Composition, isotopic fingerprint and source attribution of nitrate deposition from rain and fog at a Sub-Arctic Mountain site in Central Sweden (Mt Åreskutan)

Abstract While dry and rain deposition of nitrate (NO3−) and ammonium (NH4+) are regularly assessed, fog deposition is often overlooked. This work assesses summer fog events contribution to nitrogen deposition and availability for forest ecosystems. Rain and fog samples were collected at Mt Åreskutan, Sweden, during CAEsAR (Cloud and Aerosol Characterization Experiment), in 2014. NH4+ + NO3− represent (31 ± 25) % of total rain ion amount, and (31 ± 42) % in fog. Based on ion concentrations and the nitrate stable isotope signatures δ(15N) and δ(18O), it was possible to detect the plume generated by the Västmanland forest fire; NOx emissions from oil rigs and Kola Peninsula; and the plume of Bardarbunga volcano, Iceland. Scavenging of ions by fog was more efficient than by rain. Rain NH4+ and NO3− deposition was (26 ± 36) μmol m−2 d−1 and (23 ± 27) μmol m−2 d−1, respectively. Fog NH4+ and NO3− contributed (77 ± 80) % to total wet deposition of these species. Upscaling rain deposition fluxes to 1 year gave an inorganic nitrogen deposition of (18 ± 16) mmol m−2 a−1 ((252 ± 224) mg m−2 a−1 N equivalents), whereas fog deposition was estimated as (59 ± 47) mmol m−2 a−1 ((826 ± 658) mg m−2 a−1 N equivalents). Annual fog deposition was four times higher than previously reported for the area which only considered rain deposition. However, great uncertainty on the calculation of fog deposition need to be bear in mind. These findings suggest that fog should be considered in deposition estimates of inorganic nitrogen and major ions. If fog deposition is not accounted for, ion wet deposition may be greatly underestimated. Further sampling of wet and dry deposition is important for understanding the influence of nitrogen deposition on forest and vegetation development, as well as soil major ion loads.


Introduction
Fossil fuel burning releases significant amounts of sulfur dioxide (SO 2 ) and nitrogen oxides (NO x : NOþ NO 2 ), which cause soil and water acidification as well as damage to biodiversity. Emissions of NO x and volatile organic compounds (VOCs) contribute to tropospheric ozone (O 3 ) formation. O 3 acts as a greenhouse gas, irritates lungs, and damages vegetation (Pleijel, 1999;WHO, 2005). SO 2 and NO x are oxidized to sulfate (SO 4 2À ) and nitrate (NO 3 À ) and form secondary particulate matter, which can have deleterious health effects when inhaled (WHO, 2005). Since NO 3 À is the final product of NO x oxidation, its concentration measured in different samples can be used as a proxy of NO x emissions (Hastings et al., 2004;Hastings et al., 2009). Following wet and dry deposition, atmospheric NO 3 À is incorporated into terrestrial and ocean ecosystems. The relatively pristine northern regions, including the sub-Arctic, (50-70 N) present fragile nitrogen-limited ecosystems that can be altered by even small increases of reactive nitrogen (N r ) deposition (Atkin, 1996;Aanes et al., 2000;Rinnan et al., 2007). Both, human development and greater food demand have increased N r deposition during the last century (Lamarque et al., 2010). N r enters different ecosystems and alters the natural nitrogen cycle, which has been described as the nitrogen cascade (Mosier et al., 2002). N r is released to the atmosphere mainly as NO x in most regions, and mainly as NH 3 over India and South-East China (Bauer et al., 2007), and these N r can be transported across regional scales (Holland et al., 1999). Consequently, the study of N r wet deposition is important for the understanding of N r sources and sinks to forest ecosystems (Ogren and Charlson, 1984;Ferm et al., 2000;Hultberg and Ferm, 2003).
In addition, the 15 N/ 14 N isotope delta of NO 3 À (d( 15 N)) can be used to identify NO x sources (Moore, 1977). The 18 O/ 16 O isotope delta of NO 3 À (d( 18 O)) can be used to infer atmospheric oxidative paths of NO 3 À formation (Michalski et al., 2003). d( 15 N) values from natural and anthropogenic sources cover a wide range (Michalski et al., 2003;Hastings, 2010;Li and Wang, 2008;Felix et al., 2012). Vega et al. (2015) showed that total NO 3 À during the last 60 years found in ice cores from Svalbard is dominated by fossil fuel combustion and soil emissions due to fertilizer usage in the US and Europe. In addition, NO x emissions from forest and grassland fire activity over Siberia are also evidenced in the ice cores (Vega et al., 2015). Seasonal variations of d( 15 N) have been reported by various authors (Heaton, 1987;Freyer, 1991;Freyer et al., 1996;Hastings et al, 2004;Morin et al., 2008;Frey et al., 2009;Morin et al., 2012). Atmospheric d( 15 N) shows summer maxima and winter minima at Arctic sites, in response to source seasonality and local processes (Heaton, 1987;Freyer, 1991;Freyer et al., 1996;Hastings et al, 2004;Morin et al., 2008;Frey et al., 2009;Morin et al., 2012). d( 15 N) values below À20 & are reached during springtime, as a consequence of the oxidation of NO x photochemically emitted from the snowpack (Frey et al., 2009;Morin et al., 2012).
On the other hand, seasonal variations of d( 18 O) and 17 O excess (D( 17 O) ¼ d( 17 O) À 0.52 Â d( 18 O) described by Barkan and Luz (2007)) depend on the nitric acid (HNO 3 ) production pathway from atmospheric NO x , with higher d ( 18 O) (Hastings et al, 2004;Morin et al., 2008;Frey et al., 2009;Morin et al., 2012). Using bi-weekly measurements of d( 15 N) and D( 17 O), Morin et al. (2008Morin et al. ( , 2012 found a connection between low d( 15 N) and high D( 17 O) values as a consequence of NO x produced by local photochemical emissions from the snowpack that were later oxidized to NO 3 À by reactive halogens (e.g., BrO via Br 2 and Br), in the atmosphere during spring. The Cloud and Aerosol Experiment at Åre (CAEsAR 2014) campaign took place from June to October 2014 at Mt Åreskutan in central Sweden. Designed to investigate aerosol and clouds physico-chemical properties under orographic forcing, the campaign sampling site allowed in situ characterization of such properties at Mt Åreskutan together with extensive remote sensing sampling at the nearby valley (Zieger et al., 2015;Franke et al., 2017). During the campaign, a forest fire occurred in V€ astmanland, South-Central Sweden, which started on July 31 and was controlled by August 11 (MSB, 2015). The forest fire generated a smoke plume that was transported over the observation site with measureable impacts on the aerosol properties (Franke et al., 2017), presenting the opportunity to assess the impact of biomass burning to the local N r load and isotope composition. This work presents the ionic and isotopic composition of nitrate in rain, and cloud-water (identified hereafter as fog) samples collected during the CAEsAR 2014 campaign, complemented by back-trajectory analyses to estimate major ion, and specially N r , deposition rates and sources. In particular, we wanted to assess the contribution of frequent summer fog occurring in mountain regions to deposition and nitrogen availability for forest biological processes Zimmermann, 2002, Templer et al., 2015), that may form a significant part of soluble ion deposition (Walmsley et al., 1996;Lange et al., 2003).

