of new particle formation (NPF) from the gas phase. Here, 7 years of continuous aerosol and meteorological measurements (June 2008 to May 2015) at a remote background site of the eastern Mediterranean were recorded and analyzed to assess the impact of NPF (of 162 episodes identified) on CCN and cloud droplet number concentration (CDNC) formation in the region. A new metric is introduced to quantitatively determine the initiation and duration of the influence of NPF on the CCN spectrum. NPF days were found to increase CCN concentrations (from 0.10 % to 1.00 % supersaturation) between 29 % and 77 %. Enhanced CCN concentrations from NPF are mostly observed, as expected, under low preexisting particle concentrations and occur in the afternoon, relatively later in the winter and autumn than in the summer. Potential impacts of NPF on cloud formation were quantified by introducing the observed aerosol size distributions and chemical composition into an established cloud droplet parameterization. We find that the supersaturations that develop are very low (ranging between 0.03 % and 0.27 %) for typical boundary layer dynamics (σw ∼ 0.3 m s−1) and NPF is found to enhance CDNC by a modest 13 %. This considerable contrast between CCN and CDNC response is in part from the different supersaturation levels considered, but also because supersaturation drops from increasing CCN because of water vapor competition effects during the process of droplet formation. The low cloud supersaturation further delays the appearance of NPF impacts on CDNC to clouds formed in the late evening and nighttime – which has important implications for the extent and types of indirect effects induced by NPF events. An analysis based on CCN concentrations using prescribed supersaturation can provide very different, even misleading, conclusions and should therefore be avoided. The proposed approach here offers a simple, yet highly effective way for a more realistic impact assessment of NPF events on cloud formation.


