In Table 2 concentrations of inorganic ions, H2O2 (aq), S(IV), HMS, and DOC as well as cloud water pH are summarised for the events given in Table 1. The observed range of pH-values was from 3.6 to 5.3, with a mean of 4.3. Highest ion concentrations (on a molar basis) were observed for ammonium, followed by nitrate. Sulfate, chloride, and sodium showed considerably lower concentrations, while potassium, magnesium, and calcium were lowest. Arithmetic mean concentrations of this study are compared to literature data from clouds/fogs at other European sites in Table 3. Note that some authors report arithmetic means, while others report volume-weighted mean concentrations, which are always lower for a given data set (see Table 2). Comparability of literature pH data is even more hampered as it is either reported as arithmetic mean or derived from either arithmetic or volume-weighted mean H+ concentrations (the first approach leading to higher values than the other ones). In general, however, concentration levels in the present study are often similar to those observed in more recent campaigns at Puy de Dôme (continental non-polluted regime; Deguillaume et al., 2014), in the western Sudety Mountains (Blas et al., 2008), and at the Schmücke site in a previous campaign (Brüggemann et al., 2005). In contrast, data from the 1980s and 1990s often show much higher concentrations of sulfate and nitrate (Bridges et al., 2002; Herckes et al., 2002; Wrzesinsky and Klemm, 2000; Acker et al., 1998; Joos and Baltensperger, 1991; Lammel and Metzig, 1991), presumably due to the decline in European emissions of NOx and SO2 over the past decades (EEA, 2014). Concentrations of DOC are more sparsely available in the literature for European clouds. Mean values during HCCT-2010 compare well with data from Puy de Dôme (continental non-polluted regime; Deguillaume et al., 2014), Rax (Löflund et al., 2002), and Schmücke (Brüggemann et al., 2005). Data for H2O2(aq) and S(IV) are even more sparse. In the present study, H2O2(aq) has been found to be within the same order of magnitude as determined in similar environments (Deguillaume et al., 2014; Brüggemann et al., 2005; Löflund et al., 2002), while S(IV) is at the lower end of reported concentrations.

Average relative compositions based on volume-weighted mean concentrations (in mg L-1) are shown in Fig. 1 for the main cloud events. DOC was converted to DOM (dissolved organic matter) using a conversion factor of 1.8 as in previous studies (Giulianelli et al., 2014; Benedict et al., 2012; Straub et al., 2012; Collett et al., 2008). Solute concentrations are always dominated by the main ions sulfate, nitrate, and ammonium, explaining approx. 60–70 % of total determined concentrations (campaign average 62 %). Among them, nitrate represents the dominant species (approx. 30–50 % of total concentrations, average 35 %), while sulfate and ammonium comprise lower fractions of total solutes (averages of 14 and 13 % respectively). Organic compounds contribute approx. 20–40 % (average 28 %) and are thus another main constituent of cloud water dissolved material. These fractions are similar to what has been reported for background and anthropogenic influenced conditions at Puy de Dôme (Marinoni et al., 2004) and are – despite the different environment – strikingly similar to the 20-year mean composition of Po valley fogs with 35, 15, 18, and 25 % contributions of nitrate, sulfate, ammonium, and DOM respectively (Giulianelli et al., 2014).

The ion balance of inorganic anions versus cations (including [H+]) is shown in Fig. 2. An anion deficit is observed for nearly all samples, ranging up to 178 µeq L-1. Inorganic anions missing from the calculation are unlikely to explain the deficit, as they will have a small impact on the ion balance only (bicarbonate < 1 µM for given pH values, bisulfite < 3.2 µM based on S(IV) and HMS data). Concentrations of a large number of organic acids were measured from the bulk cloud water samples and will be presented elsewhere (van Pinxteren et al., 2016). Summing up the equivalent concentrations of the most abundant determined acids (formic, acetic, glycolic, oxalic, malonic, succinic, and malic acid) with consideration of their respective dissociation states depending on their pKa values and sample pH values gives a range of 5–82 (average of 23) µeq L-1, which explains 6–100 % (average of 56 %) of the inorganic anion deficit. In about 10 % of the samples organic acid equivalent concentrations significantly exceeded the anion deficit (up to 255 %), likely related to measurement uncertainties and/or non-determined cations. Considering that the DOC fraction likely contains many more than the analytically resolved organic acids, it can be assumed that the missing anions are predominantly organic in nature and that organic acidic material had a non-negligible impact on the cloud water acidity during HCCT-2010. Similar observations have been made before in other cloud/fog systems (Straub et al., 2012; Hegg et al., 2002; Khwaja, 1995; Collett et al., 1989).

