This study builds a 0.1° × 0.1° daily XCH₄ product covering 2020–2023. The authors do this by bias-correcting GOSAT-2 with TCCON as reference, then bias-correcting GOSAT and TROPOMI with the TCCON-informed GOSAT-2 product as reference, and then finally filling each daily 0.1° × 0.1° grid cell with the bias-corrected data from the three sensors, giving priority to GOSAT-2, then TROPOMI, then GOSAT. This fused product performs well against withheld TCCON data and provides additional coverage past that of TROPOMI in challenging retrieval environments. The presentation quality is excellent, but I encourage the authors to consider the following comments at their discretion to improve the scientific quality and significance.

General comments

1. While TCCON is an excellent reference dataset, there are downsides to using it as the basis of your bias correction (e.g., limited independent validation data remains). Do the GOSAT-2 and TCCON co-locations represent diverse enough conditions (e.g., in surface albedo) to justify its use as reference? Figure 6 shows a small range of surface albedos relative to the global distribution (cf. Figure 6 in Balasus et al., 2023). In the case of TROPOMI, a small-area approximation can be used to generate data with more diverse retrieval conditions for reference purposes, either training or validation (Lorente et al., 2021).
2. The authors have focused mostly on the delivery of the 0.1° × 0.1° product as the outcome of the study (as opposed to, for example, what their bias corrections tell us about shortcomings in any of the retrievals). Could the authors suggest some studies that could benefit from their data? In many cases, XCH₄ data alone is insufficient. In inverse modeling studies using chemical transport models, users would also need e.g. averaging kernel, prior profile, and pressure grid information (and might want data from all available sensors, not just the best one, in order to reduce random errors), though inverse modeling studies are not the only application of XCH₄ data.
3. Have the authors found it necessary to account for differences in prior profiles and vertical sensitivities of the different instruments (TCCON, TROPOMI, GOSAT, GOSAT-2) when making comparisons?

Specific comments

Line 39: misspelled reference?

Line 46: consider defining XCH₄

Line 61: Lorente et al. (2021) uses a small-area approximation, not TCCON

Line 99: please check the 30 times number (cf. Table 1 in Jacob et al., 2022)

Line 224: extra “and” at the end of the sentence

Line 253: specify northern-hemisphere/boreal summer and autumn

Figure 4: metrics (e.g. RMSE in subplots c,f,h) look much more optimistic than the LOSOCV column of Table S6

Figure 5: why not leave one season out in your cross validation if you are going to plot like this?

SI: double-check (e.g., no bolds in Tables S6 and S7, no Table S8, etc.)

Line 525: Jacob et al. (2022) no longer is in Discussion

Review of “Enhanced methane monitoring: A globally harmonized daily 0.1° XCH4 through machine learning-based fusion of GOSAT, GOSAT-2, and TROPOMI” by Keya et al.



General comment:



First of all I want to thank the authors for this well-structured and clear and concise article. In terms of grammar, I have nothing to add. However, I do have three important issues that need to be addressed.

Specific points:



The first is the issue of proper accreditation. Each of the individual TCCON datasets used, should have their data reference added to the references list (See https://tccon-wiki.caltech.edu/Main/DataLicense for TCCON citation guidelines). These individual per station citations can be found on https://tccondata.org/.



The second point pertains to the selection of GOSAT-2 as the standard to which the other satellite products are bias-corrected (step 2 in the overall process). This selection of GOSAT-2 is based on the results listed in Table-3. However, it is not clearly stated if the common data sample on which Table 3 is based, is the training sample or instead comes from the LOSOCV approach. If the first, I consider this to be a weak basis for selection, if the latter it is a stronger one as we are fundamentally interested in the performance of the bias corrected products outside the scope of the training dataset. That said, we also need to take the global distribution of the TCCON network into account, which under-samples large swaths of the globe. In that view it is hard to state, with confidence, based on the analysis performed here, that GOSAT-2 should be taken as the definitive reference. I would very much prefer it if the authors performed 3 different step2 analysis wherein in turn, GOSAT, GOSAT-2 and TROPOMI are taken as a reference. This will allow the authors to perform an intercomparison, assess the impact of this choice on a global scale, identifying regions where things converge and diverge, and make a more thorough determination on whether all 3 of these end products turn out to be valid candidates or that one is superior.

The third point addresses the rank order merging method used in step three, where each 0.1° daily grid is filled with GOSAT-2 if available, then TROPOMI if available, then GOSAT. If the ML harmonization between the satellite products performed in step 2 is successful, I see no reason why this method is superior to simply taking the median of all products in the daily grid cell. If the ML harmonization is unsuccessful, then clearly the rank order creation of the merged dataset isn’t the solution either.



minor points:

Line 164: Sha et al. 2021 is used as a source for the used validation colocation criteria. Note however that Sha et al. consider the line of sight of the FTIR instrument. To quote the paper:“An effective location of the FTIR measurement on the line of sight (i.e. at a 5 km altitude) is used to do the co-location”. This should be acknowledged.

Line 241: a space is missing between "conditions." and "Given".

Line 295: please repeat that we are building a 0.1° daily product here.

Line 308: I would not describe Xianghe as a rural site (it sits within 100 km of Beijing Centre) in a heavily industrialized and urbanized region. There is probably a mix-up with Edwards (described as urban), which 100 km radius touches the outskirts of Los Angeles but in and of itself is situated in the desert.

Paragraph 336 onwards: Here we see an improvement of the fusion product compared to TROPOMI when comparing to GOSAT-2, but all components within the fusion product are LM bias corrected towards GOSAT-2. It is thus basically telling us the same as Figure 7.

Paragraph 363 onwards: How exactly was the growth rate calculated. Also, no uncertainty values on the growth rate are given.

Line: 580: This is a AMTD reference, replace by its non-discussions final paper

Sha, M. K., Langerock, B., Blavier, J.-F. L., Blumenstock, T., Borsdorff, T., Buschmann, M., Dehn, A., De Mazière, M., Deutscher, N. M., Feist, D. G., García, O. E., Griffith, D. W. T., Grutter, M., Hannigan, J. W., Hase, F., Heikkinen, P., Hermans, C., Iraci, L. T., Jeseck, P., Jones, N., Kivi, R., Kumps, N., Landgraf, J., Lorente, A., Mahieu, E., Makarova, M. V., Mellqvist, J., Metzger, J.-M., Morino, I., Nagahama, T., Notholt, J., Ohyama, H., Ortega, I., Palm, M., Petri, C., Pollard, D. F., Rettinger, M., Robinson, J., Roche, S., Roehl, C. M., Röhling, A. N., Rousogenous, C., Schneider, M., Shiomi, K., Smale, D., Stremme, W., Strong, K., Sussmann, R., Té, Y., Uchino, O., Velazco, V. A., Vigouroux, C., Vrekoussis, M., Wang, P., Warneke, T., Wizenberg, T., Wunch, D., Yamanouchi, S., Yang, Y., and Zhou, M.: Validation of methane and carbon monoxide from Sentinel-5 Precursor using TCCON and NDACC-IRWG stations, Atmos. Meas. Tech., 14, 6249–6304, https://doi.org/10.5194/amt-14-6249-2021, 2021.