The revised manuscript has significantly improved over the original. Please see the report by Referee #3 and make the requested revisions.

The manuscript has improved significantly after revision. Referees #1, #3 and #4 still have a few additional comments. Comments by Referee #4 are given below. Please address all Referee comments in the next revision of the manuscript, or explain in detail why additional revisions are not needed.

_____
Comments by Referee #4 on revised manuscript:

I appreciate the authors' long replies to the comments including mine (this is my second review). However, I still see the two key concerns have not been addressed. I conclude that the topic is important but the analysis method is not robust.

First, the authors did not address the potential bias in the emission estimates in the revision. They state it is out of the scope of the study. Then they should clearly state their estimates are subject to potential biases, not addressed here, so that the readers can understand the estimates with correct information. Again I argue that recent many “regional” studies evaluate meteorology, which gives the estimates more confidence. Are the authors confident of the estimates? I believe there are at least a few ground-based meteorology observation stations around the tower (if the tower itself does not have one). To the authors, it may not have been straightforward to transform the met. bias to the bias in the emission estimates although it is still doable. But at least I expected the authors to evaluate the meteorology data and discuss about it. At the minimum the authors should state this study’s limitation in terms of bias analysis.

The authors did not address my previous comment about the obs-model mismatch uncertainty (covariance matrix). I see another reviewer also commented on this. The authors only say “observation–model mismatch errors were defined as RMSE value of model minus observation.” This is only valid when the model does not have a mean bias compared to the observation. But this is not the case of GHG problems because the prior emission is biased (too low or too high) or imperfect (e.g., spatially). That’s why we have two covariance matrices: one for the emission and another for the random error. Figuring out the random error (i.e., the RMSE in the posterior comparison) is a time-consuming process, but should be done properly as many recent regional studies have done and developed the method. Since the treatment of random errors is “uncertain” here and I am uncertain about the posterior estimates from this study.