Introduction
Cloud condensation nuclei (CCN) and cloud droplet formation constitutes the direct microphysical link between aerosols and clouds. Quantifying how changes in aerosols affect global clouds, precipitation and climate is limited by the large number of processes and scales 45 that need to be captured in models (Stevens and Feingold, 2009;Pöschl et al., 2010;Seinfeld et al., 2016;Cecchini et al., 2017). New particle formation (NPF), the process during which new particles are formed directly from the gas-phase, is thought to significantly shape the distribution of CCN throughout the atmosphere (Pierce and Adams 2007;Westervelt et al., 2013;Gordon et al., 2017). Although initially too small (1-2 nm; Kerminen et al., 2012) to act 50 as CCN, particles from NPF can grow to sufficient size and hygroscopicity over a period of few hours to days and eventually act as efficient CCN.
Field studies have demonstrated substantial local enhancement in CCN number from NPF. For example, Wiedensohler et al. (2009) observed that the CCN size distribution was dominated by cloud outflow regions associated with convection (e.g. Hermann et al., 2003). NPF within marine boundary layers can strongly affect CCN concentrations at all cloud-relevant 75 supersaturations (Kalivitis et al., 2015;Kalkavouras et al., 2017). When these small particles however are mixed within the boundary layer, they may subsequently grow to CCN-relevant sizes, or even act as CCN in strongly convective clouds (Fan et al., 2013;Wang et al., 2016).
A thorough assessment of NPF impacts on CCN levels requires knowledge of all events and subsequent microphysical processing that occurred throughout the path of an air-mass. 80 Observationally, this is almost impossible to carry out; one can therefore only quantify the CCN concentration perturbation, or enhancement, above "background" levels that existed prior to an NPF event (Peng et al., 2014;Wu et al., 2015;Ma et al., 2016). Although conceptually straightforward, studies differ in the approach used to define the initiation of an NPF event (e.g., a strong enhancement in total particle number, the shape of the size distribution), the pre-event 85 CCN concentration (e.g., a 30-minute or 1 hour-average CCN concentration before the initiation time) and also the metric used to quantify the CCN enhancement from an NPF event (e.g., peak enhancement, a time-averaged enhancement, and the size defining the lower limit of CCN activation). Furthermore, observational studies quantify CCN enhancements from measurements of aerosol number size distribution; the link to CCN concentrations is done by 90 using a prescribed (or calculated) "critical diameter" above which all particles act as CCN in clouds. Studies widely vary in the approach used to determine dc, so additional considerations are required between assessments. Theoretically, dc depends on the level of supersaturation that develops in clouds and the chemical composition of the particles (Seinfeld and Pandis, 2006).
Often, dc is prescribed between 50 and 150 nm, corresponding roughly to clouds with maximum 95 saturation levels between 1.0%, and 0.1%, respectively (Kerminen et al. 2012). However, clouds are not characterized by a constant supersaturation, rather exhibit variable levels that instantaneously adjust to the intensity of cloud updrafts and the CCN spectra (Nenes and Seinfeld, 2003;Hudson et al., 2014). It is clear that all the above conventions need careful consideration, as they can affect the magnitude and duration of CCN enhancement for each 100 event. Asmi et al. (2011) at the Pallas GAW station in northern Finland estimated the contribution of NPF to CCN concentration. The method adopted was to subtract the concentration of particles larger than 80 nm diameter (N80) at the end of the NPF, from the average N80 before the NPF influence (defined from the time where the NPF started up to where the nucleation mode 105 particles reach 80 nm diameter). A similar approach was used to quantify the enhancement from NPF to particles larger than 50, and 100 nm (N50, N100, respectively). The relative enhancement of N50, N80, and N100 from NPF was 160±270%, 210±110%, and 50±130%, respectively. Furthermore, it is clear that the timing of the initiation of the NPF event and the subsequent growth of particles to CCN and eventually droplets is of utmost importance, as the time delay between the different processes actually limits the time during which the albedo of clouds is affected by NPF. In reality, the total contribution of nucleation process (including indirect effects) to a present-day net short-wave radiation in the atmosphere, depends on the rate in 130 which the emissions of gas-phase compounds responsible for nucleation and subsequent growth, as well as of primary particles acting as a sink for nucleated particles, throughout an NPF day.
Although most prior observation studies linked NPF to CCN number enhancement, very few of them actually link NPF to the process of cloud droplet formation and cloud droplet number. 135 The latter distinction is important, given that droplet number in clouds exhibit a sub-linear response to aerosol increases, owing to the elevated competition for water vapor and reduction in cloud supersaturation. The understanding of NPF impacts on CCN levels may therefore provide a biased view on its potential impact on droplet number (Nd) and the aerosol indirect effect. Using cloud droplet parameterizations to interpret observed aerosol size distribution 140 data, however, may allow one to address this issue. Kalkavouras et al. (2017) illustrated this issue by using a "conventional" approach to quantify CCN enhancement (with a critical diameter at which all particles act as CCN depended on observed composition and a prescribed supersaturation) and reporting much higher CCN number enhancements (~87%) for two sites in the eastern Mediterranean (Santorini and Finokalia)  (~12%) during two consecutive NPF episodes. The reason for this 8-fold discrepancy is in the drastically different supersaturation used to quantify CCN enhancement (0.2, 0.4, 0.6, and 0.8%) than what was computed for cloud droplet number (0.10 and 0.13% for updraft velocities of 0.3 m s -1 and of 0.6 m s -1 , respectively).
This study follows up on the initial work of Kalkavouras et al. (2017) and quantifies the impact 150 of NPF on CCN levels and cloud droplet number in the Eastern Mediterranean atmosphere over 7 years of continuous measurements (June 2008 to May 2015) of aerosol number size distributions and chemical composition. From this data, we aim to (i) quantify the seasonality and contribution of atmospheric NPF to the production of newly CCN in the eastern Mediterranean marine atmosphere, (ii) characterize the differences between nucleated particles 155 and their relative contribution to the CCN budget, and, (iii) investigate the NPF impacts on cloud droplet number concentration (Nd) and on maximum supersaturation (smax) formed in clouds in the vicinity of Finokalia. In the process of addressing these goals, we consider all the issues regarding the calculation of cloud supersaturation and event characteristics that affect the NPF impact calculations. (http://finokalia.chemistry.uoc.gr/), is located at the top of a hill over the coastline, in the northeast part of the island of Crete, facing the Aegean Sea in the wide north sector. Since the site was established in 1993, Finokalia experiences two characteristic periods during the year; the dry period from April to September, and the wet one from October to April. The dry period is dominated by strong winds of N/NW direction (up to 90%, originating from Central and 170 Eastern Europe and Balkans) of speed exceeding 10 m s -1 . The wet period is characterized by limited prevalence of the N/NW sector, and significant transport from Sahara (S/SW winds; occurrence up to 20%). An extensive description of the site and prevailing meteorology can be found in Mihalopoulos et al., 1997.

