Long-term sub-micron aerosol chemical composition in the boreal forest: inter- and intra-annual variability

15 The Station for Measuring Ecosystem Atmosphere Relations (SMEAR) II is well known among atmospheric scientists due to the immense amount of observational data it provides of the earth–atmosphere interface. Moreover, SMEAR II plays an important role in large European research infrastructures, enabling the large scientific community to tackle climate and air pollution related questions, utilising the high-quality long-term data sets recorded at the site. So far, the well-documented site was missing the description of the seasonal variation of aerosol chemical composition that is crucial for understanding the 20 complex biogeochemical and -physical processes governing the forest ecosystem. Here, we report the sub-micron aerosol chemical composition and its variability utilising data measured between 2012 and 2018 using an Aerosol Chemical Speciation Monitor (ACSM). We observed a bimodal seasonal trend in the sub–micron aerosol concentration culminating in February (2.7, 1.6, 5.1 μg m for median, 25, 75 percentiles, respectively) and July (4.2, 2.2, and 5.7 μg m for median, 25, 75 percentiles, respectively). The wintertime maximum was linked to an enhanced presence of inorganic aerosol species (ca. 25 50%) whereas the summertime maximum (ca. 80% organics) to biogenic secondary organic aerosol (SOA) formation. During the exceptionally hot Julys of 2014 and 2018, the organic aerosol concentrations were up to 70% higher than the 7–year July mean. The projected increase of heat wave frequency over Finland will most likely influence the loading and chemical composition of aerosol particles in the future. Our findings suggest strong influence of meteorological conditions such as radiation, ambient temperature, wind speed and direction on aerosol chemical composition. To our understanding, this is the 30 longest time series reported describing the aerosol chemical composition measured online in the boreal region, but the continuous monitoring will be maintained also in the future. https://doi.org/10.5194/acp-2019-849 Preprint. Discussion started: 8 November 2019 c © Author(s) 2019. CC BY 4.0 License.

Aerosol Chemical Speciation Monitor (ACSM). We observed a bimodal seasonal trend in the sub-micron aerosol 23 concentration culminating in February (2.7, 1.6, 5.1 µg m -3 for median, 25 th , 75 th percentiles, respectively) and July (4.2, 24 2.2, and 5.7 µg m -3 for median, 25 th , 75 th percentiles, respectively). The wintertime maximum was linked to an enhanced 25 presence of inorganic aerosol species (ca. 50%) whereas the summertime maximum (ca. 80% organics) to biogenic 26 secondary organic aerosol (SOA) formation. During the exceptionally hot Julys of 2014 and 2018, the organic aerosol 27 concentrations were up to 70% higher than the 7-year July mean. The projected increase of heat wave frequency over 28 Finland will most likely influence the loading and chemical composition of aerosol particles in the future. Our findings 29 suggest strong influence of meteorological conditions such as radiation, ambient temperature, wind speed and direction 30 on aerosol chemical composition. To our understanding, this is the longest time series reported describing the aerosol 31 chemical composition measured online in the boreal region, but the continuous monitoring will be maintained also in the 32 future. 33

ACSM measurements 193
The Aerosol Chemical Speciation Monitor (ACSM; Aerodyne Research Inc. USA) was first described by Ng et al. in 194 2011. It was developed based on the Aerosol Mass Spectrometer (AMS) (Canagaratna et al., 2007), but simplified at the 195 cost of mass and time resolution to achieve a robust instrument for long-term measurements. The ACSM samples ambient 196 air with a flow rate of 1.4 cm 3 s −1 through a critical orifice (100 µm in diameter) towards an aerodynamic lens efficiently 197 transmitting particles between approximately 75 and 650 nm in vacuum aerodynamic diameter (Dva) and pass through 198 particles further up to 1µm in Dva with a less efficient transmission (Liu et al., 2007). After this, the particles are flash 199 vaporized at a 600 °C hot surface in high vacuum and ionised with electrons from a tungsten filament (70 eV, electron 200 impact ionisation, EI). These processes lead to substantial fragmentation of the molecules forming the aerosol particles. 201 The resulting ions are guided to a mass analyser, a residual gas analyser (RGA) quadrupole, scanning through different 202 mass-to-charge ratios (m/Q). The detector is a secondary electron multiplier (SEM). The particulate matter detected by 203 the ACSM is referred to as non-refractory (NR) sub-micron particulate matter (PM1). The word 'non-refractory' (NR) 204 is attributed to the instrument limitation to detect material flash evaporating at 600 °C thus being unable to measure heat-205 resistant material such as minerals or soot. The word 'PM1' is linked to the aerodynamic lens approximate cut-off at 1 206 µm. Importantly, the NR-PM1 reported from these ACSM measurements is a difference between the signal of particle-207 laden air and signal recorded when the sampling flow passed a particle filter (filtered air). 208

