A proton-transfer-reaction time-of-flight mass spectrometer (PTR-ToF-MS 4000X2, Ionicon Analytik GmbH), equipped with a particle inlet (Chemical Analysis of aeRosol ONline, CHARON), was utilized to measure VOC in the gas phase and semi-volatile compounds in the particle phase. The PTR-ToF-MS 4000X2 employed here has an ion funnel, which reduces ion loss and enhances sensitivity (Pugliese et al., 2020). The CHARON inlet integrates a gas-phase denuder, an aerodynamic lens with an inertial sampler, and a thermo-desorption unit, all coupled to a PTR-TOF-MS. The gas-phase denuder removes gas-phase analytes, while the aerodynamic lens focuses the aerosol particles. The inertial sampler then concentrates the particle-enriched sample flow. Finally, the particles enter a thermal desorption unit where they are volatilized before being detected by PTR-MS. The inlet system demonstrated a particle enrichment factor of 18±2 (Fig. S3), as discussed in detail in our previous study (Song et al., 2024). The vaporizer (TDU) was operated at 150 °C and 7–8 mbar absolute pressure. The CHARON inlet was described in detail elsewhere (Eichler et al., 2015; Müller et al., 2017; Gkatzelis et al., 2018a; Gkatzelis et al., 2018b).

In this campaign, the CHARON-PTR-MS automatically switched between gas and particle phase measurements. The sequence was: 5 min of high-efficiency particulate air (HEPA) filter measurement for particle background, 10 min of particle-phase measurement, 3 min transition, 10 min of gas-phase VOC measurement, and another 2 min transition. The PTR drift tube was kept at 2.7 mbar, 470 V, and 100 °C, with the ion funnel at 45 V. These settings correspond to an electric field (E) to gas molecule number density (N) ratio of ∼100 Td for VOC measurement. For particle measurement, the PTR was automatically adjusted to 60 Td.

During gas-phase measurement, ambient air was sampled continuously from a 2 m PFA tube (4 mm inner diameter) at 8 L min⁻¹, with 40–100 mL min⁻¹ directed to the PTR-ToF-MS through a polyetheretherketone (PEEK) tubing at 80°. The instrumental background was determined weekly at noon by introducing pure dry nitrogen (N₂, 99.9999 %) into the inlet for 5–15 min to exclude ambient air and quantify the instrumental baseline (Fig. S4). For particle-phase measurement, ambient particles were sampled with a PM_2.5 inlet (Comde Dendra) through a vertical electropolished 3.45 m stainless-steel tube (12 mm inner diameter) at 16.7 L min⁻¹, with 550 mL min⁻¹ directed to the CHARON inlet. Particle background was automatically determined using the HEPA filter.

Gas calibrations were performed at the beginning, middle, and end of each campaign periods using a gas cylinder (Ionicon Analytik GmbH) with 15 VOC, including toluene, trimethylbenzene, xylene, alpha-pinene, acetone, acetonitrile, benzene, and isoprene (accuracy 5 % at ∼1000 ppb). The sensitivity ranged from ∼1000 to 4000 cps ppb⁻¹ for different compounds across different seasonal calibration periods (Fig. S5). Raw data from CHARON-PTR-MS were processed using IONICON Data Analyzer software (IDA 2.2.0) following Müller et al. (2013), Lannuque et al. (2023), and Peron et al. (2024). The CHARON inlet enrichment factor was determined via external calibration with size-selected (DMA, TSI) ammonium nitrate particles (100–500 nm) measured by a condensation particle counter (CPC, TSI).

A high-resolution time-of-flight Aerosol Mass Spectrometer (HR-ToF-AMS) (Aerodyne Research Inc.), equipped with a PM_2.5 aerodynamic lens, was used to measure non-refractory PM_2.5 (NR-PM_2.5) components, including organic aerosol, nitrate, sulfate, ammonium, and chloride, with a time resolution of 1 min (DeCarlo et al., 2006; Canagaratna et al., 2007; Williams et al., 2013). The operation and calibration procedures of the AMS are detailed in our previous publications (Huang et al., 2019; Song et al., 2022). The instrument's capability to provide high-resolution mass spectra enables detailed chemical composition analysis of aerosol particles in real time. Briefly, ambient air was sampled through a PM_2.5 inlet (flow rate of 1 m³ h⁻¹) that was shared with the CHARON system via a 3.45 m stainless-steel tube. A subset of this flow was then directed to the HR-ToF-AMS at a flow rate of 84 cm³ min⁻¹. The aerosol particles were focused into a narrow beam by a PM_2.5 aerodynamic lens, which effectively transmits particles with vacuum aerodynamic diameters (dva) ranging from ∼70 to ∼2500 nm. The particles were subsequently heated by a vaporizer at 600 °C, causing the non-refractory components to volatilize. The volatilized particles were then ionized by electron impact at 70 eV, which is a standard method for ionizing organic and inorganic compounds, ensuring fragmentation patterns that are well-characterized and suitable for identification. The AMS was calibrated using dried ammonium nitrate aerosol particles with sizes of 100–500 nm to determine ionization efficiency and instrument response. Calibration particles were size-selected using a DMA, and signal stability was verified prior to and during each measurement period. Data from the AMS were processed and analyzed using the SQUIRREL 1.65G and PIKA 1.25G software packages. To address the issue of particle bounce losses, a chemical-composition-based collection efficiency (CE) of approximately 0.5 was applied to calculate the particle mass concentrations (Docherty et al., 2011; Middlebrook et al., 2012). For the elemental analysis of organic aerosol (OA), including the hydrogen-to-carbon ratio (H : C) and the oxygen-to-carbon ratio (O : C), we utilized the improved ambient method (Aiken et al., 2008; Canagaratna et al., 2015).