Aerosol composition and size distribution
175 Number size distribution of particles having mobility diameters from 9 to 848 nm (scanned range) were measured with a 5 min time resolution, using a custom-built scanning mobility particle sizer (SMPS; TROPOS-Type, Wiedensohler et al., 2012). The system is a closed-loop, with a 5:1 ratio between the aerosol and sheath flow, and it comprises a Kr-85 aerosol neutralizer (TSI 3077), a Hauke medium differential mobility analyzer (DMA), and a TSI-3772 180

Page 5
Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1254 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 11 December 2018 c Author(s) 2018. CC BY 4.0 License. condensation particle counter (CPC). The sampling was made through a PM10 sampling head and the sample humidity was regulated to a relative humidity below 40% using Nafion ® dryers in both the aerosol and sheath flow, and thereafter charged via Kr-85 neutralizer, and introduced into the DMA. By setting different voltages in the DMA, particles of different electrical mobility are selected and their particle number concentration can be measured. The fluctuation 185 of voltage yields an electrical particle mobility distribution, which can be inverted into a particle number size distribution. The recorded number size distributions were corrected for particle losses by diffusion on the various parts of the SMPS following the recommendations by Wiedensohler et al. (2012). Three different types of calibration were performed for the SMPS, namely DMA voltage supply calibration, aerosol and sheath flows calibrations and size 190 calibrations.
The complete dataset of particle size distributions was checked for the presence of NPF events, identified by a sudden of nucleation-mode particles (i.e. those with diameters below 25 nm), that lead to a continuous increase in larger particle concentrations over a short period of time (usually less than 4h). The NPF event progression is characterized by the relative changes of 195 the three particle modes, "nucleation" (diameter less than 25 nm), "Aitken" (diameter between 25 and 100 nm), and "accumulation" (diameter larger than 100 nm). The concentration of particles in each mode is obtained from each SMPS size distribution using an algorithm to parameterize each particle's mode with a multi log-normal distribution function (Hussein et al., 2005), as follows: (2) where ( ) is the aerosol number size distribution, ΔNi is its binned approximation from the SMPS data and i9, i100, i848 are the SMPS size bins that correspond to particles of 9, 100 and 848 205 nm, respectively. The upper and lower sizes are limits of size detection for the particular SMPS. and cations (K + , Na -, NH 4 + , Mg 2+ , Ca 2+ ), using the procedure of Bardouki et al. (2003).

Page 6
Atmos. Furthermore, the PM10 quartz filters were analyzed for organic and elemental carbon (Carbon Aerosol Analysis Lab Instrument, SUNSET Laboratory Inc.) using the EUSAAR 2 protocol of 215 analysis (Cavalli et al., 2010). For the estimation of the fine particulate matter fraction (PM1) chemical composition, the respective concentrations of sulfates, organics, and ammonium from the bulk PM10 are considered using the approach presented in Bougiatioti et al. (2009)

Cloud Condensation Nuclei (CCN)
Measurements of cloud condensation nuclei (CCN) concentration (cm -3 ) between 0.2 and 1.0% 225 supersaturation, were conducted using a Droplet Measurement Technologies (DMT) constant flow streamwise thermal-gradient CCN counter (CFSTGC; Roberts and Nenes, 2005), from November 2014 to May 2015. The CFSTGC is composed of a cylindrical diffusion chamber in which supersaturation is generated and controlled by the air flow rate, pressure, and a streamwise temperature gradient maintained by a heater and a set of thermoelectric coolers 230 (Roberts and Nenes, 2005;Lance et al., 2006). The air flow rate used was 0.5 L min -1 with a sheath-to-aerosol flow ratio of 10:1, and a top-bottom column difference, ΔΤ, between 4 and 15 K. Concentrations were measured at each supersaturation (0.2, 0.38, 0.52, 0.66, and 0.73%) for 15 min, yielding a CCN spectrum consisting of 5 different supersaturations approximately every hour. Calibration of the instrument supersaturation was performed by determining the 235 minimum diameter of classified ammonium sulfate aerosol that activates at given chamber flow rate, ΔT and chamber pressure, following the procedure of Bougiatioti et al. (2009). The CCN instrument was calibrated numerous times throughout the campaign. For the lower supersaturation, the relative variability between calibrations did not exceed 1%, whereas for the highest supersaturation the variability was under 4%. As CCN concentrations during the 240 measurement period rarely exceeded 5,000 cm -3 , no correction for water vapor depletion inside the CFTGC chamber was deemed necessary (Lathem and Nenes, 2011).

