Measurement report: Sources of carbonaceous aerosol in a South-East European metropolis

Platt, Stephen M.; Davidović, Milos; Bartonova, Alena; Ćirović, Željko; Eckhardt, Sabine; Evangeliou, Nikolaos; Gundersen, Hans; Jovanović, Maja; Jovašević-Stojanović, Milena; Močnik, Griša; Petrović, Bojana; Schneider, Philipp; Yttri, Karl Espen

doi:10.5194/egusphere-2026-1708

Preprints

https://doi.org/10.5194/egusphere-2026-1708

Preprints

10 Aug 2026

| 10 Aug 2026

Status: this preprint is open for discussion and under review for Atmospheric Chemistry and Physics (ACP).

Measurement report: Sources of carbonaceous aerosol in a South-East European metropolis

Stephen M. Platt, Milos Davidović, Alena Bartonova, Željko Ćirović, Sabine Eckhardt, Nikolaos Evangeliou, Hans Gundersen, Maja Jovanović, Milena Jovašević-Stojanović, Griša Močnik, Bojana Petrović, Philipp Schneider, and Karl Espen Yttri

Abstract. Carbonaceous aerosol measurements are scarce in the Western Balkans. We present year-round observations (2023 to 2025) from the urban-background site Ada Marina, Belgrade (44.790166° N, 20.4169° E; 71 m a.s.l.), combining three source apportionment (SA) approaches (organic tracers, the aethalometer model, and a novel application of non-negative matrix factorization, NNMF, to aethalometer data) with FLEXPART modelling and gridded emissions to identify source regions and transport pathways.

The annual mean equivalent black carbon (eBC) concentration was 1.42 ± 1.40 µg m⁻³, indicating a moderate eBC burden, while PM₁₀ (27.7 µg m⁻³) remained below current EU and Serbian limits but above WHO guidelines and upcoming EU limits. All SA methods consistently show strong winter to summer seasonality driven by residential wood combustion (RWC). The aethalometer model attributes 64 % of eBC to solid fuel (eBC_SF) in winter versus 18 % in summer, NNMF yields 51 % versus 6 %, and offline analysis of tracers show elemental carbon from biomass burning (EC_BB) contributions of 48 % versus 3 %.

Organic carbon (OC) shows similar seasonality, with biomass burning contributing 59 % in winter and 4 % in summer. A substantial wintertime secondary OC component linked to RWC is evident, with secondary OC (SOC) strongly correlated with EC_BB and eBC_SF (R² = 0.78 to 0.86), alongside enhanced brown carbon absorption (29 % annually; 44 % in winter).

FLEXPART indicates pollution episodes are dominated by regional transport within Serbia and the Pannonian Basin, while correlations between modelled residential emissions and observations demonstrate strong model skill.

Received: 26 Mar 2026 – Discussion started: 10 Aug 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 2374 KB)

Supplement (685 KB)

Download & links

Status: open (until 16 Nov 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2026-1708', Anonymous Referee #1, 14 Oct 2026 reply
This manuscript presents year-round observations of carbonaceous aerosol at an urban background site in Belgrade, Serbia. The study combines a comprehensive suite of measurements, including filter-based OC/EC and organic tracers, eBC and BrC from AE33, with source apportionment and FLEXPART transport model. The results highlight the pronounced wintertime influence of residential combustion on carbonaceous aerosol concentrations. The manuscript is generally well written, and its main value lies in addressing the limited availability of long-term carbonaceous aerosol measurements in the Western Balkans. However, several methodological inconsistencies need to be resolved, and some conclusions are not fully supported by the current analysis, as detailed below. I therefore recommend publication following major revision.
Major Comments
Section 2.3.2. NNMF is used to separate the measured absorption spectra into two factors. I think the mathematical analysis is thorough with stable factors, but more caution is needed in interpreting the results because distinct absorption spectra do not necessarily correspond to unique emission sources. The synthetic test in Fig. S6 is useful but does not fully represent ambient conditions, where absorption spectra may deviate from simple power laws, as the authors acknowledge in Fig. 6, and may change during atmospheric processing. Furthermore, agreement with the aethalometer model does not provide fully independent validation because both approaches rely on the same measurements for source separation. For example, enhanced secondary BrC absorption during winter could be attributed to the solid fuel component by both methods because of its stronger wavelength dependence. The strong correlation between the NNMF solid fuel factor and independently estimated ECBB provides more convincing evidence. However, solid fuel combustion is not equivalent to biomass burning, and neither necessarily identifies residential wood combustion exclusively (detailed in Comment 4). These distinctions and the associated uncertainties in source attribution need to be discussed more carefully.

