Evaluation and Analysis of the Seasonal Cycle and Variability of the Trend from GOSAT Methane Retrievals
, Hannakaisa Lindqvist 1, Janne Hakkarainen 1, Marko Laine 1, Ralf Sussmann 2, Aki Tsuruta 1, Rob Detmers 3, Nicholas M. Deutscher 4, Edward J. Dlugokencky 5, Frank Hase 6, Otto Hasekamp 3, Rigel Kivi 7, Isamu Morino 8, Justus Notholt 9, David F. Pollard 10, Coleen Roehl 11, Matthias Schneider 6, Mahesh Kumar Sha 12, Voltaire A. Velazco 4, Thorsten Warneke 9, Debra Wunch 13, Yukio Yoshida 8 and add
Show full author list
Round 1
Reviewer 1 Report
his paper presents the seasonal cycle and the trend of XCH4 derived from three different retrieval algorithms for GOSAT, namely NIES, RemoTeC full physics, and RemoTeC proxy algorithms. The seasonal cycle and the trend were derived at 15 TCCON sites to compare them with the TCCON data. In addition, zonal mean of the trend of XCH4 was also compared with the NOAA MBL reference.
Although this paper contains interesting results which merit publication with some minor corrections, one can think of a different methodology to approach the problem of the global methane growth. The comments below are my personal and intuitive opinion, and I do not insist that the paper should be rewritten accordingly.
The authors used bias-corrected XCH4 data provided by the two RemoTeC products, On the other hand, in my understanding, the NIES product version 2.72 is not bias-corrected. Why did the authors mix bias-corrected products with raw products?
A related question is that why the authors used the bias-corrected XCH4 of RemoTeC products. TCCON data were used in correcting biases of the RemoTeC products. Therefore, a comparison with TCCON data can be a self-consistency check, but might not be an independent validation.
The authors described the growth of XCH4 as a local trend + constant seasonal cycle. In this description, the seasonal cycle can be deformed by an anomalous event such as that reported by Ishizawa et al. (2016). The deformed seasonal cycle may then cause apparent anomalous local trends when there is no anomalous event. The description of a global trend + variable seasonal cycle seems to be more appropriate, because an anomalous event would be identified as an anomaly in the seasonal cycle.
Detailed comments are listed below:
Page 4, line 103:
bias correction
Page 4, line 119
‘vBecause’ should be ‘Because’.
Page 6, line 190:
“Figure 2 shows” seems to be a typo of “Figure 3 shows”.
Page 7, lines 218-219:
The authors state that they calculated daily averages for all co-located GOSAT and TCCON retrievals. In page 6, lines 191-192, however, the authors state that they considered all TCCON soundings within +-1 hour from the GOSAT sounding, which confuses me. Please give an explicit explanation for the daily averages.
Page 15, lines 425-430:
More detailed discussion on the RMS error is desired. I guess that one of the reason for the large RMS error at Tsukuba and Saga might by that there is little co-located GOSAT retrievals during the summer, suggesting that the accuracy of the dynamical co-location is limited.
Page 17, lines 514-515:
What causes the displacement? Australia?
Author Response
The point-by-point responses can be found from the attachments.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
General comment:
This paper is an interesting study of the trends of seasonal cycles of XCH4, as seen from various retrieval products. It would deserve some English-proofreading, but is rather well written. It is original, dense and provides a large wealth of information.
Major comments :
1- The approach relies on a statistical algorithm called Dynamic Linear Model, as described in Appendix B. It has alreay been applied in other contexts, but I could not understand how the model statistical parameters hve been assigned (detail is supposed to be given in a github repository, but I got lost in this site). I would think that they are quite uncertain, with correlated errors among them (i.e. the seasonal cycle term cannot be clearly separated from the trend term and the autoregressive term). I would also think that they depend on the estimated retrieval errors, that may be wrong.
2- Related to this question, is the comparability of the statistical results using different retrieval products. The text already notes a sensitivity between TCCON and GOSAT averaging kernel correction. But depending on their own statistical hypotheses, there may also be a sensitivity between averaging kernel corrections from different GOSAT retrievals. The fact that the theoretical framework (optimal estimation, Tikhnov, ...) differs among the algorithms does not help. Also note that the assigned retrieval errors may be differently wrong among the algorithms. Similarly, the suggestion made by the authors to use ensemble of retrievals (l. 476 and 578) looks like a free lunch.
3- The authors claim that TCCON is independent from the satellite retrievals (l. 125). This helps their reasoning, but has TCCON been used at all in the satellite retrieval bias-correction, or in the tuning of the retrieval uncertainty estimate? I would doubt it. In addition, some quantative numbers about the reliability of the TCCON calibration would be useful, so that we know what differences to TCCON are significant.
Minor comments:
1- l. 483: the authors note the impact of boundary layer variations at background sites, but I would think that it is marginal.
2- l. 579: given the uncertainty in the OH profiles, I would think that models would not help here.
Author Response
The point-by-point responses can be found from the attachments.
Author Response File: Author Response.pdf
Reviewer 3 Report
A very important paper that needs to be published.
Author Response
The point-by-point responses can be found from the attachments.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors have addressed my concerns well.
