Inverse anthropogenic methane emissions from large point source are challenging but significant to manage fugitive emissions and design of control strategies. The study by Dowd et al. estimates methane fluxes from a large pipeline gas leak based on high-resolution satellite and ground-based mobile observations. The authors use and compare three different models to estimate methane emissions from pipeline leaks, assess the leak’s contribution to observed concentrations at a downwind tall tower site, and further successfully help the local utility company solve the leakage problem, which is of great application value. Nevertheless, from a research article perspective, there are certain aspects that require careful re-evaluation, such as the comparison and uncertainty of different top-down methods. Additionally, the overall structure of the paper could be improved. Only after appropriately addressing the following issues can I recommend this paper for publication in AMT.

Comments to introduction

The introduction section only consists of two paragraphs, which may not have a strong sense of logic and hierarchy, and there are too many words in each paragraph. The author can rewrite it as 3-4 paragraphs. And I suggest the section should supplement relevant background of previous methods for inverting point-source emissions, and the comparison of different methods.

Comments to estimation methods

1. In the flux estimation method, the authors do not provide a clear description of the calculation process and model parameters, and how the observation data are used in this method. I suggest the author should present specific calculation formulas and parameters explanation. Besides, how the source location is determined by the GEOS-FP wind direction? I find this sentence confusing. What is the resolution of the GEOS-FP dataset, and I suggest supplementing the relevant information.
2. In the Gaussian plume model, the authors use a mean of the source locations provided by the satellite retrievals as the source location, which is different from the assumption of the above flux method. What is the formula for calculating the diffusion coefficient and atmospheric stability classification? What is the unit of the modeled downwind concentration C, is it consistent with the observation. If not, how converted? Is the height of control surface related to atmospheric stability?
3. Since the authors can obtain the UKV meteorology to drive the NAME dispersion model, why not use this high-resolution data to drive the other two models? If the author reconciles the wind fields in the three methods, it may be more persuasive to compare the difference of retrieved fluxes.
4. The section lacks a description of uncertainty for each method. What aspects of uncertainty were considered and how is the range calculated?

Additionally, I suggest the author can consider adding a flowchart framework in the section to clearly show the calculation process of the three methods.

More minor issues:

There are different fonts in Line 144, 149 and 225.

Line 289: What is “The LI-7810 data” ?

The x-label in Figure 2 should be revised to “YYYY-MM-DD”. How to plot the uncertainty range? I recommend supplementing the uncertainty estimation for each method in the method section.

Line 327: Perhaps the author can assess how much uncertainty of the wind fields. Recommend discussing the consistency of results under the same wind fields.

Line 333-335: The Gaussian plume model assumes that the wind field is constant, which only uses wind vector of a single point, so I think the winds driving the NAME and Gaussian plume model are not fully consistent.

It is confusing in the 3.3 section what “pollution event” refers to. Two events defined in line 355 and line 360 seem the same event, but their definitions and criteria are not identical. I would suggest keeping consistent with clear definition.

Line 453: The difference between the NAME and the Gaussian plume model is not only from the wind fields. The uncertainty of the Gaussian plume model also comes from the atmospheric stability parameter, the source location and other variables.

Line546: What is the “CH4_{Tall t}” ?

Notice the unit formats (quite a few instances for the full text). For example, “ms-1” in Line 179 should be “m s-1”.

Referee Comments on First validation of high-resolution satellite-derived methane emissions from an active gas leak in the UK

Note – for simplicity the line numbers in this review relate to the online version, though this reviewer was provided with a revised version of the manuscript containing a missing section 2.2.3 containing a description of the tall tower measurement methodology.

General comments

This paper presents a series of measurements undertaken to confirm a methane emission identified during a satellite measurement targeted on another source. This is a useful and important study to show independent methods confirm the identified emission and to compare the emission rate estimates obtained by the different approaches. The study deployed ground mobile sampling (MS) based concentration measurements to provide confirmation of the emission location and emission rate. A comparison to emission retrievals using a NAME model based approach applied to both the mobile and satellite concentration data was also carried out to ‘normalise’ a model approach for both observational methods. As an additional study the impact of the emission on concentration levels observed at a tall tower was also assessed to determine if the tall tower would have detected the leak.