209
The ACSM measurements for the current study were conducted within the forest canopy through the roof of an air 210 conditioned container. A PM2.5 cyclone was used to filter out big particles that could cause clogging of the critical orifice. 211 A Nafion dryer was installed in 2013 upstream the instrument ensuring a sampling relative humidity (RH) below 30%. 212 Before this, the RH was not controlled nor recorded. Thus, the RH was likely high during summer, but low during 213 wintertime. Moreover, a 3 litres per minute (Lpm) overflow, which was ejected only before the aerodynamic lens, was 214 used to minimise losses in the sampling line (length approximately 3 m). The data were acquired using the ACSM data 215 acquisition software (DAQ) provided by Aerodyne Research Inc., the instrument manufacturer. The DAQ version was 216 updated upon new releases. The ACSM was operated to perform m/Q scans with a 200 ms Th -1 scan rate in the mass-to-217 charge range of m/Q 10 Th to 140 Th. Filtered and particle-laden air were measured interchangeably for 28 quadrupole 218 scans resulting in ca. 30 minute averages. The air signal, obtained from the automatic filter measurements, was subtracted 219 from the sample raw signal, yielding the signal from aerosol mass only. The data processing was performed using ACSM 220 Local v. 1.6.0.3 toolkit within the Igor Pro v. 6.37 (Wavemetrics Inc., USA). Upon data processing, the different detected 221 ions were assigned into organic or inorganic species bins (i.e. total organics, sulphate, nitrate, ammonium and chloride) 222 using a fragmentation table (Allan et al., 2004). Moreover, the data were normalized to account for N2 signal variations 223 related to ACSM flow rate and sensitivity changes (due to SEM voltage response decay). where Cs is the concentration of species s, CE is the particle collection efficiency (see chapter 2.4 ACSM collection 229 efficiency correction), and Tm/Q the m/Q -dependent ion transmission efficiency in the RGA quadrupole mass analyser. 230 The Tm/Q is constantly recorded based on naphthalene fragmentation patterns and their comparison to naphthalene 231 fragmentation pattern in the NIST data base (75 eV EI; http://webbook.nist.gov/). Naphthalene is used as an internal 232 standard in the ACSM and is thus always present in the mass spectrum (Ng et al., 2011

Additional measurements 246
In addition to the ACSM-measurements, the SMEAR II station has a large number of other air composition related 247 measurements. In the current study, we investigate only a small fraction of them. analyser and MARGA-2S measurements were conducted within the forest canopy are described in more detail in the 262 sections below. The data availability is shown in Figure A.2. 263

DMPS 264
The Differential Mobility Particle Sizer (DMPS) measures the aerosol size distribution below 1 µm electrical mobility 265 diameter. SMEAR II holds the world record in online aerosol size distribution measurements (Dada et al., 2017), as the 266 measurements started already in 1996. The DMPS system is described in detail previously (Aalto et al., 2001). Briefly, 267 the SMEAR II DMPS is a twin DMPS setup that samples 8 m above ground from an inlet with a flow rate of 150 Lpm. 268 The measurement cycle is 10 minutes. The first DMPS (DMPS-1) has a 10.9 cm long Vienna type differential mobility 269 analyser (DMA) and a model TSI3025 condensation particle counter (CPC) that was changed to model TSI3776 after period. The DMA high voltages were also validated with a multimeter. The CPC concentrations were compared against 279 each other with size-selected ammonium sulphate particles in the 6-40 nm range as well as compared against the TSI3775 280 particle counter that measures the total aerosol particle number concentration at the station. The sizing accuracy of the 281 two DMAs were cross-compared with 20 nm ammonium sulphate particles. In addition, the accuracies of the RH, 282 temperature and pressure probes were validated each year.

