Review of: Validation of the Fugitive Emission Distributed Sampling (FEDS) system: A mobile, multi-inlet system for continuous emissions monitoring.

General: The work presented here feels a little “work in progress” rather than a true validation of the system. Whilst it is a useful step in the validation process of FEDS, I am a little concerned that it doesn’t demonstrate the capabilities in a truly rigorous way in comparison to the intended use scenarios and that a follow-up piece of work is needed. I would request that the term “validation” is removed from the title as I think more is needed – maybe “first assessment” is more appropriate?

I believe this work is worth publishing after revisions, as it lays the groundwork for the FEDS system. A part II controlled release study paper would be welcomed – possibly alongside the first site level measurements?

There are a number of significant improvements and increase in scope that I would like the authors to consider. Some should be feasible within the boundaries of the current dataset, but others I would like them to be mindful of as they design their next controlled release experiments to assess and improve the system.

1. The work should be as rigorously compared to other systems attempting to achieve the same / similar goals. The work of the Stanford and Colorado State Universities that have done significant work assessing continuous monitoring system performance should be heavily cited and compared as directly as possible through the creation of comparable performance metrics. As it currently stands, I have no direct reference in the paper as to whether this system performs well, averagely or poorly compared to other systems.
2. The use of the models chosen (and why not others) is not terribly well justified. I would have liked to have seen a comprehensive look at the possible choices and then reasoning for using these and not others. There are solution possibilities using tools such as GRAL (https://gral.tugraz.at/) or even an algorithmic approach to quantification such as that being proposed by LBL for abandoned oil and gas infrastructure assessment (https://egusphere.copernicus.org/preprints/2025/egusphere-2025-344/egusphere-2025-344.pdf)
3. There is no “non-model” baseline to compare to. Given the simplicity of the set up I would have liked to have seen a simple analysis of the results identifying when a concentration is elevated above background by a certain threshold and then plotting a back trajectory line from the node using the measured wind fields, and using that to determine source location and provide a comparative success index without the need for a model.
4. I wonder if there is room for improvement on the inlet design? There a plenty of design set ups that could have borrowed heavily from the eddy covariance community that have utilized fast 10 Hz instruments with rapid throughput of gas which would massively reduce the dead analysis period that needs to be removed and allow much faster switching between nodes. I would encourage this to be investigated and some spreadsheet calculations to work out if this is worth redesigning for in the future.
5. Somewhat adding to the previous point, the low time resolution of the system (one measurement period per node per hour) is very limiting and will cause severe problems when trying to scale up to sites with intermittent sources or variable emission rates. It may be okay for purely providing an overarching emission estimate from long duration deployment, but without the ability to also act as a mitigation tool the role of it as a method may be very limited in the context of continuous monitoring options.
6. As you have identified, understanding the wind is key. For future work, I would recommend siting a small 2D sonic at each node so that exact wind information can be obtained for each location. For any site with topography or structures, quite dramatic changes in wind can be seen across a site due to shielding and slope effects which would be important information for the modelling.
7. The potential improvement of fully utilizing the atmospheric stability is mentioned in the paper, but not followed up. It seems that adding this analysis in is within the remit of the work as it should demonstrate whether this is something that should be included in future analysis?
8. Why does the performance of the two models vary so significantly in terms of false positive detections? From an operator perspective this is a key metric as no one wants to send staff out to fix things that don’t exist. This should have more detail around it and understand why this is occurring. This should be considered a significant problem for the windtrax solution.
9. There is a hint in the conclusions that low cost sensors were also trialed – I would love to see the data on this and some more details as to how it was set up. A null result is still very useful and may help others to not continue down a similar path of method.

Detailed line by line:

Abstract:

L15: Add some details of the CR - number of sources, release rate ranges, what is being simulated.

L19: Without knowledge of the true emission rate these values are a little meaningless. If the emission rate is 100kg hr, then these are wonderful. If it is 2kg hr, less so... Maybe replace throughout with % of true emission.

L21:23: Feels more like discussion that abstract.

Main Text:

L31: efficient, accurate and transparent monitoring

L41: Discuss snapshot vs continuous

L46: Put in context of reporting requirements such as the new EU regualtions or voluntary programmes such as OGMP2.0? These are already in place and have specific requirements around needs.

L74: The CR needs to be better described here. Is it fully blind, single point or multiple point etc...?

L79: I wouldn’t consider custom Gaussian plume out of reach of commercial teams.

L83: Specify what is considered high-performance

Fig 1: Instrument labelled as uMEA

L114: The FMEA may sample at 1Hz, but what is the turnover time of the cell and therefore what is the true sampling resolution of the system?