Study site
Rain and fog samples were collected at Åre (63˚26' N, 13˚6' E), Central Sweden (Figure 1a), during 27 June 2014 and 12 September 2014 (rain), and 4 July 2014 and 12 September 2014 (fog). Sixteen fog samples were collected at Mt Åreskutan station (1250 m above sea level a.s.l.), while 16 rain samples were collected at the foot of Mt Åreskutan (400 m a.s.l.) outside the village of Åre. In terms of air masses arriving to the site, the north-west sector represents clean Atlantic air, the south-east sector is characterized by air from densely populated and industrialized areas, and air masses coming from the north-east sector pass over boreal forests (Ogren and Rodhe, 1986;Drewnick et al., 2007;Franke et al., 2017). The mountainous area itself and its surroundings are considered rural areas; therefore no large local anthropogenic N r sources are expected. The station is often within the planetary boundary layer (PBL) and surrounded by clouds, making it suitable for studying the partitioning of pollutants between rain and fog. Table 1 shows details on sample collection and analyses. The samples were collected at various intervals spanning from sub-daily collection to a continuous collection during 17 days, with an average collection interval of 5 days (Table 1). Before collection, sampling instruments and bottles were rinsed with ultrapure water (>18 MXÁcm, Millipore Co.). Collection of samples was done in either high density polyethylene (HDPE, Nalgene) bottles or laboratory grade glass bottles (Duran V R ), both materials present an inert behavior, e.g., ion exchange between the bottle material and the sample can be neglected.

