Comparing modeled and measured carbon dioxide data at five canadian peatlands revealed highly variable non-growing season fluxes

Abstract

Peatlands hold up to one-third of the global organic carbon (C) stock and are considered a critical C sink. Climate models project that the rate of warming in northern peatlands will continue throughout the 21st century, with the greatest warming occurring during the non-growing season (NGS), which is poorly represented in current in-situ measurements. As a result, the contribution of NGS fluxes to annual peatland CO₂ budgets remains uncertain. Remotely sensed and modeled data products offer a potential means to estimate year-round fluxes, but their performance in peatland ecosystems has not been fully evaluated. In this study, we used the remotely sensed and modeled Soil Moisture Active Passive Level 4 Global Daily EASE-Grid C NEE (SMAP-NEE) data product to acquire 9 years (2015–2023) of SMAP-NEE data for five peatlands. We compared these values to a subset of year-round eddy covariance NEE (EC-NEE) measurements within this time frame at each of the five peatland sites. The analyses showed that the SMAP-NEE data product reports a stronger growing season (GS) sink and a weaker NGS source than the EC-NEE measurements. Using the relationship between SMAP-NEE and EC-NEE, we produced a Corrected-SMAP-NEE dataset, which provided an estimate of seasonal and annual carbon dioxide (CO₂) budgets. Our data analyses of the Corrected-SMAP-NEE dataset showed that NGS CO₂ emissions represented a highly variable proportion (33%–256%) of the GS CO₂ uptake, that reduced the annual CO₂ sink strength proportionally. This work demonstrates the necessity of monitoring during the NGS and highlights the importance of incorporating peatland data into model training to improve CO₂ flux estimates in these environments.

1 Introduction

Peatlands are wetlands that have accumulated a layer of organic soil, or peat, of 30 cm or deeper (NWWG - National Wetlands Working Group, 1988). Despite occupying only 3% of global terrestrial land area, they are critical stores of organic carbon (C), storing 15%–30% of the terrestrial soil C pool (Gorham, 1991; Post et al., 1982; UNEP, 2022). Specifically, northern (boreal and subarctic) peatlands, representing 57% of the global distribution of peatlands (UNEP, 2022), are estimated to store up to 88% of the total peatland C stock (Yu et al., 2010), with yearly C sequestration capacity in this region as high as 96 Tg C yr^-1 (Gallego-Sala et al., 2018). Organic C accumulation in peatland ecosystems occurs because carbon dioxide (CO₂) fixation from the atmosphere by vegetation via gross primary production (GPP) exceeds CO₂ loss via ecosystem respiration (ER), which is the sum of plant (autotrophic) respiration and respiration by heterotrophic microorganisms (Chapin et al., 2006). The difference between ER and GPP is the net ecosystem exchange (NEE), (NEE = ER-GPP) (Chapin et al., 2006). NEE varies seasonally, typically switching from negative (net CO₂ uptake) during the growing season (GS) (Coffer and Hestir, 2019; Lund et al., 2010) to positive (net CO₂ release) during the non-growing season (NGS; i.e., fall, winter, and spring) (Natali et al., 2019; Rafat et al., 2021). However, due to variability in C cycling processes, both uptake and release of CO₂ could occur in either the GS or NGS (Lund et al., 2010; Lafleur et al., 2003). An understanding of C dynamics in both the GS and NGS, as well as their interaction, is crucial in predicting how peatland CO₂ emissions will respond to a changing climate.

With climate change, the rate of increase in air temperature in the Northern hemisphere is greater than the global average (Hu et al., 2021; Jeong et al., 2012). As a result, there is concern about northern peatlands’ future CO₂ sink strength due to the strong temperature sensitivity of C cycling processes (Rafat et al., 2022; Ward et al., 2019). Recent climate projections for boreal peatlands indicate higher air temperatures of about 2.5 °C–7.5 °C by the year 2,100 (Jensen et al., 2019). These rising temperatures and the associated response of peatland ecosystems could result in substantial releases of CO₂ to the atmosphere (Limpens et al., 2008; Lunt et al., 2019). At these high latitudes, warming is more pronounced during the NGS, while, at the same time, the duration of the NGS is decreasing (Rafat et al., 2022; Barnard et al., 2018; Liu et al., 2018). While there is a wealth of literature examining the impacts of temperature-based effects on CO₂ fluxes in northern peatlands, the outcomes have been highly variable across a range of peatland types. In some cases, increased precipitation and a longer GS may increase GPP, resulting in greater annual net CO₂ uptake in northern peatlands (Gallego-Sala et al., 2018; Liu et al., 2019; Lunt et al., 2019). However, rising temperatures may also accelerate soil organic C (SOC) decomposition, thereby increasing C losses by respiration, resulting in reduced annual net CO₂ uptake or net release of CO₂ (Coffer and Hestir, 2019; Lunt et al., 2019). GPP and ER are also interdependent on nutrient availability, as faster SOC decomposition makes soil nutrients more available and most of the vegetation nutrient supply comes from recycling plant detritus, fostering GPP (Hobbie and George, 2014). It is generally accepted that both GPP and ER will increase with climate change, but their relative changes and, in turn, the change in the balance between them will ultimately impact how peatlands’ annual NEE fluxes respond to climate warming (Gallego-Sala et al., 2018).

Eddy covariance (EC) is a micrometeorological technique that provides a means to derive nearly continuous net CO₂ exchange (i.e., NEE) by calculating the covariance between the fluctuations of vertical wind velocity and the CO₂ mixing ratio (Baldocchi et al., 2018). The EC setup could consist of either an open-path or closed-path infrared gas analyser (IRGA) (Haslwanter et al., 2009). Open-path analysers are typically located in the same location as the sonic anemometer, measuring gas concentrations at that location, whereas, closed-path analysers are typically located inside a weather-proof enclosure, measuring gas concentrations that come down from an intake near the location of the sonic anemometer (Haslwanter et al., 2009). Open-path analysers, while requiring less power and maintenance, are exposed to the elements influencing measurement accuracy and resulting in more data gaps when the sensor is dirty or wet (Haslwanter et al., 2009). In northern peatland sites, which are typically remote, collecting EC data may be both expensive and logistically challenging, notably during harsh winter conditions (Coffer and Hestir, 2019; Limpens et al., 2008; Lunt et al., 2019). Additionally, direct measurements of NGS NEE are still relatively rare due to the challenges of using open-path EC equipment in the winter (Dutch et al., 2024).

