the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 Sep 2026
Ammonia measurements over multiple fertilisation campaigns using eddy covariance and the integrated horizontal flux method
Abstract. Quantifying field-scale ammonia emissions following nitrogen fertilisation remains challenging, and estimates may differ substantially between measurement methods. We compared quantum cascade laser-based eddy covariance (QCL-EC) with the integrated horizontal flux method using Adapted Low-cost Passive High Absorption (ALPHA) samplers (ALPHA-IHF) during eight urea fertilisation campaigns in winter wheat (2021–2023).
Both approaches identified post-fertilisation emission periods and captured similar dynamics, but paired interval estimates differed. Only 62% of intervals yielded ALPHA-IHF flux estimates, mainly because irregular or weak concentration profiles prevented robust calculation. QCL and ALPHA concentrations were positively correlated, but QCL concentrations were, on average, 31% higher, whereas QCL-EC emission estimates were 23% lower than ALPHA-IHF estimates. Median campaign emission factors were 1.6% for QCL-EC and 3.2% for ALPHA-IHF, respectively. QCL-based estimates were more precise, and differences between the methods varied with short-term ammonia concentration variability, air temperature, friction velocity and ALPHA profile quality.
The contrasting results show that concentration-level patterns do not translate directly into flux estimates and that differences between methods vary with field conditions. QCL-EC resolved short-term net exchange but required turbulence-based quality filtering, whereas ALPHA-IHF yielded time-integrated emission estimates only when the fitted concentration profile met the quality criteria. Without an independent reference, systematic over- or underestimation could not be assigned to either method. Assessments of fertiliser-induced emissions must consider method-specific data coverage and differences in flux representation. The greater precision of QCL-based estimates and high temporal resolution of QCL-EC make it promising for detecting subtle emission differences under low-emission conditions, while improved concentration precision and resolution of weak gradients could broaden ALPHA-IHF applicability.
- Preprint
(1054 KB) - Metadata XML
-
Supplement
(646 KB) - BibTeX
- EndNote
Status: open (until 04 Nov 2026)
- RC1: 'Comment on egusphere-2026-5494', Anonymous Referee #1, 18 Oct 2026 reply
-
RC2: 'Comment on egusphere-2026-5494', Anonymous Referee #2, 21 Oct 2026
reply
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2026/egusphere-2026-5494/egusphere-2026-5494-RC2-supplement.pdf
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 251 | 170 | 44 | 465 | 62 | 52 | 55 |
- HTML: 251
- PDF: 170
- XML: 44
- Total: 465
- Supplement: 62
- BibTeX: 52
- EndNote: 55
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The authors compare QCL-based eddy covariance (QCL-EC) with the ALPHA-sampler integrated horizontal flux method (ALPHA-IHF) over eight urea fertilisation campaigns in winter wheat between 2021 and 2023. Multi-year paired measurements of this kind are rare, and the dataset is a valuable contribution to the evaluation of field-scale NH₃ emission methods.
My main concerns relate to the statistical analysis, which seems more complicated than necessary while not being explained clearly enough. Flux measurement uncertainties are not considered. The central bivariate model is only briefly described, and its conclusions depend on assumptions that are not stated. I also found several inconsistencies between the text, tables and figures. Finally, the Supplement, which is meant to document the statistical approach, is verbose and repetitive and difficult to follow. I believe these issues can be addressed, and I recommend major revision.
1. Accounting for flux uncertainties. The manuscript would benefit considerably from quantifying the flux uncertainties of both methods and carrying them through the analysis; at present neither is reported or used. For QCL-EC, I suggest reporting the random flux uncertainty and the flux detection limit (e.g. Finkelstein and Sims, 2001; Langford et al., 2015). It would also help to propagate the uncertainty arising from the damping correction (43 % mean loss) and from gap filling, and to report the fraction of gap-filled data in each paired interval. For ALPHA-IHF, the triplicate samplers and the profile-fit standard errors offer a natural basis for an uncertainty estimate, and a sensitivity analysis for the choice of background concentration and of z_p would be informative. Including these uncertainties in the regression and agreement analyses would show whether between-method differences exceed the combined measurement uncertainty. It would also provide the evidence needed to support the conclusion that QCL-EC is better suited to low-emission conditions (L749–754).