In Fig. 3a the variability of observed solute concentrations for selected ions is indicated in box plots. Variability was high both within events (max / min ratios of up to 5–8 for main ions during the longer events and up to 5–34 for minor ions), as well as in-between events (max / min ratios of median concentrations between 3 and 6 for main ions, 6–29 for minor ions). In general, cloud water solute concentration variability can be caused by (i) changes in microphysical cloud conditions, e.g. supersaturation and LWC; (ii) changes in CCN concentration, size distribution, and chemical composition; (iii) changes in gas-phase concentrations of soluble gases and corresponding phase equilibria; and (iv) chemical reactions in the cloud water. Distinctly different concentration patterns can be observed in Fig. 3a for three ion groups from similar sources, i.e. secondary ions ammonium, nitrate, and sulfate, sea-salt ions sodium and chloride, and the biomass burning and/or soil marker potassium, indicating a dominant influence of air mass history and thus CCN concentration and composition on cloud water solute concentrations. This is most obvious for sodium and chloride, which show highest concentrations during FCEs 1.1, 22.1, and 26.1+2. During these events, back-trajectory analysis revealed a stronger influence of marine emissions (residence time indices above water surfaces were between 0.3 and 0.5, as compared to < 0.2 for the remaining events; cf. Fig. S5).

To remove any influence of LWC fluctuations, CWLs are plotted in Fig. 3b. The CWL patterns resemble those of solute concentrations to a large extent, suggesting that for our data set CCN composition and concentrations of soluble gases (i.e. air mass history) have a stronger impact on cloud water solute concentrations than LWC variability. Relative standard deviations (RSDs) of solute concentrations (whole campaign) are 66, 60, and 60 % for sulfate, nitrate, and ammonium respectively and 84–125 % for trace ions. RSDs of CWLs are similar, sometimes even higher, with values of 80, 52, and 66 % for sulfate, nitrate, and ammonium respectively and 62–96 % for trace ions. Removing LWC variability, therefore, does not reduce concentration variability, at least for the LWC range in this study. This is similar to observations of Aleksic and Dukett (2010) from a much larger data set and indicates that LWC is obviously an important, but not necessarily the primary control, factor of solute concentrations.

If at all, an inverse functional relationship between solute concentration and LWC (Elbert et al., 2000; Möller et al., 1996) can only be observed during single events (i.e. when CCN concentration and composition as well as gas-phase concentrations might be regarded comparably constant) in our data set. This is shown in Fig. 4a for TIC versus LWC where the colour-coded single event data indicate more or less constantly decreasing TIC with increasing LWC for some events. Overall, however, the pattern approximates those observed for larger data sets (Aleksic and Dukett, 2010; Kasper-Giebl, 2002; Möller et al., 1996): maximum TICs are decreasing, while minimum TICs stay relatively constant with increasing LWC, leading to a range of observed TICs at any given LWC. As one and the same LWC value can result from different cloud microphysical conditions (e.g. few large drops versus more small drops) and clouds with similar LWC can form in very different air masses, this is actually an expected observation. In several other cloud/fog studies relationships between TIC and/or solute concentrations with LWC were reported to be non-existent (Giulianelli et al., 2014; Straub et al., 2012; Marinoni et al., 2004; Kasper-Giebl, 2002).

The reason for this ostensible contradiction to the conclusions of the studies by Möller et al. (1996) and Elbert et al. (2000) might lie in different assessments of the quality of fitted models. Möller et al. (1996) and Elbert et al. (2000) report power law fits with coefficients of determination (R2) of 0.27 and 0.38 respectively. Even when considering these values satisfactory (on the general usefulness of R2 especially for goodness-of-fit of nonlinear models see Spiess and Neumeyer, 2010), the presented scatter plots leave room for questioning the ability of the fitted functions to adequately represent the data.

