Three meteorological stations (MSs) were installed on the north-facing slopes of the San Francisco valley along an altitudinal transect ranging from 1960 to 3180 m in elevation. A fourth station (El Tiro, 2725 m) was installed at a mountain pass about 4 km up the valley on the Cordillera Real (Fig. ).

Rain and OP samples were collected at each station between 2004 and 2009. While rain water was collected in totaling gauges (UMS 200, made of polyethylene to warrant chemical inertia), OP was collected in 1 m2 mesh grid fog collectors designed according to Schemenauer and Cereceda's proposal . Details about rain and fog measurement techniques, calibration, and data handling are described in , , and .

Rain and OP samples were collected at almost regular weekly intervals. The samples were filtered and immediately stored in frozen state, before being sent to the laboratory for ion analyses. All samples were analysed at the University of Munich's Weihenstephan Center (TUM-WZW) for major ions (K+, Na+, NH4+, Ca2+, Mg2+, Cl-, SO42-, NO3-). Cations were analysed according to the inductivity-coupled plasma method (Perkin Elmer Optima 3000), while ion chromatography (Dionex DX-210) was used for anions. The detection limits are 0.1 and 0.2 mg L-1 for sodium and chloride, respectively.

The sea salt mixing ratio of the Monitoring Atmospheric Composition and Climate (MACC) reanalysis data set was used as a proxy for the sea salt concentration in the atmosphere, with a horizontal resolution of 0.75∘ by 0.75∘ . In this data set, the concentration of sea salt generated by wind stress on the ocean surface was determined based on a source function developed by and , accounting for sedimentation as well as wet and dry deposition processes. The sea salt concentration was integrated for three size bins (0.03–0.5, 0.5–5.0, and 5.0–20.0 µm) and calculated for 60 vertical model levels with the upper model limit at 0.1 hPa . To our knowledge, this is the only available global sea salt concentration data that span the period covered in this study. Furthermore, this reanalysis data set has performed well when compared to measured satellite and ground-based data .

Since sea spray aerosol consists mainly of chloride and sodium , we used the ion concentration of both elements in rain and OP as proxies of sea salt atmospheric inputs into the ecosystem. Because sodium is a conservative ion in sea salt aerosol, it is often used as a reference for sea salt concentration in precipitation chemistry and atmospheric chemistry modelling studies . Chloride is more unstable, as it photochemically reacts with other atmospheric ions (e.g. sulfur and nitrogen species), and it is depleted as a function of time spent in the atmosphere .

Weekly sodium and chloride concentrations in water samples from rain and OP were weighted with the total weekly precipitation volume to calculate volume-weighted monthly mean concentrations (VWMMs). With the calculated VWMMs we compiled monthly time series of sea salt concentration for a 6-year time series from 2004 to 2009, which represented the temporal variation in the concentration at each altitudinal level of the study area. To identify differences in the distribution of sodium and chloride concentrations between the sites, we created boxplots of total concentration over the 6-year evaluated period at each altitudinal level. Additionally, we computed total volume-weighted means (VWMs) to compare our observations with those from other studies.

We analysed the relationship to other ions (K+, NH4+, Ca2+, Mg2+, SO42-, NO3-) using a principal factor analysis (PFA) to locate common transport histories and the likely origin of sodium and chloride. Before conducting the PFA, the data were normalized and scaled to achieve comparable distributions.

In coastal continental areas, the Na+ / Cl- molar ratio is typically that of sea salt . This ratio was calculated using the measurements from each altitude and serves as an indicator of the origin of both sodium and chloride concentration in precipitation. The ratio changes as a function of distance from the ocean, as chloride is photochemically depleted in the atmosphere.

Finally, to asses a likely impact of anabatic flows on the sodium and chloride budget, we calculated wind direction relative frequency plots.

The HYSPLIT model was used to generate back trajectories of air masses encompassing 10 days with a resolution of 1 day . Modeling was done using the openair package for R statistical language. The wind fields were derived from the ERA-Interim reanalysis with a grid resolution of 0.75∘ by 0.75∘. All trajectories had their origins at the San Francisco River catchment in southern Ecuador. The MACC reanalysis sea salt concentration data were set as proxy of sea salt concentration in the atmosphere for air-mass transport analysis by back-trajectory techniques. To test the link between the MACC sea salt concentration and the sodium and chloride concentrations actually measured in rain and OP, both were linearly correlated. Pearson's product–moment correlation coefficients were calculated between the concentration at the two uppermost MSs (El Tiro and Cerro del Consuelo) and the MACC sea salt concentration (see Table ). Based on the correlation coefficients, the MACC data set at 700 hPa and the medium particle size (0.5–5.0 µm) was chosen as the input parameter for further examination by back-trajectory analysis, because it yielded the highest correlation coefficient and significance level.