Section 3.1. The attribution of wintertime SOC to residential wood combustion requires more caution. SOC is estimated as the residual, so uncertainties in primary OC estimates and contributions from unaccounted sources may affect the inferred SOC concentrations. Although a Monte Carlo uncertainty analysis is described, the propagated uncertainty in SOC estimates needs to be more clearly quantified and discussed. Correlations between SOC and biomass burning tracers suggest a possible association with residential wood combustion, but they do not necessarily indicate secondary formation from wood combustion precursors. Common meteorological conditions and co-varying emissions may also contribute to these correlations.

Section 3.4. The authors select the sector combination that best correlates with the observations and then interpret this agreement as evidence of strong model performance. However, the resulting correlation is not an independent measure of model performance. In addition, the Taylor analysis provides useful information on correlation and variability but does not account for mean concentration bias. I recommend a scatterplot of modeled and observed concentrations.

Throughout the manuscript, solid fuel, biomass burning, and residential wood combustion are sometimes treated interchangeably, as are liquid fuel and traffic, although these categories are not equivalent. Coal is also used for residential heating and can produce BrC with elevated AAE. Combustion of some lignites can also emit levoglucosan, complicating its interpretation as an exclusive biomass burning tracer. Mannosan and galactosan are listed in Table 1 but not reported in Table 3. Examining levoglucosan/mannosan ratios could provide additional information on wood type and potential coal contributions, which may also help constrain the assumed emission ratios. Similarly, the low AAE factor may include fossil fuel combustion sources other than traffic, such as industrial and non-road emissions.

Minor Comments
Eq. 4 is labelled as liquid fuel absorption, but the equation actually gives the solid fuel contribution. Likewise, Eq.5. Also in Eq. 5, the second term in the numerator should be B_ABS(880 nm).

Table 3. The apportioned EC and OC components do not sum to the measured totals, e.g., EC_FF+EC_BB vs. EC. Please explain whether mixing statistics were used, reconcile the results, and check their implications for Figs. 3 and 4.

L210 and L232. Offline filters are for PM10, whereas the aethalometer uses a PM2.5 inlet. Please discuss how these differences may affect the agreement among source apportionment approaches. For the AE33 measurements, please clarify whether the sampled aerosol was dried and describe the QA/QC procedures and data coverage of the hourly record. For filter sampling, please provide the sampling duration and flow rate, and clarify whether OC concentrations were corrected using field blanks.

L250. Seven tracer samples per month were selected. This is not a random sampling approach and may not preserve the annual mean of individual carbonaceous species. Please demonstrate the representativeness of the selected samples relative to the full filter dataset.

L271. This may exclude periods with substantial non-power-law BrC absorption. Please indicate the fraction of data retained and whether the derived source fractions are sensitive to this criterion.

L405. The Monte Carlo analysis assumes a uniform uncertainty of 25% for all measured parameters. Please justify this assumption.

Fig. 4. The current EU annual PM10 limit value is 40 µg/m3. 50 µg/m3 is the daily limit value.

L598. A low occurrence of negative contributions does not indicate that the selected AAEs are too low. If AAE_SF is underestimated, periods with observed AAEs exceeding AAE_SF can produce negative liquid fuel contributions. Please clarify the interpretation and describe how negative source contributions were handled in the subsequent analysis.

Technical Corrections
Line 299. Correct “demotes”

Please check figure and section cross-references throughout the manuscript for consistency.

Reply
Citation: https://doi.org/10.5194/egusphere-2026-1708-RC1
RC2:
'Comment on egusphere-2026-1708', Anonymous Referee #2, 26 Oct 2026 reply
General comments
This manuscript presents a valuable year-round characterization of carbonaceous aerosol at an urban-background site in Belgrade, Serbia. I particularly appreciate the combination of offline OC/EC and organic-tracer measurements, multi-wavelength Aethalometer observations, source-apportionment approaches, and FLEXPART simulations. The study is relevant because long-term source-apportionment data from the Western Balkans are still limited. Overall, I find the manuscript interesting and potentially suitable for publication, but several methodological points need clarification. I therefore recommend major revision.
Major comments
The manuscript links wintertime SOC to residential wood combustion based partly on the stronger correlation of OCSOC with OCBB than with OCFF. However, OCSOC is not measured independently but calculated as the residual. Because the compared quantities are mathematically related and also depend on assumed source ratios, these correlations should be interpreted with caution. I suggest softening the statement that the correlation demonstrates an RWC origin of wintertime SOC. It provides supporting evidence when considered together with levoglucosan, seasonal behaviour, optical properties, and transport information, but it does not by itself establish causality. If possible, a sensitivity analysis across the Monte Carlo realizations or using different plausible source ratios would help demonstrate the robustness of the SOC–biomass-burning relationship. Otherwise, this limitation should be discussed explicitly.