To combat these challenges, we tested NASA’s Soil Moisture Active Passive Level 4 C Net Ecosystem Exchange (SMAP-NEE) data product (Kimball et al., 2022). This data product provides daily estimates of NEE, GPP, and heterotrophic respiration (HR) from a satellite data-driven terrestrial C flux model at a 9 km spatial resolution (Kimball et al., 2022). The following variables are input into the algorithm to produce the derived C data product: 500 m resolution Moderate Resolution Imaging Spectroradiometer (MODIS)-based global plant functional type (PFT) classification (from MCD12Q1 Type 5) (Friedl and Sulla-Menashe, 2019), 500 m resolution Fraction of Photosynthetically Active Radiation (fPAR) (from VIIRS VPN15A2) (Giglio, 2024), 9 km resolution SMAP Level-4 soil moisture data (SPL4SMGP) (Reichle et al., 2025), ¼ degree resolution pre-processed global, daily averaged meteorology data from the GEOS-5 Forward Processing (GEOS-5 FP) system (Rienecker et al., 2008; Kimball et al., 2022; Jones et al., 2017). The SMAP algorithm uses a light use efficiency model to estimate GPP and a terrestrial C flux model to estimate respiration, NEE, and soil organic carbon pools (Kimball et al., 2022). GPP is calculated from environmental inputs (minimum air temperature, vapor pressure deficit, freeze-thaw status, soil moisture, fPAR, and solar radiation) using plant functional type specific modifiers that account for differences in vegetation responses to environmental conditions (Kimball et al., 2022). NEE is then calculated using GPP, soil moisture, and surface temperature (Kimball et al., 2022). The model parameters are defined using a biome properties look-up table calibrated against a global network of in-situ EC CO₂ flux measurements (2000-2008) from the FLUXNET La Thuile Collection, representing eight major plant functional types (i.e., Evergreen needleleaf, evergreen broadleaf, deciduous needleleaf, deciduous broadleaf, shrub, grass, cereal crop, and broadleaf crop) (Kimball et al., 2022; Endsley et al., 2023). While previous work has explored peatland-specific applications of a similar terrestrial C flux modelling framework (Watts et al., 2014), the SMAP model does not include a peatland-specific plant functional type. This model instead represents peatlands through revised soil properties in the soil moisture product (i.e., higher soil porosity), which impacts these inputs for the C flux model (Endsley et al., 2023). The targeted accuracy for the SMAP-NEE data product is a mean unbiased root-mean-square error (MURMSE) of 1.6 g C m^-2 day^-1, or 30 g C m^-2 year^-1, which is reported as similar to the accuracy of EC observations (Endsley et al., 2023).

The SMAP-NEE data product has been used to estimate NEE in northern regions where in-situ measurements are relatively sparse, particularly in western Siberia where there is extensive peatland coverage (Karlsson et al., 2021; Zolkos et al., 2022). Karlsson et al. (2021) found that monthly mean NEE ranged from −1.88 to 0.64 g C m^-2 day^-1 in the western Siberian lowlands, using the SMAP-NEE data product (Karlsson et al., 2021). In the Mackenzie Lowlands and Travaillant Uplands, in the Northwest Territories in Canada, Zolkos et al. (2022), estimated that NEE ranged from −0.60 to −0.82 g C m^-2 day^-1 for the period of June through August of 2016 (Zolkos et al., 2022). They highlight, however, that SMAP-NEE has higher applicability at multi-kilometer spatial scales, evidenced by their comparison of mean summertime SMAP-NEE to EC flux tower observations (Zolkos et al., 2022). While the SMAP-NEE data product has been evaluated against EC towers in the United States, Australia, and recently the Arctic Boreal Zone (Jones et al., 2017; Madelon et al., 2025), its assessment in peatlands remains limited.

In this study, we analyzed observations of net CO₂ fluxes from five Canadian peatland sites, spanning four ecoclimate zones, to determine: (1) the difference between daily SMAP-NEE estimates and daily eddy covariance NEE (EC-NEE), (2) the proportion of the net GS CO₂ uptake offset by net NGS emissions, and (3) inter-seasonal and inter-site differences in NEE trends. The quantification of NGS and GS CO₂ fluxes will improve our understanding of annual NEE dynamics in peatland ecosystems and how resilient their CO₂ sink status may be to changes in climatic conditions. This study presents the first direct comparison between EC tower measurements and the SMAP-NEE data product in peatlands, providing a critical evaluation of its performance and its suitability for estimating NEE in these ecosystems.

2 Materials and methods

2.1 Site descriptions

The AmeriFlux and Fluxnet databases were searched for peatland sites across Canada for the availability of EC-NEE data for any period aligning with the SMAP-NEE data product’s collection period for both the GS and NGS. Five peatlands had data for the period that aligned with SMAP-NEE (from 2015 onwards), including Kinoje Lake Peatland (CA-KLP) in Ontario, Robinsons Peatland Pasture (CA-RPp) in Newfoundland, Robinsons Natural Bog (CA-RPn) in Newfoundland, Delta Burns Bog (CA-DBB) in British Columbia, and Scotty Creek Bog (CA-SCB) in the Northwest Territories (Figure 1). Sites differed in their conditions, including climate and vegetation (see Tables 1, 2).

FIGURE 1

2.2 Data acquisition and processing

The eddy covariance (EC) CO₂ flux data were acquired directly from site principal investigators and the Ameriflux network. The EC instrumentation and data processing procedure differed between sites (see Supplementary Table S1). Processing of the EC data for CA-SCB and CA-DBB was performed according to the method outlined in Wang et al. (2018), as this was also how the data for CA-RPp and CA-RPn were treated prior to use for this study. This method was adopted to standardize the quality control, gap-filling, and flux partitioning procedure. The gap-filled datasets were used to calculate daily average fluxes for each site, using only days where 70% or more of the half-hourly fluxes were present. In the case of CA-KLP, the dataset was acquired as daily average fluxes pre-processed according to Humphreys et al. (2014), and as such were only post-processed by removing days with <70% of the half-hourly fluxes present. Notable data gaps, as well as the timespans used in further calculations, are presented in Supplementary Table S1. The SMAP-NEE data were acquired from the Oak Ridge National Laboratory Distributed Active Archive Center (ORNL DAAC) using NASA’s Fixed Sites Subset Tools and their satellite remote sensing and modelled data products (Kimball et al., 2022; ORNL DAAC, 2024). For each site, the SMAP pixel NEE value was assigned according to the location of the EC tower. As a result of the data product’s spatial scale, ecosystem types other than peatlands could be represented in the pixel. To account for these spatial differences, we identified the land cover class (Government of Canada, 2024) of the EC tower and determined the fractional area of this land cover class within the respective SMAP pixel (Supplementary Table S2). Then we multiplied the NEE value for the SMAP pixel by the fractional area resulting in the final SMAP-NEE variable.