Instead of LWC, Marinoni et al. (2004) report TIC in cloud water at Puy de Dôme to be a power function of effective droplet radius (Reff), even though with similarly poor R2 of 0.29. In Fig. 4b, TIC during HCCT-2010 is plotted against Reff, which was determined by the PVM as well. In contrast to LWC, both maximum and minimum TIC values are decreasing with increasing Reff in this plot and the relationship comes indeed closer to a functional one (best fit for simple linear regression; R2 increases from 0.14 with LWC to 0.52 with Reff as explanatory variable). There is, however, still substantial unexplained TIC variation, likely arising from different broadness and/or skewness of the droplet size spectrum and from processes like phase equilibria and/or aqueous-phase reactions.

In Fig. 4c and d the relationships of DOC with LWC and Reff are shown, which are very similar to the ones observed for TIC. Herckes et al. (2013) examine total organic carbon concentrations against LWC for a number of different sites worldwide. A simple relationship explaining the variation across all locations could not be identified by the authors. However, their plot looks remarkably similar to the plots of TIC versus LWC from the larger data sets referenced above (decreasing spread of concentrations with increasing LWC), indicating that the main factors controlling the organic content of fog and cloud water are the same as the ones determining inorganic ion concentrations (likely nucleation scavenging and some additional gas-phase uptake).

As a further means to study the various influences on solute concentrations, principal component analysis was performed on cloud water solute concentrations and pH, back-trajectory RTIs, LWC, and Reff. Factor loadings of four extracted principal components after Varimax rotation are shown in Table 4. The first factor is highly correlated to air mass residence times above the oceans and cloud water concentrations of sea-salt constituents sodium, magnesium, and chloride. The second factor shows high loadings for all four main cloud water solutes (sulfate, nitrate, ammonium, DOC), representing typical main particulate components in aged continental air masses. The third factor is highly correlated to potassium and calcium concentrations and air mass residence times above agricultural lands and likely represents a mixed soil/biomass burning influence. The fourth factor mainly includes the variability of air mass residence times above urban areas, with no strong correlation to cloud water constituents. pH shows a weak anticorrelation to this factor, which could indicate an impact of acidic pollutants in comparably fresh air masses.

LWC has a much smaller impact on the marine factor than air mass residence time above water and its loading on factor 2 is weak as well (in contrast to Reff, which has a significant impact on this factor). This further supports the conclusion of LWC variability impacting solute concentrations to a lesser extent when several clouds with different air mass histories are considered.

In summary, the discussion in this section shows that no single factor is available to adequately describe the complex processes controlling solute concentrations of both inorganic and organic material in bulk cloud water. If a simple functional relationship is needed, Reff might be a somewhat better choice than LWC. The probabilistic approach of Aleksic, however, seems more appropriate: for any given LWC (and probably Reff as well), solute concentrations exhibit a (non-linear) distribution, as they depend on several other variables at the same time.

In parallel to the bulk cloud water sampling, a CVI separated droplets from the interstitial phase and enabled the chemical characterisation of residual particles from filters and online with an AMS (Sect. 2.2). The resulting CWLs of main solutes (normalised to standard conditions) are compared to the ones obtained from bulk cloud water samples in Fig.  5. As can be seen, the temporal trends are often similar from both time-resolving samplers (CASCC2 and CVI-AMS), while absolute values can differ. During FCEs 11.3, 22.1, and 26.1+2, the ratios between CASCC2 and CVI-AMS CWLs are close to 1, especially for ammonium and sulfate (see Fig. S6 for ranges of CWL ratios). During FCEs 1.1, 11.2, and 13.3, this ratio is close to 2 (median), while it can be even higher for nitrate. Time-integrated mean CWLs from CVI filters are mostly close to the values from the CVI-AMS for sulfate and nitrate (with the exception of FCE1.1), while for ammonium, they are substantially lower during four out of the six events shown. CWL deviations for DOC (for residual particle data calculated as AMS organics divided by a conversion factor of 1.8 as above) tend to be lower than for the ions and CASCC2/CVI-AMS ratios are even below 1 during FCEs 1.1, 11.2, and 26.1+2 (Fig. S6). DOC CWLs from CVI filters are not given due to unreliable data from the small masses sampled on the filters.