The mass concentrations of PM_2.5 and PM₁₀ and particle size distributions (0.2–18 µm) were determined using an optical particle counter (Fidas200, Palas). Particle number concentrations (>2.5 nm) were monitored using a water-based condensation particle counter (Model 3789, TSI Inc.) and a butanol-based condensation particle counter CPC3776 (TSI Inc., USA). Particle number size distributions ranging from 10 to 410 nm were measured using a nanoparticle sizer (Nanoscan 3910; TSI Inc.) at time resolution of 1 min in winter, and distributions ranging from 13.6 to 763.5 nm were measured using a scanning mobility particle sizer (SMPS; DMA 3081; TSI Inc.) at a time resolution of 7 min in Summer. Equivalent black carbon (eBC) levels were monitored with an aethalometer (AE33, Magee Scientific) at a 1 min resolution. Ammonia (NH₃) concentrations were measured via cavity ring-down spectroscopy (G2103, Picarro Inc.). The gas was sampled via a Teflon tube (4 mm i.d.; 3.1 s residence time). Ozone (O₃) and nitrogen dioxide (NO₂) concentrations were tracked using the O₃41M and AS32M gas analyzers (both from Environment S.A.), respectively. Meteorological parameters such as temperature, relative humidity, wind speed and direction, global radiation, and precipitation were recorded with a compact weather sensor (WS700, Lufft) installed on the container roof. Additional meteorological data was measured by the Institute of Meteorology of the LMU on top of the institute at about 30 m a.g.l. (https://www.meteo.physik.uni-muenchen.de/request-beta/, last access: 23 April 2026). The vertical aerosol distribution was assessed using a scanning aerosol Lidar (Rametrics Inc., Type: LR111-ESS-D200, named KASCAL). Fiber laser-induced fluorescence (FILIF) detects formaldehyde (HCHO) by exciting molecules at 353 nm and measuring fluorescence above 370 nm. The technique alternates between on-peak and off-peak laser frequencies to determine HCHO concentration with high specificity. Calibrated against FTIR standards, FILIF achieves ±27 pptv precision (10 %–15 % accuracy) at 10 Hz sampling frequency, unaffected by humidity. All the information could be found in previous studies (Ye et al., 2021). Unless otherwise specified, all measured values are reported as mean ± standard deviation. Please note that this work is linked to the recently established low-cost sensor network in downtown Munich, monitoring especially O₃, NO₂ and PM_2.5 (Chen, 2025). This included frequent comparison measurements between reference instruments on a bicycle platform, the instruments in our container, a LfU regulatory monitoring station, and the low-cost sensor network in downtown Munich. Back trajectories were calculated using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Stein et al., 2015) at 500 m altitude with 72 h backward duration to identify potential source regions and transport pathways influencing air quality at the measurement site.

The PMF receptor model is a bilinear algorithm that separates air pollutant time series into sources characterized by factor profiles, time series, and residuals (Paatero and Tapper, 1994; Paatero, 1999). It is widely used to identify particle and VOC sources (Kim et al., 2004; Reyes-Villegas et al., 2016; Huang et al., 2018; Gkatzelis et al., 2021).

PMF has also been applied to inorganic aerosol components, elemental composition of particulate matter (PM), VOC, PAHs, black carbon, and size-resolved particle data, showing its versatility as a receptor model across various pollutant types (Äijälä et al., 2019). PMF has been used for analyzing high-time-resolution elemental data in PM_2.5 and PM₁₀ to identify urban and industrial sources and to separate fine and coarse particle sources (Reizer et al., 2021). Reviews and guidelines from European air quality agencies explicitly mention the use of PMF applied to particle mass, ions, metals, and gaseous species (like VOC and PAHs) in addition to OA for comprehensive source apportionment studies (Belis et al., 2013). In summary, PMF receptor modeling is broadly used and suitable for source apportionment of multi-pollutant data sets.

To investigate the sources of VOC, SVOA, and OA, positive matrix factorization (PMF) was applied separately to (i) VOC mass spectra measured by PTR-MS, (ii) semi-volatile aerosol species measured by CHARON-PTR-MS, and (iii) OA measured by HR-TOF-AMS. PMF analyses were conducted independently for each dataset and measurement campaign rather than using a combined dataset. Separate analyses were chosen because the campaigns were performed in different seasons under substantially different atmospheric conditions and instrument states. PMF assumes temporally stationary factor profiles and comparable uncertainty structures within a dataset; these assumptions are not fulfilled when combining measurements across seasons with distinct source contributions and atmospheric processing regimes. In addition, VOC and SVOA datasets were analyzed separately because they represent chemically and physically distinct phases with different atmospheric processing timescales, variability patterns, and measurement uncertainty structures. Combining these datasets in a single PMF analysis would violate the assumption of consistent covariance structure and could lead to mixed or non-interpretable factors. Diagnostic evaluations supporting solution stability and factor selection for each dataset are provided in the Supplement (Figs. S8–S20). For VOC-PMF, inputs were derived from PTR-TOF-MS data following the methodology of Song et al. (2024), involving comprehensive preprocessing including mass spectral deconvolution, background subtraction, and error matrix calculation where uncertainties were determined as [(0.1 × concentration)² + (0.5 × LOD)²]^1∕2 (Kajos et al., 2015). SVOA-PMF inputs were obtained through two complementary measurement modes: HEPA-filtered air samples (representing background signals) and direct atmospheric sampling (Dir) capturing real-time conditions, with ions (primarily C_xHy+, C_xH_yOz+, and C_xH_yO_zNn+). Both datasets underwent rigorous quality control procedures. Low-molecular-weight species (m/z≤60) were excluded to eliminate potential interferences from fragments and common atmospheric gases that may not be representative of specific emission sources (Zhang et al., 2011). Additionally, compounds with ≥20 % missing data points were removed to ensure statistical robustness in the PMF analysis, following established protocols for PMF data preparation (Ulbrich et al., 2009; Zhang et al., 2011). The analyses were initially processed using SoFi Pro 9.0 (Datalystica Ltd.) for exploratory factor analysis and data visualization. The source apportionment of organic aerosols (m/z12–120) was then conducted through unconstrained AMS-PMF analysis using the PMF Evaluation Tool (v3.08C) within IGOR Pro (v8.04), which maintains legacy algorithms specifically optimized for AMS datasets, particularly for handling nonlinear m/z signal-uncertainty relationships in typical AMS operating conditions.

