In this work, the authors examine the 2007 – 2020 trends in PM2.5 and its composition in the Pearl River Delta area of China.  This time period saw dramatic decreases in PM2.5 concentrations and changes in the PM2.5 composition driven by successful regulatory actions.  These air quality and associated emission trends have been discussed by others; however, in this work the authors explore the causes of the trends including changes in the oxidation rates of SO₂ and NO₂ to sulfate and nitrate respectively.  I think this is an important contribution to our understanding of the aerosol in this region and the underlying causes of their trends.  I also found no major technical issues in the work and recommend publication after the authors address a number of minor comments.

Specific comments

• Line 40, properly define PM2.5
• Line 49, define ALWC
• Lines 74 – 83. The IMPROVE equation was developed by the US National Park Service with some support from the EPA. There are two IMPROVE equations. The first, IMPROVE equation 1 (EPA, 2003), was based on the work in Malm et al., (1994).  This was replaced in 2007 with the IMPROVE equation 2 (Pitchford et al., 2007) based on the work of Malm et al., (2007) and Hand et al., (2007).  The IMPROVE equation 1 uses constant scattering efficiencies base on fixed size distributions for the different aerosol components.  In IMPROVE equation 2 the scattering efficiencies are dependent on the aerosol concentrations.  Specifically, the scattering efficiencies are a weighted average of the scattering efficiency derived from a small and large size distribution and the weights are proportional to the aerosol concentration (Pitchford et al., 2007). Which IMPROVE equation is used in this work is not discussed and needs to be clarified.
• Line 77-79, “The hygroscopic growth factor (f(RH)), which has been suggested to depend on secondary inorganic fractions (e.g., sulfate, nitrate, and ammonium), sea salt components, and water-soluble organic carbon, is solely a function of relative humidity (RH) in the algorithm” If IMPROVE equation 2 is being used then this is an incorrect statement.
• Line 85, “IMPROVE program in the United States, initiated in 1985, tracks visibility trends and their driving factors (Epa, 2011)”. The EPA reference is not in the reference list. There are many journal article reporting on the purpose of the IMPROVE program and use of the data and should be used as the reference instead of an EPA report.  For example, see Hand et al., (2024) and references there in.
• Line 143, “anions (.e. CL-, NO3- and SO42-) were analyzed with an ion-chromatography system…” The authors should note that some ammonium nitrate volatilizes from the quartz fiber filters during sampling and handling causing underestimations in NH4+ and NO3- concentrations (Yu et al., 2006).  In addition, if possible, provide an estimate of the underestimation, which should be carried over into the discussion of particulate nitrate concentrations.
• Measuring long-term trends is very challenging, and seemingly, small changes in sampling and analysis protocols can introduce discontinuities in the PM trends. This is particularly true for thermal optical carbon analyses, since the measured OC and EC are operationally defined and sensitive to changes in the method. The authors should note any changes in the monitoring protocols over the 14-year time span and discuss any evidence for or against these changes introducing discontinuities in the trends.
• Line 162, “Bayesian Inference Approach and suggested it had significant advantages in accurately estimating POC and SOC”.  All of methods to estimate POC and SOC from OC and EC data are highly uncertain.  This should be conveyed in the paper.  For example, instead of saying “suggested it had significant advantages in accurately estimated...”  could use “suggested it more accurately estimated…”
• The annual bar chart in Figure 2 provides the change in the absolute concentrations overtime. It is difficult to see the trends in the changing PM2.5 composition.  I suggest the authors include a graph similar to Figure 2a of the annual relative contributions of aerosol components to PM2.5 in the main document or supplemental information.
• In Figure 2, could some indication of the SOC and POC concentrations be included?
• Line 228, “Despite the slight decline in their concentrations, our result showed that the relative reductions of EC (-9% yr-1)…” This is confusing.  What is meant by “slight decline”?  A 9% decrease per year is not slight.
• Line 244, “When NO2 levels are low, the accumulation of nitrate is hindered due to volatilization losses.” This is a confusing sentence.  What are the volatilization losses?  Those from the filter?  Also, ammonium nitrate volatilization from the filter is not really dependent on NO2 levels, but is dependent on temperature and relative humidity.
• Line 259, “As shown in Fig 5a, a dramatic increase in SOR was observed during 2007–2020 (p < 0.05). The SOR value in 2020 (0.24 ± 0.09) was twice as high as that in 2008 (0.12 ± 0.07)”. Comparing 2008 to 2020 is a bit of cherry picking.  There is a lot of variability in the data and 2007 SOR are only 30-40% smaller than 2020.  I suggest a robust trend line, e.g. use Theil regression, is calculated from the data, and then use the slope of the trend line to estimate the change over time.
• Line 280, “This meant that the conversion of NO2 to NO3- became weaker, resulting in a greater reduction in NO3- compared to NO2.” This is not obvious to me.  How do you know that the difference in the trends is not driven by changes in the partitioning of NO3 between the gas and particle phase?
• Line 319, “Aerosol pH increased from 1.51 ± 1.07 to 3.29 ± 1.43, at a rate of 0.06 yr–1 (p < 0.05).” In figure 6a, the largest pH is about 2.7 not 3.29.  Also, I do not see a trend in these data.  Similar to the suggestion for figure 5, it would be best to fit a robust trend line through the data and use its slope to define the change over time.
• Line 364, which IMPROVE equation is being used to estimate the light extinction?
• Line 367, “Our results indicated that the IMPROVE equation tend to overestimate bext in elevated pollution periods.” The authors did not provide any evidence that one bext equation was better than the other was, so they should only say that the IMPROVE equation had higher bext than that estimated from equations 5-8 for high PM2.5 concentrations.  This is important, because many studies have shown that scattering efficiencies of secondary aerosols are correlated with concentrations.  It was this observation that drove the development of the IMPROVE equation 2.  Therefore, it is quite possible that the fixed scattering efficiencies in equation 5 would cause an underestimation of the bext.

