Non-methane hydrocarbons (NMHCs) and OVOCs were measured employing complemental online and offline techniques described in the following. The inlets were about 3 m above ground level (a.g.l.). Table 1 resumes the characteristics of the methods carried out during the campaign and indicates a list of the monitored VOCs.

At a time resolution of 30 min, 20 VOCs, including C2 to C10 anthropogenic VOCs and C10 BVOCs, were measured using two automated gas chromatographs (GCs, Chromatotec, Saint-Antoine, France) fitted out with a flame ionization detector (FID). A detailed description of both instruments (ChromaTrap and AirmoVOC), along with the sampling setup, technical information (pre-concentration, desorption–heating times, type of traps, column types) and the calibration procedure were given in Debevec et al. (2017). Very satisfactory detection limits (as 3σ of the baseline) were found with values below 104 and 17 ppt for ChromaTrap and for AirmoVOC, respectively. Relative uncertainties of VOCs measured with the ChromaTrap analyzer typically ranged from 14 % (ethane) to 73 % (propene) and from 18 % (benzene) to 53 % (o-xylene) for VOCs measured with the AirmoVOC (Debevec et al., 2017). Note that the two GCs were deployed at CAO from January 2015 to February 2016, allowing for direct comparisons of BVOC levels recorded in March 2015 to summertime values.

Additional VOCs were measured at a time resolution of 10 min using an online high-sensitivity proton transfer reaction – quadrupole mass spectrometer (PTR-QMS, Ionicon Analytik GmbH, Innsbruck, Austria; Lindinger et al., 1998), which allowed the detection of protonated OVOCs (alcohols, aldehydes, ketones and carboxylic acids), aromatics (sum of C8 and C9) and BVOCs (e.g., isoprene and the sum of monoterpenes). This instrument has been extensively described in recent reviews (Blake et al., 2009 and references therein) and a description of the analytical setting implemented here and calibration procedure were given in Debevec et al. (2017). The detection limit of the 16 protonated compounds typically ranged from 11 to 203 ppt, and relative uncertainty was evaluated between 18 % and 44 % (Debevec et al., 2017). Note that nighttime isoprene concentrations discussed in Sect. 3.1 and 3.2 could be due to other compound fragments such as 2-methyl-3-buten-2-ol (MBO).

Additionally, more than 400 offline 3 h integrated air samples were collected on sorbent cartridges – multi-sorbent and DNPH (2,4-dinitrophenylhydrazine) cartridges – using an automatic clean room sampling system (ACROSS, TERA Environment, Crolles, France). C1–C16 organic compounds were sampled for 3 h via a 0.635 cm diameter 4 m length PFA line and then trapped into one of the two types of cartridges: a multi-sorbent cartridge composed of carbopack C (200 mg) and carbopack B (200 mg) (carbotrap 202, Perkin-Elmer, Wellesley, Massachusetts, USA) and a Sep-Pak DNPH-Silica cartridge (Waters Corporation, Milford, Massachusetts, USA). These techniques are described in Detournay et al. (2011), and their setup in the field is further presented in Detournay et al. (2013) and Ait-Helal et al. (2014). Briefly, 39 C5–C16 NMHCs, including alkanes, alkenes, aromatics and nine BVOCs, along with six C6–C11 n-aldehydes, were sampled at a flow rate of 200 mL min-1 on the multi-sorbent cartridges that were preliminarily conditioned during 24 h with purified air at 350 ∘C and 10 mL min-1 flow rate, using a RTA oven (French acronym for “régénérateur d'adsorbant thermique” – TERA Environment, Crolles, France). A total of 10 additional C1–C8 carbonyl compounds were sampled in parallel with the DNPH cartridges at a flow rate of 1.5 L min-1. During the sampling, different ozone scrubbers have been used in order to avoid any possible ozonolysis of the monitored compounds: a MnO2 ozone scrubber was employed for the multi-sorbent cartridges, while a KI ozone scrubber was installed upstream of the DNPH cartridges. In addition, stainless-steel particle filters of 2 µm diameter porosity (Swagelok) were used to prevent any sampling of particles. Samples were later analyzed in the laboratory by GC-FID (with TurboMatrix 650 ATD, Perkin-Elmer, Wellesley, USA; for the multi-sorbent cartridges) or high-performance liquid chromatography coupling with ultra violet detection (HPLC-UV; for the DNPH cartridges). The reproducibility of the analysis was checked regularly by the analysis of a standard, leading to the plotting of a control chart for each compound, which allowed the reproducibility of each instrument to be checked. The detection limit of the VOCs measured with offline techniques was typically below 5 ppt for the multi-sorbent cartridges and ranged from 6 to 27 ppt for the DNPH cartridges. Relative uncertainty was evaluated between 3 % and 26 % for the multi-sorbent cartridges and between 11 % and 37 % for the DNPH cartridges (Ait-Helal et al., 2014).

α-Pinene and β-pinene measured by both online GC-FID and offline techniques were selected to cross-check the quality of the results recorded during the campaign. Online measurements were additionally averaged on a 3 h timescale to allow direct comparison with offline measurements and reported in Fig. S1 in the Supplement. α-Pinene showed a better determination coefficient than β-pinene (r2 of 0.69 and 0.47 for α-pinene and β-pinene, respectively) and a slope closer to 1 (1.08 for α-pinene and 0.67 for β-pinene). The results from α-pinene and β-pinene investigated in this paper were taken from the GC-FID measurements due to a higher time resolution and a better analytical performance of AirmoVOC. Additionally, the sum of eight monoterpenes collected by multi-sorbent cartridges was also used in comparison to the non-speciated monoterpenes measured by PTR-MS (Fig. S1), yielding to similar variability and consistent ranges of concentrations (r2: 0.73; slope: 0.79).