This paper is a valuable addition to the literature on methane emission detection and quantification.

I do have a slight concern over the title and the use of the term validation, which although it doesn’t have a formal definition, does imply performance evaluation and independent assessment.  There are three aspects of the satellite measurements that this paper covers;  leak detection, the identification of leak location, and leak emission rate determination. As noted by the authors, this provides one of the first assessments of the satellite with a real world ‘live’ leak. With some additional information the paper could provide more evidence to assess and validate the satellite results.

The paper also assesses whether the leak is detectable in the time series of concentration measurements at one of the UK DECC network tall tower stations at RGL. The is a very interesting addition to the study, however, it does not add much information to the validation of satellite derived measurements. The authors could consider whether this would be best separated into two papers – one addressing the satellite validation, and the other assessing the potential methods for detecting such a leak, including tall tower sites, satellites, and combined observation systems.

Specific comments and suggestions

The following suggests some comments on leak detection, leak location and emission rate quantification. Addressing these in turn:

Leak detection

The performance claimed for the satellite is 50% probability of detection for a 100kg/h emission at wind speeds of 3m/s.  The paper does not provide the meteorological information (e.g as a minimum the wind speed and direction) obtained from the different sources used by the different techniques, for the different days that the leak was measured. This should be included, at least in the SI.

Some indication of how sensitive the detection probability is to wind speed, leak rate, and other factors such as land cover would be informative and the expected detection threshold for the actual surface type (inhomogeneous farmland), and range of wind speeds during this study would be very helpful.

Leak detection is clearly demonstrated, as the presence of a leak from a distribution pipeline was confirmed by the relevant utility company. The initial identified plume is plotted in Figure1.  This and all subsequent satellite plots presented in the paper are roughly centred on the emission. The targeted landfill consists of a number of potential methane sources – a composting and household waste (plus closed landfill) (~1.2 km to the south of the leak) and an active area of landfill tipping (which at the time of the leak the active tipping area appears to have been 900m to the southeast of the emission) and other potential sources such as closed landfill area and a biogas plant.  In order to understand the ‘screening’ or leak detection capability of the satellite it would be useful to show/describe what area was within the original targeted/ tasked region.

Therefore, it would be useful to add information on the following points:

• What was the full field and centre of the original target area of the tasked satellite survey – is that equivalent to the areas shown in Figure 1
• Was the satellite re-tasked/focussed to the leak area?
• Was the detection/identification of the emission plume manual or was a threshold automatically triggered?

The paper also compares the potential for detection of the leak by two continuously screening approaches, the MARS approach utilising the Tropomi satellite which provides detection of leaks above > 25,000 kg h-1 and the observation of enhanced concentration using the Tall Tower network in the UK.  In order to place detection into context the paper could also provide a comparison against the ‘baseline’ leak detection mechanism in the UK for distribution networks – which relies on the fact that natural gas in the medium/low pressure network is odourised (the leak was apparently from a  Medium Pressure (MP) 350 mbar – 2 bar pipeline which would contain odorised gas). Given the close location of a road, it seems likely the leak would be reported by the public – it could be informative to understand from the utility operator if any reports had been received.

Other pipeline leak screening approaches are already used in different countries for example the DVGW Set of Rules used in Germany that mandates screening of pipelines by ground or airborne surveys. While this may be beyond the scope of this paper, if a screening approach is being suggested, it should be compared to other existing approaches.

I do have a further very speculative suggestion – what was the original driver for the selection of the landfill for the satellite tasking?  Was it selected because of pre-existing concerns about emissions? A quick Google search indicated press articles on complaints related to the odour from the landfill – an odour described as rotten eggs. Notably the mercaptan/sulphide based New Blend odorant used in the UK also has a ‘rotten egg’ odour. Could it be that local odour issues had been miss-attributed to the landfill when it was in fact the gas leak? This would have two implications – one, the leak had been occurring for quite a long period, and secondly the satellite provides a very useful tool for targeted investigation of such reports. But it would imply the underlying detection was indirectly down to the odorant in natural gas.

Note – the reference (Wales & West Utilities, 2023) links to a website with a generic description the utility company’s upgrade programme and does not give information on the pipe identified as the source of the leak – the utility company should be asked directly if this pipe was an older metal pipe or not.