References

EPA, 2003. Guidance for Tracking Progress Under the Regional Haze Rule.  https://www.epa.gov/visibility/guidance-tracking-progress-under-regional-haze-rule

Hand, J.L., Malm, W.C., 2007. Review of aerosol mass scattering efficiencies from ground-based measurements since 1990. J. Geophys. Res. 112, D16203.

Hand, J. L., A. J. Prenni and B. A. Schichtel (2024). "Trends in Seasonal Mean Speciated Aerosol Composition in Remote Areas of the United States From 2000 Through 2021." Journal of Geophysical Research-Atmospheres 129(2): 22.

Malm, W.C., Sisler J.F., Huffman D., Eldred, R.A., and Cahill, T.C., Spatial and seasonal trends in particle concentration and optical extinction in the U.S. J. Geophys. Res. 99(D1):1347- 1370, 1994

Malm, W.C.; Hand, J.L. An Examination of the Physical and Optical Properties of Aerosols Collected in the IMPROVE Program; Atmos. Environ., 2007, 41, 3404-3427.

Pitchford, M., W. Malm, B. Schichtel, N. Kumar, D. Lowenthal and J. Hand (2007). "Revised algorithm for estimating light extinction from IMPROVE particle speciation data." Journal of the Air & Waste Management Association 57(11): 1326-1336.

Yu, X. Y., T. Lee, B. Ayres, S. M. Kreidenweis, W. Malm and J. L. Collett (2006). "Loss of fine particle ammonium from denuded nylon filters." Atmospheric Environment 40(25): 4797-4807.

This work presents long-term data on PM2.5 chemical composition for a regionally representative site in the Pearl River Delta region of China. The authors report decreases in PM2.5 and its chemical constituents. They find an increase in secondary species. They add an analysis investigating the sulfur and nitrogen oxidation rate and find that sulfur oxidation rates increase while nitrogen oxidation rates decrease. Other physicochemical properties are assessed.

There are many studies already published that are similar to the one presented here, and there is not much new information. The novel part of this study is the investigation into sulfur and nitrogen oxidation rates, and for that reason, I think this paper is of value to add to the literature. I have some concerns that should be addressed before I can recommend it for publication.

Major

There is a lack of placing into context the findings in this study to other prior studies that have found similar phenomena. Specifically, I noted that the discussion of the impact of aerosol water on the organic aerosol species lacks many references to other studies that initially identified aerosol water as a major influencing factor on organic aerosol concentrations, some of which are from over 10 years ago. This is especially true in the paragraph starting in Line 302 and Section 3.3. A few papers that should be noted include:

Attwood, A. R., et al. (2014), Trends in sulfate and organic aerosol mass in the Southeast U.S.: Impact on aerosol optical depth and radiative forcing, Geophys. Res. Lett., 41, 7701–7709, doi:10.1002/2014GL061669.