The final number of factors for each dataset was selected based on a combination of diagnostic criteria, including the evolution of Q/Qexp values, inspection of scaled residual distributions, physical interpretability of factor profiles, and temporal behavior. Physical interpretability was evaluated using dominant marker compounds in each factor mass spectrum together with their temporal patterns, allowing source attribution based on established source signatures (e.g., traffic-related factors identified by toluene/xylene dominance and rush-hour diurnal peaks). Solutions with increasing factor numbers were examined to identify the point beyond which additional factors primarily resulted in factor splitting without meaningful improvement in residual structure. Detailed diagnostics, including Q/Qexp evolution, residual analysis, Fpeak sensitivity tests, and correlations with characteristic compound markers and diurnal patterns, are provided in the Supplement (Sect. S3.2, Figs. S7–S19, Table S4–S9).

The summer period was characterized by frequent sunny weather with moderate to high temperatures and significant precipitation events mainly at the beginning and end of the summer campaign. The late winter/spring period shows typical winter conditions until 8 March, according to temperature, precipitation followed by more spring-like conditions thereafter.

Wind measurements at 3 and 30 m above ground (Fig. 1a and b) revealed consistent vertical gradients across all seasons, with average speeds of 0.7±0.4 and 2.6±1.4 m s⁻¹, respectively. Meteorological conditions varied significantly between seasons (Fig. 1c and d), creating distinct chemical environments. The higher summer temperatures (21.9±5.8 °C) promoted enhanced biogenic emissions and photochemical activity, while lower winter temperatures (8.4±3.0 °C) favored primary pollutant accumulation. Relative humidity and total precipitation showed minimal seasonal variation. NO₂ concentrations were 2.5–2.8 times higher in winter/spring (26.1±9.9/28.4±10.8 µg m⁻³) than summer (10.2±7.7 µg m⁻³) due to increased heating emissions and reduced photolysis under shallow boundary layers (339±298 m/510±477 m vs. 516±453 m, variability). The BLH values are derived from ERA5 reanalysis and represent regional-scale atmospheric mixing. They do not resolve street-canyon ventilation. In a street canyon, buildings can limit horizontal and vertical dispersion, especially under stable winter conditions. Therefore, pollutant accumulation at the measurement site may be stronger than suggested by ERA5 BLH alone. This limitation is particularly relevant for reactive pollutants such as NO₂. In addition to dispersion, NO₂ is strongly influenced by local traffic emissions and rapid photochemical processes (e.g., photolysis and O₃ titration). As a result, short-term variations in NO₂ may reflect a combination of street-canyon trapping, advection, and chemical transformation. This can differ from more inert traffic-related species such as BC, which are primarily governed by physical mixing. The ERA5 BLH should therefore be interpreted as an indicator of large-scale mixing conditions rather than street-level ventilation. The LfU station showed similar winter (27.2±9.2 µg m⁻³) and early spring (27.8±9.2 µg m⁻³) mean concentrations to Theresienstrasse, but exhibited weak temporal correlation, reflecting the strong spatial variability of NO₂. Notably, summer NO₂ at LfU (23.4±10.1 µg m⁻³) was 2.3 times higher than Theresienstrasse, yet temporal correlation improved substantially (R=0.5) in Fig. S6. This seasonal pattern suggests that summer's higher boundary layer enhanced vertical mixing and local photochemical reactions, increasing spatial differences in mean NO2 levels. However, stronger synchronized daily photochemical cycles improved temporal agreement between the two sites. The diurnal evolution of NO₂, O₃, and BLH reveals seasonally distinct coupling between photochemical processing and boundary-layer mixing (Fig. S7). During summer, increasing BLH coincides with decreasing NO₂ concentrations and enhanced daytime O₃ formation, indicating efficient vertical mixing and active photochemical processing. Spring exhibits intermediate behavior, whereas winter shows weaker BLH development and persistently elevated NO₂ levels consistent with reduced photolysis and limited dispersion. Gaseous NH₃ showed elevated concentrations across all seasons, with winter (5.4±1.5 µg m⁻³) and spring (4.0±1.8 µg m⁻³) levels comparable to summer values (4.8±2.6 µg m⁻³). O₃ concentrations at Theresienstrasse exhibited strong seasonal variation, with highest levels in summer (67.1±26.6 µg m⁻³), followed by early spring (52.2±19.3 µg m⁻³), and lowest in late winter (32.4±20.2 µg m⁻³). The LfU station showed similar seasonal patterns (summer: 59.3±25.5 µg m⁻³; late winter: 30.8±17.3 µg m⁻³; early spring: 45.9±17.2 µg m⁻³) with strong temporal correlations (R=0.8 in August 2023; R=0.7 in March 2024) in Fig. S6, indicating consistent O₃ behavior across both sites driven by regional photochemical processes.

Particle size distributions revealed contrasting seasonal patterns driven by different formation mechanisms (Fig. 1e and f). Summer showed the highest total particle number concentrations (10000±5100 cm⁻³) with a substantial variability dominated by ultrafine particles (<100 nm: 80.6±8.2 %), indicating a strong contribution of ultrafine particles during this period. Winter exhibited lower total number concentrations (8900±3500 cm⁻³) and a lower fraction of ultrafine particles (33.6±10.1 %), consistent with a greater relative contribution of aged and accumulation-mode particles under stable conditions. Spring showed intermediate behavior with high ultrafine fractions (89.7 %) but lower absolute number concentrations. The seasonal PM mass concentrations showed opposite trends to the particle number concentrations: winter PM_2.5 peaked at 13.0±7.4 µg m⁻³ with nitrate as the dominant component due to low temperatures, shallow boundary layers and increased residential heating. In contrast, summer (6.7±3.7 µg m⁻³) and spring (4.2±3.1 µg m⁻³) showed lower PM_2.5 levels dominated by organic aerosols, despite higher particle numbers, highlighting that ultrafine particles can dominate number concentrations in urban environments while contributing comparatively little to total particulate mass.

