Field measurements were conducted at Xichong site (22.48° N, 114.56° E; Fig. S1 in the Supplement) on the Dapeng Peninsula in Shenzhen, China, from August to September 2022. The analysis period used in this study was from 27 August to 9 September 2022, corresponding to the overlapping operation time of all deployed instruments. Located about 60 km from the city center, Xichong site is surrounded by the sea and is distant from urban areas and industrial sources, with over 90 % forest coverage. Due to minimal local anthropogenic interference, Xichong site serves as a regional atmospheric background station in South China.

During the field campaign, an aethalometer (AE31, Magee Scientific, USA) operating at seven wavelengths (370, 470, 520, 590, 660, 880, and 950 nm) and a photoacoustic extinctiometer (PAX, Droplet Measurement Techniques, USA) measuring at 532 nm were utilized to detect the online optical properties of particles. A Monitor for AeRosols and Gases in Ambient air (MARGA; Metrohm-Applikon, Netherlands) was used to detect the online water-soluble ion concentration (NH4+, Na+, K+, Ca2+, Mg2+, SO42-, NO3-, Cl-). Detailed information regarding the instrumentation and measurement uncertainties of the aethalometer is provided in the Supplement (Sect. S1). In this study, the time resolution of all online data was standardized to 1 h. Offline filter sampling was also carried out simultaneously during the field campaign. A high-volume sampler (XT1025, XTrust Analytical Instruments, China) with a flow rate of 1 m3min-1 was used to collect PM_2.5 samples on the pre-baked quartz filters with a sampling period of 24 h for each filter. Details on the filter pretreatment procedures are available in the Supplement (Sect. S2). The filters were further analyzed to measure the BrC mass, optical properties, and molecular characteristics.

Other measurements, including the mass concentration of PM_2.5 and O3 and the meteorological factors (temperature, relative humidity, wind speed, and wind direction), were conducted at the sampling site. The Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT-4) model developed by the Air Resources Laboratory (ARL) of the National Oceanic and Atmospheric Administration (NOAA; USA) was employed to compute 24 h air mass back-trajectories at a 50 m arrival height.

The mass absorption cross-section (MAC; m2g-1) of BrC can be calculated according to the following equation: 1MAC(λ)=babs,BrC(λ)[BrC], where babs,BrC(λ) is the light absorption coefficient (Mm-1) of BrC at a given wavelength λ, derived by subtracting the corresponding absorption coefficient of BC from the total particle absorption coefficient. Here, we used both online and offline methods to calculate the MAC of BrC.

A high-performance liquid chromatographer (HPLC) equipped with a photodiode array (PDA; G7117C, Agilent, USA) detector and a high-resolution mass spectrometer (HRMS; G6545A, Agilent, USA) was utilized to identify the molecular composition, determine the relative abundance, and measure the corresponding light absorption of BrC. The HPLC was equipped with a C18 column (EC-C18, 3mm×150mm, 2.7 µm particles, Agilent, USA), using mobile phases of 0.1 % formic acid–water (A, HPLC grade) and 0.1 % formic acid–acetonitrile (B, HPLC grade). Gradient elution for each sample was performed with the A–B mixture as 0–1 min hold at 95 % A, 1–20 min linear decreased to 5 % A, 20–27 min hold at 5 % A, and then 27–30 min hold at 95 % A. The HRMS was set with a soft electrospray ionization (ESI) source in full scan, operating in both positive and negative ion modes. Raw data from mass spectrometry were processed using MassHunter Qualitative Analysis (v10.0). Molecule concentrations were semi-quantified based on the intensity from mass spectrometry (Kruve, 2019; Zhang et al., 2023).

The absorbance of the PDA at 370 nm was selected as the intensity of light absorption of BrC (Hecobian et al., 2010; Wen et al., 2021). Using a peak extraction algorithm in MassHunter Qualitative Analysis, approximately 20 absorption peaks per sample were identified. The absorption intensity at a specific retention time was determined by subtracting the blank absorption from the sample absorption. Peaks were grouped by overlapping retention times, with molecules in each group recorded along with their absorbance detected by the PDA. We employed a partial least-squares regression (PLSR) model to attribute individual molecular absorbance (Sect. S4), clarifying the relationship between absorbing molecules and the absorbance of individual peaks (Zhang et al., 2023).

