This paper is a pleasant-to-read update of the possible role of a molecular H2 economy on the climate through chemical modulation of the greenhouse gases CH4, H2O, and O3.  The authors do a great job of discussing complete scenarios in terms of a life-cycle assessment involving the swap out of fossil fuels for H2.   It is a single-model result that is valuable but it does not help us choose or consolidate the large range of published results on this topic over the last two decades. It does present an excellent analysis of why Fuglestvedt's short-term approach for H2 did not work, but it does not really contain a "new methodology".

The model simulations are all credible and well designed, but I find three major problems with the write-up of their results.  These could be readily fixed if the authors choose to do so.

(1) Abstract: "We find that increases in hydrogen result in increases in methane, tropospheric ozone and stratospheric water vapour,…"  We all knew this 20 years ago (see T. K. Tromp et al., Science 300, 1740 (2003) and M. J. Prather, Science, 302, 581 (2003), also R. G. Derwent et al, Climatic Change, 49, 463 (2001)) and even the old papers estimated the climate impacts.  The claim here in the abstract, and as written through the paper, is as though this is your discovery.  I am sure that you knew the approximate answers before you began this modeling, so please recognize the literature and your place in it.

(2) This paper purports to show new results, yet it heavily references many of the approaches and results in a UK government report listed as Warwick et all. 2022.  It seems as though much of the current work appeared already in the 2022 report.  Is this double publication?  I have great sympathy with authors who may have been required to deliver the government report, but then the referencing to it should be much clearer than it is now, and it should note that that 75-page report contains material that is reproduced in the standard scientific literature here.  I greatly favor this work coming out in ACP, but make it clear to the readers.

(3) Abstract & Methodology. "We derive a new methodology for determining indirect Global Warming Potentials from steady-state simulations …" The authors apparently do not understand the relationship between steady-state perturbations and integrated pulse-emission transients which is clearly spelled out in Fuglestvedt 2010 ("Generally, so-called steady-state [GWPs] are used. They are calculated assuming constant emissions and steady-state conditions. For compounds that are removed by linear processes, this is equivalent to assuming an emission pulse and integrating over the entire decay of the compound (Prather, 1996, 2002).").  The two approaches give identical results, and the scaling factor is the steady-state lifetime of the perturbing species (H2 in this case).  The mistake made by Fuglestvedt was in assuming that a 1-year constant emission of the perturbing gas reached steady state. For NOx and some VOCs, it did.  For CO is did not quite and for H2 it certainly did not.  That was a good catch and those H2 GWPs need to be revisited.  I am not sure that any modern H2 GWPs rely on the shorthand calculations used in Fuglestvedt 2010. 

The equivalence between steady-state and integrated transient is exact, but the steady-state must

include all species.  To avoid a 100-year calculation to get CH4 in steady-state, most modelers force a steady state in a few years (CO time scale, and maybe O3 in the stratosphere) with H2 and CH4 fixed/perturbed at the lower boundary.  We then use the change in budget, along with feedback factors to project the CH4 and H2 in steady state.  This could/should have been estimated for the 1-year emission runs in F 2010.  Adding transients to the short-term 1-year steady-state is questionable.  I am not sure it gives the correct true steady-state answer or the correct integration of the transient following a single, annual pulse.  The derivation here is not convincing as the perturbation must be decomposed into all the chemical modes it excites.

Methodology:  L345.  Correct, we agree it is the budget lifetime of the emitted/perturbing species that determines the scaling.  However, the perturbation timescale of the secondary species such as CH4 determines the amplitude of that perturbation.

L 353:  This is incorrect. If you have a steady state then you should NOT be using the subsequent decay.  Period.  Look up the steady-state equivalence derivation again (i.e., CH4 must also be in steady-state.  If you simply have a pulse emission (even over 1 year), then you must follow ALL transients. 

L 357-375: If you want to just follow the transients, then the time constants you have here are wrong.  The 'atmospheric lifetime' for H2 and CH4 are the budget times: Burden/Loss.  But, the decay times here are the chemical mode times (need an eigenvalue decomposition or else an estimate of the perturbation time scale from the feedbacks) and are longer than the lifetime.  You do use 'perturbation time' in Figure 5, but not in these equations?  You know that H2 changes OH, and so it also has a perturbation time (given the soil sink, this is much closer to the standard lifetime).