Calculation of CCN concentrations from size distribution data
As in numerous prior studies, CCN concentrations can be calculated from the observed number size distributions by integrating the SMPS data from a characteristic diameter dc to the largest 245 size particles measured:
where idc is the SMPS size bin that contains dc and 848 is the bin with the largest particles measured by the SMPS. Instead of prescribing dc (as done in other studies), we link it to a desired supersaturation level, sc, using κ-Köhler theory: 250 where Mw is the molar mass of water, σw is the surface tension of water, R is the universal gas constant, T is the temperature, and ρw is the density of water. CCN concentrations are then taken as being equal to the concentration of particles with diameter above dc (Kalkavouras et al., 2017). The aerosol hygroscopicity parameter, κ, is calculated assuming that it is a mixture of 255 an organic and inorganic component with volume fraction εorg, εinorg and characteristic hygroscopicity κorg, κinorg respectively. Past studies at Finokalia have suggested that assuming κorg=0.16 and κinorg=0.6 reproduce CCN to within 2% on average, but exhibit some size dependence (Bougiatioti et al., 2009;. For 4 NPF days during August and September 2012, the combined processing of the concurrent CCN and ACSM data during NPF events 260 provides the size-resolved κ (Fig. S1), which can be used to assess the validity of using a common κ for all sizes (supersaturations). For supersaturations below 0.2%, the size-resolved κ from the CCN data is higher by 23% compared to the bulk κ from the ACSM data, while for supersaturations between 0.2 and 0.4%, CCN-derived values agree quite well with bulk chemical composition data (slope 0.94), but with considerable scatter. For supersaturations 265 above 0.4% κ derived from the chemical composition data exhibits on average an overestimation bias of 38.5%. Altogether, the κ trends suggest that the composition of particles tends to increasingly deviate (or vary) from the bulk as they get smaller (i.e., with higher supersaturation)indication of enrichment by organics, often observed for NPF-derived particles (e.g., Cerully et al., 2011). The large scatter at around 0.4% supersaturation can be 270 attributed to chemical composition fluctuations, given that concentrations are affected by both the fresh organic-rich and aged sulfate-rich modes, more at least than found in the higher or lower supersaturation CCN. Overall however, this level of hygroscopicity error, is not expected to induce substantial errors in CCN concentration predictions, as demonstrated in the closure study below; a size-dependent consideration of hygroscopicity is therefore deemed 275 unnecessary.
We subsequently test the aforementioned approach for calculating CCN from chemical composition and size-distribution measurements (Eq. 4) against direct CCN measurements (Section 2.3) collected from September 2014 to March 2015. The degree of "CCN closure" is assessed with 5 minute-averaged data at 0.38, 0.52, 0.66, and 0.73% supersaturation (Fig.S2). 280 The measured values of CCN at each supersaturation correlate strongly with the predicted values, when considering all the available data. With increasing supersaturation, s, the value of  (Table S1). For the lowest supersaturations (0.38 and 0.52%), there is an overestimation (22%) of predicted CCN concentrationsconsistent with the fact that using bulk κ, which is higher than the "real" size-dependent κ, 285 would lead to slight overestimations in CCN. Interestingly enough, although these κ biases increase with decreasing size, the overestimation and scatter in CCN is decreased, for the higher supersaturations (0.66 and 0.73% -estimated and measured values agree within 10%) because an increasingly larger fraction of the aerosol activates so the error in absolute CCN number is diminished. Regardless of supersaturation, CCN prediction errors and scatter do not seem to 290 exceed 40%; these are considered minor, especially within the context of droplet number calculationsbecause the former exhibit a strongly sub-linear response to CCN changes in the eastern Mediterranean (e.g., Kalkavouras et al., 2017;Bougiatioti et al., 2016) which means that CCN errors translate to much smaller errors in CDNC.

Cloud droplet formation calculations 295
From knowledge of the aerosol hygroscopicity, size distribution and cloud vertical velocity, we can determine the droplet number (Nd) and maximum supersaturation for clouds forming in the vicinity of Finokalia, during all NPF events. Such calculations are useful to directly link aerosol with cloud droplet number in NPF-influenced clouds, and, determine the "cloud-relevant" supersaturations for which CCN perturbation calculations are relevant. For such calculations 300 we use the droplet parameterization based on the "population splitting concept" of Nenes and Seinfeld (2003), later improved by Fountoukis and Nenes (2005), Barahona et al. (2010), and Morales and Nenes (2014). These formulations provide a rapid and accurate calculation of droplet number that forms in cloud updrafts, and largely captures the droplet numbers that form in ambient clouds (e.g., Ghan et al., 2011;Morales-Betancourt et al., 2011). When calculating 305 Nd, the size distribution is described using a sectional representation (Nenes and Seinfeld, 2003) derived directly from the SMPS distribution data, similar to what was done in Kalkavouras et al. (2017). Observations of updraft velocity are not available at Finokalia for the time period sensitivity test also considers a more vigorous boundary layer (w = 0.6 m s -1 ), following Kalkavouras et al. (2017).