Sample collection and chemical analyses
Rain samples were collected in a 5 L plastic bottle using a wet-only sampler (M.I.C., Canada). Fog samples were collected in a pre-cleaned glass bottle with a custom-built single stage Caltech Active Strand Cloudwater Collector (Collett Jr. et al, 1990;Demoz et al., 1996) (Figure 1b), made with a stainless steel housing connected to fan rotating at 2000 min À1 . Air was pumped with a 5 m 3 min À1 flow through Teflon strings placed inside the housing to trap the fog drops by inertial impaction. The theoretical lower size cut-off of the instrument is 3.5 lm, based on droplet diameter (Collett et al., 1990). A switch turned on the fan only when relative humidity (RH) was above 80 %. After collection, rain and fog samples were filtered using Munktell quartz microfiber filters (T293, 47 mm diameter), poured in pre-cleaned plastic bottles and kept frozen (À20 C) until chemical analyses. Blanks for the collection of fog (N ¼ 2), rain (N ¼ 3), and for the filtration processes (N ¼ 1) were also obtained following the same procedure as for the samples (but with the fan turned off), using ultrapure water.
The samples were transported frozen (À20 C) to the Department of Earth Sciences, Uppsala University, in order to analyze major ion concentrations (Na þ , NH 4 þ , K þ , Ca 2þ , Mg 2þ , Cl À , Br À , NO 3 À , and SO 4 2À ) using a ProfIC850 Metrohm ion chromatograph (IC). Samples and standards were handled under a class 100 clean air hood, using powder-free gloves. Standards were prepared before analysis and stored frozen at À20 C. Shortly prior to analysis, samples and standards were melted at room temperature (with vial lids closed), and then filtered using 0.2 lm polyethersulfone (PES) filters. Samples were placed in the IC auto-sampler, covered with aluminium foil to avoid any dust contamination. A minimum of 5 mL H 2 O per sample was required to measure cations and anions. Calibration curves (R ! 0.9998) were constructed using 12 standards in the concentration range 1-1000 mg L À1 . Samples with ionic concentrations greater than the highest standard were diluted automatically using the Dosino module, which is part of the ProfIC850 chromatographic system. Three sample blanks consisting of ultrapure water were analysed at the beginning and the end of every sample batch. Check samples consisting of melted bulk snow collected in Uppsala, Sweden, were analysed every 10 samples to assess the reproducibility of measurements within a given batch. The reproducibility was within 5 % for each ion. Limits of Blank (LoB) (Armbruster and Pry (2008)) were calculated as the mean concentration plus 1.645 times one standard deviation (1r) of six separate blanks, i.e., LoB ¼ mean blank þ 1.645r blank , and were below 1 lmol L À1 for each ion, while the Limits of Detection (LoD) (Armbruster and Pry (2008)) were calculated as LoB plus 1.645 times 1r of six separate low concentration samples, i.e., LoD ¼ LoB þ 1.645r low concentration sample , and were below 1 lmol L À1 for each ion. Non-sea-salt (nss-), and sea-salt (ss-) fractions were calculated using the mean seawater composition with Na þ in seawater as reference ion, as follows: ( ᭢ ) Samples in which nitrogen stable isotopes are measured; ( Ã ) volume corresponds to a fraction of total volume, i.e., the 5 L collection bottle was over-filled and part of the water was discarded; (n.a.) not available; and ( ÃÃ ) R eff below the cut-off size of the instrument. Samples in italics indicate that they were collected during the timespan of the forest fire in central Sweden. where X is an ion different from Na þ , and calculated using the standard mean chemical composition of seawater with a practical salinity of 35. Due to the relatively low concentrations of NH 4 þ and NO 3 À in surface seawater (Summerhayes and Thorpe, 1996), NH 4 þ and NO 3 À were not separated into nss-and ssfractions (i.e., these ions were assumed to have a nss-origin only). d( 15 N), d( 17 O) and d( 18 O) of nitrate were analysed at the Stable Isotope Laboratory, University of East Anglia, using the denitrifier-thermal decomposition method (Casciotti et al., 2002;Kaiser et al., 2007). This method uses denitrifying bacteria (Pseudomonas aureofaciens) to convert NO 3 À into nitrous oxide (N 2 O), followed by thermal decomposition in a gold tube at 880 C, which is then introduced into a mass spectrometer to determine d( 15 N), d( 17 O) and d( 18 O). The method requires a minimum of 10 nmol of NO 3 À (20 nmol being optimum) in at most 13 mL of sample. Consequently, samples R11-R14, which presented low NO 3 À concentrations, were pre-concentrated using lyophilisation following the procedure described by Vega et al. (2015). Considering that the observed range of d( 15 N) in the samples is about 20 &, and that the concentration factors of the lyophilized samples were <10, any changes in d( 15 N) due to sample pre-concentration are negligible compared to the environmental variability and can be disregarded. All samples were then filtered using 0.2 lm PES filters, collected in clean plastic tubes and kept frozen (À20 C) until isotopic analysis.

Rain deposition
Since rain samples were collected at different sampling intervals, mean daily precipitation (P d ) was estimated using daily precipitation data provided by the Swedish Meteorological and Hydrological Institute (SMHI, http:// opendata-download-metobs.smhi.se) for three meteorological stations (Medstugan, Digern€ aset, and Vallbo) located nearby the study site ( Figure 1a). Of the 77-day campaign period, only 58 days had precipitation (Table  S1, in the supplementary material). The daily ionic rain deposition (D rain , in mmol m À2 d À1 ) was then calculated as follows: where c i is the ionic concentration, in mmol L À1 , measured in rain samples, and P d the mean daily precipitation, in mm d À1 (Table S1). Even though we can approximately know the amount of daily precipitation by using the station data, since each rain sample corresponds to various days in which precipitation was collected, it is not possible to know if the ion concentrations vary from one collection day to another. Consequently, the calculation of D rain was done assuming that the ionic concentration during different days, within the same collection interval and rain sample, remained constant.