L 425: the mass-weighted OH mean is a truly terrible metric for anything.  It simply does not describe the CH4 or H2 loss.  You should be calculating the OH weighted by exp(-1775/T) and the CH4 abundance

L 427:  Your CH4 feedback factor here is very high, I suspect that this is calculated from only the tropospheric OH loss.  You need to add in the stratospheric loss and soil sink (which have no feedbacks).  This Stevenson derivation is really awkward and not the original one used in the SAR and TAR.  Using the dCH4 perturbation, one calculates the change in loss of CH4:

            sensitivity = d ln(L-CH4) / d ln(CH4) , ie, for a 10% CH4  increase at the lower boundary, the denominator is ln(1.1).  This sensitivity is < 1 because OH decreases with CH4 (or H2) increases.  The feedback factor is then formally ff = 1/(1 – sens).

For CH4 and H2, the L-CH4  = trop-OH loss + soil loss + strat loss.

Minor comments:

L39: please go back to the original references here.

L175:  Very good point about the shift in fossil fuel related SLCFs.  Nicely done.  This Life cycle assessment is needed.

L186: As above, we really do not care about the mass-weighted OH change, it does not tell us about CH4 loss or even CO loss.

Figure 1: Plot the blue dots on top of the orange to see the fit?

L204: As above, does the CH4 lifetime include OH or also strat and soil, otherwise 8.5 is VERY short! as it implies a total lifetime of 7.5 years vs 9.

How does strat O3 change in response to the H2O affect photolysis and trop O3?

L260 why not also Fig 1c?  blue vs red

L271-283:  This whole section on calculating the impact of OH changes on aerosol formation is so model-dependent that the results and even the sign of the results are simply not robust.  I have no doubt that the authors tried to simulate aerosols in their model and these are their results.  But, let us face the facts:  the processes that drive these aerosol results occur on space and time scales that are in no way resolved with any global model.  The processing of aerosols and SO2 as in liquid clouds in the boundary layer, or in being transported out of the boundary layer is complex and I doubt that many models have the same mechanisms or agree on how to model this.  If given the same task (increase CH4, which changes OH and HO2 and O3), a MIP were to reproduce these results, I would be shocked.

For this model result, there are several problems already.  Where are the other aerosols and how do they respond (nitrates? SOAs?).  One of the competition for SO2 is surface deposition and that is long known to compete with OH oxidation to SO3.  This depends on boundary layer mixing not discussed here.  Further, the SO3 probably attaches to existing aerosols in the polluted boundary layer and does not form new particles.  The analysis here argues about mass-weighted OH changes, but it is really only the boundary layer that matters here.  Once in the free troposphere, all of SO2 will probably react with OH since it is unlikely to be pulled into a cloud. 

The chemistry discussion here is confusing:  "The increase in HO2 outweighs the decrease in OH" makes no sense.  HO2 >> OH and so the HO2:OH ratio is driven by the decrease in OH, not the increase in HO2. e.g., a ratio of 1000:10 goes to 1001:9

I do not doubt the results for the UKESM1 model in O'Connor et al (2022) but I doubt that any other model, given the complexity of competing forces, would give the same or close to the same answer.

L284ff:  Since the UKESM is being run here as an atmosphere model with SSTs, it generates it own weather and cloud systems.  The weather/clouds are chaotic and not simply determined by the SSTs.  Thus with aerosol and heating rate perturbations, how many decades did you have to run to get a robust signal for things like CRE?  Is this whole section on aerosol-cloud forcing meaningful? It is not convincing to me.

L300ff:  It would be nice to have tabulated need individual components of ERF and other chemical changes for others to compare with.

General Comments

This paper reports important results on the climate impacts of hydrogen leakage to the atmosphere, using a set of experiments with the UKESM1 model. The study is well designed and presented, and will be of wide interest to both scientists working on this problem and policymakers considering the pros and cons of using hydrogen as a future fuel.

I have a few queries and suggestions for clarification and improvement (details below). Once these are addressed, I am happy to recommend publication.

Specific Comments

Abstract. A key result not mentioned in the abstract is the large cloud adjustment found in the ERF calculations. Whilst uncertain, this result seems potentially very important, as it increases the climate impacts by about 50%, and I think should be included here. I also think the net climate impact (including CO₂ reductions associated with the uptake of a H₂ economy) should also be clearly stated in the abstract, even though these are only done simply here, and not a crucial part of the study.

