the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 May 2025
Hot spots, hot moments, and spatiotemporal drivers of soil CO2 flux in temperate peatlands using UAV remote sensing
Abstract. CO2 emissions from peatlands exhibit substantial spatial and temporal variability due to their heterogeneous nature, presenting challenges to identify their underlying drivers and to accurately quantify and model CO2 fluxes. Here, we integrated field measurements with Unmanned Aerial Vehicle (UAV)-based multi-sensor remote sensing to investigate soil respiration across a temperate peatland landscape. Our research addressed two key questions: (1) How do environmental factors control the spatial-temporal distribution of soil respiration across complex landscapes? (2) How do hot spots and hot moments of biogeochemical processes influence landscape-level CO2 fluxes? We find that dynamic variables (i.e., soil temperature and moisture) play significant roles in shaping CO2 flux variations, contributing 43 % to seasonal variability and 29 % to spatial variance, followed by semi-dynamic variables (i.e., NDVI and root biomass) (19 % and 24 %). Relatively static variables (i.e., soil organic carbon (SOC) stock and C/N ratio) have a minimal influence on seasonal variation (2 %) but contribute more to spatial variance (10 %). Additionally, predicting time series of CO2 fluxes is feasible by using key environmental variables (test set: R2 = 0.74, RMSE = 0.57 μmol m-2 s-1), while UAV remote sensing is an effective tool for mapping daily soil respiration (test set: R2 = 0.75, RMSE = 0.54 μmol m-2 s-1). By the integration of in-situ high-resolution time-lapse monitoring and spatial mapping, we find that despite occurring in 10 % of the year, hot moments contribute 28 %–31 % of the annual CO₂ fluxes. Meanwhile, hot spots – representing 10 % of the area – account for 20 % of CO2 fluxes across the landscape. Our study demonstrates that integrating UAV-based remote sensing with field surveys improves the understanding of soil respiration mechanisms across timescales in complex landscapes, providing insights into carbon dynamics and supporting peatland conservation and climate change mitigation efforts.
- Preprint
(1817 KB) - Metadata XML
-
Supplement
(751 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on "Hot spots, hot moments, and spatiotemporal drivers of soil CO2 flux in temperate peatlands using UAV remote sensing"', Anonymous Referee #1, 27 Jun 2025
The work presented in this manuscript is an impressive culmination of data informing models of the spatiotemporal variations in CO2 flux in temperate peatlands. While the study is rich in data with a compelling model, the current state of the manuscript does not acknowledge two of the most well documented sources of carbon respiration in peatlands. Further, I question the use of UAS TIR imagery to effectively determine soil temperature in a peatland that is so richly covered in vegetation. These concerns are reflected in my two main comments below:
Main Comment 1: I am very happy to see the use of UAS thermal infrared imagery in this study. However, the details into how this method effectively captures the soil temperature of the peatland are not described. Flight conditions of the UAS were not described in the methods (clear, cloudy, after-rain, etc.) which could significantly impact the data. The TIR imagery would also need a thermal emissivity value to be calibrated and ensure accurate temperatures of the surface of the peatlands. This leads me to my issue with the vegetation cover of the peatlands interfering with the soil temperatures. Vegetation cover would have significantly higher thermal emissivity values compared to soils along the peat surface. In the article cited below, Harvey et al. (2019) describes a calibration for the thermal emissivity of water. I believe you should do something similar for vegetation versus soil along the surface of the peatland at the very least. If you can upscale this further by vegetation type, which has already been mapped in your 2024 publication, this would yield even more novel results for your study. Another route with this is offered by calibration in TIR image processing software, where you can apply a static thermal emissivity value based on other published values. This could further improve the model by breaking up the peatland into subsections and again ensuring the accuracy of your temperature values.
Harvey, M. C., Hare, D. K., Hackman, A., Davenport, G., Haynes, A. B., Helton, A., ... & Briggs, M. A. (2019). Evaluation of stream and wetland restoration using UAS-based thermal infrared mapping. Water, 11(8), 1568.