Leak location

The general identification of the leak location is again confirmed by the utility company finding a leak in the area identified by the satellite. The paper presents the determination of the location of the leak from the satellite as (51.95088N, 2.09962W)  reported in Line 94. Google maps shows this to the west of the railway line and ~50 m north of the road. The location of the repair of the pipe, and therefore presumably the location of the actual underground leak, is clear from Picture S1 and the aerial image (S3). The rough location of the repair as per Picture S1 is (51.95068N, 2.09902W) again from Google maps In the SI a picture is also presented of vegetation die back (S2) which could be an indicator of the location where the sub-surface leak entered the atmosphere. This is some distance from the repair, but is possible, as this is along the underground route of the pipe (as available from the utility company). I have estimated, based on the Picture S2, the region of vegetation die back is at ( 51.95063N, 2.09943W) (from Google maps). Note, I did not find any discussion or mention of S2 in the paper.

These locations are ~ 40-50 metres from the initial satellite leak location, which could be discussed in the context of the location uncertainty claimed for the satellite. The utility company may also be able to confirm the precise location of the leak and whether they surveyed the area and found the point the leak entered the atmosphere.

The location is described as being determined from the initial satellite measurement when the leak was first identified, however, later in the paper the location used for NAME model, which is the same ( 51.95088N, 2.09962W), is described as being based on the average of four of the five source locations identified by the satellite. More information on the locations identified for each satellite measurement should be provided. The references for the satellite state ~25m location accuracy, but the location accuracy is also expected to vary with various factors (presumably meteorology, emission rate, surface type). It would be useful to provide and discuss the locations provided by the satellite and the uncertainty in these for each of the five passes that identified the emission.   

Throughout the paper the location identified from the satellite data is used, by the satellite team and also by the NAME model and the MS retrieval. A discussion on the the sensitivity of each of the emission estimation approached to the location would be useful as a separate discussion. At various points throughout the paper (e.g Lines 180, 416, 471) uncertainties due to the source location are mentioned, so it would be useful to investigate the uncertainties in the location in more depth.  When investigating the sensitivity of the NAME model results to the location a series of NAME model runs with the location varied by 10 m were carried out. The reason for the choice of 10m should be stated, the uncertainty in the satellite location determination could have been used, or some variation based on the range of locations determined by the satellite. Similar sensitivity analysis of the effect of the location on the other emission estimation methods could also be carried out.

It is also mentioned that the repair work may have made the emissions more diffuse, however, if the leak had been migrating underground before entering the atmosphere, the repair work may also have moved the effective leak location. More detail on the individual location information reported for each measurements would be informative in discussing this. The implication of a more diffuse emission is also not discussed in detail.

Emission rate

Three different approaches are used for quantification:

Satellite concentration data with IME method, using GEOS-FP wind.

MS concentration data with a gaussian plume model using either onboard wind data or the Met Office’s UKV model.

NAME Lagrangian dispersion model applied to Satellite and MS concentration data, wind data from Met Office UKV model.

These sections should include more information on the sources of uncertainty in each emission estimation method and the impact of assumptions made.

To provide validation of the satellite emission data using alternate methods (MS and the NAME model) when no direct measurement of the emission rate is possible, it is important that all sources of uncertainty in all the methods are assessed. In particular this should address what sources of uncertainty may be correlated between the methods. In this study, the methods used are not independent – both MS and NAME make use of the location provided by the satellite in their retrievals. In addition, all the methods (apart from one MS measurement) use wind data derived from models. Given the importance of wind direction and wind speed on the results it would be very useful for the paper to assess the meteorological data in more depth. The local scale of the measurements, and the local impact of features such as the rail bridge,  and the effect that the meteorological data has on the retrievals.

While the uncertainties in the different estimated emissions are given, it would be useful to state if these are expanded uncertainties and with what degree of confidence these are quoted, and what coverage factor is used. 