Aerosol particle mass composition varies significantly by season, reflecting different formation mechanisms. Organic aerosol particles dominate in summer due to high photochemical activity (Fig. 2), averaging 4.3±2.9 µg m⁻³, higher than 3.3±1.7 µg m⁻³ in winter and 1.8±1.8 µg m⁻³ in spring. Nitrate is the main component in winter (4.5±3.2 µg m⁻³) due to low temperatures favoring particle phase and increased heating emissions providing abundant NH₃ and NO_x precursors for ammonium nitrate formation. Nitrate concentrations drop significantly in spring (1.0±1.3 µg m⁻³) and summer (0.3±0.2 µg m⁻³) as rising temperatures shift the equilibrium toward the gas phase. Ammonium follows similar patterns, peaking in winter (1.8±1.1 µg m⁻³) and remaining low in warmer seasons (spring: 0.4±0.4 µg m⁻³; summer: 0.3±0.2 µg m⁻³), consistent with reduced ammonium nitrate formation.

The average diurnal variations of key volatile organic compounds (VOC) are examined below, focusing on major biogenic and anthropogenic species, including BTEX – a group of aromatic hydrocarbons comprising benzene, toluene, ethylbenzene, and xylene. The VOC selected here comprise 32.3±13.1 % of all VOC detected in summer, 41.0±12.9 % in winter and 37.9±12.5 % in spring.

Figure 3 illustrates the diurnal variations of six VOC across three seasons, revealing distinct patterns tied to their sources and environmental influences. Isoprene and monoterpenes, both biogenic VOC, exhibit the highest mixing ratios in summer. However, Isoprene displays a bimodal pattern in summer and spring, with concentration peaks during the morning and evening hours, suggesting a potential link to traffic activity during the morning and evening hours. This is consistent with direct measurements of isoprene in vehicle exhaust by Borbon et al. (2001). Additionally, fragmentation of higher-carbon aldehydes and cycloalkanes from anthropogenic sources may contribute to the signal of isoprene, further complicating source attribution in urban environments (Coggon et al., 2024b). Monoterpene exhibited early morning peaks in summer and spring, but was almost flat in winter. In contrast, their concentrations are minimal in winter due to suppressed biological activity. Anthropogenic VOC like benzene, toluene, xylene, and trimethylbenzene show different seasonal behaviors. Benzene peaks in winter, likely due to lower BLH and reduced dispersion, with dips at midday possibly from increased OH radical oxidation under sunlight. Toluene, xylene, and trimethylbenzene display notable spikes in spring, particularly during morning and evening hours, suggesting strong contributions from traffic emissions and industrial activities.

We included 117 VOC ions (Table S2) for PMF analysis in summer and 97 VOC ions (Table S3) in winter/spring. Ions with extremely high signal intensities that could disproportionately influence the PMF solution, as well as very low signal ions with minimal contribution, were excluded. The excluded high-abundance ions are common hydrocarbon and oxygenated VOC fragments, such as m/z 33.034 (C₁H₅O⁺), 41.039 (C₃H5+), 43.054 (C₃H7+), 45.034 (C₂H₅O⁺), 57.070 (C₄H9+), and 59.049 (C₃H₇O⁺). These ions were removed because their high concentrations may lead to over-representation of specific factors, whereas low-signal ions were excluded to reduce noise interference (Song et al., 2024). The average VOC molecules mixing ratio is 3.4±1.0 ppb in summer, 2.1±0.7 ppb in winter, and 3.4±1.7 ppb in spring. This study thoroughly examined factor profiles, diurnal patterns, and correlations with tracers (Figs. 4, S8 and Table S4) for summer and for winter (Figs. S9, S10 and Table S5). Five factors were identified based on measured VOC as the optimal interpretable solutions for summer and winter, respectively.

In summer, the first factor was characterized by high contributions and strong correlations with aromatic hydrocarbon ions, such as C₇H9+ (m/z 93.07), C₈H11+ (m/z 107.086), and C₉H13+ (m/z 121.102). These compounds correspond to toluene, xylenes, and C₉-aromatics, commonly used as vehicular emission markers (Squires et al., 2020; Jain et al., 2022). We assigned it to traffic VOC. These ions show a correlation coefficient (R) of approximately 0.9 with traffic emissions (Table S3). According to the diurnal profiles of traffic (Fig. 4) in summer, it shows distinct peaks during morning and evening hours.

The second factor was classified as weakly oxidized BVOC. It was identified by oxidation products of monoterpenes, specifically C₉H₁₅O1+ myrcenol (m/z 139.112), C₁₀H₁₃O1+ carvone (m/z 149.097), weakly-oxidized molecules of monoterpenes C₁₀H₁₅O1+ (m/z 151.112), C₁₀H₁₇O1+ (m/z 153.128) and C₁₀H₁₅O2+ (m/z 167.107) (Li et al., 2020a). These ions show strong correlations with the Oxidized BVOC factor, with R values of 0.93, 0.93, 0.93, 0.91, and 0.87, respectively. Oxidized BVOC have a nighttime peak in the diurnal profile and drop to nearly zero by noon, likely due to enhanced dilution and quicker oxidation to oxidation states not detectable by PTR-MS.

The third factor was identified as BVOC due to the predominance of C₁₀H17+ monoterpene (m/z 137.133) and its fragmentation C₆H9+ (m/z 81.07) ions in this VOC factor, with correlations of 0.98 and 0.86, respectively. Sesquiterpenes also show a strong correlation with this factor (R=0.80). The average diurnal behavior of BVOC shows an early morning peak, because BVOC, especially monoterpenes, are often stored in vegetation and released at the start of the morning due to sunlight and slight temperature increases (Malik et al., 2023), combined with a shallow boundary layer that keeps emissions near the surface.

The fourth factor showed a good correlation with butyric acid C₃H₅O3+ (m/z 89.024) and C₄H₅O3+ (m/z 101.024), with R values of 0.75 and 0.72, respectively. The time series of O₃ and this factor showed a good correlation during certain periods. The diurnal cycle of it exhibits a daytime peak, indicating involvement in photochemical oxidation processes. We classified it as aged VOC.