Page 9
Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1254 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 11 December 2018 c Author(s) 2018. CC BY 4.0 License. Furthermore, we determine the relative contribution of aerosol chemical composition, εκ, and aerosol number concentration, εNtotal, to variations in droplet number using a propagation of variance (Sullivan et al., 2016;Bougiatioti et al., 2016; (Bougiatioti et al., 2016;. The relative contribution of κ, and Ntotal to the Nd droplet number variation is estimated only during periods with high temporal resolution in chemical composition in order to capture the diurnal variability of κ (ACSM measurements, June 2012 to May 2015). were used for the analysis (Stein et al., 2015), initialized with meteorological conditions from GDAS (0.5 o resolution), were calculated at several heights (100, 500, and 1000 m above ground 335 level (a.g.l.)), with a duration of 48 hours. The back-trajectories are important for understanding the provenance of the different air masses and how they related to the occurrence and evolution of NPF events. Meteorological parameters, as wind speed and direction, temperature, relative humidity, and solar radiation were also continuously monitored during the study period, by the automatic weather station installed at Finokalia at 2 m a.g.l., and the time resolution for all of 340 the measurements was 5 minutes (http://finokalia.chemistry.uoc.gr/).

Aerosol chemical composition and hygroscopicity during NPF events
162 NPF episodes were recognized (Kalivitis et al., 2018) and the chemical composition of submicron particulate matter during these episodes was primarily composed of sulfate, 345 contributing on average by 39±8% to the total estimated PM1 mass as derived from the respective bulk PM10 24-h quartz fiber filters, and by 51±12% as derived from the ACSM highresolution measurements, respectively. Moreover, regarding the organic material the contribution was found to be in the order of 38±10% and 44±12%, respectively indicating that  Figure S3. It can be seen that κ tended to decrease throughout the early morning hours (6:00 to 9:00 LT) for each critical supersaturation probably owing to the downward 365 transport of secondary organic aerosol (SOA) during the boundary layer mixing, whilst at some point after noon, κ begun to augment probably linked to the formation of particulate sulfate during this period. As expected, lower supersaturation levels are associated with higher κ values, indicating that smaller particles were much less hygroscopic than larger ones, with the difference being of 0.2 κ units between the lower (under 0.2%) and the maximum 370 supersaturation (0.6-0.7%). This feature has been attributed to the enrichment of organic material in sub-100 nm particles (Kalivitis et al., 2015). The chemically-derived κ from the ACSM measurements generally does not present any remarkable fluctuation, and it seems to converge with the CCN-derived κ values of lower supersaturations. This constant character of the chemically-derived κ, may be an evidence that using prescribed levels of supersaturation or 375 critical diameters to calculate CCN concentrations can provide a biased influence of NPF events on CCN, since there is a clear dependence between the chemical composition and the size of a particle.