Cascade impactor 284
The PM1 and PM2.5 (particulate mass of aerosol particles with an aerodynamic diameter below 2.5 µm) mass 285 concentrations measured between 2012-2017, which were included in the current study, were retrieved from the cascade 286 impactor measurements. This gravimetric PM10 impactor, produced by Dekati Ltd., is a three-stage impactor with cut-287 points at 10, 2.5 and 1 µm. The collection is conducted on greased (Apiezon vacuum grease diluted in toluene) Nuclepore 288 800 203 25 mm polycarbonate membranes with 30 Lpm flow rate, approximately 5 m above ground level. The filter 289 smearing was performed to avoid losses due to particle bouncing. The filters were weighed manually every 2-3 days and 290 stored in a freezer for possible further analysis. 291

PTR-MS 292
The monoterpene concentration was measured using the proton transfer reaction quadrupole mass spectrometer (PTR-293 MS) manufactured by Ionicon Analytik GmbH, Innsbruck, Austria (Lindinger and Jordan, 1998). The monoterpene 294 measurement setup is described in detail previously . Shortly, the PTR-MS was placed inside a 295 measurement cabin on the ground level and the sample air was drawn down from a measurement mast to the instrument 296 using a 157 m long PTFE tubing (16/14 mm o.d./i.d.). The sampling line was heated and the sample flow was 45 Lpm. 297 However, the sample entering the PTR-MS was only 0.1 Lpm. During the study period, the primary ion signal H3O + 298 (measured at isotope m/Q 21 Th) varied slightly around 5-30 × 10 6 c.p.s. (counts per seconds). The instrument was 299 calibrated every 2-4 weeks using three different VOC standards (Aper-Riemer) and the instrumental background was 300 measured every third hour using VOC free air, produced by a zero air generator (Parker ChromGas, model 3501). 301 Normalised sensitivities and the volume mixing ratios were then calculated using the method introduced previously 302 (Taipale et al., 2008). For example, the normalized sensitivity of alpha-pinene (measured at m/Q 137 Th) varied between 303 2 and 5 n.c.p.s. p.p.b. -1 over the study period. Only the signal of monoterpenes at m/Q 137 Th were analysed in the current 304 study. 305

Aethalometer 306
The concentration of equivalent black carbon (eBC) in the PM1 size range was measured by using two different Magee 307 Scientific Aethalometer models: AE-31 during 2012-2017, and AE-33 in 2018. The sample air was taken through an 308 inlet equipped with a PM10 cyclone and a Nafion dryer, and a PM1 impactor. Aethalometers determine the concentration 309 of eBC by collecting aerosols on a filter medium and measuring the change in light attenuation trough the filter. Both of 310 the Aethalometers quantify eBC concentration optically at seven wavelengths (370, 470, 520, 590, 660, 880 and 950 nm). 311 Only the eBC concetration determined at 880 nm was used in the current study. AE-31 data was corrected for a filter 312 loading error with a correction algorithm derived previously (Collaud Coen et al., 2013). A mass absorption cross section 313 of 4.78 m 2 g -1 at 880 nm was used in the eBC concentration calculation. The AE-33 used a "dual-spot" correction is 314 described previously (Drinovec et al., 2015). 315

