The study area is the Stordalen mire in northern Sweden (68°22^′ N 19°03^′ E; Fig. 1). The mean annual temperature of the area was -0.5 °C and mean annual precipitation sum was 204 mm for period 1985–2023, based on measurements from a nearby meteorological station (Polarforskningssekretariatet and SMHI, 2025a, b). The area contains a permafrost thaw gradient ranging from palsa with intact permafrost to partly and fully thawed permafrost, each of which includes hydrological conditions and vegetation compositions typical for each thaw stage. The permafrost is largely intact in drained palsa with ericaceous shrubs (e.g., Empetrum nigrum, Andromeda polifolia, Rubus chamaemorus), hummock-forming Eriophorum vaginatum, mosses (e.g., Dicranum sp.) and lichens (Holmes et al., 2022; Hough et al., 2022; Malhotra and Roulet, 2015). In the partly thawed hydrologically perched stage, the vegetation is dominated by Sphagnum spp. and feather mosses as well as sedges, such as Eriophorum vaginatum and Carex rotundata, typical for ombrotrophic bogs (Holmes et al., 2022; Malhotra and Roulet, 2015). The fully thawed stage includes wet minerotrophic fens that have generally higher pH and their dominant vegetation includes Eriophorum angustifolium, Equisetum fluviatile and Carex rostrata (Holmes et al., 2022; Johansson et al., 2006; Malhotra and Roulet, 2015). In general, moisture increases from intact to fully thawed permafrost stages, due to improved connections to the ground water (2007–2017 mean annual water table level ± standard deviation at partly thawed stage: 9.1 ± 4.6 cm below peat surface, fully thawed stage: 2.1 ± 4.4 cm above peat surface) (Crill et al., 2023; Holmes et al., 2022; Malhotra and Roulet, 2015). The peat depth reaches down to 3 m and is underlain by silt (Johansson et al., 2006).

We took peat cores using a push corer (d= 12 cm, length = 45 cm) to obtain plant material for belowground trait measurements. Despite destructive sampling being restricted, we were able to obtain three cores per thaw stage (n= 9 cores in total). In addition, peat coring was not possible next to the CH₄ flux chambers and within the eddy covariance footprint, so we estimated the vegetation composition (vascular plant and moss species) in the chamber collars with vegetation surveys (point intercept method; Sect. A1, Table C1) and found representative vegetation patches for each chamber outside of the footprint. In the plot selection, emphasis was put on vascular plant species and secondarily on moss species cover. The peat cores were collected on the 21 July 2023. The tightly packed samples were put into a cooler with ice packs and stored frozen for four days. The samples were transported from Sweden to University of Zurich, Switzerland (11 h), and stored in a freezer in -20 °C until further processing.

To compare the chamber and peat coring plots, and to provide a non-destructive aboveground plant trait comparison to the belowground traits, we estimated species-specific vascular plant green area (GA; Tables 1 and C2). We randomly assigned three (peat coring plots) to five (chamber plots) subquadrats (16.8×16.8 cm) of the vegetation survey grid (Sect. A1) for GA measurements using the function sample in R (R Core Team, 2025). We measured green leaf and stem dimensions (leaf width and length, stem length and diameter) per species within each subquadrat. We measured 10 leaves per species per subquadrat, with the exception of fully thawed stage chamber plots where we measured all leaves and stems. For E. nigrum, we estimated the total number of leaves by counting all the leaves of three stems roughly representing the majority of the stems within the subquadrat. We calculated species-specific leaf and stem area (m2) based on geometric formulae, and GA was estimated based on the average number of green leaves and stems per m² multiplied by green leaf and stem area per m² (Wilson et al., 2007). Since the leaves of Rubus chamaemorus did not fit into a simple geometric formula, we estimated their green area using photos taken above the survey grid and ImageJ.JS v0.5.8 (Ouyang et al., 2019; Schneider et al., 2012) (see mean GA per species in Table C2).

In order to compare the abiotic environmental characteristics of the peat coring plots with the chamber plots, we measured peat temperature (°C), pH, water table level (cm) and active layer depth (cm) at the individual plots (Table 2). Peat temperature was measured 2–3 times a day (08:00 a.m., 12:00 p.m., 05:00 p.m.) at 10 cm depth over a three-day period (23–25 July 2023) using a soil temperature probe (Testo 720, SN 03679207, Testo, Titisee-Neustadt, Germany). Active layer depth was measured with a metal rod.

We measured pH on 25 July 2023. For the intact and partly thawed plots, we took ca. 8.5 g of peat from 0.5 m distance from the chamber collar at ca. 5 cm depth (intact) or 7–10 cm depth (partly thawed). The pH was measured using a pH meter (826 pH mobile, SN 1826001012160, Metrohm, Herisau, Switzerland) and pH electrode (Polyplast, Ref 238380, SN13208, Hamilton, Reno, USA) by mixing the peat sample with CaCl₂ solution (10 mL 0.01 mol L⁻¹ CaCl₂ per sample). At fully thawed plots, pH was measured directly from open water at ca. 5 cm depth. The electrode was always kept in liquid and when not in measurement solution, it was kept in a protective tube with KCl.