The NGS was initially defined using a method based on Rafat et al. (2022), where the start of the NGS was defined as the first day of three consecutive days when the daytime air temperature fell below 1 °C, and the end of the NGS was defined as the first day of three consecutive days where the temperature rose above 1 °C. Using this method, we found that CA-DBB did not have a NGS for each year of study. Since there were periods with lower temperatures and solar radiation aligning with periods of CO₂ release occurring during the winter, we adapted our definition.

The SMAP-NEE was used to determine the GS start and end dates, based on when NEE transitions from positive to negative (Barnard et al., 2018), due to the extensive gaps in the EC-NEE data. The period inside these dates represents the GS, whereas the period outside these dates represents the NGS. To determine these dates, we used three different methods and then took the mean (Barnard et al., 2018; Zeileis and Grothendieck, 2005; Grolemund and Wickham, 2011; Wickham et al., 2023a; Wickham et al., 2023b). The first method was the variable length regression (VLR) method which calculates linear regressions on subsets of daily SMAP-NEE from three to 7 days in length during Spring (March 1 to June 21) and Fall (September 1 to December 1). The day with the steepest positive slope (SMAP-NEE to the day of year (DOY)) marks the start of the GS, whereas the day with the shallowest negative slope (SMAP-NEE to DOY) marks the end of the GS. The second method is the low smooth threshold method (LST), which marks the start of the GS as the day when an interval averaged NEE crosses and remains below a threshold (zero) across 5 days, whereas the end of the GS is marked when the interval averaged NEE remains above the threshold (zero) for 5 days, after the last possible GS start of the year. The third method is the high smooth threshold method (HST), which is functionally identical to the LST method, but for 10 days. The 9-year time series (2015–2023) was separated into years from the first GS start to the GS start of the following year, including only one GS and one NGS per year. The average start dates for both seasons are shown in Supplementary Table S3.

2.3 Data analysis

All data analyses were performed in R (R Core Team, 2023). After splitting the data into the NGS and GS, based on SMAP-NEE, the dataset was split into training and validation subsets using a stratified random sampling approach stratified by site and season, and also using only the dates where EC-NEE was available (no gaps) (Kuhn, 2008). The training data represented 75% of the EC-NEE data without gaps, whereas the validation subset represented 25%. The training data was used to run a linear regression to identify the relationship between SMAP-NEE and EC-NEE on a site-by-site basis separated seasonally. The resulting equations were applied to the original SMAP-NEE dataset to generate a Corrected-SMAP-NEE dataset, reflecting the substantial mismatch between the SMAP-NEE and EC-NEE (Table 3). This approach was adopted to quantify and correct for biases observed in the SMAP-NEE data product, consistent with our objectives of evaluating its performance in peatland ecosystems, rather than developing a new predictive model. A linear correction was selected as a transparent and interpretable method appropriate for the number of study sites and was applied on a site- and season-specific basis to capture spatial and temporal variability in model bias, thereby improving the accuracy of subsequent analyses. The validation data was used to determine how well the SMAP-NEE/Corrected-SMAP-NEE data product estimates EC-NEE, by first calculating the difference between daily SMAP-NEE/Corrected-SMAP-NEE and daily EC-NEE for all available days and then taking the average. As an additional validation metric, the root mean squared error (RMSE) was calculated, given the large sample size, to compare with the target mean unbiased RMSE for the SMAP-NEE/Corrected-SMAP-NEE data product using:where n is the number of data points, y_i is the estimated value (SMAP-NEE and Corrected-SMAP-NEE) for the ith data point, and ŷ_i is the measured value (EC-NEE) for the ith data point. Additionally, the Kruskal Wallis and Wilcoxon comparisons were used to determine the differences between different measures of daily NEE, partitioned by site, and year.

The resulting Corrected-SMAP-NEE dataset was used for the remainder of the analyses where the cumulative fluxes of the NGS to the GS were compared by calculating the percentage of net GS NEE to net NGS NEE ((net NGS NEE/net GS NEE) × 100%). The Kruskal Wallis and Wilcoxon tests were also used to determine if there were differences in annual and seasonal estimates of Corrected-SMAP-NEE between sites and to determine if there were differences in Corrected-SMAP-NEE between seasons. Statistical significance was assessed at p < 0.05. To evaluate the relative variability of respective measures of NEE, we calculated the coefficients of variability (CV), which is defined as the standard deviation divided by the mean and reported as a percentage. We then used the Kruskal–Wallis and pairwise Wilcoxon tests to determine if there were differences between these measures.

3 Results

3.1 Direct comparison of daily SMAP-NEE and EC-NEE

Comparing the daily GS SMAP-NEE to the daily GS EC-NEE (Figures 2a–e), with all sites considered, on average SMAP-NEE values were greater than EC-NEE values by 0.37 g C m^-2 day^-1 (95% confidence interval (CI)=(-0.10,0.84)) (Table 3). Greater SMAP-NEE values relative to the EC-NEE values were also found at each of the five sites (Table 3). When comparing the daily NGS SMAP-NEE to the daily NGS EC-NEE (Figures 2f–j), with all sites considered, on average SMAP-NEE values were smaller than EC-NEE values by 0.20 g C m^-2 day^-1 (95% CI= (−0.41,0.02)) (Table 3). All sites during the NGS, except CA-KLP, share this finding of SMAP-NEE values being smaller than EC-NEE values, whereas at CA-KLP, SMAP-NEE values were greater on average than EC-NEE values by 0.02 g C m^-2 day^-1 (95% CI=(-0.08,0.12)) (Table 3). With all sites considered the RMSE for the daily fluxes for SMAP-NEE during the GS was 1.06 g C m^-2 day^-1, whereas during the NGS the RMSE was 0.62 g C m^-2 day^-1. Both RMSE values are within SMAP’s target mean unbiased RMSE indicating that the model performs within their target.