OCEC-analyser 316
Organic carbon (OC) and elemental carbon (EC) concentrations were measured using a semi-continuous Sunset OCEC 317 analyser (Bauer et al., 2009) produced by Sunset Laboratories Inc. (USA). The aerosol sampling was conducted through 318 the same container roof as the ACSM. The inlet length was approximately the same as for the ACSM (ca. 3 m). The 319 sample flow was guided through a PM2.5 cyclone and a carbon plate denuder to avoid collection of large particles and a positive artefact introduced by organic vapours. In the OCEC, the sample is collected on a quartz-filter for 2.5 hours with 321 an 8 Lpm flow rate. The sampling procedure is followed by the analysis phase. The analysis phase includes thermal 322 desorption of PM from the filter following the EUSAAR-2 protocol (Cavalli et al., 2010), and introducing the aerosol 323 sample to inert helium gas that is used to carry the OC to a MnO2 oxidising oven. This leads to OC oxidation to CO2, 324 which is then quantified, with a non-dispersive infrared (NDIR) detector. Afterwards the remaining sample is introduced 325 to a mixture of oxygen and helium enabling EC transfer to the oven. The resulting CO2 from EC desorption and 326 combustion is also quantified using the NDIR detector. An additional optical correction was used to account for the 327 amount of pyrolysed OC during the helium phase. EC was also quantified using a laser installed in the analyser. This unit used for the current study is described in more detail previously (Makkonen et al., 2012). 342

ACSM collection efficiency correction 343
The ACSM data processing includes correcting for the measurement collection efficiency (CE) that is estimated to be 344 approximately 0.45-0.5 in average for AMS-type instruments (Middlebrook et al., 2012). The reduction is caused by 345 particle bouncing at the instrument vaporizer (Middlebrook et al., 2012). Middlebrook et al. (2012) provide a method to 346 estimate the CE, based on aerosol chemical composition. However, this method was not applicable to our data set due to 347 low, and thus noisy, ammonium signals that were most of the time near the instrument detection limit. Thus, we chose to 348 calculate the collection efficiency based on the ratio between the NR-PM1 (total mass concentration measured by the 349 ACSM) and a Differential Mobility Particle Sizer (DMPS)-derived mass concentration (after subtracting the equivalent 350 black carbon, eBC). The DMPS-derived mass concentration is determined as follows: 1) Calculation of aerosol volume 351 concentration (m 3 m -3 ) of the ACSM detectable size range, where the aerodynamics lens transmission is most efficient 352 (50-450 nm in electrical mobility = ~75-650 nm in vacuum aerodynamic diameter) and assuming spherical particles, 2) 353 Estimating aerosol density based on the ACSM-measured chemical composition (ρ(NH 4 ) 2 SO 4 = 1.77 g cm -3 , ρNH 4 NO 3 = 1.72 354 g cm -3 , ρOrg= 1.50 g cm -3 , ρBC = 1.00 g cm -3 ), 3) Calculating the mass concentration (µg m -3 ; mass concentration = density 355 × volume concentration). As direct scaling of ACSM data to the DMPS-derived and eBC subtracted mass concentration 356 is strongly not recommended, we chose to use two-month running medians of the ratio between the NR-PM1 and eBC-357 subtracted DMPS-derived mass concentration. The two-month running median approach diminishes the effect instrument 358 noise in the DMPS-derived mass concentration that could otherwise be introduced as additional uncertainty into the 359 ACSM-data. The two-month median CEs were within 10% of the annual mean values in years 2013-2018. In 2012 the 360 CE had stronger seasonal variation (16% variation around the mean, peaking in summer) likely due to the lack of the 361 aerosol dryer in the sampling line. The magnitudes of the CEs can be obtained from Figure  concentration and impactor PM1 is better after CE correction both due to increased correlation coefficients (R 2 ) and slopes 383 (k), the (two month running median) DMPS-based CE correction is justified. Hereafter, all the ACSM data presented and 384 discussed are CE corrected. We refer to it as NR-PM1. The CE correction method applied importantly also ensures more 385 quantitative year-to-year comparability of the ACSM data acquired as it also corrected for the overestimated calibration 386 values obtained during 2013-2015. 387