Main Comment 2: This manuscript in its current form does not describe the importance of hydrology or atmospheric pressure on carbon respiration in peatlands. Depths of the water table are well documented in numerous articles to impact carbon respiration, with drier climates and lower water tables leading to enhanced carbon production. This deserves its own paragraph in the introduction and needs to be discussed in the methods in the background of the site. No details are provided about the hydrologic setting of the peatland study site and the difference between minerogenous groundwater contributions versus ombrogenous precipitation contributions are not clarified. The peatland needs to be described in some way (as either a bog, fen, heath, etc.) to understand the expected flows within the peat matrix that change how CO2 is released and generated. I would also point to Figure 2 as the CO2 respiration modeled parallels the rise in the water table of the peatland mid-year, which would be expected in these environments. The role of atmospheric pressure also needs to be addressed by the authors in the manuscript in the introduction and discussion sections. I would like to highlight the carbon respiration occurring in Figure 3 that could be attributed to shifts in atmospheric pressure prior to precipitation. The model appears to fail to quantify the change in atmospheric pressure transitioning from the winter to the spring in the northern hemisphere. This period is also well documented to release carbon rapidly in peatlands due to the thaw occurring and seasonal change in pressure locally. Pressure would be a great variable to incorporate with this model. I believe your correlation with temperature would be secondary to this new variable, if included.
For these reasons, I recommend this manuscript to be reconsidered after major revisions. Additional minor grammatical and queries are provided in the attached “mark-up” version of the manuscript.
-
AC1: 'Reply on RC1', Yanfei Li, 28 Aug 2025
We thank the reviewer for the positive comments and valuable feedback. We have now incorporated the suggested changes and provided a deeper exploration of our study. Below, we outline our responses to the main comments, while detailed responses to the specific points are provided in the attached document.
Response to Main comments 1:
Thank you for your constructive comments and valuable suggestions regarding the calibration of UAV-based thermal infrared (TIR) imagery. We have now included a detailed description of the methodology and calibration steps in the revised manuscript.
During each UAV TIR flight, the camera emissivity was set to the default value (100%). To calibrate the raw land surface temperature (LST) data, we placed two reference panels (i.e., one hot and one cold, with temperature sensors underneath) on the ground for every flight. The raw LST data were first calibrated using the linear regression between the recorded panel temperatures and corresponding pixel values in the imagery. This initial calibration procedure is now described in the main text (Section 2.6 UAV imagery processing). Following your suggestions, we further refined the LST calibration by incorporating spatially upscaled thermal emissivity maps based on the land cover map of our previous work (Li et al., 2024). The emissivity values for each land cover class were obtained from published literature and applied during image processing to generate double-calibrated LST datasets:
The raw thermal infrared video streams were converted into RJPG images using ThermoViewer version 3.0.26 (TeAX, 2022). Subsequently, the thermal images were processed with the Pix4D mapper to generate land surface temperature (LST) maps (resolution: 12 cm). To calibrate the LST of each date (Figure 2a), we first applied linear regressions of temperature obtained by camera and temperature of 2 targets on the ground (Text S1) to create a correction formula. Next, we mapped the spatial variations of surface emissivity using the classification-based approach (Li et al., 2013; Snyder et al., 1998), based on land cover data from our previous work (Figure 1b; Li et al. (2024)) and emissivity values of each class from literature (Snyder et al., 1998). Finally, we converted the LST to thermal radiance using Planck’s law, applied an emissivity-based correction, and then converted the radiance back to obtain calibrated LST.
References
Li, Y., Henrion, M., Moore, A., Lambot, S., Opfergelt, S., Vanacker, V., Jonard, F., Van Oost, K., 2024. Factors controlling peat soil thickness and carbon storage in temperate peatlands based on UAV high-resolution remote sensing. Geoderma, 449, 117009. https://doi.org/10.1016/j.geoderma.2024.117009.
Li, Z.-L., Wu, H., Wang, N., Qiu, S., Sobrino, J.A., Wan, Z., Tang, B.-H., Yan, G., 2013. Land surface emissivity retrieval from satellite data. International Journal of Remote Sensing, 34, 3084-3127. 10.1080/01431161.2012.716540.
Snyder, W.C., Wan, Z., Zhang, Y., Feng, Y.Z., 1998. Classification-based emissivity for land surface temperature measurement from space. International Journal of Remote Sensing, 19, 2753-2774. 10.1080/014311698214497.
