key: cord-0154339-krokry52 authors: Chen, Gang; Canonaco, Francesco; Tobler, Anna; Aas, Wenche; Alastuey, Andres; Allan, James; Atabakhsh, Samira; Aurela, Minna; Baltensperger, Urs; Bougiatioti, Aikaterini; Brito, Joel F. De; Ceburnis, Darius; Chazeau, Benjamin; Chebaicheb, Hasna; Daellenbach, Kaspar R.; Ehn, Mikael; Haddad, Imad El; Eleftheriadis, Konstantinos; Favez, Olivier; Flentje, Harald; Font, Anna; Fossum, Kirsten; Freney, Evelyn; Gini, Maria; Green, David C; Heikkinen, Liine; Herrmann, Hartmut; Kalogridis, Athina-Cerise; Keernik, Hannes; Lhotka, Radek; Lin, Chunshui; Lunder, Chris; Maasikmets, Marek; Manousakas, Manousos I.; Marchand, Nicolas; Marin, Cristina; Marmureanu, Luminita; Mihalopoulos, Nikolaos; Movcnik, Grivsa; Nkecki, Jaroslaw; O'Dowd, Colin; Ovadnevaite, Jurgita; Peter, Thomas; Petit, Jean-Eudes; Pikridas, Michael; Platt, Stephen Matthew; Pokorn'a, Petra; Poulain, Laurent; Priestman, Max; Riffault, V'eronique; Rinaldi, Matteo; R'o.za'nski, Kazimierz; Schwarz, Jaroslav; Sciare, Jean; Simon, Leila; Skiba, Alicja; Slowik, Jay G.; Sosedova, Yulia; Stavroulas, Iasonas; Styszko, Katarzyna; Teinemaa, Erik; Timonen, Hilkka; Tremper, Anja; Vasilescu, Jeni; Via, Marta; Vodivcka, Petr; Wiedensohler, Alfred; Zografou, Olga; Minguill'on, Mar'ia Cruz; Pr'evot, Andr'e S.H. title: European Aerosol Phenomenology -- 8: Harmonised Source Apportionment of Organic Aerosol using 22 Year-long ACSM/AMS Datasets date: 2022-01-03 journal: nan DOI: nan sha: 57226243a007519a6e418b05eac53c5b67b377fa doc_id: 154339 cord_uid: krokry52 Organic aerosol (OA) is a key component to total submicron particulate matter (PM1), and comprehensive knowledge of OA sources across Europe is crucial to mitigate PM1 levels. Europe has a well-established air quality research infrastructure from which yearlong datasets using 21 aerosol chemical speciation monitors (ACSMs) and 1 aerosol mass spectrometer (AMS) were gathered during 2013-2019. It includes 9 non-urban and 13 urban sites. This study developed a state-of-the-art source apportionment protocol to analyse long-term OA mass spectrum data by applying the most advanced source apportionment strategies (i.e., rolling PMF, ME-2, and bootstrap). This harmonised protocol enables the quantifications of the most common OA components such as hydrocarbon-like OA (HOA), biomass burning OA (BBOA), cooking-like OA (COA), more oxidised-oxygenated OA (MO-OOA), and less oxidised-oxygenated OA (LO-OOA). Other components such as coal combustion OA (CCOA), solid fuel OA (SFOA: mainly mixture of coal and peat combustion), cigarette smoke OA (CSOA), sea salt (mostly inorganic but part of the OA mass spectrum), coffee OA, and ship industry OA could also be separated at a few specific sites. Oxygenated OA (OOA) components make up most of the submicron OA mass (average = 71.1%, a range of 43.7-100%). Solid fuel combustion-related OA components (i.e., BBOA, CCOA, and SFOA) are still considerable with in total 16.0% yearly contribution to the OA, yet mainly during winter months (21.4%). Overall, this comprehensive protocol works effectively across all sites governed by different sources and generates robust and consistent source apportionment results. Our work presents a comprehensive overview of OA sources in Europe with a unique combination of high time resolution and long-term data coverage (9-36 months), providing essential information to improve/validate air quality, health impact, and climate models. The comparison of SA results from different sites is challenging, as they are not always equivalent due to the subjective decisions made by the analysts during the different PMF analysis steps, especially regarding the optimal number of factors, their identification, and validation. For that reason, a standardised protocol (Section 2.3) has been developed to guide the PMF analyses to more streamlined directions to ensure comparability between results obtained from different sites/instruments. According to this protocol, rolling PMF is performed following the latest and most advanced statistical features present within the Source Finder Professional (SoFi Pro) package (Datalystica Ltd., Villigen, Switzerland, (Canonaco et al., , 2013 ) integrated into the Igor Pro software (WaveMetrics Inc., Lake Oswego, OR, USA). Although subjective judgements cannot be avoided entirely, the developed protocol aims to minimise the number of decisions to be made by the user. This highly time-resolved information of OA sources in Europe could substantially improve the development, validation, and prediction of regional/global air quality/climate models by providing This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 4 extra independent information. These results could also be helpful to health-related studies when trying to accurately predict the toxicity of atmospheric aerosol since OA has significantly different health impacts depending on its origin (Daellenbach et al., 2020) . Multiple years of data are finally needed to assess the impact of particulate matter sources on morbidity and mortality due to chronic exposure (Liu et al., 2019; Yang et al., 2019) . Eventually, this work will provide valuable information for policymakers to take the most effective mitigation measures for aerosol-related environmental problems. Overall, this study presents a comprehensive overview of OA sources across Europe by following a thoroughly-designed and harmonised protocol (Section 2.3). Specifically, the seasonal/spatial variability of OA sources regarding time series and source profiles are unfolded in the following sections. This study is the main outcome of the Chemical On-Line cOmpoSition and Source Apportionment of fine aerosoL (COLOSSAL) project (https://www.costcolossal.eu/), based on measurements performed within ACTRIS. In total, 22 year-long datasets were used here from 14 different countries via ACSM/AMS since 2013 (Fig. 1 ). This study includes data from 18 Q-ACSM (quadrupole ACSM, (Ng et al., 2011b) ), 3 ToF-ACSM (Time-of-Flight ACSM (Fröhlich et al., 2013) ), and 1 C-ToF-AMS (compact time-of-flight AMS (Drewnick et al., 2005) ). This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 5 Fig. 1 . ACSM/AMS measurement periods considered in this study. Overall, the ACSM and AMS considered here apply similar techniques. Briefly, the air is passing through a critical orifice into an aerodynamic lens, where atmospheric aerosol is focused and accelerated (the smaller the aerodynamic size, the higher the velocity) into a vacuum chamber (10-5 Torr) and impacts on the surface of a standard vaporiser heated at 600 °C. The resulting vapours are then ionised by electron impact (electric ionisation at 70 eV), and these ions are further extracted into the detector to be characterised using the mass spectrometer. Compared to the AMS, the ACSM is more robust, affordable, and easier to operate, making it suitable for long-term monitoring purposes. However, it cannot measure the size-resolved chemical composition and its time and mass resolutions are poorer compared to the AMS. Most of the ACSMs deployed here participated in an inter-comparison activity conducted by the Aerosol Chemical Monitor Calibration Centre (ACMCC) at SIRTA (https://sirta.ipsl.fr/) and reported consistent results as long as proper calibrations were conducted (Crenn et al., 2015; Freney et al., 2019) . One of the This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 6 objectives of the COLOSSAL project is to deliver a harmonised standard operating procedure (SOP) for ACSM (COLOSSAL, 2021) , and most of the 22 datasets were collected by following this SOP. However, the recommended relative ionisation efficiency (RIE) calibration procedures have varied over the long-time span of these datasets (2013) (2014) (2015) (2016) (2017) (2018) (2019) . For instance, some of the datasets conducted RIE calibration only on specific m/z values (jump scan) as recommended earlier, instead of scanning the entire mass range (10 to 150 amu) of the mass spectrometer (Freney et al., 2019) . Considering the nitrate interference on the CO2 + signal at m/z 44 (so-called Pieber effect) is time-dependent (Freney et al., 2019; Fröhlich et al., 2015; Pieber et al., 2016) and m/z 44 is not measured in jump scan RIE calibrations, it is thus impossible to do a post-correction