Possible reasons for these deviations are manifold and include (i) different sampling locations in the cloud (tower versus inlet at house wall); (ii) different cut-off and detection characteristics (all dissolved bulk material analysed from CASCC2, while AMS measures non-refractory submicron residual particles only); (iii) different assumptions/corrections for sampling efficiency (assumption of constant sampling efficiency across droplet size spectrum for CASCC2, correction of CVI sampling efficiencies based on particle number size distributions); (iv) measurement uncertainties of analytical methods, AMS, and PVM for LWC measurement; (v) (for DOC) uncertainty in the OM to OC conversion factor (1.8) and inclusion of undissolved organic matter in the AMS residual organics concentration; (vi) (for filter samples) potential negative artifacts from evaporation of semi-volatile particle constituents during sampling as well as uncertainty from blank correction especially for short sampling times and low sampled masses; and (vii) (very important for some species) different droplet “pretreatment”, i.e. liquid collection in the bulk sampler versus evaporation of water and volatile constituents such as ammonia, nitric acid, and dissolved VOCs in the CVI. Given all these uncertainties and systematic differences, a general agreement between CWLs obtained from the different samplers within a factor of 2 appears acceptable. A notable exception with much less agreement is nitrate during FCE11.2, where bulk cloud water CWLs are about a factor of 3.5 higher than CVI concentrations. The reason for the large deviation during this event is likely an enhanced concentration of nitric acid, which is taken up as nitrate into the bulk cloud water, but can be (partly) released back to the gas phase during droplet drying in the CVI (see also the following section).

SEs were calculated by two different approaches. “In-cloud” SEs are based on cloud water loadings and interstitial particle concentrations (both being normalised to STP) and are calculated as follows: SEin-cloud=CWLCWL+cint, where SEin-cloud is in-cloud scavenging efficiency, CWL is cloud water loading in µg m-3, either from bulk cloud water (CASCC2) or from droplet residual concentrations (CVI-AMS and CVI-Filter), and cint is interstitial particle concentration in µg m-3 (INT-AMS or INT-Filter)

“Upwind” SEs, in contrast, are based on a comparison of STP normalised CWLs and upwind concentrations, calculated as SEupwind=CWLcupw, where SEupwind is upwind scavenging efficiency, CWL is cloud water loading in µg m-3 from bulk cloud water (CASCC2), and cupw is upwind concentration from MARGA measurements in µg m-3, either particulate only or total aerosol (particulate + gaseous concentration)

The results of these calculations are shown in Fig. 6. In-cloud SEs calculated from the different samplers usually agree well except for cases where sampler intercomparison was poor (Sect. 3.3.3). Comparison with upwind SEs, however, reveals substantial differences, which are summarised as event means in Table 5 (for residual in-cloud SEs only the ones based on CVI/INT AMS data are given here to avoid redundancy). Mean in-cloud SEs for sulfate are usually > = 0.9 except for FCE11.2 and FCE13.3, where substantial fractions (21–44 %, depending on data used) of in-cloud sulfate reside in interstitial particles. During these events particle activation curves obtained from comparing measured particle number size distributions upwind and in-cloud were comparably shallow and the critical activation diameter was larger than during other events (Fig. S7), consistent with larger fractions of submicron sulfate not being activated to cloud droplets due to cloud microphysical conditions. Consistent with our data, in-cloud SEs of sulfate between 0.52 and 0.99 have been reported for clouds at Puy de Dôme, Brocken, and Mt. Sonnblick (Sellegri, 2003; Acker et al., 2002; Hitzenberger et al., 2000; Kasper-Giebl et al., 2000), with larger values being more typical.