FIGURE 2

The correction (Corrected-SMAP-NEE) performed on a site-by-site basis greatly reduced the variability of the SMAP-NEE fluxes, particularly during the NGS. After the correction, with all sites considered for the GS, on average Corrected-SMAP-NEE were smaller than EC-NEE values by 0.01 g C m^-2 day^-1 (95% CI= (−0.05,0.03)) (Table 3). CA-KLP, CA-RPp, and CA-DBB also share this tendency, whereas at CA-RPn and CA-SCB Corrected-SMAP-NEE values were greater than EC-NEE values by 0.03 (95% CI=(-0.19,0.24)), and 0.01 (95% CI=(-0.05,0.07)) g C m^-2 day^-1, respectively (Table 3). After the correction, with all sites considered for the NGS, on average Corrected-SMAP-NEE were smaller than EC-NEE values by 0.002 g C m^-2 day^-1 (95% CI= (−0.08,0.07)). CA-KLP, CA-RPn, and CA-DBB share this tendency, whereas at CA-RPp, and CA-SCB Corrected-SMAP-NEE values were greater than EC-NEE values by on average 0.09 (95% CI=(-0.01,0.19)), and 0.01 (95% CI=(-0.02,0.04)) g C m^-2 day^-1, respectively (Table 3). Additionally, the RMSE with all sites considered for the daily fluxes for the Corrected-SMAP-NEE during the GS was 0.94 g C m^-2 day^-1, whereas during the NGS 0.57 g C m^-2 day^-1. These RMSE values are lower than those for the SMAP-NEE dataset, representing a more accurate model estimation of measured fluxes. For the coefficients of variation, we found that SMAP-NEE after correcting for LULC within the pixel had the highest CV followed by unprocessed SMAP-NEE (Supplementary Table S4). The variability of the SMAP-NEE dataset was greatly reduced when we implemented the site-by-site correction, but it better reflected the variability of the EC-NEE dataset (Supplementary Table S4).

The target mean unbiased RMSE for SMAP-NEE is less than or equal to 1.6 g C m^-2 day^-1, which they report as similar to the accuracy of EC observations (Endsley et al., 2023). The unprocessed SMAP-NEE dataset meets this target where the RMSE for the GS is 1.06 g C m^-2 day^-1, but for the NGS is 0.62 g C m^-2 day^-1. However, given the lower productivity of peatlands, reaching these targets could still lead to erroneous source or sink attributions at these sites. We found that based on mean daily NEE, SMAP-NEE generally captured the interannual variability in shifting from sink to source (Supplementary Figure S1). Based on annual NEE SMAP-NEE, these differences in measures of NEE have a cumulative effect in which SMAP-NEE does not capture the sink or source trend of that site in that year (Supplementary Figure S1). The correction (Corrected-SMAP-NEE) reduced the RMSE in the GS to 0.94 g C m^-2 day^-1 and to 0.57 g C m^-2 day^-1 for the NGS. Overall, the correction reduced the difference between the SMAP-NEE and EC-NEE fluxes, resulting in seasonal and annual budgets that better reflect the in-situ environment (Table 4).

3.2 NGS emissions compared to GS uptake for corrected-SMAP-NEE

The cumulative NEE for the entire study period, as well as the mean cumulative and daily Corrected-SMAP-NEE fluxes of CO₂ for the five peatland sites are presented in Table 4. These fluxes are further divided into NGS, GS, and annual fluxes. The GS overall consistently represented a period of CO₂ uptake, whereas the NGS overall consistently represented a period of CO₂ output, despite there being some variability at the daily scale (Table 4). At CA-KLP, net NGS emissions accounted for on average 33.0% (95% CI=(22.5,43.7)) of net GS uptake resulting in the site functioning as an annual CO₂ sink for all years studied. At the CA-RPn site, net NGS emissions accounted for on average 255.9% (95% CI=(223.4,288.4)) of net GS uptake, resulting in the site functioning as an annual CO₂ source for all years studied. Moreover, at the CA-RPp site, net NGS emissions accounted for on average 74.5% (95% CI=(65.2,83.8)) of net GS uptake, resulting in the site functioning as an annual CO₂ sink for all years studied. At CA-SCB, net NGS emissions accounted for on average 110.7% (95% CI=(93.5,127.9)) of net GS uptake, and the site functioned as an annual CO₂ sink from 2015-2018, then functioned as an annual CO₂ sink from 2018-2023. Finally, at CA-DBB, net NGS emissions accounted for on average 45.6% (95% CI=(42.2,49.0)) of net GS uptake, resulting in the site functioning as an annual CO₂ sink for all years studied.

3.3 Inter-seasonal and site differences for corrected-SMAP-NEE

The boxplots for Corrected-SMAP-NEE separated by period (GS, NGS, and annual fluxes) for the five peatland sites are presented in Figure 3. Some notable patterns were that Corrected-SMAP-NEE during the GS was always negative, whereas Corrected-SMAP-NEE during the NGS was always positive. However, annual Corrected-SMAP-NEE fluctuated between being positive or negative (Figure 3). Notably, with all sites considered, the Corrected-SMAP-NEE fluxes were always significantly more positive during the NGS, compared to the negative values during the GS (Wilcoxon rank sum, p < 0.001, W = 0) (Table 3). There were significant differences in Corrected-SMAP-NEE between sites during both the GS and NGS, as well as annually (Kruskal–Wallis and Pairwise Wilcox Test: F = 39.7, p < 0.0001; F = 40.2, p < 0.0001; and F = 35.0, p < 0.0001, respectively).

