The measurement site (49∘6′10.54′′ N, 8∘24′26.07′′ E) is in the upper Rhine valley 175 m north of the village of Leopoldshafen and 10 m east of the tramline no. S1/11 described by Hagemann et al. (2014) (marked with a star in Fig. 1). The Rhine valley often acts as a wind channel, leading to wind directions mainly from southwest or northeast. As shown in Fig. 1, the measurement site is located ∼11 km north of downtown Karlsruhe and ∼5 to 15 km northeast of an industrial area, including a refinery with 15.5 Mt yr-1 capacity, a 1450 MW hard-coal-fired power plant, and a 365 MW gas-fired power plant (Hagemann et al., 2014). Therefore, the measurement site is potentially subject to urban and industrial emissions, especially for southerly wind directions. Apart from that, there are several nearby biogenic emission sources, e.g., forests, sunflower fields, and other agricultural areas (Hagemann et al., 2014; Huang et al., 2019). Local traffic emissions from the minor road near the measurement site were identified by measurements with high time resolution. Its contributions to the aerosol load are small due to the generally low traffic.

The instrumentation for this campaign has been described in detail in previous publications (Huang et al., 2019; Shen et al., 2019a). Briefly, the data we use here are from the 6-week (26 July to 31 August 2016) deployment of a laser ablation aerosol particle time-of-flight mass spectrometer (LAAPTOF; AeroMegt GmbH) and a high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS; Aerodyne Inc.; hereafter AMS). These instruments were used to provide real-time information on size and chemical composition for individual particles and bulk samples, respectively. The LAAPTOF is a commercially available single-particle mass spectrometer and has been well described in recent publications (Gemayel et al., 2016; Marsden et al., 2016; Ramisetty et al., 2018; Shen et al., 2018). In brief, aerosol particles with a size range of 70 nm to 2.5 µm in vacuum aerodynamic diameter (dva) are sampled with a flow rate of ∼ 80 cm3 min-1, then focused and accelerated via an aerodynamic lens (ADL). Afterwards, the particles of ∼ 200 nm to 2.5 µm dva can be detected with two laser diodes (λ=405 nm) mounted in the particle time-of-flight (PTOF) chamber. Once a single particle is detected successively by both of the detection lasers, its dva is calculated from the time of flight and an excimer laser (λ=193 nm) is triggered to desorb and ionize the refractory and non-refractory species of the particle. The resulting cations and anions are analysed with a bipolar ToF mass spectrometer. The AMS measures the particle size-resolved bulk chemical composition, e.g. organic compounds, ammonium, nitrate, sulfate, and chloride, in mass concentrations with high time resolution and sensitivity (DeCarlo et al., 2006). Similarly as the LAAPTOF, aerosols are sampled with a flow rate of ∼84 cm3 min-1 via a similar ADL (Peck et al., 2016) in the size range 70 nm to 2.5 µm dva and then pass through a PTOF chamber. At the end of the PTOF chamber, particles encounter a 600 ∘C heater that vaporizes the non-refractory species. The corresponding vapours are ionized by electron impact with high energy of 70 eV, generating positive ions that are analysed with a ToF mass spectrometer.

All sampling inlets (PM2.5, PM10, and total suspended particulates) were positioned at the top of a measurement container (3.7 m above ground level, a.g.l.). Particle number (size >2.5 and >7 nm; CPC3776 and CPC 3022A, TSI Inc.), particle mass (PM2.5, PM10, and total suspended particulates; FIDAS, PALAS GmbH), black carbon (AE33-7 Aethalometer, Magee Scientific Co.), and the trace gases O3 (O341M, Environment SA), SO2 (AF22M, Environment SA), NO2 (AS32M, Environment SA), and CO2 (NGA2000, Rosemont Inc.) were measured. The meteorological parameters temperature, relative humidity, solar radiation, wind direction, and wind speed (WS700, Luft GmbH) were measured 10 m a.g.l. and complemented by meteorological data from a 200 m tower located at the KIT campus north, ∼2 km east of the measurement site (49∘5′43.0′′ N, 8∘25′46.0′′ E).