Half-hourly peat temperature data was measured with thermocouples installed at each chamber and acquired with CR10x data logger (Campbell Scientific Inc., UT, USA) (Bäckstrand et al., 2008; McCalley et al., 2014; Varner et al., 2026). As continuous soil moisture and water table level data were not available for the productive season in 2023, we used 10-year (2007–2017) water table depth (WTD) means and standard deviations per thaw stage from Crill et al. (2023) for a rough background estimate in soil moisture conditions. To estimate relationships between peat moisture and CH₄ fluxes, we also obtained soil moisture (volumetric water content, %VWC; Biasi et al., 2026) estimates for each chamber using soil moisture sensors (TMS-4 Standard, TOMST s.r.o., Prague, Czech Republic; Wild et al., 2019) at ca. 8 cm depth at ca. 1 m distance from the chambers. The soil moisture values were calibrated to peat using the myClim R package (Man et al., 2023; Wild et al., 2019). To obtain a representative estimate of the central tendency of the dominant peat moisture conditions at each chamber, we used overall chamber medians of daily-aggregated values (non-normal and distributions varied between chambers) from the May–August 2024 and 2025 periods.

We cut the frozen peat cores horizontally into minimum 10 cm increments, and then further into four vertical parts (ca. 25 %) using a bandsaw. We randomly chose one of the quarters as a subsample for trait measurements. We thawed the subsamples in a refrigerator over 2–3 d and separated the aboveground vascular plants (see species-specific biomass in Table C3). We picked all living roots (light in color and elastic) and rhizomes (stiff, firm bark) from the subsample using tweezers and jeweler's glasses (2.5× magnification). Due to the very high amount of shrub roots, we took a further subsample (ca. 15 g fresh weight) from which we picked all shrub roots. We cleaned the roots and rhizomes with deionized water and grouped them into herbaceous and shrub PFTs based on a combination of branching patterns, color, and thickness (e.g., shrub roots were generally thinner and had higher branching density than herbaceous roots). The forb Rubus chamaemorus was grouped into shrub PFT (e.g., Blume-Werry et al., 2019), due to root identification challenges. We did not separate coarse and fine roots but instead focused on the average root system properties. The picked roots were stored in 70 % ethanol in a refrigerator (4 °C, max. 4 months).

We put the cleaned roots and rhizomes on a transparent tray filled with deionized water and scanned them with an Epson Perfection V700 Photo scanner (Epson, Japan), and analyzed them for root length, surface area and volume using WinRhizo Pro software (v. 2003b, Regent Instruments Inc., Québec, Canada). We used 1400 dpi for shrub roots with very fine roots (< 1 mm), 600 dpi for herbaceous roots, and 100–300 dpi for rhizomes. When rhizome length and diameter were not accurately calculated by WinRhizo, we measured them manually using ImageJ.JS. Then, all roots and herbaceous rhizomes were dried in 70 °C and woody shrub rhizomes in 102 °C for 4 d and weighed at 0.00001 g precision to obtain root and rhizome dry weight. To obtain a proxy of belowground gas transport potential (i.e., porosity), we calculated root and rhizome tissue density (TD, g cm⁻³) by dividing their dry weight by fresh volume. For shrub rhizomes, the TD represents all the different rhizome tissues (e.g., bark, wood and pith). While this is not a direct measurement of the gas transport ability of a woody rhizome, low tissue density could indicate increased rhizome aeration (e.g., pith cavities) and possibly gas transport (Vroom et al., 2022). Specific root length (SRL, m g⁻¹) was calculated by dividing total root length by its dry weight.

Since peat core depths varied between plots, we standardized root total length (m m⁻²), biomass (g m⁻²) and surface area (SA, m² m⁻²) to 30 cm depth by taking into account core volume and peat bulk density. We also standardized the traits to 10 cm depth for depth-specific analyses. In addition, shrub root and rhizome biomass, length and SA were scaled from the sub-subsample to the increment level and to 30 cm standardized depth. We calculated root and rhizome volumes by summing diameter-class-specific volumes from WinRhizo (Rose, 2017). All traits were calculated per PFT but also summed (biomass, length, and SA) or biomass-weighted (TD, SRL, diameter) to get an across-PFTs trait estimate. We used the rest of the peat core sample (remaining ca. 75 %) to determine peat bulk density and gravimetric water content (GWC). We thawed the sample in a refrigerator over 2–3 d, removed the aboveground vascular plants, took a 5 ± 0.2 g subsample, and took its fresh and dry weight to obtain bulk density and GWC.