FIGURE 3

4 Discussion

4.1 Limitations of the SMAP-NEE data product

The findings in this study highlight the discrepancy between in-situ NEE fluxes measured in northern peatlands and those estimated by the SMAP data product. Conversely, in this study the SMAP-GPP values during the GS and NGS had generally good agreement with the values of EC-GPP (Supplementary Figure S2), consistent with global SMAP-GPP estimates matching well with the literature (Endsley et al., 2023). There are a number of uncertainties that could contribute to errors in the SMAP data product, including errors stemming from inputs and uncertainties associated with model parameterization, initialization, and calibration (Kimball et al., 2022). The SMAP derived C fluxes have been derived by calibrating the SMAP C cycling model outputs via comparison against EC tower measurements from global FLUXNET sites representing major plant functional type classes (Kimball et al., 2022). Neither wetlands nor peatlands are explicitly represented in the plant functional type classes used to calibrate SMAP, as well as other land cover types may be represented within the large 9 by 9 km pixel size modelled by SMAP (Jones et al., 2017). The coarse spatial resolution of SMAP introduces a mismatch with EC flux tower footprints, which typically sample much smaller and more homogenous areas. As a result, sub-pixel heterogeneity in SMAP pixels may mask the individual contributions of specific land cover types (i.e., peatlands) leading to the observed discrepancies between SMAP-NEE and EC-NEE that, when aggregated, lead to erroneous sink or source attributions (Supplementary Figure S1). Version 7 of the validation assessment report for the SMAP data product (Endsley et al., 2023) detailed that they specifically validated the data product in peatlands by comparing the derived fluxes to peatland FLUXNET tower sites and reported that SMAP data generally reproduces the patterns of peatland sites, whereby peatlands have lower seasonal NEE amplitudes compared to non-peatland sites. Our finding that the SMAP-NEE values were higher than EC-NEE values during the GS and lower during the NGS demonstrates that using the uncorrected SMAP-NEE dataset would result in peatlands being estimated to be stronger GS CO₂ sources and weaker NGS CO₂ sinks. We propose that further validation exercises of the SMAP-NEE data product leverage the wealth of publicly available globally distributed peatland EC-NEE data and make these comparisons available to end users. Moreover, the EC-derived NEE data may introduce additional sources of uncertainty and potential bias. Although efforts were made to standardize processing, the data were subjected to site-specific preprocessing prior to inclusion in AmeriFlux. Differences among sites in record length, data completeness, and the relative representation of GS versus NGS conditions may influence model performance and lead to site-dependent variability in results. These factors may lead to site-dependent variability in model performance and could bias comparisons across sites. As a result, part of the observed variability in model performance may reflect differences in data quality and representativeness rather than model performance alone. These uncertainties should be considered when interpreting the results.

4.2 Correcting SMAP-NEE is valuable, permitting data availability

The SMAP data product is advantageous because it (1) enables the estimation of peatland C fluxes throughout both the GS and NGS, with no limitations due to equipment failure during the harsh conditions of the NGS in cold regions, and (2) allows for inexpensive data collection throughout the NGS and GS and across years in hard-to-access high latitude peatlands. However, our findings show that the C fluxes derived by the SMAP data product have their limitations. Additionally, while the correction of the SMAP data improved the accuracy of the SMAP dataset, it is likely that several years of in-situ data per site would be needed to have the correction be meaningful. If there were several years of EC-NEE data, as well as additional meteorological information resulting from the EC method, a better route for long-term assessment of C cycling trends would be to generate models on a site-by-site basis. Moreover, while this study leveraged EC flux data from Canadian peatland sites temporally coinciding with the SMAP-NEE observation period, future work could extend this approach by incorporating global peatland observations. An area for future extension would be, upon the inclusion of additional peatland sites, using the relationship between EC-NEE and SMAP-NEE at all sites to correct SMAP-NEE (Supplementary Table S5; Supplementary Figure S3), and extrapolating this correction relationship to additional peatland sites. The relative similarity of the linear regression parameters between the EC-NEE and SMAP-NEE used for the SMAP data correction across the 5 sites suggests that such a general correction relationship for peatland sites may be possible. However, the agreement between the Corrected-SMAP-NEE and EC-NEE was better on a site-by-site basis. Future work should consider implementing a spatial or temporal blocking approach to provide a more independent evaluation of model performance. Such approaches were not feasible with the dataset used in this study and may result in more conservative performance estimates than those reported here. Our findings suggest that the SMAP-NEE data product should be corrected prior to generating estimates for peatland sites. In addition, machine learning approaches could be explored in future work to better capture potentially non-linear relationships between SMAP-NEE and EC-NEE. This approach may improve upon the relatively limited explanatory power of the linear correction applied here, despite its improvement over the uncorrected product.

4.3 CO₂ sink strength varies with NGS emissions

The data analyses in this study showed that net NGS CO₂ release represented between 33% and 256% of the net GS CO₂ uptake. This range was higher than that reported by Zhao et al. (2016), who found that NGS CO₂ emissions from four studies accounted for 16%–80% of GS CO₂ uptake in boreal peatlands. Specifically, at CA-KLP, CA-RPp, and CA-DBB GS CO₂ uptake offset by NGS CO₂ emissions fell within the range reported by Zhao et al. (2016), whereas CA-SCB was slightly higher than their reported range (110.7%), and CA-RPn was over 3X higher than the maximum range value. The two Newfoundland Sites (CA-RPp and CA-RPn) had the highest daily NGS emissions (0.62, and 0.45 g C m^-2 day^-1, respectively) despite having the shortest NGS period. We hypothesize this result comes from climatic differences as the two Newfoundland sites had substantially higher mean annual snowfall (Tables 1, 2). Snow can be insulating resulting in higher soil temperatures that facilitate higher microbial activity, and thereby CO₂ emissions (Gouttevin et al., 2012; Zhao et al., 2022). CA-RPp offset these high NGS emissions due to fast growing vegetation, resulting in a proportion that falls within the expected range, however CA-RPn had typical bog vegetation and therefore resulted in the high emissions as they were not offset during the GS (−1.21, and −0.26 g C m^-2 day^-1, respectively) (Wang et al., 2018). The observed site-level differences are not primarily attributable to the SMAP-NEE algorithm’s sensitivity to vegetation and climate. Notably, sites located within neighbouring SMAP pixels do not exhibit distinct SMAP-NEE, despite clear differences in EC-NEE fluxes. Instead, these differences are largely driven by the site-specific correction applied in this study.

The estimated annual mean cumulative NEEs were ranged from −60.7 to 58.4 g C m^-2 yr^-1 across the 5 sites (Table 4), which are in line with measurements determined using EC towers in the range of −90–100 g C m^-2 yr^-1 reported by Limpens et al. (2008). These estimates are also in line with the mean annual NEE of −48 g C m^-2 yr^-1 (95% CI=(-171,75) calculated from EC measurements in high latitude peatlands (>45°N) (Strack et al., 2023). Additionally, the analyses in this study showed that the five peatlands differed in their annual CO₂ sink capacity throughout the study periods (Table 4; Figure 3). This finding contrasts with a study by Lund et al. (2010) that synthesized EC data for 12 arctic wetland sites, maintaining that all sites acted as annual sinks for CO₂. These results suggest that applying a correction to the SMAP-NEE data product improves its ability to capture the range and variability of peatland CO₂ exchange reported in EC-based studies. However, this improvement is contingent on the availability of in situ EC data and is therefore primarily applicable in regions where sufficient observations exist to support this correction. Even with the correction, substantial mismatches remain, indicating that the product may not be suitable for generating reliable site-level estimates in peatlands.