Single-particle data were analysed by using the Igor LAAPTOF data analysis software (version 1.0.2, AeroMegt GmbH). The corresponding basic procedures (e.g. mass calibration and fuzzy c-means clustering) and particle mass quantification method have been described in detail in our previous studies (Shen et al., 2018, 2019a). The AMS data analysis was done with the standard software written for Igor, including SQUIRREL (version 1.60C) and PIKA (version 1.20C). To account for a particle bouncing effect, we applied a composition-dependent collection efficiency (CE, the product of net particle transmission and detection efficiency) to the entire AMS dataset (Canagaratna et al., 2007; Middlebrook et al., 2012). In order to investigate the potential sources of organic aerosols, we have done a positive matrix factorization (PMF) analysis for AMS high-resolution (HR) spectra of organic compounds by using the PMF Evaluation Tool (PET version 3.00D) (Ulbrich et al., 2009). Detailed protocols were reported by Ulbrich et al. (2009) and DeCarlo et al. (2010). In brief, the HR input matrices without isotopes were created using PIKA. A five-factor PMF solution with FPEAK = 0 was chosen according to the following criteria, including least residuals (unexplained data), Q/Qexp≈1 (the quality-of-fit parameter Q is the sum of the squares of the scaled residuals; Qexp is the expected Q), and distinguishable features (e.g. spectra and diurnal patterns) of the components. It should be noted again that although fuzzy clustering and PMF analyses are commonly used methods in source apportionment, they cannot provide information on aerosol origin and the interaction between aerosol transport and transformation processes, which can be studied with aerosol transport models.

The regional transport model COSMO-ART is used in this study to simulate an episode with a high mass concentration of organics (up to ∼25 µg m-3 measured by AMS). Detailed model descriptions can be found in previous studies (Bangert et al., 2011; Knote et al., 2011; Stanelle et al., 2010; Vogel et al., 2009). Therefore, we will only give a brief review here. COSMO-ART is a fully online coupled model system, which uses the same spatial and temporal resolution for all scalars, such as temperature, gas, and aerosol particle concentrations. The meteorological module is based on the operational weather forecast model COSMO of the German Weather Service (Deutscher Wetterdienst, DWD). When atmospheric chemistry and aerosol dynamics are coupled to COSMO, it allows for the online calculation of the transport of reactive trace gases and aerosol particles as well as their interaction with the state of the atmosphere. Physical processes, e.g. nucleation, coagulation, condensation, and sedimentation, are included in the model. The aerosol particles are represented by five interactive lognormal modes representing particles consisting of internally mixed sulfate, ammonium, nitrate and organic compounds, pure soot, and internal mixtures of them all. The hourly biogenic emissions are calculated online according to land use data, the simulated temperatures, and radiative fluxes (Vogel et al., 1995). Anthropogenic emissions, including traffic emissions, large point sources, and area sources such as industrial and household areas, were precalculated by Johannes Bieser at the Helmholtz Zentrum Geesthacht (Backes et al., 2016; Bieser et al., 2011). In our study, the simulation domain is covered by a 350×330 grid with a horizontal resolution of 1×1 km2 and 50 vertical atmospheric layers (nonlinear increase up to 22 km a.g.l.). Most of the simulation results shown in this paper correspond to the lowest atmospheric level (10 m a.g.l.) and we indicate if results are from higher levels. The hourly emissions of SO2, NOx, CO, NH3, anthropogenic volatile organic compounds (VOCs), and primary organic aerosols (POA) were precalculated. As shown in Fig. S1 in the Supplement, VOCs, NOx, and POA are mainly emitted at a lower atmospheric level in urban areas, such as the cities of Karlsruhe, Stuttgart, Mannheim, and Strasburg, as well as the corresponding highways. SO2 hot spots typically correspond to the stack heights of the power plants emitting, such as the one in the western part of Karlsruhe.