The updated coefficients and contributions of double-calibrated LST data (i.e., contributions increased by 5% after the second calibration) in the model for mapping soil temperature are now provided in the support material (see details in the attached file).
In addition, details of UAV flight conditions have been added to the Supplementary Material (Text S1: UAV Campaigns):
The UAV flight missions were carried out between 10 h 00 and 14 h 00, at a frequency of every two weeks in summer and monthly interval in other seasons. The flight patterns and altitudes used for UAV missions were similar to our previous work (Li et al., 2024). The RGB images were captured at a flight height of 100 m. The side and frontal overlap ratios were set to 70 % and 80 %, respectively, resulting in a spatial resolution of 2.05 cm. The multispectral and thermal infrared flights were conducted at above take-off point altitude of 90 m and a speed of 7.1 m/s simultaneously using a dual gimbal connector. Both side and frontal overlap ratios were set to 80 %. In this case, the spatial resolutions of the multispectral and thermal infrared images are approximately 6 cm and 12 cm, respectively. A MicaSense calibrated reference panel with known reflectance values was used immediately to calibrate the multispectral camera before and after each flight. The TeAX thermal infrared camera combines FLIR Tau2 cores and ThermalCapture hardware that allows the user to store raw infrared video streams directly on a local USB memory stick, together with additional information like position and time from GPS. In addition, TeAX technology makes heated shutters provide evenly a uniform temperature across the shutter and maintains this temperature throughout the duration of its operation. During the flight mission, the emissivity setting of the thermal infrared camera was set to 100 %. To further correct the differences between the true surface temperature of the ground and that measured by the sensor due to emissivity effect, two homemade thermal calibration panels (50 cm x 100 cm, one hot and one cold that fills with ice) were used on the ground with a known temperature to adjust any offsets in the thermal images and to understand the temperature changes throughout the duration of the flight. To enhance the LiDAR signal penetration, we chose the triple-echo mode with a sampling frequency of 160 kHz, maintaining a flight height of 50 m above the take-off point at a speed of 6 m/s. During the flight mission, the ground sampling distances varied between 1.16 cm and 2.18 cm per pixel. The IMU calibration procedures were conducted automatically at the beginning, during the mission, and after flight routes to ensure inertial navigation accuracy.
The RGB and LiDAR flights were conducted in RTK positioning mode using a D-RTK 2 base station (DJI, Shenzhen, China). The base station was set up at a known point and was used to provide real-time positional corrections throughout the flight. For the multispectral and thermal infrared cameras, nine ground control points (GCPs) were used (50 cm × 50 cm targets). The GCPs were made of white laminated board stuck with aluminum foil in the diagonal area and were distributed across the study site during the flight mission. Their position was measured using an Emlid Reach RS2 GPS device, utilizing a post-processing RTK solution with the Belgian WALCORS network.
We believe these improvements address your concerns regarding calibration accuracy and vegetation cover effects on TIR measurements, and they strengthen the robustness of our soil temperature mapping approach.
Response to Main comments 2:
Our study site is an ombrogenous bog hillslope, characterized by distinct SE-NW oriented topographic units (i.e., summit, topslope, shoulder, backslope, and footslope). Additional details about the landscape have been added to the revised manuscript.
We agree that the water table plays an important role in regulating soil respiration in peatlands, and we have included this in both the introduction and discussion. Annual water table dynamics of the study site are now presented in the results (Table 2 in the attachment). Five water table sensors were installed along the middle transect of the landscape (i.e., five slope positions (summit, topslope, shoulder wet, backslope, footslope), 5 sites in total), close to locations where we conducted soil respiration measurements. However, water table data were only available from June 2023 to October 2024, whereas our soil respiration measurements (i.e., the Licor system, at 6 slope positions (summit, topslope, shoulder dry, shoulder wet, backslope, and footslope), 33 sites in total) were conducted from February 2023 to March 2024. Including water table as a model variable would therefore reduce the number of CO2 observations by approximately half (from 666 to 336) and will not capture a full annual cycle.