In contrast to in-cloud SEs, sulfate upwind SEs were mostly ≪ 0.9, indicating incomplete mass conservation between the sites. From previous studies at the Schmücke (Brüggemann et al., 2005; Herrmann et al., 2005) and results on aerosol processing presented in a forthcoming companion paper, it is known that various physical loss processes, such as scavenging of cloud droplets by trees and/or entrainment of cleaner air masses from aloft can reduce observed concentrations of all particle constituents along the air path from upwind via Schmücke towards the downwind site. Upwind SEs being smaller than in-cloud SEs support these conclusions of physical particulate mass losses from the upwind to the in-cloud site. Only during FCE13.3 are upwind SEs found to be higher than in-cloud SEs, indicating additional sulfate mass within the cloud, which could result from chemical production, uptake of gaseous H2SO4 (Roth et al., 2016; Harris et al., 2013, 2014), and/or other processes (e.g. entrainment). Similar to sulfate, ammonium shows in-cloud SEs typically > 0.9, except for FCE13.3 (large activation diameter). Upwind SEs are similarly large when only upwind particulate ammonium concentrations are considered, but they drop to mean values between 0.4 and 0.7 when gaseous upwind ammonia – which is likely to be taken up by the cloud water at least partially – is included in the balance. Consistent with the conclusions from sulfate, the lower overall upwind SEs thus likely reflect the impact of physical loss processes at the sites.

For nitrate and DOC, these comparisons look different. While in-cloud SEs are again > 0.9 in most cases, upwind SEs are > 1 in most cases, indicating additional nitrate and DOC at the in-cloud site (note that event mean DOC upwind SEs in Table 5 were calculated using water-soluble organic carbon concentrations from impactor samples, as the MARGA analyses inorganic ions only). For DOC, this most likely results from uptake of water-soluble VOCs (e.g. acids, aldehydes, ketones) into cloud droplets. The highest value was observed for FCE11.2, where the inorganic anion deficit was highest as well (Fig. 2), indicating that a significant amount of organic material taken up from the gas-phase must have been acidic or – alternatively – neutral compounds were oxidised to organic acids upon dissolution in the cloud droplets. It is noted that the main organic acids mentioned above explain only less than 10 % of the inorganic anion deficit for this event.

For nitrate, upwind SEs stay similarly high or even higher than in-cloud SEs even after considering any upwind HNO3 measured by the MARGA. Especially when considering that nitrate likely experiences similar physical mass losses as ammonium and sulfate (which typically were on the order of 10–40 % at the downwind site, data not shown here), this would imply a nitrate budget at the cloud site substantially larger than the sum of particulate and gaseous nitrate at the upwind site. Given that aqueous-phase oxidation of NOx to nitrate can be considered negligible (Seinfeld and Pandis, 2006) and a potential positive nitrate artefact from hydrolysis of N2O5 in the cloud water can be assumed to be present in similar magnitude in the wet rotating denuder samples of the MARGA system (Phillips et al., 2013), such a large budget increase of nitrate at the cloud site seems unrealistic. In addition, a comprehensive data analysis focussing on aerosol processing during FCEs (manuscript in preparation) does not yield indications for increased nitrate at a site downwind of the cloud either on average over all FCEs or specifically during FCE11.2, where nitrate enrichment was highest. Any additional nitrate in the cloud water thus needs to evaporate back to the gas phase upon cloud dissipation.

The most likely explanation for the observed discrepancy is a severe underestimation of nitric acid by the MARGA system. Accurate nitric acid determination is known to be challenging due to the “stickiness” of the molecule (Rumsey et al., 2014) and adsorption in the inlet was reported to be strongly increased when sampling air – as during FCE sampling – is near 100 % RH (Neuman et al., 1999). As the inlet HDPE tubing during HCCT-2010 was approx. 3.5 m long (from PM10 head to denuder), significant losses of HNO3 before denuder sampling seem likely. In a (yet) unpublished intercomparison of nitric acid between the MARGA unit as used during HCCT-2010 and a separate batch denuder with inlet tubing reduced to a minimum, concentration ratios between the MARGA and the reference denuder were typically between 0.17 and 0.98 (10th and 90th percentile; G. Spindler and B. Stieger, personal communication, 2015). Using a value of 0.25 (lower quartile of the intercomparison) as a correction factor for nitric acid measured during HCCT-2010 (i.e. multiplying measured apparent concentrations by 4) yields upwind SEs for total nitrate between 0.7 and 1.2 (as event means), which would be more consistent with the values obtained for ammonium and sulfate.

An enrichment of cloud water nitrate has previously been observed in several studies and has usually been related to the uptake of nitric acid as the most probable explanation (Prabhakar et al., 2014; Hayden et al., 2008; Brüggemann et al., 2005; Sellegri et al., 2003; Cape et al., 1997), which is in agreement with our considerations described above.

