The article titled "Satellite-based modeling of wetland methane emissions on a global scale (SatWetCH4 1.0)" presents a pioneering approach to understanding and quantifying methane emissions from wetlands across the globe. The authors have developed an innovative model that leverages satellite-based data to simulate methane emissions with a high degree of spatial and temporal resolution. This study is particularly significant given the substantial contribution of wetlands to global methane emissions and the critical role of methane as a greenhouse gas.

The manuscript is well-structured, with a clear abstract that succinctly summarizes the study's objectives, methods, and findings. Overall, this study represents a significant advancement in the field of wetland methane emissions modeling. It is my pleasure to recommend this paper for publication in GMD journal.

My comments here are relatively minor, but I hope may be useful for the authors to consider.

1. L165-L171: Is the reason for using ERA5's lay2 soil moisture because of the high correlation coefficient between the site data and lay2 soil moisture? In this article (DOI: 10.1126/sciadv.aba2724), the soil moisture used is from 0-289cm depth, and I haven't seen how the interannual variation of soil moisture at other layers correlates with the site observations. As far as I know, the interannual variation of soil moisture in lay1 and lay2 is similar, and lay1 is closer to the surface soil moisture. Why not use the data from lay1?
2. L206-L210: Are you suggesting that the green single-peaked line (Albuhaisi et al., (2023))is more in line with physical laws but the formula is more complex and prone to non-convergence, which is why such a complex approach was not adopted? So, your point is that the methods of Albuhaisi et al., (2023) should be more reasonable, rather than the others?
3. Fig 2: By the way, could you explain why there are three lines in WetCHARTs, and what do they represent, respectively? What is the significance of such classification? Your results seem to be mainly close to the one with Q10=3.
4. L221: I understand that when you were assessing with the site data, you set f_wto 1, and later when generating the product, you utilized WAD2W and TOPMODEL? Or after considering the two different wetland areas, why didn't you compare with this quantity? Was it because you considered the spatial resolution of the gridded data to be too coarse?
5. In Figure 3c, the simulation appears to be quite similar to the observations in the low-value area, but there is a significant difference in the high-value area. Could you briefly explain the reason for this? Is it related to the wetland areaf_w being set to 1?
6. L227-L230: From a statistical relationship perspective, it is true that the correlation coefficient between the tropical region and the site data is low. Could you briefly explain the reason? Since your model uses MODIS observations similar to NPP, there should theoretically be more data in the tropical regions. Is it because the emissions are high in the tropics and the mechanisms of methane production activities are not well understood?
7. In Figure 4, should the time period range of the annual average be indicated in the figure title? Is it 2003-2020? In line 249, it is mentioned that WAD2M is from 2003 to 2020, is TOPMODEL climatological? Please clarify.
8. Line 261: The absolute magnitude of the C_substratevalue is small, and the text provides the corresponding explanation. My question is, after normalization, compared to the other two products (SoilGrids and HWSD), the high-value areas of C_substrate in the region north of 60°N and more northerly, could you briefly explain this reason?
9. L293-294: The overall smaller values of C_substrateare due to the lower values obtained from MODIS. Do you think these comparatively smaller values are reasonable? Please explain.
10. L356-358: Why did you previously mention that the global distribution is related to the distribution of wetland areas, yet the wetland distributions in the Southern Hemisphere are opposite for the two, yet the methane emission fluxes from wetlands are similar? In Equation 1, temperature and wetland area are directly proportional to methane emission fluxes; why is temperature considered the dominant parameter in the Southern Hemisphere?

Review of Satellite-based modeling of wetland methane emissions on a global scale (SatWetCH4 1.0). Bernard et al.

The authors use surface eddy covariance data for wetland methane fluxes from the global FLUXNET-CH4 Community Product v1.0 to calibrate an empirical model that is applied to produce a globally upscaled wetland methane emissions product. The study is at the forefront of global wetland methane modeling and responds to the scientific imperative to understand the global methane budget in a time of rapid change (accelerating growth of atmospheric methane concentrations since 2014). With revisions focused on reconciling findings of the author's study (SatWetCH4 1.0) with the first and only published global wetland methane upscaling product (UpCH4 v1.0; McNicol et al. 2023) and providing more justification of the model calibration approach, the study is likely to become an important contribution to wetland methane upscaling science. 