As an alternative, we have reported the model performance (including water table) in the support material of the revised version (Table 3 in the attachment). In this model, we found that water table explained 10 % of seasonal variation in CO2 fluxes, while NDVI and soil temperature were the most important variables, contributing 22 % and 21 %, respectively. The relatively small contributions from water tables might reflect (i) the limited number of monitoring locations, (ii) the shorter monitoring period, (iii) the generally low water table across the year (Table 2 in the attachment), particularly at the footslope, backslope, and summit, where maximum water tables remained > 9 cm below ground surface. This maintained aerobic layers that support soil respiration, thereby reducing the influence of their fluctuations on CO2 fluxes. Increasing spatial coverage and temporal resolution of water table observations across the landscape would likely improve our ability to examine its influence on CO2 emissions.
We appreciate your suggestion addressing the role of atmospheric pressure in peatland CO2 fluxes and have now included relevant discussion in both the introduction and discussion of the revised manuscript.
Atmospheric pressure can influence gas fluxes via pressure pumping by three mechanisms: (i) atmospheric turbulence, (ii) longer-period barometric changes associated with frontal passages, and (iii) quasi-static pressure fields induced by wind over irregular topography (Ryan and Law, 2005). In our study, when atmospheric pressure was included as a predictor in the model (Table 3, Model 3), it explained only 1 % of seasonal variability in soil respiration, whereas NDVI and temperature were dominant contributors. Examination of high-frequency time series data (i.e., hourly CO2 flux from the eosFD probes) showed that at the daily scale, the diurnal pattern of CO2 fluxes did not follow atmospheric pressure fluctuation (Figure 1 in the attachment). At longer time scales, the two variables displayed only weak correlations. Instead, the CO2 flux showed clear correlation with temperature dynamics. Moreover, we observed that declines in atmospheric pressure were often followed by precipitation events, which in turn were associated with decreases in both air temperature and CO2 flux, or slight CO2 fluxes increases (Figure 1 in the attachment).
This suggests that atmospheric pressure may indirectly influence soil respiration by affecting precipitation patterns, rather than exerting a strong direct control. This limited direct effect of atmospheric pressure in our study area is likely due to (i) the absence of abrupt winter–spring thaw events which are typical of high-latitude peatlands, and (ii) the predominantly aerobic status of the surface peat in this hillslope bog, where maximum water tables remained below the surface most cases of the year (Table 2 in the attachment). In saturated peatlands, falling atmospheric pressure has been shown to trigger methane (CH4) ebullition by releasing trapped gas bubbles (Baird et al., 2004; Tokida et al., 2005; Tokida et al., 2007), while in our study site, such bubble formation and ebullition are likely minimal. Consequently, the potential for direct pressure-driven CO₂ release is relatively low. Another contributing factor maybe the limitations of our observations: (i) the lower temporal resolution of biweekly chamber measurements (i.e., the li8100 A system) and (ii) high temporal frequency but short monitoring period at each slope position (i.e., the eosFD sensors) along the middle transect may have limited our ability to detect short-lived CO₂ flux responses to atmospheric pressure fluctuations.
References
Ryan, M., Law, B. Interpreting, measuring, and modeling soil respiration. Biogeochemistry 73, 3–27 (2005). https://doi.org/10.1007/s10533-004-5167-7.
Baird, A.J., Beckwith, C.W., Waldron, S., Waddington, J.M., 2004. Ebullition of methane-containing gas bubbles from near-surface Sphagnum peat. Geophysical Research Letters, 31. https://doi.org/10.1029/2004GL021157
Tokida, T., Miyazaki, T., Mizoguchi, M., 2005. Ebullition of methane from peat with falling atmospheric pressure. Geophysical Research Letters, 32. https://doi.org/10.1029/2005GL022949
Tokida, T., Miyazaki, T., Mizoguchi, M., Nagata, O., Takakai, F., Kagemoto, A., Hatano, R., 2007. Falling atmospheric pressure as a trigger for methane ebullition from peatland. Global Biogeochemical Cycles, 21. https://doi.org/10.1029/2006GB002790
-
AC1: 'Reply on RC1', Yanfei Li, 28 Aug 2025
-
RC2: 'Comment on egusphere-2025-1595', Anonymous Referee #2, 03 Aug 2025
II. General comments
This manuscript presents a comprehensive multi-annual dataset on peatland carbon dynamics, combining UAV-based spatial observations with ground-based temporal CO₂ flux measurements. Overall, the paper is clearly written and well-structured, with the main findings effectively highlighted. The integration of spatio-temporal information via UAV surveys is a notable strength and adds interesting insights. However, there are three major issues that should be addressed before considering publication.