Fog deposition
Ion deposition by fog (D fog , in mmol m À2 d À1 ) was estimated as: where 86400 is a conversion factor for units (e.g., number of seconds in one day (s d À1 )), c(H 2 O) is the mean liquid water mass concentration in air (or liquid water content; LWC) (in g m À3 ) corresponding to each fog event (Table  S2), q(H 2 O) is the density of liquid water (in kg L À1 ), v is the mean deposition velocity of fog droplets (in m s À1 ) during each fog event, and c i is the ionic concentration (in mmol L À1 ) measured in each fog sample. LWC was obtained using laser-diffraction measurements at 5-min averages, and readings were only considered during periods in which the sampling location was within the cloud. D fog depends on droplet size through the v parameter. The effective radius (R eff ) of fog droplets corresponding to each fog event was obtained using a particulate volume monitor (PVM; Gerber, 1991) at a 1-min time resolution. Mean R eff values are shown in Table 1. Two fog samples (F7 and F11) had mean R eff below the fog collector cut-off size, therefore, we did not considered those samples in the D fog calculations. We used modelled cloud droplet deposition velocities reported by Matsumoto et al. (2011) for particles with radii larger than 1.75 lm (cut-off of the fog collector) and smaller than 15 lm (i.e., v ¼ 0.047 m s À1 for 1.75 lm<R eff <2.5 lm; v ¼ 0.101 m s À1 for 2.5 lm<R eff <5 lm; v ¼ 0.135 m s À1 for 5 lm<R eff <10 lm; and v ¼ 0.156 m s À1 for 10 lm<R eff <15 lm). We then assigned deposition velocities to each R eff value measured at 1-min resolution.

NITRATE DEPOSITION FROM RAIN AND FOG 5
The LWC and v values were then averaged and used in equation (5). The D fog values reported in this study contain uncertainties related to the variables in equation (5), especially the selection of v values, in addition to the intrinsic fog collection efficiency of the instrument used in this study.

Air mass back-trajectory analyses
We used a reconstruction of air mass back-trajectories with the NOAA Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) (Draxler and Hess, 1998;Draxler, 2004) in order to understand the medium-and long-range transport of aerosols and the influence of different source regions of air masses arriving to the sampling site. The calculations were based on meteorological data from NCEP's (National Weather Service's National Centres for environmental Prediction) Global Data Assimilation System (GDAS) with 1-degree resolution. Air masses were modelled at an elevation of 500 m above ground level with 7-day back-trajectory runs with a restart interval of 48 h. The length of the back-trajectories was chosen based on the estimated lifetime of NO x in Arctic regions, i.e., $10 days (Liu et al., 1987), and the global mean lifetime of NO 3 À aerosol, i.e., 4-6 days (Xu and Penner, 2012).

Results and discussion
3.1. Ion concentrations and nitrate stable isotope values Table 2 shows mean molar concentrations of major ions (ss-and nss-fractions), standard deviation (1r), and the 25 th , 50 th and, 75 th percentiles for rain and fog samples. The 50 th percentile corresponds to the median and the 25 th and 75 th percentiles, i.e., lower and upper percentiles are located half-way between the median and the data extremes presented in Table 2, represent a robust and resistant measure of central tendency and spread of the data (Wilks, 2006). For the sake of comparativeness, we conducted our analysis using mean concentration values. It is worth saying that some of the references used in this study only report mean concentration or deposition values and the 25 th , 50 th , and 75 th percentiles (e.g., Jung et al., 2013), mean, standard deviation, and the 25 th , 50 th , and 75 th percentiles (Franke et al., 2017), median and extreme values (e.g., Ogren and Rodhe, 1986), mean and  extreme values (Zimmermann and Zimmermann, 2002;Lange et al., 2003), or did not report any statistics, either robust or nonrobust, to account for spreading of the data (e.g., Ogren and Charlson, 1984;Ferm et al., 2000). Our data sets are physically constrained to lie above a minimum value (i.e., precipitation, ion concentrations and deposition fluxes, cannot take negative values), therefore, the datasets would be more likely positively skewed. To test the symmetry of our data sets, we employed the Yule-Kendall index, which is a robust and resistant alternative to the skewness coefficient (Wilks, 2006). The results of the test show that major ion (Na þ , Cl À , Mg 2þ , NH 4 þ , Br À , NO 3 À , SO 4 2À , K þ , and Ca 2þ ) concentration data for rain and fog have Yule-Kendall index values above zero, which indicates that the data sets are rightskewed as expected. In rain, molar mean values were in the range of 0.1-17 lmol L À1 , with the exception of Br À that showed average concentrations two orders of magnitude lower than the other ions. Meanwhile, in fog, mean values were in the range of 1-53 lmol L À1 , with Br À also exhibiting mean concentrations two order of magnitude lower than the other ions. The fog/rain ratio in Table 2 shows that mean concentrations of most ions were 2-8 times higher in fog than in rain, with the exception of nssK þ , and nssCa 2þ which showed similar values in both rain and fog. NO 3 À made up a larger proportion of total inorganic nitrogen ions (NO 3 À þ NH 4 þ , NO 2 À was not analyzed) than NH 4 þ in both rain and fog samples. Figure 2 shows the contribution of each ion to the total load of ions in rain and fog samples. For both rain and fog samples, total inorganic nitrogen contributes about (31 ± 25) % and (31 ± 42) % of the total ion load, respectively. nssCa 2þ was the dominant ion in rain samples (27 ± 31) %, with NO 3 À , NH 4 þ , nssSO 4 2À , Na þ , and Cl À also having individual contributions above 10 %. In fog, Na þ was the dominant ion (27 ± 75) %, followed by NO 3 À , Cl À , NH 4 þ , and nssSO 4 2À with individual contributions above 10 %. Figure 3 shows the sea-salt molar ratios referred to Na þ in bulk seawater, rain, and fog (ratios were calculated using mean ion concentrations). The Cl À /Na þ ratio in rain and fog is lower than in seawater. For the other ions, ratios in rain and fog are higher than in seawater indicating a dominance of non-sea salt sources, which can be also observed in the mean ion concentrations presented in Table 2.
For stable isotopes of nitrogen and oxygen in nitrate, there is a difference of (6 ± 8) & between mean d( 15 N) values in fog (i.e., (À8 ± 2) &) compared to values in rain