ACSM chemical speciation validation 388
To validate the ACSM chemical speciation process, the ACSM organics (Org) and sulphate (SO4) were compared against 389 the organic carbon (OC) measured by a Sunset OCEC-analyser (see section 2.3.5 OCEC-analyser for instrument) (  and Turpin, 2002;Russell, 2003). The linear regression is calculated using all the overlapping data from year 2018, when the OCEC was well functioning after instrument service. Regarding the water-soluble inorganic ions, only the SO4 2-398 concentration (in PM2.5), retrieved from the MARGA measurements, was used for the current analysis for ACSM data 399 validation purposes. The nitrate time series, for example, are known to be different between the two measurements at 400 SMEAR II (Makkonen et al., 2014). The scatter between the nitrate measurements, visualised also here, in Figure  First, we state that the ACSM data set was not long enough to provide sufficient statistics for investigation of long-term 441 trends of NR-PM1 or its components' loading, hence no analysis of such is presented here. In this section we discuss the 442 inter-and intra-annual variation in sub-micron non-refractory aerosol chemical composition at SMEAR II in 2012-443 2018. We first introduce the monthly scale behaviour and year-to-year variability. We briefly introduce two case studies, 444 one linked to elevated sulphate loading at the station due to a lava field eruption in Iceland, and another one discussing 445 the effect of heatwaves on PM1 loading and composition. Hereafter, we introduce the overall median diurnal profiles of 446 individual chemical species observed in the NR-PM1, and finally the chemical composition observations linked to wind 447 speed and direction observations above the forest canopy. 448

Inter-and intra-annual variation 449
The monthly median seasonal cycles of NR-PM1 and PM2.5 show bimodal distributions as the PM loading has two 450 maxima: one peak in February, and another one in summer (June, July, and August), the latter one being more significant.  Figure 5d). However, uncertainties can be attributed to the eBC concentration as scattering coatings, such as salts or even 510 photochemically aged SOA can also generate a lensing effect leading to an overestimation of eBC (Bond and Bergstrom,511 2006; Zhang et al., 2018). Such effect could lead to a substantial overestimation of eBC especially in summertime, when 512 the organic loading is highest. Some certainty of the BBOA and BC enhancement with high temperatures can however 513 be retrieved from Figure 5c, where carbon monoxide (CO) mixing ratio also increases at relatively high ambient 514 temperatures (T > 15 °C). CO is known to be emitted from incomplete combustion processes. Nonetheless, as the increase 515 in eBC visualised in Figure 5d is less drastic than for monoterpenes, we suggest the biogenic SOA production as the major 516 organic aerosol source in summertime. While the quantification and separation of BBOA from SOA will be the topic of 517 an upcoming independent publication centred on the analysis of organic aerosol mass spectral fingerprints at SMEAR II, 518 we briefly introduce the behaviour of f60. f60, which equals the contribution of m/Q 60 Th signal to the total organic signal 519 recorded by the ACSM, is a marker for levoglucosan-like species originating from cellulose pyrolysis in biomass burning 520

