Review of “Inter-comparison of multiple two-way coupled meteorology and air quality models   (WRF v4.1.1-CMAQ v5.3.1, WRF-Chem v4.1.1, and WRF v3.7.1-CHIMERE v2020r1) in eastern China,” by Gao et al., submitted to Geoscientific Model Development.

This paper intercompares several two-way nested coupled meteorological and air quality models in eastern China. This is the first such comparison that I am aware of. The paper is thorough and should be published following some modest revisions.

Introduction. “The feedbacks of aerosols to meteorology have been widely investigated by two-way coupled meteorology and air quality models in the past two decades.”  Two-way coupled meteorological and air quality models have been developed and applied for almost three decades (Jacobson, 1994; 1997; 1998, 2001).

Table 1. what is the vertical resolution of the boundary layer in each model (how many layers in the bottom 1 km and what is the bottom-layer thickness?

Table 1. How many aerosol size bins and components per bin? Do you use a modal or discrete bin approach?

Table 1. Does photolysis account for clouds? How are clouds treated for radiative transfer calculations?

Table 1. What height is the model top and how are model-top boundary conditions treated?

The authors evaluate with RMSE, which is an absolute quantity for each variable. However, normalized gross error (absolute value of differences between model and data, divided by data, summed over all locations and normalized by the number of locations, is a more useful metric since it gives error relative to the data values rather than an absolute amount. It is similar to NMB, but with absolute values taken, since NMB cancels out large errors of the opposite sign. Also, it would be useful to see some time-series plots of model results versus data.

A lot of comparisons are performed, but what are the most relevant comparisons with data? Ozone and PM2.5 calculations? Please focus the discussion of results more. Right now the results section is crammed with lots of information that is not easy to determine from what is important and not important.

References.

Jacobson, M. Z., Developing, coupling, and applying a gas, aerosol, transport, and radiation model to study urban and regional air pollution. Ph. D. Thesis, Dept. of Atmospheric Sciences, University of California, Los Angeles, 436 pp., 1994.

Jacobson, M. Z., Development and application of a new air pollution modeling system. Part III: Aerosol-phase simulations, Atmos. Environ., 31A, 587–608, 1997.

Jacobson, M. Z., Studying the effects of aerosols on vertical photolysis rate coefficient and temperature profiles over an urban airshed, J. Geophys. Res., 103, 10,593-10,604, 1998.

Jacobson, M. Z., GATOR-GCMM: A global through urban scale air pollution and weather forecast model. 1. Model design and treatment of subgrid soil, vegetation, roads, rooftops, water, sea ice, and snow, J. Geophys. Res., 106, 5385-5401, 2001.

I have some major concerns of using FDDA for a study like this and try to objectively learn from model performance. When using FDDA, nudging is forcing the model to the same observations that are being used for evaluation (NCEP reanalysis includes probably all observations that are using for evaluation). Additionally, FDDA makes conclusions for feedback studies very problematic, since the physics parameterizations react in very different ways. Furthermore, you are using different physics and chemistry routines in all models. This makes this an "apples to oranges" comparison. Do you know what happens in Morrison microphysics when the mix-activation routine is not called (no chemistry)? My first thought was to reject the paper, however, the authors have done an immense amount of work and present some useful results that can be used by some of the developers. In turn I will propose accepting but with major revisions. These major revisions should be focused on the interpretation of the results. Abstract and conclusions should clearly say that this work is NOT to decide which model is better or worse since employed setups are very different and furthermore, it is not clear how FDDA runs influences feedback studies. Also, the authors need to be clear on what is used by Radiation (R) and MicroPhysics (MP) if feedback is off versus on. Are you just using a constant droplet number? A climatology? WRF-Chem has a lot of options, how come you decided to use different physics than in WRF-CMAQ? Chimere is way behind in the WRF version used, which makes that one even harder to compare. I am not asking you to rerun this monster simulation, but you will need to rephrase some of your abstract, conclusion, and results description. Since this reviewer is not asking for additional runs, this should not be a major effort. I really appreciate the work you folks put into this paper!

It would be interesting, maybe in a later additional paper, to compare the feedbacks in the different models for a shorter run that does not use FDDA. Maybe picking one or several interesting 5 day periods from the long runs that you used for this paper.