Given the spatial resolution of the data set (0.75∘ by 0.75∘), the outcome of this analysis can only provide evidence of synoptic transport pathways and source–receptor relationships for medium- to long-distance sources for an area of approximately 80 by 80 km2 in the southern Ecuadorian Andes. Local-scale transport is not represented by the used trajectory models.

We first aimed to identify the potential geographic origin of the sea salt concentration over this wider area covering our receptor site in southern Ecuador. For this particular purpose we used source–receptor modelling, as it has been successfully applied to determine likely geographic origins of pollutants and aerosols e.g.. Here, two different hybrid receptor models were used for comparison: the potential source contribution function (PSCF) and the adjusted concentration-weighted trajectory (CWT) running on a grid that covers the domain of the 2192 generated trajectories between 2004 and 2009. Given the high seasonality of synoptic air mass transport, we calculated the models on a seasonal basis .

The PSCF calculates the probability that a source of aerosol or pollutant observed at the ground measurement site is located at a specific cell in the geographic space and is defined by PSCFij=mijnij, where nij is the number of trajectory points that passed through cell (i,j) and mij is the number of times that trajectory points passing through the cell (i,j), and correspond to a high sea salt concentration (above an arbitrary threshold) at the time of the trajectory's arrival at the receptor site. The function is based on the premise that, if a source is located at that specific location, the air masses represented by the trajectory passing through the collocated cell are likely to collect and transport the material along the trajectory until the receptor site. In this study, we defined two concentration thresholds: the 75th percentile for moderate-to-high concentration and the 90th percentile for high concentration.

The adjusted CWT function uses a grid domain to calculate a grid-wise logarithmic mean concentration of an aerosol or pollutant and is defined by ln⁡C‾ij=1∑k=1Ntij∑k=1Nln⁡C‾ktijk, where i and j are the grid indices, k the trajectory index, N the total number of trajectories, C‾k the pollutant concentration measured upon arrival of trajectory k, and tijk the residence time of trajectory k in grid cell (i, j). In this method, a weighted concentration is assigned to each pixel in the domain. This concentration is the average of the sample concentrations at a given receptor that have associated trajectories crossing the respective cell.

In a second step, we assessed the contribution of the main transport pathways of sea salt to the observed concentration over southern Ecuador. For this purpose, we integrated the MACC sea salt data to the back-trajectory cluster analysis. As for the source–receptor modelling approach, cluster analysis was applied on a seasonal basis to group similar air mass histories. This revealed general circulation patterns and, subsequently, enabled to post-process concentration data in relation to cluster origin and pathways. A partitioning algorithm based on spherical k means was used to define the appropriate number of trajectory clusters, as well as prior knowledge of the main wind systems affecting the receptor site. We tested different k values and chose the maximum number of cluster objects (i.e. back trajectories) that most closely reproduced the known conditions. The cosine distance was used as measure of similarity/dissimilarity between different trajectories. Afterwards, the frequency of trajectories represented by each cluster was determined.

To estimate the contribution of the different seasonal transport pathways to the observed sea salt concentration, the single trajectories belonging to each cluster object (cluster mean trajectory) were related to the sea salt concentration in the nearest neighbouring pixel within the study area. In this way, the contribution of each cluster object to the sea salt concentration above the study site could be statistically evaluated. Likewise, to analyse sources and sinks of sea salt along the cluster mean trajectories, we extracted the mean seasonal concentration values from the MACC data pixels (from 2004 to 2009) that matched the cluster mean trajectories in location, time, and height. Seasonal sea salt concentrations maps were calculated to further interpret the concentration along the cluster mean trajectories. For this, the MACC sea salt data were vertically integrated between 875 hPa (the minimum height of the trajectory clusters) and 500 hPa (maximum height). Additionally, based on findings by showing that burning biomass is a contributor to chloride emissions, the Copernicus atmosphere monitoring system's (CAMS) global fire assimilation system (GFAS) was used to create seasonal maps of NOx biomass burning fluxes over South America. To estimate the Na+ and Cl- contribution from the biomass-burning aerosol mass, we calculated the Na+ and Cl- mass ratio to NOx based on emission factors from . In this way, seasonal fractions of biomass-burning Na and Cl relative to sea salt were calculated.