Characteristics and interpretation of the Finokalia NPF events
In all studies to date (summarized in the introduction), NPF impacts on CCN concentrations is 380 based on analysis of the evolution of the aerosol size distribution over time, to quantify i) how long it takes before freshly-formed particles in a given air-mass reach CCN-relevant sizes, and, ii) the degree to which CCN concentrations are augmented from the NPF. Here we present in detail the corresponding methodology used to interpret the NPF data from Finokalia, by applying to a "representative" type-I NPF event ( (Fig. 1). Regarding the intensity, it was a strong episode, since according to Zhang et al. (2004) Ntotal (where Ntotal is the number of particles larger than 9 nm) exceeds 10,000 cm -3 for at least 1h (in Fig. 3, Ntotal exhibits values exceeding 10,000 cm -3 between 10:00 to 15:30 LT). Subsequent growth of the aerosol generates a characteristic "banana shape" in the time-series of diurnal particle number concentration (Fig.  390   1a). The episode was characterized by a burst in particle number concentration in the 9 to 25 nm diameter range (nucleation mode), and enables a robust determination of the starting time (tstart) of the NPF event. Following Leino et al. (2016), we calculated half-hour median concentrations of the nucleation mode particles from the measurement data, since the half-hour median concentration was deemed sufficient to determine the tstart. When plotting the time series 395 of the intermediate nucleation mode particles, the NPF is distinctly visible as the particle concentrations rapidly increase from 3,850 to just over 17,000 cm -3 over a 2.5 h period starting at 8:30 LT (Fig 1b). The nucleation mode particles peak at 11:00 LT (see Fig. 1c), without any visible change in Aitken-mode concentrations until after 11:30 LT. This increase, in conjunction with the decrease of the nucleation mode particles in number, strongly suggests the 400 transfer of nucleation-mode to Aitken-mode particles from condensation and coagulation. The NPF event is said to terminate when the nucleation mode particles start to decrease. The appearance and formation of the nucleation mode particles are linked to the onset of solar radiation (Fig. 2). Afterwards, particles continued to grow in size for several hours, exceeding 100nm in diameter at 21:30 LT. Following the methodology of the mode-fitting (Hussein et al., 405 2004;Kulmala et al., 2012) the nucleation mode particles exhibited a growth rate of 3.7 nm h -1 , while the formation rate value of particles in the nucleation mode was 2.0 cm -3 s -1 (Kulmala et al., 2012), which are well in the range of the representative values reported by Kalivitis et al. (2018) at Finokalia site.
To quantify the impact of NPF on CCN concentrations, the following approach is used. From  in order to estimate the initiation of the influence on the potential CCN due to NPF, and is termed the "decoupling time", tdec. We determine tdec, and therefore the period (i.e. start and end) of intense NPF impact on CCN spectrum, based on the temporal evolution of the relative dispersion (RD) of the Rs for all supersaturations (Fig. 3b). RD was calculated by diving the standard deviation of the instantaneous values of Rs (at 0.38, 0.52, 0.66, 0.73, and 1.00% 430 supersaturation) with their average value. RD is useful, at it is highly sensitive to the introduction and evolution of particles from NPF as they transit the distribution over the resolved supersaturation range. It is said that NPF influences the CCN as long as the RD exceeds the envelope of (low) values seen during the initial stages of the NPF event. Indeed, from 08:30 to 13:30 LT, the RD is low (less than 0.1), and rapidly increases at 13:30 LT and Subsequently, we calculate the evolution of Rs before and after tdec for each supersaturation on 29 August 2012 (Fig. 3a). Specifically, "before" is the time period between tstart and tdec, whereas "after" is the period from the tdec until the end of CCN production (21:30 LT). This variation of 445 the Rs can be equivalent to the percentage contribution of CCN owing to NPF. The Rs was estimated to be 0.94±0.08, 1.02±0.09, 1.04±0.09, 1.03±0.09, and 0.99±0.08 prior to the starting of the CCN production (i.e. between 8:30 and 13:30 LT), and 1.09±0.60, 1.21±0.52, 1.25±0.43, 1.26±0.40, and 1.39±0.32 for 0.38, 0.52, 0.66, 0.73, and 1.0% supersaturation, respectively after 13:30 LT until the end of the production. The time intervals and tdec are driven by the processes 450 that affect the aerosol number distributions (i.e. coagulation and condensation), and hence affect the CCN population. Assuming a constant growth rate (3.7 nm h -1 ) for particles with diameter smaller than 100 nm, we approach the time which the new particles after the tstart are able to grow to the respective dc (35 to 67 nm for s 1.0 to 0.38%) and act as CCN. This time fluctuates from 2.7 to 10.5 h in the 1.0-0.38% supersaturation range, showing that larger particles (67 nm) 455 start to "feel" the influence from NPF late in the afternoon (19:00 LT). tdec is later for supersaturations below 0.7%, and this difference may occur due to the change of the growth