Concerning OVOCs, acetaldehyde, acetone and methyl ethyl ketone (MEK) were monitored by both PTR-MS and an offline technique. According to Fig. S2, these OVOCs showed good determination coefficients (r2 of 0.81, 0.90 and 0.84 for acetaldehyde, acetone and MEK, respectively) with slopes close to 1 for each compound (1.16, 0.87 and 1.04 for acetaldehyde, acetone and MEK, respectively) and relatively low intercepts (77 ppt for acetaldehyde, 86 ppt for acetone and 9 ppt for MEK). Acetaldehyde, acetone and MEK measurements presented in the following are those which resulted from PTR-MS by reason of a finer time resolution.

As a consequence, recovery of the different techniques, frequent quality checks and an uncertainty determination approach have allowed us to assure a satisfying robustness of the dataset, and cross-check comparisons have shown comparable results for the different techniques used (within the range of uncertainties).

A large set of real-time atmospheric measurements was performed by the DLI at the CAO, in order to characterize trace gases (NO, NO2, O3, CO and SO2). These latter are presented in more detail by Kleanthous et al. (2014). The time resolution was 5 min for each analyzer. The results examined in this study are the hourly averages.

Particle size distribution measurements were performed using a setup of a custom-made differential mobility particle sizer (DMPS, TSI Inc., model 3080; Villani et al., 2008) completed by a particle size magnifier (PSM, Airmodus, model A09; Vanhanen et al., 2011). The DMPS consists of a bipolar charger to charge the aerosol particle population to the equilibrium charge distribution, a 28 cm differential mobility analyzer (DMA) in a closed sheath-air loop and a condensation particle counter (CPC, TSI Inc., model 3010). This instrument was operated to measure the aerosol size distribution over 20–200 nm diameter size range from 8 to 11 March and over the 10–250 nm size range from 12 to 27 March, with a time resolution of 460 s. NDPMS was used in this paper to refer to total number concentrations of particles obtained by integrating the DMPS measurements. Total number concentrations of particles larger than 1 nm diameter (NPSM) were measured with a PSM using diethylene glycol (DEG) as the working fluid at a fluid flow rate of 1 standard liter per minute. A PSM can grow particles as small as 1 nm to larger than 90 nm, after which a CPC is used to count the grown particles. Considering its time resolution (1 s), PSM data were filtered from local pollution spikes due to the local anthropogenic activity on the CAO site (between 07:00 and 17:00 LT during weekdays). Finally, the particle cluster and sub-10 nm particle (between 1 and 10 nm) concentrations were calculated as the difference between the total particle concentration derived from the DMPS and the PSM concentrations (NPSM–NDPMS).

The charged cluster size distributions were recorded with a neutral cluster and air ion spectrometer (NAIS). This spectrometer is a modified version of the AIS instrument (Airel Ltd, Mirme et al., 2007; Mirme and Mirme, 2013) which is an instrument capable of measuring mobility distributions of sub-3 nm charged aerosol particles and clusters. Controlled charging, together with the electrostatic filtering, enables it to additionally measure the neutral aerosol particles' distribution. The measurement principle of the NAIS is based on two independent spectrometer columns, one of each polarity, where the ions are classified by a DMA. More details are given in Manninen et al. (2011). The mobility range is 3.2–0.0013 cm2 V-1 s-1, corresponding to particle Milikan diameter between 0.5 and 50 nm.

The chemical composition of non-refractory submicron aerosol (NR-PM1) has been continuously monitored by deploying a quadrupole aerosol chemical speciation monitor (Q-ACSM, Aerodyne Research Inc., Billerica, Massachusetts, USA), which has been fully characterized by Ng et al. (2011). This instrument shares the same general structure with the aerosol mass spectrometer (AMS) except that it has been specifically intended for long-term monitoring purposes. The Q-ACSM instrument was operating continuously with 30 min time resolution during the whole duration of the campaign totaling 1292 valid data points (corresponding to a time recovery of 95 %). The ACSM dataset was validated by comparison with co-located PM1 chemical composition results obtained by integrated daily (24 h) time resolution filter-based measurements. Instrument settings, field operation, calibration and data processing are those reported in Petit et al. (2015).

Black carbon (BC) was calculated using the 880 nm channel of a seven-wavelength (370, 470, 520, 590, 660, 880 and 950 nm) aethalometer (AE31 model, Magee Scientific Corporation, Berkeley, CA, USA) with a time resolution of 5 min. Presuming difference in the absorption Ångström exponent between fossil fuel and biomass-burning-derived aerosol, the BC originating from these two sources was apportioned following the method described by Sandradewi et al. (2008).