Our study area is characterized by the complex topography of the Andes (see Fig. ). Hence, temporal variation and distribution is of interest in our study of sodium and chloride concentrations in rain and OP at different altitudes and topographical locations.

Figure  (left column) depicts the time series of sodium and chloride concentrations at different altitudes. Cerro del Consuelo MS, situated at 3180 m, demonstrated the clearest temporal pattern in concentration, where the highest peaks occurred almost regularly between September and February (Fig. a). The highest concentrations of sodium were recorded at Cerro del Consuelo and El Tiro (2825 m) (Fig. a and b). Contrarily, chloride concentration peaks in OP were highest at the lowest MS, Estación Científica San Francisco (ECSF) (Fig. d).

To compare the respective distributions, the boxplots in Fig.  (right column) show the concentration of sodium and chloride for both rain and OP at every MS. Overall, no essential variations between the concentration at each MS could be observed except for chloride in OP and rainwater at ECSF (Fig. d), where reported values were much higher than those measured at other elevations. Regarding ion concentration in sodium and chloride species, a clear difference could be observed with chloride concentration in the interquartile range extending between 0.22 and 0.51 mg L-1, and sodium concentration extending between 0.06 and 0.20 mg L-1.

In rainwater samples, the concentrations of chloride were considerably higher than those of sodium at each MS. A larger range and higher extreme values were observed in the chloride concentration. Differences in sodium concentrations between MSs at different altitudes were negligible. The concentrations showed a slight increase at transect 1 (TS1) (median of 0.14 mg L-1) and El Tiro (median of 0.13 mg L-1) and decreased again at the highest station, Cerro del Consuelo (median of 0.07 mg L-1). Compared to the rain samples, OP contained a higher mean concentration of sodium and chloride but also a greater range in its distribution (Fig. , left column). The concentration of chloride was also considerably higher than that of sodium, with the highest mean concentration (median of 0.62 mg L-1) at the lowest MS, ECSF. Sodium concentration peaked at El Tiro (0.17 mg L-1). At TS1, the mean concentration was lowest (median Na+ 0.09 mg L-1 and Cl- 0.3 mg L-1) increased once again at the highest elevations, El Tiro and Cerro del Consuelo (median Na+ between 0.11 and 0.17 mg L-1 and Cl- between 0.33 and 0.35 mg L-1).

A PFA of every major ion concentration was conducted to gain insights into the origin of sea salt inputs for each MS (Table ). This analysis indicated four components that explain the majority of the variability in the data set. The load of sodium and chloride had a considerable bearing on either factor 1 or 2, depending on the altitude and location of the MS, and the precipitation type. These two factors explained at least 29 % of the variability in the system (Table ).

Sodium and chloride explained the greatest variability in the system's rain samples, except for TS1 (2660 m).

At Cerro del Consuelo, sodium, chloride, and potassium dominated the variability in rain, given that they loaded to factor 1. In OP samples, biomass-burning compounds such as nitrate, sulfate, and ammonium had a stronger signal, loading to factor 1. Factor 2 was loaded by sodium and chloride only. No other compounds loaded to this factor, meaning that sodium and chloride most likely originated in sea salt from Atlantic and Pacific air masses.

At El Tiro, sea salt sources were exclusively present in factor 1 for rain and factor 2 for OP, similar to Cerro del Consuelo. As for Cerro del Consuelo, sea salt explained most of the variance in rain, followed by biomass-burning compounds. The opposite was true for OP, where biomass-burning compounds dominated, followed by sea salt.

The situation at TS1 was more complex than at Cerro del Consuelo and El Tiro, given the combined influence of the local mountain-valley breeze system and the synoptic system. In rain, biomass-burning compounds dominated the variability, with significant loadings on factor 1. Factor 2 was only loaded with sodium and chloride. In OP, factor 1 represented sea salt and crustal material, as sodium, chloride, and calcium loaded to this factor.

In the rain samples at ECSF, sodium and chloride loaded to factor 1 together with K+. In the OP samples nitrate, SO4-, and K+ dominated the variability loading to factor 1, while sodium and chloride loaded to factor 2.