During our sampling period, wind at the Xichong site predominantly came from two directions: northwest and northeast (Figs. 1a and S4). Air masses from the northwest primarily originated from inland areas of the peninsula, while those from the northeast were from the sea (Fig. S1). High ozone levels were observed at times, mainly during the daytime, and were associated with relatively high nitrate concentrations (Fig. 1b and c). Additionally, water-soluble potassium, as a marker for biomass-combusted aerosols (Zhai et al., 2015), also exhibited a time of elevated levels during our observation (Fig. 1c). Based on distinct meteorological and pollutant concentration characteristics, we selected three typical cases for detailed analysis of their optical properties. The selection criteria for each case were as follows: Case 1, the ozone case, with (1) the concentration of O3>100 ppb, (2) the concentration of [NO3-]>0.6 µgm-3, (3) wind speed <3 ms-1, and (4) consecutive duration >6 h (red shading in Fig. 1); Case 2, the transport case with (1) wind direction >270°, (2) wind speed >4 ms-1, and (3) consecutive duration >6 h (blue shading in Fig. 1); and Case 3, the combustion case, with (1) the concentration of [K+]>0.2 µgm-3 and (2) consecutive duration >6 h (yellow shading in Fig. 1).

Meanwhile, the HYSPLIT 24 h air mass backward trajectories indicate that, during Case 1, air masses predominantly originated from areas close to the sampling site. Combined with low wind speeds (<3 ms-1), meteorological conditions limited atmospheric dispersion, promoting ozone accumulation and secondary pollutant formation (Fig. S5). In Case 2, the air mass trajectories were from the inland region, with strong wind speeds that facilitated the transport of pollutants to the sampling site. For Case 3, the air mass trajectories also originated from the interior region but were associated with lower wind speeds than in Case 2.

Polar plots of wind direction, wind speed, and MAC_BrC,370 were further analyzed for cases 1–3 (Fig. 2). In Case 1, the average wind speed was 2.06 ms-1, with pollution mainly from local sources (Fig. 2a). During Case 1, the average PM_2.5 concentration was 8.05 ± 2.67 µgm-3, and the average MAC_BrC,370 was 1.25 ± 0.56 m2g-1 (Fig. 2d). For Case 2, the wind primarily came from the northwest, passing over the peninsula, with an average wind speed of 7.81 ms-1 (Fig. 2b). The average PM_2.5 concentration and MAC_BrC,370 for Case 2 were 4.87 ± 2.36 µgm-3 and 2.68 ± 0.30 m2g-1, respectively. In Case 3, the wind speed averaged 2.19 ms-1, with erratic wind directions (Fig. 2c). The average PM_2.5 concentration and MAC_BrC,370 for Case 3 were 5.05 ± 1.32 µgm-3 and 3.42 ± 0.41 m2g-1, respectively.

Among the three cases, although the average PM_2.5 concentration in Case 1 was the highest, its MAC_BrC,370 was the lowest, indicating that the light-absorbing ability of BrC in this high-ozone scenario was relatively weak. The low wind speed in Case 1 limited the influx of transported pollutants. High concentrations of ozone and [NO3-] indicated that the aerosols in Case 1 were primarily secondary and highly aged. However, Case 3, characterized by a high concentration of potassium and identified as a plume from combustion sources, had the highest MAC_BrC,370 in our observations, indicating the strongest light-absorbing ability of combusted BrC among the cases.

In real-world atmospheric environments, BrC aerosols often originate from a mixture of sources, and complete source separation is rarely achievable in field studies. While our case-based framework aimed to identify periods with a dominant emission influence, we acknowledge that source mixing may still occur and introduce variability in the retrieved optical parameters. As such, the reported MAC and AAE values should be interpreted as reflecting source-dominant conditions rather than pure-source characteristics. Nevertheless, the clear contrasts in chemical composition and optical responses across cases suggest that dominant sources exert a meaningful influence on BrC absorption. This reinforces the relevance of our findings for understanding BrC behavior under realistic ambient conditions, despite the limitations of bulk sampling and complex source environments.