consistently. Therefore, none of the datasets was corrected for the Pieber effect. However, it is assumed that such artefact would not represent more than 25% (at a higher maximum, e.g., during severe ammonium nitrate pollution episodes) of total OA readings by AMS/ACSM used for the present study. Complementary to ACSM/AMS measurements, equivalent black carbon (eBC) was also monitored at all sites using filter-based absorption photometers. It was typically measured using Multi Angle Absorption Photometer (MAAP) (Thermo) or Aethalometer Model AE33 (Magee Scientific) devices and predominantly with default settings as proposed by the manufacturer (i.e., with no extra data correction procedures). In the case of multi-wavelength instruments (e.g., AE33 or former AE31 devices), eBC concentrations were reported from measurements at 880 nm, and solid/liquid fuel-burning eBC subfractions (noted eBCwb and eBCff, respectively, hereafter) were distinguished from each other based on the application of the so-called Aethalometer model (Sandradewi et al., 2008; Zotter et al., 2017) . This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 7 The 22 sampling sites are classified based on their geographic locations as urban (13 sites, including four flagged as suburban: Athens DEM, Lille, Paris, and Bucharest) or non-urban (9 sites, Table S1 ). The chemical composition of some of these 22 datasets have already been reported Bressi et al., 2021) . This study focuses on the overview of OA source apportionment results and includes new sites in our analysis. More details about source apportionment results at some of these sites can be found in the following published papers: Canonaco et al., 2021; Farah et al., 2021; Heikkinen et al., 2021; Minguillón et al., 2015; Poulain et al., 2020; . PMF has been customarily performed to conduct source apportionment of ambient aerosol data (e.g., ACSM/AMS data) in many previous studies (Lanz et al., 2007; Ulbrich et al., 2009; Zhang et al., 2011) . PMF model was first introduced by (Paatero and Tapper, 1994) as follows: where is the elements of the matrices for the measurements, is the factor time series, is the factor profiles, and is the PMF residuals. The subscripts i, j, and k represent time, m/z, and a discrete factor, respectively. The superscript, p represents the number of factors. PMF finds the model solution by using the least-squares algorithm by iteratively minimising the following quantity: This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 8 where σij is the measurement uncertainty. However, the PMF model does not provide a unique solution, which is usually referred to as rotational ambiguity (Paatero et al., 2002) . Specifically, the model can deliver the exact same quantity of Q for a combination of the matrices G (time series) and F (factor profiles) and for a combination of their rotations and ( ℎ = • and = • ). In this case, Q is the same, despite the solution being possibly entirely different. Even though the non-negativity constraints limit the number of allowed rotations, there are still many possible rotations and thus solutions. In order to reduce the rotational ambiguity of the PMF model, Paatero and Hopke (2009) proposed a multilinear engine (ME-2) algorithm, which allows the addition of a priori information into the model (e.g., source profiles or time series of external data) to prevent the model from unrealistic rotations and to generate more unique solutions. Using a priori information allows the user to guide the model towards an environmentally reasonable solution. Canonaco et al. (2013) implemented an ME-2 solver (Paatero, 1999) into the Igor-based software package, Source Finder (SoFi). SoFi enables the users to have enhanced rotational control over the factor solutions by imposing constraints via, e.g., the a-value approach on one or more elements of F and/or G (Paatero and Hopke, 2009) . For instance, the a value (ranging from 0 to 1) is the tolerated relative deviation of a factor profile ( ) or time series ( ) from the chosen a priori input profile ( ) or time series ( ) during the iterative least-square minimization, as demonstrated in Equations 3a and 3b (Canonaco et al., 2013) : This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 9 = ± • (3b) Conventional PMF is conducted over the whole dataset, with the assumption that the OA source profiles are static, which can lead to high errors when it comes to long-term datasets considering that OA chemical fingerprints are expected to vary over time (Paatero et al., 2014) . For instance, showed a substantial seasonal variability of oxygenated organic aerosol (OOA) factor profiles. Parworth et al. (2015) first proposed to run PMF analysis on a smaller time window (e.g., 14 days) to roll over the whole dataset with a certain step. This technique was further refined by Canonaco et al. (2021) . The rolling PMF window mechanism allows the PMF model to adapt the temporal variations of the source profiles (e.g., biogenic versus biomass burning influences on OOA), which usually provides well-separated OA factors. In addition, with the help of the bootstrap strategy (Efron, 1979) and the random a-value approach, users can estimate the statistical and rotational uncertainties of the PMF results ). This work presents a standardised protocol to identify main OA components. These guidelines work well for all 22 datasets despite the various pollution sources and OA levels at the different stations. All 22 datasets were analysed by following the protocol described in this study to minimise users' subjectivity. Figure 2 provides the general working flow of this protocol, while the more detailed step-by-step guideline is summarised in Table 1 . Detailed explanations of each step are unfolded in the following subsections. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 10 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 11 Table 1 . Step-by-step protocol for running rolling PMF. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 12 To effectively implement rolling PMF analysis, knowing potential sources for each season for any given site is crucial. Seasonal PMF pre-tests allow us to retrieve reasonable seasonal PMF results (so-called base case). The first step of the pre-tests is to conduct unconstrained PMF with varying factors from 2 to 8 for each season. Based on the time series, diel patterns, factor profiles, and correlation with external data, the number of factors for each season could potentially be pre- and COA). In this section, the approach used to identify each source type effectively is covered in the following paragraphs. For a site potentially impacted by traffic, it is suggested to constrain the HOA mass spectrum from Crippa et al. (2013) using a tight a value (0.05-0.1) with a narrower range of factors (3-7). Typically, the HOA factor has a pronounced diel pattern with distinct morning and evening rush hour peaks, and the HOA factor profile is typically similar among different sites (Crippa et al., 2014) . However, when the diel pattern does not show the typical variability expected from traffic emissions, one should consider using the HOA mass spectrum from unconstrained PMF runs or loosen the a value for HOA from Crippa et al. (2013) . When a BBOA-like factor exists in both unconstrained and HOA-constrained runs, the "local" BBOA spectrum retrieved from these runs are recommended to be used as a constraint/reference profile in the next step of the PMF analysis. This is because of the relatively large spatial variabilities of BBOA factor profiles (Crippa et al., 2014) . The expected BBOA factor usually has a pronounced contribution of m/z 29, 60, and 73 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 13 signals and a distinct diel pattern with high concentration during nighttime. If the BBOA factor was not present in previous steps, it should be checked if f60 (i.e., the fraction of m/z 60 to the total organic mass) is above the background level of 0.3% (Cubison et al., 2011) , also having a clear temporal pattern beyond the noise. However, it should not be the only criterion to determine the existence of BBOA (the background level of f60 is instrument-dependent). Constrained PMF runs with a reference BBOA spectrum Ng et al., 2011a) and a relatively high a value (0.3-0.5) need to