Figure a shows the Na+ / Cl- molar ratio calculated in data from OP and rain water samples collected at each MS along the altitudinal gradient studied. The typical molar Na+ / Cl- ratio in precipitation for areas close to the sea is 0.86 (dotted line in the figure), according to . The International Co-operative Programme on Assessment and Monitoring of Air Pollution Effects on Forests (ICP) recommends an acceptable range of values between 0.5 and 1.5 molar units (dashed line in Fig. ). Outliers were likely due to samples with concentrations too close to the detection limits. When approaching the lower limits of the concentration, the ratio becomes more unstable and tends towards more extreme values.

The highest stations show a stronger marine influence, particularly in OP. The ratio fluctuates within the acceptable range and is mostly close to the standard value of sea salt in precipitation from coastal areas. This influence diminishes as the altitude decreases, especially for OP. Median ratios of 0.7, 0.8, 0.5, and 0.3 for Cerro del Consuelo, El Tiro, TS1, and ECSF, respectively, also reflect a greater influence of sea salt at higher altitudes.

A somewhat stronger seasonal behaviour was identified at the two highest stations (grey columns show the period from September to February, with the highest frequency of intrusion by the easterlies). Figure b depicts the frequency of trajectories on a yearly basis. In the first 3 years (2004–2006) the occurrence of westerlies was more frequent and at the same time the Na+ / Cl- ratio was close to that of fresh sea water (local Pacific influence). During 2007–2009, when easterlies were more frequent, the Na+ / Cl- ratio increased due to the increasing influence of distant Atlantic sources (chloride is depleted during transport) and the likely contribution of forest and agricultural fires .

Locally driven winds, such as thermally induced anabatic winds, can contribute to the transport of local sodium and chloride from the valley to the upper parts of the catchment. In a previous study, however, showed that anabatic winds do not impact MSs located on mountain tops, where synoptic winds predominate. Figure a and b show relative frequencies of the wind direction at ECSF and Cerro del Consuelo, respectively. At the lower altitudes (ECSF) a typical mountain-valley breeze circulation system exists, while at the crest (Cerro del Consuelo) north-easterly wind directions predominated.

In the previous section, we analysed the temporal and altitudinal variation of sodium and chloride concentrations in deposition driven by rain and OP. This section addresses the remaining question of where the sodium and chloride source areas are located geographically.

The synoptic wind system over South America is driven by strong seasonal circulation patterns. Because the air mass transport to the receptor site is directly linked to the seasonal cycle of the large-scale circulation system and thus, sources of sea salt concentration and their intensity may vary with the seasons, we examined seasonal patterns present in the sources and dominant air mass trajectory clusters.

We first evaluated potential contributory sources to sea salt concentration at the receptor site for each season. For this purpose, the two hybrid receptor models were used as shown in Fig. . In accordance with a sensitivity analysis done with back trajectories starting at different altitudes, these functions were applied to 3180 m, the altitude of the MS Cerro del Consuelo, on top of the highest peak in the catchment. Trajectories starting at lower altitudes have greater uncertainty because local flows are driven by the complex topography and cannot be reproduced. Those starting at higher altitudes provided no further information as they have coincidental source areas.

Figure  shows the spatial distribution of potential sources calculated by the PSCF (a, b) and the CWT (c), for DJF, MAM, JJA, and SON at 3180 m starting height at the receptor. When we compare the spatial distribution of sources between the two models (PSCF and CWT), similar locations in the Atlantic and Pacific oceans are indicated. The highest likelihood (concentrations above 5e-9 for Fig. a and b and above 5e-9 for Fig. c) is an equatorial Pacific location, which points to stronger sources of sea salt in that region contributing to the high concentration at the receptor site.

Strong sources of sea salt are expected from either the Pacific or the Atlantic. To judge from the high probability that the concentration stems from the oceans, the results of the PSCF (90th percentile concentration threshold, Fig. b) and the CWT (Fig. c) performed best in discriminating between potential geographic sources that contributed to moderate and high sea salt concentrations over southern Ecuador. In contrast, when using the 75th percentile as the concentration threshold (Fig. a) the PSCF only detected the transport pathways for sea salt irrespective of the intensity of the source contribution to the concentration.

Seasonal source contributions that had the greatest impact, i.e. responsible for high sea salt concentrations at the receptor site, occurred between September and February (September, October, and November (SON) and December, January, and February (DJF)). During SON, the equatorial Pacific was the dominant source, while in DJF both the Pacific and Atlantic sources contributed to the concentration. Yet, the Pacific sources still appeared stronger, as indicated by the large number of high values in that area. Furthermore, chlorine-containing compounds related to sea salt and biomass burning were likely co-linearly transported during that season, as shown by the high concentration over the northern portion of South America during DJF, and the coincidence of the biomass-burning season in that area (Fig. ). On the other hand, during austral autumn (MAM) and winter (JJA) the models identified no relevant potential sources of high sea salt concentration.

