Increase of secondary organic aerosol over four years in an urban environment

. The evolution of fine aerosol (PM 1 ) species as well as the contribution of potential sources to the total organic aerosol (OA) at an urban background site (Palau Reial, PR, 80 m a.s.l) in the western Mediterranean basin (WMB) was 10 investigated. For this purpose, an aerosol chemical speciation monitor (ACSM) was deployed to acquire real-time measurements for two one-year periods: May 2014 - May 2015 (period A) and Sep 2017 - Oct 2018 (period B). Total PM 1 concentrations showed a slight decrease (from 10.1 to 9.6 µg·m -3 from A to B), although the relative contribution of both inorganic and organic compounds varied significantly. Regarding inorganic compounds, SO 42- , black carbon and NH 4+ showed a significant decrease from period A to B, whilst NO 3- 15 concentration was found higher in B. Source apportionment revealed OA was 46% and 70% of secondary origin (SOA) in periods A and B, respectively. Two oxygenated secondary sources (OOA) were differentiated by their oxidation status (i.e. aging): less-oxidized (LO-OOA) and more-oxidized (MO-OOA). Disregarding winter periods, where LO-OOA production is not favoured, LO-OOA transformation into MO-OOA was found more effective in period B. The highest MO-OOA-to-LO-OOA ratio (1.5) was found in September-October 2018, implying an accumulation effect after the high temperature and solar 20 radiation conditions in the summer season. In addition, SOA was found sensitive to a NO x -polluted ambient and to other pollutants, especially to ozone, which could be enhancing its production specially during afternoon hours. The anthropogenic primary OA sources identified, cooking-like OA (COA), hydrocarbon-like OA (HOA), and biomass burning OA (BBOA), decreased from period A to B in both absolute concentrations and relative contribution (as a whole, 44% and 40%, respectively). However, their concentrations and proportion to OA grow rapidly during highly-polluted episodes. 25 The influence of certain atmospheric episodes on OA sources was also assessed. Both SOA factors seem linked with long and medium-range circulations, especially those coming from inland Europe and the Mediterranean (triggering mainly MO-OOA) and summer breeze-driven regional circulation (triggering mainly LO-OOA). In contrast, POA pollution is enhanced either during air-cleaning episodes or stagnation anticyclonic events. diel concentrations are flatter in MSY 270 because mountain sites are less affected by NO titration, leading to high daily O 3 average concentrations. Therefore, the highest difference between MSY and PR, the more ozone has been reacting at PR. Enhanced reactivity of ozone in period B could be attributed to the reaction with higher VOCs concentrations resulting in the production of SOA. Also, contrasting O 3 and LO-OOA diel cycles, afternoon hours it can be seen how the LO-OOA increase coincides with the O 3 minimum.


30
Atmospheric particulate matter (PM) is regarded by the World Health Organization as one of the most harmful air pollutants to human health (WHO, 2016(WHO, , 2018. Fine particles (PM1, those with aerodynamic diameter <1 µm) have a significant impact on human health (Trippetta et al., 2016;Yang et al., 2019, climate (Shrivastava et al., 2017), and visibility (Shi et al., 2014). Organic Aerosol (OA) is the main constituent of fine aerosol in the atmosphere (Zhang et al., 2007) and it can be classified regarding its origin as primary OA (POA), consisting of directly emitted solid or liquid OA; or secondary OA 35 (SOA), resulting from chemical transformation of pre-existing particles, nucleation or gas-to-particle condensation. https://doi.org/10.5194/acp-2020-1244 Preprint. Discussion started: 16 December 2020 c Author(s) 2020. CC BY 4.0 License.
The ACSM was operated with 24 scans per measurement (alternating sample/filter scan) with a scan speed of 200 ms·amu -1 , resulting in a 30-minute time resolution. Data acquisition software (versions 1.4.4.5, 1.5.2.1 and 1.6.0.0 depending on the period) and analysis software (version 1.6.1.1) implemented in Igor Pro (WaveMetrics, Inc.) were provided by Aerodyne Research Inc. Data were corrected to account for flow rate changes and for response decay by using the N2 signal. The 80 composition-dependent collection efficiency (CE) (Middlebrook et al., 2012) corrections were applied. University of Barcelona. In all data analysis, time will always be UTC based, unless otherwise specified.