The fifth factor, biomass burning VOC, is dominated by propylene glycol C₃H₉O2+ (m/z 77.06) and orthoacetic acid C₂H₇O3+ (m/z 79.039), with strong correlations of 0.85 and 0.83, respectively. Additionally, a ring fragment of oxidized syringol, C₄H₇O4+ (m/z 119.034), and ring fragments of oxidized guaiacol and syringol molecules C₅H₃O2+ (m/z 95.013), C₅H₇O4+ (m/z 131.034) and C₄H₇O3+ (m/z 103.04) contribute smaller fractions but display even higher correlations, with values of 0.90, 0.91, 0.92 and 0.86 (Yee et al., 2013). Compared with the other 4 factors, the biomass burning VOC mass spectrum has more oxidized compounds.

In late winter and early spring, the first factor was identified as traffic VOC using the same method as in summer, showing high contributions and strong correlations with aromatic hydrocarbon ions, including toluene C₇H9+ (m/z 93.07), C₉ aromatics C₉H13+ (m/z 121.102), and cymene C₁₀H13+ (m/z 133.102) in Figs. S7, S8 and Table S5.

The second factor, classified as terpenes VOC, was marked by the predominance of monoterpene C₁₀H17+ (m/z 137.133) and its fragment C₆H9+ (m/z 81.07) ions with correlations of 0.90 and 0.89, respectively. Sesquiterpenes also show a strong correlation (R=0.92). The diurnal variation shows both morning and evening peaks, with the morning peak being higher than the evening peak. This pattern differs from typical traffic trends. This pattern is similar to that of limonene from shower gels (Yeoman et al., 2020), which shows higher concentrations in the morning compared to the evening. It correlates with the traffic factor because people typically shower in the morning before leaving home, leading to higher detection. By the evening, the limonene has dissipated.

The biomass burning VOC factor included a lot of smaller oxygenated VOC (OVOC) such as 1,3-propanediol C₃H₉O2+ (m/z 77.06), acetic anhydride C₄H₇O3+ (m/z 103.04), 2-furoic acid C₅H₅O3+ (m/z 113.024), maleic acid C₄H₅O4+ (m/z 117.019), butanedioic acid C₄H₇O4+ (m/z 119.034), benzoic acid C₇H₇O2+ (m/z 123.045) (Lemieux et al., 2004) and salicylic acid C₇H₇O3+ (m/z 139.04). They all show strong correlations (R=0.85–0.90) with this factor. Notably, C₅H₅O3+ and C₇H₇O3+ are recognized as tracers of biomass burning VOC (Li et al., 2020b; Romanias et al., 2024). It displays a peak during the daytime, possibly due to biomass burning activities in certain areas (e.g., cooking or outdoor grilling), which increase VOC emissions and lead to a daytime peak in the VOC diurnal profile.

The fourth factor was dominated by xylene, represented by C₈-aromatic hydrocarbon C₈H11+ (m/z 107.086), with a correlation of 0.96. This factor also shows a strong correlation with 3-ethyl-pyridine C₇H₁₀N1+ (m/z 108.081) and the C₉-aromatic hydrocarbon VOC C₉H13+ (m/z 121.102), with correlations of 0.98 and 0.91, respectively. Xylene and trimethylbenzene are components of flue gases from fossil fuel combustion VOC (Niu et al., 2021). It also displays morning and evening peaks in its diurnal cycle, leading to its identification as a traffic2 VOC.

The fifth factor has minimal correlation with most masses, only showing a correlation of 0.73 with benzene (C₆H7+) and weak correlations with various oxygenated compounds. Therefore, this factor may originate from different low-concentration emission sources. Its diurnal variation shows a small evening peak, likely influenced by traffic or industrial emissions. This factor accounts for 54 % ± 9 % in late winter, decreasing to 23 % ± 16 % in early spring due to the increased temperature (Fig. 5), and as background VOC are commonly found to have a high proportion, we classify it as background VOC.

The source contributions of VOC vary significantly across seasons, reflecting shifts in emission patterns and atmospheric conditions. In summer, the largest contributor is aged VOC at 42±18 %, indicating the dominance of secondary pollutants formed through atmospheric photochemical oxidation processing. Traffic emissions account for 22±14 %, while oxidized biogenic VOC make up 13±13 %, and fresh BVOC contribute 7±7 %. Biomass burning plays a notable role at 16±15 %, likely due to outdoor barbeque in warmer months. Winter exhibits a different profile, with background sources dominating at 54±9 %, suggesting stable atmospheric conditions. Traffic-related emissions split into two categories – traffic1 (26±6 %) and traffic2 (7±6 %) – possibly reflecting different vehicle types or fuel usage patterns. Monoterpenes VOC remain low (4±4 %), consistent with reduced biogenic activity and from anthropogenic source like personal care products (Wu et al., 2024), while biomass burning contributes 8±6 %, potentially from residential heating. In spring, the contributions are traffic1 22±7 %, monoterpene VOC increased to 9±9 % as vegetation becomes more active, biomass burning VOC 33±14 %, traffic2 13±8 %, and background 23±16 % (Fig. 5).

We included 153 SVOA ions (Table S6) for PMF analysis for summer and 171 SVOA ions (Table S7) for winter/spring, excluding abundant ions that could skew results and very low ion signals with minimal impact (Song et al., 2024). The average semi-volatile organic aerosol concentration was 0.3±0.2 µg m⁻³ in summer, 0.14±0.08 µg m⁻³ in winter and 0.07±0.07 µg m⁻³in spring. Through a comprehensive analysis of factor profiles, diurnal variations, and tracer correlations in summer (Figs. S11, S12, and Table S8) and in winter (Figs. S13, S14, and Table S9). Five semi-volatile organic aerosol (SVOA) factors were identified as the best interpretable solution for different seasons.