In conclusion, the comparison of upwind and in-cloud scavenging efficiencies reveals that (i) nucleation scavenging typically removed > 80 %, often close to 100 % of soluble material from the particle phase upon cloud formation, (ii) uptake of gaseous ammonia, nitric acid, and water-soluble VOCs had an additional significant impact on observed cloud water concentrations, and (iii) particulate material is clearly lost or diluted to some extent between the upwind and the in-cloud site, likely due to physical processes such as droplet scavenging by trees and/or entrainment of cleaner air masses.

In Fig. 7 volume-weighted mean (VWM) concentrations per cloud event are shown for ions, H2O2, and DOC within the droplet size classes of the three-stage collector. Even though the nominal cut-off diameters of the three stages are given in Fig. 7, it has to be noted that in reality significant mixing of droplets between the nominal size classes occurs due to the relatively broad collection efficiency curves (Straub and Collett, 2002). Concentrations in a given droplet size class are thus influenced by droplets from other size classes to a significant extent and the size distributions can only reflect an approximate picture of the real pattern. Volumes of cloud water collected per stage were between 5.9 and 240 mL with typically lowest volumes on the intermediate stage (16–22 µm) and highest volumes in the smaller or larger size class, depending on the sample (see Fig. S8 for details).

Volume-weighted mean concentrations per event were calculated to reduce the complexity of the data set, even though information on the temporal evolution of size-resolved concentrations is lost by the averaging. Data for all individual samples taken with the three-stage collector are given in the Supplement (Fig. S9–S18). As can be seen there, concentrations levels of individual cloud water constituents can vary significantly within one cloud event while the general patterns of concentrations in the three droplet size classes are often quite persistent during an event (exceptions will be noted below). For the major ions sulfate, nitrate, and ammonium, two main profiles of size-resolved cloud water concentrations can be observed in the VWM data: (i) decreasing concentrations with increasing drop size for FCEs 1.1, 11.2, 11.3, and 13.3 and (ii) profiles with minimum concentrations in medium-sized droplets on stage 2 (“U”-shaped profiles) for FCEs 22.1 and FCE26.1+2. Only for nitrate during FCE1.1 a profile of increasing concentrations with increasing drop size is observed. Concentration differences between highest and lowest values are usually within a factor of 2 with the exception of FCE11.2, where concentrations of sulfate and ammonium in large drops were a factor of 3–4 lower than in small drops (on VWM basis). The two types of profiles reflect the dominant profiles of major ions in the individual samples (Fig. S9–S11) for most of the events. Only during FCE1.1 and mainly for sulfate and ammonium does the VWM profile not adequately represent the individual profiles, which were rather variable during the first half of this 15 h event and stabilised to a profile of increasing concentrations with increasing drop size during the second half of the event. As sampled water volumes were comparably low during the second half of the event, however, their weight to the volume-weighted mean profile is rather low. Literature data from three-stage cloud water collectors is very sparse. Raja et al. (2008) report decreasing concentrations of main ions with increasing drop size for fog samples in the US Gulf Coast region, obtained with the same collector as in the present study. Collett et al. (1995) observed U-type, profiles in cloud samples obtained with a different three-stage collector (different nominal cut-offs) from two sites in North Carolina and California, USA.

The VWM profiles of low concentration ions (chloride, sodium, magnesium, calcium, and – in part – potassium) were found to be markedly different from the major ion profiles. Concentrations were usually increasing with increasing drop size, especially for events with elevated concentrations (FCE1.1, 22.1, and 26.1+2) due to elevated impact of marine emissions on sampled air masses (cf. Sect. 3.3.2). Also, observed concentration differences in different drop size ranges tended to be larger (up to a factor of 10) as compared to major ion concentrations. Available literature data for minor ions in three drop size ranges reveal diverse profiles, depending on species and location (Raja et al., 2009; Collett et al., 1995).

In contrast to the ionic data, concentrations of H2O2 in different collector stages were comparably homogeneous, with maximum differences of 25 % (or a factor of 1.3). This is likely related to the different incorporation pathway (uptake from gas phase as compared to nucleation scavenging for the ions), which is expected to yield more similar concentrations in differently sized cloud drops, at least if equilibrium conditions are assumed (Hoag et al., 1999).