Additional measurements and instrumentation
Standardized protocols of quality control (COST Action CA16109 COLOSSAL Chemical On-Line cOmposition and Source Apportionment of fine aerosoL, 2019) were carried out in both periods. The sum of all NR-PM1 species (OA + SO4 2-+ NO3 -+ NH4 + + Cl -) and BC was compared with co-located PM1 measurements. This analysis assumes that the PM1 contribution of the mineral and sea salt tail from the coarse PM or the trace elements are negligible. Also, species concentrations obtained by 100 Q-ACSM were compared with the same components from off-line PM1 determination except for ACSM OA, which was compared with organic carbon (OC).

Source apportionment of OA
Source apportionment of the organic mass fraction was conducted applying the Positive Matrix Factorization (PMF) method (Paatero and Tapper, 1994), using the multilinear engine (ME-2) (Paatero, 1999), to OA mass spectra. It ranged from 12 to 105 120 Th and excluded higher m/z ions which accounted for minor fraction of total signal (<3% on average), presented low S/N ratio and were interfered by the naphthalene signal. The PMF model decomposes the bulk OA matrix by iterating where X is the m·n measured OA matrix, m is the number of rows of X, which consist of the number of time steps, and n is its number of columns, which account for the number of m/z from 12 to 120. G is the matrix that denotes the contributions of a 110 factor p at a time step i and F represents the contribution of ion j to factor p mass spectrum. The matrix E is the residual matrix containing the unexplained information from X. (2) is the optimization variable. In order to avoid to conduct this function to not a global but a local minimum the whole n·m space 115 should be covered, and to this end rotational tools are vital to converge to mathematically coherent solutions.
PMF allows to introduce a-priori information to guide the model, e.g. by using the a-value approach. Both factor profiles and time series can be restricted to resemble to introduced reference ones, named anchor profiles or anchor series. By means of a parameter a, the percentage of freedom of a given factor respect to their constraint can be varied.
Each dataset was separated in four periods: April-May, June-August, September-October and November-March. Modifications 120 from standard seasons were based on the monthly-averaged concentration of BBOA markers (f60, f73), only revealing a potential BBOA source from November to March (Fig. 4), and meteorological variables.
The OA source apportionment was carried out following the methodology described by Crippa et al. (2013) and COLOSSAL guidelines (work in progress), using the SoFi (Source Finder) toolkit version 6.8k developed by Datalystica Ltd. Unconstrained

125
PMF was performed with 3 to 8 factors runs and posteriorly, PMF with constraints was run with 3 to 5 or to 6 factors depending on the season. Differences between solutions of different number of factors for each season are shown in Table S1, Fig. S 2 and chosen seasonal profiles in Fig. S3. The constraints applied through ME2 for HOA and COA sources used the anchor profile of (Crippa et al., 2013), as the solutions anchored with them correlated better with the external tracers than those anchored with Mohr et al., 2012 (Table S2). The profile anchor used for BBOA was from Ng et al., 2010. In all cases, they all 130 were subject to a sensitive analysis with a-values ranging 0-0.5 and steps of 0.1 to choose the best a-value combination for these three factors. Optimization of the number of factors and a-value combinations implied considering (Table S1): i) variation of the ratio between Q/Qexp (Q exp = m · n − p · (m + n)), which should present a steady descent from p to p+1, p+2 factors,

Classification of atmospheric episodes.
Classification of atmospheric episodes was performed with the HYSPLIT model (Stein et al., 2015). Air mass back-trajectories for 120 h at three heights (750, 1500 and 2500 m a.s.l) were computed, with vertical flux modelling, for each day of measurements and interpreted to be classified regarding its predominant transport provenance into Atlantic North (AN), Winter 145 Anticyclonic (WA) (from October to March), Europe (EU), Mediterranean (MED), North African (NAF) and Summer Regional (SREG, from April to September), characteristics of which are discussed in previous works (Pey et al., 2010;Ripoll et al., 2014Ripoll et al., , 2015.