Diurnal variation of NR-PM1 composition 623
The year-to-year variation in the NR-PM1 monthly median seasonal cycles shows rather consistent behaviour throughout 624 the measurement period and even the overall 10 th percentile of the PM-data suits the bimodal trend discussed in the 625 section above. The 10 th percentile also agrees with the seasonal trends associated with individual NR-PM1 chemical 626 species, i.e. organics, sulphate and nitrate as well as their precursors (Figure 4) 7b&d). As wintertime PM is presumably mostly long-range transport, its components' diurnal patterns are less obvious 638 due their cumulative build-up in the atmosphere. For example, as sulphate aerosols, the most prominent inorganic species, 639 are long lived due to their low volatility, we do not expect sulphate to have diurnal variation in wintertime because of the 640 lack of major SO2 sources at SMEAR II's proximity. The ammonium mass concentration lacks diurnal pattern as well 641 and peaks at the same time of the year as sulphate. The degree of aerosol neutralisation by ammonia can be estimated by 642 the ratio between the measured ammonium and the amount of ammonium needed to neutralise the anions detected by the 643 ACSM (termed "NH4 predicted") (Zhang et al., 2007b). The overall ratio was 0.66 hinting towards moderately acidic 644 ammonium sulphate aerosols ( Figure A.7&Figure 7h), though the uncertainty in this value is high due to the low loadings 645 of ammonium at SMEAR II. We also acknowledge that the ratio between measured and predicted ammonium 646 concentration is not fully accurate for acidity estimations, and if such are needed, a better estimation could be provided 647 with thermodynamic models. The temporal variation of the ammonium balance does not show diurnal variability either, 648 but a very modest decrease during January (Figure 7h), when the ambient temperature was the lowest. In the case of 649 wintertime organic aerosol, the lack of a diurnal trend (Figure 7c to show modest diurnal variability. The ratio between organic and inorganic aerosol chemical species (OIR) exhibits diurnal variability from March to October, when also ambient temperature has strong diurnal variation. The OIR achieves 660 its minimum during daytime and maximum during night (Figure 7b). In other words, particles have the highest organic 661 fraction during night-time and early mornings. The organic aerosol mass concentration increases during night (Figure 7c), 662 likely due to more efficient partitioning of semi-volatile species into the aerosol phase. This effect is seen even more 663 clearly in the nitrate concentration (Figure 7e), with a strong diurnal pattern largely tracking the diurnal temperature trends 664 over the year. 665

666
The nature of particulate nitrate can be estimated via fragmentation ratios of NO + and NO2 + ions detected by the ACSM 667 as described by (Farmer et al., 2010) for the AMS. A higher ratio (> 5) generally means a greater presence of organic 668 nitrates and a lower ratio (2-3) indicates inorganic ammonium nitrate. As the ACSM has a low mass resolving power, we 669 here estimate the ratio between m/Q 30 and m/Q 46 Th as a proxy for the NO + : NO2 + -ratio. We note that there is possible 670 interference of organic mass fragments at these m/Q-ratios. Nonetheless, we observe that the wintertime nitrate resembles 671 ammonium nitrate and the summertime nitrate hints towards the presence of organic nitrates (Figure 7g). This is in line 672 with the recent study stating that more than 50% of the nitrates detected in the sub-micron particles at SMEAR II are 673