Page 13
Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1254 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 11 December 2018 c Author(s) 2018. CC BY 4.0 License. rate, which has been reported to increase with an increasing particle diameter (Paasonen et al., 2018). The Rs exhibits almost similar mean values after the tdec until 21:30 LT, for 0.38, 0.52, 0.66, and 0.73% supersaturation. Thus, the number of the newly-formed particles which reach 460 the CCN-size (dc varying from 43 to 67 nm) is independent from s, indicating that the newlyformed particles in this size range may exhibit similar chemical composition (internal mixture), or could be merely that particle number in the size range between 43-67 nm particles increased more or less to the same extent, after the tdec. The subsequent percentage contribution of NPF into CCN population at the same time period was calculated to be 16, 19, 20, 22, and 40% for 465 the above-mentioned supersaturations, respectively. The first four contributions are similar, since the dc spans from 67 nm for 0.38% to 43 nm for 0.73%, respectively. Regarding the s of 1.0%, the aerosol sizes are even smaller (~35 nm), and the contribution of NPF on CCN increases considerably. These contributions are suggestive of the convolution of NPF with condensational growth of both fresh and preexisting ("background") particles to produce CCN 470 size range particles, introducing a bias which can reach up to 50% regarding the exact activation of particles solely originating from the NPF. The amount of the "background" particles, which take place into the processes of activation from newly-formed particles to CCN, was calculated by subtracting the mean value of the concentration of particles in the nucleation mode from tstart until 11:30 LT (the formation of nucleation mode particles ceased - Fig. 1a) and the respective 475 mean value2 hours prior to the tstart.
The procedure outline in section 3.2 is repeated for the all the 161 remaining NPF episodes to determine the relative contribution of the NPF episodes to the Rs and subsequently to the CCN budget. The comprehensive results are presented in Table S2, and an extensive seasonal analysis in the Supplementary Material 3.3 (SM 3.3). Altogether, when considering all 162 NPF 480 episodes we found that, the average contribution of NPF to the CCN budget over eastern Mediterranean varied from 39 to 69% in the 0.38-1.0% supersaturation range, and displayed a seasonal variation (Fig. 4). In winter, tstart was observed during daytime (median 11:00 LT), followed by tdec 2.5 hours later. The contribution on CCN production due to atmospheric NPF and growth was estimated to be 47, 47, 48, 50, and 54% for 0.38, 0.52, 0.66, 0.73 and 1.0% 485 supersaturation (Fig. 4), respectively. For spring and summer, tstart exhibited a median value at 10:00 LT, and 9:00 LT, respectively, whilst the tdec was on average 2.5 hours after the tstart. The CCN production for associated with the nuclei growth to larger sizes increase by almost 40% for both seasons (Fig. 4), and for the aforementioned supersaturations. Finally, throughout autumn, tstart was detected in the morning (median 9:30 LT), followed by tdec on average 3.5 490 hours after the tstart, whereas the NPF episodes elevated the CCN numbers by 46, 47, 52, 55, and 69% (Fig. 4)   well as the starting of the CCN influence due to NPF (13:30 LT), the arrival of the air mass is 505 followed by a depression in Nd (relative mean decrease 7.9±2.9% for w=0.3 m s -1 and 13.5±3.9% for w= 0.6 m s -1 ). Concurrently, there is a slight increase in the maximum supersaturation (relative mean increases4.7±2.1% for w=0.3 m s -1 and 6.9±2.3% for w= 0.6 m s -1 ). Both trends are related to decreases in accumulation mode aerosol number, related to processes other than NPF (growth of the boundary layer, and dry deposition)as the latter has 510 not had the chance to influence particles that act as CCN in clouds. Nd exhibits the minimum value at 17:25 LT (Fig. 5) and corresponds to when droplet formation begins to "feel" the particles generated from NPF. Hereafter, this time will be expressed as tNd (Fig. 5). There is a time lag between tdec and tNd, since particles formed in an NPF event need sufficient time to grow into CCN-relevant sizes, and subsequently into a cloud droplet. After tNd, smax is negatively 515 correlated with Nd for both updraft velocities, owing to the increasing competition for water vapor from the growing number of CCN. For both updraft velocities, the increase of Nd until the sunset (around 21:30 LT) was similar and on the order of 21.9±6.5%, leading to a simultaneous decrease of smax by 11.8±2.7%. Water vapor competition effects can be assessed by comparing Nd at sunset with the CCN for smax at tNd (where competition effects from the 520 NPF-generated particles are vanishingly small). Using this approach, we find that competition effects suppress Nd by 20% for w=0.3 m s -1 and 12.3% for w=0.3 m s -1 . It is worth noting that, if smax did not vary over the period of Nd influence, the increase of Nd from the tNd until 21:30 LT was similar for both w and merely of 5.5±2.5%, since the competition for water vapor is restricted considerably. This clearly shows that the prescription of a constant supersaturation in 525 the CCN analysis may lead to biased results regarding the impact of NPF on regional clouds.

Impact of NPF on droplet number and cloud formation
Since Nd does not increase significantly until midnight, it is clear that most of the impact of the NPF is on nocturnal clouds, which carries important implications for the formation of drizzle and structure of the boundary layer in the following day.
The degree to which Ntotal and κ variations influences Nd variability can be expressed by 530 calculating the relative contribution of the total aerosol number, and the hygroscopicity to the droplet number using the equations (4), (5), and (6) in section 2.5. The results are displayed in Table S4. We find that Nd varies from tNd to midnight by 30 cm -3 for w equal to 0.3 m s -1 , and 35 cm -3 for w equal to 0.6 m s -1 . 68% of this variance can be attributed to aerosol number and the remaining 32% to changes the chemical composition. The above procedure, when carried 535 out for the 161 remaining NPF episodes, provides consistently similar results (Results depicted in Table S3)

