Oxidative potential apportionment of atmospheric PM<sub>1</sub>: A new approach combining high-sensitive online analysers for chemical composition and offline OP measurement technique

Camman, Julie; Chazeau, Benjamin; Marchand, Nicolas; Durand, Amandine; Gille, Grégory; Lanzi, Ludovic; Jaffrezo, Jean-Luc; Wortham, Henri; Uzu, Gaëlle

doi:https://doi.org/10.5194/egusphere-2023-1441

Preprints

https://doi.org/10.5194/egusphere-2023-1441

Preprints

28 Jul 2023

| 28 Jul 2023

Oxidative potential apportionment of atmospheric PM₁: A new approach combining high-sensitive online analysers for chemical composition and offline OP measurement technique

Julie Camman, Benjamin Chazeau, Nicolas Marchand, Amandine Durand, Grégory Gille, Ludovic Lanzi, Jean-Luc Jaffrezo, Henri Wortham, and Gaëlle Uzu

Abstract. Source apportionment models were widely used and/or atmospheric chemical processes. These technics are necessary to target the sources affecting air quality and to design effective mitigation strategies. More, the evaluation of the toxicity of airborne particulate matter gains interest as the PM concentrations classically measured appear insufficient to characterise the human health impact. Oxidative Potential (OP) measurement have recently been developed to quantify the PM capability to induce an oxidative imbalance in lungs. As a result, this measurement unit could be a better proxy than PM mass concentration to represent PM toxicity. In the present study, two source apportionment analyses were performed using Positive Matrix Factorization (PMF) from organic aerosol (OA) mass spectra measured at 15 min time resolution using a Time of Flight-Aerosol Chemical Speciation Monitor (ToF-ACSM) and from 19 trace elements measured on an hour basis using an online metals analyser (Xact). The field measurements were carried out in summer 2018. While it is common to perform PMF studies individually on ACSM and more recently on Xact datasets, here we used a two-step methodology leading to a complete PM₁ source apportionment. The outputs from both OA PMF and Xact PMF, the inorganic species concentrations from the ACSM and the black carbon (BC) fractions (fossil fuel and wood burning) measured using an Aethalometer (AE33) were gathered into a single dataset and subjected to a combined PMF analysis. In overall, 8 factors were identified, each of them corresponding to a more precise source than performing the previous single PMF analyses. The results show that besides the high contribution of secondary ammonium sulfate (28 %) and organic nitrate (19 %), about 50 % of PM₁ were originated from combustion processes (traffic, shipping, industrial, cooking and biomass burning emissions). Simultaneously, PM₁ filters were collected during the experimental period on a 4 hours sampling basis. On these filters, two acellular OP assays were measured (dithiothreitiol; OP_DTT and ascorbic acid; OP_AA) and an inversion method is applied on factors issued from all PMFs to assess contributions of the PM sources to the OP. This work highlights the sensitivity of OP_AA toward industrial and dust resuspension sources and those of OP_DTT toward secondary ammonium sulfate, shipping and biomass burning.

Received: 29 Jun 2023 – Discussion started: 28 Jul 2023

Download & links

Preprint (PDF, 1858 KB)

Supplement (2381 KB)

Download & links

Julie Camman et al.

Status: final response (author comments only)

RC1:
'Comment on egusphere-2023-1441', Anonymous Referee #1, 21 Aug 2023

Research on the composition and sources of particulate matter is of interest given its potential impact on health. There is growing concern about submicron particles, as they can penetrate deep into the body. Knowledge of the potential toxicity of fine particles emitted from different sources and formed by different processes is essential to apply appropriate PM pollution reduction measures. The oxidative potential (OP) is currently being evaluated as a parameter to quantify the ability of PM to oxidize molecules through the production of reactive oxygen species. Recent studies concluded that OP may be a better parameter than PM concentrations to monitor the health impact of particulate matter pollution.

The current study has performed the source apportionment (SA) analysis of PM1 by applying PMF to a pooled data set obtained from high-resolution (<60 min) measurements performed during the summer of 2018 (7-week period) in Marseille, in a background site with industrial influence. At the same time, the OP (OPDTT and OPAA) was determined offline in 4-h filters, with a relatively high frequency of time resolution for OP analyses. The contribution from the identified sources was used to apportion their contribution to OP by applying multilinear regression analysis; where the dependent variable OP is explained by a linear combination of the contribution of the sources of PM as independent variables. Thus, this method makes it possible to evaluate the oxidation capacity of the identified PM1 sources.

For the PM1 source apportionment, the authors applied a double PMF method based on the PMF2 proposed by Petit et al. (2014), where PMF was applied to a dataset combining the results of the SA studies. Thus, Pettit et al, applied PMF to a data set composed of the result of applying PMF to the OA data set (obtained by ACSM), the BCff and BCwb concentrations (estimated by applying the aethalometer model to the AE33 measurements) and the ion concentrations measured by the ACSM.

In the present study, in addition to these data sets, the outputs of applying PMF to online measurements of metals (every 60 minutes) by using an online Xact XRF analyzer were also used. Thus, in a first step, PMF was applied separately to the OA dataset (5 factors) and the Xact dataset (5 factors). In a second step, the results of these PMF analyzes were combined with the BC sources and the ions measured by the ToF-ACSM. This method allowed to identify 8 factors

The paper is of interest and deserves to be published in ACP although there are some aspects that can be improved. My main concern is about the application of the PMF to the data set of metal concentrations before the final PMF-PM1. The results obtained from the PMF metals are not entirely satisfactory. The diurnal pattern of the tire/brake factor is difficult to explain (despite frequent spikes). The so-called regional background factor, contains typical brake tracers, such as Cu and Sn, also Sb. This possible misidentification is reflected in Figure S14 and may affect the analysis of PMF-PM1. I understand that the PMF-OA can simplify the application of the final PMF-PM1. However, in my opinion, the prior grouping of the metals into 5 factors decreases the information for the PMF-PM1 and makes interpretation more difficult. Have you tested the PMF-PM1 run using BC sources, OA sources, ions and metal concentrations? If not, I suggest doing so and comparing the results between the two approaches.