1. Representativeness and definition of key terms:
The study site, while referred to as a “landscape,” seems to comprise several hectares — yet no clear definition of what landscape scale entails is provided. The authors should clarify whether and how this site is representative of a broader peatland landscape or region in Belgium, and discuss the limitations if not. Similarly, the notion of a “complex” landscape is repeatedly invoked without explanation; if this adjective refers to the spatio-temporal variability of CO₂ flux patterns, this should be explicitly defined early on. Crucially, the type of peatland studied is not described, substantially limiting the transferability of the findings to other regions. The introduction should provide clearer context and definitions to justify the representativeness of the study.
2. UAV methods and robustness of spatial thermal data:
The UAV-based thermal campaign is insufficiently described and omits discussion of common limitations (e.g. emissivity settings, light conditions, flight altitude, duration). Without any comparison to in-situ temperature data and evaluation of the resulting LST, the robustness of the spatial thermal dataset is highly questionable. While the spatial approach remains valuable for identifying relative patterns and hotspots across the site, the derivation of absolute CO₂ budgets from UAV-derived LST seems unjustified in its current methodological description. The authors should therefore refine the methods and expand the discussion to include these uncertainties and assumptions.
3. Clarity and accessibility of methods:
The workflow is currently difficult to follow due to key methodological details being split between the main text and supplementary materials. For example, the use of the TWI to infer VWC is only explained in the supplement. I strongly recommend consolidating essential information in the main text and providing a clear workflow diagram — ideally one for the spatial and one for the temporal modelling approach — to improve clarity.
II. Specific comments
l. 46-47 the two paragraphs are poorly linked- you should introduce the term soil respiration already in the first paragraph
l. 59 this statement is misleading, because the weather conditions in peatland are not more variable than elsewhere; you should rather refer to the higher sensitivity to meteorological variability regarding CO2 fluxes compared to other land cover types
l. 68 shortly define hot moments and hotspots here (not only in the methods)
l. 202 should be presumably “used for generating daily mean soil temperature”, at least when referring to Text S1
l. 272 unclear, why and how TWI is in the model- see my general comments
l. 291-294 names for the different slope positions should be marked in the map or at least explained in the methods
l. 295-302 unclear whether the species were retrieved from the UAV orthomosaics (Figure 1b) or determined in-situ at the measurement locations
l.312 should be rewritten the other way around (different letters refer to significant differences)
l. 323 there is an interesting finding about NDVI that is currently not mentioned: the sign of the coefficient is positive for the seasonal patterns and negative for the spatial patterns; should be explained and discussed
l. 355 mention this station with coordinates in the methods
l. 389-391 not clear whether water table remained stable; there are feedbacks between soil temperature and soil moisture- they are not completely independent from each other)
l. 523-541 I missed any implications for peatland restoration from these findings; Can the success of the restoration be monitored with this methodology? It would be good to write 1-2 final sentences as an outlook in this perspective
l. 540-541 this is an interesting point (high-frequency hotspots vs. sporadic hotspots), but I could not find this being mentioned or presented elsewhere in the manuscript which is a requirement for a conclusion to be drawn
III. Technical comments
l. 52 please check grammar of this sentence
Citation: https://doi.org/10.5194/egusphere-2025-1595-RC2 -
AC2: 'Reply on RC2', Yanfei Li, 28 Aug 2025
We appreciate your positive comments and suggestions. We believe these changes will enhance the clarity and accuracy of our manuscript. Below, we outline our responses to the main comments, while detailed responses to the specific points are provided in the attached document.
