the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 Jul 2026
Technical note: Soil MIR-FTIR spectral signatures reveal a measurable link to soil greenhouse gas flux potential
Abstract. The growing reliance on nature-based solutions for climate change mitigation has increased the need for robust methods to assess soil greenhouse gas (GHG: CO2, CH4, N2O) emissions. A key challenge is the upscaling of site-specific observations to improve predictions of the spatial distribution and magnitude of soil GHG fluxes. Upscaling is hindered by the laborious, costly, and time-consuming nature of soil flux measurements at the scale needed to generate sufficiently large datasets. Fourier-transform infrared (FTIR) spectroscopy is a cost-effective, high-throughput method that has demonstrated its ability to predict soil physicochemical properties. Since the magnitude of soil GHG fluxes is strongly influenced by these properties, FTIR may also serve as a basis for predicting soil GHG fluxes. In this study, we evaluated whether FTIR calibration models can be developed to predict soil GHG fluxes and organic matter decomposition rates directly from mid-infrared spectra. FTIR calibration and cross-validation were performed using soil samples collected across 20 wetland study sites in six European countries and linked to measured soil GHG fluxes and decomposition rates at the corresponding sampling sites. The results demonstrated that FTIR can be calibrated to reliably predict soil decomposition rates and heterotrophic respiration standardised at a given soil temperature. For decomposition rates, the root mean square error (RMSE) of FTIR-predicted values was 0.01 dd⁻1, corresponding to approximately 17–53 % of the interquartile range of observed rates. For heterotrophic respiration, the RMSE of 18 mg CO2–C m⁻2 h⁻1 corresponded to 21–64 % of the interquartile flux range. The findings also indicate FTIR's potential to identify biogeochemical risk areas or potential hotspots of CH4 and N2O emissions, effectively mapping the substrate-driven 'ultimate' controls even when 'proximate' environmental triggers (e.g., water table fluctuations) are absent.
- Preprint
(1138 KB) - Metadata XML
-
Supplement
(547 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
CC1: 'Comment on egusphere-2026-3623', Jonathan Sanderman, 27 Jul 2026
-
AC1: 'Reply on CC1', Aldis Butlers, 28 Jul 2026
Thank you for your thoughtful comments. We agree that these analyses would further strengthen the mechanistic interpretation of the MIR models, and we appreciate these suggestions.
1) We agree that evaluating FTIR capability on easily measured soil properties, such as organic matter content and pH, would provide valuable insight into the additional predictive value of MIR spectroscopy. In parallel with this study, we have developed FTIR calibration models for several soil physicochemical properties, including pH, total carbon, total nitrogen, and HNO3-extractable K, Ca, Mg, and P, using a much larger dataset of approximately 9,000 soil samples. External validation using samples collected at the study sites showed promising performance, particularly for pH, total carbon, and total nitrogen.
However, these calibrations were developed from a substantially different dataset than the one used for the GHG models presented here. Including these results would require a comprehensive description of the additional sampling design, laboratory analyses, calibration procedures, and validation strategy, which would considerably broaden the scope of the manuscript. We intended to keep this contribution focused on demonstrating the feasibility of directly predicting soil GHG-related variables from MIR spectra using samples collected within the study sites.
We agree that an inclusion of MIR-based predictions on conventional soil properties provides important context, and we can consider incorporating such analyses either in a substantially expanded version of this manuscript; however, in that case, we probably should have aimed for a format of a research article; or we can consider presenting these results in a dedicated follow-up study.
2) Thank you for this suggestion. We agree that analysing variable importance and comparing the spectral regions contributing to GHG prediction with those responsible for predicting organic matter would provide valuable mechanistic insight into the features responsible for predictions.
Among the co-authors, we have discussed this aspect extensively. We believe that interpreting the spectral features responsible for model performance requires dedicated chemometric analyses and careful attribution of absorption bands to specific soil constituents. Such analyses represent a substantial research topic in their own right and were beyond the intended scope of this proof-of-concept study, which focused primarily on evaluating predictive performance.
We fully agree that investigating the spectral features driving successful predictions is an important direction for future research and appreciate this suggestion.
Citation: https://doi.org/10.5194/egusphere-2026-3623-AC1
-
AC1: 'Reply on CC1', Aldis Butlers, 28 Jul 2026
-
RC1: 'Comment on egusphere-2026-3623', Anonymous Referee #1, 26 Aug 2026
Review of a Technical Note: Soil MIR-FTIR spectral signatures reveal a measurable link to soil greenhouse gas flux potential
The manuscript evaluates whether mid infrared (MIR) diffuse reflectance FTIR spectroscopy of soil samples can predict soil GHG flux potential (CO2/heterotrophic respiration, CH4, N2O) and organic matter decomposition rates, using a dataset of twenty wetland sites across six European countries. The authors report good predictive performance for heterotrophic respiration (particularly decomposition rate), weaker performance for CH4, and largely unsuccessful calibration for field-measured N2O, though incubation-derived N2O showed moderate predictive skill. The paper proposes the useful conceptual framing of GHG Flux Potential as a standardized, spectroscopically accessible soil trait. Overall, this is a well-structured and interesting proof-of-concept study with a clear narrative and a genuinely novel angle (extending FTIR from soil-property prediction to direct GHG-flux prediction). I recommend minor revisions before acceptance, primarily to address the points below.
