Machine learning estimates of eddy covariance carbon flux in a scrub in the Mexican highland

Guevara-Escobar, Aurelio; González-Sosa, Enrique; Cervantes-Jiménez, Mónica; Suzán-Azpiri, Humberto; Queijeiro-Bolaños, Mónica Elisa; Carrillo-Ángeles, Israel; Cambrón-Sandoval, Víctor Hugo

doi:https://doi.org/10.5194/bg-18-367-2021

Articles | Volume 18, issue 2

https://doi.org/10.5194/bg-18-367-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/bg-18-367-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 18, issue 2

Research article

|

18 Jan 2021

Research article |

| 18 Jan 2021

Machine learning estimates of eddy covariance carbon flux in a scrub in the Mexican highland

Aurelio Guevara-Escobar, Enrique González-Sosa, Mónica Cervantes-Jiménez, Humberto Suzán-Azpiri, Mónica Elisa Queijeiro-Bolaños, Israel Carrillo-Ángeles, and Víctor Hugo Cambrón-Sandoval

Download

Final revised paper (published on 18 Jan 2021)
Preprint (discussion started on 10 Jan 2020)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'Review', Anonymous Referee #1, 16 Jan 2020
- AC3: 'Author response to Referee #1', Mónica Cervantes-jimenez, 28 Apr 2020
AC1: 'Author response to Referee #1', Mónica Cervantes-jimenez, 20 Feb 2020
RC2: 'Review', Anonymous Referee #2, 28 Feb 2020
- AC2: 'Author response to Referee #2', Mónica Cervantes-jimenez, 23 Mar 2020

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

ED: Reconsider after major revisions (28 Apr 2020) by Paul Stoy

AR by Mónica Cervantes-jimenez on behalf of the Authors (08 Jul 2020) Manuscript

ED: Referee Nomination & Report Request started (16 Jul 2020) by Paul Stoy

RR by Anonymous Referee #2 (22 Jul 2020)

Suggestions for revision or reasons for rejection

Overall, this new version of the paper is much stronger than the last one. The authors have considerably expanded their analysis, now including a comparison of various regression models to predict GPP. The Bernal site now becomes a study site for a study with a much broader impact for the community. The comparison between EC and MODIS is very interesting and makes a compelling case for a more careful use of remotely sensed GPP in semi-arid grasslands.

I do think that the paper could be strengthened by clearly embracing the GPP comparison. Indeed, I find that the title, abstract, and even to some degree the introduction are not fitting this new version of the paper anymore, and might actually do a disfavor to the article, making these new interesting results more difficult to find. I think that rewriting the abstract and introduction to focus on the GPP comparison and thinking of the Bernal site as a study site would, in the end, be a more powerful narrative here, and be a better fit to the results and discussion sections as they are.

I have a few other minor comments below:

Page 1, L16: remotely sensed
L19 despite it being in an overgrazed condition
Page 2, L3: rural-urban migration being an important driver…
Page 3, L13: what’s is “inter-converted”?
Page 4, L11: we hypothesized whether or not…
Page 5 L33: do we know why the shrub was cleared?
Page 5, L35: two verbs in one sentence “The soil has a clay loam texture are Vertisols with abundant sub rounded basaltic stones…”, please rephrase
Page 6, L2: the vegetation corresponds to
P12, L27: Sections 3.1 and 3.2 present
P17, L19: for the Bernal site
P17, L21: gurant? typo
P17, L21-23: However, during… exchange at that time: This sentence is grammatically incorrect and confusing, please rephrase.
P22, L4: predicted GPP
P23, L31: complicates…the calculation?
Section 4.3 (pages 24-25) seems like it really belongs to the Results section, and not the discussion
Section 4.4 looks to me like the “discussion” part of “result” section 4.3
Page 25, L25-26: rephrase this sentence, it’s difficult to understand

Hide

RR by Anonymous Referee #3 (18 Aug 2020)

Suggestions for revision or reasons for rejection

Guevara-Escobar measure eddy covariance fluxes from a scrub ecosystem in Mexico and explore the ability of machine-learning and other models, including MODIS, to predict measured GPP. Many aspects of the analysis are quite rigorous, but the manuscript needs comprehensive improvement to explain why particular analyses were used, why they are important (the Introduction needs to be comprehensively rewritten) and how to interpret results (the Discussion often reads like a qualitative Results section rather than a synthesis of what was learned). Other parts of the manuscript like the abstract and for the most part the Methods were informative and well-written. By focusing on the major ideas and clearly communicating the important parts of the analysis, the manuscript will make an interesting contribution to the literature. I also have some concerns about how GPP was estimated given that the u* threshold seems anomalously low. More details on this would be forthcoming.