Regarding the factors identified by PMF-PM1, the diurnal cycles identified for the Biomass Burning factor are not clear. Furthermore, considering that this source was not identified from the PMF-OA, the relatively high contribution obtained for this factor (5% of PM1) in July is surprising. As was done with the PMF-metals, it would be useful to perform NWR analysis for the final PMF-PM1 factors.

Finally, I think the conclusions section could be improved. The discussion can be extended on the advantages/disadvantages of the proposed method and the comparison with previous PMF analyzes carried out in the area.
Minor corrections

Line 12. Check first sentence of the Abstract

Line 15. Check verb person: OP measurement have…

Line 20: use Xact 625i (as in the main text) instead of Xact,

Lines 26-27: Sulfate and nitrate formed from SO2 and NOx also originated by combustion processes.

Line 29: replaced “is” by “was”

Line 62: Chen et al., 2021

Line 85: replace “AE33 data” by “Aethalometer (AE33) data”

Line 87: replace “aethalometer by “AE33”

Section 2.1. Please, can you clarify the duration of the sampling period for OP? 7 weeks or 15 days?

Line 227: I wonder about the selection of Br for PMF; this element provides little information as tracer of sources. I would exclude it from the PMF dataset and I would try 6 sources with XactPMF

Line 254; Replace “toxic lead metals” by “toxic lead forms” or “toxic lead compounds”

Line 280; section 2.4.3. Please, provide more information about BCff and BCwb estimation (AAE used…)

Line 358-362: This paragraph can be simplified.

Line 385-386: check sentence

Figure 4 (and Figure 5). These figures are difficult to understand as they are now. Improve figure legends. Explain the axes and legend in Figure 4a and 5a.

Line 399: Why is it limited to public construction? No private construction?

Line 474-476. SO42 also tracer of the shipping profile. This is the second factor explaining variation of BCwb after the BB.

Lines 486-490: may you explain better the differences/similarities with previous studies?

Citation: https://doi.org/10.5194/egusphere-2023-1441-RC1
- AC1:
  'Reply on RC1', Julie Camman, 06 Dec 2023
  We would like to thank the referees for their time to evaluate our manuscript and for their positive and constructive feedbacks, which helped improve the quality of the paper. Our responses to the comments are presented below.
  Minor revisions: All grammatical and cross-referencing errors in the text were corrected (listed below). Thank you very much to our referees.
  Line 12. Check first sentence of the Abstract
  
  Response: A part of the sentence had unfortunately been truncated. It has been corrected by the following sentence: « Source apportionment models were widely used to successfully assign highly-time resolved aerosol data to specific emissions and/or atmospheric chemical processes. ».
  Line 15. Check verb person: OP measurement have…
  
  Response: The sentence was corrected in the manuscript by the following sentence: « OP measurement has […] ».
  Line 20: use Xact 625i (as in the main text) instead of Xact,
  
  Response: Corrected.
  Lines 26-27: Sulfate and nitrate formed from SO2 and NOx also originated by combustion processes.
  
  Response: The sentence has been modified in the manuscript as following: « The results show that besides the high contribution of secondary ammonium sulfate (28%) and organic nitrate (19%), about 50% of PM₁ originated from distinct combustion sources, including emissions from traffic, shipping, industrial activities, cooking, and biomass burning. ».
  Line 29: replaced “is” by “was”
  
  Response: Corrected.
  Line 62: Chen et al., 2021
  
  Response: Chen et al., (2021) refers to an OA source apportionment study in Switzerland, whereas Chen et al., (2017) is quoted to support the potential health effects of ambient PM₁.
  Line 85: replace “AE33 data” by “Aethalometer (AE33) data”
  
  Response: Corrected.
  Line 87: replace “aethalometer by “AE33”
  
  Response: Corrected.
  Section 2.1. Please, can you clarify the duration of the sampling period for OP? 7 weeks or 15 days?
  
  Response: The duration of the sampling period for OP was clarified by the following sentence: « Finally, PM₁ collection for OP analysis was performed for 15 days (from 11^th July and 25^th July 2018) every 4 hours on 150 mm diameter quartz fibre filters (Whatman Tissuquartz) pre-heated at 500°C during 8 hours), using a high-volume aerosol sampler (HiVol, Digitel DA80) at a flow rate of 30 m³.h⁻¹ ».
  Line 227: I wonder about the selection of Br for PMF; this element provides little information as tracer of sources. I would exclude it from the PMF dataset and I would try 6 sources with XactPMF
  
  Response: We agreed with the reviewer that, in general, Br is not attributed to a specific source. However, due to its significant variability and concentrations with 99.8% of data points above the MDL, we deemed Br to be of interest for performing a PMF. Nevertheless, we tried to increase our solution to 6 factors and excluding the Br element (see Figure A1 in the supplement). The new resolved factor is interpreted as a split of the dust resuspension factor, dominated by Cu. Despite Cu often being associated with brake lining, this factor presented no correlation with traffic tracers (BC_FF, NO_x, HOA) or any related diurnal patterns. Consequently, this solution did not offer further information and was not retained in the study.
  Line 254; Replace “toxic lead metals” by “toxic lead forms” or “toxic lead compounds”
  
  Response: Corrected.
  Line 280; section 2.4.3. Please, provide more information about BCff and BCwb estimation (AAE used…)
  
  Response: We added the following details to the main text, lines (287 – 289): « BC_WB and BC_FF were deconvolved based on the model of Sandradewi et al., (2008). We used the 470 and 950 nm wavelengths with a constant absorption Angström exponent of 1.68 and 1.02 for pure wood burning and traffic, respectively, as recommended by Zotter et al., (2017) and Chazeau et al., (2021) ».
  Line 358-362: This paragraph can be simplified.
  
  Response: The paragraph has been simplified as follows: « Spearman coeficients (r_s) between PM₁ mass measured by FIDAS and OP display some differences (r_s PM₁ vs OP_v^AA = 0.23 (p<0.01) and r_s PM₁ vs OP_v^DTT = 0.63 (p<0.001)) where PM₁ is much more associated to OP_v^DTT than to OP_v^AA. These Spearman coefficients are close to those found by in ’t Veld et al., (2023) on PM₁ all year long in a similar urban coastal environment (Barcelona) (r_s PM₁ vs OP_v^AA = 0.29 (p<0.001) and r_s PM₁ vs OP_v^DTT = 0.73 (p<0.001)) ».
  Line 385-386: check sentence
  
  Response: The sentence was checked and rephrased accordingly: «The PMF_metals solution is investigated with the factor profiles and time series presented in Fig. 4, along with the factor relative diurnal cycles and contributions shown in Fig. S8. ».
  Figure 4 (and Figure 5). These figures are difficult to understand as they are now. Improve figure legends. Explain the axes and legend in Figure 4a and 5a.
  