Sources
In order to assess possible sources explaining the total variance in the ion concentrations from both rain and fog samples, a principal component analysis (PCA) was applied to the different ion series. For the PCA analysis, ion concentrations were log-normalised by taking their logarithms, subtracting the mean of the data series from each data point and then dividing the result by the standard deviation of the data series. PCA analyses were performed using ion concentration of samples listed in Table  1, at a resolution corresponding to each sampling interval. The sum of the variances of the first three principal components (PC1, PC2, and PC3) was ! 80 % of the total variance of the original rain and fog series. Principal components are shown in Table 3. PCA results are consistent between rain and fog samples, and show that all selected ions contribute similarly to most of the variance in the samples (PC1). The loadings in PC2 for rain, show that an independent source of NO 3 À and NH 4 þ (e.g., the effect of the large forest fire occurred in central Sweden during the sampling period, industrial NO x emissions and/or soil emissions form fertilized areas) could explain on the order of 12 % of the total variance, while the loadings in PC3 show that an independent source of Br À (not linked with sea-salts) could explain near 5 % of the total variance, e.g., from methyl bromide used for agricultural fumigation purposes. In the case of the fog samples, PC1 is similar as for rain samples, while PC2 suggest a source of NH 4 þ . PC3 shows a source of nssMg 2þ dominating this component, likely the input of mineral dust from the mountaintop. Figure 4 shows NH 4 þ , NO 3 À , and Na þ concentrations, and Figure 5 shows Br À , nssMg 2þ , nssSO 4 2À , and K þ concentrations, in individual rain and fog samples collected during the CAEsAR campaign. Note that the concentrations were plotted in such a way that each collection interval for rain overlaps the correspondent collection period for fog (e.g., R1 was collected in the same period as F1, R5 in the same period as F7, and no fog samples were collected during the collection period of sample R0). In case of samples R6 and R7, both overlap with sample F8 (the overlapping is indicated with horizontal arrows in Figures 4 and 5); while samples R9 and R10 together overlap with F10 and F11. We divided the samples into periods in which NH 4 þ and NO 3 À concentrations were more than double the mean (high), within twice the mean and half the mean (medium), or less than half the mean (low) ( Table 4). In general, most of rain samples presented NH 4 þ concentrations classified as medium (samples R0, R2-R10, and R15). One sample had NH 4 þ concentration classified as high (sample R1), concentrations ranging from medium to high (Table 4), had a common origin over industrialized regions or over the North Atlantic Ocean (Figure 6a, b, and  were also categorised as high; for R9, all ions but nssK þ and nssSO 4 2À had concentrations classified as high. For R15, only nssSO 4 2À concentrations were considered high. Therefore, we consider ion concentrations in samples classified as high to have mainly an anthropogenic influence. This is in agreement with pollution episodes evidenced by Franke et al. (2017) during July 3-12 corresponding to samples R1 and F1 (Figure 5a), which also show large rain daily deposition fluxes of nssSO 4 2À (Figure 7b), and the pollution event registered during September 7-10 corresponding to sample F15 explained further.
On the other hand, rain samples with NO 3 À concentrations classified as low, and NH 4 þ concentrations ranging from low to medium (Table 4), originated from three sectors: the Atlantic Ocean, the Arctic, and the Baltic area ( Figure S1). In addition to low inorganic nitrogen concentrations found in these samples, Br À , nssMg 2þ , and nnsSO 4 2À concentrations were also generally low or medium. Therefore, we consider samples with inorganic nitrogen concentrations classified as low to have marine ( Figure S1: samples R3, R4, and R14) and Arctic ( Figure  S1: R11, Figure 6: R12, and R13) origins.  Back-trajectories for samples with NO 3 À and NH 4 þ concentrations classified as medium (Table 4) showed inland air masses traveling within Scandinavia ( Figure  S2: samples R2, R6, and R7), with some of the trajectories reaching over the Atlantic Ocean ( Figure S2: samples R0, R5, and Figure 6: sample R8). This air mass classification based on ion concentration thresholds agree with the results by Franke et al. (2017) for air masses arriving Mt Åreskutan during the CAEsAR campaign period.
3.2.1. Forest fire. Franke et al. (2017) reported that the plume of a forest fire originated in V€ astmanland, Central Sweden (Figure 1a), which started on July 31 and was controlled by August 11 (MSB, 2015). The plume of the forest fire reached the mountain top during August 2-5, evidenced by high organic carbon (OC), elemental carbon (EC), and K þ concentrations measured in particulate matter sampled at the top of Mt Åreskutan. Daily rain deposition ( Figure  7) showed high NH 4 þ and NO 3 À deposition during August 4-6 (corresponding to sample R8), which could be associated to the forest fire plume. nssSO 4 2À and Br À rain deposition fluxes were also high during this event. Conversely to the findings by Franke et al. (2017), rain deposition of K þ did not show maximum values during the mentioned period. Figure 8 shows (Hastings, 2010;Fibiger and Hastings, 2016). NO 3 À measured in Svalbard ice cores has been linked to Siberian forest and grassland fires evidenced by a marked increase in d( 15 N) values (i.e., less negative values corresponding to an enrichment in the 15 N isotope) (Vega et al., 2015).  Figure 8b).  (Table 4), and air mass back-trajectories for the period covered by the samples had a north to north-east origin (e.g., Figure 6e and f). These back-trajectories show that air masses travelled over Kola Peninsula, Russia, where several industrial NO x sources exist (Prank et al., 2010;Hongisto, 2015). In addition, Franke et al. (2017) reported high EC to OC mass concentration ratios during August 20-23, associated with air masses arriving from the north, with PBL residence times above oil rigs off the Norwegian coast. These rigs emit NO x (Lee et al., 2015) (Figure 8b), which could be due to formation of NO 3 À via oxidation of NO x with hydroxyl radicals (OH) in Table 4. Rain and fog sample classification considering concentrations thresholds; more than double the mean (high), less than half the mean (low), and within twice the mean and half the mean (medium).