Only 84 exposed filters were selected for organic-tracer analysis, corresponding to seven samples per month. The samples were chosen at specific points of the monthly distribution, including the minimum, maximum, and several percentiles of a composite chemical metric. I understand the motivation for covering a broad range of chemical conditions, but this selection is not equivalent to random or time-representative sampling. In particular, systematically including the minimum and maximum each month could influence seasonal or annual arithmetic means if all selected samples are given equal weight. Please explain more clearly how representative seasonal and annual means were obtained from this subset. Were the selected samples weighted according to their corresponding portions of the monthly distribution? If not, I think the potential bias introduced by the sampling strategy should be quantified or at least discussed, especially because several reported source fractions are based on these samples.

Several very high correlations are reported between source-apportioned components, including values close to 0.99. However, some of these variables are not independent. For example, ECBB and OCBB are both calculated from levoglucosan using assumed emission ratios, while OCFF is derived from ECFF using an OC/EC ratio. I therefore think the manuscript should distinguish more clearly between comparisons that provide independent validation and correlations that are partly expected because the variables share common inputs or calculation procedures. This is particularly important where terms such as “good agreement” or “validation” are used. A short discussion of this non-independence would make the interpretation more balanced.

Minor comments
The statistical terminology should be checked carefully. Table 6 is described as presenting Pearson correlation coefficients, while the text repeatedly refers to “Pearson’s R²”. These are not interchangeable. This is particularly important for the Taylor diagram, where the angular coordinate conventionally represents the correlation coefficient R, not R². Please verify which quantity was calculated and plotted and make the terminology consistent throughout the text, tables, figures, and captions.

Section 2.4 states that the FLEXPART simulations extended 30 days backward in time, whereas the caption of Figure 10 describes 21-day backward simulations. Please clarify which duration was actually used and correct the manuscript accordingly.

The manuscript correctly notes that every-second-day sampling does not allow a formal comparison with the regulatory daily exceedance criterion. However, I find the statement that the observed frequency is “indicative of annual exceedance rate below the allowed limit” too strong. Since approximately half of the days were not sampled, exceedances on unsampled days cannot be excluded. I suggest removing this inference or making it more cautious.

At several points, the manuscript discusses local residential wood-combustion emissions, while the FLEXPART analysis indicates important regional transport from Serbia and the Pannonian Basin. These findings are not contradictory, but I think the distinction between source type and source location should be clearer. Statements attributing SOC or BrC specifically to “local emissions” should therefore be supported more quantitatively or phrased more cautiously.

Technical corrections
Figure 4 caption: correct the current annual PM10 limit from 50 to 40 µg m⁻³ and modify the plotted reference line accordingly.

Figure 5 caption appears to skip panel “e” when describing the panels (“d” followed by “f”). Please check the panel lettering.

Figure 7 caption: “magenta subsets” appears to be missing a closing parenthesis.

Check consistency in the notation BABS/BAbs and eBCSF throughout the manuscript.

Reply
Citation: https://doi.org/10.5194/egusphere-2026-1708-RC2

Supplement

https://doi.org/10.5194/egusphere-2026-1708-supplement

Model code and software

Source apportionment of multi-wavelength aerosol absorption using NNMF and aethalometer-based methods (Igor Pro code) Stephen M. Platt https://doi.org/10.5281/zenodo.19221204

Viewed

Total article views: 423 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
230	144	49	423	63	48	44

HTML: 230
PDF: 144
XML: 49
Total: 423
Supplement: 63
BibTeX: 48
EndNote: 44

Views and downloads (calculated since 10 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	62	38	9	109
Sep 2026	72	59	23	154
Oct 2026	96	47	17	160

Cumulative views and downloads (calculated since 10 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	62	38	9	109
Sep 2026	72	59	23	154
Oct 2026	96	47	17	160

Viewed (geographical distribution)

Total article views: 358 (including HTML, PDF, and XML) Thereof 358 with geography defined and 0 with unknown origin.

Country	#	Views	%

Cited

Latest update: 29 Oct 2026

Short summary

We present year-round (2023 to 2025) carbonaceous aerosol measurements from Ada Marina, Belgrade, combining tracers, the aethalometer model, a new NNMF application, and FLEXPART. All methods show residential wood combustion is important, driving winter increases in eBC and OC, enhanced secondary OC, and brown carbon absorption. FLEXPART highlights regional transport within Serbia and the Pannonian Basin, with good model–observation agreement, underscoring the need to reduce RWC emissions.


Total:	0
HTML:	0
PDF:	0
XML:	0

Measurement report: Sources of carbonaceous aerosol in a South-East European metropolis

Supplement

Model code and software

Viewed

Viewed (geographical distribution)

Cited

2 citations as recorded by crossref.