  Response: The legends for Figure 4 and Figure 5 were improved as recommended.
  Line 399: Why is it limited to public construction? No private construction?
  
  Response: The sentence has been modified in the manuscript by the following sentence: « The construction work influence is supported by […]. ».
  Line 474-476. SO42 also tracer of the shipping profile. This is the second factor explaining variation of BCwb after the BB.
  
  Response: Indeed, we add the following sentences: « This factor further accounts for a noticeable variation of sulfate (11.6% of the total sulfate concentration). This is in agreement with the results from Chazeau et al., (2021) indicating that during 25% of the days in summer 2017, sulfate concentrations were prominently influenced by the nearby harbor. ».
  Lines 486-490: may you explain better the differences/similarities with previous studies?
  
  Response: The differences and similarities with previous studies were clarified as following: « In overall, the present PMF approach successfully identified various sources of PM₁ during the summer season, consistent with previous studies in Marseille. These sources include traffic (El Haddad et al., 2013; Bozzetti et al., 2017; Salameh et al., 2018), cooking (Bozzetti et al., 2017), and a minor contribution from biomass burning (Bozzetti et al., 2017; Salameh et al., 2018). However, this study marks the first identification of an ON-rich factor.
  Previous source apportionment of PM_2.5 markers by Salameh et al. (2018) highlighted the dominant contribution of ammonium sulfate in summer (35%) and identified a dust factor with a metal composition similar to the current study (Cu, Fe, Ca). While they identified a fossil fuel factor attributed to mixed harbor and industrial emissions, our results provide new insights by distinctly separating industrial and shipping emissions simultaneously advected onsite by sea breeze. ».
  -------------------------------------------------------
  Response to anonymous referee #1 comments :
  Referee comment: The paper is of interest and deserves to be published in ACP although there are some aspects that can be improved. My main concern is about the application of the PMF to the data set of metal concentrations before the final PMF-PM1. The results obtained from the PMF metals are not entirely satisfactory. The diurnal pattern of the tire/brake factor is difficult to explain (despite frequent spikes). The so-called regional background factor, contains typical brake tracers, such as Cu and Sn, also Sb. This possible misidentification is reflected in Figure S14 and may affect the analysis of PMF-PM1. I understand that the PMF-OA can simplify the application of the final PMF-PM1. However, in my opinion, the prior grouping of the metals into 5 factors decreases the information for the PMF-PM1 and makes interpretation more difficult. Have you tested the PMF-PM1 run using BC sources, OA sources, ions and metal concentrations? If not, I suggest doing so and comparing the results between the two approaches.
  Response: We agree with the reviewer that the interpretation of the PMF_metals results needed some clarifications. It is commonly understood that elements such as Cu, Sb, Sn, and Zn are typically assigned to brake lining and tire wear emissions. While it has already been pointed out for the coarse (PM_10-2.5) and intermediate (PM_2.5-1) fraction, it is less clear for the fine PM₁ fraction. Visser et al. (2015b) demonstrated that elements such as Cu, Sn, Sb are mainly found in the coarse mode at the “Marylebone Road” kerbside site. Hays et al. (2011) described similar trends for a near-highway site, with only Zn being significantly present in the fine mode. Additionally, these mentioned elements are not found to be correlated with each other. Such results suggest the existence of significant alternative source for these elements, potentially mixed here in a regional-scale background factor. Since most trace elements in the fine mode are non-volatile, they can undergo long-range atmospheric transport (Morawska and Zhang, 2002).
  This lack of identification was also described by (Visser et al., 2015a), who identified only a mixed traffic-related factor for the PM1-03 fraction, whereas a brake wear factor was resolved for the PM_2.5-1 and PM_10-2.5 fraction. In our study, the tire/brake wear factor is mainly constituted of Zn and Sb, which are among the most represented tracers for these sources in the fine mode. This statement was clarified into the manuscript (lines 429-433) : « Visser et al. (2015b) demonstrated that elements usually assigned to brake lining and tire wear emissions (e.g. Cu, Sb, Fe or Sn) are mainly found in the coarse mode at the "Marylebone road" kerbside site, and Hays et al. (2011) reported similar trends for a near-highway site in Raleigh, with Zn being the only element significantly present in the fine mode. Such results suggest the existence of significant alternative source for these elements, potentially mixed in the regional-scale background factor. » and lines 436-437: « Since most trace elements in the fine mode are non-volatile, they can undergo long-range atmospheric transport (Morawska and Zhang, 2002). »
  We also emphasize in the conclusion that incorporating the measurement of additional elements, such as Ba, S, Cl, and Si, could be an interesting feature to refine some sources. Regarding the PMF methodology, we followed the reviewer’s recommendation and we performed a PMF-PM1 using BC sources, OA sources, ions and metals concentrations. We added a dedicated paragraph on the comparison between the two approaches in the main text as follows (lines 521-533):
  « To assess the robustness of the PMF² solution, the results were compared to a PMF solution utilizing the OA factors from PMF_organics, ACSM inorganic species (SO₄^2-, NO₃^-, NH₄⁺, Cl^-), BC sources and metals concentrations as the input dataset. Consistent with the PMF² method, constrains, instrument weighting, criteria selection and bootstrap analysis were applied and are reported in the Supplement section. This alternative approach successfully identified the same 8 factors (Fig. S18, exhibiting comparable mass contributions and very high correlations with the PMF² factors time series (Table S5), all exceeding a R² of 0.9, except for shipping (R²=0.81).
  The biomass burning and shipping factors accounted for slightly higher concentrations in the PMF² solution, due to slightly elevated contribution of SO₄^2-, NH₄⁺ and MOOA concentrations which dominate the PM₁ mass. The metals composition found in the factors from this alternative PMF approach is in agreement with the metals profiles from the PMF_metals solution. Note that Zn and Sb, the most prominent elements in the tire/brake metals factor were mainly present in the traffic source. However they displayed again some mixing with other factors (dust resuspension, AS-rich and cooking), suggesting additional sources unresolved by the current PMF solutions. Previous studies suggested that Zn may originate from waste incineration or other industrial processes (Belis et al., 2019; Visser et al., 2015a; Manousakas et al., 2022). Comparable results in terms of explained variability were observed, emphasizing the suitability of both methods for such study. »
  We also provide further details in the supplements, with the Figure A2 (Figure S18 in the main text) and Table A1 (Table S5). Figure A2 and Table A1 are available in the supplement. Lastly, we further discuss the comparison between PMF methodologies in the response to reviewer #2.
  We added this following paragraph in the final version of the supplementary information (SI) :
  "Preparation of the PMF_PM1 with OA factors + metals + ions + BC dataset:
  Among the 8 identified factors, 4 were not systematically resolved across the several preliminary runs (cooking, biomass burning, industrial and shipping factors). The solution was constrained using base case profiles from the 10 factors-solution for industrial, the 11 factors-solution for cooking and shipping, and the 12 factors-solution for biomass burning. Note that for each run we applied the same C-values for the instrument weighting than PMF² solution. A bootstrap analysis was performed for 100 runs and the accepted runs based on the pre-defined list of criteria (the correlation with base case profiles for the constrained factors and the monitoring of the dominant variable intensity for the unconstrained factors) were averaged into a definitive solution."
  Referee comment: Regarding the factors identified by PMF-PM1, the diurnal cycles identified for the Biomass Burning factor are not clear. Furthermore, considering that this source was not identified from the PMF-OA, the relatively high contribution obtained for this factor (5% of PM1) in July is surprising. As was done with the PMF-metals, it would be useful to perform NWR analysis for the final PMF-PM1 factors.
  Response: As mentioned by the reviewer, unfortunately a BBOA was not resolved for the PMF-OA. The absence of local domestic heating emissions during this season has made its identification difficult. While El Haddad et al. (2013) didn’t resolve this factor with the c-ToF-AMS neither, Bozzetti et al. (2017) with the offline AMS technic and Salameh et al. (2018) using PM_2.5 offline markers identified a low biomass burning contribution in summer, with 5% of the total OA concentration and 2% of the PM_2.5 concentrations, respectively. In the current study, the PM₁ biomass burning factor was mainly constituted by MOOA (64% of the factor) and to a lesser extent by BC_WB (19%). Numerous studies identified a biomass burning factor for OA that exhibits characteristics of an oxidized OA profile, with enhanced signal at m/z 29, m/z 44 (Belis et al., 2019; Bougiatioti et al., 2014). In the PMF-OA, a portion of MOOA may account for a secondary biomass burning origin (e.g. wildfire, agricultural activities), as the main BBOA fingerprints, m/z 60 and m/z 73 were both predominantly attributed to this factor (40% and 39%, respectively).
  The influence of more secondary process for the biomass burning factor was stated lines 463–467 in the manuscript:
  «While no primary biomass burning organic aerosol (BBOA) factor was resolved with the PMForganics analysis in summer, the presence of a significant MOOA contribution reflects the influence of secondary process in this biomass burning factor. The low concentration of this factor is in agreement with minor regional emissions linked to agricultural activities, wildfires and cooking practices such as BBQ, transformed through oxidation processes during regional transport and aging (Chazeau et al., 2022; Cubison et al., 2011) ».
  Moreover, we followed the reviewer’s suggestion and the NWR analyses for the PMF-PM₁ factors are displayed in Figure A3 (Figure S17 in the main text), available in the supplement. The following lines were modified accordingly:
  -lines 458 – 460 : « The full PM₁ source apportionment solution is explored in this section with the average factor profiles (Fig. 5a), the time series (Fig. 5b), the pie chart of mass contributions (Fig. 5c), the average diurnal profiles (Fig. 5d) and the NWR analyses (Fig. S17). »
  -lines 467 – 470 : «The NWR analysis in Fig. S17 showed biomass burning concentrations associated with higher wind speed than sources with a local origin (traffic, shipping, cooking and ON-rich), corresponding to south-westerly winds from the Mediterranean Sea. Additionally, the north-east land breeze advected these aged emissions back to the sampling site. »
  -line 505 : « […] This factor displayed an origin from the North to East within the land. »
  -line 519 : « […] This interpretation is supported by the NWR analysis presented in Fig. S17. »
  Referee comment: Finally, I think the conclusions section could be improved. The discussion can be extended on the advantages/disadvantages of the proposed method and the comparison with previous PMF analyzes carried out in the area.
  Response: We thank the reviewer for the suggestion. We added the following discussion to the conclusions in lines 668-676: « The PMF² approach successfully identified 8 well-resolved sources (AS-rich, traffic, ON-rich, cooking, shipping, biomass burning, industrial and dust resuspension), a solution not achievable through single PMFs conducted separately on OA and metals datasets. The method enabled the assignment of OA factors, which typically described components arising from a mixture of sources and chemical processes rather than a single emission source, to more specific PM₁ sources. Additionally, this approach allowed to assess both the primary and secondary origin of anthropogenic sources, such as traffic and cooking. However, a limitation of this method is that non-explained variability and uncertainties of the factors from the first step PMFs will propagate into the PMF² results and therefore need to be carefully assess. The inclusion of additional elements measurements, such as Ba, S, Cl, and Si to the PMF_metals, could be an interesting feature to refine some sources and address this limitation. »
  A comparison with previous PMF analyzes carried out in the area was also detailed in the section 3.2.3 (lines 536-543): « In overall, the present PMF approach successfully identified various sources of PM1 during the summer season, consistent with previous studies in Marseille. These sources include traffic (El Haddad et al., 2013; Bozzetti et al., 2017a; Salameh et al., 2018), cooking (Bozzetti et al., 2017a), and a minor contribution from biomass burning (Bozzetti et al., 2017a; Salameh et al., 2018). However, this study marks the first identification of an ON-rich factor. Previous source apportionment of PM_2.5markers by Salameh et al. (2018) highlighted the dominant contribution of ammonium sulfate in summer (35%) and identified a dust factor with a metal composition similar to the current study (Cu, Fe, Ca). While they identified a fossil fuel factor attributed to mixed harbor and industrial emissions, our results provide new insights by distinctly separating industrial and shipping emissions simultaneously advected onsite by sea breeze. ».
  