The statement on page 1 line 28 could use a reference, I have seen other publications that have put this value at 40%. It could also be combined with the next statement to read '...ecosystems, but almost 2 million km2 (50%) in Mexico, mainly the...'

The first paragraph of the as a whole should be rewritten to be more simple and lead more clearly from the geographic extent of semi-arid and arid ecosystems to their degradation. Breaking the paragraph into 2 paragraphs will help.

Delete this statement, it is unimportant in this context: Substantial biosphere–precipitation feedback is often found in regions that are transitional between energy and water limitation, such as semi-arid or monsoonal regions (Green et al., 2017). The introduction as a whole sounds like a literature review in need of more structuring. Move from general information to the specific problem of measuring GPP in degraded semiarid ecosystems. At the moment there is too much information and it is confusing to read.

The hypothesis is not testable. I recommend restructuring it around the notion of what was explored and why. I see that a previous review suggested a hypothesis, but if it cannot be statistically tested it is not a useful addition.

For the site description, I assume that 20,717 means 20.717, i.e. decimal degrees?

'We assumed that the footprint of the EC was measuring only the patchy scrub vegetation'. This can be tested and non-target-ecosystem measurements removed. Alternately, it may not be important if structures are sufficiently far away. How far away is far away?

Was the u* threshold really 0.033 m/s? This is very low, probably too low, but I'm also confused about the wording that describes this application. I am concerned that many fluxes with insufficient turbulence are being included in the final output, which would cause respiration to be too low and carbon uptake estimates to be too high. GPP estimates would then be too high which may be why the fit in Figure 6 is so poor. How do the GPP estimates compare against other ecosystems? This is a topic for the Results section and I am mostly interested in knowing if the GPP results are defensible rather than a comprehensive comparison of GPP at semi-arid ecosystems.

eq. 1 and 2 (and 3): please use the multiplication sign and not the star in formal equations.

I happen to find the model comparison with Santa Rita interesting: how much variability at a site can be explained by a model from another site?

page 10 line 10: the authors of the R package should be cited.

The first paragraph of the Results should be deleted. Perhaps as a first paragraph of a Discussion section but even then probably unneccesary.

Section 3.4 should come earlier in the results section. Reviewers need to be convinced that fluxes are defensible before modeling is described.

The IVI analysis (Table 3) would be more convincing if it was placed in the context of the flux footprint. Do these numbers come from across the entire site including wind directions that are uncommon?

Section 4.1: how would one interpret these results and why was EML best?

‘is generally considered to be 1 km2’ is incorrect. The footprint dimensions depend on sensor height and the Obukhov length.

Section 4.3 is interesting but discussing water leakiness without measurements is unnecessary.

4.4 call this section ‘overgrazing’ or ‘ecosystem management’ instead. Discussing erosion is outside of the context of the manuscript. The manuscript has too much information, much of it extraneous.

5. Conclusions: reasons why machine learning fit best and what can be learned from this for ecosystem analysis are still lacking.

Hide

ED: Reconsider after major revisions (18 Aug 2020) by Paul Stoy

AR by Mónica Cervantes-jimenez on behalf of the Authors (22 Sep 2020)

ED: Referee Nomination & Report Request started (29 Sep 2020) by Paul Stoy

RR by Anonymous Referee #3 (12 Oct 2020)

ED: Publish subject to technical corrections (15 Nov 2020) by Paul Stoy

AR by Mónica Cervantes-jimenez on behalf of the Authors (22 Nov 2020) Author's response Manuscript

Short summary

All vegetation types can sequester carbon dioxide. We compared ground measurements (eddy covariance) with remotely sensed data (Moderate Resolution Imaging Spectroradiometer, MODIS) and machine learning ensembles of primary production in a semiarid scrub in Mexico. The annual carbon sink for this vegetation type was −283.5 g C m⁻² y⁻¹; MODIS underestimated the primary productivity, and the machine learning modeling was better locally.