Summary and Conclusions
The aerosol particle number size distributions along with chemical composition and 540 meteorological parameters were studied at a remote background site in the Eastern Mediterranean over a 7-year period in order to quantify how regional new particle formation (NPF) events modulate the concentration of aerosol, cloud condensation nuclei (CCN), droplet number and maximum supersaturation developed in clouds of the region.
Overall, 162 NPF episodes were recorded with the majority occurring during spring and 545 summer (32 and 30.8%, respectively), few during winter (14.8%) and the rest (22.4%) during autumn. The timing and duration of NPF influences on the CCN spectrum and cloud droplet number were accurately determined using a set of new statistical metrics derived from the observational data. Wintertime NPF events were found to start around 11:00 LT and begin affecting the CCN spectrum 3 hours into the event, while in springtime were initiated one hour 550 earlier and increased CCN concentrations 2.5 hours into the event. During summer, recorded NPF events started the earliest (9:30 LT) and the impact on CCN concentrations occurred roughly 2 h after, while in autumn NPF episodes occurred between 9:30 and 10:00 LT, but with the largest delays in observing CCN impacts -3h 30 min after the start of the event. Overall, when accounting for all NPF episodes, we found that the average increase on CCN levels (0.38-555 1.0% supersaturation) from NPF over eastern Mediterranean ranged from 37 to 69%.
When the observed size distributions and chemical composition are used in conjunction with a cloud droplet parameterization, the impact of NPF on Nd differs considerably from the CCNbased analysis. Regardless of season, we find that the maximum supersaturation developed in typical boundary layer clouds (updraft velocities of the order of 0.3 m s -1 ) vary between 0.07% 560 and 0.12%, giving cloud droplet number increases of 7% to 13%. This 4 to 10-fold decrease in Nd sensitivity to NPF (compared to what is deduced from the CCN analysis) is primary from the actual cloud supersaturation being much lower than the prescribed levels in the CCN analysis. Nd sensitivity to NPF however is further reduced during the evolution of NPF events

Page 16
Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-1254 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 11 December 2018 c Author(s) 2018. CC BY 4.0 License. owing to their increased competition for water vapor when forming cloud droplets (the droplet 565 response can be suppressed by almost 1/5 compared to assuming constant supersaturation throughout the NPF). The lowest impact on Nd is observed during summer, as this season exhibits the highest aerosol concentrations prior to NPF events -that either act as CCN or grow to become so during an event. Pre-existing particles have been estimated to contribute up to 50% of the activated CCN during summer, denoting the importance of background conditions. 570 A striking consequence of the low cloud supersaturations is that NPF impacts on Nd are observed much later in the event, typically in the late afternoon (after 16:00 LT), and that Nd is relatively insensitive to increases in CCN during the course of an event owing to the competition effects for water vapor. Thus, the impacts of NPF events on eastern Mediterranean clouds occur during the late evening and nighttime. Although such Nd enhancements may limit 575 the short-term impact of NPF on shortwave cloud forcingit may reduce cloud drizzle and promote stabilization of the marine boundary layer with potentially important implications for the overall boundary layer structure (e.g., Rosenfeld et al., 2006) in days following NPF events.
Perhaps one of the most important findings of this study is the importance of constraining the levels of supersaturation that are generated in ambient clouds, and the diurnal characteristics of 580 the influence during NPF events. Choosing prescribed levels of supersaturation or diameters to define CCN concentrations can provide substantially biased or incomplete insights on the influence of NPF events on regional clouds, the hydrological cycle and climate. The approach presented here offers a simple and highly effective paradigm for quantifying the potential impacts of NPF events on clouds, with tools available to interested researchers upon request. 585

Author contributions
AK, NK and NM contributed measurements. AN and AB conceived the study and developed the analysis tools, AN, AB, PK and NM carried out the analysis and wrote the paper. All authors commented on the manuscript.

Acknowledgments 590
This research is co-financed by Greece and the European Union     and calculated cloud droplet number concentrations (Nd) (left axis) for updraft velocities of w= 0.3 m s -1 (bottom), and w=0.6 m s -1 (top) during the "representative" new particle formation event captured at Finokalia on 29 August 2012. The size of the circles corresponds to the number concentration of Nd, while tdec is the "decoupling time" (13:30 LT), and tNd is the time when the number of droplets start to "feel" the NPF (17:25 LT), according to the approach described in the main text.