L125: Expand this section so that the reader can understand the lags and data invalidation periods. 60s seems like a very long time to clear a portion of line if the lines to each node are being continually pumped. I’d like to see some of this analysis in the main paper.

L129: Is this really sufficient for traceability? I think this quality of calibration is quite poor and I would have expected better QC on this. Can this be replicated in the lab now with more cal gases to demonstrate that instrument performance is as expected over the measurement range seen.

L169: I do have questions around wind persistence and whether it would become less of a controlling problem if you weren’t having to average everything to hourly timesteps. I suspect that there isn’t sufficient data to dig into this further, but I would consider looking at minute by minute enhanced methane readings and correlating with wind direction to see if that gives better data for locating an emission source.

L200: As this is stated to be a validation study, I would like to see more information and justification around the design of the experiment – especially given that it is stated that this system will be used for landfill quantification, among other things.

Fig 3. Can the wind direction be plotted as dots so that there is no 360 jumps in the data.

L260 section: I’d like to see a probability of detection metric determined for the set up if at all possible.

Fig 4. This is not at all intuitive and I would give serious consideration to redefining how this is presented and instead look at methane excess over background as the primary metric rather than just atmospheric mole fraction as I think it would provide much more immediate understanding

L434: Reference issue

L458-462: Whilst these models may perform to the stated stats, I don’t get a sense of understanding as to why they are performing well or poorly (with the exception of the discussion of the wind persistence). Is there more to the modelling than this / more nuance?

L497: Surely measuring more methane downwind from the source can’t be considered a conclusion?

L513: Give % errors so that we can see how performance was without having to know what the release rates were.

Conclusions: General – I’d suggest shortening and tightening up once the major corrections are sorted. It is a little long and unconcise as it stands.

Validation of the Fugitive Emission Distributed Sampling (FEDS) system: A mobile, multi-inlet system for continuous emissions monitoring

Shaw et al.

egusphere-2025-1451

The paper describes the validation of a monitoring system to be used to quantify methane emissions over long-term periods from large, aerodynamically complex sources (e.g. natural gas network, landfill, and waste treatment sites).  The validation was conducted using four controlled emissions (point source; 20-hour duration; fixed emission rate 1 kg CH₄ h^-1) with the methane mixing ratio measurements being made a short distance (~35 m) downwind of the release over a flat grass fetch.  The measurement campaign lasted 18-days in March/April 2022.  Results show good agreement between calculated emissions and the controlled emission rate.

The paper is written well but similar results have been published by other studies, research in this field has moved on considerably.

Firstly, there is little novelty in the work.  Many systems have been developed that can quantify methane emissions from point sources a relatively small distance away over an aerodynamically simple wind field.  New research in this field either investigates using low-cost/low-power technology that can be deployed remotely or novel dispersion modelling approaches.  The authors claim the USP of this paper is that it uses a single expensive analyzer connected to multiple sampling locations.  However, methane analyzers, such as the one used here (no details were given on the spec and I cannot find it on ABB’s website), are expensive (USD 20,000 - $50,000), require mains power to operate and a controlled climate environment to operate in.  Moreover, the methane mole fraction data were used to generate an hourly average – raising the question of whether a trace methane analyzer is actually needed.  As a result, those quantifying methane emissions to justify GHG emission estimates (e.g. landfill and O&G operators) are using alternative instruments that are lower cost, lower power and can be adapted to operate in a range of environmental conditions.   The data are presented but there is no synthesis or research questions asked other than - does this approach work in the simplest of emissions scenarios.

Secondly, claims made that the FEDS could be used to quantify emissions of gases from other sources are not validated by the data presented in the paper.  The controlled release experiments are very basic when compared against emission scenarios generated at FluxLab (https://fluxlab.ca/) or METEC (https://metec.colostate.edu/).  The authors do not describe why they have chosen a single, continuous point source emission of 1 kg CH₄ h^-1, released at ~2 m above the ground for 20 hours or what real-world emission scenario this is meant to represent.  This emission not representative of those from either landfills (heterogenous, area source emission) or wastewater treatment sites (both point and area source emissions).  This could simulate a source on the natural gas network (usually point source), but in real-life emissions are typically intermittent (e.g. cycling with separator dumps).  The aerodynamic complexity of the experiment is a real weakness, and I find it difficult to imagine any real-world emissions scenario that this would represent.

Overall, I would recommend: 1. that the authors reanalyze their data to present something more novel (e.g. down-sampling the mole fraction data to assess the highest sensor detection limits that still yields the same results – how does this affect the cost of the system and are there available sensors that could be used instead of the LGR?); and 2. be very clear in the conclusions what the limitations of these validation experiments are and place the controlled emission scenario you have used in a realistic, real-world context.