  Citation: https://doi.org/10.5194/egusphere-2023-1441-AC1
RC2:
'Comment on egusphere-2023-1441', Anonymous Referee #2, 08 Sep 2023

This work aims at studying the oxdative potential (OP) of the submicron aerosol sources affecting the air quality in Marseille during summer, and thus providing information about the relative toxicity of the different PM¹ sources. This type of studies provides important information to the decision makers to design more effective mitigation strategies targeted at decreasing the PM toxicity rather than its concentration. For this purpose the authors collected a rich dataset consisting of ACSM data (OA mass spectra and ions), Aethalometer data (BC_traffic/BC_{wood burning}), Xact XRF data (metals) with high time resolution (≤ 1h). The dataset is completed by OP_AA and OP_TTC measurements conducted on filter samples with a 4h time resolution. The authors adopted a PMF² approach to perform a source apportionment of the PM₁ aerosol fraction. This approach consisted in performing two initial PMF analyses respectively on OA (ACSM data) and on metals (Xact data). The outputs of these two PMFs (time series of OA and metal aerosol sources) have been used as inputs for the PM₁ source apportionment together with BC (aethalometer) and ACSM ions time series. In order to determine the the OP of each resolved PM source, a multilinear regression of OP time series has been performed. Overall the topic is of interest and deserves a publication on ACP after considering a couple of aspects which should be discussed more in details. More specifically, the interpretation of the oxidative potential of the resolved PM₁ sources deserve a deeper investigation, and the authors should justify the adoption of a PMF² approach for the source apportionment.
Regarding the interpretation of the OP results:
The OP_AA and OP_DTT time series displayed in figure 3a and 6a seems to show a weak correlation. This is also reflected by the OP multilinear regression results displayed in figure 7 and table 1, where some PMF factors e.g. dust resuspension and industrial emissions are positively correlated with OP_AA and negatively correlated or not correlated with OP_DTT. Vice-versa the shipping emission factor and the AS-rich factor are positively correlated with OP_DTT and negatively correlated or not correlated with OP_AA. I would discuss more in depth the differences of the results obtained from OP_AA and OP_DTT. What's the physiological representativeness of OP_AA and OP_DTT? Which one is more relevant for human health? Both DTT and AA are reductant substances. It seems that AA is more sensitive to Cu and other elemental impurities, while DTT is sensitive to other oxidative species, therefore it seems that OP_DTT and OP_AAare related to different oxidative pathways. Could you briefly elaborate on the physiological representativeness of these two pathways and which one is more relevant and the specific relevance of each one? Do OP_AA and OP_DTT provide complementary information or one is more representative than the other of the real oxidation processes occurring in-vivo? If OP_DTT and OP_AA provide complementary information, do the authors suggest to always perform both the analyses? Moreover, from the results displayed in figure 7, where the sources are ranked by their contribution to OP_DTT, OP_AA and PM₁, it seems that there's a certain correlation between the sources contributions to PM₁ and OP_DTT. On the opposite, such a correlation is completely missing between sources contribution to OP_AA and PM₁ mass. Does it suggest that AA is more sensitive to the chemical composition of the sources, while DTT is more sensitive to the aerosol concentration, and therefore less representative of the real oxidative potential of an aerosol source? Without a critical discussion on these aspects, the results displayed in figure 7 might lead to contradictory conclusions, for example the industrial factor can be considered as toxic or non toxic if looking respectively at OP_AA or OP_DTT results.
Regarding the source apportionment strategy:
I suggest the authors to justify the adoption of the PMF² approach. This approach utilized the outputs of the OA and metals source apportionments as input for a comprehensive PM₁ source apportionment. Alternatively a unique PMF analysis could have been performed using the ACSM and Xact raw data as direct inputs for the PM₁ source apportionment. The PMF² approach has two drawbacks. Firstly, the uncertainties of the first PMF analyses and their unexplained variability are propagated into the PM¹ source apportionment. Secondly, the OA and metal aerosol sources, which had been already resolved by the first PMFs (on ACSM and Xact data), are then reapportioned and potantially re-mixed into different PM¹ factors. This is observed in figure 5a for the traffic and cooking factors, where a non-negligible contribution from LOOA and MOOA is observed. Similarly, the brake/tire factor resolved by the metals' PMF, is splitted into 4 PM¹ factors (traffic, AS-rich, industrial, and biomass burning). This suggests that either the PM₁ source apportionment hasn't been fully-optimized, or the input factor time series were already not well resolved from other sources, and therefore the error of the OA and metal PMFs have been propagated into the final PM₁ source apportionment. Instead, using the OA ACSM raw data as input for the PM₁ source apportionment might help resolving a better traffic profile, because the OA ACSM mass spectra contain many hydrocarbon fragments which are typically related to traffic exhaust.
Minor comments:
In the references, 3 publications from Bozzetti et al., are cited. All of them from 2017. In the text is not clear which one is referenced and when because they are all identified as Bozzetti et al., 2017.
Plot 3b: missing x-axis
Line 423: missing year of publication of Salameh et al.
Line 453-454: the sentence stating that the industrial contribution to PM1 found in Marseille is comparable to the contribution observed in other cities might lead to misleading conclusions. The industrial contribution being similar and low in different cities could be merely casual or due to the distance of the sampling stations from the emission spots, and on type of the industrial processes involved. I think this similar and low industrial contribution to PM1 among different cities only demonstrates that the urban background stations are typically scarcely affected by industrial emissions because of their geographical location.
Line 566: missing reference.

Citation: https://doi.org/10.5194/egusphere-2023-1441-RC2
- AC2:
  'Reply on RC2', Julie Camman, 06 Dec 2023
  We would like to thank the referees for their time to evaluate our manuscript and for their positive and constructive feedbacks, which helped improve the quality of the paper. Our responses to the comments are presented below.
  Minor revisions: All grammatical and cross-referencing errors in the text were corrected (listed below). Thank you very much to our referees.
  In the references, 3 publications from Bozzetti et al., are cited. All of them from 2017. In the text is not clear which one is referenced and when because they are all identified as Bozzetti et al., 2017.
  