150
Comparison of ACSM output with co-located measurements can be found in Table S3. OPC shows the best agreement with ACSM (NR-PM1+BC vs. OPC PM1), with Squared Pearson correlation coefficients R 2 =0.71 and R 2 =0.76 and slopes of the https://doi.org/10.5194/acp-2020-1244 Preprint. Discussion started: 16 December 2020 c Author(s) 2020. CC BY 4.0 License. orthogonal distance fit 0.989 ± 0.007 and 1.285 ± 0.006 for periods A and B, respectively. The slopes over 1 when comparing ACSM with other instruments can be partially attributed to differences in the particle size range measured. Moreover, overestimation of primary OA sources by ACSM by a factor of 1.2 to 1.5 has also been reported (Reyes-Villegas et al., 2018;155 Xu et al., 2018) as a consequence of source-unspecific OA RIE application, leading in turn to PM1 overestimation by ACSM.
Very good agreement is shown between off-line analysis of SO4 2-, NO3and NH4 + (R 2 >0.85) and also between OA and organic carbon (OC) (R 2 =0.73 in A and R 2 =0.86 in B). The anion Clis not considered due to the very low concentrations and potential determination problems (Tobler et al., 2020). OA-to-OC ratio is estimated from the slope in the scatterplot between these two variables, resulting in a value of 2.69 and 2.96 in periods A and B, respectively, a 68% and 85% higher than the 1.6 value 160 found in Barcelona (Minguillón et al., 2011), calculated as in (Aiken et al., 2008). This OA-to-OC ratios would point, similarly to the previous findings at Montsec  and Montseny , to the aforementioned unspecific-source OA RIE overestimation in ACSM and the artifact evaporation of semi-volatile compounds of OC in filters.

Submicron aerosol composition
Data overview for periods A and B is shown in Table S4. Time series of NR-PM1 species, co-located gases and meteorological Average PM1 concentrations (± standard deviation) resulting from the sum of NR-PM1 components and BC are 10.1 ± 6.7 μg·m -³ during campaign A and 9.6 ± 6.6 μg·m -³ during campaign B (reduction of a 5% from A to B). A decrease of a 5%, 21%, 9% and 18% is shown for of OA, SO4 2-, NH4 + and BC, respectively, although NO3and Clincreased an 8% and a 20% from period A to B.

175
Seasonally averaged mean PM1 species concentrations are displayed in Fig and August, for instance, the ratio of occurrence of SREG/AN is lower in period A than in period B, which is coherent with lower PM1 concentrations recorded in July A with respect to July B.
OA seasonal trend behaves similarly as bulk NR-PM1 due to same causes, but also photochemical enhancement in warm months due to higher sun irradiation, except for summer 2014. SO4 2concentrations are in both cases higher during warmer 185 than colder months, opposite to the NO3concentrations trend (except for September-October 2014) as has been widely reported in previous studies Pey et al., 2009;Ripoll et al., 2015). In period A, higher NH4 + concentrations happen during winter months whilst they are seasonally-stable through period B. BC shows the lowest concentrations in summer months in both periods, although this decrement is more pronounced in period A.   and AMS (Mohr et al., 2012(Mohr et al., , 2015 in which same OA sources were identified. Just for the cold season in period A, a single OOA factor was extracted. Supporting the solutions chosen, mean scaled residuals (Fig. S 7) are sharply centered to zero in both cases although histogram A shows higher spread and skewness, meaning worse match 195 between OA measured and the sum of OA factors concentrations.