In summer, the first factor showed distinct signals from primary oxidation products of terpenes, including oxidized monoterpenes C₁₀H₁₅O1+ (R=0.93) and C₁₀H₁₇O1+ (R=0.89) (Li et al., 2020a). While the expected monoterpene oxidation products are typically C₉H₁₅O1–5+, we found that C₉H₁₃O1+ and C₉H₁₃O2+ also correlated well with this factor, with correlation coefficients R=0.91 and R=0.85, respectively. This suggests that the C₉H₁₅O1–2+ compounds may lose hydrogen, forming C₉H₁₃O1+ and C₉H₁₃O2+ (Gkatzelis et al., 2018b). Additionally, nitrogen-containing species like C₉H₁₃N₁O₁, C₉H₁₂N₁O₂, and C₉H₁₃N₁O₅ displayed high intensity and significant correlation with this factor, which may indicate they are fragments of monoterpene-derived organic nitrates, such as C₉H₁₅N₁O_5–7 (Massoli et al., 2018). The diurnal cycle shows a nighttime peak (Fig. S10c), which we have attributed this factor to weakly oxidized biogenic organic aerosol (Weakly OBOA) based on the presence of a monoterpene oxidation tracer and the observed diurnal pattern.

The second factor includes clear markers of isoprene oxidation products, specifically C₅H₈O_n species (Li et al., 2020a), with C₅H₉O6+ (R=0.83) and C₅H₉O3+ (R=0.82) showing strong correlations (Riva et al., 2016). Additionally, the ion with the highest fraction, C₃H₅O4+ (R=0.85), likely originates from the oxidation of ISOPOOH (Rios, 2018). However, benzoic acid C₇H₇O₂? (R=0.95), showing a strong signal, and phthalic anhydride C₈H₅O3+ (R=0.83) are both known emissions from biomass burning aerosols (Bruns et al., 2017; Koss et al., 2018). C₆H₈O₅ (R=0.84) (Molteni et al., 2018) is an oxidation product of benzene, while C₇H₁₀O₆ (R=0.84) as reported by Nakao et al. (2011), is an oxidation product of o-cresol, an emission from oxygenated aromatic BBOA. Therefore, we define this factor as mixed oxidized isoprene OA and BBOA.

The third factor was identified as BBOA due to its strong correlation (R=0.88) with vanillic acid C₈H₈O₄ (Fleming et al., 2020) and its prominent fragment (C₈H₆O₄) with even higher correlation (R=0.96), which constitutes a significant portion of this factor. Additionally, another BBOA tracer, syringic acid (C₉H₁₀O₅) (Wan et al., 2019), and its fragment (C₉H₆O₃, R=0.95), likely formed by the loss of two H₂O molecules, or as a direct BBOA emission (Fleming et al., 2020), further support this identification. Other ions, such as C₆H₆O₅ and C₇H₈O₅, are oxidation products of guaiacol (Yee et al., 2013). The diurnal cycle of this factor, peaking during the daytime, suggests that barbecue events may be contributing sources during the summer.

The fourth factor shows no correlation with other compounds and exhibits a distinct nighttime peak (Fig. S10c). However, it correlates strongly with nitrate detected by AMS (R=0.71). This suggests that the factor represents regional background, with its nighttime increase likely influenced by local accumulation effects.

The fifth factor was identified as more oxidized monoterpene OA, comprising highly oxidized monoterpene products such as C₁₀H₁₅O3-5+ (R=0.86, 0.85, 0.84), diacetin C₇H₁₃O5+ (R=0.79), oxidized molecules of monoterpenes C₈H₁₃O2+ (R=0.81), C₈H₁₃O4+ (R=0.80), and its fragment (C₈H₁₁O3+, R=0.83). Compounds like C₉H₁₃O3-4+ (R=0.81, 0.78) likely arise from fragments of the C₉H₁₄O_n series, representing more oxidized monoterpene products that have lost hydrogen. Compared to the diurnal pattern of less oxidized terpene products, this factor's peak displays a delay, suggesting that primary oxidation products form initially and then undergo further oxidation, resulting in more oxidized BOA (More OBOA).

In winter and spring, the first factor is identified as night-time aged BBOA due to the presence of distinct BBOA tracers, C₆H₆N₁O4+ and C₇H₈N₁O4+ (Fig. S14a), which do not appear in the mass spectra of other factors. Nitrocatechols (C₆H₅N₁O₄) originates from anthropogenic activities, including biomass burning and vehicle emissions. Meanwhile, methyl-nitrocatechols (C₇H₇N₁O₄) are specific markers for BBOA, as they are formed from m-cresol released during biomass combustion and diesel exhaust (Kourtchev et al., 2016). These tracers also exhibit the same strong correlation of 0.97 with this factor (Table S8). Syringic acid (C₉H₁₀O₅) also correlates well (R=0.70) with this factor. In terms of its daily cycle did not show peaks during morning or evening rush hours; instead, a peak was observed around 20:00 LT, suggesting it aligns more with BBOA. This factor showed a strong correlation with highly oxidized nitrogen-containing compounds and secondary organic aerosols (syringic acid). This factor is more related to highly oxidized nitrogen-containing secondary organic aerosols, indicating that nitrogen-containing compounds produced by BBOA may have undergone significant oxidation by NO3 radicals. Therefore, I identified it as night-time aged BBOA.

The second factor was characterized by a high fraction of oleic acid C₁₈H₃₅O2+ (R=0.90), C₁₆H₃₅O3+ (C₁₆H₃₃O₂(H₂O)⁺) (R=0.88), C₁₆H₃₃O2+ corresponding to palmitic acid (R=0.54), and C₁₆H₃₁O1+ is the fragment of C₁₆H₃₃O2+ (R=0.54). C₁₈H₃₄O₂ is identified as the cooking tracer oleic acid, while C₁₆H₃₃O2+ serves as another cooking aerosol tracer, corresponding to palmitic acid (Reyes-Villegas et al., 2018; Wang et al., 2020b; Huang et al., 2021). C₁₆H₃₁O1+ is thought to originate from C₁₆H₃₃O2+ due to fragmentation involving the loss of one H₂O molecule. These findings strongly suggest that this factor represents cooking aerosol.