be performed to seek more proof of its existence by (i) comparing the solution without a BBOA factor; (ii) by checking the correlation factor between HOA vs eBCff, If different slopes in f55 vs f57 plots at different hours of the day point towards the presence of COA (Mohr et al., 2012) , it should be constrained tightly using the corresponding spectrum from Crippa et al. (2013) with an a value of 0.05-0.1. Then, it should be checked if the diel pattern (mass concentration or mass fraction) is reasonable, i.e., it peaks during the time of expected cookingrelated activities (noon/afternoon and evening peaks). The COA factor is typically only present in urban environments close to residential and commercial areas. For an environment with potential coal combustion sources, looking for a "local" CCOA factor mass spectrum from unconstrained or HOA-constrained PMF runs is recommended. However, it is typically challenging to identify the CCOA factor with ACSM data when its contribution is not significant because of the relatively low m/z resolution and since the mass range (up to 100/120 for Q-ACSM) does not include polyaromatic hydrocarbons (PAHs). In addition, the similar spectrum pattern of hydrocarbon ions (e.g., CnH2n-1 + and CnH2n+1 + ) between HOA, BBOA and This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 14 CCOA makes it challenging for PMF to resolve these two factors (Sun et al., 2016) . Therefore, the existence of CCOA should be justified at least by the most significant contribution of m/z 115 (mainly C9H7 + ) and the absence of the morning rush hour peak. Sometimes, PMF can also help picking up a site-specific factor with special fingerprints in previous steps (besides some common POAs and OOAs). In that case, the key ions should be checked to investigate potential sources, then the fragmentation table to understand how these key ions are calculated. Most importantly, OOA components should not be constrained as they are Scaled residuals should be monitored throughout the PMF analysis. The scaled residuals' daily cycle structure might indicate missing or badly separated sources. In addition, spikes and structural patterns in the time series of the scaled residuals always require extra attention, as they suggest some high uncertainties of the current model for these points/periods or instrument issues. Also, if the constrained profiles cause systematic patterns in the residuals, they should be reconsidered or tested with a larger a value. Last but not least, the a-value sensitivity analysis for constrained factors should be conducted to optimise (i.e., towards enhanced correlation with externals, reasonable factor profiles, and small scaled residual, etc.) the constrained factor to the dataset of This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 15 interest. Then, once the reasonable PMF solution has been determined (so-called base case result), the mass spectra for all constrained factors are input factors for bootstrap analysis in the next step. A bootstrap resampling strategy (Efron, 1979) is recommended to test the stability of the base case solutions. Therefore, all the mass spectra of constrained factors (i.e., POAs and site-specific factors) should be constrained using the random a-value technique with an upper a value of 0.4-0.5 and repeats of 100-1000. Next, the same technique mentioned in Section 2.3.4 should be used to filter out "incorrect" solutions (not environmentally reasonable). As a next step, the quality of the averaged solution of selected PMF runs should be checked in terms of the uncertainties of factor profiles, time series, and percentage of selected runs. If the bootstrap solution shows significant uncertainties, the base case needs to be re-evaluated. The mass spectra for all constrained factors resulting from the bootstrap phase should be saved and used for the rolling PMF analysis in the next step. The mass spectra from bootstrapped solutions are recommended as the reference profiles to constrain POA and site-specific factors during rolling PMF. BBOA is known as the most spatiotemporally variable factor compared to all other POA factors (Crippa et al., 2014) . Also, considering that the highest mass concentrations of BBOA occur in winter, the BBOA mass spectrum from the bootstrapped winter solution is recommended as the constraint, as it is more representative of the dataset. Alternatively, the published profiles (HOA and COA from and BBOA from Ng et al. (2011a) or Crippa et al. (2013) ) could be used . Canonaco et al. (2021) suggested using a random a-value technique (randomly select a This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 16 values for each constraint within a certain range) and bootstrap resampling strategy to estimate the rotational uncertainties of rolling PMF. Based on the seasonal bootstrap analysis, the upper a value for the site-specific factors can be determined by the seasonal variation and uncertainties. Canonaco et al. (2021) showed that an upper value of 0.4 for POAs is sufficient to cover the temporal variabilities. However, an upper a value of 0.5 for BBOA is suggested when high temporal variabilities of the BBOA mass spectrum are expected. When the number of factors is not identical for all the seasons, the rolling PMF should be conducted with both n and n+1 factors over the entire dataset. With the large number of PMF runs expected for rolling analysis (e.g., >15,000 runs for a one- year dataset with 30-min time-resolution), inspecting each single PMF run is not feasible. Therefore, a criterion-based selection should be used to (i) evaluate the quality of the PMF runs quantitatively and relatively objectively and (ii) sort out the unconstrained factors in the same order for further averaging. The criteria-based selection has been explicitly explained in Canonaco et al. (2021) . In short, SoFi Pro enables the user to define criteria based on the time series and/or factor profiles to select environmentally reasonable solutions. In addition, this criterion-based selection function can also serve to reposition unconstrained factors as unconstrained factors can appear in random order in a different iteration of the PMF. The inexact sorting criteria can result in a mixing of the unconstrained factors. Therefore, it is crucial to use the most representative sorting criterion, i.e., f44 for the MO-OOA (criterion #7 in Table S2 ), as suggested by . At the same time, it is also recommended to monitor f44 in MO-OOA and f43 in LO-OOA to reject zerovalues of these two criteria. The criterion of f44/f43 for MO-OOA is not recommended because it This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 17 could accept PMF solutions with smaller f44 in MO-OOA than LO-OOA when f43 is extremely small. Statistical tests such as the t-test for time series based criteria (#4, #5 and #7 in Table S2 ) and correlation-based criteria (#1, #2 and #3 in Table S2 ) should be performed to minimise subjective decisions. With a p value ≤ 0.05, it is possible to select PMF runs with statistically significantly higher scores compared to the same criterion for other factors (for time series based criteria) or statistically higher correlation with each other (for correlation-based criteria). This technique allows the user to select environmentally reasonable PMF solutions with minimum subjective judgements. Last but not least, the optimum length of the time window should be determined by minimising non-modelled data points and Q/Qexp while applying the same criteria and thresholds to these PMF runs with different time windows (7, 14 or 28 days) . Based on our study, a 14-day window size is the most commonly selected one, which is consistent with previous studies ). For all the 22 ACSM/AMS datasets, this standardised protocol works well in general. However, different numbers of factors at different periods remain challenging for rolling PMF. There are three special cases that this protocol could not cover. Specifically, the BBOA factor is not present in the warm period for the Barcelona, Cyprus, and Marseille datasets. However, this protocol could not cover such situations with a proper criterion to objectively include/exclude certain OA factors (e.g., BBOA) in certain time periods. The distribution of the correlation between BBOA and eBCwb was utilised