  Response: There was a duplicate among the three publications. The two « Bozzetti et al. (2017) » publications have been differentiated by the addition of (a) and (b).
  Plot 3b: missing x-axis
  
  Response: The x-axis has been added on the Figure 3b.
  Line 423: missing year of publication of Salameh et al.
  
  Response: It has been corrected by: « Salameh et al. (2018) »
  Line 453-454: the sentence stating that the industrial contribution to PM1 found in Marseille is comparable to the contribution observed in other cities might lead to misleading conclusions. The industrial contribution being similar and low in different cities could be merely casual or due to the distance of the sampling stations from the emission spots, and on type of the industrial processes involved. I think this similar and low industrial contribution to PM1 among different cities only demonstrates that the urban background stations are typically scarcely affected by industrial emissions because of their geographical location.
  
  Response: This statement, as written, is indeed unclear and prone to misinterpretation, thanks for noticing it. We agreed the distance between the industrial area and the urban site may account for the observed very low concentrations in comparison to more local sources. However, it is important to note that industrial plumes are transported by sea breeze conditions, which prevail almost daily in the Marseille area during the summer.
  
  The low PM₁ mass concentration for this source is expected as the size of the industrial particles generally belongs to the ultrafine mode (<100nm) (Riffault et al., 2015). Chazeau et al. (2021) and El Haddad et al. (2013) already described that plumes originated from the main industrial area of Fos-Berre are mainly attributed to ultrafine particles and thus influence the mass only to a minor extent. This clarification is now articulated in the main text as follows, lines 478-483: « The factor contributes little to the PM₁ composition (3.2%), which is expected as the size of the industrial particles generally belong to the ultrafine mode (<100nm) (Riffault et al., 2015). Chazeau et al., (2021) and El Haddad et al., (2013) already described that plumes originated from the main industrial area of Fos-Berre are advected onsite by sea breeze conditions and are mainly attributed to ultrafine particles, influencing the mass concentrations only to a minor extent. Similar contributions were found in another Mediterranean coastal city, Barcelona (4%; Via et al. (2023)), and in some French urban sites in the vicinity of an industrial area (Weber et al. 2019)».
  Line 566: missing reference.
  