The third factor was identified as aged combustion. In this factor, the aromatic hydrocarbon C₇H9+ accounts for a relatively higher fraction and exhibits a strong correlation (R=0.79). Although C₈H13+ does not contribute significantly as C₇H9+, it also shows the same strong correlation. These compounds are attributed to combustion sources (Wang et al., 2022). The ion C₇H₁₀N1+, potentially a fragment of acridine (C₇H₁₂N1+) originating from coal combustion (Wang et al., 2021), exhibits a distinct signal and the highest correlation (R=0.94). Similarly, C₆H₁₅?O3+ (R=0.93) corresponds to 2-methoxyethyl ether (MXEE), a product of fuel combustion. Monoterpene oxidation products, such as C₁₀H₁₇O1+ (R=0.86) and C₉H₁₃?O1+ (R=0.82), are also observed. This is likely because monoterpenes, serving as biofuel components in engines and boilers, form oxidized monoterpene products when combusted and oxidized (Dagaut et al., 2024). Additionally, benzocaine C₉?H₁₂?N₁?O2+? exhibits strong correlation (R=0.80) with this factor and is likely a product of oxidized combustion processes.

The fourth factor has C₆ carboxylic acids (C₆H₇O5+) which is the phenol oxidation products with OH radicals in the low-NO_x system, and C₇H₉O5+ denoted as the guaiacol with OH adduct. Propanedioic acid (C₃H₅O4+), pentanedione (C₄H₅O3+), and 2-oxopentanedioic acid C₅H₇O4+ are fragments derived from the OH oxidation products of biomass burning VOC (BBVOC) such as guaiacol and syringol (Yee et al., 2013). Additionally, C₆H₆O₂ is an OH radical oxidation product from phenol. A higher fraction of 2,5-di-(hydroxymethyl)furan C₆H₉O3+ was observed, which may result from C₆H₇O2+ binding with a water molecule. These photochemical products, formed from BBVOC oxidation by OH radicals, confirm that this factor represents aged BBOA. All the aforementioned ions exhibit a strong correlation of approximately 0.85 with this factor, as shown in Table S2. Furthermore, its diurnal pattern, with a peak at 15:00 LT during the day, leads to its identification as day-time aged BBOA.

The fifth factor shows a very strong correlation with stearic acid C₁₈H₃₇O2+ (R=0.95), C₁₆H₃₃O₂(H₂O)⁺ (R=0.91), C₁₈H₁₅O1+ (R=0.90), C₁₉H₁₅O₁ (R=0.88), and oleic acid C₁₈H₃₅O2+ (R=0.88). However, these compounds do not account for significant fractions in this factor's mass spectrum as they do in factor 2's mass spectrum. Notably, trimethoxy methane C₄H₁₁O3+ is a key ion in this factor but contributes minimally to others and shows no correlation with the factor (R=0.02). While the composition includes unrelated compounds, its strong correlation with sulfate detected by AMS (R=0.74) suggests a regional background origin, with the nighttime peak indicating local accumulation effects.

During summer, the composition of SVOA includes 11±15 % weakly OBOA, 16±15 % oxidized isoprene and BBOA (mixture of oxidized isoprene OA and BBOA), 25±21 % BBOA, 30±22 % attributed to regional background sources, and 18±16 % to More OBOA (Fig. 6). Although cooking-related SVOA ions were present during summer, the higher contribution of biogenic oxidation products in overlapping mass spectral regions likely reduced the statistical separability of a distinct cooking factor in the SVOC PMF analysis (Zhu et al., 2018; Coggon et al., 2024a). In winter, nighttime aged BBOA accounts for 10±9 %, cooking OA contributes 4±5 %, aged combustion OA makes up 9±10 %, daytime aged BBOA constitutes the largest fraction at 55±21 %, and regional background sources contribute 23±17 %. In spring, nighttime aged BBOA accounts for 12±12 %, cooking OA contributes 11±12 %, aged combustion OA makes up 24±18 %, daytime aged BBOA constitutes the largest fraction at 46±25 %, and regional background sources contribute 7±9 %.

This study thoroughly examined source factor profiles, diurnal patterns, and correlations with tracers (Figs. S15–S20) for OA particles. The average organic aerosol concentration was 4.3±2.9 µg m⁻³ in summer, 3.4±1.7 µg m⁻³ in winter and 1.8±1.8 µg m⁻³ in spring. Five OA factors were identified as the optimal interpretable solutions for summer and winter/spring, respectively. Alternative PMF solutions with different factor numbers were systematically evaluated based on Q/Qexp behavior, factor interpretability, and tracer correlations; details are provided in the Supplement (Fig. S14).

