The study investigated size-resolved particle effective density and its relationship with chemical composition, which has implications for air quality, climate, and health. The use of AAC-SMPS in tandem with machine learning (ML) techniques, such as SHAP analysis, demonstrates a commendable level of innovation. However, I have two concerns: first, I am somewhat worried about the accuracy of the measurements from the entire system. Second, one of the highlights of the paper is using size-resolved effective density to infer particle composition, but the paper does not provide composition information for different sizes for comparison. I recommend a major revision of the paper.

Comments:

Why does the effective density of ammonium sulfate increase with particle size, while ammonium nitrate remains relatively stable (Fig. S2 )? I’m curious about the measurement results of this system for the effective density of PSL (polystyrene latex) spheres with different particle sizes. Since PSL spheres are essentially regular spherical shapes, their effective density should theoretically equal the material density (1.05 g/cm³). If the effective density of PSL spheres with different sizes cannot remain stable, I suspect that the observed relationship between effective density and particle size may not reflect the true situation but rather a bias in the observation system.

Line 85 Please specify the resolution (R_s) of AAC, the sheath of DMA, and the scanning settings of SMPS(such as diameter range, time for one scan, how many SMPS scans in 5 min). The Rs and DMA sheath could both influence the uncertainty of the results. Besides, After the authors provide these parameters, it is recommended that they perform an uncertainty analysis.

Line 98 The effective density of BC (could be smaller than 0.5 g/cm³  for fresh BC) is totally different from BC’s material density (~1.8 g/cm³), Is it reasonable to use the material density of black carbon to calculate the overall effective density?

Line 128-143 Another possibility for the unimodal distribution is that the minimum particle size setting for this observation was Dae = 200 nm, and the aerodynamic diameter of most fresh black carbon particles is smaller than this, so the bimodal distribution could not be detected.

Line 165 Please improve the clarity of Figure 3, especially as the figure legend is a bit blurry.

Line 190 I have two questions regarding the measurement and discussion of large particles (Dae = 531 nm):

(1) It was mentioned earlier (line 92) that particles with a mobility diameter exceeding 600 nm were not used in Gaussian fitting. How significant is the impact of this on the effective density determination at 531 nm?

(2) The number concentration of large particles is relatively small. After AAC selection, can the size distribution measured by SMPS still be fitted with a Gaussian function? I would like to see the size distribution scans by SMPS at different Dae values."

The current manuscript presents the characteristics of the ambient aerosol density of one-month online measurements in Hangzhou. The ms is generally written well. I have mainly questions for clarification.

Do the SHAPs work for particles larger than 350 nm? If not, it deserves to be mentioned in the abstract.

Lines 33–34: Please specify the size range.

The current system uses a dryer before the instruments. When RH is high enough, ambient particles may exist as aqueous droplets. Is there any effect of the particle phase state on density measurements?

Please define the sub-density mode.

The relationship between density distribution and particle mixing state is unclear. Please elaborate on what the bimodal distribution means.

The authors conclude no significant diurnal variation in the density values. However, Figure 4a noted a gradual increase in the density for 531 nm particles, which deserves more discussion and comparisons with other studies.

In this study, only small particles showed a strong correlation between feature values and the prediction. Please specify the feature values. Does the weak correlation for big particles mean that the chemical composition is invariable compared to that of small particles?

Minor comments:

Line 58: Please rewrite “machine learning (ML) methods, such as ozone pollution and potential aerosol sources.” Ozone pollution and potential aerosol sources are not ML methods.

Line 111: Please define “RF.”

Line 148: Fig.2 -> Fig.3.