Meteorological parameters (temperature, pressure, relative humidity, wind speed, wind direction and radiation) were monitored every 5 min using a weather station (Campbell Scientific Europe, Antony, France) located on the rooftop of the CAO building, at approximately 5 m a.g.l. Additionally, planetary boundary layer (PBL) assimilated data were generated by the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim global atmospheric reanalysis at the location corresponding to the Troodos station (34.92∘ N, 32.88∘ E; ∼20 km west of the CAO station). The ERA-Interim model, setup and dataset are detailed in Sect. S1 in the Supplement. Even if these assimilated PBL data were not provided for the CAO station but for the Troodos one, they were only used in this study to qualitatively investigate PBL height effect on BVOC concentration levels and variations.

Classification of air mass origins has been based on the analysis of the retroplumes computed by the FLEXible PARTicle (FLEXPART) Lagrangian model (Stohl et al., 2005) considering CAO as the receptor site. The FLEXPART model simulates trajectories of user-defined ensembles of particles released from three-dimensional boxes. The classification was based on hourly resolution model simulations going back in time to 5 days, taking into account only the lowest 100 m a.g.l. (footprint plots), even if the 3 km was modeled. These backward retroplumes were classified within eight source regions, similar to Kleanthous et al. (2014), identified by a custom-made algorithm combined with visual inspection. The source region map is depicted in Fig. 2 based on the residence time of particles over each source region. During March 2015, the CAO station was mostly under the influence of continental air masses originating from southwest Asia (cluster 7 – 31 %), northwest Asia (cluster 4 – 28 %), west of Turkey (cluster 5 – 10 %) and Europe (cluster 3 – 11 %) together with by marine air masses (cluster 2 – 14 %). Note that air masses categorized as “local” (cluster 0) occurred only on 23 and 24 March, and may rather be considered as a transitory state between periods of air masses originating from northwest Asia and west of Turkey. It is worth noting that March 2015 was characterized by an unusually high contribution of southwest Asian air masses in the detriment of European air masses compared to the period 1997–2012 investigated in Kleanthous et al. (2014).

The most relevant variables for identifying NPF events are the formation rate (Ji, expressed in cm-3 s-1) at a given diameter (i, in nm) and the growth rate (GR, in nm h-1), which is defined as the diameter rate of change due to particle population growth. The growth rate between two size classes was calculated considering the method defined by Hirsikko et al. (2005) which is based on the time corresponding to the maximum concentration in each size class of the selected size range by fitting a normal distribution to the size class concentration.

Formation rates were especially evaluated for the very first steps of the formation process, i.e., between 1 and 3 nm (Kontkanen et al., 2017). As previously mentioned, the PSM was measuring in a total mode during the studied period, which did not allow any growth and nucleation rate calculation at such diameters. The total particle formation rate was thus calculated at 3 nm (J3) from the total particle concentration measured in the size range 2–4 nm by the NAIS (N2-4), by using the growth rate in the size range 1.5–3 nm (GR1.5-3, in nm h-1), and the loss of particles by coagulation scavenging of 3 nm particles on larger pre-existing particles (CoagS3, in s-1) both derived also from the NAIS measurements. The growth rate of the corresponding size range is then obtained by a linear least square fit through the time values previously found. The total particle formation rate at 3 nm was finally calculated according to Eq. (1), from Kulmala et al. (2012): J3=dN2-4dt+CoagS3⋅N2-4+12nmGR1.5-3⋅N2-4.

CS denotes the ability of the particle size distribution to remove condensable vapor from the atmosphere and hence describes the loss rate of the condensable vapors onto the pre-existing particles (Pirjola et al., 1999). This variable is proportional to the surface area density of an aerosol particle and has been calculated based on size distribution measured with DMPS as proposed by Kulmala et al. (2001).

SO2 produces ambient H2SO4, which is currently thought to be the most likely nucleation precursor candidate as well as contribute to the growth of newly formed particles (Kulmala et al., 2013; Sipilä et al., 2010). To study the connection between NPF and H2SO4, an empirical proxy for H2SO4 concentration was calculated from the SO2 concentration according to Eq. (2), adapted from Mikkonen et al. (2011) which is based on previous work by Petäjä et al. (2009): H2SO4calc=2.468×10-11⋅GlobRad⋅SO21.385CS⋅RH1.03, where GlobRad is the global radiation in W m-2, [SO2] is the sulfur dioxide concentration in molec cm-3, CS is the condensation sink in s-1, and RH is the relative humidity. The coefficients used in Eq. (2) were calculated from sulfuric acid measured with a CIMS instrument (Sellegri et al., 2016). This proxy was constructed for radiation higher than 10 W m-2 but the predictive ability is significantly raised for radiation exceeding 50 W m-2 (Rose et al., 2015).