Table 1 and Fig. S2 give an overview of meteorological conditions, trace gases, and particle concentrations for the entire measurement campaign. In summer 2016, the major wind direction was southwest (174±75)∘ and the average wind speed was (1.3±0.9) m s-1 at the measurement site in the upper Rhine valley. This wind speed allows trace gases and particles to be transported over a distance of 10 km within 2 h, e.g. from the city of Karlsruhe to the measurement site. During the measurement time sunrise and sunset were at around 06:30 and 20:30 CEST (Central European Summer Time), respectively. Temperature and relative humidity (RH) showed relatively stable anti-correlated diurnal patterns. Given the frequent clear nights, night-time temperature inversions occurred for 32 out of 37 nights during the entire measurement period (Fig. S3). Among the trace gases, CO2 and O3 showed clear diurnal patterns, with ozone peaking in the afternoon. NO2 concentrations were higher in the morning hours and in the late evenings and were generally anti-correlated with O3. NO2 and particle number showed spikes at similar times (more frequently during the daytime), mainly due to local emissions from nearby road traffic. The averaged particle number concentration of (1.2±0.8)×104 cm-3 for particles larger than 2.5 nm measured by CPC3776 in this study is comparable with that observed at the same site in previous tram-based measurements (Hagemann et al., 2014). SO2 concentrations showed peaks with low frequency (up to >10 ppb) but only when the wind was coming from the direction of the industrial area in the southwest. As shown in Fig. S2, these SO2 peaks correspond to higher particle number concentrations at around noontime when the wind speed is higher (∼1.7 m s-1) and vertical mixing intensified. This is in agreement with the downward mixing of stack emissions from industrial sources, which can be validated by model calculations using COSMO-ART (refer to Sect. 3.2).

In the entire measurement period, the total organic compounds, sulfate, nitrate, and ammonium were analysed to contribute on average (58±12) %, (22±7) %, (10±1) %, and (9±3) % to the total non-refractory PM2.5 mass measured by AMS, respectively. Given the AMS-PMF results shown in Fig. S4, five OA factors (components) were determined with characteristic mass spectra, corresponding elemental ratios (O : C and H : C), and distinct temporal variations. These were (9±7) % HOA, (16±11) % SV-OOA, and (75±15) % LV-OOA, i.e. (16±15) % LV-OOA1, (37±17) % LV-OOA2, and (21±13) % LV-OOA3, contributing to the OA mass. The HOA and OOA are linked to primary and secondary organic aerosol (POA and SOA), respectively (Ulbrich et al., 2009). Thus, SOA dominated the OA mass at this measurement site. Further descriptions of PMF results can be found in the Supplement (Fig. S4) and more discussions are given in Sect. 3.4.2, especially regarding the organic compounds.

In a previous publication (Shen et al., 2019a), we could identify seven major particle classes at the measurement site among ∼3.7×105 single particles with sizes between 200 nm and 2.5 µm dva. In brief, class 1 is identified as calcium-rich and soil-dust-like particles (calcium–soil) with marker peaks at m/z 40 Ca+, 56 CaO+, 57 CaOH+, 75 CaCl+, 96 Ca2O+, and 112 (CaO)2+. This particle class shows a good correlation (Pearson's correlation coefficient γ≥0.6) with the laboratory-derived reference spectra of soil dust particles. Class 2 is identified as aged-soot-like particles (aged soot) with characteristic Cn± patterns and is mixed with sulfate and nitrate. Class 3 is identified as sodium-salt-like particles (sodium salts) with marker peaks at m/z 23 Na+, 81 / 83 Na2Cl+, 139 / 141 Na3Cl2+, 165 Na3SO4+, 35 / 37 Cl-, 93 / 95 NaCl2-, 115 Na(NO2)2-, 131 NaNO2NO3-, and 147 Na(NO3)2-. Class 4 includes secondary-inorganic-rich and amine-containing particles (secondary inorganic–amine) with prominent ammonium, nitrate and sulfate markers, and amine marker peaks at m/z 58 C2H5NHCH2+, 59 (CH3)3N+, 86 (C2H5)2NCH2+, 88 (C2H5)2NO/C3H6NO2+, and 118 (C2H5)2NCH2+. Class 5, aged-biomass-burning and soil-dust-like particles (biomass burning–soil), has aged-biomass-burning marker peaks at m/z 39 K+, 213 K3SO4+, 46 NO2-, 62 NO3-, and 97 HSO4-, as well as peaks related to aromatic compounds (50 C4H2+, 63 C5H3+, 77 C6H5+, 85 C7H+, 91 C7H7+, 95 C7H11+, 104 C8H8+, 115 C9H7+). Furthermore, class 5 shows a strong correlation with reference spectra of soil dust particles. The representative spectra of class 6, aged-biomass-burning and organosulfate-containing particles (biomass burning–organosulfate), also features aged-biomass-burning particles and has organosulfate marker peaks at m/z 141 C2H5OSO4-, 155 C2H3O2SO4-, and 215 C5H11OSO4-. Class 7 particles contain all the marker peaks from the other six classes and show a good correlation with the reference spectra of mineral and soil dust particles; they are consequently labelled as “mixed/aged dust” (short for mixed/aged and dust-like particles). These particle types were not always present in similar fractions but showed distinct periods with one or two types dominating the particle mass (Shen et al., 2019a). Here, we will mainly discuss two particle classes that dominated the total particle mass, i.e. sodium salts and biomass burning–soil particles, contributing on average (36±27) % and (30±18) % to the PM2.5 mass measured by the LAAPTOF, respectively. We will also discuss one characteristic particle class, i.e. secondary inorganic–amine particles, which showed relatively intensive and distinct peaks three times during this campaign (Fig. S2) as well as several weaker peaks.