5 Conclusion

We tested the remotely sensed and modeled SMAP-NEE data product to estimate NEE for five Canadian peatland sites but found it necessary to correct the data product based on a subset of eddy covariance data, due to data mismatch of the tower data and the original product. Based on the corrected data, we found the NGS to represent a period of net CO₂ emission, whereas the GS represented a period of net CO₂ uptake. The proportion of NGS emissions compared to GS uptake varied substantially on a site-by-site basis ranging from 33% to 256%. Our focus on seasonality and identifying different correction factors for either season represents a novel approach, and our findings highlight the importance of considering the NGS when evaluating annual peatland C budgets. Furthermore, while peatlands are likely sinks in the long term, their annual sink status is dynamic, and multi-year measurements are crucial for reducing the uncertainty in the estimates of a site’s source/sink status. Peatland specific plant functional types should be incorporated into the model in order for the product to be usable in this application, and further validation exercises of the SMAP-NEE data product should leverage the wealth of publicly available globally distributed peatland EC-NEE data and make these comparisons available to end users.

References

• 1

  BaldocchiD.ChuH.ReichsteinM. (2018). Inter-annual variability of net and gross ecosystem carbon fluxes: a review. Agric. For. Meteorol.249, 520–533. 10.1016/j.agrformet.2017.05.015

  • CrossRef
  • Google Scholar

• 2

  BarnardD. M.KnowlesJ. F.BarnardH. R.GouldenM. L.HuJ.LitvakM. E.et al (2018). Reevaluating growing season length controls on net ecosystem production in evergreen conifer forests. Sci. Rep.8, 17973. 10.1038/s41598-018-36065-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3

  ChapinF. S.WoodwellG. M.RandersonJ. T.RastetterE. B.LovettG. M.BaldocchiD. D.et al (2006). Reconciling carbon-cycle concepts, terminology, and methods. Ecosystems9, 1041–1050. 10.1007/s10021-005-0105-7

  • CrossRef
  • Google Scholar

• 4

  ChristenA.KnoxS. (2021). AmeriFlux BASE CA-DBB Delta burns bog, ver. 2-5, AmeriFlux AMP. 10.17190/AMF/1543378

  • CrossRef
  • Google Scholar

• 5

  ChristenA.JassalR. S.BlackT. A.GrantN. J.HawthorneI.JohnsonM. S.et al (2016). Summertime greenhouse gas fluxes from an urban bog undergoing restoration through rewetting. Mires Peat17, 1–24. 10.19189/MaP.2015.OMB.207

  • CrossRef
  • Google Scholar

• 6

  CofferM. M.HestirE. L. (2019). Variability in trends and indicators of CO₂ exchange across arctic wetlands. J. Geophys. Res. Biogeosci.124, 1248–1264. 10.1029/2018jg004775

  • CrossRef
  • Google Scholar

• 7

  DutchV. R.RutterN.WakeL.SonnentagO.Hould GosselinG.SandellsM.et al (2024). Simulating net ecosystem exchange under seasonal snow cover at an arctic tundra site. Biogeosci21, 825–841. 10.5194/bg-21-825-2024

  • CrossRef
  • Google Scholar

• 8

  D’AcunhaB.MorillasL.BlackT. A.ChristenA.JohnsonM. S. (2019). Net ecosystem carbon balance of a peat bog undergoing restoration: integrating CO₂ and CH₄ fluxes from eddy covariance and aquatic evasion with DOC drainage fluxes. J. Geophys. Res. Biogeosci.124, 884–901. 10.1029/2019JG005123

  • CrossRef
  • Google Scholar

• 9

  EndsleyK. A.KimballJ. S.KundigT.ReichleR. H.ArdizzoneJ. V. (2023). Validation assessment for the soil moisture active passive (SMAP) level 4 carbon (L4_C) data product version 7. Boulder, Colorado USA: NASA National Snow and Ice Data Center Distributed Active Archive Center. Available online at: https://nsidc.org/sites/default/files/documents/technical-reference/nasa_smap_l4c_assessment_v7.pdf (Accessed January 13, 2025).

  • Google Scholar

• 10

  Environment and Climate Change Canada (2023). 1981-2010 climate normals & averages. Climate. Available online at: https://climate.weather.gc.ca/climate_normals/index_e.html (Accessed January 13, 2025).

  • Google Scholar

• 11

  Environmental Protection Agency (2024). Ecoregions of North America. EPA. Available online at: https://www.epa.gov/eco-research/ecoregions-north-america (Accessed January 13, 2025).

  • Google Scholar

• 12

  FriedlM.Sulla-MenasheD. (2019). “MCD12Q1 MODIS/terra+Aqua land cover type yearly L3 global 500m SIN grid V006,” in NASA land processes distributed active archive center. 10.5067/MODIS/MCD12Q1.006

  • CrossRef
  • Google Scholar

• 13

  Gallego-SalaA. V.CharmanD. J.BrewerS.PageS. E.PrenticeI. C.FriedlingsteinP.et al (2018). Latitudinal limits to the predicted increase of the peatland carbon sink with warming. Nat. Clim. Change8, 907–913. 10.1038/s41558-018-0271-1

  • CrossRef
  • Google Scholar

• 14

  GiglioL. (2024). VIIRS/NPP active fires 6-Min L2 swath 375m V002. Sioux Falls, SD: NASA Land Processes Distributed Active Archive Center. 10.5067/VIIRS/VNP14IMG.002

  • CrossRef
  • Google Scholar

• 15

  GorhamE. (1991). Northern peatlands: role in the carbon cycle and probable responses to climatic warming. Ecol. Appl.1, 182–195. 10.2307/1941811

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 16

  GouttevinI.MenegozM.DominéF.KrinnerG.KovenC.CiaisP.et al (2012). How the insulating properties of snow affect soil carbon distribution in the Continental pan-arctic area. J. Geophys. Res. Biogeosci.117, G02020. 10.1029/2011JG001916

  • CrossRef
  • Google Scholar

• 17

  Government of Canada (2024). 2020 land cover of Canada. Open Gov. Portal. Available online at: https://open.canada.ca/data/en/dataset/ee1580ab-a23d-4f86-a09b-79763677eb47 (Accessed January 13, 2025).