Nine BVOCs, namely α-/β-pinenes, α-/γ-terpinenes, limonene, myrcene, camphene, 3-carene and isoprene, have been detected and quantified at the CAO, and their mean levels during March 2015 are presented in Fig. 3. Statistical analysis, uncertainties and detection limits of the BVOC measurements are presented in Table S1 in the Supplement. The average concentration of the sum of terpenoids during March 2015 was 282±307 ppt. Among BVOCs monitored during the intensive field campaign, the most abundant were monoterpenes. The average concentration of monoterpenes during the intensive field campaign was 236±294 ppt with a maximum up to 4500 ppt (recorded during the night of 10 March). Monoterpenes exhibited high daily amplitude, with a mean mixing ratio of 154 ppt during the daylight hours against 329 ppt during the nighttime hours (Fig. 3). Higher concentrations of monoterpenes (estimated by the concentrations of the sum of α-pinene and β-pinene) were observed during the summertime (307 ppt on average; Fig. 3) but maximum concentrations were at the same order of magnitude (e.g., a peak up to 3600 ppt was recorded during the night of 31 July – not shown here). The dominant monoterpenes observed during the field campaign were β-pinene (61±142 ppt) and α-pinene (58±131 ppt), followed by limonene (27 ppt), camphene (25 ppt), Δ3-carene (11 ppt), myrcene (6 ppt), α-terpinene (3 ppt) and γ-terpinene (below 1 ppt). α-Pinene and β-pinene accounted together for 62 % of the total monoterpene concentration. Average concentration of isoprene was quite low (46 ppt) in March 2015 but it was higher by a factor of 3 in the summertime (Fig. 3) due to higher temperatures. As a matter of fact, isoprene, α-pinene and β-pinene are the major BVOCs emitted by the Mediterranean vegetation (Owen et al., 2001). In addition to their high emission rates by vegetation, they are the least reactive isoprenoids with OH radicals and ozone (Atkinson and Arey, 2003) and therefore tend to accumulate (for short periods) in the atmosphere. Other more reactive compounds, such as α-terpinene and limonene, are removed very quickly following their emission, thus exhibiting lower concentrations in the atmosphere.

An overview of BVOC concentrations at different background locations in the Mediterranean is depicted in Fig. 3. As for CAO, Cape Corsica and Finokalia are representative remote sites with Mediterranean shrublands, and primary BVOC concentrations recorded at these sites have similar seasonal behaviors and concentration levels. Speciated monoterpenes measured at remote/rural Mediterranean sites were predominantly composed of α-pinene and β-pinene with a higher proportion of camphene and 3-carene observed only at CAO. Contrarily, α-terpinene was observed in higher proportion during the summer field campaign performed at Cape Corsica.

As shown in Fig. 4, the diurnal variations of isoprene and monoterpenes present opposite diurnal evolution. Daily amplitude is of 317 ppt on average for monoterpenes, with a decrease during daytime hours, a significant increase after sunset (17:00–18:00 LT), high concentrations throughout the night and decrease after sunrise (06:00 LT). The monoterpene average diurnal patterns indicated that their emissions were solely dependent on temperature (Geron et al., 2000a and references therein) and lower, but still significant, emissions occurred throughout the night. A similar pattern with nighttime maxima has been observed at other locations in the Mediterranean (e.g., in Portugal by Cerqueira et al., 2003; in France by Detournay et al., 2013; in Italy by Kalabokas et al., 1997 and Davison et al., 2009; and in Greece by Harrison et al., 2001) and was assigned to nocturnal emissions of monoterpenes stored in the understory vegetation (Niinemets et al., 2004; Schurgers et al., 2009). These nighttime maxima are enhanced by the low removal processes (i.e., low oxidizing species concentrations) and the shallow nocturnal boundary layer which concentrate close to ground level the monoterpenes emitted by vegetation. Furthermore, the prevalent nocturnal winds at CAO (originating from the southwestern to southeastern sectors) may have also contributed to these nighttime maxima with air masses enriched with biogenic emissions from the forests located in the Troodos Mountains (Debevec et al., 2017; Galvin, 2014). The major vegetation types covering these mountains are pine forests, composed of Calabrian pines (Pinus brutia – from foothills to the high mountains up to 1200 m) and black pines (Pinus nigra – on the highest peaks at altitudes from 1400 to 1951 m); and oak forests, mostly composed of golden oaks (Quercus alnifolia – found with Pinus brutia or in inland maquis between an altitude of about 800 and 1500 m) and Kermes oaks (Quercus coccifera – up to 1400 m elevation) (Fall, 2012 and references therein). These coniferous species are considered as very strong emitters of monoterpenes (Aydin et al., 2014a, b). More specifically, oaks are usually classified as predominantly isoprene emitters (Helmig et al., 2013), whereas Quercus coccifera are considered as significant emitters of α-pinene and β-pinene in Owen et al. (2001). A considerable fraction of the monoterpene emission from pines originates from large storage structures (Niinemets et al., 2004) and, as such, they continue to emit at a higher relative level during the nighttime compared to oaks as long as nocturnal temperatures remain sufficiently high (Laothawornkitkul et al., 2009; Owen et al., 1997).

