Towards standardized processing of eddy covariance flux measurements of carbonyl sulfide

Kohonen, Kukka-Maaria; Kolari, Pasi; Kooijmans, Linda M. J.; Chen, Huilin; Seibt, Ulli; Sun, Wu; Mammarella, Ivan

doi:https://doi.org/10.5194/amt-13-3957-2020

Articles | Volume 13, issue 7

https://doi.org/10.5194/amt-13-3957-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/amt-13-3957-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 13, issue 7

Research article

|

22 Jul 2020

Research article |

| 22 Jul 2020

Towards standardized processing of eddy covariance flux measurements of carbonyl sulfide

Kukka-Maaria Kohonen, Pasi Kolari, Linda M. J. Kooijmans, Huilin Chen, Ulli Seibt, Wu Sun, and Ivan Mammarella

Download

Final revised paper (published on 22 Jul 2020)
Supplement to the final revised paper
Preprint (discussion started on 07 Oct 2019)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC1: 'Review of amt-2019-313', Georg Wohlfahrt, 31 Oct 2019
- AC1: 'Final response to Reviewer 1 comments', Kukka-Maaria Kohonen, 28 Feb 2020
RC2: 'Review of Erkkilä (Kohonen) et al 2019', Anonymous Referee #3, 05 Nov 2019
- AC2: 'Final response to Reviewer 2 comments', Kukka-Maaria Kohonen, 28 Feb 2020

Peer-review completion

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

AR by Kukka-Maaria Kohonen on behalf of the Authors (27 Mar 2020) Author's response Manuscript

ED: Referee Nomination & Report Request started (02 Apr 2020) by Christof Ammann

RR by Georg Wohlfahrt (05 Apr 2020)

Suggestions for revision or reasons for rejection

Review by Georg Wohlfahrt:

General comments:
It is great to see that the authors have ironed out most of the issues that I addressed in my review of the first version of this manuscript. I think the second version is much better and actually lives up to what the title suggests and makes a significant contribution to the literature. There are however two more major issues, in addition to a few minor ones, that need to be fixed before I can recommend publication in AMT.
(1) Cumulative sums with and without gap-filling: This is something that I have commented on before – it is misleading in my view to report cumulative sums over some period if there are gaps in the data. This is why need gap-filling! I am trying to illustrate my point with an example that we have a better feeling for than COS flux, namely precipitation: Average annual precipitation at Hyttiälä is around 700 mm – if due to data gaps the sum would be only 350 mm – would you think it is useful to report that number? What do we learn from that number apart from that a continuous record is required for deriving annual sums as otherwise reported numbers are biased? What the authors could do though is to compare the mean or median of the non-gap-filled and gap-filled data – this would tell them something about what gap-filling does to the distribution of the data, e.g. if it is mostly nighttime data that are missing than I would expect the mean/median to increase (i.e. become less negative).
(2) Gap-filling: This section has been considerably expanded, yielding interesting insights, but also a few concerns. First, I think the current presentation of these results is way too qualitative – we need hard facts on how the various gap-filling algorithms compare, i.e. the usual statistics and when comparing algorithms with differing numbers of parameters a metric that takes this into account needs to be used, e.g. the AIC. Along these lines, Fig. 7 would also profit from showing the uncertainty/range of the measured/simulated values. Second, Fig. 7 reveals a substantial diurnal bias, especially during the night and the first half of the day when the three algorithms grossly overestimate COS uptake (at 05:00 it looks like measured uptake is -11 pmol/m2s, while all three algorithms calculate an uptake of around -18 pmol/m2s), which interestingly does not show up in Fig. S9a which reports the largest bias at the highest radiation values.
Given the form of the algorithm:
Fcos = a*I/(I+b) + c*D + d
it would appear to me that a much better fit could be obtained by choosing a less negative d, which causes a parallel upward shift of the simulated values, and by decreasing the value of b in exchange in order to increase the response to radiation. Overall, the overestimation is less prominent during the afternoon, in particular when RH/VPD is included, which suggests this to be a behavior resulting from asymmetry in stomatal behavior, discussed already in Kooijmans et al. (2019) – the addition of RH/VPD however appears unable to fully account for this. I think the authors should try to improve the diurnal bias and at least discuss it.

Detailed comments:
l. 13: what does „normally“ mean?
l. 14: isn’t this a repetition from above?
Table 1: what does the “-“ mean – that the information is not provided? Please note that Spielmann et al. (2019) used the approach as outlined in Gerdel et al. (2017) and thus corrected for storage by assuming the concentration changes at the EC height to representative of all heights below the EC sensors.
l. 90: “acquired by” instead of “gathered to”
l. 127: processing of what?
l. 227: it is not so much of a “drop”, but rather a “decrease”
l. 293: “accounts for” instead of “smooths”
l. 352: here and on many other accounts – be careful when using low/high or increase/decrease with negative fluxes as often you get the direction wrong, as in this line; maybe resort to saying things like less/more negative and alike
l. 412: what does “normally” mean?
l. 467-468: I accept that the authors want to show the magnitude of the frequency response corrections and thus report fluxes without these, BUT they should not mix up the difference between corrected and uncorrected values with the difference between the two spectral correction approaches, which they do in this sentence! In the first part you discuss differences between different lag time approaches, while in the second part the difference between uncorrected and corrected fluxes. I think you should report the difference between the two spectral corrections in the second part of the sentence, which is just 2.7 % and thus relatively minor.

