Dear David Sayres et al.,

Both referees were very critical on the data processing procedure used to calculate the fluxes. Thus the paper will be reconsidered after major revision.

The data should be reprocessed by a flux calculation method that does not discard low frequency contributions, as suggested by the referees.

Also, the method to relate the fluxes to surface characteristic could be improved, as pointed out by referee #1.

Sincerely,
Janne Rinne
Co-editor

Dear David Sayres et al.,

I find the manuscript improved in the revision. Your reply to the main objection, i.e. the apparent neglect of low-frequency contribution, shows that the low frequencies are indeed taken into account. The modified revised version of the manuscript explains this more explicitly. Thus, I regard the manuscript to be publishable after some minor revisions.

Your reply 2.1 discusses the FFM to extensively. As the FFM is not well known in the flux community, could you add shortly the key points of this, and about assumptions, into the manuscript, e.g.:
-Its statistical approach to flux calculation (somewhat similar to disjunct eddy covariance approach)
-Difference to RFM and wavelet approaches
Also, it would be good to explicitly discuss the assumptions behind FFM, e.g.:
-that there are no systematic flow patterns over different land cover classes)
-that the land cover can be divided into distinct, internally homogenous, land cover types (this relates to point 3 as well

You could also mention that comparing FFM and wavelet is an interesting question but not tackled in this paper.

Changes to MS, 6.2: can’t find the latter addition from the revised manuscript.

Reply 10.1: I do not see the reasoning not to do high-frequency correction when the highest measured wavenumber is in inertial sub-range. This is no quarantee that the high frequency losses wouldn’t be significant. What you mean by data starvation test, can you describe this in more detail.

Comment 13, and answers: It would be good to have a comparison on how wide the 2D footprint is compared to typical land cover class fraction. This would maybe justify not using 2D footprint.

Furthermore, when calculation the membership to certain class, did you weigh the contribution by footprint function, or did you take certain %-area as footprint area and use that (non-weighed) in membership calculation. I’d assume the former, but couldn’t find it explained in the text.

Page 11, lines 14-19: Here you discuss on the representativity of the tower fluxes based on flights on one day. How about the other days with flight near tower, do they show the same pattern?

Page 12, lines 4-7: “For the flight speed of the Centaur 5 at low altitude and wind conditions during the flights, the footprint length for each 60-m fragment varied between 0.5 and 5 km, though the part of the footprint whose probability of contributing more than 90% of the flux was between 100 and 800 m long”. How do you define the footprint length of 0.5 - 5 km? Theoretically, there is no absolute boundaries of footprint function.

Technical corrections

Page 8, line 17: “We use this to establish transects at altitudes typically 10 m to 30 m above ground; low as safely possible” would be better as “We use this to establish transects flown at altitudes typically 10 m to 30 m above ground; as low as safely possible”.

Page 12, lines 1-2: “Increasing this threshold increases the link between the calculated flux and a single land class, but reduces the number of footprints available for the analysis thus loosening the confidence interval”. Would this be better as “Increasing this threshold increases the link between the calculated flux and a single land class, but reduces the number of footprints available for the analysis thus widening the confidence interval”?

Sincerely,
Janne Rinne
Co-editor

Dear David Sayres et al.,

I find that most of my comments have been answered in a satisfactory manner.

However, I am not sure if I agree with the conclusion drawn from data starvation test. Subsampling the original full 10 Hz time series at lower sampling frequencies should not lead to significant systematic flux underestimation, at least in the cases where the averaging period is much longer than the integral time scale of w’s’ (e.g. Lenschow et al., 1994; Bosveld and Beljaars, 2001; Rinne et al., 2008; Rinne and Ammann, 2012; Turnipseed et al., 2009). The data starvation actually simulates disjunct eddy sampling, rather than high frequency loss by inadequate response time. In order to properly estimate the effect of inadequate instrumental response time, the original time series should be filtered with low pass filter, rather than just subsampled. While you are probably right that the effect of high frequency loss on fluxes is relatively small, it should be shown in more rigorous way.

Sincerely,
Janne Rinne
Co-Editor

References

Bosveld FC, Beljaars ACM (2001) The impact of sampling rate on eddy-covariance flux estimates. Agric For Meteorol 109:39–45

Lenschow DH, Mann J, Kristensen L (1994) How long is long enough when measuring fluxes and other turbulence statistics? J Atmos Ocean Technol 11:661–673

Rinne J, Douffet T, Prigent Y, Durand P (2008) Field comparison of disjunct and conventional eddy covariance techniques for trace gas flux measurements. Environ Pollut 152:630–635

Rinne J, Ammann C (2012) Disjunct Eddy Covariance Method. In Aubinet M, Vesala T, Papale D, (Eds) Eddy Covariance – A Practical Guide to Measurement and Data Analysis. Springer ISBN 978-94-007-2350-4. pp. 291-307.

Turnipseed AA, Pressley SN, Karl T, Lamb B, Nemitz E, Allwine E, Cooper WA, Shertz S, Guenther AB (2009) The use of disjunct eddy sampling methods for the determination of ecosystem level fluxes of trace gases. Atmos Chem Phys 9:981–994

Dear David Sayres,
The description on the interpolation of the sub-sampled data series improved the description of the high frequency testing. I am pleased to accept the revised manuscript for publication in APC.
Sincerely,
Janne Rinne