Response to main comments 1:
Thank you for your constructive suggestions. Our study site is a bog hillslope, covering an area of 33 ha in the Belgian Hautes Fagnes, which represents an important ecosystem for studying peatland carbon fluxes due to its sensitivity to climate change and hydrological dynamics, as we mentioned in the last paragraph of the introduction. We now have added more context about the study area and defined the landscape scale and explained the complexity in section 2.1 Study area of the revised manuscript:
This ombrotrophic bog is mainly fed by precipitation and covers an area of approximately 32 hectares. The landscape exhibits complex structures, characterized by distinct SE-NW oriented topographic units (i.e., summit, topslope, shoulder, backslope, and footslope), along with diverse microtopographic features, spatiotemporal varying thermal-hydrological conditions, differences in peat thickness and carbon storage, and a range of vegetation types (Henrion et al., 2024; Li et al., 2024; Sougnez and Vanacker, 2011). More specifically, the summit is a low-relief, southeast-facing plateau at 675 - 680 m elevation, which transitions downslope into the topslope and concave shoulder slope positions (Figure 2a above). The northwest-facing backslope is relatively steeper (average slope grade: 4.98°; elevation range: 645 - 670 m) compared to these upper units, while the footslope lies in the northwestern hillslope adjacent to Hoëgne River. The peat thickness varies spatially from 0.20 to 2.10 m across the landscape, with deeper deposits in the footslope and shallower peat at the topslope (Henrion et al., 2024; Li et al., 2024). The estimated soil organic carbon (SOC) stocks (i.e., top 1 m layer) range from 176.13 t ha-1 to 856.57 t ha-1, with significantly higher storage at the summit, shoulder, and footslope (Li et al., 2024). Due to the pronounced topographic gradients and microtopography, the landscape exhibits great spatiotemporal variability in rootzone soil volumetric water content (range: 0.1 – 1 cm3 cm-3) and water table dynamics (range: -80 – 5 cm) (Henrion et al., 2025). The study site was drained and planted with spruces in 1914 and 1918, while the plantations were progressively cleared between 2000 and 2016. Since 2017, the site has been under restoration and now primarily covered by Vaccinium myrtillus, Molinia caerulea, Juncus acutus, and native hardwood species (e.g., Betula pubescens and Quercus robur) (Figure 2b in the attachment).
Because we utilized the UAV remote sensing to monitor the surface environmental dynamics, extending coverage to a larger area would require substantial additional resources and time. Our study focused on demonstrating UAV-based mapping of soil respiration with high spatial resolution down to centimeter level, whereas this comes at the expense of spatial coverage. We now have discussed this limitation in the discussion part of the revised manuscript:
In addition, the key environmental variables used for mapping soil respiration were estimated by UAV data, which inevitably introduce uncertainties into the prediction processes. For instance, because daily UAV imagery was unavailable, the predictors (i.e., air temperature, LST, and NDVI) for modelling the spatiotemporal dynamics of soil temperature were linearly interpolated between acquisition dates, potentially adding uncertainty to the model results. Moreover, flight conditions and preprocessing of the raw UAV data (e.g., georeferencing, resampling, the calibration of LST, downscaling air temperature) may have further introduced errors into the soil temperature estimates. The corrected daily TWI maps were also subject to uncertainty, as they relied on in-situ soil VWC observations, which were only available in the middle transect of the landscape. Similarly, uncertainties in SOC stock mapping arose from the peat thickness estimation and soil sampling strategy, as discussed in our previous work (Li et al., 2024).
Response to main comments 2:
Thank you for your comments. We now have refined the methods and expanded related discussion of these uncertainties. Please see our response to the Main comments 1 of reviewer 1.
Response to main comments 3:
Thank you for your suggestions. We now have placed the methodology about mapping soil temperature and corrected daily TWI in the main text (Section 2.7 & 2.8). We also added one new section about model performance evaluation in the revised manuscript (Section 2.9). Besides, we provided two workflow diagrams for clarity (Figure 4 in the attachment):
-
AC2: 'Reply on RC2', Yanfei Li, 28 Aug 2025
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 444 | 73 | 22 | 539 | 30 | 20 | 37 |
- HTML: 444
- PDF: 73
- XML: 22
- Total: 539
- Supplement: 30
- BibTeX: 20
- EndNote: 37
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1