With daily amplitude of isoprene of 43 ppt on average, the observed isoprene pattern followed an usual diel profile controlled by temperature and solar radiation (Geron et al., 2000b; Owen et al., 1997). As depicted in Fig. 4, isoprene concentrations started to increase immediately at sunrise (06:00 LT), indicative of local biogenic emissions. However, isoprene concentrations did not decrease immediately at sunset (17:00–18:00 LT) but remained rather constant at the beginning of the night (considering the upper end of the whiskers and mean values of hourly box plots depicted in Fig. 4) and followed a slow decrease until reaching a minima (03:00 LT). Isoprene levels showed their minimum levels during the night, although levels up to 200 ppt could be noticed during the night of 10 March and coincided with the highest concentrations of monoterpenes recorded during the field campaign. This finding suggests that air masses were enriched with biogenic emissions, the main contributors of monoterpenes, which were also partially isoprene emitters as reported previously (Detournay et al., 2013). This finding is also in agreement with the source apportionment reported in Debevec et al. (2017). Two VOC biogenic sources were identified; they were both composed of different primary biogenic species (pinenes and isoprene for factor 1 and factor 2, respectively) and showed distinct temporal variabilities and geographic origins supporting the division of BVOC sources into two factors. Biogenic factor 1, driven by pinene emissions, also explained a small proportion of isoprene (15 %; Debevec et al., 2017). Nonetheless, the contribution of MBO to the isoprene signal of PTR-QMS m/z 69 cannot be discarded (Karl et al., 2012; Kim et al., 2010). Isoprene usually dominates over most other BVOCs in many places and these interferences are often shown to be minor (Karl et al., 2004; Misztal et al., 2011; Warneke et al., 2010). However, measurements, particularly in coniferous ecosystems, can be of greater analytical challenge due to the concomitant emission of isoprene and MBO (Kim et al., 2010; Schade and Goldstein, 2001). The emissions of MBO could require light as isoprene (Harley et al., 1998) or could be mainly temperature dependent (Hellén et al., 2018; Tarvainen et al., 2005). Isoprene and/or MBO that were emitted during the late afternoon could not be fully oxidized photochemically, as OH concentrations begin to fall, and could remain in the nighttime atmosphere.

The days during the measurement period (i.e., 8–27 March) were classified with respect to whether or not NPF was observed. The NPF event days were identified using PSM and DMPS measurements and based on the criteria and methodology reported by Dal Maso et al. (2005). Briefly, a day was classified as an NPF event if (1) a clear increase in the fine particulate mode was observed, followed by (2) a sustained growth for at least a couple of hours until it reached a relevant particle size to form cloud condensation nuclei (CCN), resulting in a well-known “banana shape” as depicted in Fig. 8. Out of 20 observation days, such events were observed on 14 days (i.e., 8–10, 14, 16–18, 20–23 and 25–27 March) during daytime. The beginning of each event corresponds to the time when a new nucleation mode appeared, and its end is defined as the stop of the subsequent growth. On most NPF event days (8–10, 14, 16, 18, 23 and 25–27 March), we observed a steep increase in particle number (i.e., NDMPS increased up to 25 000 particles cm-3) initiated during the morning. On each NPF event day, a clear increase of the nanoparticle concentration, calculated as the difference between NPSM and NDMPS, was observed prior to the NDMPS number concentration increase. This indicates that the NPF events observed at CAO are initiated in the vicinity of the measurement site at the same time as at the regional scale, as the growth of these clusters are measured continuously over several hours as they are transported a further distance from the measurement site. On the 17, 20 and 22 March, cluster concentration increases were observed, indicating that nucleation was occurring in the local environment, but they were not followed by newly formed particle growth and hence not observed at the regional scale. The remaining 6 days out of the 20 observation days were classified as non-event days (i.e., 11–13, 15, 19 and 24 March).

The event days were classified further into subclasses (Ia, Ib, II and apple) according to the classification proposed by Yli-Juuti et al. (2009) based on previous work by Hirsikko et al. (2007) and Vana et al. (2008). This classification depends on the event applicability to growth and formation rate analysis. Class I represents the days when the formation and growth rate are determined with a good confidence level. Class I is divided into Class Ia and Ib. The Class Ia event (14, 18 and 23 March) has clear and strong particle formation with little or no pre-existing particles, while a Class Ib event is any other Class I event (8–10, 20 and 25 March) where the particle formation and growth rate can still be determined. The Class II event (16–17 and 22 March) represents the events where the accuracy of formation rate calculation is questionable due to data fluctuation even though the banana shapes are still observable. Class apple event (21 and 26–27 March) refers to an increase in the fine particulate mode, but newly formed particles do not show clear growth, and hence there is a clear gap between the new mode and other modes. Note that particulate formation and growth rates discussed in this study were only calculated for Ia and Ib events.

To investigate the factors governing daytime NPF processes, it was decided to partition the NPF days into four classes (NPF1–NPF4) based on prevailing atmospheric conditions. More precisely, this classification was based on the transport pathways of the air masses arriving at the CAO site along with mean daily concentrations of atmospheric parameters presented in Table 2. This table also presents mean daily values of property indicators for NPF events (i.e., CS, particulate formation and growth rates) for event and non-event days. Figure 9 depicts diurnal cycles of particle number NPSM and NDMPS and accumulated diel variations of PM1 contributions. Figure 10 shows diurnal cycles of parameters with suspected influence on NPF, averaged over event and non-event days. Note that no clear trend was observed on 21 March. In addition to this statistical vision of the results, time series of particle number NPSM and NDMPS, CS, PM1 and suspected parameters controlling NPF events are provided in Fig. S4.

NPF1 event days (i.e., 18 and 25–27 March) and NPF2 event days (i.e., 8–10 and 23 March) mainly concerned class I and apple type events. The CAO station may receive pollution of local/regional origins since winds during NPF1 and NPF2 event days were mainly from northeastern, eastern and southeastern directions, and the station received air masses originating from southwest and northwest Asia (Fig. S4). In the course of these NPF event days and especially for NPF2 ones, high concentrations of PM1, HOA, BC and NO2 (Table 2), which are recognized as tracers of anthropogenic sources, suggested that NPF events were of anthropogenic origin. High concentrations of inorganic compounds (SO4, NO3 and NH4; Table 2) were also observed, suggesting the air masses sampled during NPF1 and NPF2 events were both polluted and aged. Additionally, higher BVOC concentrations (i.e., isoprene, MVK + MACR and monoterpenes; Table 2) were observed during NPF2 event days than the ones during NPF1. Consequently, NPF1 event days were characterized mainly by anthropogenic origin, while NPF2 event days were of mixed origins (anthropogenic and biogenic).