OA source apportionment
COA accounts for a 18% and 14% (period A and B, respectively) of OA (Fig. 2). COA profiles (Fig. 3) reflect the expected pattern of signals related to the oxygenation of fatty acids due to cooking activities (m/z 29, 55, 41, 69) (He et al., 2010).
Correlation between COA and HOA is not negligible, as profiles show a Pearson coefficient of R 2 ≥0.71 (Table S2). However, Figure 5 also reveals a much higher dinner peak than the lunch peak along almost all seasons, probably related to a much 205 thinner planet boundary layer (PBL) at 9 PM than at 1 PM.
HOA consists mainly of ions stemming from diesel exhaust from recondensed engine lubricating oil compounds (Canagaratna et al., 2010;Chirico et al., 2010). Average contributions in periods A and B are, respectively, 20% and 12% ( BBOA mass spectra is alike in both cold periods (Fig. 3) characterized by ions at m/z 29, 43, 60 and 73, fragments of anhydrosugars, such as levoglucosan, products of cellulose pyrolysis combustion reactions (Alfarra et al., 2007). BBOA is only abiding by the coldest period, from November to March (Fig. 4) and accounting for 14% and 12% of total OA in A and 220 B, respectively. Diel cycles present the same trend in both periods (Fig. S8), staying flat throughout the day and ascending from 7 PM to 0 AM, pointing to a relation to nocturnal domestic heating or breeze-driven transported pollutants from forest or agricultural fires (Reche et al., 2012) and enhanced by a narrower boundary layer. Correlations with ions m/z 60 and m/z 73 are better in period B (R 2 =0.91 and 0.61, respectively) than in period A (R 2 =0.60 and 0.55, respectively) (Table S2). withdraw from the expected triangle proposed by (Ng et al., 2010) due to a worst definition of SOA factors, besides some might be untrustworthy because they are close to the OA detection limit. Atmosphere is inferred to be more oxidized in period 230 B, especially in summer months as clouds of points present lower f43 and higher f44 than in A. This is reflected in the OOA factors discussed below.
LO-OOA mass spectra differs significantly from period A to B. The m/z 43-to-m/z 44 ratio is 0.37 in period A and 1.05 in period B (Fig. 3), indicating a LO-OOA profile with a higher oxidation degree in period A. Contribution of LO-OOA to 235 apportioned OA increased from period A to B although the period average is not directly comparable, as period A lacks this factor in the cold period, hence considered as zero and leading to lower LO-OOA concentrations (Fig. 2). Considering only the three seasons excluding the Nov-Mar period, LO-OOA shows a slight decrease from period A to B (1.6 µg·m -3 to 1.5 µg·m -3 ). Seasonally, in all cases except for the already discussed summer 2014, the amount of this source decreased from period A to B. In both cases, LO-OOA concentrations rise towards warmer months, which suggests SOA formation pathways might be 240 linked to photochemical oxidation and breeze regimes. This can also be inferred by the diel patterns of LO-OOA (Fig. S 8), which are flat except for a valley of around a factor of -30% from 2 to 8 PM, when temperature and irradiation reach their maximums, but also related to PBL widening and maximum sea breeze speeds at the time. The rise in LO-OOA at night could be also explained by nighttime SOA formation via NO3 radical (result of anthropogenic NOx reactions) oxidation of VOCs, but especially due to the land breeze prevailing during the night and transporting previously formed OA from inland areas to 245 the coast. The valleys around 3PM of seasonal diel cycles do seem to be more profound in April-May and June-August and the peak-to-valley ratio more pronounced in period B (Fig. 6).
MO-OOA mass spectra are similar for periods A and B, even including the OOA profile from the cold subperiod in A.
Excluding the Nov-March subperiod, MO-OOA shows an increase of a 90% (from 1.0 µg·m -3 to 1.9 µg·m -3 in A and B, respectively). The m/z43-to-m/z44 ratios are of 0.28 and 0.21, respectively, and hence, a less oxidized MO-OOA in period A  Concentrations of the secondary factors as a function of temperature and NOx and CO concentrations for winter (DJF) and summer (JJA) are shown in Fig. S11. A tendency of higher concentrations of SOA towards higher temperatures and NOx concentrations in winter and summer can be observed. Temperature has been reported to enhance SOA photochemical 260 pathways from VOCs; these graphs indicate, as pointed out in , that these reactions are also favored over a certain NOx concentration threshold. However, high concentrations of SOA are simultaneously high with high temperature and CO levels, inferring SOA might be enhanced not only by a highly NOx-polluted ambient, but the ensemble of pollutants during severely-contaminated episodes.

