General comments

This manuscript uses total column methane mixing ratios (XCH4) from the IASI instrument onboard the MetOp-B satellite to estimate the amount of CH4 released from the Nordstream pipeline leaks in September 2022. There have been previous studies that have attempted to estimate the source from ground-based observations, but this appears to be the first study using satellite retrievals for the source estimate. Based on two methods, the Integrated Mass Enhancement (IME) approach and a Bayesian inversion using a Eulerian atmospheric transport model, the authors find a source of 215-390 Gg for the first 2 days following the first pipe rupture. The lower end of this range overlaps with bottom-up and other observation-based estimates.

This study provides another constraint on this enormous (but also short-lived) CH4 source and is a nice example of how satellite observations can generally provide constraints on CH4 emissions. On the other hand, there are a number of shortcomings which I think need to be addressed before being accepted for publication in ACP.

The first is that the IME method, at least in the way it is used here, does not really seem suitable to the problem since the plume is not well-defined and is partially obscured by cloud. However, it could be interesting to combine this method with the Lagrangian model (HYSPLIT) simulations in a point-source inversion. This would entail using HYSPLIT to model the plume using a prior estimate of the source, and then the IME could be compared to the modelled IME for areas of the plume identified (e.g. Fig. 2) and then the difference between these estimates could be used to optimize the source. In this way, the problem of cloud could be circumvented, and a more complete estimate of the source could be obtained. It should be more comparable then to the estimates obtained using the Eulerian transport model, TOMCAT.

The second is that a rather poor estimate of the prior emissions from the pipelines is used. It would not require much effort to come up with a physical model for the emissions, which would give a more accurate representation of the amount of CH4 released and of the release rate. I do not understand why this was not done.

The third is that errors in the representation of atmospheric transport by the relatively coarse TOMCAT model are not really addressed. The authors mention this as a problem but do not attempt to reduce the effect of these errors. The authors perform one inversion where the mean of enhanced CH4 is used rather than the individual retrievals, which gave a rather different result for the emissions. I suggest the authors explore an intermediary approach using averaged retrievals, where the averaging is sufficient to reduce errors due to incorrect positioning of the plume, but where there is still some spatial information. Also, I think the authors should examine more carefully why there is a big discrepancy between the modelled and observed mixing ratios at three of the four ICOS sites, and check if the most recent ICOS data were used, since the first data submission missed some of the peaks (see specific comments below).

Specific comments

L20: Suggest authors specify the date of the first explosion rather than just “September 2022” since in L33 the author’s state they incorporate data up to 28 September, so for reader’s it would be interesting to know how many days this is after the start of the leaks.

L146: I think there are only 46 atmospheric stations in the ICOS network (see: https://www.icos-cp.eu/observations/atmosphere/stations). If the authors include flux sites in the 140 stations that they mention, this is inconsistent with the fact that they label these as “tall tower monitoring stations”, also, the flux sites are not relevant to this study.

L158: In the first submission of ICOS data, some high values of CH4 associated with the Nordstream leaks were filtered out in the automatic quality control. This was corrected in a subsequent submission. Could the authors please confirm which version of ICOS data they used, and if this version was the corrected one?

L194-195: The grid cells do not all have exactly the same area, so do the authors take the area-weighted average for the calculating the additional CH4 burden?

L236-238: Why was a constant release rate used as a prior estimate when very clearly the release was not constant. The first pipe rupture occurred early on the morning of 26/09, while the subsequent 3 ruptures occurred in the evening of 26/09, so this information should be reflected in the prior. The calculation of the second prior release rate also seems rather primitive, why not use a physical calculation of the release based on the length and inner diameter of the pipes and the initial pressure, and combined with what is known about the timing of the ruptures?

L246-247: If I understand correctly, only the Nordstream leak source was optimized. This assumes that the estimates for the non-leak sources are accurate. What is the uncertainty from this assumption? Could the authors estimate this uncertainty by trying different prior source estimates? Or include these sources also in the optimization?

L247-248: Only one overpass on morning of 28/09 was assimilated, so there is little constraint actually on the temporal evolution of the leak source, thus, it would be even more important to have a good temporal evolution in the prior estimate.

L266: It’s not clear to me what these 24 different inversions are. Could these be summarized in a table?

L295: Most of the plume must have been obscured on 27/09, the actual CH4 enhancement should not have been so much lower compared to that on 26/09 and 28/09.

L363: Same comment as for L158, the authors should check which version of the ICOS data they used.

L408: I think the authors could explore a bit further the method of comparing modelled mean XCH4 values over given regions with observed means, which would be perhaps a way of reducing the impact of the transport uncertainty, which appears to be quite significant (but not altogether unexpected considering the resolution of the model and that the source is only in the order of 100s meters in diameter). Perhaps this could be done by comparing the model with averages of retrievals?

L450: A peak emission rate in the night of 26-27/09 is actually to be expected because the subsequent 3 ruptures of the pipeline occurred at around 19:00 local time on 26/09. Thus it is only a few hours after 3 of the 4 leaks started.

Please see the comments in the attachment.