NPF3 event days (i.e., 14, 16–17 and 22 March) and NPF4 event days (i.e., 20 March) concerned all class II events and some class I events. During these NPF event days, the CAO station received winds which were mainly from northwestern, western and southeastern directions and the station was mainly influenced by maritime air masses (Fig. S4) and continental ones which have not been newly in contact with anthropogenic sources. This finding was confirmed by the low anthropogenic tracer concentrations observed during these NPF days which were similar to the ones observed during non-event days (Table 2). Moreover, BC did not exceed 0.5 µgm-3 (Fig. S4), supporting the consideration of atmospheric conditions as clean conditions according to Cusack et al. (2013). Higher isoprene concentrations than the ones characterizing non-event days were only noticed during NPF4 events even if MVK + MACR mean concentrations of NPF4 event days were slightly lower than the ones observed on the non-event days (Table 2). As a result, NPF3 event days were characterized mainly by marine origin, while NPF4 event days were probably of biogenic origin. Note that PM1 concentrations during NPF4 event days were of the same range as the ones recorded during NPF event days of anthropogenic origin (Table 2) due to higher concentrations of inorganic compounds (i.e., SO4 and NH4) and a higher contribution of low-volatility oxygen-like organic aerosol (LV-OOA) to OM concentration. These findings suggest highly processed (aged) regional background pollution transported to the receptor site during NPF4 event days.

The intensity of global solar radiation was higher on event days (275 W m-2 on average; Table 2) than on non-event days (203 W m-2). This finding is consistent with the literature (Cavalli et al., 2006; Cusack et al., 2013; Guo et al., 2008; Hamed et al., 2007). Solar radiation is known as an important parameter in the initial step of atmospheric nucleation, since photochemical reactions among various chemicals were facilitated by stronger solar radiation, leading to the production of the nucleating and/or condensing species involved in NPF (Harrison et al., 2000).

The hourly average RH followed the opposite temporal pattern of that of the intensity of global radiation (Fig. 10). The RH was hence lower on event days (61 % on average; Table 2) than on non-event days (80 %) as observed elsewhere (Boy and Kulmala, 2002; Guo et al., 2012; Hamed et al., 2007). This could be partly explained by the fact that lower RH days usually have fewer clouds causing more solar radiation and subsequently producing more OH radicals to form more condensable vapors (Hamed et al., 2007). Another possible reason could result from an increase in particle hydration with increasing RH, which leads to larger pre-existing particle surface areas and thereby larger condensation sinks, resulting in a possible inhibition of the nucleation (Birmili et al., 2003; Hamed et al., 2011).

Temperatures were also higher on event days (i.e., NPF1-NPF3: 13.8 ∘C on average; Table 2) than on non-event days (11.2 ∘C) and especially during NPF1 event days (15.4 ∘C). As for RH, higher temperatures might be linked to higher solar radiation associated with higher photochemistry, and it is difficult to separate the two effects with our datasets. Higher temperatures have been associated with the nucleation events in Germany (Birmili et al., 2003), in Italy (Hamed et al., 2007) and in Atlanta (Woo et al., 2001), which may be induced by higher BVOC emissions (Guo et al., 2008). Contrarily, lower temperatures have been associated with the nucleation events in Finland (Boy and Kulmala, 2002; Vehkamäki et al., 2004) and in Hong Kong (Guo et al., 2012). These later findings could be due to lower temperatures at the start time of the NPF events which may enhance the nucleation of H2SO4 with water vapor (Guo et al., 2012).

NPF1 and NPF2 event days occurred at CAO at high levels of SO2 (both at 0.7 ppb on average and up to 2.2 ppb; Table 2 and Fig. 10) compared to SO2 ones on non-event days (0.2 ppb on average and up to 0.4 ppb), suggesting its implication in nucleation formation. Conditions seem to be even more favorable for H2SO4 production during NPF2 event days since higher H2SO4 concentrations were observed on NPF2 event days (1.4×108 molec cm-3 on average; Table 2) than on NPF1 event days (6.3×107 molec cm-3). NPF1 and NPF2 also occurred at high CS (up to 0.4 s-1; Fig. S4) compared to non-event days (up to 0.2 s-1). A high condensation sink might suppress the particle formation and the growth of the newly formed particles as a substantial fraction of the vapors that could be condensing on the larger particles. Furthermore, the condensation sink is proportional to the coagulation sink of nucleation-mode particles on pre-existing particles. Therefore, at high CS, a high growth rate is required for the newly formed particles to survive and grow to larger sizes instead of being scavenged by coagulation (Kulmala et al., 2005). Thus, for NPF1 and NPF2, a high condensable source rate had to compensate for the high CS.

What further distinguished NPF2 from NPF1 was that isoprene concentrations were particularly high during NPF2 event days (Table 2 and Fig. 10) than its concentrations during non-event days due to more favorable temperatures (Fig. 10). Similar diurnal variations were also observed between isoprene, temperature and NDMPS-PSM during NPF2 event days (Figs. 9 and 10), suggesting that isoprene and H2SO4 can both play a role during NPF2 event days. Contrarily, isoprene concentrations during NPF1 event days were similar to those observed during non-event days which would suggest that NPF1 was mainly induced by H2SO4.