As shown in Fig. 2a, total particle mass concentrations have clear diurnal patterns (the data are averaged over the entire campaign), showing higher night-time values and a peak during the morning rush hour (∼ 09:00 local time). Higher particle mass values mainly correspond to southern wind directions, with one exception of extremely high masses associated with northwesterly winds (Fig. 2c). Specifically, sodium-salt-containing particles have higher fractions during daytime and correspond to southwesterly wind directions (Fig. 2b). Such wind was observed mainly during daytime and with higher wind speeds than at night. Biomass burning–soil particles enrich during night-time and mainly correspond to southeast wind (Fig. 2d). Secondary inorganic–amine particles typically rise in the morning and correspond to south and southwest wind directions (Fig. 2). When we compare weekdays (Monday to Friday) and weekends (Saturday and Sunday) in Fig. S5, the sodium salt particles were always correlated with southwest wind directions. Obvious differences can be found for secondary inorganic–amine and biomass burning–soil. Secondary inorganic–amine-type particles were mainly advected by south and southwest wind during weekdays. However, they were much less abundant during weekends and did not show a clear relationship with wind directions (only a small peak for northwest). For biomass burning–soil particles the observations are a bit more complex.

The biomass burning–soil particles measured during weekdays were correlated with south and southwest wind as well, indicating potential sources from the city of Karlsruhe and its industrial area. Please note that the representative spectra of these particles show aromatic ion peaks, which may arise from aromatic compounds such as toluene and xylene due to motor vehicle emissions. This type of particle also contains some metals such as barium and lead that are related to coal combustion from industry (Hudson et al., 2004; Xu et al., 2018). These features correspond to urban and industrial sources. During the weekends, biomass burning–soil particles had higher mass concentrations than during weekdays and were correlated with wind directions from the east, southeast, and northwest, where there are mainly forests and agricultural areas. Therefore, we can infer that there were various sources for biomass burning–soil particles and these sources differed between weekdays and weekends. This could be e.g. barbecue fires during weekends, while the weekday source is likely to be associated with that for secondary inorganic–amine particles.

In order to refine this type of particle, we compared them with black carbon particles measured by an Aethalometer (AE33-7). The equivalent black carbon concentration was calculated by the mass absorption cross-sectional value of 11.7 m2 g-1 (Herich et al., 2011). They seem to show a similar trend in the time series with a triplet of peaks from 25 to 28 August (refer to Figs. S2 and S6). We found that biomass burning–soil and black carbon concentrations are well correlated only for the weekends, with Pearson's correlation coefficients γ≥0.6 (Fig. S6b), and poorly correlated for weekdays (γ<0.3). This points to different major sources for weekdays and weekends. In addition, the black carbon instrument measuring the light absorption at different wavelengths (λ) can be used to determine the absorption Ångström exponent (α) (Drinovec et al., 2015). For the entire campaign, the average α was 0.96±0.24, calculated by using a pair of wavelengths and their corresponding mass absorption cross sections (MAC = 11.7 and 23.6 m2 g-1 for λ=950 and 470 nm, respectively; Herich et al., 2011). Herich et al. (2011) and Zotter et al. (2017) have assigned α=∼0.9 to BC from road traffic emissions, while the α values for wood-burning BC were typically higher at ∼1.68 (Zotter et al., 2017). Our measurements indicate that traffic was the dominating source of BC. Considering the good correlation between BC and biomass burning–soil particles during the weekend, we could infer the corresponding traffic emissions that contributed to biomass burning–soil particles. All observations support the conclusion that this particle class has significant anthropogenic sources, which may vary between weekends and weekdays. It should be noted that there were no significant biomass burning organic aerosols (BBOAs) identified by AMS-PMF for the entire measurement period. The marker f60 (the mass fraction of m/z 60 to the total organics) of BBOA is very close to the background fraction of 0.003 (Crippa et al., 2014; Cubison et al., 2011). In addition to biomass burning, other potential sources for potassium- and sulfate-rich particles detected by LAAPTOF must also be taken into consideration, such as coal combustion (Xu et al., 2018). Therefore, further analysis is required to identify the different sources contributing to this particle class.