After locating the potential geographic sources of sea salt, trajectory cluster analysis was applied to identify the main representative air mass transport patterns, and thus the transport pathways of sea salt (Fig. ). Here, the easterlies were dominant. In the air mass transport, fast flowing east trajectories dominated (from approximately 83 to 97 % of the trajectories, in DJF and JJA, respectively), and slower-moving trajectories from the west appeared rather sporadically (between approximately 2.8 and 17 %). The occurrence of bow-shaped trajectories was common (Fig. , MAM and DJF) and characterized the coastal wind system associated with the Humboldt current .

Westerlies mostly evolved during SON and DJF, and to a lesser extent during MAM. Meanwhile, north easterlies were absent during the austral winter (JJA) following the displacement of the ITCZ to the north. The eastern clusters exhibited no seasonal pattern, because they represent the prevailing wind directions throughout the year.

Table  summarizes the mean sea salt concentration over southern Ecuador related to each cluster object reaching the receptor site for each season. High concentrations of sea salt are associated with westerly and north-easterly trajectories mainly occurring between September and May (Table ), whereas easterly air masses that passed over the Atlantic and continental South America before arriving at the receptor site showed intermediate to lower concentrations. In addition to the cluster-associated mean sea salt concentration, values in parenthesis in Table  describe the proportion (in percentage) that each cluster contributes to the total concentration during the study period. Cluster C2, representing the north easterlies, was associated with the highest contributions in DJF and MAM. In SON, cluster C2 represented the easterlies and was likewise associated with the highest contributions. SON is the main biomass-burning season in the Brazilian Amazon, which likely contributed to the overall budget. Furthermore, easterlies' transport from September to May was associated with approximately 75–80 % of the total concentration. The remaining 15–20 % were attributed to air flows passing over the Pacific before reaching the receptor site. These highly loaded seasonal Pacific flows took place in the Southern Hemisphere's late spring and summer as easterlies weaken due to the southward shift of the ITCZ. Atlantic air masses contributed to the concentration constantly over the year, also in austral winter (JJA), when the Pacific flows were negligible. However, transport from the Pacific clearly dominated the high peaks at the end and the beginning of each year. The sea salt concentration associated with the easterly clusters was much weaker. However, due to its high frequency, it persistently contributed to the transport from the Atlantic with likely additions from Amazon fires. Similar seasonal patterns were also identified in the measurements as illustrated in Fig. , the most clearest of which occurred at the highest station, Cerro del Consuelo (Fig. a). That means that the observed patterns in the measurements can be explained by the large-scale atmospheric circulation patterns.

Figure  depicts the sea salt concentration along the cluster mean trajectories and the trajectory height for each season. The north equatorial Atlantic was a great source of sea salt during DJF (Fig. , column 1), according to the sea salt concentration along the trajectory clusters. Nonetheless, the concentration rapidly decreased as the air masses travelled over the continent. Compared to westerly air masses, easterly air masses were lower in elevation, which increased the probability that they were loaded with aerosols from surface emissions. Those air masses then ascended abruptly as they approached the Andean range. In comparison, the equatorial Pacific is a less significant source for sea salt. Because of its vicinity to the receptor site and because the air masses spent most of the time over the ocean, the concentration did not sink significantly over time (Fig. , C4–C6). The concentration peaks in C1 and C2 were due to sea salt intrusions from the Caribbean Sea and canalized by the Andean cordillera, as depicted in Fig. a. The season DJF was also characterized by frequent forest and agricultural fires in northern South America, which likely contributed chlorinated compounds from biomass burning to the budget as well (Fig. ).

A similar situation also occurred in MAM (Fig. , column 2), where C1–C3 are north-easterly air masses and C4–C6 represent westerly pathways. In C2 the intrusion of sea salt from the Caribbean was also present, but less pronounced than in DJF (Fig. ).

For SON (Fig. , column 3) most of the budget was transported from the Atlantic and the continent: clusters C1, C2, and C3. Because this period coincided with the Brazilian biomass-burning season (see Fig. ), a considerable quantity of sodium and chloride from the Atlantic (Fig. a) and from fire emissions were probably transported to the receptor site.