Sample Rain (Fog)
Rain Fog On the other hand, samples R1, and R9-R10 show medium to high NH 4 and NO 3 , but no positive d 15 N. In addition, sea-salt (Na þ ) concentrations measured in these samples have medium (R10) to high (R1 and R9) values (Figure 4a). Back-trajectories for these samples show air masses travelling from the south-west sector to the study site (Figure 6a and b, and Figure S2f). Figure 7 in Franke et al. (2017) shows major offshore oil rig installation over Northern Europe. We hypothesize that medium to high inorganic nitrate in these samples may originate from oil rigs located in the Northern Sea, which would explain also the enrichment in sea-salts observed in these samples. Grahn et al. (2015) reported the presence of sulfur gases detected on September 10 at V€ asterbotten county, approximately 250 km to the north east of Åre due to degassing of the Icelandic Bardarbunga volcano. nssSO 4 2À concentrations in rain and fog (samples R15 and F16, Figure 5a and b), and rain nssSO 4 2À deposition showed high values during September 11 (Figure 7b), which could be attributed to the passing of the volcanic plume over Åre. This is supported by the air mass back-trajectory for that period (Figure 6d) which shows transport pathways from Iceland to central Sweden.

Scavenging of ions
As noted in Section 3.1, mean concentrations of ions showed higher values in fog than in rain (Table 2), which suggest that scavenging of ions per unit of cloud water deposited by wet deposition was more efficiently done by fog than by rain (with the exception of Ca 2þ ), as expected for water-soluble ions (Lange et al., 2003;Gilardoni et al., 2014). NH 4 þ and NO 3 À showed similar (statistically Fig. 7. Stacked bar plot of (a) NH 4 þ , and NO 3 À , and (b) Br À , nssMg 2þ , and nssSO 4 2À daily rain deposition obtained as explained in Section 2.3. Note that Br À concentrations have been scaled to improve visualization.
NITRATE DEPOSITION FROM RAIN AND FOG 13 significant at 95 % confidence interval when using a Wilcoxon rank-sum test (Gibbons and Chakraborti, 2003)) fog/rain ratios of mean concentrations ( Table 2). The highest partition between fog and rain was registered for seasalts Na þ (fog/rain ratio: 8 ± 22), ssK þ (10 ± 36), ssCa 2þ (10 ± 36), ssMg 2þ (8 ± 21), and ssSO 4 2À (8 ± 22), most likely associated with the presence of these ions in coarse form, as for example in sea-salt aerosols acting as fog condensation nuclei (Sasakawa et al., 2003), supporting the assumption that Na þ in the samples was mostly originated by sea-salts (Section 2.2). Jung et al. (2013) have found that there is a preferential behavior for coarse particles to act as condensation nuclei in sea fog, with NO 3 À being scavenged more efficiently than NH 4 þ . Table 5 shows fog/rain ratios of mean ion concentrations of samples classified as high, medium, or low (explained in Section 3.2.), and associated to three main ion sources: anthropogenic (i.e., high NO 3 À concentrations), inland (i.e., medium NO 3 À concentrations), and marine Arctic (i.e., low NO 3 À concentrations) (also described in Section 3.2.). Sea-salt ions present similar ratios within each group, with ratios associated to anthropogenic sources being larger than ratios of marine Arctic sources, and larger than inland sources (with the exception of Cl À that shows lower values than the other sea-salt ions in samples associated with anthropogenic and inland sources). The lower Cl À ratio could be due to the volatilization of sea-salt Cl À as HCl, which would be enhanced by the presence of HNO 3 and H 2 SO 4 ; consequently, it could be expected that Cl À loss would be enhanced by the presence of anthropogenic aerosol. There is a clear difference between the ratios for sea-salt fractions and non-sea-salt fractions (Table 5). In samples associated to anthropogenic sources, ratios for the same ion are higher in the sea-salt fraction than in the non-seasalt fraction. This could indicate that fog scavenges nonsea-salt fractions less efficiently than sea-salt fractions in air masses with anthropogenic origin. Non-sea-salt fractions of K þ , Ca 2þ , and Mg 2þ are often associated with crustal emissions. Or, alternatively, that crustal aerosols were below the altitude at which the fog developed at Mt Åreskutan, hence, those would not be scavenged by fog, and concentrations would be depleted in fog samples compared to rain, with rain washing out particles below cloud level.