for the Marseille data, which appears to have a bimodal Fisher distribution. Thus, the 10 th percentile results from the separated distributions were used as thresholds to define the existence of BBOA . For the Barcelona and Cyprus dataset, the criterion to This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 18 decide the existence of BBOA is the explained variation of f60 by BBOA. A t-test was conducted with the null hypothesis that the variation of f60 explained by BBOA is not significantly larger than that of other factors. The presence of BBOA is only considered when the p value was ≤ 0.05. In addition, as discussed before, a different number of factors often suffer from more uncertainties at the edge of the transition period by averaging different numbers of factor solutions simultaneously. One strategy to avoid averaging over different numbers of factor solutions is to unselect any data point in a range of edge ± window length/2. However, it could potentially lead to relatively more non-modelled points during the transition period. Therefore, keeping a static number of factors in the rolling analysis as much as possible is recommended if that is environmentally feasible. Thus, it remains challenging to objectively define the transition point to an improved source apportionment for rolling PMF analysis with a different number of OA factors. stations across Europe. Overall, the total PM1 (sum of OA, eBC, nitrate (NO3), sulfate (SO4), ammonium (NH4), and chloride (Chl)) mass concentration has an average of 9.7 ± 7.9 µg/m 3 , with generally higher concentrations at urban stations (brown circles, avg = 12.2 ± 9.3 µg/m 3 ) compared to non-urban ones (green circles, avg = 6.2 ± 3.3 µg/m 3 ). Kraków is the most polluted site (40.4 µg/m 3 ), and Birkenes is the cleanest (1.3 µg/m 3 ). The OA contribution to the total submicron aerosol ranges from 21 to 75%, which is consistent with previous results based on shorter campaigns . For other main chemical species, eBC, NO3, SO4, NH4, and Chl exhibit an average contribution to the total PM1 of 10.0%, 15.0%, 16.2%, 9.9% and 1.2%, This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 19 respectively. eBC and NO3 show higher contributions at the urban sites (12.1% and 16.2%) than at the non-urban ones (6.8% and 13.3%). Primary and secondary OA factors are identified for each station in Fig. 3 . Overall, the well-known POA factors have been resolved, including HOA and BBOA. All datasets identify HOA except Hyytiälä (non-urban). BBOA is resolved in 19 datasets (12 urban and 7 non-urban sites) except for Hyytiälä, Puy de Dôme, and Helsinki. In general, both HOA and BBOA present considerable components of the total submicron aerosol with average contributions of 5.0% and 5.6%, respectively. Also, they both show higher contributions at the urban sites (HOA: 5.7% and BBOA: 6.3%) than at the non-urban ones (HOA: 4.0% and BBOA 4.5%). COA is identified in three southern European cities (i.e., Athens, Marseille, and Barcelona), a megacity (London), and a central European city (Zürich). It has an average contribution to the total PM1 of 6.3%. CCOA is resolved in Kraków and Melpitz (non-urban) with contributions of 5.8% and 6.9%, respectively. SFOA, which likely originates from peat and coal combustion, appears at the two Irish sites (Dublin (urban) and Carnsore Point (non-urban), with contributions of 12.2% and 6.0%, respectively. In addition, local factors (particular of the monitoring site) are highlighted in this study: an m/z 58-related OA (58-OA) in Magadino (1.2%); a coffee roaster OA factor in Helsinki (3.0%); a sea salt factor at Carnsore Point (1.5%); cigarette smoking OA (CSOA) in Zürich (Qi et al., 2019; Stefenelli et al., 2019) ; and a mixed ship-industry factor in Marseille (2.2%, . This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 20 drastic differences among urban (16.7% and 10.6%) and non-urban sites (29.6% and 17.2%), which is expected since more primary sources are present in the urban environment. When summing up MO-OOA and LO-OOA (Total OOA), Fig. 3 suggests that secondary OA is the main contributor to total submicron PM (average = 34.5%, range from 11.7 to 62.4%) and dominates OA (average = 71.1%, range from 47.3 to 100%) across Europe. In addition, the resolved OA factors have been validated using available external data (Table S3) . Regarding the PMF errors (Equation (6) Table S4 . In general, POAs often have smaller PMF errors than OOA factors since they are always constrained. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 21 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 22 To understand how the OA composition changes under different loadings, each dataset is divided into ten bins containing the same number of points based on the OA mass concentrations. As shown in Fig. 4 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 23 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 24 The highly time-resolved long-term ACSM/AMS data allow investigating the diel cycles of the OA components. Figure 5 shows the averages (solid lines) and standard deviations ( HOA shows a distinct pattern at urban sites with characteristic morning and evening rush-hour peaks. By contrast, at non-urban locations, HOA does not follow the same pattern, indicating that this factor is likely associated with transported traffic emissions or with non-traffic primary hydrocarbon emissions at these sites. COA, as mentioned resolved at six urban sites, shows distinct noon and evening peaks with minor standard deviations, which suggests small spatial variabilities of cooking emissions. BBOA has a similar diel cycle at urban and non-urban sites with reduced values during the day and a marked evening peak, which indicates that most likely, (residential) heating emissions are the main contributor to BBOA. However, at the urban sites, the evening peak of BBOA is more pronounced than at the non-urban sites, which suggests that the urban BBOA is more local and synchronised to domestic heating than the non-urban sites. The delayed morning peaks and broader span of non-urban BBOA diel profiles further indicate relatively distant BBOA sources at non-urban sites. The MO-OOA diel trend at both urban and non-urban sites shows the most stable pattern. A slight decrease is observed during the night starting after 23:00 and continuing until 06:00-07:00, probably due to the decrease in the formation rates of MO-OOA in the absence of photochemical This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 25 activity. By contrast, the MO-OOA concentrations increase slightly from morning to afternoon at the non-urban sites, potentially due to photochemistry. Moreover, the long-range transported origin could also play a role in this stable diel trend of MO-OOA. The diel cycles of LO-OOA at the urban and non-urban sites reveal nighttime maxima with a slight decrease at noon, suggesting local production or enhanced vapour partitioning onto pre-existing aerosol in the shallow nocturnal boundary layer. Urban LO-OOA shows much stronger evening peaks than non-urban sites, potentially caused by nighttime chemistry yielding urban OOA from POA oxidation. Kodros et al. (2020) and Tiitta et al. (2016) demonstrated how dark ageing of BBOA could potentially yield substantial amounts of OOA. Such ageing is likely to explain some of our observations at the urban sites (Zhang et al., 2020) . Both urban and non-urban sites show small spatial variability for the Total OOA, but the diel cycle for urban sites has a more substantial evening peak again due to the strong LO-OOA increase. In general, the POA factors show more temporal variability than the OOA factors at both urban and non-urban sites. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 26 . The POA factors generally show more pronounced variability than the OOA factors. In addition, the urban sites have stronger patterns with less spatial variability than the non-urban sites. In addition, Fig. S4 shows the normalised weekly cycles for the non-urban and urban sites separately (with Birkenes and Hyytiälä datasets included among the non-urban sites). Compared with the diel cycles, the weekly ones are much weaker in general. In addition, POA shows a stronger variability compared to the OOA factors, similar to the averaged diel cycles (Fig. 5) . HOA shows decreased values during the weekend than the weekdays at urban (-2.6%) and nonurban (-3.5%) sites. Except for HOA, the non-urban weekly cycles are much less pronounced than the ones at the urban sites. Specifically, BBOA increases during the weekend with 19.9% and 12.3% higher than the weekdays for urban and non-urban sites, respectively. This is because more This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 27 wood-burning activities (e.g., open fire grills and residential heating) are expected during weekends (Fuller et al., 2014) . COA shows a similar trend as it increases during the weekends (+18.8% at urban sites), suggesting that cooking activities are more pronounced. All OOA factors (MO-OOA, LO-OOA, and Total OOA) do not present strong weekly cycles (<6.3% difference between the weekday and weekend), with relatively larger spatial variability for OOA at the urban sites, indicated by larger standard deviations. The time series of daily-averaged OA fractions for each site is shown in Fig. S1 , which presents a big picture of the entire source apportionment result. It indicates a significant spatial and temporal variabilities of OA contributions across Europe. To study the seasonal variation of OA and its sources, data was divided into four seasons: winter (DJF: December, January, and February), spring (MAM: March, April, and May), summer (JJA: June, July, and August), and autumn (SON: September, October, and November). Figure 6 indicates a relatively small spatial variability of the relative OA contributions at both urban and non-urban datasets, and there is no clear pattern between the OA fraction and PM1 loading. Urban sites have higher POA contributions in OA than non-urban sites. However, each dataset shows an apparent seasonal variability with higher POA contributions and mass concentrations in cold seasons than in warm ones ( Fig. 6 and Fig. S5 ). The contributions of POA also appear to be higher when the total PM1 mass concentration increases at all non-urban sites, as shown in Fig. 6 . Specifically, urban sites show higher HOA contributions (overall average contribution of 12.7 ± 2.9%) than non-urban sites (7.4 ± 2.7%), which is expected due to more traffic emissions in urban areas. Moreover, for both urban and non-urban sites, the HOA contribution shows a distinct This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 28 seasonality with the lowest contribution in summer (8.8 ± 3.4%) and the highest in winter (12.0 ± 4.8%), and similar contributions in spring (10.9 ± 4.3%) and autumn (11.5 ± 4.6%). This might be due to the lower boundary layer with stagnant conditions in the cold season favouring the accumulation of primary and local pollutants and/or reduced photochemistry. In addition, the heating related sources (i.e., BBOA, CCOA, and SFOA) are obviously more pronounced during the cold seasons than the warm ones. Specifically, BBOA has an average contribution of 8.3 ± 4.7% in summer but 16.9 ± 8.4% in winter. CCOA was only found in Kraków and Melpitz, and its contributions varied from season to season with substantially enhanced contribution during winter (Kraków: 18.2% and Melpitz: 23.1%) compared to summer (Kraków: 4.5% and Melpitz: 8.7%). The drastic seasonal variations in Kraków are due to the widespread use of coal-burning for residential heating purposes in winter. In Melpitz, the coal combustion is less dependent on local sources, but most likely, emissions from Poland or other eastern European countries are rapidly transported by advection in winter, leading to the observed seasonality. For the Dublin dataset, the SFOA (heavily affected by both peat and coal combustion sources (Lin et al., , 2018 ) shows an enhanced contribution in winter (32.9%) and a decreased contribution in summer (13.2%) . The SFOA in Carnsore Point station shows less seasonality because of the absence of local sources but still has a higher contribution in winter (13.3%) than in summer (9.2%). The COA contribution shows almost no seasonality ranging from 14.3 ± 2.7% in summer to 15.4 ± 3.3% in autumn at the six urban sites where this factor is resolved, suggesting that the cooking emission contribution is constant and non-negligible in European cities in all seasons. The specific contributions of all OA components for all datasets and all seasons are summarised in Table S5, in which all POA except HOA, BBOA and COA are summed up as "other OA". This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 29 Fig. 6 . Relative contributions of all OA components at each station grouped by season (see text). The stations are categorised into non-urban (left) and urban sites (right), where each subset is ordered by PM1 mass concentration. Canonaco et al. (2015) analysed the seasonal variability of major ion intensities (i.e., m/z 44 and m/z 43) in the OOA factors in Zürich. They suggested that biomass burning emissions, the most important precursor of SOA in that city during winter, cause the LO-OOA factors to be at the left half of Sally's triangle (Ng et al., 2010) within the f44 vs f43 space, as shown for biomass burning This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 30 emissions by Heringa et al. (2011) . On the contrary, biogenic SOA from terpene oxidation may "push" the LO-OOA factors to the right side of the triangle in summer (Pfaffenberger et al., 2013) . This study explores the seasonality of f44 vs f43 for MO-OOA, LO-OOA and the Total OOA factors for both seasons at all sites, as shown in Fig. 7 . It shows that the rolling PMF provides a good separation between MO-OOA and LO-OOA in both winter and summer (Fig. 7a) , which is consistent with what Canonaco et al. (2021) and reported for two Swiss datasets. However, the positions of MO-OOA for the different stations in the f44 vs f43 space show large spatial variability (Fig. 7a) , mainly attributed to the complex and various ageing processes in different locations under different meteorological conditions. In general, all LO-OOAs shift to the right side, though by a different extent during the summer (JJA) compared to the winter (DJF) months. This is more apparent when summing up the LO-OOA and MO-OOA into the Total OOA factors (Fig. 7b) . The shift of LO-OOA and the Total OOA factors is most likely due to the enhanced biogenic emissions with higher temperatures during summer seasons . This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 31 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 32 BBOA contribution throughout the year (18.2% in summer). Considering biomass burning oxidises rapidly, the biomass burning influence is more pronounced in the MO-OOA factor during the whole year. That is why the LO-OOA factor in winter remains on the right side of the triangle. Hyytiälä is located in the boreal forest with significant biogenic SOA formation in summer Yli-Juuti et al., 2021) , which explains the high f43 in summer. In addition, no POA factors were deconvolved from this dataset following the presented protocol. By utilizing machine learning techniques, Heikkinen et al. (2021) resolved a slightly aged POA factor that could neither be further separated into HOA nor BBOA. This factor appeared only in winter and coincided with a LO-OOA drop to near-zero loadings when utilizing k-mean clustering approach . Therefore, it is very likely that the Hyytiälä LO-OOA shown in this study is influenced by aged POA in winter, which keeps the LO-OOA f43 high. Also, the potential terpene emission from Korkeakoski sawmills (ca. 7 km NE of the monitoring station) could also keep the LO-OOA on the right side (Äijälä et al., 2017) . The combined effects of high summertime biogenic SOA contribution to LO-OOA and wintertime POA mixing to LO-OOA could explain why LO-OOA in Hyytiälä always stays on the right side of the triangle. When considering the rest of the sites, both the LO-OOA and Total OOA factors generally shift to the right side of the triangle during summer compared to winter. This study also investigates key ions' spatial and monthly variabilities in common OA factors (i.e., m/z 55 and m/z 57 for HOA and COA; m/z 60 and m/z 73 for BBOA; m/z 44 and m/z 43 for MO-OOA, LO-OOA, and the Total OOA). All monthly intensities of these key ions in these OA factors were averaged across the 22 datasets to see possible monthly trends (Fig. 8) . Overall, the key ions This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 33 for HOA, COA, MO-OOA and Total OOA factors barely show a monthly trend. In contrast, f60 and f73 in BBOA are significantly higher in the cold months compared to the warmer months. The main reason is likely that levoglucosan and thus m/z 60 is not stable in the warm season as well as the change in the biomass burning source during different seasons (e.g., residential heating and outdoor open fire) (Bertrand et al., 2018; Bougiatioti et al., 2014; Xie et al., 2014) . The most dominating ions (m/z 44 and m/z 43) in the LO-OOA factor show a relatively strong monthly trend compared with MO-OOA and Total OOA. Specifically, f44 is smaller, and f43 is higher in LO-OOA during the warm months, which is a further indication that the enhancement of biogenicallyformed SOA could increase the intensity of f43 in LO-OOA when the temperature is increased (which eventually "pushes'' the LO-OOA factors to the right side of the triangle as presented in Fig. 7 , ). Monthly trends of these key ions of these OA factors for each station are shown in Fig S6. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 34 Fig. 8 . Monthly average relative intensities of key ions in the corresponding factor profiles over all the datasets. In order to compare the spatial variations of key ions in the different OA factors, the interquartile range (IQR) of the monthly site-averaged intensities has been normalised by their median ( Table 2 ). In general, f55 for HOA and COA shows a relatively small IQR/median ratio with averages of This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 35 0.15 ± 0.03 and 0.16 ± 0.07, respectively. The f57 for HOA and COA shows similar consistency across 22 sites with averages of 0.20 ± 0.03 and 0.09 ± 0.05, respectively. Overall, the HOA and COA factors are generally consistent across different locations. It agrees well with the previous findings reported by Crippa et al. (2014) . However, the most essential fingerprint ions of BBOA, f60 and f73, appear to have the largest spatial variability among the POA factors with IQR/median ratio in a range of 0.28-0.39 and 0.20-0.36, respectively. This is expected since the type of wood and burning conditions, as well as chemical ageing, can affect the BBOA mass spectrum significantly (Bertrand et al., 2018; Bougiatioti et al., 2014; Grieshop et al., 2009; Heringa et al., 2011; Weimer et al., 2008; Xie et al., 2014) . Thus, as discussed in a previous section, retrieving site-specific BBOA factor profiles using unconstrained PMF analysis instead of published BBOA factor profiles is strongly recommended. Table 2 . Normalised spatial variations for key ions using the interquartile range (IQR) divided by the medians of average monthly intensities across the 22 datasets. The most dominating ion in MO-OOA, f44, has a somewhat smaller IQR/median ratio, ranging from 0.16 to 0.34, and the f43 in MO-OOA has an IQR/median ratio in a range of 0.45-0.83. This is because different datasets could have significantly variable ageing processes, precursors, and This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 36 meteorological conditions, which appear to affect the degrees of oxygenation (the relative intensities of m/z 44 and m/z 43). In addition, MO-OOA remains dominated mainly by f44 (average f44 = 0.25), but with a much smaller f43 intensity (average f43 = 0.04). Therefore, the slight changes in the intensity of f43 could have a larger effect on the IQR/median ratio for this ion. Due to enhanced biogenic emissions, the f43 in LO-OOA shows consistently larger intensities in the warmer months (May-September), as shown in Fig. 8 . Consequently, the f44 in LO-OOA decreases during the warm seasons and is related to the seasonal differences in f44 vs f43 of Fig.7 . In addition to the dynamic monthly trends observed in LO-OOA, there are also strong spatial variabilities for these two key ions (i.e., m/z 44 and m/z 43). The IQR/median ratios of f44 and f43 in LO-OOA are in the range 0.38-0.94 and 0.39-0.60, respectively. This is expected considering LO-OOA has never been constrained, and various ageing processes, precursors, and meteorological conditions could contribute to the large spatial variabilities. When we sum up the LO-OOA and MO-OOA to the Total OOA factor, m/z 44 is still the dominating ion with an average of 0.21, but it is smaller than that of MO-OOA alone due to the much smaller f44 in LO-OOA with an average of 0.14. Overall, f43 in the Total OOA factor still shows an increasing trend during warm seasons like the LO-OOA factor, which indicates that the effect of enhanced biogenic emissions on the intensity of f43 in the Total OOA might be rather considerable. Moreover, f43 in the Total OOA shows relatively smaller spatial variabilities (compared to MO-OOA and LO-OOA) with an IQR/median ratio of 0.26-0.41. The f44 in the Total OOA has slightly larger spatial variabilities than MO-OOA with an IQR/median ratio of 0.21-0.40, but it is still more stable than LO-OOA. It is expected considering the large spatial variabilities in LO-OOA. However, the sum of the OOA factors shows little monthly trends except for the increasing f43 (thus, decreasing in f44) in warmer months (May-September). This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 37 A state-of-the-art standardised protocol for source apportionment of long-term ACSM/AMS organic aerosol mass spectrum datasets was developed. Our protocol was validated systematically and strictly applied to 22 sites with year-long measurements. It demonstrates the consistency of this protocol with comprehensive source apportionment results, even though each dataset was analysed by each research group individually. Our source apportionment strategy has been significantly improved compared to conventional seasonal PMF by utilising rolling windows, bootstrap, and ME-2 techniques, which were first introduced by Canonaco et al. (2021) . As addressed by and , this strategy allows us to retrieve robust source apportionment results by considering temporal variations of source profiles. Importantly, the success of the rolling mechanism is an essential step to make real-time source apportionment possible (Chen et al., 2022) . However, the current protocol/strategy remains challenging to objectively define the transition point to an improved source apportionment for rolling PMF analysis when a different number of OA factors is necessary for different periods. More tools should be tested to address this challenge in the near future. Overall, this work provides a comprehensive overview of the OA sources in Europe with highly This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 38 smoke OA factor in Zürich. The OOA factors together constitute the main contributor (47.3-100%) to OA and generally show more stable diel and weekly cycles than POA factors. The contributions of POA increase with increasing total OA mass concentration in most of the polluted regions. It suggests that the control of primary emissions could help mitigate OA mass concentration or at least decrease the likelihood of highly polluted episodes. Also, most POA factors show enhanced contribution/mass concentrations during cold seasons compared to warm seasons due to residential heating. Lower boundary layer heights (lower temperature) combined with stagnant conditions can readily cause the accumulation of pollutants. In particular, HOA (traffic emissions) is a nonnegligible OA source with a rather consistent contribution across different stations (10.7 ± 3.8%). Six urban sites display a significant COA factor (i.e., two Athens datasets, Zürich, London, Barcelona, and Marseille) with an overall average contribution to OA of 14.5 ± 2.5%. Moreover, most of the datasets show a resolved BBOA component (except Hyytiälä, Puy de Dôme, and Helsinki) with important contributions to OA (annual average:12.4 ± 6.9%), which increases significantly in winter with a contribution of 16.9 ± 8.4%. Melpitz and Kraków present a CCOA factor (annual average: 14.7 ± 0.8%, winter average: 20.6 ± 3.5%), while SFOA is found to be heavily affected by peat and coal combustion sources at Carnsore Point and Dublin (annual average: 19.3 ± 10.4%, winter average: 22.7 ± 14.4%). All of these results