By compiling BrC light-absorption measurements reported in 20 studies, Saleh et al. (2020) classified BrC into four classes, each with characteristic MAC_BrC,550 and AAE values: very weakly absorbing BrC (VW-BrC; MAC_BrC,550 of 1.3×10-3–1.3×10-2, AAE of 7–10), weakly absorbing BrC (W-BrC; MAC_BrC,550 of 1.3×10-2–0.13, AAE of 5–8), moderately absorbing BrC (M-BrC; MAC_BrC,550 of 0.13–1.3, AAE of 2.5–5), and strongly absorbing BrC (S-BrC; MACBrC,550>1.3, AAE of 1.5–2.5). The optical properties defining these BrC classes are expected to be associated with their corresponding physicochemical properties, such as molecular size, volatility, and solubility. In this study, the AAE values for both online measurements of BrC and filter-based offline measurements of WSOC were calculated in the wavelength range of 370 to 550 nm, referred to as AAE_370–550.

For online measurements of BrC, the optical results show an approximately linear correlation. In Case 1, results fall into both the W-BrC and M-BrC categories, whereas results for Case 2 and Case 3 fall primarily into the M-BrC category (Fig. 3a). In Case 1, where the ozone concentration is high, BrC shows weaker light-absorbing ability and stronger wavelength dependence compared to cases 2 and 3. BrC in Case 3 exhibits a high light-absorbing ability with low wavelength dependence. For Case 2, a portion of the optical results overlaps with those from Case 3, possibly due to the transported air mass originating from a similar source to that in Case 3.

For the filter-based offline measurements of WSOC, the trend of AAE_{370–550,WSOC} and MAC_WSOC,550 is consistent with the online results, showing an inverse correlation (Fig. 3b). The sample in Case 1 shows the highest wavelength dependence and the lowest light-absorbing ability of WSOC. It is worth noting that, although the ozone concentration was also high during Case 3, its optical results did not exhibit the same high wavelength dependence as observed in Case 1. The possible reason could be that primary WSOC produced by combustion has stronger light absorption, which dominated the optical behavior of WSOC during Case 3.

The offline MAC values based on WSOC extractions do not account for Mie scattering effects due to the lack of particle-phase interactions in liquid measurements (Liu et al., 2013; Zeng et al., 2020). Moreover, because particle size distribution and particle mixing state information was not available during the sampling period, Mie model corrections were not performed in this study. Therefore, direct quantitative comparisons between offline and online MAC values may involve uncertainties. Nevertheless, while the absolute MAC values from the two methods are not directly comparable, the observed trends between the two approaches are generally consistent. This consistency provides additional confidence in the robustness of the observed variations in BrC optical properties across different cases.

Saleh et al. (2020) suggested that VW-BrC primarily originates from secondary BrC, W-BrC mainly comes from smoldering BrC, and M-BrC is mainly associated with high-temperature BrC. However, in our observations, we found that Case 1, which we consider to be dominated by secondary BrC, still falls within the W-BrC or even M-BrC regions for both online airborne measurements and filter-based offline analysis. Possible reasons could be as follows: (1) unlike laboratory studies, field environments have greater diversity and uncertainty in BrC sources, and (2) differences in measurement methods may lead to variations in the results (Bond and Bergstrom, 2006; Saleh, 2020). Although the results for Case 2 fall solely within the S-BrC category, we believe that particles during this period are transported from inland urban areas, where the sources are more complex, including contributions from traffic emissions, industrial combustion, and secondary sources.

The optical properties of BrC can be affected by atmospheric aging processes such as photochemical bleaching and secondary browning (Zhao et al., 2015). Previous studies have demonstrated that such transformations can occur over timescales of several hours to 1 d, depending on oxidant levels, radiation intensity, and humidity (Forrister et al., 2015; Washenfelder et al., 2015). In this study, although we did not explicitly isolate aging effects, back-trajectory and chemical evidence suggests that the BrC observed was predominantly regionally influenced, with estimated transport times generally within this aging-relevant range. Therefore, the reported MAC and AAE values likely represent moderately aged BrC.