Slightly lower PM1 concentrations were measured during NPF1 event days (9.7 µgm-3; Table 2) compared to PM1 concentrations during NPF2 event days (12.9 µgm-3) since lower OM concentrations were observed on NPF1 event days (4.3 µgm-3 vs. 6.8 µgm-3 during NPF1 and NPF2 event days, respectively; Table 2), while similar concentrations of SO4 and NH4 were noticed during these NPF event days (2.9–1.9 and 3.3–2.1µgm-3 for SO4–NH4 concentrations during NPF1 and NPF2 event days, respectively). Moreover, semi-volatile oxygen-like organic aerosol (SV-OOA) contributed to OM more intensively only during NPF2 event days (up to 86 %; Debevec et al., 2017), while LV-OOA contributions were in the same range during NPF1 and NPF2 event days (1.7–1.8µgm-3; Table 2). As a conclusion, isoprene contribution during NPF2 event days coincided with higher PM1 concentrations due to a higher contribution of SV-OOA.

NPF3 and NPF4 event days occurred at low levels of SO2 (0.3 and 0.2 ppb for NPF3 event days and NPF4 event days, respectively) inducing low H2SO4 concentrations (Table 2; Fig. 10) that could be not sufficient to initiate nucleation formation. These NPF events started at low CS (0.05 s-1; Fig. 10), highlighting a lower uptake of species by condensation and favoring nucleation processes under clean atmospheric conditions. Note that daily CS observed during NPF4 event days was high compared to CS during NPF3 event days and non-event days (Table 2; Fig. 10) but CS was below 0.1 s-1 when an NPF4 event was initiated as depicted in Fig. 10. Additionally, the increase in CS followed the increase in NDMPS-PSM, suggesting that the CS time variation was mainly driven by the growth by condensation of newly formed particles.

A steep increase in particle number NDMPS-PSM up to 70 000 particles cm-3 was recorded between 08:00 and 10:00 LT on 20 March (i.e., NPF4 event day) when solar radiation and temperature were observed to be intense. Isoprene concentrations increased from 06:00 LT on 20 March and remained relatively high (60–110 ppt) during the morning, suggesting a role in NPF formation. Contrarily, isoprene concentrations recorded during NPF3 event days were lower than NPF4 event days and non-event days, suggesting that a marine source could be involved in NPF3 events.

On 23 March (mixed NPF2 event type), the NPF event occurred at similar levels of biogenic tracer (isoprene concentrations between 60 and 110 ppt; Fig. S4) than the ones observed on 20 March (biogenic NPF4 event type), while higher mean particulate formation and growth rates were met during the selected NPF2 event (J3: 8.97 cm-3 s-1 – GR1.5-3: 3.18 nm h-1) compared to the mean rates characterizing the NPF4 event day (J3: 8.13 cm-3 s-1 – GR1.5-3: 1.93 nm h-1). These findings suggest polluted air mixed with high concentrations of biogenic compounds induced more intense particulate formation and faster growth.

Additionally, on 23 March during the NPF2 event, concentrations of H2SO4 (below 0.2×108 molec cm-3) were close to the ones observed on 27 March (anthropogenic NPF1 event type). Characterized as an apple event type, the particulate formation and growth rates were not calculated for the NPF1 event on 27 March and were hence not compared with the ones associated with NPF event on 23 March. Note that higher strength (i.e., maximum particle cluster number concentration) was noticed for the selected NPF2 event day (maxima NPSM-DMPS of 63 000 cm-3 on 23 March and of 41 000 cm-3 on 27 March). Moreover, at higher H2SO4 concentrations than ones observed on 23 March, NPF1 event days have shown contrasted particulate formation rates (J3: 5.00–49.60 cm-3 s-1 on 18–25 March, respectively) and both higher mean growth rate and mean CS (GR1.5-3: 4.75 nm h-1 – CS: 0.12 s-1) than the mean ones associated with NPF2 event days (GR1.5-3: 3.70 nm h-1 – CS: 0.09 s-1). It seems that polluted air masses, observed at the receptor site during NPF1 event days, were characterized by a high amount of condensable species involved in the particle growth, permitting to overcome the increased CS and allowing also a fast growth. At higher H2SO4 and BVOC concentrations, NPF2 event days occurring on 8–10 March have shown higher particulate formation rates than the one of the NPF event on 23 March (J3: 12.23 cm-3 s-1 on average ±5.62 cm-3 s-1 on 8–10  March and J3: 8.97 cm-3 s-1 on 20 March). This finding again would confirm polluted air mixed with high concentrations of biogenic compounds can induce more intense particulate formation.

As a result of these comparisons, chemical or photochemical reactions involving biogenic and anthropogenic species form new compounds which may be involved in nucleation. Several field studies have found an enhancement of biogenic SOA under the influence of anthropogenic emissions (e.g., Carlton et al., 2010; Shilling et al., 2013). The enhancement can be a result of increased gas-to-particle partitioning, increased oxidant concentrations or a change in the reaction pathways (Hoyle et al., 2011; Kanakidou et al., 2000). Laboratory experiments have also shown higher SOA formation levels in mixtures of VOCs compared to a single VOC (Ahlberg et al., 2017; Flores et al., 2014). Ahlberg et al. (2017) even found that isoprene did not produce much SOA mass in single VOC experiments but contributed to the mass in the cases of VOC mixtures.