  • Google Scholar

• 18

  GrolemundG.WickhamH. (2011). Dates and times made easy with lubridate. J. Stat. Softw.40, 1–25. 10.18637/jss.v040.i03

  • CrossRef
  • Google Scholar

• 19

  HaslwanterA.HammerleA.WohlfahrtG. (2009). Open-path vs. closed-path eddy covariance measurements of the net ecosystem carbon dioxide and water vapour exchange: A long-term perspective. Agric. For. Meteorol.149, 291–302. 10.1016/j.agrformet.2008.08.011

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 20

  HelbigM.ChasmerL. E.DesaiA. R.KljunN.QuintonW. L.SonnentagO. (2017). Direct and indirect climate change effects on carbon dioxide fluxes in a thawing boreal forest–wetland landscape. Glob. Change Biol.23, 3231–3248. 10.1111/gcb.13638

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 21

  HobbieJ. E.GeorgeW. K. (2014). Alaska's changing arctic: ecological consequences for tundra, streams, and lakes. Oxford University Press.

  • Google Scholar

• 22

  HuG.ZhaoL.WuT.WuX.ParkH.FedorovA.et al (2021). Spatiotemporal variations and regional differences in air temperature in the permafrost regions in the northern hemisphere during 1980–2018. Sci. Total Environ.791, 148358. 10.1016/j.scitotenv.2021.148358

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 23

  HumphreysE. R.CharronC.BrownM.JonesR. (2014). Two bogs in the Canadian Hudson Bay lowlands and a temperate bog reveal similar annual net ecosystem exchange of CO₂. Arct. Antarct. Alp. Res.46, 103–113. 10.1657/1938-4246.46.1.103

  • CrossRef
  • Google Scholar

• 24

  JensenA. M.WarrenJ. M.KingA. W.RicciutoD. M.HansonP. J.WullschlegerS. D. (2019). Simulated projections of boreal forest peatland ecosystem productivity are sensitive to observed seasonality in leaf physiology. Tree Physiol.39, 556–572. 10.1093/treephys/tpy140

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 25

  JeongJ.-H.KugJ.-S.KimB.-M.MinS.-K.LinderholmH. W.HoC.-H.et al (2012). Greening in the circumpolar high-latitude may amplify warming in the growing season. Clim. Dyn.38, 1421–1431. 10.1007/s00382-011-1142-x

  • CrossRef
  • Google Scholar

• 26

  JonesL. A.KimballJ. S.ReichleR. H.MadaniN.GlassyJ.ArdizzoneJ. V.et al (2017). The SMAP level 4 carbon product for monitoring ecosystem land-atmosphere CO₂ exchange. IEEE Trans. Geosci. Remote Sens.55, 6517–6532. 10.1109/TGRS.2017.2729343

  • CrossRef
  • Google Scholar

• 27

  KarlssonJ.SerikovaS.VorobyevS. N.Rocher-RosG.DenfeldB.PokrovskyO. S. (2021). Carbon emission from Western Siberian inland waters. Nat. Commun.12, 825. 10.1038/s41467-021-21054-1

  • CrossRef
  • Google Scholar

• 28

  KimballJ. S.JonesL. A.EndsleyA.KundigT.ReichleR. (2022). SMAP L4 global daily 9 km EASE-grid carbon net ecosystem exchange, version 7. Boulder, Colorado USA: NASA National Snow and Ice Data Center Distributed Active Archive Center. 10.5067/3K9F0S1Q5J2U

  • CrossRef
  • Google Scholar

• 29

  KuhnM. (2008). Building predictive Models in R using the caret package. J. Stat. Softw.28, 1–26. 10.18637/jss.v028.i05

  • CrossRef
  • Google Scholar

• 30

  LafleurP. M.RouletN. T.BubierJ. L.FrolkingS.MooreT. R. (2003). Interannual variability in the peatland-atmosphere carbon dioxide exchange at an ombrotrophic bog. Glob. Biogeochem. Cycles17, 1036. 10.1029/2002gb001983

  • CrossRef
  • Google Scholar

• 31

  LeeS. C.ChristenA.BlackA. T.JohnsonM. S.JassalR. S.KetlerR.et al (2017). Annual greenhouse gas budget for a bog ecosystem undergoing restoration by rewetting. Biogeosci14, 2799–2814. 10.5194/bg-14-2799-2017

  • CrossRef
  • Google Scholar

• 32

  LimpensJ.BerendseF.BlodauC.CanadellJ. G.FreemanC.HoldenJ.et al (2008). Peatlands and the carbon cycle: from local processes to global implications – a synthesis. Biogeosci5, 1475–1491. 10.5194/bg-5-1475-2008

  • CrossRef
  • Google Scholar

• 33

  LiuQ.PiaoS.JanssensI. A.FuY.PengS.LianX.et al (2018). Extension of the growing season increases vegetation exposure to frost. Nat. Commun.9, 426–428. 10.1038/s41467-017-02690-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 34

  LiuL.ChenH.LiuX.YangZ.ZhuD.HeY.et al (2019). Contemporary, modern and ancient carbon fluxes in the zoige peatlands on the Qinghai-Tibetan Plateau. Geoderma352, 138–149. 10.1016/j.geoderma.2019.06.008

  • CrossRef
  • Google Scholar

• 35

  LundM.LafleurP. M.RouletN. T.LindrothA.ChristensenT. R.AurelaM.et al (2010). Variability in exchange of CO₂ across 12 northern peatland and tundra sites. Glob. Change Biol.16, 2436–2448. 10.1111/j.1365-2486.2009.02104.x

  • CrossRef
  • Google Scholar

• 36

  LuntP. H.FyfeR. M.TappinA. D. (2019). Role of recent climate change on carbon sequestration in peatland systems. Sci. Total Environ.667, 348–358. 10.1016/j.scitotenv.2019.02.239

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 37

  MadelonR.KimballJ. S.EndsleyK. A.De LannoyG. J. M.SonnentagO.AlcockH.et al (2025). Assessing the SMAP level-4 carbon product over the arctic and subarctic zones. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.18, 9850–9864. 10.1109/JSTARS.2025.3555850

  • CrossRef
  • Google Scholar

• 38

  NataliS. M.WattsJ. D.RogersB. M.PotterS.LudwigS. M.SelbmannA.-K.et al (2019). Large loss of CO₂ in winter observed across the northern permafrost region. Nat. Clim. Change9, 1005. 10.1038/s41558-019-0644-0

  • CrossRef
  • Google Scholar

• 39

  NWWG - National Wetlands Working Group (1988). Wetlands of Canada. Ecological land classification series no. 24. Sustainable development branch, environment Canada. Ottawa, Ontario, Canada: Polyscience Publications, Inc.

  • Google Scholar

• 40

  ORNL DAAC (2018). Fixed sites subsetting and visualization tool. Oak Ridge, Tennessee, USA: ORNL DAAC. 10.3334/ORNLDAAC/1567

  • CrossRef
  • Google Scholar

• 41

  ORNL DAAC (2024). Fixed sites subsets tool. MODIS/VIIRS subsets. Available online at: https://modis.ornl.gov/sites (Accessed January 13, 2025).