Figure 1. NB the figure has capital A, B, C, D, but the caption has lower case (also other figures). Suggestion for Figure 1d: the large deviations of water vapor between scenarios below about 15 km (i.e. partly from the troposphere) are not useful, so why not start the profiles a bit higher (~15 km)? Also, it would be interesting to see the water vapor profiles for the simulations with adjusted methane (and O₃ precursors) as well as those shown, since the simulations with adjusted methane are the ones that show the most complete overall impacts of H₂ emissions.

L205. Is the methane lifetime of 8.5 years a total lifetime (i.e. with respect to OH, Cl, soils and stratospheric losses; Prather et al., 2012) or just with respect to OH (and if the latter, is it both tropospheric and stratospheric OH)? Stevenson et al. (2013) Table 7 reports values for total lifetime.

L253. Interestingly (and the opposite of what you say), Figure 2c (750 ppm H₂) actually shows a small net decrease in ozone (this can also be seen in Fig. 1c and Table S1). Is this a significant result (one would expect an increase, based on the other experiments), or is this just illustrating the level of meteorological noise in the simulations?

L254. Do you mean “compare Figure 2b with Figure 2f” (rather than 2a)?

L255. I would argue that Figures 2h and 2j are the relevant figures to compare with Figure 2b, and not 2g and 2i, for the context described, since 2g and 2i have fixed methane. (Similar comment on Figure 2 caption, L275.)

L286 OH₂ should be HO₂

L300 “change in” should be “amount of”?

L305 I think it is worth stressing that although these ERF adjustments are apparently only available from UKESM, they do suggest potentially important climate effects via cloud adjustments, and it is important that other modelling groups look at this. The Cloud Radiative Effect seems to add about an extra 50% to the RFs.

Section 3.3 and Figure 3. Is it important to stress here that the ERFs calculated do not include changes in CO₂ emissions (but do include changes in CH₄ emissions and all O₃ precursors, as well as H₂)? Later, you do discuss the overall impacts of the scenarios in the context of CO₂ emissions, pointing out the overwhelmingly beneficial outcome in terms of impact on CO₂/climate of moving to a hydrogen economy (at least a green hydrogen economy). But this should perhaps be absolutely clear in this section too. (It is all too easy for some policymakers to get the wrong end of the stick about H₂ and just see a positive ERF or large GWP value and conclude it is a major problem).

L334 A Global Warming Potential (GWP) with a time horizon of zero years is indeterminable and is meaningless. This is perhaps more a criticism of Ocko and Hamburg (2022), but if this work wants to refer to it, it should warn readers that it is meaningless. The “GWP” calculated in Ocko and Hamburg with a constant emission rate is not the same metric as a GWP (their Figure 3b incorrectly labels it as a GWP), and that should also be clarified, as it is another important source of confusion in the literature.

Section 3.4.1 and Figure 5. The formulae for the GWP calculations is certainly complex and opaque (and hence difficult to follow and check), but I have every confidence that the authors have done this correctly, and is probably the best that can be done given the experimental set-up. Some of my confidence comes from a comparison that I feel is worth adding, of Figure 5 and Derwent et al. (2020) Figure 1, which is an equivalent result from a different model that does not constrain methane concentrations with a lower boundary condition. Figure 5 shows a peak response of methane of about 15 ppb for a H₂ pulse of 1500 ppb (i.e. ratio of CH₄ response to H₂ added of ~1/100), with the peak about 5 years after the pulse. Derwent et al. added a pulse of size 1.67 Tg (~6 ppb) H₂ to a model with free-to-respond methane – the response is shown in their Figure 1: a peak methane response of ~0.08 ppb (i.e. ratio of ~1/75) after about 3.5 years. So the response represented by the complex equations is similar to that found in a full model simulation (albeit a model with different set-up and different parameters such as H₂ and CH₄ lifetime, etc.).

L497 (and elsewhere). IPCC recommends referring to individual chapters by their lead authors, rather than to the whole reports.

L498. The methane feedback factor for UKESM1 of 1.49 is suspiciously large, and is actually outside the range of models quoted in Thornhill et al. (2021) (the individual models have values of 1.32, 1.31, 1.43, 1.30, 1.26 and 1.19 – and the first one of those is for UKESM1). This makes me think it may have been incorrectly calculated. It is conceivable that the set-up of UKESM1 used here is sufficiently different to change this factor compared to that used in the analysis of Thornhill et al., but that seems unlikely. How important is this feedback factor for the calculated GWP?

L509. 2104 -> 2014.