As mentioned before, the dominating particle classes varied for different time periods (Shen et al., 2019a). In the following we focus on an episode with particles rich in sodium salts to find out if sea salt particles can significantly influence the aerosol particle composition at this central European location about 800 km inland. Furthermore, we discuss another episode with relatively high organic mass concentrations to find out how these mass loads could form and exist for a relatively long time period. Both episodes are easily recognized from the time series of particles measured by LAAPTOF and the organic compounds measured by AMS (Figs. S2 and 3; see also Fig. 4 in Shen et al., 2019a), as well as the time series of OOA, in particular for SV-OOA, LV-OOA1, and LV-OOA2 (Fig. S4). Apart from the difference of dominating particle classes and particulate organics, the particle acidity indicator value, organonitrate fraction, and organic carbon oxidation state, as well as meteorological conditions were quite different between these two episodes. For particle acidity, the molar ratio of measured to predicted NH4+ from AMS measurement is commonly used as an indicator (Zhang et al., 2007). The predicted NH4+ value is the mass concentration required to fully neutralize sulfate, nitrate, and chloride measured by AMS. A lower ratio value represents higher acidity under certain conditions, i.e. inorganic species are dominating the particle mass and crustal elements have relatively low concentrations (Zhang et al., 2007). It should be noted that organic acids and crustal elements such as sodium can bias the results (Hennigan et al., 2015, and the references therein). Despite the large uncertainty for indicating acidity, we can still use such ratios to indicate special events, e.g. sodium-nitrate-rich particles will provide nitrate, leading to lower values. As shown in Fig. 3, this ratio reaches the lowest value during the sodium-salt-rich episode, while the value in the organic-rich episode is at the average level. If we assume the organic acids only have a minor influence on the particle acidity, the anti-correlation between the acidity indicator value and organic concentrations may be related to acid-catalysed organic particle formation. For the fraction of organonitrates, the AMS-derived ratio of NO+/NO2+ is commonly used as an indicator (Farmer et al., 2010; Kiendler-Scharr et al., 2016). In the episode rich in sodium salts, particles contain the highest organonitrate fraction, while the particles in the organic-rich episode contain the lowest fraction (refer to Fig. S7). It should be noted that NaNO3 can also result in higher NO+/NO2+ ratios (5 to 7; Allan et al., 2004) than NH4NO3 (2.3 in our calibration). However, the NO+/NO2+ values here are much higher (see Fig. S7) and thus we can expect a major contribution by organonitrates. Higher fractions of organonitrates suggest a stronger influence of anthropogenic emissions on the sodium-rich aerosol. The averaged carbon oxidation state (OS‾C) of the organic compounds is a robust metric of its degree of oxidation or processing (Canagaratna et al., 2015; Kroll et al., 2011). It turns out that the organics were more oxidized in the organic-rich episode, with relatively higher OS‾C value and less variability (-0.10±0.20), than that for the entire campaign (-0.19±0.23) (refer to Fig. S7 and Table 2). In addition, the major wind with higher speeds was from the southwest in the episode rich in sodium salt particles, and we will show in Sect. 3.3 that this is related to the long-range transport of sea salt particles. For the organic-rich episode, the wind speeds were lower most of the time. The different features of these two episodes are summarized in Table 2. We will discuss the episodes in more detail in Sect. 3.3 and 3.4, but we first demonstrate the usefulness of the transport model to identify an industrial point source.