2. Regression with errors in both variables. Section 2.5 does not state which regression was used, and it appears to be ordinary least squares. OLS treats the x-variable as error-free, which is not the case here. With ALPHA-IHF on the x-axis (Fig. 4b), the variable with the larger error, the slope is expected to be biased toward zero (Wu and Yu, 2018). Based on the variance components in Table 4, the expected attenuation is roughly 0.43. The slope of 0.23 and the finding that it "differed significantly from 1" may therefore partly reflect the regression method rather than a real difference between the methods. I would suggest an errors-in-variables approach on the same (log) scale as the Bland–Altman analysis. If per-interval uncertainties become available for both methods (see comment 1), York regression (York et al., 2004) or orthogonal distance regression weighted by these uncertainties would be well suited. It would also help to clarify which data the two R² values refer to: 0.67 at L383 and 0.30 at L399.
3. Description and assumptions of the bivariate model. The bivariate model underpins two key conclusions: that QCL is more precise, and that the covariate dependence of the emission ratio is driven by ALPHA-IHF. Yet it is described only briefly in the main text. I was able to reproduce all entries of Table 4 with Grubbs' moment estimators (Grubbs, 1948). If this is indeed the approach, the model assumes that both methods respond to the true signal with a slope of 1 on the log scale. With two methods and no replicate measurements, a larger total variance of ALPHA cannot be uniquely attributed to larger random error, because a stronger proportional response of ALPHA to the true signal would produce the same pattern. The conclusions mentioned above therefore follow from this assumption rather than being shown by the data. I encourage the authors to give the model equations and assumptions in the main Methods and, if possible, to test the slope assumption. For example, the ALPHA triplicates could provide an independent estimate of the ALPHA repeatability error.
In this context, the "error SD ratio" of 1.6 and 1.4 at L414–416 appears to be the ratio of the total SDs. The ratios of the method-specific error SDs implied by Table 4 are about 9 and 2.4. If this is the case, the results of the extended model in Fig. 8 seem difficult to reconcile with Fig. 6. In Fig. 8a, QCL and ALPHA have nearly identical QCL SD slopes, implying little effect on the log ratio, yet Fig. 6a shows QCL SD as the strongest and most significant effect on that ratio. If the difference arises from including the measurement level in one model but not the other, this should be explained.
4. Data selection and comparability. The data flow would be easier to follow with a clear account, per campaign, of the number of intervals sampled, those with valid concentrations, and those with valid emissions. The 12 pairs excluded as "missing or unreliable" (L325) arguably should not be counted among the initial 98. The denominators for the "missing ALPHA-IHF" percentages in Table 2 seem inconsistent, and the overall 62 % matches neither 59/98 nor 59/86. For Table 3, please clarify whether the QCL-EC values cover the full gap-filled campaign or only the paired intervals. If they cover the full campaign, the comparison with ALPHA-IHF, which includes valid intervals only, is not like-for-like. It would also help to explain why interval emissions are "23 % lower" with QCL-EC, while cumulative emissions are "about twice as high" with ALPHA-IHF. These describe different quantities, and the cumulative sums are dominated by a few high-emission intervals.
Several steps in the ALPHA-IHF processing rely on visual inspection or case-by-case manual decisions (L237–241, L261, L558). Defining objective, reproducible acceptance criteria, describing the factors that guided the decisions, and reporting how many data were excluded at each step would greatly improve reproducibility.
A related point concerns the vertical extent of the profiles. With a 70 m fetch, the NH₃ plume above the source area is likely several metres deep (roughly x/10, i.e. about 7 m). However, the highest sampler was at 3.2 m, and wind speed was measured only up to 2 m. A substantial part of each IHF flux may therefore come from extrapolating the fitted concentration and wind profiles up to z_p. Reporting the fraction of the integrated horizontal flux that lies above the highest sampling height, and discussing how this extrapolation affects the comparison with QCL-EC, would strengthen the analysis.
5. Length and focus. The Discussion repeats a good deal of material from the Methods and Results. For example, the damping-factor statement at L186–187 reappears almost verbatim at L686–688, and much of Sect. 4.4 restates results. Some paragraphs read as general statistical explanation rather than interpretation of the data (e.g. L469–472, L495–500). Tightening these parts would make the key messages clearer.