Major comments
The manuscript does not clarify whether the repeated calibration/validation cycles are with respect to site-level (or even sampling-point-level) independence. Soil spectra from the same site are likely to be highly autocorrelated (similar mineralogy, organic matter composition, land-use history), so if spectra from a given site can appear in both the calibration and validation subsets across different cycles, the observed metric values would be biased, essentially measuring within-site interpolation rather than genuine out-of-site prediction. Given that the ultimate motivation of the paper is spatial upscaling to new, unsampled locations, this is a critical methodological point. I would ask the authors to explicitly state how validation folds were built and, ideally, to report performance under a leave-one-site-out validation scheme in addition to the current random cross-validation.
1) The calibration aim for HR combines directly measured (trenched) respiration with respiration estimated by multiplying total CO2 flux at vegetated points by a fixed factor of 0.5. This ratio is well known to vary substantially across ecosystems, seasons, and vegetation types, and using it introduces structured (not just random) noise into the reference dataset. The authors acknowledge this qualitatively (lines 258-261) but do not quantify it. It would strengthen the paper to report (a) what proportion of the calibration dataset relied on this approximation versus direct trenched measurements, and (b) a sensitivity analysis comparing model performance with and without the approximated subset.
2) Since MIR absorbance behavior differs substantially between highly organic and mineral matrices, pooling both soil types in a single calibration could mask stronger relationships that might emerge from stratified models. Given that soil type is already recorded in the appendix, a sensitivity analysis (or at least a discussion of why pooling was preferred) seems feasible and would considerably strengthen the paper's conclusions.
3) It is already acknowledged by the authors and I don't consider the absence of independant validation data as a fatal flaw for a technical note but given how central the proof claim is to the paper's contribution, I'd suggest moving this caveat earlier rather than only in the Discussion, so that the reader's expectations are calibrated from the outset.
Minor comment
- Ligne 20 in the abstract: the RMSE of FTIR-predicted values was 0.01 dd for decomposition rate: consider adding the mean/median decomposition rate for scale, since 0.01 dd⁻¹ is uninformative without context (it is clarified later via IQR, but the abstract would benefit from one anchoring number).
Citation: https://doi.org/10.5194/egusphere-2026-3623-RC1 -
AC2: 'Reply on RC1', Aldis Butlers, 28 Aug 2026
Reply to a Review of a Technical Note: Soil MIR-FTIR spectral signatures reveal a measurable link to soil greenhouse gas flux potential
The manuscript evaluates whether mid infrared (MIR) diffuse reflectance FTIR spectroscopy of soil samples can predict soil GHG flux potential (CO2/heterotrophic respiration, CH4, N2O) and organic matter decomposition rates, using a dataset of twenty wetland sites across six European countries. The authors report good predictive performance for heterotrophic respiration (particularly decomposition rate), weaker performance for CH4, and largely unsuccessful calibration for field-measured N2O, though incubation-derived N2O showed moderate predictive skill. The paper proposes the useful conceptual framing of GHG Flux Potential as a standardized, spectroscopically accessible soil trait. Overall, this is a well-structured and interesting proof-of-concept study with a clear narrative and a genuinely novel angle (extending FTIR from soil-property prediction to direct GHG-flux prediction). I recommend minor revisions before acceptance, primarily to address the points below.
We thank the reviewer for the constructive assessment of our manuscript and for recognising the novelty.
The manuscript does not clarify whether the repeated calibration/validation cycles are with respect to site-level (or even sampling-point-level) independence. Soil spectra from the same site are likely to be highly autocorrelated (similar mineralogy, organic matter composition, land-use history), so if spectra from a given site can appear in both the calibration and validation subsets across different cycles, the observed metric values would be biased, essentially measuring within-site interpolation rather than genuine out-of-site prediction. Given that the ultimate motivation of the paper is spatial upscaling to new, unsampled locations, this is a critical methodological point. I would ask the authors to explicitly state how validation folds were built and, ideally, to report performance under a leave-one-site-out validation scheme in addition to the current random cross-validation.