CH₄ flux data were obtained from automated chambers (Varner et al., 2026). The details of the chamber measurement system in general are presented in Bäckstrand et al. (2008). In 2012, the chambers were replaced to similar design as in Bubier et al. (2003). In each thaw stage (intact, partly and fully thawed), three transparent automated gas flux chambers (basal area: 0.2 m², height: 15–75 cm) measured CH₄ and CO₂ every three hours with 5–8 min enclosure time and 5 min air flushing before and after closure (Bäckstrand et al. 2008; Holmes et al. 2022). The CH₄ concentrations were measured with a Los Gatos Research Fast Greenhouse Gas Analyzer (Los Gatos Research Inc, Los Gatos, CA, USA) and the timing control and data acquisition was done with Campbell CR10x data logger (Campbell Scientific, Logan, UT, USA), which was located inside a temperature-controlled cabin. The CH₄ flux data were fit using linear regression. The data were quality-checked using regression diagnostics where CH₄ fluxes outside of significance thresholds at 95 % and 90 % confidence levels were filtered out (Snedecor and Cochran, 1989), as well as manual quality flags that were used to filter out data points affected by known instrumental issues.

We aggregated the higher frequency CH₄ flux data to daily scale by taking the daily CH₄ flux median. Median was chosen to estimate the central tendency of CH₄ fluxes due to skewed and non-normal flux data distributions, and to avoid the disproportionate influence of short-lived high CH₄ flux events that may not be related to belowground plant-mediated CH₄ cycling processes that were the main perspective of this study. Then, we subset the daily-aggregated chamber CH₄ flux data to contain the dates of the productive season in 2023. We defined the productive season as the period between the first and last passing of daily median net ecosystem CO₂ exchange (NEE) past zero, as this can be considered as the plant active season relevant for plant-mediated CH₄ fluxes (Körner et al., 2023). For NEE, we utilized the half-hourly gap-filled NEE (NEE_VUT: estimated across 40 friction velocity threshold realizations for each year according to FLUXNET data processing protocols; quality flags = 3 filtered out) from the ICOS Abisko-Stordalen Palsa Bog (SE-Sto) eddy covariance data processed with FLUXNET standards (Lundin et al., 2023; Pastorello et al., 2020). The resulting productive season used for CH₄ flux data subsetting was 19 May 2023 to 30 August 2023 (Fig. B1). As a background proxy for potential plant carbon substrate provision and plant-mediated gas transport (Knox et al., 2021), we also utilized the half-hourly gap-filled gross primary production (GPP; GPP_NT_VUT: nighttime partitioning at the daily scale and GPP_DT_VUT: daytime partitioning with light response curves at the hourly scale, with variable friction velocity thresholds implemented as an ensemble of 40 friction velocity thresholds by the FLUXNET team; see Lundin et al., 2023 and Pastorello et al., 2020 for details) data from Lundin et al. (2023). However, it is important to note that the EC tower footprint covers only intact and partly thawed stages and does not estimate the NEE and GPP for the fully thawed stage (Łakomiec et al., 2021). To explore the possible diurnal-scale plant-mediated CH₄ transport as a supporting analysis, we also aggregated CH₄ fluxes to the hourly scale using means and medians (Fig. B1).

To explore the variation in the trait-CH₄ flux relationships during the productive season, we calculated CH₄ flux period values (early, middle and peak CH₄ flux) by fitting Gaussian generalized additive models (GAM) with identity link function for daily median CH₄ flux data per chamber with function gam from package mgcv (Wood, 2011, 2017). In the models used for determining CH₄ flux period values, CH₄ flux (response variable) was transformed to approximate normality using inverse hyperbolic sine and date was included as an explanatory variable in a thin plate regression spline. Following an approach similar to TIMESAT (Eklundh and Jönsson, 2015; Jönsson and Eklundh, 2002) and Delwiche et al. (2021), CH₄ flux period values were calculated as follows: (1) peak CH₄ flux: the highest predicted CH₄ flux value during the productive season, (2) early CH₄ flux: the point where the predicted daily CH₄ flux reached 10 % of the peak CH₄ flux for the first time, and (3) middle CH₄ flux: the point where the predicted daily CH₄ flux reached 50 % of the peak CH₄ flux for the first time. We also calculated the whole season median to represent the whole productive season CH₄ flux estimate per chamber. Since one chamber at intact permafrost thaw stage had GAM R2 < 0.1, we used a moving 7 d median to estimate the early, middle and peak CH₄ season values. See details of GAM fit in Table C4.