  Response: The reference “Weber et al. (2021)” has been added.
  ---------------------------------------------------------
  Response to anonymous referee #2 comments :
  Referee comment : This work aims at studying the oxdative potential (OP) of the submicron aerosol sources affecting the air quality in Marseille during summer, and thus providing information about the relative toxicity of the different PM¹ sources. This type of studies provides important information to the decision makers to design more effective mitigation strategies targeted at decreasing the PM toxicity rather than its concentration. For this purpose the authors collected a rich dataset consisting of ACSM data (OA mass spectra and ions), Aethalometer data (BC_traffic/BC_{wood burning}), Xact XRF data (metals) with high time resolution (≤ 1h). The dataset is completed by OP_AA and OP_TTC measurements conducted on filter samples with a 4h time resolution. The authors adopted a PMF² approach to perform a source apportionment of the PM₁ aerosol fraction. This approach consisted in performing two initial PMF analyses respectively on OA (ACSM data) and on metals (Xact data). The outputs of these two PMFs (time series of OA and metal aerosol sources) have been used as inputs for the PM₁ source apportionment together with BC (aethalometer) and ACSM ions time series. In order to determine the the OP of each resolved PM source, a multilinear regression of OP time series has been performed. Overall the topic is of interest and deserves a publication on ACP after considering a couple of aspects which should be discussed more in details. More specifically, the interpretation of the oxidative potential of the resolved PM₁ sources deserve a deeper investigation, and the authors should justify the adoption of a PMF² approach for the source apportionment.
  Regarding the interpretation of the OP results:
  The OP_AA and OP_DTT time series displayed in figure 3a and 6a seems to show a weak correlation. This is also reflected by the OP multilinear regression results displayed in figure 7 and table 1, where some PMF factors e.g. dust resuspension and industrial emissions are positively correlated with OP_AA and negatively correlated or not correlated with OP_DTT. Vice-versa the shipping emission factor and the AS-rich factor are positively correlated with OP_DTT and negatively correlated or not correlated with OP_AA. I would discuss more in depth the differences of the results obtained from OP_AA and OP_DTT. What's the physiological representativeness of OP_AA and OP_DTT? Which one is more relevant for human health? Both DTT and AA are reductant substances. It seems that AA is more sensitive to Cu and other elemental impurities, while DTT is sensitive to other oxidative species, therefore it seems that OP_DTT and OP_AAare related to different oxidative pathways. Could you briefly elaborate on the physiological representativeness of these two pathways and which one is more relevant and the specific relevance of each one? Do OP_AA and OP_DTT provide complementary information or one is more representative than the other of the real oxidation processes occurring in-vivo? If OP_DTT and OP_AA provide complementary information, do the authors suggest to always perform both the analyses?
  Response : Thank you for the feedback, which has enabled us to provide further details on the interpretation of the OP results and add a section “3.3.4. Discussion” to the manuscript. Indeed, the association between OP^AA and OP^DTTis moderate to weak (r_s = 0.41; p<0.001) and illustrates the different sensitivity of the two OP tests to chemical constituents found in ambient air. As shown in the Figure 7, the contribution of the PM sources to OP is dependent on the OP test used. This reflects the different oxidation pathways involved in the oxidation of the two probes (AA and DTT). A detailed answer for the physiological representativeness of OP_AA and OP_DTT has been added in the manuscript, in the section “3.3.4 Discussion”, lines (632 – 639):
  “AA is naturally present in the lungs, and its predominant anionic form in solution (HA^-) is oxidised by various mechanisms facilitated by OH^•, O₂^•-, HO₂^• and other radicals, and by transitions metals as Cu (II) or Fe (III) (Campbell et al., 2019). DTT has a disulfide bond and is considered as a chemical substitute for cellular reducing agents such as nicotinamide adenine dinucleotide phosphate oxidase (NADPH) or protein thiols (Verma et al., 2015; Borlaza et al., 2018). Protein thiols play an important role in major oxidative stress, restoring the redox balance by eliminating free radicals (Baba and Bhatnagar, 2018). Many studies have linked these two probes (AA and DTT) to transition metals (Cu, Fe, Mn, Zn), EC and OC (Gao et al., 2020). In addition, the different sensitivity of AA and DTT to both organic compounds and transition metals has been evidenced in Calas et al., 2018, Gao et al., 2020 and Pietrogrande et al., (2022).
  Although no consensus has been reached on an OP test that is more representative of health impact, epidemiological studies have mainly associated OP^DTT with health endpoints, which has not been demonstrated with OP^AA. The community currently recommends the complementary use of these two tests. A detailed response has been added in the manuscript, in the section “3.3.4 Discussion”, lines (640 – 650):
  “Today, no consensus has yet been reached on which OP test is most representative of health impact, and the community still recommends the complementary use of OP tests, in particular the association of both AA and thiol-based (DTT or GSH) assays (Moufarrej et al., 2020). This association is today the unique way of assessing the full panel of the most oxidising compounds of PM. However, recent studies have shown positive associations between OP^DTT and various acute cardiac (myocardial infarction) and respiratory endpoints, supporting the interest of the OP^DTT assay for this purpose (Abrams et al., 2017; Weichenthal et al., 2016; He and Zhang, 2023). On the contrary, several studies did not associate OP^AAto health endpoints including early-life outcomes, respiratory and cardiovascular mortality, cardiorespiratory emergencies and lung cancer mortality (Borlaza et al., 2023; Marsal et al., 2023). Nonetheless, a recent study has associated OP^AA with oxidative damage to DNA (Marsal et al., 2023). These results so far may suggest that OP^AA provides partial information on the link between OP and adverse health effects, and further epidemiological studies are needed to determine whether OP^AA should be considered as a proxy for health impact.”
  Referee comment : Moreover, from the results displayed in figure 7, where the sources are ranked by their contribution to OP_DTT, OP_AA and PM₁, it seems that there's a certain correlation between the sources contributions to PM₁ and OP_DTT. On the opposite, such a correlation is completely missing between sources contribution to OP_AA and PM₁ mass. Does it suggest that AA is more sensitive to the chemical composition of the sources, while DTT is more sensitive to the aerosol concentration, and therefore less representative of the real oxidative potential of an aerosol source? Without a critical discussion on these aspects, the results displayed in figure 7 might lead to contradictory conclusions, for example the industrial factor can be considered as toxic or non toxic if looking respectively at OP_AA or OP_DTT results.
  Response : Thank you for your pertinent comment. Indeed, as in many other studies referenced in section “3.1 “OP results” lines (371-372), PM₁ is more associated with OP^DTT than with OP^AA (r_s PM₁ vs OP_v^AA = 0.23 (p<0.01) and r_s PM₁vs OP_v^DTT = 0.63 (p<0.001)). Associations values were mentioned in the manuscript in lines (366-367), but these values have not been discussed in depth. These values reflected the sensitivity of DTT to a wider range of chemical compounds, implying a stronger association with aerosol concentration, whereas AA displays a heightened sensitivity to chemical composition (which exhibit robust specificity). Indeed, OP_v^AA is known to be more sensitive to some PM components as Cu(II) or Fe(II) but also some quinones (Calas et al., 2019, Campbell et al., 2019; Pietrogrande et al., 2022). In addition, the state-of-the-art highlighted PM concentration as a significant predictor of OP_v^DTT in univariate models (Janssen et al., 2014; Weber et al., 2018). For the moment, we need to keep a critical eye on the results since a multitude of sources have been identified by the two OP tests and therefore deserve to be considered. A detailed response explaining the observed correlation coefficients has been added in the manuscript, section 3.1 “OP results”, lines (366 – 373):
