This study combines longterm AOD observations and detailed ground-based PM2.5 measurements to comprehensively evaluate GEOSCCM performance in surface aerosol, column AOD, and aerosol optical properties. It identifies key factors for improving performance of GEOSCCM and potentially other models. I recommend this manuscript to be published after minor revisions that address questions below. 

Specific comments:

L159-160: The ‘Southern Asia’ region is huge and include polluted regions with different behaviors (South Asia vs East Asia). Should justify why they can be combined in this analysis. 

Section 4.2: 

surface PM2.5 is notably underestimated for most regions except for NA and Aus, despite that AOD is in good agreement with satellite. Could add some discussions about why surface PM tend to be underestimated by GEOSCCM. Are there any previous work on the aerosol vertical distribution that can explain the bias? What could be done in the future for improving the agreement near the surface? 

Section 4.3: 

(1) the under estimated BC in Figure 6 and good agreement in Figure 7 seem to be suggesting anthropogenic BC not being capture in the model (emission inventory). 

(2) There is consistent seasonality bias. And according to IMPROVE, the bias in background dust is getting higher. What could be contributing to this bias and trend? Inland intrusion of dust from the Atlantic Ocean can explain the bias in southeast US and maybe Southern US, but no so much in southwestern US. 

Minor comments:

L27: ‘relate more to simulated aerosol mass’ - would be more accurate to say ‘aerosol speciation’

~L300: eqn 1, sigma simple shows up as a question mark in the pdf document.

Table 1: ‘Washington U. S. Louis’ - it is not common to abbreviate ‘St. Louis’ as ‘S. Louis’. 

Figure 6: y axis label is cut off for the first and fourth rows. 

Table 2: could be moved to the supplements as it is not discussed a lot and figure 8 is illustrative enough. 

L484: ‘Dame’ should be ‘Same’

L501: ‘except Europe’ could be removed since the model doesn’t obviously perform worse here. 

This study presents a comprehensive evaluation of aerosol optical properties (AOD) and PM2.5 mass concentration simulations using the GEOSCCM model, combined with multi-source observational data (satellite AOD, ground-based PM2.5 measurements, etc.). The work is organized and written in a very clear way. The long-term simulation (1960-2020) and the analysis of seasonal variations are impressive. I recommend the publication of this work after minor revisions.

(1) The manuscript compares model outputs with MODIS NNR, AERONET, and IMPROVE datasets and have adequately discussed the reasons of model biases, including emission inventory, and the method simulating inorganic nitrates etc. I suggest that the authors add a few sentences on the uncertainties in observations that might affect the model evaluation, e.g., would MODIS NNR retrievals exhibit biases in biomass burning regions due to differences in smoke optical properties and thus contribute to the model’s underestimation of AOD?

(2) As RH plays an important role in simulating aerosol scattering, I am curious how to improve RH performance in GEOSCCM?

(3) The current “Discussion and Conclusions” section is  long and the readers may miss the main points delivered. I suggest divide section 5 into two sections for discussion and conclusion separately. 

(4) Is the simulation presented in Fig. 8(b) and Fig. 10(a) also at RH=35 %? Better clarify this.