Thank you for the observation. We can clarify in the manuscript that the cross-validation performed does not account for site- or sampling-point-level grouping. All spectra was treated as a single dataset, meaning that spectra originating from the same site may occur in both calibration and validation subsets, while individual spectra are excluded from calibration when used for validation. We agree that leave-one-site-out validation would provide an assessment of transferability. However, the objective of this Technical Note is to demonstrate a proof of concept rather than to provide a ready-to-use calibration with established accuracy for application in specific regions. Given that the dataset comprises sites from six countries, measured by different research teams under contrasting environmental conditions, leave-one-site-out validation would already approach an external site-level test. We do not consider such re-analysis necessary at this stage, particularly because the manuscript already identifies independent validation as a task for further method development and, as noted by the Referee in Comment 3, its absence is not considered a critical flaw of this proof-of-concept study. For practical application, the calibration would in any case need to be validated using independent sites representative of the region and soil conditions in which the method is intended to be applied. We can clarify this limitation and scope more explicitly in the manuscript.
1) The calibration aim for HR combines directly measured (trenched) respiration with respiration estimated by multiplying total CO2 flux at vegetated points by a fixed factor of 0.5. This ratio is well known to vary substantially across ecosystems, seasons, and vegetation types, and using it introduces structured (not just random) noise into the reference dataset. The authors acknowledge this qualitatively (lines 258-261) but do not quantify it. It would strengthen the paper to report (a) what proportion of the calibration dataset relied on this approximation versus direct trenched measurements, and (b) a sensitivity analysis comparing model performance with and without the approximated subset.
We fully acknowledge that the 0.5 factor is not constant in reality and that the heterotrophic contribution to total soil respiration can vary among ecosystems, seasons and vegetation types. This was also the reason we considered the good calibration performance obtained despite this additional uncertainty to be an important result rather than simply a limitation. In our dataset, around 40% data represented directly measured HR; we can specify this. Direct measurements of heterotrophic respiration are often unavailable in existing datasets, while total soil CO2 flux measurements are common, and some form of partitioning or adjustment is therefore frequently needed when estimating soil carbon balance. Our intention was to test whether a simple and widely applicable transformation could still provide reference data of sufficient quality for MIR calibration. We agree that repeating the calibration using only directly measured HR would be informative. However, this would remove a large proportion of the calibration dataset and would not reflect the practical situation in which such recalculated data will likely be needed to maximise the use of available observations. Furthermore, a robust assessment of this question would ideally require both calibration and independent validation based on directly measured HR. We therefore consider it more valuable at this proof-of-concept stage to demonstrate that useful calibration performance can still be achieved despite this source of uncertainty. We can clarify this rationale more explicitly in the manuscript and add the suggested sensitivity analysis as an important option for further method development.
2) Since MIR absorbance behavior differs substantially between highly organic and mineral matrices, pooling both soil types in a single calibration could mask stronger relationships that might emerge from stratified models. Given that soil type is already recorded in the appendix, a sensitivity analysis (or at least a discussion of why pooling was preferred) seems feasible and would considerably strengthen the paper's conclusions.
We agree that pooling mineral and organic soils may mask stronger relationships that could emerge from more stratified calibrations. This is already recognised in the Discussion, where we propose separate calibrations for contrasting soil groups as an important direction for further method development. Our intention at this proof-of-concept stage was to test whether a measurable relationship could be detected across a deliberately heterogeneous dataset rather than to optimise calibration performance for individual soil groups. We consider stratified calibration particularly valuable once larger and more diverse datasets become available, as subdivision of the present dataset would substantially reduce the amount of calibration data within individual soil categories. We can clarify this rationale and the proposed sequence of further method development more explicitly in the manuscript. We also hope that demonstrating the potential of this approach will encourage further inter-institutional and cross-country collaboration, allowing sufficiently large datasets to be assembled for robust soil-group-specific calibrations.
3) It is already acknowledged by the authors and I don't consider the absence of independant validation data as a fatal flaw for a technical note but given how central the proof claim is to the paper's contribution, I'd suggest moving this caveat earlier rather than only in the Discussion, so that the reader's expectations are calibrated from the outset.
We agree and will move this caveat earlier in the manuscript. We can also look for an appropriate place to define the scope of the study more explicitly, particularly that the intention is to present and evaluate the concept rather than to provide a ready-to-use or immediately transferable calibration method for routine laboratory application. This should also make clearer why several refinement steps, including independent validation and more stratified calibrations, are proposed as subsequent stages of method development rather than being addressed within the present study and its limited dataset.
- Ligne 20 in the abstract: the RMSE of FTIR-predicted values was 0.01 dd for decomposition rate: consider adding the mean/median decomposition rate for scale, since 0.01 dd⁻¹ is uninformative without context (it is clarified later via IQR, but the abstract would benefit from one anchoring number).
Thanks, we will do this.