confirm that the reduction of solid fuel-burning for residential heating is one of the key leverages to mitigate fine PM levels in Europe, especially in winter. This study also provides a comprehensive overview of the spatial and temporal variabilities of commonly resolved OA factors (i.e., HOA, COA, BBOA, MO-OOA, LO-OOA, and Total OOA). Overall, the MO-OOA, LO-OOA, and Total OOA factors vary significantly spatially and seasonally. This is expected since the ageing processes, abundances/types of precursors and This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 39 meteorological conditions can differ both temporally and spatially. Regarding the seasonality of the LO-OOA factor, most of the datasets agree well with the findings reported by Canonaco et al. 2015) , with increasing f43 (thus, decreasing f44) intensity in warm seasons likely due to the enhanced biogenic emissions. Moreover, with the help of the rolling PMF technique, time-dependent OA factor profiles have been retrieved. Therefore, this study also investigates the monthly trends (averaged over 22 datasets) and corresponding variabilities (IQR/median ratio) across sites for key ions in the commonly resolved OA factors. While these key ions barely show monthly trends in HOA, COA, MO-OOA, and Total OOA, BBOA key ion (f60 and f73) intensities increase during the cold seasons due to the abundance of biomass burning sources and the lower reactivity of levoglucosan. The increased f43 of LO-OOA during warm seasons is most likely due to enhanced biogenic SOA formation. In terms of spatial variabilities, key ions for HOA and COA factors show a small IQR/median ratio with a range of 0.06-0.27 due to both factor contributions being consistent (if present) as reported in previous studies. However, the key ions for the BBOA factor show a relatively larger spatial variability with the IQR/median ratio ranging from 0.20 to 0.39, which suggests potentially different combustion conditions, type of woods, etc., contribute to this variability in the real-world scenarios. In addition, the f43 intensities in MO-OOA and LO-OOA show large spatial variability, with the IQR/median ratio ranging from 0.45 to 0.83 and 0.39 to 0.60, respectively. The f44 in LO-OOA has a large IQR/median ratio range of 0.38-0.94, but the f44 in MO-OOA is rather less varied across sites with a relatively small IQR/median ratio range of 0.16-0.34. This is expected since OOA factors are never constrained combined with the complexities of ageing processes in different locations and meteorological conditions. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 40 With the help of this state-of-the-art source apportionment protocol, this study has retrieved highly time-resolved, long-term (>9 months), and robust OA source information consistently, with minimum subjective judgements. This highly time-resolved comprehensive OA source information can be useful inputs/constraints to improve/validate climate, health, and air quality models. Finally, this work suggests that policymakers should address the major combustion sources (i.e., biomass burning, coal combustion, and peat) that affect air quality in Europe. Besides, more attention should be paid to the traffic source, even though it is quite constant across places because it is significant in organic aerosols and is also a proxy for non-exhaust emissions. Due to regional transport, decreasing the emissions of primary factors would also decrease the secondary factors observed at the non-urban sites. This work would not have been possible without the following contributions: All co-authors were supported by the cost action of Chemical On-Line cOmpoSition and Source Apportionment of fine aerosol (COLOSSAL, CA16109). GC and AT were supported by a COST related project of the Swiss National Science Foundation, Source apportionment using long-term Aerosol Mass This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 41 the French National Research Agency (ANR) through the PIA (Programme d'Investissement d'Avenir) under contract ANR-11-LABX-0005-01, and the CLIMIBIO project, both financed by the Regional Council "Hauts-de-France" and the European Regional Development Fund (ERDF). This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 44 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 3 Fig. S1 . Time series of the daily-averaged OA contribution for each factor. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 4 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 5 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 10 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 12 Table S4 . PMF errors (std/mean conc.) of major OA factors (estimated by logarithmic probability density functions (pdf) of the standard deviations of each time point i divided by the mean concentration of each time point i for corresponding OA factors, Equation (6) in Canonaco et al. (2021) This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 13 Table S5 . Contribution (in %) of major OA components for each season. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 14 Resolving anthropogenic aerosol pollution types -deconvolution and exploratory classification of pollution events The importance of sesquiterpene oxidation products for secondary organic aerosol formation in a springtime hemiboreal forest Evolution of the chemical fingerprint of biomass burning organic aerosol during aging Processing of biomass-burning aerosol in the eastern Mediterranean during summertime SoFi, an IGOR-based interface for the efficient use of the generalized multilinear engine (ME-2) for the source apportionment: ME-2 application to aerosol mass spectrometer data Seasonal differences in oxygenated organic aerosol composition: implications for emissions sources and factor analysis This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International A new method for long-term source apportionment with time-dependent factor profiles and uncertainty assessment using SoFi Pro: application to 1 year of organic aerosol data Organic aerosol source apportionment by using rolling positive matrix factorization: application to a Mediterranean coastal city Measurement report: Fourteen months of real-time characterisation of the submicronic aerosol and its atmospheric dynamics at the Marseille-Longchamp supersite Real-time Source Apportionment of Organic Aerosols in Three European Cities Time-dependent source apportionment of submicron organic aerosol for a rural site in an alpine valley using a rolling positive matrix factorisation (PMF) window, Atmospheric Chemistry and Physics Guidelines for comparison of ACSM measurements with co-located external data Organic aerosol components derived from 25 AMS data sets across Europe using a consistent ME-2 based source apportionment approach This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International Wintertime aerosol chemical composition and source apportionment of the organic fraction in the metropolitan area of Paris Effects of aging on organic aerosol from open biomass burning smoke in aircraft and laboratory studies A New Time-of-Flight Aerosol Mass Spectrometer (TOF-AMS)-Instrument Description and First Field Deployment Bootstrap Methods: Another Look at the Jackknife Urban increments of gaseous and aerosol pollutants and their sources using mobile aerosol mass spectrometry measurements Air quality in Europe 2021 Altitude Aerosol Measurements in Central France: Seasonality, Sources and Free-Troposphere/Boundary Layer Segregation The second ACTRIS inter-comparison (2016) for Aerosol Chemical Speciation Monitors (ACSM): Calibration protocols and instrument performance evaluations Sarda-This manuscript is a preprint and has not been peer-reviewed yet ACTRIS ACSM intercomparison -Part 2: Intercomparison of ME-2 organic source apportionment results from 15 individual, co-located aerosol mass spectrometers The ToF-ACSM: a portable aerosol chemical speciation monitor with TOFMS detection Contribution of wood burning to PM10 in London Laboratory investigation of photochemical oxidation of organic aerosol from wood fires 2: analysis of aerosol mass spectrometer data Eight years of sub-micrometre organic aerosol composition data from the boreal forest characterized using a machine-learning approach Investigations of primary and