6. Supplement. In its current form, the Supplement is long and repetitive and does not give a clear account of the models actually fitted. I would encourage the authors to rewrite it as a concise methods description. It should contain the model equations with all symbols defined, the underlying assumptions, the estimation method and software, and ideally a worked example linking the equations to the values in Table 4. The annotated R code is welcome but cannot replace such a description. The essentials of the bivariate model would be better placed in the main Methods section.
Specific commentsL19–21: It would help to state that the emission factors are expressed as % of applied N.
L70: Please specify what the biases refer to, and relative to what reference.
L91–92: The authors may wish to cite additional recent QCL-EC NH3 studies, e.g. Sun et al., 2015
L106: The phrase "study site changing annually" is unclear. Were these different fields, and how far apart were they? Since soil texture differs between years, site and year are confounded, which is worth acknowledging.
L133–137: It would be clearer to state explicitly that only one EC tower and one (or two?) IHF mast were used.
L148–153: Was instrument drift checked, for example with zero measurements? The inlet response time would also be useful to report.
L182–201: A short summary of how the Wintjen et al. (2020) method was applied would be helpful, together with the fractions of data removed by quality control, the resulting gap fractions, and the performance of the gap-filling model on the test data (e.g. R2, RMSE).
L237–241: Please describe how the wind profiles were inspected and how many data points were excluded.
L324: The campaign duration of 8–20 days differs from the "six to 25 days" given at L171.
L357–362: It would be informative to report the variability (range or SD) of peak timing and plateau duration, not only the means. How exactly was a "sustained decrease" defined?
L364–366: A figure showing the mean diurnal cycles of fluxes and concentrations would support this statement.
L376–377: The sign of the reported difference is opposite to that in Table 3.
L380–381: Campaign 2022-II, with an EF of about 2.6 %, does not appear to be a "very low emission" campaign.
L396–397: The meaning of "does not imply close correspondence in absolute values" could be made more explicit.
L462–463: Which instrument does "contaminations" refer to, and what evidence supports this?
L510–511: What is the practical implication of the best agreement occurring at QCL SD = 3.6 µg m–3?
L529–530: Unequal error variances indicate differences in precision (under the model assumptions), but not by themselves a dependence on measurement conditions. The two statements could be separated.
L548–551: Please state the objective criteria used to classify an IHF estimate as failed.
L563–566 and Fig. 9: How was R² calculated for the nonlinear profile fit (centred or uncentred)? Negative values indicate a fit worse than a constant, which raises the question of whether such profiles should be used at all.
Table 5: AIC should be introduced in the Methods. A table of the individual coefficients (estimate, SE, p-value, VIF) would be a useful addition, possibly in the Supplement.
Figures: Uncertainty bars and markers for fertilisation and rain events would make Figs. 2 and 3 more informative. Fig. 4 would benefit from stating n and the regression type, and the y-axis of Fig. 5 should indicate the log scale. Fig. 9 has a truncated axis.
ReferencesFinkelstein, P. L. and Sims, P. F.: Sampling error in eddy correlation flux measurements, J. Geophys. Res.-Atmos., 106, 3503–3509, https://doi.org/10.1029/2000JD900731, 2001.
Grubbs, F. E.: On estimating precision of measuring instruments and product variability, J. Am. Stat. Assoc., 43, 243–264, https://doi.org/10.1080/01621459.1948.10483261, 1948.
Langford, B., Acton, W., Ammann, C., Valach, A., and Nemitz, E.: Eddy-covariance data with low signal-to-noise ratio: time-lag determination, uncertainties and limit of detection, Atmos. Meas. Tech., 8, 4197–4213, https://doi.org/10.5194/amt-8-4197-2015, 2015.
Sun, K., Tao, L., Miller, D. J., Zondlo, M. A., Shonkwiler, K. B., Nash, C., and Ham, J. M.: Open-path eddy covariance measurements of ammonia fluxes from a beef cattle feedlot, Agric. For. Meteorol., 213, 193–202, https://doi.org/10.1016/j.agrformet.2015.06.007, 2015.
Wu, C. and Yu, J. Z.: Evaluation of linear regression techniques for atmospheric applications: the importance of appropriate weighting, Atmos. Meas. Tech., 11, 1233–1250, https://doi.org/10.5194/amt-11-1233-2018, 2018.
York, D., Evensen, N. M., López Martínez, M., and De Basabe Delgado, J.: Unified equations for the slope, intercept, and standard errors of the best straight line, Am. J. Phys., 72, 367–375, https://doi.org/10.1119/1.1632486, 2004.