This manuscript studied the impact of meteorological changes and emission reductions on PM2.5 pollution in the Pearl River Delta (PRD) during the cold seasons from 2015 to 2017. The authors aimed to explain why PM2.5 levels in the PRD remained high despite of significant emission reductions in PRD and its upwind regions in East China. By applying the regional models, they found that transport contributions to PM2.5 levels rose from 70% in 2015 to 78% in 2017, while local emissions declined. And they concluded that the meteorology change was the dominant driver of the multiannual variations of PM2.5 during the studied period. And the meteorological change was likely driven by large scale climate variability, namely the transition from a strong El Niño in 2015 to a weak/moderate La Niña in 2017. This study also pointed out that the meteorological impact should be taken into consideration when the emission control policies were assessed.

The manuscript is well organized and written clearly, the description is precise, and the discussion is fruitful. I recommend publishing it after minor revision. Below are my comments referring to lines (L), equations (Eqs.) and figures (Fig.).

L160: spinning -> spanning

L149: (1) Maybe it is better to specify the names of nine cities here when this concept is first mentioned.

(2) It seems that not all the nine cities are shown in Fig. 1.

Fig. 1: (1) The x tick labels and y tick labels need to show the unit, e.g., 112 °E.

(2) The orange line is not introduced in the caption.

Fig. 2: (1) The latitude and longitude labels are too small.

(2) "The black boxes are the simulation domains for WRF, while the nested areas indicate the simulation domains for CMAQ.": Does it mean that d01, d02, and the domain larger than d01 are all simulated by WRF? And CMAQ only simulates d01 and d02? Please specify it clearly.

L190: Oct and Dec are selected in the simulations, but in L159, the cold season is defined as the period from Oct to Jan. Will this affect the simulation results?

Eqs. 5 - 10: Why the contribution of S_Emis_O,15/16 is not calculated by C_L15O16M15 - C_Base15? And similar questions for S_Meteo,15/16, S_Emis_O,16/17, and S_Emis_L,16/17?

Fig. 4: The x tick labels and y tick labels need to show the unit, e.g., 110 °E.

Fig. 5: The legend of "Observations" is point instead of point+line.

L412: What does 'population-weighted mean' represent?

Fig. Graphic Abstract: Why the decrease of local emissions is marked with "Enhanced"?

This manuscript presents a valuable investigation into the complex interplay between emission reductions and meteorological variability on PM2.5 pollution in the Pearl River Delta from 2015 to 2017. The study is well-designed, employing the WRF/CMAQ model with a comprehensive set of sensitivity analyses to deconstruct the drivers of the observed persistent pollution. The paper is clearly written, and the conclusions are well-supported by the analysis.I enjoyed reading this manuscript. This paper is suitable for publication after some minor revisions.

1. A minor suggestion for the title. While "Persisted" is accurate, a word like "Anomalous" or "Resurgent" might better capture the unexpected increasement of PM2.5 during 2015-2017. This is just a suggestion for the authors to consider.
2. In Section 2.3 (line 213), the manuscript introduces the Factor Separation Method (FSM). Could the authors first confirm if this method is fully described by Equations 1-4? Second, a brief discussion on the advantages of FSM over the more traditional Brute Force Method (BFM), such as its ability to quantify the interactive effects between different emission regions, would be beneficial for readers less familiar with the technique.
3. Regarding the background contribution (Fbg), the definition could be more explicit. Please clarify in the text that this term represents all influences from outside the d02 simulation domain, as determined by the chemical boundary conditions used in the model.
4. Could the authors provide a more explicit discussion on the specific meteorological drivers for the changes in sulfate andnitrate? For instance, was the significant change in nitrate primarily driven by shifts in temperature and humidity (impacting chemical formation) or by changes in wind patterns (impacting transport)? Disentangling these influences would provide a clearer mechanistic understanding.
5. In Sections 4.3 and 4.4, the text describes complex trends from the source apportionment and budget analyses. While the figures and tables are comprehensive, the narrative would be much easier to follow if key quantitative values were integrated directly into the descriptive text. This would reduce the need for the reader to constantly switch between the text, figures, and tables.
6. The conclusion that unfavorable meteorology can mask emission control benefits is powerful and has significant policy implications. This point could be further nuanced by adding a brief discussion in the final section about the differing timescales of these effects. While meteorological variability can clearly dominate inter-annual changes, it is crucial to emphasize that sustained, long-term emission reductions remain the fundamental and most effective strategy for improving air quality.