For the MS – at least 12 passes along the road are made for each measurement some more details would be useful - for example the time period of the traverses, how the plume average transect presented in the SI is determined and whether the emission retrieval is carried out on these averaged concentration data or on each run, and if the average is used.  For measurements this close to the source, would it be expected that local wind fields, the effect of local terrain such as the trees and railway bridge would have a significant impact, and the concentrations observed by the MS be sensitive to small changes in local wind. Also the fact that the sampling height of the MS will vary as the car traverses the bridge may also have an effect. The discussion on the NAME model states the UKV wind data has a resolution of 1.5 km and an hourly temporal variability, some discussion of the potential influence of this on the very local wind field relevant to the local dispersion would be useful.

The NAME model provides an alternative set of mass emission estimates for both approaches using a common model. The mass emission estimation methodology is tailored for each observation approach – for the satellite total column data is produced to match the satellite observations. It is noted on line 209 that the modelled plume and the satellite observations do not overlap well. Section 2.4 discusses the determination of model bounds, however, these relate to the definition of the plume area to be integrated for a given model output. The NAME model was also run with different locations to test the sensitivity to source location. Interestingly for most runs the perturbed model runs are all higher than the default location. This seems counter intuitive and it would be worth checking this is correct and discussing what might explain this.

Some further investigation of the sensitivity of the NAME model plumes to other input parameters would be interesting, to provide more understanding of the uncertainty in this process. This would allow the differences between the model and satellite emissions to be put into context. It would also be worth exploring whether any of the constraints used to match the model to the observed satellite plumes could also introduce correlations between the two approaches.  This could help understand how independent the NAME approach is. This is important as the NAME time series is used to support the suggestion that the emissions are varying with time.

Similarly, the NAME model was configured to provide  concentration path data to match the MS measurements – constrained to be at a height of 2m. This does not take account of the different elevations for the MS sampling point due to the terrain (e.g rail bridge).  Some investigation of the influence of relevant parameters in the model for the mobile monitoring would again be useful.

The time series of measurements is presented clearly with the graphic in Figure 2, the other events mentioned in the text could also be added, the data the utility company were informed and the data the investigation and repair work began. Any further information from the utility company on the leak, and whether they did vary the pressure in the pipe after being informed of the leak would be important information to also aid the interpretation of the results.

The discussion notes the gas leak is well within the detection threshold of the satellite quoting 42 kg/h, however, in other areas of the paper a detection threshold of 100 kg/h is used, and the references given state 200kg/hr as the smallest detection.  As mentioned earlier a discussion on the theoretical detection limit for the satellite for the conditions of these measurements would be useful.

In line 91 the impact of wind speed is mentioned as likely cause of differences in the flux observed by the mobile system – however, as mentioned the wind data is not reported in the paper and should be added. It is suggested that the difference between the mobile and satellite measurements could be due to real changes in the emission. However, the two mobile measurements in May and June overlap with their uncertainties as do the two corresponding satellite measurements. Both techniques show a reduction in emissions from May to June – which may be real – though for both measurement techniques the difference between pairs of measurements lie well within their uncertainties.  The difference between the two methods does look to be more systematic and probably not due to variations in emissions (though this can't be ruled out). Some discussion on whether there could be any reason for systematic under reading by the satellite or over reading by the MS approach would be informative. It is also worth noting that the NAME model approach gave lower results (in Table 1) for both the satellite and MS data on both the May and June results, with the mobile data again resulting in higher emission rates. This either supports the suggestion that the real emission rate is varying, or implies some systematic effect impacting both the NAME model and the other method retrievals. Some more detailed investigation/discussion on this would be informative.

The paper would benefit from a more detailed review of the potential influence factors that might affect the methods, in particular any effects that might systematically affect all the methods. The sparse and temporally non-overlapping measurements does make it hard to draw firm conclusions on the performance of the satellite.

Minor comments

Check consistency of form of units e.g.. in Line 117 both  kg h^-1 and  m/s are used. 

Typo Line 462 “We assessed the frequency of pollution events during our both NAME_spring and NAME_long simulations and found a low number of ‘leak pollution events’.  “ Check word order – should it be …during both our …

For the data from the 22nd May (438 ± 215 kg h -1 ) and  26th May (998 ± 377 kg h-1 ) the uncertainties do overlap – however, in Figure 2 the error bars do not appear to overlap for these two results – please check.

For the NAME results for the satellite data  on the 22/05/2023 the reported result 384 is not within the bounds provided [173, 292], for all other results the model bounds are above and below the reported number – is this correct?