Carlton, A. G. and Turpin, B. J.: Particle partitioning potential of organic compounds is highest in the Eastern US and driven by anthropogenic water, Atmos. Chem. Phys., 13, 10203–10214, https://doi.org/10.5194/acp-13-10203-2013, 2013.

Ervens, B., Turpin, B. J., and Weber, R. J.: Secondary organic aerosol formation in cloud droplets and aqueous particles (aqSOA): a review of laboratory, field and model studies, Atmos. Chem. Phys., 11, 11069–11102, https://doi.org/10.5194/acp-11-11069-2011, 2011.

Thien Khoi V. Nguyen, Shannon L. Capps, and Annmarie G. Carlton. Decreasing Aerosol Water is Consistent with OC Trends in the Southeast U.S. Environmental Science & Technology 2015 49 (13), 7843-7850, DOI: 10.1021/acs.est.5b00828

Second, I noted that the timeframe of the study includes 2020, yet there is no acknowledgement of the potential impacts of this anomalous year on the long-term trend. The authors need to at least make note of the fact that large emissions changes during 2020 may impact the trend, and where possible, quantify the uncertainty that brings to their analysis.

Minor

Line 157: “EC is a product of carbon fuel-based combustion processes and is exclusively associated with primary emission…” I am a bit confused on this paragraph. Are the authors referring to the EC measured by the OC/EC analyzer? If so, then this statement needs more clarification and discussion of uncertainties. Instrument-measured OC and EC are determined optically, and are thus more operational definitions than true determinants of carbon sources. The cutoff temperatures between OC and EC varies by instrument type and network protocol followed, and errors for the cutoff tend to be large. The authors should specify what temperatures were used for their analysis, and quantify where possible the uncertainty associated with their chosen protocol. See the following references:

Martina Giannoni, Giulia Calzolai, Massimo Chiari, Alessandra Cincinelli, Franco Lucarelli, Tania Martellini, Silvia Nava, A comparison between thermal-optical transmittance elemental carbon measured by different protocols in PM2.5 samples, Science of The Total Environment, Volume 571, 2016, Pages 195-205, ISSN 0048-9697, https://doi.org/10.1016/j.scitotenv.2016.07.128.

Khan, B., Hays, M. D., Geron, C., & Jetter, J. (2011). Differences in the OC/EC Ratios that Characterize Ambient and Source Aerosols due to Thermal-Optical Analysis. Aerosol Science and Technology, 46(2), 127–137. https://doi.org/10.1080/02786826.2011.609194

Line 160: Please provide more details on the EC-tracer method and how it differs from your Bayesian Inference Approach. In addition, please describe the Bayesian approach in more detail. What is the significance of the K values?

Line 170: It is unclear why the conversion from SOC to SOA is needed.

Line 183: The authors refer to annual average PM2.5 concentrations, but earlier state that most measurements were done from October to December. Are the concentrations shown here true annual averages, or are they the averages from October to December (wintertime)? How many of the samples fall outside this October to December range?

Paragraph starting Line 202: There are several undefined acronyms (POA, SOA, SIA) – this is true in other areas of the manuscript as well (ALWC, SOR, NOR). These need to be defined at their first usage.

Line 210-212: where the authors state that the Bayesian Inference approach is more reliable. On what basis are you making this claim? To what are you comparing the approaches to determine reliability?

Line 230: What control measures have been put in place for biomass burning and dust?

Line 244-247: Nitrate accumulation would be more impacted by temperature than NO2 concentrations.  

Section 3.4: The authors should specify which IMPROVE extinction equation they are using. There is an updated one from 2023 (see https://vista.cira.colostate.edu/Improve/wp-content/uploads/2023/10/IMPROVE_Data_User_Guide_24October2023.pdf, Section 8.1). Is that the equation used here? In addition, what is the local parameter scheme?

Line 368: Does the IMPROVE equation overestimate, or does the local parameterization underestimate? It is not clear that this can be said with any certainty. It is probably better to just state the differences between the methods.

Figures

Figure 3: What do the error bars represent?

Supplemental Information

Text S2 should be moved to the main document.

Figure S6 may benefit from an additional line showing the uncorrected changes in Cl-.

Figure S7: The colors of the lines do not match the colors in the legend.

Figure S8: Define SOR.