Data of porewater CH₄ and CO₂ concentrations, δ13CH₄ and δ13CO₂ were obtained from Wilson et al. (2026). The porewater samples were collected from the field near the automated flux chambers using stainless steel sippers on 25 and 26 July 2023. The bottom of the sippers is perforated, and a syringe is attached to the top of the sipper via a stopcock. The sipper was inserted in the peat to the desired depth (0–4, 10–14, 20–24, 30–34, 40–44, 50–54, 60–64, 70–74, and 80–84 cm below peat surface) and the syringe was used to carefully draw up 30 mL of porewater. The porewater was then injected into pre-evacuated sealed glass serum vials. Phosphoric acid (1 mL, 10 %) was added to each vial to preserve the samples and ensure that all dissolved inorganic carbon was in the form of dissolved CO₂ in equilibrium with gaseous CO₂. Porewater CO₂ and CH₄ concentrations and isotopes were analyzed simultaneously via headspace analysis, corrected for solubility using Henry's Law, on a Finnigan Mat Delta V Isotope Ratio Mass Spectrometer coupled to a gas chromatograph (Merritt et al., 1995). The analytical uncertainty based on repeated measurements of a standard was < 0.15 ‰ for δ13CO₂ and δ13CH₄ and <5 % for CO₂ and CH₄ concentrations. To obtain a robust estimate of the isotopic fractionation between CO₂ and CH₄, we calculated the isotope fractionation factor (αC) (e.g., Hodgkins et al., 2014; McCalley et al., 2014) with the following formula: 1αC=δ13CO2+1000δ13CH4+1000

To confirm the similarity of the peat core (n= 9) and CH₄ flux plots (n= 9), we used principal component analyses (function PCA from package FactoMineR; Lê et al., 2008), PERMANOVA (function adonis2 from package vegan; Oksanen et al., 2025), and Euclidean distances between plots within thaw stages, using normalized plot herbaceous and shrub plant cover and GA, mean peat temperature and pH (see Fig. B2).

For assessing general differences in daily median CH₄ flux between thaw stages across the productive season, we built linear mixed effects models where CH₄ flux was the response variable and thaw stage (factor) was the explanatory variable, and chamber (n= 3 per thaw stage) was the random effect. To account for the normality of residuals and chamber-dependent temporal autocorrelation, we applied inverse hyperbolic sine transformation to CH₄ fluxes, and modeled temporal autocorrelation with autoregressive autocorrelation structure of order 1 (corAR1). We built three different models, where each had either intact, partly or fully thawed permafrost stage as the reference level and fit the model with restricted maximum likelihood and obtained p-values (α= 0.05) from Wald t-test. Linear mixed effects models were built with function lme from package nlme (Pinheiro and Bates, 2000; Pinheiro et al., 2023).

Due to the small sample size (n= 3 per thaw stage), we were unable to test for statistical differences in belowground traits between thaw stages and sampling depths (traits, porewater CH₄ concentration and isotopic data) and relied on descriptive methods. However, to explore the general direction and strength of the relationships between root and rhizome traits and CH₄ fluxes (early, middle, peak, and season median) across the thaw stages, we used linear regressions with centered and scaled trait and CH₄ flux values. Only models which did not violate assumptions of homoscedasticity, linearity and normality of residuals were included in the analyses. In cases with little trait variation within each thaw stage, pseudoreplication was considered too severe for reliable model estimates, and these model results were not reported. We focused on model R2, slopes, and model p-values. In cases where linear regressions were unreliable or violated the model assumptions, we explored the relationship strength with Kendall τ correlation coefficients. We acknowledge that the model and correlation estimates can be unstable with small sample size and thus use them as descriptive exploratory tools for estimating the changes in the general directions of the trait-CH₄ flux relationships across thaw stages during the productive season. All data processing and statistical analyses were conducted in R v4.5.1 (R Core Team, 2025).

Belowground traits varied between thaw stages, and the biomass-weighted traits followed shifts from shrub- to herbaceous-dominated plant communities (Fig. 2, Table C1). Summed root biomass and surface area (SA) were 22.2× and 17.0× lower, respectively, at the fully thawed than intact stage, but herbaceous root biomass (mean: 238.3 ± 172 g m-2) and SA (mean: 19.9 ± 13 m² m⁻²) were highest at partly thawed stage. Total shrub root length reached a very high mean of 178.62 km m⁻² and a maximum of 199.7 km m⁻² in the intact stage, substantially higher values than for herbaceous roots (mean: 19.79 km m⁻² and max 33.33 km m⁻² in the partly thawed stage) (Fig. B4). While shrub rhizome biomass and SA decreased from intact to partly thawed stage, herbaceous rhizomes showed an opposite trend (e.g., mean SA 0.13 ± 0 to 3.41 ± 0.08 m² m⁻², 26-fold increase). Notably, herbaceous root and rhizome tissue density (TD) were 1.5× and 2.4× lower in fully thawed than in intact permafrost, respectively (mean ± standard deviation root 0.004 ± 0.0008 to 0.003 ± 0.0002 g cm⁻³; rhizome 0.2 g cm⁻³ ± 0 to 0.08 ± 0.01 g cm⁻³), and these decreases were reflected in the biomass-weighted values (root: 1.5× lower, rhizome: 7.4× lower). While specific root length (SRL) was highest at partly thawed stage (mean shrub 193.6 ± 32 m g⁻¹, 1.4× higher than at intact; mean herbaceous 71.6 ± 35 m g⁻¹, 2.3× and 1.2× higher than at intact and fully thawed, respectively), the biomass-weighted means showed a 2.2-fold decrease from intact to fully thawed stage.