Hide

RR by Anonymous Referee #3 (21 Apr 2020)

Suggestions for revision or reasons for rejection

Kohonen et al have made substantial changes to their manuscript to address the reviewer concerns, and I think the paper is much improved. It will make a useful addition to the literature. I have only a few remaining suggestions, noted below.

1. The bolding of recommended options in Fig. 1, while a good idea, seems a bit subtle to me. I would like to see also an additional row in Table 1, giving the options recommended by this paper, which would serve as a quick and easy summary.

2. Regarding the flux uncertainties, I thank the authors for clarifying that only a subset of the considered methods were used in the assessment of processing uncertainty. I see now that the manuscript states that briefly in Section 2.4.6, but I think Section 3.7 needs to make it clear as well. Currently, after reading through several subsections about how this or that processing choice impacts the flux, the reader comes to the final Results subsection (3.7) and finds an overall comparison of the relative impacts of processing and random uncertainty on the flux. At this point the reader should be informed that many of the processing choices they have just been learning about are not included, and why.

3. The authors seem to have overlooked my related concern, that the “total uncertainty” defined in Section 2.4.6 (and discussed in Section 3.7 and displayed in Fig. 8) blurs the distinction between stochastic noise, which can be averaged out, and systematic bias, which cannot. As EC flux measurements are always aggregated in some way (e.g. via daily averages, annual sums, regressions, or models), the sum of the random and systematic uncertainties for a single EC flux measurement is not of any practical relevance. If a reader wants to quantify how systematic processing uncertainty will contribute to the overall uncertainty in a result derived from EC data, they must first calculate how the random uncertainty in each individual flux measurement propagates through their method of aggregation (i.e. how it averages down), and then compare the random uncertainty in the aggregate result to the systematic processing uncertainty. The values of random and processing uncertainty presented in the manuscript are useful for that, but their sum and their relative contributions to that sum are not. So the “total flux uncertainty” in the manuscript is at best irrelevant and at worst misleading as to the relative importance of the random and systematic uncertainties to real EC analyses. I would remove it from the paper, and I would also add text calling the reader’s attention to the fact that the random uncertainty reported is for a single flux measurement, and that it will average down in real EC analyses while the systematic processing uncertainty will not.

Hide

ED: Publish subject to minor revisions (review by editor) (27 Apr 2020) by Christof Ammann

AR by Kukka-Maaria Kohonen on behalf of the Authors (07 May 2020) Author's response Manuscript

ED: Publish subject to minor revisions (review by editor) (28 May 2020) by Christof Ammann

In the revised version, the authors have addressed most of the referee comments in an adequate way. However, there are a few minor issues that need to be corrected before the manuscript is acceptable for publication. They are listed in the following.

Best regards
Christof Ammann

----------------------------------
Line 4:
This is syntactically incorrect. Change to "... and we provide a method ..."

Fig. 1: For a better readability of the paper, the frequently used abbreviations for the different processing options (e.g. BA, RF, etc.) should also be indicated in Figure 1.

Line 425-428:
This is not an adequate discussion of the problem in my view. Filtering itself does not lead to a bias but rather the subsequent gap filling. So it needs to be discussed whether the gap filling of calm nights (below the u* threshold) based on data of windy nights (above the u* threshold) leads to a bias in the gap-filled COS fluxes. If a dependency on the concentration gradient is not excluded, it would need to be taken into account in the gap filling procedure.

Line 496-499 (see last comment of referee G. Wohlfahrt):
This corrected formulation is still misleading as it mixes (compares) the variation of different alternative lag methods with the systematic effect of the necessary high frequency correction. The 2.7% variation in high-frequency correction method is much smaller than the 15.9% variation in the lag methods. Therefore, the sentence "The largest differences in the final fluxes came from time lag determination and spectral corrections" is inadequate or misleading, since a flux without high-frequency correction cannot be considered as final.

Reviewer#2, Comment 2 (Line 469):
The author response to this comment was not fully satisfying. The authors should explain why they only used the coordinate rotation and the detrending effect to quantify the processing uncertainty.

Line 510:
Change to: "When averaging fluxes over time, the relative random uncertainty ..."

Line 511:
This sentence is unclear. Change to: "Relative processing uncertainty decreased from single 30 min COS fluxes to daily averages but remained at the same level for longer averaging periods."

Hide

AR by Kukka-Maaria Kohonen on behalf of the Authors (05 Jun 2020) Author's response Manuscript

ED: Publish as is (24 Jun 2020) by Christof Ammann

AR by Kukka-Maaria Kohonen on behalf of the Authors (25 Jun 2020) Manuscript

Short summary

Biosphere–atmosphere gas exchange (flux) measurements of carbonyl sulfide (COS) are becoming popular for estimating biospheric photosynthesis. To compare COS flux measurements across different measurement sites, we need standardized protocols for data processing. We analyze how various data processing steps affect the calculated COS flux and how they differ from carbon dioxide (CO₂) flux processing steps, and we aim to settle on a set of recommended protocols for COS flux calculation.