Wind direction dependence 689
The wind direction plays a key role together with other meteorological conditions determining the aerosol chemical 690 composition at SMEAR II. While the sections above focus more on the role of radiation and temperature on sub-micron 691 aerosol composition, this section explains the role of wind direction and speed. We want to stress that this section does 692 not include any definite geographical source analysis of the NR-PM1 components. A detailed trajectory analysis is a 693 better tool for understanding the actual footprint areas of air pollutants as wind direction analysis might lead to a 694 systematic bias in the pollutant origins due to prevailing weather patterns. The highest organic aerosol loading was observed during summer for all of the wind direction bins (I -IV) with rather 703 modest diurnal variability, perhaps due to the coarse time resolution used (Figure 8, left panels). The greatest organic 704 aerosol concentration was associated with sector II that covers the direction of the Korkeakoski sawmills located 6 -7 705 km to the SW (Figure 8c). Moreover, the February peak in organic aerosol was also most distinguishable from sector II 706 ( Figure 8c). Sulphate aerosol in turn was mostly detected with winds from sector I and II (Figure 8, right panels). Sector 707 I shows a general wintertime enhancement (Figure 8b), whereas sector II shows a clear maximum during February (Figure  708 8d). The westerly sectors (III&IV) were associated with cleaner air (Figures 8e-h). Organic aerosol concentration at SMEAR II increased with S-SE winds as already visualised also in Figure 8c (Figure  731 9a). The monoterpene mixing ratio also peaked, with a more narrow range of wind directions, analogous with the direction 732 of the nearby Korkeakoski sawmills (Figure 9d). With higher wind speeds, monoterpenes were also observed from a 733 wider span of wind directions. Organic aerosol showed wind speed dependence with S-SE winds with lower 734 concentrations associated with wind speed exceeding 25 km h -1 (ca. 6.9 m s -1 ). A possible explanation is that the 735 monoterpene emissions from the sawmills did not have time to oxidise and form SOA with such high wind speeds before 736 reaching SMEAR II. Organic aerosol concentration was relatively constant outside the sawmill interference, though the 737 lowest loadings were detected when air masses arrived with wind speeds exceeding 20 km h -1 (ca. 5.5 m s -1 ) from the 738 NW sector. In contrast, monoterpene mixing ratio was rather constant with varying wind directions and wind speeds, 739 obviously again apart from the sawmills direction (approximately 130°). Similar observations of the wind direction 740 dependence of monoterpene mixing ratios have been reported before, with a subsequent organic aerosol mass 741 concentration increase at SMEAR II with SW winds (Eerdekens et al., 2009;Liao et al., 2011). ). Such wind speed dependence can be observed with long-range transported air pollutants: their transport is generally 776 more efficient with higher wind speeds. The results presented here are also consistent with hygroscopicity measurements 777 conducted at SMEAR II (Petäjä et al., 2005), where the hygroscopic growth factor was greatest when SO2 rich air arrived 778 fast to the station from the NE. 779 780 NE and SE represent the major SO2 sources in February. The NE SO2 was detected with lower wind speed dependence 781 than generally observed (Figures A.8b&e). The lifetime of SO2 is dependent on wet and dry deposition, and oxidation to 782 sulphate (photochemistry or aqueous phase chemistry in cloud droplets). These factors influence the likelihood of 783 detecting SO2 from distant sources. The higher wintertime concentrations are also linked to the atmospheric boundary 784 layer dynamics, as discussed earlier. The SW and SE-S winds with wind speeds exceeding 16 km h -1 (ca. 4.4 m s -1 ) were 785 associated with sulphate during February ( Figure A.8b). Sulphate was detected also with a wide range of wind directions 786 during low wind speeds. In the case of low wind speeds, it is hard to determine the wind direction accurately. However, 787 it was clear that sulphate was not associated with W-NW winds, as shown previously in the paper (Figure 8, right panels). The nitrate concentration field visualised in Figure 9c was highest when wind blew from SE-SW. No wind speed 805 dependence could be attributed to the nitrate from E-SE, whereas for SW, nitrate concentration clearly elevated when 806 wind speed exceeded 20 km h -1 (ca. 5.5 m s -1 ). NOx concentration, in turn, was not significantly elevated with SW winds 807 regardless of the wind speed, but shows similar behaviour to nitrate with SE-S winds (Figure 9f). The nitrates arriving 808 with SW likely spend more time in the atmosphere than in the case of SE-S source. A previous study focusing on organic 809 nitrates at SMEAR II linked their occurrence to SE winds (Kortelainen et al., 2017). They suggest night-time nitrate 810 radical oxidation of sawmill BVOCs as their major source. The same study attributes inorganic ammonium nitrate with 811 SW winds. The study was conducted in spring-time. Also our results suggested an increased organic nitrate presence in 812 spring compared to wintertime (Figure 7g). 813 814 In February, the nitrate concentration field resembles the overall concentration field depicted in Figure 9c, but highest 815 loadings were typically associated with low wind speeds from S-SE ( Figure A.8c). The reason for not observing nitrate 816 with high wind speeds could be the fact that there is not enough time for nitrate aerosol formation. NOx concentration 817 was overall elevated between NE and SE, and the clean SE-N sector had negligible NOx loading ( Figure A.8f). Despite 818 the NOx availability in the North, no nitrate aerosol was observed. This could be due to limited ammonia availability in 819 winter time. Most NOx was detected with E-SE winds when wind speed was 8-16 km h -1 (ca. 2.2-4.4 m s -1 ). 820 821 In July, SW winds blew most of the nitrate to SMEAR II ( Figure A.9c). However, also slightly elevated concentrations 822 can be observed with S-SE winds ( Figure A.9c). The nitrate associated with SW winds again requires high wind speeds. 823 The NOx concentration was significantly lower in July compared to February, as already shown in Figure 4f  The clean NW sector shows bright values for the ammonium balance field. Here, the ammonium balance exceeds one 843 due to the noisiness of the data introduced by both ammonium and nitrate used in the ammonium balance calculation 844 being below their detection limits during NW winds. percentiles, respectively) of which 68% was organics, 20% sulphate, 6% nitrate, and 6% ammonium. Chloride 855 concentrations in the non-refractory sub-micron particles were negligible (< 1%). As many factors, such as ambient temperature, solar radiation, atmospheric boundary layer height and wind influence the aerosol particle concentrations 857 and trace gas emissions, oxidation and volatility, we observed a clear seasonal cycle in NR-PM1 loading and composition. 858 859 During warm months, biogenic VOC emissions increase, and upon oxidation, produce SOA which represents a major 860 source of PM at SMEAR II. Organic aerosol mass concentration achieved its annual maximum in July (3.3, 1.7, and 4.6 861 µg m -3 for median, 25 th and 75 th percentiles, respectively) that further lead to the annual maximum in the total NR-PM1 862 loading (4.2, 2.2, and 5.7 µg m -3 for median, 25 th , 75 th percentiles, respectively). Organics on average made up 80% of 863 the NR-PM1 in summer. During the exceptionally hot Julys of 2014 and 2018, the organic aerosol concentrations were 864 up to 70% higher than the 7-year July mean. Most of the mass could be associated with increased biogenic SOA 865 production. The projected increase of heat wave frequency over Finland (and in general) will most likely influence the 866 loading and chemical composition of aerosol particles, and subsequently affect the Earth's radiative balance. Also from 867 this perspective, continuing the long-term measurements at SMEAR II is essential. 868