In summer, the first factor was identified as cooking organic aerosol (COA), based on its high H : C ratio (1.82), low O : C ratio (0.19) (Fig. S17a), higher contributions at m/z 55 (C₃H₃O⁺) and m/z 57 (C₃H₅O⁺), and the mass spectrum dominated by hydrocarbon ions. These characteristics are consistent with reported COA in urban areas (Elser et al., 2016; Äijälä et al., 2017; Liu et al., 2018). Additionally, fatty acid (C₁₆H₃₅O3+) detected by CHARON-PTR showed a moderate correlation (R=0.54) with the COA time series (Fig. 17b). The second factor is characterized by a mass spectrum also dominated by hydrocarbon ions, a similar O : C ratio, and a high O : C ratio, but with higher contributions at m/z 55 (C₄H7+) and m/z 57 (C₄H9+) (Elser et al., 2016). These features are consistent with the characteristics of hydrocarbon-like organic aerosol (HOA) which is related to traffic. additionally, HOA shows a good correlation (R=0.66) with the eBC time series. The third factor is assigned to Biogenic secondary organic aerosol (BOA), based on its H : C ratio of 1.63 and O : C of 0.44. The O / C ratio falls within the range typically associated with semi-volatile oxidized organic aerosol (SVOOA), which is approximately 0.35±0.14 (Setyan et al., 2012). And it has a good correlation of 0.86 with monoterpene oxidation products (pinonaldehyde) in CHARON-PTR. BOA's diurnal pattern (Fig. S17c) shows an increase starting at 20:00 LT, reaching its peak at 07:00 LT. This trend aligns with the nighttime oxidation process of monoterpenes, supporting its identification as BOA. The fourth factor was identified as aged biomass burning organic aerosol (BBOA) because its O / C ratio of 0.597 is slightly higher than that of fresh BBOA (0.15–0.5) and falls within the range of aged BBOA (0.5–0.87) (Ortega et al., 2013). Its diurnal shows a small peak at 15:00 LT and a higher peak at 22:00 LT, the nighttime peak indicates that fresh BBOA happened rapid dark aging process (Kodros et al., 2020). And it shows a strong correlation (R=0.93) with BBOA tracer levoglucosan detected by CHARON-PTR. therefore, the fourth factor is identified as aged BBOA. The fifth factor has the lowest H : C ratio (1.43) and the highest O : C ratio 0.75, which falls within the range of low-volatility oxygenated organic aerosol (LVOOA) (0.6–1.0) (Jimenez et al., 2009). LVOOA is dominated by CO⁺ and CO2+ and shows a strong correlation (R=0.70) with O₃, indicating that it is associated with photochemical oxidation processes (Kumar et al., 2016). Its diurnal pattern shows a daytime peak. These characteristics align well with the typical properties of LVOOA, leading to its identification as LVOOA. To ensure that every tracer detected by CHARON-PTR and correlated with AMS-PMF factors is representative, we compared different related fatty acids, toluene and trimethyl benzene, different monoterpene oxidation products, and different BBOA tracers. All of them showed good time-series correlations (Fig. S19).

In winter, we identified all the factors using the same method as in summer. The difference is that the traffic diurnal cycle (Fig. S18c) exhibits distinct peaks during the morning and evening rush hours. Fresh BBOA shows an evening peak, indicating evening residual heating activities. Aged BBOA is characterized by its correlation with C₆H₇O5+, which we identified as an oxidation product of BBOA in the CHARON-PTR section. its diurnal pattern is very flat, suggesting it may be associated with regional background levels or long-range transport. for LVOOA, NH4+ is used as an indicator, as it correlates strongly with inorganic aerosol during autumn and winter (Freney et al., 2011). Other different OA factors time series shown in Fig. S20.

In summer, OA contributions are as follows: traffic 8±8 %, cooking 11±13 %, BOA 22±14 %, aged BBOA 25±21 %, and LVOOA 33±20 %. In winter, the contributions are: traffic 5±4 %, cooking 12±12 %, fresh BBOA 13±9 %, aged BBOA 36±12 %, and LVOOA 33±17 %. In spring, the contributions are: traffic 9±7 %, cooking 19±16 %, fresh BBOA 27±17 %, aged BBOA 37±19 %, and LVOOA 9±10 % (Fig. 7).

Previous PMF studies in Stuttgart and Karlsruhe identified consistent summer source patterns: traffic-related OA contributes less than 10 %, SV-OOA accounts for approximately 16 %, while LV-OOA dominates at around 75 % of OA mass. This LV-OOA prevalence indicates strong biogenic influence from regional vegetation and photochemical processes enhanced by higher summer temperatures. In Stuttgart, elevated winter PM_2.5 stems primarily from biomass burning and residential heating, evidenced by enhanced levoglucosan and nitrated phenol signals, alongside increased traffic-related primary OA (POA). Both traffic OA and BBOA contribute substantially, though LV-OOA remains significant. Karlsruhe experiences similar source patterns with increased residential heating, traffic emissions, and coal combustion from industrial sources, but benefits from less severe inversion and stagnation conditions, resulting in lower overall concentrations. Munich's intermediate winter pollution levels suggest a source composition transitioning between these profiles, warranting detailed PMF analysis to characterize its specific emission contributions and compare with the established patterns in Stuttgart and Karlsruhe.

PMF source apportionment reveals strong seasonal dependencies in organic aerosol composition, with summer conditions driving the majority of aerosol burden in Munich. Summer demonstrates the highest concentrations across all source categories, dominated by aged BBOA (2.3 µg m⁻³) as the largest single contributor due to barbeque charcoal combustion. Photochemical LV-OOA (1.5 µg m⁻³) and biogenic emissions (1.4 µg m⁻³) emerge exclusively during summer, reflecting enhanced secondary aerosol formation under high temperatures and intense solar radiation that promote both biogenic emissions and photochemical processing. In contrast, late winter and spring show dramatically reduced organic aerosol levels, with late winter contributions limited primarily to regional background (0.4 µg m⁻³) and minimal aged BBOA (0.4 µg m⁻³). Traffic and cooking emissions remain consistently minor throughout all seasons (≤0.7 µg m⁻³), suggesting these local primary sources are less significant compared to secondary formation processes and regional biomass burning influences.

SVOA and OA composition exhibits distinct seasonal patterns driven by varying source emissions and atmospheric processing. Summer periods show elevated OA concentrations due to enhanced photochemical oxidation under high solar radiation, promoting secondary organic aerosol formation (Fig. 8). Biogenic sources contribute significantly during summer through increased vegetation emissions. Rapid atmospheric oxidation transforms fresh BBOA to aged BBOA, explaining the predominance of aged BBOA despite active barbecue activities. In late winter, residential heating dominated fresh BBOA emissions. During early spring, both heating and barbecue emissions contributed to fresh BBOA, with the latter increasing as temperatures rose. Consequently, Fresh BBOA concentrations increased from late winter to early spring. However, Fresh BBOA became less detectable in peak summer due to accelerated aging processes under higher temperatures and enhanced photochemical activity. Late winter and spring show substantial aged BBOA from residential heating activities. Cooking and traffic emissions remain consistent year-round sources, though their oxidation efficiency increases significantly in summer compared to winter and early spring when photochemical processes are less active.