In the case of nssMg 2þ , the ratio found in air masses associated with marine-Arctic origins, was almost twice the ratio for the sea-salt fraction. This could be attributed to sources of nssMg 2þ with no crustal origin, due to the fact that other crustal ions do not show high fog/rain ratios (e.g., nssK þ nor nssCa 2þ ). Kindbom et al. (1993) reported point sources of Mg 2þ (i.e., steel industry, forest industry, and cement industry) and areal sources (i.e., wood/bio-fuel, oil, and coal) in Sweden. Marine Arctic air masses passed over large mining and industrial areas located in Northern Sweden; therefore, it is plausible that nssMg 2þ aerosol generated by those activities could had been transported to Mt Åreskutan and scavenged preferably by fog than rain.
Ratios for Br À are also high for marine Arctic air masses, which could be due to non-sea-salt sources. There are multiple sources of bromine that could eventually generate Br À in the troposphere. Wetlands are a natural source and sink of methyl bromide (CH 3 Br) to the atmosphere, with net uptake found for wetlands located in Northern Sweden (Hardacre et al., 2009). In addition, anthropogenic organic and inorganic brominated  compounds are mainly used in agriculture and paper-cellulose industry, respectively, as fungicide and algaecide (KEMI-report, 2014). However, it is not possible to assess with the current data if Br À measured in rain and fog is mainly from natural, agricultural, or industrial sources; this question remains for future work on the area, considering the presence of wetland, forest plantations, and at least four cellulose factories north-east of the sampling site (SkogsIndustrierna, 2018).

Rain deposition
Mean precipitation during the sampling period was (2 ± 3) mm (with 58 registered precipitation events within the 77-day sampling period, Table S1). A precipitation event was considered as "large" when the precipitation amount was above the mean precipitation value plus twice the standard deviation (i.e., 8 mm). Therefore, according to the station data, large precipitation events were observed in the region during July 17, August 4-6, and August 18. Table 6 shows mean and 1r of rain deposition, in lmol m À2 d À1 , of major ions sampled at Mt Åreskutan. Mean rain deposition of NH 4 þ and NO 3 À was (26 ± 36) lmol m À2 d À1 and (23 ± 27) lmol m À2 d À1 , respectively, representing (36 ± 37) % of the total major ion rain deposition flux. Ca 2þ contributed with the largest individual percentage to the total flux (25 ± 30) %, while Na þ and Cl À contributed together with (22 ± 26) % of the total rain deposition flux, and SO 4 2À contributed with (12 ± 15) %, with the remaining ions, K þ , Mg 2þ , and Br À , contributing less than 10 % altogether. Daily rain deposition of NH 4 þ , NO 3 À , Br À , nssMg 2þ , and nssSO 4 2À are shown in Figure 7. Rain deposition was highly variable during the campaign depending on ion concentration and precipitation amount. Of the three large precipitation events registered in the station data (i.e., more than 10 mm per day, Table S1), only two of them showed increased NH 4 þ and NO 3 À fluxes (i.e., on July 17, and August 4-6, Figure 7). The largest rain deposition events (i. e., on July 7, and August 4-6, Figure 7) correspond to anthropogenic and inland air mass back-trajectories. Those two events represented 8 % and 27 % of the total load of NH 4 þ and NO 3 À during the whole campaign period.