secondary particulate matter of different wood combustion appliances with a high-resolution time-offlight aerosol mass spectrometer Aerosol Technology: Properties, Behavior, and Measurement of Airborne Particles AR6 Climate Change 2021: The Physical Science Basis Development of an Aerosol Mass Spectrometer for Size and Composition Analysis of Submicron Particles This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International Rapid dark aging of biomass burning as an overlooked source of oxidized organic aerosol Source apportionment of submicron organic aerosols at an urban site by factor analytical modelling of aerosol mass spectra Characterization of aerosol chemical composition with aerosol mass spectrometry in Central Europe: An overview Wintertime aerosol dominated by solid-fuel-burning emissions across Ireland: insight into the spatial and chemical variation in submicron aerosol Extreme air pollution from residential solid fuel burning Ambient Particulate Air Pollution and Daily Mortality in 652 Cities Chemical characterization of submicron regional background aerosols in the western Mediterranean using an Aerosol Chemical Speciation Monitor Identification and quantification of organic aerosol from cooking and other sources in Barcelona using aerosol mass spectrometer data Real-Time Methods for Estimating Organic Component Mass Concentrations from Aerosol Mass Spectrometer Data Organic aerosol components observed in Northern Hemispheric This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 49 datasets from Aerosol Mass Spectrometry An Aerosol Chemical Speciation Monitor (ACSM) for Routine Monitoring of the Composition and Mass Concentrations of Ambient Aerosol The Multilinear Engine-A Table-Driven, Least Squares Program for Solving Multilinear Problems, Including the n -Way Parallel Factor Analysis Model Methods for estimating uncertainty in factor analytic solutions Rotational tools for factor analytic models Understanding and controlling rotations in factor analytic models Positive matrix factorization: A non-negative factor model with optimal utilization of error estimates of data values Long-term measurements of submicrometer aerosol chemistry at the Southern Great Plains (SGP) using an Aerosol Chemical Speciation Monitor (ACSM) The link between organic aerosol mass loading and degree of oxygenation: an α-pinene photooxidation study Inorganic Salt Interference on CO 2 + in Aerodyne AMS and ACSM Organic Aerosol Composition Studies Multi-year ACSM measurements at the central European research station Melpitz (Germany) -Part 1: Instrument robustness, quality assurance, and impact of upper size cutoff diameter This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International Organic aerosol source apportionment in Zurich using an extractive electrospray ionization time-of-flight mass spectrometer (EESI-TOF-MS) -Part 2: Biomass burning influences in winter Using Aerosol Light Absorption Measurements for the Quantitative Determination of Wood Burning and Traffic Emission Contributions to Particulate Matter Sources and processes that control the submicron organic aerosol composition in an urban Mediterranean environment (Athens): A high temporal-resolution chemical composition measurement study Organic aerosol source apportionment in Zurich using an extractive electrospray ionization time-of-flight mass spectrometer (EESI-TOF-MS) -Part 1: Biogenic influences and day-night chemistry in summer Primary and secondary aerosols in Beijing in winter: Sources, variations and processes Transformation of logwood combustion emissions in a smog chamber: formation of secondary organic aerosol and changes in the primary organic aerosol upon daytime and nighttime aging Characterization of non-refractory (NR) PM1 and source apportionment of organic aerosol in Kraków Interpretation of organic components from Positive Matrix Factorization of aerosol mass spectrometric data Increase in secondary organic aerosol in an urban environment Organic aerosol mass spectral signatures from wood-burning emissions: Influence of burning conditions and wood type World Health Organization, 2021. WHO global air quality guidelines. Particulate matter (PM2.5 and PM10), ozone, nitrogen dioxide, sulfur dioxide and carbon monoxide This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International Gas/Particle Partitioning of 2-Methyltetrols and Levoglucosan at an Urban Site in Denver Laboratory characterization of an aerosol chemical speciation monitor with PM 2.5 measurement capability Source Apportionments of Aerosols and Their Direct Radiative Forcing and Long-Term Trends Over Continental United States Short-term and long-term exposures to fine particulate matter constituents and health: A systematic review and meta-analysis Significance of the organic aerosol driven climate feedback in the boreal area Trends, composition, and sources of carbonaceous aerosol at the Birkenes Observatory Ubiquity and dominance of oxygenated species in organic aerosols in anthropogenically-influenced Northern Hemisphere midlatitudes Understanding atmospheric organic aerosols via factor analysis of aerosol mass spectrometry: a review Substantial brown carbon emissions from wintertime residential wood burning over France Six-year source apportionment of submicron organic aerosols from near-continuous highly time-resolved measurements at SIRTA (Paris area, France) Prévôt 1*** 1 Laboratory of Atmospheric Chemistry 11 School of Physics, Ryan Institute's Centre for Climate and Air Pollution Studies The importance of sesquiterpene oxidation products for secondary organic aerosol formation in a springtime hemiboreal forest A new method for long-term source apportionment with time-dependent factor profiles and uncertainty assessment using SoFi Pro: application to 1 year of organic aerosol data Organic aerosol source apportionment by using rolling positive matrix factorization: application to a Mediterranean coastal city Measurement report: Fourteen months of real-time characterisation of the submicronic aerosol and its atmospheric dynamics at the Marseille-Longchamp supersite Time-dependent source apportionment of submicron organic aerosol for a rural site in an alpine valley using a rolling positive matrix factorisation (PMF) window, Atmospheric Chemistry and Physics Long-term sub-micrometer aerosol chemical composition in the boreal forest: inter-and intra-annual variability Wintertime aerosol dominated by solid-fuel-burning emissions across Ireland: insight into the spatial and chemical variation in submicron aerosol Evaluation of Composition-Dependent Collection Efficiencies for the Aerodyne Aerosol Mass Spectrometer using Field Data The impact of biomass burning and aqueous-phase processing on air quality: A multi-year source apportionment study in the Po Valley 17167-This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International Sources and processes that control the submicron organic aerosol composition in an urban Mediterranean environment (Athens): A high temporalresolution chemical composition measurement study Characterization of non-refractory (NR) PM1 and source apportionment of organic aerosol in Kraków Increase in secondary organic aerosol in an urban environment Trends, composition, and sources of carbonaceous aerosol at the Birkenes Observatory, northern Europe Six-year source apportionment of submicron organic aerosols from near-continuous highly time-resolved measurements at SIRTA This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 52 U., Szidat, S., Prévôt, A.S.H.H., Mocnik, G., Hüglin, C., Baltensperger, U., Szidat, S., Prévôt, A.S.H.H., Močnik, G., Hüglin, C., Baltensperger, U., Szidat, S., Prévôt, A.S.H.H., 2017. Evaluation of the absorption Ångström exponents for traffic and wood burning in the Aethalometer-based source apportionment using radiocarbon measurements of ambient aerosol. Atmos. Chem. Phys. 17, 4229-4249. https://doi.org/10.5194/acp-17-4229-2017 This manuscript is a preprint and has not been peer-reviewed yet. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 7 This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. This manuscript is a preprint and has not been peer-reviewed yet. The manuscript is under review for publication in Environment International. 9 Table S1 . Description of Each Dataset.