Both uptake pathways can in principle occur for DOC (VOC uptake and/or dissolution of CCN organic material). The size-resolved concentration pattern in Fig. 7, however, resembles those of major ions, suggesting nucleation scavenging as the major path of DOC incorporation into cloud water during this study.

Mean pH values per event (based on VWM concentrations of H+) are shown in Fig. 8a. A similar pattern of slightly (approx. 0.1 pH units per stage) increasing values with increasing drop diameter can be observed for nearly all events and collector stages. In individual samples (Fig. S19) differences between stages can be somewhat higher (up to approx. 0.5 pH units), but the general patterns look similar to the VWM event averages. Qualitatively, increasing pH with drop size is consistent with (i) coarse (and typically less acidic) CCNs leading to larger droplets (cf. elevated concentrations of coarse particle mode constituents) and (ii) reduced (diluted) concentrations of potentially acidic constituents (sulfate, nitrate, DOC) in larger drops (Collett et al., 1994).

These observations highlight the complexity of solute concentration drop size dependencies. Even for the comparably uniform conditions of the present study (same site, same season, similar air mass origins, similar heights within the cloud), different profiles can result for one and the same ion. This becomes even more obvious from individual samples (e.g. sulfate during FCE1.1, Fig. S9), where – as stated above – a number of different profiles can occur during the same cloud event. Considering that these individual samples represent volume-weighted averages over 2 h, it is easy to imagine that with a higher time resolution of sampling the variability of observed profiles would even increase. Without detailed numerical modelling (which is beyond the scope of this study), a quantitative understanding of these profiles and their variations seems impossible. In addition, the sampler characteristics (few stages with broad collection efficiencies) together with changing droplet size distributions in a cloud might influence the observed size dependencies. Even though drop volume size distributions were usually similar both between events (Fig. S20) and between individual samples within the events (Fig. S21), subtle changes, e.g. in the broadness of the distribution or in the abundance of large (> 30 µm) drops, can – together with the broad mixing of differently sized drops – lead to artificial modifications in the observed volume-weighted concentrations on the three stages (Moore et al., 2004a). Despite these difficulties, two broad conclusions from the three-stage sf-CASCC ion data can be drawn: (i) main ions (sulfate, nitrate, ammonium) have similar solute concentration drop size dependencies (consistent with their presumed strong internal mixing in CCNs) and are often enriched in smaller sized droplets (even though other, especially U-type profiles do occur as well), and (ii) increasing concentrations with increasing droplet sizes, which might be expected based on the consideration of the simple Ogren et al. (1992) model (see Sect. 1), are mainly observed if a strong coarse mode in upwind particles is present for a given constituent (e.g. for sodium, magnesium, chloride, and nitrate- during FCE1.1; cf. Figs. S22 and S23 for size distributions of inorganic ions at upwind site during FCEs). These findings are consistent with the availability of coarse CCN being an important prerequisite for such an inverse concentration – size relationship to develop (Schell et al., 1997), although other factors likely contribute to these observations as well.

Size-resolved concentrations of ions and H2O2 from the five-stage collector are given in Fig. 9 in the same way as described above for the three-stage data (event VWM and normalised data). Collected cloud water volumes were from 0.55 to 15 mL, with smallest volumes typically in the 4–10 µm droplet size range and largest ones mostly for droplets > 30 µm (see also Fig. S24). Concentration profiles of individual samples are shown in Figs. S25–S33. The number of events is smaller, as this sampler was not operated during FCE1.1 and FCE26.1+2. Due to the relatively low volume of cloud water the five-stage collector is sampling, DOC analysis could not be performed from these samples. For major ions, the patterns are broadly consistent with the profiles of decreasing concentrations with increasing drop size observed from the three-stage collector for FCEs 11.2, 11.3, and 13.3, with FCE22.1 showing some similarity to a U shape (even though the concentration increase towards larger drops is observable on stage 2 only, not on stage 1 collecting the largest drops). Concentration differences between smallest and largest droplets are somewhat more pronounced (typically a factor of about 2) as compared to the three-stage collector (typically smaller than a factor of 2), illustrating the higher efficiency of the five-stage collector in separating small and large drop populations. Sharpest concentration differences are usually observed between stages 4 and 5 (small droplets). This is true for basically all of the individual samples as well (Figs. S25–S33). Concentration patterns on stages 1–4, however, can vary somewhat within a single event, depending on the development of the cloud. For example, nitrate shows constantly decreasing concentrations with increasing drop sizes during the first half of FCE11.2 (Fig. S26), while during the second half concentrations in larger drops tend to increase. Similarly, ammonium concentrations develop from a maximum in medium-sized drops for the first sample to notably homogeneous concentrations across all five collector stages (difference of only about 30 % between smallest and largest drops) during FCE11.2 (Fig. S27). The observed profiles differ from those reported from a hill cap cloud at Whiteface, NY, USA, using the same five-stage collector (Moore et al., 2004a): U-type profiles with highest concentrations in largest drops were observed for ammonium and nitrate, while sulfate showed increasing concentrations with increasing drop size through all five stages. The same study reports five-stage concentration profiles from a fog event in Davis, CA, USA, which are more similar to those in this study, with decreasing concentrations with increasing drop size (Moore et al., 2004a).