Most of the root biomass was found in the top 10 cm for shrub (coefficient of variation CV and mean for intact: 69 %, 766.0 ± 155 g m⁻²; partly thawed: 67 %, 65.1 ± 28 g m-2) and herbaceous PFTs (CV and mean intact: 56 %, 17.5 ± 9 g m⁻², partly thawed: 45 %, 86.4 ± 89 g m⁻², fully thawed: 73 %, 27.7 ± 20 g m⁻²) (Fig. B3). Shrub roots reached the maximum total length of 167.3 km m⁻² at top 10 cm at the intact stage (Fig. B4). SRL and root TD varied less with depth across thaw stages (max CV SRL: 51 %, TD: 31 %; partly thawed herbaceous). All herbaceous rhizomes were present in the top 10 cm (mean biomass intact: 11.5 ± 0 g m⁻², partly thawed: 101.9±109 g m⁻², fully thawed: 107.8 ± 12 g m⁻², Fig. B5). Most of the shrub rhizomes were located in the top 10 cm (CV and mean biomass intact: 78.5 %, 392.4 ± 227 g m⁻²; partly thawed: 66.7 %, 93.9 ± 15 g m⁻²), but they were also found in 10–20 cm depth (mean intact: 109.4 ± 154 g m⁻², partly thawed: 28.9±0 g m⁻²).

The relationships between root traits and CH₄ fluxes (CH₄ flux difference between intact, partly and fully thawed p<0.01, fully and partly thawed p>0.05, Sect. A2) differed between PFTs but with little temporal variation. Increasing root TD was associated with decreasing CH₄ flux for herbaceous (min standardized regression beta coefficient, β at peak CH₄, max β at middle CH₄) and shrub PFTs (min β at middle CH₄, max β at CH₄ season median) (Fig. 3). Shrub root diameter had negative (min β at early CH₄, max β at CH₄ season median) and SRL positive (min β at peak CH₄, max β at CH₄ season median) associations with CH₄ fluxes. In general, the direction and strength of the relationships between root traits and CH₄ fluxes did not vary strongly across CH₄ flux season values (Figs. 3, B6).

PFTs combined, all root traits except diameter showed negative trends, where the strongest relationships were for root TD (min β at early and peak CH₄, max β at middle CH₄) (Fig. 3). In addition, summed root biomass and SA showed relatively strong negative relationships, with biomass showing slightly stronger negative relationships than SA (biomass: min Kendall's τ at early CH₄, max τ at middle CH₄; SA: min τ at early CH₄, max τ at peak CH₄) (Fig. B6).

Herbaceous and shrub rhizome traits showed contrasting relationships with CH₄ fluxes (Figs. 4, B7). Herbaceous rhizome biomass and SA showed positive relationships with CH₄ fluxes, with stronger trends for SA (SA: min β at peak CH₄, max β at CH₄ season median). Increases in herbaceous rhizome diameter were also associated with slightly higher CH₄ fluxes (min τ at early to peak CH₄, max τ at CH₄ season median). In contrast, shrubs had negative trends between rhizome biomass and CH₄ fluxes (min τ at early, middle and season median CH₄, max τ at peak CH₄; Fig. B7). While summed rhizome SA did not show clear trends (Fig. 4), higher summed rhizome biomass was associated with lower CH₄ fluxes (nonsignificant β at early CH₄, max β at peak CH₄; Fig. B7), while the relationship between rhizome diameter and CH₄ flux was positive (min τ at early and middle CH₄, max τ at peak and season median CH₄) (Fig. 4).

In comparison with belowground trait relationships, CH₄ fluxes were weakly and positively correlated with plant green area (GA) and peat temperature (Fig. B8). GA, which also did not show clear trends with permafrost thaw (Fig. B9), showed no clear relationships with CH₄ fluxes (max τ for shrub at peak CH₄; Fig. B8). Peat temperature did not show clear relationships with CH₄ fluxes (max τ at middle CH₄; Fig. B8). However, increases in peat moisture were generally associated with higher CH₄ fluxes (τ= 0.73 at early, middle and season median CH₄; Fig. B8), with the highest CH₄ fluxes occurring in the wettest plots in the fully thawed stage (median VWC = 100 %), and lowest CH₄ fluxes in the drier intact stage (median VWC = 43.4 %).