As shown in Fig. 4, we have modelled O3, NO2, SO2, and particle number for the organic-rich episode with COSMO-ART. Model simulations show comparable results and/or trends with field observations. For SO2, there is a modelled peak at 10:00 CEST on 26 August. As mentioned above, in the upwind direction of our measurement site is a coal-fired power plant, from which SO2 is emitted from a stack at 230 m a.g.l. Therefore, we compared SO2 concentrations in an area including the power plant and our measurement site (refer to Fig. 5a1 and a2), as well as time series of their vertical profile at the measurement site (Fig. 5b). The model results show that SO2 emissions started at 02:00 CEST from the stack of the power plant, continuously transported by southwest wind (from 06:00 CEST; not shown here). Around noon, the SO2 was transported downward due to enhanced vertical mixing. Similar to SO2, we found corresponding vertical transport of NOx and primary particles (refer to Fig. 5c and d), as well as secondary inorganic particles, which will be discussed in Sect. 3.3.3. In addition to the power plant emissions, particle emissions from the Karlsruhe refinery (e.g. around 08:00 CEST at 122 m a.g.l.) and downtown Karlsruhe (at 10 m a.g.l.) were observed in the model results, which are visualized via videos of the evolution of particle number at these two altitudes (refer to the Video Supplement). Given these findings, we have verified industrial contributions to the particles measured at this rural site. However, the particle numbers calculated by the model are 10 times lower than the measured values. In this context, it should be noted that the observations relate to particles larger than 2.5 and 7 nm. The modelled particles include Aitken- and accumulation-mode particles with median diameters at around 30–80 and 300–400 nm, respectively. The model does not represent the ultrafine particles to their full extent, especially e.g. the contribution of organic compounds or ammonia to nucleation.

Sodium salt particles accounted on average for (35.6±26.7) % of the particle mass measured by LAAPTOF during the whole campaign, and they dominated several time periods (refer to Fig. S2). However, not all the peak events were the same. We chose a unique episode when sodium chloride signatures, e.g. m/z 81 Na2Cl+, showed the strongest signal, as plotted in Fig. 6a. In this episode, the sodium salts accounted for (70±24) % of the total mass of the single particles. To study the origin of these particles, we used the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT; Stein et al., 2015) on the READY website (http://www.ready.noaa.gov, last access: 17 October 2019; Rolph et al., 2017) for back-trajectory analysis. The corresponding results suggest that these particles came from the Atlantic Ocean, which is more than 800 km west of our measurement site (refer to Fig. S8a). Therefore, we can constrain the type of sodium salt particles as sea salts with an age of about 1.5 to 3 d.

In this episode, sea salt particles retained strong chloride signatures in spite of the long-range transport due to relatively high wind speeds (e.g. the average wind speed was ∼6 m s-1 from the ocean to our measurement site at noontime on 21 and 22 August based on a 72 h back-trajectory analysis). Therefore, we interpret them as relatively fresh sea salt, which is the first subclass of sodium salt particles identified with the fuzzy clustering analysis of the single-particle mass spectra. As shown in Fig. 6a, there is an obvious decrease in m/z 81 Na2Cl+ at the end of this episode, while the sodium nitrate signature at m/z 108 Na2NO3+ increases. Such trends can also be found in similar but much less prominent sodium-chloride-rich episodes (see the red guiding arrows in Fig. 6a). This indicates the ageing process of sea salt particles, namely heterogeneous replacement of chloride by nitrate on sea salt particles (Gard et al., 1998). We assigned the second subclass of sodium salts to more aged sea salt particles, which are bigger than the relatively young sea salt (Fig. 6b). A 72 h back-trajectory analysis suggests that these particles came from the Atlantic Ocean and were carried with southwesterly wind of relatively lower speed (∼3.5 m s-1) to reach our measurement site on 23 August, thus they spent more time (∼3 d) over the continent and are more aged. The representative spectra of sodium salt particles and the averaged spectra of their subclasses can be found in Fig. S9. The signal difference between aged and young sea salt spectra (aged – young) is shown in Fig. 6c, highlighting their signatures. Similar relationships as for m/z 81 Na2Cl+ and 108 Na2NO3+ can also be found for the other signature peaks (refer to Fig. S10). It is worth noting that there are also nitrate-related peaks in young sea salt, such as m/z 104 NaClNO2-, verifying that these young sea salts were not fresh. In addition, m/z 92 Na2NO2+ and 131 NaNO2NO3- increase earlier than 108 Na2NO3+ and 147 Na(NO3)2-. We can therefore infer the ageing degree from a series of signature peaks in LAAPTOF spectra: 81Na2Cl+→104NaClNO2-→92Na2NO2+ and 131NaNO2N3-→108Na2NO3+ and 147 Na(NO3)2- (from relatively fresh to more aged, corresponding to chloride to nitrite to nitrate salts). However, this requires further systematic studies to be validated.