The patterns of trace ions also show some similarity with the ones observed from the three-stage collector, mainly in that concentrations tend to increase from medium-sized towards larger droplets for most ions and events as well. There are, however, two distinct features in the five-stage data which are not captured by the three-stage collector. First, similar to the main ions, the concentration increase towards larger droplets is often (though not always) observable on stage 2 only, with decreasing concentrations on stage 1 (largest drops). Second, all trace ions show a very pronounced concentration increase in smallest droplets (stage 5), with often a factor of 5–10 difference to stage 4 concentrations, which is usually not seen in the three-stage data, where smallest droplets are mixed with much larger ones on stage 3, leading to more diluted concentrations. Literature data on size-resolved trace ion concentrations from five-stage collectors is available only for calcium, for which a pronounced U-type profile with highest concentrations in largest drops was reported (Moore et al., 2004a), while sodium, potassium, and chloride ions were mentioned to have very similar profiles.

Compared to ionic content, the concentrations of H2O2 are more homogeneously distributed between the collector stages (maximum deviation < 50 %) – similar to what was observed from the three-stage collector data – and a general pattern cannot be observed from the (few) data available.

Event-averaged pH values from the five-stage collector are given in Fig. 8b (for individual samples in Fig. S34). Highest values were mostly observed in smallest droplets (stage 5) with a significant decrease towards the next droplet size range (stage 4) at least during three out of the four events. From collector stage 4 towards stage 2 (increasing drop sizes) pH values tend to increase, similar to what is observed from the three-stage collector (Fig. 8a), while in largest drops (stage 1) they decrease again (to different extents). Overall, pH variations between different drop size classes are not too large for the sampled clouds with maximum differences of about 0.6 pH units on event-averaged basis.

These observations are generally consistent with the findings from the three-stage collector. However, they also highlight the higher efficiency of drop population separation of the five-stage collector as compared to the three-stage collector, as ratios between minimum and maximum concentrations are larger and the sharp concentration increase towards the smallest droplets (especially for trace ions) is only observed here (for volume size droplet distributions during five-stage sampling see Fig. S35). In addition, the observation of often decreasing concentrations from stage 2 (second-largest drops) to stage 1 (largest drops) might reflect the transition from region II (condensation growth) to region III (coalescence growth) in the Ogren et al. (1992) model (Sect. 1), even though it must be noted that collection efficiency curves of these two stages are overlapping to a comparatively large extent (Straub and Collett, 2002). Compared to the study of Moore et al. (2004a) stressing the importance of cloud age (drop growth time) by comparing two different types of clouds/fogs, our data from more similar cloud systems highlight the impact of the size distributions of CCN constituents on the development of size-resolved concentration patterns. Both parameters were predicted to be relevant from detailed model sensitivity studies (Sect. 1, Schell et al., 1997). In addition, despite the considerable mixing of droplets with different sizes occurring in the samplers, the data reveal the substantial differences which can exist in different droplet size classes as well as the variability of observed solute concentration profiles even under comparably similar cloud conditions. As such differences impact both chemical reactions in cloud drops and deposition efficiencies and can thus modify atmospheric sink and/or source strengths of PM constituents (Moore et al., 2004b), further observational and modelling studies on size-resolved droplet compositions seem important.