The vegetation composition in Stordalen follows the changing hydrology and soil nutrient availability. The vegetation shifts from shrub-dominated (intact stage; e.g., R. chamaemorus, B. nana, and E. nigrum) palsas to increasingly more graminoid- and Sphagnum-dominated semi-wet bogs (partly thawed; e.g., V. oxycoccos, C. rotundata, and S. capillifolium) and wet fens (fully thawed; E. angustifolium, E. fluviatile, and S. riparium; Tables C1 and C2) (Hough et al., 2022; Johansson et al., 2006; Malhotra and Roulet, 2015; Varner et al., 2022). Following these vegetation transitions, plant traits reflect plant adaptation to the changing environmental conditions (Dong et al., 2020; Moor et al., 2017; Pan et al., 2019). While studying plant trait variation and its influence on ecosystem functioning often requires species-specificity to capture changes in diversity (Kichenin et al., 2013; Violle et al., 2014) and separation into fine and coarse root fractions (Freschet et al., 2021b; McCormack et al., 2015), CH₄ flux models often utilize PFTs which have been found to adequately represent the collective trait effects on CH₄ fluxes in peatlands (Gray et al., 2013; Laine et al., 2022). Thus, we here focused on root and rhizome traits at the PFT (herbaceous and shrub) level. This approach provides empirical constraints on plant-mediated carbon substrate provision and CH₄ transport that can inform parameterization and sensitivity analyses in process-based wetland CH₄ models, such as LPJ-GUESS (Smith et al., 2001, 2014), and HIMMELI (Raivonen et al., 2017).

Following vegetation community shifts, most belowground shrub biomass and total length were found in the intact stage while herbaceous biomass increased from intact to fully thawed stage. The shrub root biomass values were lower than previous estimates for R. chamaemorus, E. nigrum and A. polifolia (total > 2000 g m⁻²) in Stordalen (Wallén, 1986), but are much higher than recent belowground biomass estimates (∼ 280 g m⁻²; Hough et al., 2022). In addition, Hough et al. (2022) reported higher total root biomass at the fully thawed (∼ 218 g m⁻²) than partly thawed stage (∼ 146 g m⁻²), whereas this study showed the opposite (Fig. 2). These differences could arise from different root biomass standardization and sampling methods, a problem highlighted in root ecology (Freschet et al., 2021a): Wallén (1986) calculated fine root biomass based on ¹⁴C activity which can lead to overestimations (Wallén, 1986), while Hough et al. (2022) sampled a larger area (625 cm²) than in our study (∼ 79 cm²), leading to differences particularly at the fully thawed stage, where graminoids root more vertically than laterally. In addition, our root samples may have consisted of a different mix of species than in Hough et al. (2022), which together with the low sample size, may have influenced these results. Nonetheless, root biomass decreased by depth, as is common particularly for shrubs (Iversen et al., 2015; Wallén, 1986). As we observed R. chamaemorus roots up to 30 cm depth, similar to Wallén (1986) and Keuper et al. (2017) in Stordalen and by Metsävainio (1931) in Finnish peatlands, and as E. angustifolium and E. fluviatile roots have been found at 40 cm depth (Iversen et al., 2015; Metsävainio, 1931; Shaver and Billings, 1975), the maximum rooting may occur deeper than our sampling depth (ca. 30 cm). In general, these findings, and particularly the remarkably high total shrub root length values, showed that permafrost-affected peat soils are rich in plant roots and likely contribute to the carbon cycling of these ecosystems (Bardgett et al., 2014; Iversen et al., 2015).

Variation in tissue density (TD) and specific root length (SRL) (Fig. 2) with thaw suggest changes in plant gas transport and resource acquisition. Increasing soil moisture with thaw (via higher water table level; Table 2 and Fig. B8) and subsequent aerenchyma formation (e.g., Jackson and Armstrong, 1999) may have decreased TD. The increase in herbaceous rhizome SA may reflect greater aerenchyma proportion, confirming rhizome SA as a proxy for plant gas transport capacity. SRL was highest and root diameter lowest at the partly thawed stage, which could reflect increased resource acquisition particularly for shrubs (Bergmann et al., 2020), but further research with larger sample sizes is needed to investigate the potential changes in shrub resource acquisition . As the peat C / N ratio is highest in the partly thawed stage (Hodgkins et al., 2014), the high SRL may reflect low N availability, shifts in species and species-specific resource acquisition (Hewitt et al., 2019; Keuper et al., 2017). However, as fine root SRL is more meaningful for resource acquisition (Bergmann et al., 2020; McCormack et al., 2015), our results based on the whole root system provide only a rough estimate. Nonetheless, 99% of total shrub root length was < 2 mm in diameter (vs. 31 %, 75 %, and 63 % in intact to fully thawed stages, respectively, for herbaceous), and thus within the traditional fine root definition (McCormack et al., 2015).