  • Google Scholar

• 42

  PostW. M.EmanuelW. R.ZinkeP. J.StangenbergerA. G. (1982). Soil carbon pools and world life zones. Nature298, 156–159. 10.1038/298156a0

  • CrossRef
  • Google Scholar

• 43

  R Core Team (2023). _R: a language and environment for statistical computing_. Vienna, Austria: R Foundation for Statistical Computing. Available online at: https://www.R-project.org (Accessed January 13, 2025).

  • Google Scholar

• 44

  RafatA.RezanezhadF.QuintonW. L.HumphreysE. R.WebsterK.Van CappellenP. (2021). Non-growing season carbon emissions in a northern peatland are projected to increase under global warming. Commun. Earth Environ.2, 111. 10.1038/s43247-021-00184-w

  • CrossRef
  • Google Scholar

• 45

  RafatA.ByunE.RezanezhadF.QuintonW. L.HumphreysE. R.WebsterK.et al (2022). The definition of the non-growing season matters: a case study of net ecosystem carbon exchange from a Canadian peatland. Environ. Res. Commun.4, 021003. 10.1088/2515-7620/ac53c2

  • CrossRef
  • Google Scholar

• 46

  ReichleR.De LannoyG.KosterR. D.CrowW. T.KimballJ. S.LiuQ.et al (2025). SMAP L4 global 3-hourly 9 km EASE-grid surface and root zone soil moisture geophysical data. (SPL4SMGP, version 8). NASA Natl. Snow Ice Data Cent. Distributed Act. Archive Cent.10.5067/T5RUATAQREF8

  • CrossRef
  • Google Scholar

• 47

  RieneckerM. M.SuarezM. J.TodlingR.BacmeisterJ.TakacsL.LiuH.-C.et al (2008). The GEOS-5 data assimilation system - documentation of versions 5.0.1, 5.1.0, and 5.2.0 in Technical report series on global modeling and data assimilation, 27.

  • Google Scholar

• 48

  SatriawanT. W.NybergM.LeeS.-C.ChristenA.BlackT. A.JohnsonM. S.et al (2023). Interannual variability of carbon dioxide (CO₂) and methane (CH₄) fluxes in a rewetted temperate bog. Agric. For. Meteorol.342, 109696. 10.1016/j.agrformet.2023.109696

  • CrossRef
  • Google Scholar

• 49

  SonnentagO.QuintonW. (2021). AmeriFlux BASE CA-SCB scotty creek bog, ver, 2-5. Berkeley, CA: AmeriFlux AMP. 10.17190/AMF/1498754

  • CrossRef
  • Google Scholar

• 50

  StrackM.LupascuM.ByrneK. A.RouletN.TuretskyM.WaddingtonJ. M. (2023). “Peatlands, greenhouse gas exchange and climate change,” in Peatlands and climate change. 2nd Edition (Jyväskylä, Finland: International Peatland Society).

  • Google Scholar

• 51

  UNEP (2022). Global peatlands assessment – the state of the World’s peatlands: evidence for action toward the conservation, restoration, and sustainable management of peatlands. in Summary for policy makers. Global peatlands initiative (Nairobi, Kenya: United Nations Environment Programme).

  • Google Scholar

• 52

  WangM.WuJ.LafleurP.LuanJ.ChenH.ZhuX. (2018). Can abandoned peatland pasture sequestrate more carbon dioxide from the atmosphere than an adjacent pristine bog in Newfoundland, Canada?Agric. For. Meteorol.248, 91–108. 10.1016/j.agrformat.2017.09.010

  • CrossRef
  • Google Scholar

• 53

  WardE. J.WarrenJ. M.McLennanD. A.DusengeM. E.WayD. A.WullschlegerS. D.et al (2019). Photosynthetic and respiratory responses of two bog shrub species to whole ecosystem warming and elevated CO₂ at the boreal-temperate ecotone. Front. For. Glob. Change2, 54. 10.3389/ffgc.2019.00054

  • CrossRef
  • Google Scholar

• 54

  WattsJ. D.KimballJ. S.ParmentierF. J. W.SachsT.RinneJ.ZonaD.et al (2014). A satellite data driven biophysical modeling approach for estimating northern peatland and tundra CO₂ and CH₄ fluxes. Biogeosci11, 1961–1980. 10.5194/bg-11-1961-2014

  • CrossRef
  • Google Scholar

• 55

  WickhamH.FrançoisR.HenryL.MüllerK.VaughanD. (2023a). Dplyr: a grammar of data manipulation. R package version 1.1.2. Available online at: https://CRAN.R-project.org/package=dplyr (Accessed January 13, 2025).

  • Google Scholar

• 56

  WickhamH.VaughanD.GirlichM. (2023b). Tidyr: Tidy messy data. R package. Available online at: https://CRAN.R-project.org/package=tidyr (Accessed January 13, 2025).

  • Google Scholar

• 57

  YuZ.LoiselJ.BrosseauD. P.BeilmanD. W.HuntS. J. (2010). Global peatland dynamics since the last glacial maximum. Geophys. Res. Lett.37, L13402. 10.1029/2010GL043584

  • CrossRef
  • Google Scholar

• 58

  ZeileisA.GrothendieckG. (2005). Zoo: S3 infrastructure for regular and irregular time series. J. Stat. Softw.14, 1–27. 10.18637/jss.v014.i06

  • CrossRef
  • Google Scholar

• 59

  ZhaoJ.PeichlM.ÖquistM.NilssonM. B. (2016). Gross primary production controls the subsequent winter CO₂ exchange in a boreal peatland. Glob. Change Biol.22, 4028–4037. 10.1111/gcb.13308

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 60

  ZhaoZ.De FrenneP.PeñuelasJ.Van MeerbeekK.FornaraD. A.PengY.et al (2022). Effects of snow cover-induced microclimate warming on soil physicochemical and biotic properties. Geoderma423, 115983. 10.1016/j.geoderma.2022.115983

  • CrossRef
  • Google Scholar

• 61

  ZolkosS.TankS. E.KokeljS. V.StrieglR. G.ShakilS.VoigtC.et al (2022). Permafrost landscape history shapes fluvial chemistry, ecosystem carbon balance, and potential trajectories of future change. Glob. Biogeochem. Cycles36, e2022GB007403. 10.1029/2022GB007403

  • CrossRef
  • Google Scholar