For sodium salt particles, a weak but characteristic peak at m/z 129 C5H7NO3+ is likely to be an organonitrate signature (Shen et al., 2019a). As shown in Fig. S10, m/z 129 C5H7NO3+ has a similar trend as m/z 131 NaNO2NO3- and it is more pronounced in the LAAPTOF spectra of aged sea salt. In the sodium-chloride-rich episode, in general m/z 129 C5H7NO3+ has a similar trend as organonitrate mass concentrations estimated by AMS results and the results of a molecular analysis using FIGAERO-HR-ToF-CIMS (a filter inlet for gases and aerosols coupled to a high-resolution time-of-flight chemical ionization mass spectrometer) in a companion study by Huang et al. (2019). The method for organonitrate estimation can be found in Farmer et al. (2010) and the parameters used here are the same as those used by Huang et al. (2019). We have briefly described this method in the Supplement after the caption for Fig. S10. In addition, our simulation chamber study has shown that this m/z 129 C5H7NO3+ arises from the recombination of inorganic nitrate and organics due to the laser desorption and ionization (LDI) matrix effect in single-particle MS (Murphy, 2007; Reilly et al., 2000; Reinard and Johnston, 2008; Zenobi and Knochenmuss, 1998). It is even stronger when it arises from organonitrates formed under NOx-rich conditions (refer to Fig. S11). Taken together, we can conclude that the m/z 129 C5H7NO3+ signal mainly results from organonitrates, but we still cannot rule out the LDI effect. Furthermore, our simulation chamber studies also revealed that m/z 129 C5H7NO3+ in LAAPTOF spectra could be a signature for anthropogenic sources (refer to Fig. S11 and the corresponding detailed description in the Supplement). Therefore, m/z 129 C5H7NO3+ on sea salt particles indicates interactions between natural sea salt and anthropogenic emissions. It is worth noting that another weak but characteristic peak was observed at m/z 165 Na3SO4+. This signature peak for sea salt sulfate can be found with a stronger signal in young sea salt than in aged sea salt (Gard et al., 1998).

Apart from that, in high-resolution AMS spectra we have observed the peak at 78.99 CH3SO2+, which is the marker for methanesulfonic acid (MSA) from marine or coastal areas (Huang et al., 2015), although other sources cannot be excluded (Daellenbach et al., 2017). Based on the method developed by Huang et al. (2015), we have also estimated the MSA fraction (fMSA, signal fraction of MSA to total organics and sulfate) and mass concentrations for the entire campaign (refer to Fig. S12). It turns out that during the sodium-chloride-rich episode, the fMSA is enhanced compared to the other low-sodium-salt time period, such as the organic-rich episode. This also supports the fact that sodium salt particles measured by the LAAPTOF had a sea salt origin.

At the end of the measurement period, we observed three subsequent organic peaks (see Figs. 3, S2, and S7): one peak on Friday 26 August during daytime and two peaks during the following two nights. Organics measured in this episode exhibited the highest mass concentrations of (12.9±3.9) µg m-3, which accounted for (77±6) % of the total non-refractory particle mass and is almost 3 times more than organic masses measured in other periods of the entire AMS measurement period. Please note that some secondary inorganic components were also measured in this episode, showing some peaks at similar times as the high organic peaks. As for the sodium-chloride-rich episode, we have also done a HYSPLIT back-trajectory analysis for the organic-rich episode. Figure S8b shows these results: the air mass was coming from the continent during this episode.

In order to better interpret the measurements and to understand the impact of transport processes for the particles measured in the organic-rich episode, we have done COSMO-ART simulations covering the time period from 25 to 28 August 2016. Figure 7 shows the particulate organic compounds, the secondary inorganics (ammonium, nitrate, and sulfate) measured by AMS, and the result calculated using COSMO-ART (panels d to g). The corresponding meteorological conditions measured are shown in Fig. 7a to c. In general, the trends and also the amplitudes of the model results compare quite well with the observations. In the following, we will combine field observation and transport model simulation results to address these questions: (1) were the organic-rich aerosols due to local formation, regional transport, or vertical mixing? (2) Were all three organic peaks due to the same factors? (3) Did the secondary inorganics originate from the same sources as the organics? We will start the discussion with the inorganics; we will then go to the organics and end with a typical particle type in this episode.