  “Spearman coeficients (r_s) between PM₁ mass measured by FIDAS and OP display some differences (r_s PM₁ vs OP_v^AA = 0.23 (p<0.01) and r_s PM₁ vs OP_v^DTT = 0.63 (p<0.001)) where PM₁ is much more associated to OP_v^DTT than to OP_v^AA. These Spearman coefficients are close to those found by in ’t Veld et al., (2023) on PM₁ all year long in a similar urban coastal environment (Barcelona). The higher association between OP_v^DTT and PM₁ compared to OP_v^AA and PM₁ has already been observed in other studies conducted on PM₁₀(Calas et al., 2019; Weber et al., 2021; Janssen et al., 2014).This phenomenon is attributed to AA's heightened sensitivity to chemical composition, exhibiting robust specificity. Moreover, DTT demonstrates superior sensitivity to aerosol concentration owing to its more balanced sensitivities to chemical constituents (Gao et al., 2020).”
  Referee comment : Regarding the source apportionment strategy:
  I suggest the authors to justify the adoption of the PMF² approach. This approach utilized the outputs of the OA and metals source apportionments as input for a comprehensive PM₁ source apportionment. Alternatively a unique PMF analysis could have been performed using the ACSM and Xact raw data as direct inputs for the PM₁ source apportionment. The PMF² approach has two drawbacks. Firstly, the uncertainties of the first PMF analyses and their unexplained variability are propagated into the PM¹ source apportionment. Secondly, the OA and metal aerosol sources, which had been already resolved by the first PMFs (on ACSM and Xact data), are then reapportioned and potantially re-mixed into different PM¹ factors. This is observed in figure 5a for the traffic and cooking factors, where a non-negligible contribution from LOOA and MOOA is observed. Similarly, the brake/tire factor resolved by the metals' PMF, is splitted into 4 PM¹ factors (traffic, AS-rich, industrial, and biomass burning). This suggests that either the PM₁ source apportionment hasn't been fully-optimized, or the input factor time series were already not well resolved from other sources, and therefore the error of the OA and metal PMFs have been propagated into the final PM₁ source apportionment. Instead, using the OA ACSM raw data as input for the PM₁ source apportionment might help resolving a better traffic profile, because the OA ACSM mass spectra contain many hydrocarbon fragments which are typically related to traffic exhaust.
  Response : We thank referee #2 for the insightful comments. One of the objectives of the current study was to assess the PM sources contributions to OP through three scenarios: first, an OP apportionment using only OA factors from the PMF_organics; second, an OP apportionment using only metals factors from the PMF_metals; and third, to follow an harmonized methodology, we explored the third scenario (OP apportionment using PM₁ factors) by combining together the factors from both PMF_organics and PMF_metals as inputs for the PMF_PM1. The PMF² approach was the most suitable method for this purpose, emphasizing the importance of considering all PM₁ fractions to apportion OP.
  Moreover, using OA PMF factors as inputs allows to quantify the primary/secondary OA contribution to the PM₁ sources. A limitation of performing PMF on OA mass spectra from ACSM/AMS is the resolution of the SOA origin. SOA factors are usually reported as either a single factor or two factors separated by their degree of oxygenation (LOOA/MOOA) rather than in terms of sources. The PMF² approach enables a more accurate identification of SOA sources, addressing this limitation. Given that several studies highlighted the role of SOA in oxidative potential, it is important to include an accurate quantification of this fraction in the PM₁ sources, a step not achievable using the raw OA mass spectra. We added to the Introduction section the justification of using PMF² method (lines 87-92): « A known drawback of performing PMF on OA mass spectra from ACSM/AMS is the resolution of the secondary organic aerosol (SOA) origin. SOA factors are usually reported as either a single factor or two factors separated by their degree of oxygenation rather than in terms of sources. A PMF² approach using previous OA factors combined with other species and/or PMF factors may enable a more accurate identification and quantification of the SOA fraction in the PM sources. The current study addresses this challenge by intending the PMF² method for the PM₁ fraction measured with online analysers (i.e. ToF-ACSM, Xact 625i and AE33) at high time resolution (<1h). »
  We fully agree with the reviewer about the first drawback. Since we performed bootstraps for the two first PMFs we were able to statistically estimate uncertainties of the factors. These uncertainties are incorporated in the error inputs for the PMF_PM1 analysis. It is true also that the non-explained variability of the first PMFs is propagated into the PM₁ source apportionment, representing a notable inconvenience of a multi-step PMF approach. This is now explicitly stated as a limit of the methodology in the conclusion (lines 673-675): « However, a limitation of this method is that non-explained variability and uncertainties of the factors from the first step PMFs will propagate into the PMF² results and therefore need to be carefully assessed. »
  HOA and COA were constrained in the PMF_organics using reference profiles accounting for primary traffic and cooking emissions in an urban environment. Therefore, the contribution of the fast oxidation of freshly emitted primary OA is expected to be included in the SOA factors, as demonstrated in Chazeau et al. (2022). This explains why some LOOA and MOOA fractions are attributed to the traffic and cooking sources. The SOA contribution to the traffic source was previously mentioned lines 491-493: « It should be emphasized that 23% of the traffic source was constituted of SOA (LOOA and MOOA) meaning that primary traffic contribution is mixed with secondary aerosol concentrations attributed to fast oxidation of freshly emitted particles (Chirico et al., 2011). ».
  The tire/brake factor displayed the highest unexplained variation, probably due to some mixing with other sources as suggested by the reviewer. It was previously noted in lines 490-491 and is now further discussed in the new paragraph comparing the two PMF² approach.
  We agree with the reviewer that including OA raw data as PMF inputs is a very interesting method to explore the PM₁ sources as it was already performed by Belis et al. (2019). There are many possibilities in combining datasets, whether in their raw format or as PMF factors, that would need further investigation to establish a more standardized protocol for PM₁ source apportionment. Despite this statement, the scope of the present manuscript is not to inter-compare alternative PMF methodologies, which could be the focus of a fully dedicated paper.
  However, we inspected correlations between some hydrocarbon OA fragments related to traffic exhaust and the other PM₁ compounds (metals, BC, SO4, NO3, NH4 and Cl) in Figure A4, available in the supplement. The results did not show any better correlation than comparing with the HOA factor. For these reasons, we do not think it would be appropriate to present results from a PMF analysis based on OA fragments + inorganic compounds.
  Nevertheless, we performed PMF on OA factors and metals, as suggested by the referee #1, and compared it to the PMF² approach to support our assessment. The results were relatively similar and are detailed in lines 522- 535 in the section 3.2.3.
  