As a summary, NPF can occur at various condensational sinks and both under polluted and clean atmospheric conditions. Some NPF events can occur at CAO at low SO2 concentrations and low CS under clean atmospheric conditions. High calculated H2SO4 concentrations coupled with high BVOC concentrations seem to be one of the most favorable conditions to observe NPF at CAO in March 2015. Relatively high particulate formation and growth rates were associated with NPF event days of mixed origins, suggesting an intense particulate formation and a fast growth. Higher strength was noticed for an NPF2 event day under mixed influence (anthropogenic and biogenic – 23 March) than the ones observed both during NPF1 and NPF4 event days, under anthropogenic and biogenic origins, respectively, for the same levels of precursors (anthropogenic and biogenic, respectively), suggesting combination of biogenic and anthropogenic species form new compounds which may be involved in nucleation. The next part of this section is focused on 8–10 March (mixed NPF event type) to better understand how the interaction of BVOC species with anthropogenic compounds can initiate nucleation and contribute to early growth of nucleated particles.

Firstly, we will focus our discussion on the behavior of selected parameters during NPF events observed between 8 and 10 March (represented by yellow periods in Fig. 11), corresponding to the NPF2-type events. The three successive NPF events were all initiated at 08:00 LT occurring around 2 h after sunrise when isoprene concentrations started to increase in agreement with temperature and Sun radiation. Daytime variation (06:00–17:00 LT) of NPSM-DMPS was consistent with the MVK + MACR one, which could represent the oxidized species producing new particles by nucleation. Maximal NPSM-DMPS (between 30 000 and 40 000 cm-3) were observed at 11:00–12:00 LT in agreement with daily maximal solar radiation and isoprene concentrations (between 100 and 160 ppt). However, daily maximal NPSM-DMPS decreased from day to day, while slightly higher maximal isoprene daily concentrations were noticed on 9 and 10 March compared to 8 March. This finding might be linked with higher CS recorded at the beginning of NPF events on 9 and 10 March compared to 8 March, which could reduce formation of new particles. As stated before, SO2 and H2SO4 concentrations in the 8–10 March period were among the highest ones observed during the campaign (Figs. 10–11). H2SO4 and SO2 concentrations started to increase at 09:00 LT on 8 March and at 10:00 LT on 9 March, which also correspond to the increase in NPSM-DMPS concentrations (Fig. 11), suggesting that H2SO4 also play a role in NPF. Additionally, daily maximal H2SO4 and SO2 concentrations also decreased from 8 to 10 March, as did particle cluster concentrations. Monoterpene concentrations remained low during these NPF events (below 400 ppt) but higher concentrations were observed at night, suggesting their possible contribution in the growth of newly formed particles. Contrarily, monoterpene oxidation products were shown to produce new particles by nucleation more efficiently than the isoprene oxidation products (Bonn et al., 2014; Spracklen et al., 2008). This finding could suggest that isoprene alone may not contribute to particle nucleation, while isoprene combined with anthropogenic species can be involved in nucleation. Additionally, oxidation products of monoterpenes, such as pinonaldehyde or nopinone, may nucleate and condense at an early stage of the new particle formation (Sellegri et al., 2005b).

As a summary, during these specific NPF event days of the campaign, nucleation-mode particles may be formed by the combination of high H2SO4 and isoprene oxidation product concentrations under favorable meteorological conditions (high temperature and solar radiation and low relative humidity; Fig. 5), resulting in an increase in SV-/LV-OOA contributions. A similar trend was observed between CS and isoprene concentrations during NPF events only, suggesting that these compounds do not only contribute to the nucleation mode for particles but also to aerosol growth directly after nucleation.

The focus is now shifted to the variability of selected parameters during nighttime succeeding these NPF events (periods represented by blue color in Fig. 11) partly to highlight the role of monoterpenes. These nighttime periods were characterized by relatively high CS (up to 0.16 s-1), nocturnal temperatures among the highest ones observed during the campaign and relative humidity between 50 % and 100 %. Note that high isoprene (and/or MBO) concentrations have still been observed even a few hours after sunset consistent with temperature variation (Sect. 3.1.2) and could also play a role during nighttime.

Blue periods in Fig. 11 highlight periods characterized by a clear increase in SV-OOA contributions, while inorganic aerosols concentrations and LV-OOA contributions remained stable. The increase in SV-OOA contributions occurred at high isoprene concentrations, sometimes at high monoterpene concentrations and at high CS, in favor of BVOC condensation onto pre-existing particles. Moreover, after 19:00 LT on 10 March, the highest monoterpene concentrations observed during the campaign (up to 4 ppb) coincided with an elevation of SV-OOA contributions up to 7.3 µgm-3, consistent with the fact that oxidation products of monoterpenes are known to contribute to particle growth (Birmili et al., 2003).

As a summary, BVOCs observed at night at CAO potentially play a role in particle growth by condensing onto pre-newly formed aerosols and significantly influence levels and variations mainly of SV-OOA. The relationship between BVOCs and OA stated in Debevec et al. (2017) was hence confirmed, highlighting the importance of the local contribution.