The increasing peat moisture and associated changes in belowground plant traits contributed to the increasing CH₄ fluxes along the permafrost thaw gradient. Given our exploratory results (see Sect. 3.2 and Fig. B8) and earlier studies conducted at Stordalen (Holmes et al., 2022; Malhotra and Roulet, 2015), the increasing peat moisture was likely one of the primary drivers of the increasing CH₄ fluxes along the thaw gradient. However, due to the low sample size we were unable to separate the effects of plant traits and peat moisture in multivariate models, and larger sample sizes are recommended to analyze the interactions between peat moisture and plant belowground traits in future studies. Nonetheless, the univariate relationships between plant belowground traits and CH₄ fluxes indicated (1) maintained carbon substrate provision and (2) increasing plant-mediated CH₄ transport across the permafrost thaw gradient. In Stordalen, permafrost thaw has led to increasing plant-derived labile carbon availability and acetoclastic methanogenesis (Hodgkins et al., 2014; Holmes et al., 2022; McCalley et al., 2014; Ström and Christensen, 2007), but plant-mediated CH₄ transport has so far been mostly assumed based on literature and plant aboveground characteristics and activity, such as number of shoots and gross photosynthetic rates (Öquist and Svensson, 2002).

Shrub roots suggested increasing carbon substrate provision. The higher CH₄ fluxes with increasing shrub SRL (and smaller diameter; Bergmann et al., 2020) may be related to increased root exudation (Guyonnet et al., 2018; Wen et al., 2022). The negative root TD-CH₄ flux relationships support this, as higher root TD may be associated with lower root exudation (Wen et al., 2022). In contrast, herbaceous SRL was weakly associated with CH₄ fluxes, possibly due to decoupling of root exudation and SRL in some Cyperaceae species in nutrient-limited environments (Wen et al., 2022), and the SRL also including coarse roots (Picon-Cochard et al., 2012). However, based on the ¹³C-depleted δ13CH₄ and high αC, hydrogenotrophic methanogenesis may have dominated in the topmost peat, and most of the dissolved, less bioavailable organic carbon possibly originated from Sphagnum in the partly thawed stage (Hodgkins et al., 2014; McCalley et al., 2014; Mondav et al., 2014; Wilson et al., 2022). Nonetheless, the lower αC, more enriched δ13CH₄, and higher CH₄ concentrations below the root sampling depth (30–44 cm) indicated acetoclastic methanogenesis (Fig. B11) (e.g., by Methanosarcina; McCalley et al., 2014), possibly fueled by root exudates (Saarnio et al., 2004; Ström et al., 2003; Ström and Christensen, 2007). Indeed, based on photosynthetic rates, the partly thawed stage has been estimated to have more labile carbon for methanogenesis (Öquist and Svensson, 2002), and E. angustifolium and E. vaginatum have high potential root exudation rates in Stordalen (Ström and Christensen, 2007) and graminoid root exudation rates can exceed that of shrubs in the partly and fully thawed stages (Mollenkopf et al., 2026). Based on these previous studies, it is possible that the SRL based on the whole heterogeneous root system did not represent the potential root exudation accurately for herbaceous plants. Thus, roots (including herbaceous) may have exuded labile carbon, but with shrubs having a lesser effect on the net CH₄ flux.

Herbaceous plant coverage increased with thaw, which was reflected in traits indicating plant-mediated CH₄ transport (Fig. 2, Table C1). Higher herbaceous rhizome SA was positively associated with greater CH₄ fluxes, which, together with the positive diameter-CH₄ flux trends, could indicate enhanced CH₄ transport via larger and more porous rhizomes (Määttä and Malhotra, 2024). Rhizome-mediated CH₄ transport could be particularly relevant for plants with pressurized gas transport, such as Equisetum spp., where pressure gradients between the atmosphere, leaves, and rhizomes can drive CH₄ emission via old leaves and broken stems (Vroom et al., 2022). As shrub rhizomes were not associated with CH₄ fluxes (Figs. 4, B8), they may not drive plant–mediated CH₄ cycling in Stordalen, though standardized measurements of, e.g., rhizome length and biomass are needed to confirm this (e.g., Freschet et al., 2021a).

Plant-mediated CH₄ transport may have been pronounced in the fully thawed stage with high peat moisture, herbaceous plant coverage, and low porewater CH₄ concentrations (Table C1 and Figs. B8, B11), indicating plant CH₄ uptake. As low root and rhizome TD indicates higher porosity (Ye and Ryser, 2022), the negative relationships between root TD and CH₄ flux suggest increased plant-mediated CH₄ transport in response to increasing peat moisture. The isotopic evidence (heavier δ13CH₄ and lower αC) may further suggest plant-mediated transport of acetoclastically-produced CH₄ at the fully thawed stage (Chanton, 2005). This supports previous findings of higher herbaceous shoot numbers increasing CH₄ flux in the fully thawed stage (Öquist and Svensson, 2002). However, a recent study conducted in the same productive season found that 30 % of the CH₄ flux in the fully thawed stage consisted of graminoid plant-mediated CH₄ transport and the rest of CH₄ production stimulated by graminoid root exudation, while in the partly thawed stage plant-mediated CH₄ transport dominated the CH₄ flux (80 %) (Mollenkopf et al., 2026). Nonetheless, both this and the study by Mollenkopf et al. (2026) together suggest increased herbaceous plant-mediated CH₄ transport and carbon substrate provision in the fully thawed stage. In Stordalen, most of the plant-mediated CH₄ transport likely occurs via diffusion, typical for Carex spp., E. angustifolium, and E. vaginatum (e.g., Ge et al., 2025; Noyce et al., 2014; Bhullar et al., 2013). CH₄ diffusion may be more efficient through E. angustifolium with low root branching in the fully thawed than at partly thawed stage with more densely rooted Carex species (Schimel, 1995; Shaver and Billings, 1975; Table C1). However, pressurized CH₄ transport via Equisetum fluviatile (in fully thawed stage) could be negligible and requires further investigation as we did not separate the belowground material into species (Hyvönen et al., 1998; Kankaala and Bergström, 2004). As graminoid (e.g., E. angustifolium) roots have low bioavailability in the fully thawed stage (Hough et al., 2022), and root porosity and decomposability may be decoupled (Pan et al., 2019; Ye and Ryser, 2022), the low root and rhizome TD were unlikely to lead to increased decomposition and carbon substrate provision.