Conclusions
Total inorganic nitrogen ions (NH 4 þ þ NO 3 À ) amount to (31 ± 25) % and (31 ± 42) % of the total load of ions in rain and fog samples, respectively. Based on an air mass back-trajectories and principal component analysis, three main air mass origins were identified for the sampling period: anthropogenic, inland, and marine Arctic, in agreement with previous findings by Franke et al. (2017) during the CAEsAR campaign. Anthropogenic air masses were associated with high NH 4 þ and NO 3 À concentrations; however, not necessarily with the largest influx of inorganic nitrogen through rain deposition. Marine Arctic air masses showed generally low ion concentrations; however, they presented high fog/ rain ratios for Br À and nssMg 2þ . High NH 4 þ and NO 3 À rain deposition fluxes during August 4-6 (corresponding to sample R8) could be associated to the plume of the forest fire occurred in V€ astmanland, Central Sweden, which started on late July and was controlled by mid-August. This plume was also detected by Franke et al. (2017) as high concentrations of carbonaceous aerosols sampled at Mt Åreskutan during these particular days. In addition, less negative d( 15 N) values found in sample R8 could be linked to the influence of NO 3 À enriched in 15 N, fingerprint previously reported in the forest fire plumes (Hastings et al., 2010). A sharp shift in d( 15 N) toward positive values was observed in rain samples with marine Arctic origin, during the period August 22-September 10 (samples R12-R14), likely due to the influence of both Kola Peninsula and oil rig NO x emissions transported from the north to the study site. However, the d( 15 N) of such emissions is not known. Evidence of pollution events of this kind has been also reported by Franke et al. (2017). These findings indicate that polluted air masses with different sources and origins can be traced using d( 15 N) at this site, independent of the NO 3 À concentration present in the plume and samples. In addition, nssSO 4 2À concentrations in rain and fog (samples R15 and F16), and rain deposition showed high values during September 11, which can be attributed to the passing of the plume generated by Bardarbunga volcano in Iceland (Grahn et al., 2015). In general, fog scavenging (per unit of cloud water) of major ions at Mt Åreskutan was more efficient than by rain. Rain deposition fluxes for NH 4 þ and NO 3 À were estimated to be (26 ± 36) lmol m À2 d À1 and (23 ± 27) lmol m À2 d À1 , representing approximately (36 ± 37) % of the total ion deposition by rain. NH 4 þ and NO 3 À total deposition by fog was estimated as (77 ± 80) % of total wet deposition (i.e. rain deposition þ fog deposition), which is four times higher than the contribution by rain deposition to total inorganic nitrogen; therefore, acting as a significant source of nutrients to the area. However, it is worth saying that due the great uncertainty on the calculation of D fog (sections 2.4 and 3.5), associated to the deposition velocity selection used in the calculation of D fog , these results should be considered bearing in mind those large uncertainties attached to the calculations. Scaling the rain and fog deposition fluxes obtained during the CAEsAR campaign, annual rain deposition of Table 6. Mean and standard deviation (1r) for rain and fog deposition of selected ions considering modelled deposition velocities reported by Matsumoto et al. (2011).

Ion
Rain deposition (lmol m À2 d À1 ) Fog deposition (lmol m À2 d À1 ) NH 4 þ and NO 3 À at the sampling site would be approximately (18 ± 16) mmol m À2 a À1 N equivalents ((252 ± 224) mg m À2 a À1 N equivalents), while annual fog deposition of total inorganic nitrogen would be approximately (59 ± 47) mmol m À2 a À1 N equivalents ((826 ± 658) mg m À2 a À1 N equivalents). When considering the percentage of contribution of fog deposition to total wet deposition, the corrected upper end of the range for wet deposition reported by Bertills and N€ ashol (2000) (i.e., now corrected to include fog deposition) would be 870 mg m À2 a À1 N equivalents which is close to the annual value estimated in this study for Mt Åreskutan in this study.
In conclusion, the results presented in this study suggest that fog deposition cannot be neglected in this area, at least for the particular time of the year of the study. If deposition by fog is not accounted for, wet deposition of different compounds such as inorganic nitrogen may be greatly underestimated. This in turn may have implications for biogenic cycle of forests and other vegetation. Consequently, further sampling of wet and dry deposition is important for understanding the influence of atmospheric nitrogen deposition to the site and its influence on forest and vegetation development, as well as major ion concentrations in the soil.