McNicol, G., Fluet-Chouinard, E., Ouyang, Z., Knox, S., Zhang, Z., Aalto, T., et al. (2023). Upscaling wetland methane emissions from the FLUXNET‐CH4 eddy covariance network (UpCH4 v1.0): Model development, network assessment, and budget comparison. AGU Advances, 4(5). https://doi.org/10.1029/2023av000956

I led the analysis and writing of the study referenced above which was published in October 2023. I am aware of the ethical considerations regarding promotion of one's own work during the review process, but I would like to explain why I consider referencing of my study as not merely justified, but essential.

Given the lack of a true global wetland methane emissions benchmark dataset, our understanding advances via careful model/product inter-comparison. The absence of a comparison to our 2023 study greatly reduces the potential insights that could be gained. For instance, Bernard et al. estimate a global annual total wetland methane source of around 70-86 TgCH4 y-1. This is remarkably low; much lower than the spread of bottom up and top down models within the Global Carbon Project methane budget ensemble or WetCHARTS. I think the estimates may be different due to Bernard's inclusion of an explicit substrate availability term in the empirical function, over the temperature-dominated learned function in McNicol et al., or the inflexibility of the temperature response term (see section below), but this needs to be explored in some detail by the authors before a clear insight can be gained into the pros and cons of using this substrate dependent formulation.

Another key distinction that would be valuable to explore via inter-comparison with the UpCH4 product is the choice to use an empirically defined model itself, rather than a purely data-driven (random forest) model as in UpCH4. As a community we are being encouraged to combine our ecosystem science knowledge of good process representation with a full utilization of flux observations, enabled with machine learning. I imagine this study could become a useful baseline study of a purely empirical model, which arrives at one calibrated parameter set (one model) from optimization on all ground surface methane flux data, much as UpCH4 is intended as a baseline for a purely data driven model, which optimizes on the same dataset to identify, in effect, an ensemble of highly conditional predictive models. Neither method is anywhere close to perfect, and future advances are most likely in uniting the two, yet Bernard et al. make no mention of this modeling issue within their limitations section. The behavior of the purely data driven models is of particular concern during expansive extrapolations where models may behave in an unconstrained (by data) manner, yet in McNicol et al. 2023, we arrived at a more plausible global total of 146 TgCH4 y-1. A discussion of model equifinality and extrapolation is entirely absent and would b readily facilitated by comparison to our published study.

A more specific concern not considered by the present study related to this issue of empirical model calibration is that recent work has demonstrated that the temperate dependency of methane flux varies in space and time, such that a single temperature dependency is almost certainly an erroneous model framework assumption (Chang et al. 2021; Yuan et al. 2024). This is not addressed at all in the present study, nor are these two recent and high profile paper cited.

Chang, K.-Y., Riley, W. J., Knox, S. H., Jackson, R. B., McNicol, G., Poulter, B., et al. (2021). Substantial hysteresis in emergent temperature sensitivity of global wetland CH4 emissions. Nature Communications, 12(1), 2266. https://doi.org/10.1038/s41467-021-22452-1

Yuan, K., Li, F., McNicol, G., Chen, M., Hoyt, A., Knox, S., et al. (2024). Boreal-Arctic wetland methane emissions modulated by warming and vegetation activity. Nature Climate Change, 14(3), 282–288. https://doi.org/10.1038/s41558-024-01933-3

I am also concerned that the concept, structural elements, and visualization choices are remarkably similar to our 2023 study, despite no citation being present. In addition to our project leads, our study included over 50 co-authors to honor the terms and spirit of a data policy agreement designed to provide fair credit to a large and growing international community of eddy covariance scientists. I was heavily involved in the acquisition of data in FLUXNET-CH4 v1.0 and a common concern voiced by international investigators outside Europe and North America, whose contributions would do much to address the data and community biases present across the global contributor network, was that their data may be used in high profile global synthesis studies for which they would not receive fair credit. While Bernard et al. do cite Delwiche et al. 2021, which is the dataset release paper, citing our study would ensure that the spirit of the data policy of the original FLUXNET-CH4 synthesis was not undermined by omission on downstream FLUXNET-CH4 upscaling work. The authors should also include, as indicated on the FLUXNET website, the second of the following to CC-BY data attribution requirements to include the DOIs of the sites contributing data:

Data use should follow these attribution guidelines for CC-BY-4.0:

• Cite the data-collection paper, for example, for FLUXNET2015, cite Pastorello et al. 2020¹
• List each site used by its FLUXNET ID and/or per-site DOIs in the paper (these DOIs are provided with download)

Fluxnet.org/data/data-policy/ accessed July 6, 2024

​

Juliette Bernard et al. developed a simple empirical model (SatWetCH4) to simulate global wetland methane emissions by leveraging large-scale remote sensing data. This model addresses uncertainties in wetland methane emissions and proposes a new approach for estimating substrate availability using MODIS data. Calibrated with eddy covariance flux data, the model aims to provide a simpler, faster alternative to more complex Land Surface Models. While I appreciate the authors' development of this new model and its contribution to global wetland CH4 monitoring, I have several major comments and concerns:

Major Issues:

1. Novelty and Approach:

One of the novel aspects of this paper is directly modeling large-scale substrate availability for methanogens. It uses NPP and soil organic carbon turnover (Eqn. 2) to estimate the carbon substrate mass balance. However, I have two questions regarding this approach:

Methanogens use CO2 or acetate as substrates. What exactly does C_substrate represent? Is there evidence that the modeled C_substrate correlates with actual substrates for methanogens, making it a valid proxy variable? In other remote-sensing-based approaches (e.g., Bloom’s WetCHARTS), soil respiration rate (CO2 flux) is used as a proxy for substrate availability.

How is C_substrate validated at sites in terms of its control over wetland CH4 emissions? Does eddy covariance site data show a strong relationship?

Validation Concerns:

The validation of C_substrate (Figure 5) is not convincing for two reasons. SoilGrids and HWSD provide benchmarks for upland soil carbon stock but not wetland C_substrate. Additionally, HWSD/SoilGrids provide total soil carbon stock, which does not necessarily turn over quickly. However, the defined residence time of C_substrate is less than 5.5 years (section 2.1).

2. Q10 Parameter:

Q10 is a key parameter in the proposed modeling approach. However, Q10 is a complicated variable for wetland methane emissions due to different temperature sensitivities for various methanogens and methanotrophs. The emergent Q10 has been found to be small when constrained by satellite CH4 concentrations and inversions (Shuang Ma et al., 2021). This work uses a simply calibrated value of Q10 (2.99). I suggest a more in-depth discussion of the high Q10 value in existing literature, including reasons for discrepancies, implications, and biogeochemical processes.

3. Model Simplicity and Missing Processes:

While I appreciate the simplicity of the model equation, capturing the major dynamics of wetland CH4 emissions, some important processes are not represented. Vegetation phenology (Carole Helfter et al.), which has remote sensing data available (EVI or NDVI), and atmospheric pressure, which controls the bubbling processes of wetland CH4, are significant. Including or at least discussing these variables/processes in the next version would be valuable.

4. Validation Performance:

Related to comment 4, the performance of SatWetCH4 at validation sites (Figure 3) is not satisfactory. This suggests that temperature and C_substrate alone may not be sufficient to capture the observed CH4 emission variability at the scale of eddy covariance sites. More effective calibration or the inclusion of missing dominant control variables in Eq. 1 may be necessary.

5. Global Emission Estimates:

SatWetCH4 extrapolates site parameters to global wetlands. When compared with other products (GCP or WETCHARTS), SatWetCH4’s emissions appear significantly lower, particularly tropical emissions throughout the year and boreal/arctic emissions in June/July/August. This may be due to SatWetCH4’s low bias at sites with high emissions (Figure 3C).

6. Summary

In summary, while I appreciate the authors' attempt to model global wetland CH4 emissions, I must point out that major variables and processes are missing in SatWetCH4. The site-level model calibration is not effective, and the global estimate of wetland CH4 emissions is significantly lower than values in the literature. I look forward to seeing an improved version during revision.