265
Ozone has been reported to be an atmospheric oxidant inducing SOA production, whose period-averaged concentrations are substantially higher in period A than in B (Tables S4). Still, Figure S12 points out that ozone is more reactive in the period B, as the area of the difference between PR concentrations and those in the regional background site MSY (Montseny Natural Main contributors to high pollution episodes can be targeted monitoring the proportion of each factor to the total OA concentration as in Fig. 7 (Zhang et al., 2019b). In period A, concentrations of POA increase along with OA concentrations.
However, in period B, with reduced POA concentrations respect to A, SOA represents the main pollutant even during OA growth, and more concretely, LO-OOA increases at expenses of MO-OOA. This points out that whilst POA was more responsible of high pollution episodes in A, this tendency was inverted over period B, as SOA is reasonably constant in B   northeastern directions, i.e. from residential areas nearby the site. Regional production seems to be the main cause of LO-OOA as well as northern and northwestern advections, transporting air masses inland. Eastern and north-eastern winds have been reported to recirculate pollutants from the outlying industrialized areas which are canalized through the Llobregat river basin.
MO-OOA long-range transport is deduced from the plots, since highest concentrations are driven by stronger wind speeds from northeast in period A, and northwest, southwest and east in period B. Note that the main directions for both LO-OOA 300 and MO-OOA in period B coincide, while the origin for LO-OOA and MO-OOA in period A seems to differ. This could be due to additional/missing source foci or changes in their advective pathways.
Regarding air masses, in Fig. 9, the proportion of OA of each source per episode is shown. Proportion of SOA is highest for SREG, coinciding with high temperatures, and coherent with the enhanced oxidation over time during these air mass recirculation episodes. High proportion is also recorded for EU and MED in period B. For the rest of episodes there is not a 305 clear variation in terms of SOA contribution. Regarding absolute concentrations, the pattern is different for periods A and B (Fig. S9b). The highest concentrations are reached during WA episodes in period A, linked to stagnation conditions, and lowest during SREG, due to major occurrence in an aforementioned anomalously cold, wet summer (Servei Meteorològic de Catalunya, 2020a, 2020b). For period B, the highest concentrations correspond to MED and EU episodes, while the lowest were recorded during AN due to associated northern strong winds. Highest proportion of LO-OOA occurs for AW and SREG 310 episodes, the first carrying pollutants which have crossed the whole Iberian Peninsula and the latter recirculating them along the breeze regime. MO-OOA is dominant in the highly polluted MED episodes, western advections and EU episodes, which provide long-range-driven pollutants from continental Europe.
While bulk PM1 decreased slightly from period A to B, with consistent decreases of all its main constituents except for NO3 -, the average OA concentrations remain similar in both periods (4.2, 4.0 µg·m -³). Nevertheless, the relative contribution of the different OA sources varies significantly. The severe reduction of POA found from period A to B (-31%) is mainly driven by a significant decrement of HOA (-40%). The simultaneous reduction in BC and NOx concentrations of 18% and 4% point to an effect of traffic-restriction implemented policies. On the other hand, SOA concentrations increased from A to B both in 320 absolute and relative terms, representing a 45% and 60% of total OA in periods A and B, respectively. The predominance of the SOA over total OA in period B remains also for the highly polluted episodes, where there is an increase of the relative POA contributions, but they still remain below 50% of total OA. Digging into SOA composition, it is more aged in period B, This agrees with the increase of reactivity of ozone in the study area, as determined by the variations of urban and regional locations.
The temperature and solar radiation seem to enhance SOA concentrations, as shown by the pronounced seasonality, growing from cold to warm months (Fig. 5). The ratio LO-OOA-to-MO-OOA decreases strongly from period A to period B (mean pointing to local SOA formation as the summer breeze recirculation might enhance fresh SOA. On the contrary, the lowest LO-OOA-to-MO-OOA ratio is recorded during EU and MED, indicative of aged SOA transport.

Conclusions
Characterization of non-refractory fine aerosol (NR-PM1) in the urban background of Barcelona, including organic aerosol retrieved by an optical counter, although slopes over 1 suggest an overestimation likely caused by the use of the default relative ionization efficiency for OA, which could be lower than the actual one.  Nevertheless, O3 has become more reactive on period B, therefore becoming a probable promoter of OA oxidation.
To the authors' knowledge, this is one of the first times NR-PM1 chemical composition and OA sources have been studied