  Citation: https://doi.org/10.5194/egusphere-2023-1441-AC2
- AC3: 'Reply on RC2', Julie Camman, 06 Dec 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-1441/egusphere-2023-1441-AC3-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-1441-AC3

Julie Camman et al.

Supplement

https://doi.org/10.5194/egusphere-2023-1441-supplement

Julie Camman et al.

Viewed

Total article views: 503 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
339	142	22	503	38	7	7

HTML: 339
PDF: 142
XML: 22
Total: 503
Supplement: 38
BibTeX: 7
EndNote: 7

Views and downloads (calculated since 28 Jul 2023)

Month	HTML	PDF	XML	Total
Jul 2023	121	44	5	170
Aug 2023	92	38	3	133
Sep 2023	31	21	4	56
Oct 2023	47	21	4	72
Nov 2023	22	8	0	30
Dec 2023	26	10	6	42

Cumulative views and downloads (calculated since 28 Jul 2023)

Month	HTML	PDF	XML	Total
Jul 2023	121	44	5	170
Aug 2023	92	38	3	133
Sep 2023	31	21	4	56
Oct 2023	47	21	4	72
Nov 2023	22	8	0	30
Dec 2023	26	10	6	42

Viewed (geographical distribution)

Total article views: 495 (including HTML, PDF, and XML) Thereof 495 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 10 Dec 2023

Short summary

Fine particle pollution is a major health issue in the city of Marseille which is subject to numerous pollution sources. Sampling carried out during the summer enabled a fine characterization of the PM₁ sources and their oxidative potential, a promising new metric as a proxy for health impact. PM₁ came mainly from combustion sources, secondary ammonium sulfate and organic nitrate, while oxidative potential of PM₁ came from these sources and also from resuspended dust in the atmosphere.


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

Oxidative potential apportionment of atmospheric PM1: A new approach combining high-sensitive online analysers for chemical composition and offline OP measurement technique

Supplement

Viewed

Viewed (geographical distribution)

Oxidative potential apportionment of atmospheric PM₁: A new approach combining high-sensitive online analysers for chemical composition and offline OP measurement technique