869
Winter months indicate low amounts of solar radiation and a shallow boundary layer. NOx and SO2, the main precursors 870 for particulate nitrate and sulphate, respectively, achieved their maximum mixing ratios during the darkest months while 871 emitted into the shallow boundary layer during the period of low photochemical activity. These species are generally 872 emitted in combustion processes that lead to high wintertime concentrations both due to the additional need of residential 873 heating as well as the shallow boundary layer prohibiting their vertical mixing. The maximum wintertime NR-PM1 874 concentration was most commonly detected in February, and explained by an enhancement of inorganic aerosol species. 875 The particulate sulphate and nitrate peaked in February, which was later than their precursors, as a combined result of 876 wind patterns, deposition mechanisms and photochemistry affecting their formation and removal rates. The contribution 877 of inorganic aerosol species was ca. 50% of the total NR-PM1 (2.7, 1.6, 5.1 µg m -3 for median, 25 th , 75 th percentiles, 878 respectively) in February of which 30% was sulphate, 10% nitrate and 10% ammonium. Importantly, much of these 879 inorganic aerosol species were most likely from long-range transport. If emission regulations regarding SO2 and NOx 880 become stricter in the future in Europe, and especially in Russia, the wintertime NR-PM1 might decrease significantly at 881 SMEAR II. 882 883 To our understanding, this is the longest time series reported describing the aerosol chemical composition measured on-884 line in the boreal region. Long-term monitoring of changes introduced by emission regulations together with the changes 885 introduced by the changing climate, are crucial for understanding the aerosol-sensitivity of the (boreal) climate. Thus, we 886 keep the ACSM measurements on going at SMEAR II to obtain an even longer data set. The data presented here will be 887 publicly available, and we welcome collaborative work in utilising this information for broadening the understanding of 888 the boreal environment. 889

Data availability 890
The ACSM data are available at EBAS database (http://ebas.nilu.no/). The trace gas and meteorology data are available at the SMART 891 SMEAR data repository (https://avaa.tdata.fi/web/smart). Other data are available upon request from the corresponding authors.