In this study, the authors did lots of work to estimate AOD changes during spring 2020, based on satellite remote sensing products, and to evaluate AOD responses to emission reductions in several CovidMIP models. Unfortunately, the different satellite instruments gave such a large spread in AOD changes, even with dust (as one of the natural species) excluded in the retrievals. Strong regional dependence of the robustness of observational estimates and model performance is found, but drivers behind this are unclear. The analysis of CovidMIP models does not add much to the literature, beyond the original CovidMIP paper (Jones et al., 2021) and other published studies. These are the major concerns leading to my hesitation to recommend the current manuscript for publication. The CanESM5 sensitivity tests are more interesting and potentially revealing. The novelty and science significance of this paper may be increased by focusing more on in-depth analysis of the sensitivity experiments on the roles of meteorological factors and possibly microphysical processes driving the response of aerosols to emission reductions in spring 2020.         

Below are a few more specific comments:

• The abstract lacks quantitative results either from the observational estimates or model analyses.
• Line 4 (and several other places): strictly aerosol optical depth rather than aerosol burden is estimated in this study, which should be made clear.
• Line 39-41: This statement is inaccurate. Jones et al. (2021) did specifically compare regional AOD changes among the participating Earth system models.
• Line 46: Please be more specific about what kinds of observed changes being used for model evolution purposes. Global or regional climate models use many different observational data for the evaluation purpose.
• Line 61-64: depending on the purpose of obtaining aerosol concentrations, it can be a big problem of using a column optical property (AOD) as a proxy for aerosol concentration or aerosol burden mentioned in the first science question.
• Line 78 (and section 2.1): It is too vague and generic to name meteorological conditions as one of the determining factors of AOD. In addition to the emissions, one should at least speak to the transport and sink terms of atmospheric aerosols such as dry and wet deposition in aerosol budget equation, although the detailed aerosol chemical and microphysical processes are sometimes even more important, depending on the aerosol types.   
• Table 1: There might be too few models. Are they outliers among the 12 CovidMIP models?
• Line 228-231: How was the first assumption tested?
• Line 284: This statement about contribution of dust to total AOD is inaccurate and can be misleading. It highly depends on season and region. Globally, dust contributes to less than 25% of annual total AOD.

The manuscript “How well do Earth System Models reproduce observed aerosol changes during the Spring 2020 COVID-19 lockdowns?” use the COVID lockdown and the following emission reduction for model evaluation. Modelled changes in aerosol optical depth (AOD) due to COVID restrictions and satellite retrieved AOD in March, April, May (MAM) 2020 are compared. The Earth System Models and observations show consistent results in Europe and India, where India is the only region considered with a significant reduction in AOD in MAM 2020 in the observations. In China and Northern Hemisphere as a whole, the modelled reduction in AOD is overestimated. Using one model, a systematic assessment of the influence of meteorology, baseline emissions, size of COVID emission reductions are done. The spread in the observations of AOD is a limiting factor of further constraining the models.

The manuscript is well structured and presented, and I have only a few comments.

The uncertainties in the satellite AOD products precludes a further constraint on the models responses to emission reduction. As this method outlined here, also can be used to evaluate response to future emission reductions, a bit more on future direction of satellite AOD evaluation would have been good.

In the abstract:

L4: “observed regional aerosol burdens during” It is not aerosol burden that is assessed, but AOD (as a proxy for aerosol burden).

Consider swapping section 2.3 and 2.2?

Table 1: The mineral dust column can be misread as if models include mineral dust or not in the simulations. Maybe replace “Mineral dust?” by “od550dust” or “Mineral dust output”.

Section 3.2: Could be useful with a table of the satellite AOD products.

ACROS-C is used in Figure 1, but not mentioned in section 3.2.1.

L253: “the Northern Hemisphere as a whole” From figure 1 and text elsewhere, the “as a whole” is not entirely correct as it is 0N-70N. Replace “as a whole” with (0N-70N).

Figure 1 (and 2 and 3) contain a lot of information. It could be useful to add more information to the legend, maybe first present what is included in the timeseries (the six models and the observations with symbol and black line). Then, as a separate box or just below, the 2020 values (Square: control, diamond: covid pert, MMEc). For MMEc maybe only show the square and not the line, as I was looking at the time series when I first looked at the plot. See also if filled, black outline, opaque/semi-transparent can be indicated inside the figure as well. I am not able to see if the results are plotted opaque or semi-transparent. Possible to use filled or not filled symbols instead?

Figure caption: Delete “horizontal offset for visual clarity” Already mentioned that the right side of the panel was for 2020 and “2020 values” are the titles of the subpanels.   

L313: It is hard by eye to see the difference in the trend between observations and models for the reference period (2015-2019).

L359-362: This was a bit unclear. Just note that the MMEc is as in Figure 2.