Citation: https://doi.org/10.5194/egusphere-2026-3623-AC2
-
AC2: 'Reply on RC1', Aldis Butlers, 28 Aug 2026
-
RC2: 'Comment on egusphere-2026-3623', Anonymous Referee #2, 03 Sep 2026
Review of egusphere-2026-3623
Technical Note: Soil MIR-FTIR spectral signatures reveal a measurable link to soil greenhouse gas flux potential
This manuscript addresses a timely and relevant topic by evaluating the potential of MIR-FTIR spectroscopy of dry soil samples (measured using the DRIFT method) to predict soil greenhouse gas fluxes and decomposition-related processes across a diverse set of wetland soils.
However, at this stage, the study represents mainly a promising step towards the development of a predictive tool, rather than a method ready for broader application, as also acknowledged by the authors in the conclusions. The principal problem is not the FTIR concept itself, but that the present validation strategy and rather high prediction error does not yet demonstrate the general predictive ability claimed in several parts of the manuscript.
The manuscript is generally clearly written and understandable, but several methodological aspects should be clarified to better assess the robustness, reproducibility, and potential applicability of the presented approach. The authors used specific spectroscopy software for model development, while future users may need to reproduce or implement the models in other software. One of my main concerns is therefore that the underlying FTIR spectral data, which are essential for model reproduction, independent validation, and future application, are not provided.
Major comments
- FTIR data should be made available
The current Data Availability statement is insufficient. Measured versus FTIR-predicted values alone do not constitute the underlying dataset required to reproduce the analysis or independently evaluate the developed models. This is particularly important given the unclear sample structure and cross-validation procedure. The reference data are provided in the supplementary Excel file, but not the underlying FTIR spectra – only the file names that are not of any use as such. I thus consider it very important that the raw MIR-FTIR data (before preprocessing), together with reference data and relevant metadata, are made available. Providing these data is essential not only for transparency and reproducibility and for supporting the FAIR principles, but also for possible independent validation and the potential future application of the approach.
- Mineral vs. organic soils
The authors state that both mineral and organic soils were included, but it is unclear how many samples of each soil type were used and how they performed in the models. This information is particularly important for MIR-FTIR modelling because mineral and organic soils can have substantially different spectral characteristics. Without a clear description of the soils included in the calibration, it is difficult to define the applicability and transferability of the developed models.
It would also be useful to distinguish mineral and organic soils in the plots, for example using different colours. This would provide a simple indication of whether the models perform similarly or differently for both soil groups.
The supplementary Excel file should also include soil type (e.g. mineral vs. organic/peat soil) for each sample. This information is provided in the Appendix Table A1, but including it directly in the dataset would allow easier filtering and evaluation of the data.
- Spectral pre-processing
The authors tested several preprocessing methods and spectral regions and picked the best-performing model for each variable and soil depth. However, the final preprocessing method and spectral region used for each reported model are not provided. Please report these model specifications, including the number of PLS factors/components, to allow reproducibility of the calibrations.
- Validation
The cross-validation procedure needs more description, particularly with respect to data independence. Each soil sample was measured in three replicate subsamples, but it is unclear whether replicate spectra from the same sample were always kept together within the same calibration or validation subset. If not, the models may have been calibrated and validated using laboratory replicates of the same soil samples. It is also unclear whether samples originating from the same sampling point or study site could occur in both calibration and validation sets – possibly yielding overoptimistic results.
The authors should clearly describe how validation subsets were constructed and at which level independence was maintained. Ideally, model performance should be evaluated using site-grouped or leave-one-site-out cross-validation to demonstrate transferability to independent sites. I do not consider it essential that the authors rerun the models, but if site-level independence was not maintained, this limitation should be clearly stated in the manuscript and the reported model performance should be considered potentially optimistic when applied to independent sites.
Minor comments
The supplementary Excel file should be corrected. In column I (“Site”), sites from Latvia currently show #REF!
I would also suggest adding more sample information to the supplementary dataset to allow easier filtering, at least a column identifying mineral and organic/peat soils.
Citation: https://doi.org/10.5194/egusphere-2026-3623-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 178 | 56 | 31 | 265 | 29 | 25 | 19 |
- HTML: 178
- PDF: 56
- XML: 31
- Total: 265
- Supplement: 29
- BibTeX: 25
- EndNote: 19
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This paper presents an interesting use of mid infrared spectroscopy. I would like to make two suggestions aimed at demonstrating that the MIR models are actually directly predicting the processes in question:
1) The authors should compare the MIR model results to a simple model with just OM (and maybe OM and pH) to see if the MIR data actually outperforms these other simple to measure soil properties. I make this suggestion because many emergent soil biological properties (such as GHG flux and decay rates) are heavily dependent on organic matter content. Hopefully the MIR model has significantly more predictive power but right now the reader doesn't know.
2) Present the variable importance. Also build models to predict OM content and compare the important variables. Do the important model features make sense for predicting the particular response? Comparing the important variables between the models for GHG fluxes and OM content will show whether or not the model is doing anything besides predicting OM content which we know is an important driver of these biological properties.