The water-soluble organic carbon (WSOC) species were ionized using ESI+ and ESI- ionization modes to detect the organic compounds. The identified molecules were categorized into groups based on atom composition: CHO, CHON, CHOS, and CHONS. The van Krevelen (VK) diagram is a widely used graphical method that plots H/C ratios against O/C ratios in molecular formulas to qualitatively identify the major chemical species in WSOC (Kim et al., 2003). In this study, the VK space is divided into seven regions based on previous studies: (1) lipid-like (O/C = 0–0.3, H/C = 1.5–2.0), (2) aliphatic/protein-like (O/C = 0.3–0.67, H/C = 1.5–2.2), (3) carbohydrate-like (O/C = 0.67–1.2, H/C = 1.5–2.4), (4) unsaturated hydrocarbons (O/C = 0–0.1, H/C = 0.7–1.5), (5) lignins/carboxylic-rich alicyclic-molecule-like (CRAM) (O/C = 0.1–0.67, H/C = 0.7–1.5), (6) tannin-like (O/C = 0.67–1.2, H/C = 0.5–1.5), and (7) condensed aromatics (O/C = 0–0.67, H/C = 0.2–0.7) (Feng et al., 2016; Ohno et al., 2010; Zeng et al., 2024). The sizes of scatters in Fig. 4 are proportional to the absorbance.

For each case, filters were selected to coincide with the core pollution periods, characterized by stable meteorological conditions and elevated pollutant concentrations. Although the number of samples was limited, the chemical results are considered reasonably representative of the dominant source influences during these periods. In Case 1, CHO compounds, which account for 36.5 % of the absorbance, are the most abundant form of BrC. These CHO compounds likely contain carboxyl or hydroxyl functional groups. The light-absorbing CHO compounds may originate from biomass burning smoke (Desyaterik et al., 2013; Chen et al., 2022a, 2023; Zhou et al., 2022) and have also been detected in water-soluble organic carbon (WSOC) and cloud water (Bianco et al., 2018; Kourtchev et al., 2016). Secondary CHO compounds, including typical dimers of α-pinene and diterpenoid derivatives, have also been detected in previous studies (Kourtchev et al., 2014; Kristensen et al., 2014; Gómez-González et al., 2012). The CHOS group in Case 2 contributes the highest relative absorbance (29.5 %). The CHOS compounds are considered to contain long aliphatic carbon chains with low aromaticity and are typically derived from anthropogenic emissions, such as diesel vehicles (Tao et al., 2014), coal combustions (Song et al., 2019), and vessels (Cui et al., 2019). The CHON group in Case 3 exhibits the highest relative absorbance (43.2 %). It is worth noting that, although CHON is not the most abundant group in terms of molecular abundance, its relative absorbance is the highest, suggesting that CHON compounds have a strong molecular absorption capacity. The CHON compounds have been found to be mainly derived from biomass burning species, such as nitrophenols, nitrocatechols, and nitroguaiacols. (Kourtchev et al., 2015; Zhang et al., 2013; Song et al., 2018). Several CHON species consistent with indole-derived structures, including isatin (C8H5NO2) and nitroindole (C8H6N2O2, C8H7NO4), were detected in the mass spectra. The identification of these compounds supports the attribution of the observed BrC to biomass burning sources (Baboomian et al., 2023; Chen et al., 2023; Mayorga et al., 2022; Jiang et al., 2019; Montoya-Aguilera et al., 2017) and highlights the complexity of nitrogen-containing brown carbon species in ambient aerosols. To provide a clearer overview of the molecular-level characteristics of BrC identified in this study, we summarized the major light-absorbing organic compounds and structures detected in the field samples in Table S1 in the Supplement. We further conducted a correlation analysis between the relative absorbance of CHON and the MAC of WSOC throughout the whole sampling period, finding that, as the relative absorbance of CHON increases, the MAC of BrC also becomes larger (Fig. S7). The measurement of chemical molecules provides support for the results corresponding to our optical observations in different cases.

In the VK diagram, aliphatics, lignins, and carbohydrates dominate in all three cases. In Case 3, WSOC shows a higher proportion of absorbance from lignins, which are commonly attributed to biomolecules and biomass burning species (Kitanovski et al., 2014). The differences in the chemical molecular compositions of BrC across the different cases during our observations led to variations in the light absorption of organic matter.