Simultaneously to plant-mediated CH₄ transport, herbaceous roots may have enhanced rhizospheric CH₄ oxidation. Previous studies at Stordalen have reported shifts in methanotroph communities with thaw as well as an increase in CH₄ consumption, which may result from greater plant-mediated rhizospheric oxidation (i.e., more O₂ transport with lower root TD) and acetoclastic methanogenesis (Perryman et al., 2020; Singleton et al., 2018). In addition, Sphagnum spp. can host methanotrophic CH₄-consuming bacteria (e.g., Larmola et al., 2010; Putkinen et al., 2014), which may have increased CH₄ consumption in the more O₂-rich intact and partly thawed stages dominated by S. balticum and S. capillifolium, with intact stage chambers possibly having higher CH₄ consumption rates than peat cores with no S. balticum (Table 1). However, the average δ13CH₄ emitted from the partly thawed chamber plots was -79.6 ± 0.9 ‰ in 2011, similar to or ¹³C-depleted relative to the porewater CH₄ at the same sites, indicating little CH₄ oxidation (McCalley et al., 2014). Thus, the heavier δ13CH₄ in the fully and partly thawed stages could also indicate increased CH₄ oxidation close to O₂-rich air at the topmost peat or in the oxidized rhizosphere at ca. 20–44 cm depth (Perryman et al., 2020; Singleton et al., 2018). In addition, during CH₄ oxidation, methanotrophs preferentially consume lighter ¹²CH₄, enriching the remaining CH₄ pool in ¹³C while producing isotopically light CO₂ (Whiticar, 1999). However, isotopically light CO₂ could also indicate less methanogenesis instead of increased methanotrophy, and rhizospheric δ13CH₄ enrichment can also result from faster plant ¹²CH₄ uptake and increased plant-mediated CH₄ transport (Chanton, 2005). Thus, based on the lower αC (i.e., maintained acetoclastic methanogenesis) at 20–44 cm depth, and the strongly negative relationships between root TD and CH₄ fluxes (i.e., decreasing amount of aerenchyma is related with lower CH₄ fluxes and vice versa), it seems that the CH₄-oxidizing effect of herbaceous plants particularly in the fully thawed stage is overshadowed by increased carbon provision for methanogenesis and enhanced plant-mediated CH₄ transport.

The relationships between the static belowground traits and CH₄ fluxes (early, middle, peak, and season median) did not vary strongly within the productive season. This may result from the lack of temporal root and rhizome sampling within the productive season. However, as an exploratory analysis, our results may indicate higher herbaceous plant-mediated CH₄ transport at the middle CH₄ flux period (Fig. 3). In contrast to our hypotheses, shrub root traits were more strongly related (i.e., higher linear regression β and R2) with peak and season median CH₄ fluxes whereas herbaceous root (TD) and rhizome (SA) traits were strongly related with CH₄ fluxes throughout the season. This could result from the lack of temporal sampling, but also from low temporal variation in coarse root diameter (diameter is used for determining root TD; see Sect. 2.5) (Picon-Cochard et al., 2012), possibly contributing to the lack of temporal trends between root diameter and CH₄ fluxes. The root TD-CH₄ flux relationships may have reflected higher CH₄-transporting herbaceous root biomass at the middle and peak CH₄ flux period (July–September; Shaver and Billings, 1977) instead of temporal TD and diameter variation. Indeed, diffusive plant-mediated CH₄ transport via Carex spp. could be high during the mid- and peak-growing season, as for example Carex rostrata (present at the partly and fully thawed stages but not captured by our aboveground measurements; see Table C1 and e.g., Hough et al., 2022) has been found to transport most CH₄ at high summer and early autumn (Ge et al., 2025; Noyce et al., 2014). Furthermore, a study conducted in Stordalen during the same productive season found that plant-mediated CH₄ transport and carbon substrate provision via graminoids was highest in August and September, further confirming these results (Mollenkopf et al., 2026). In addition, gross primary production (GPP), a proxy for plant-mediated gas transport (Knox et al., 2021), peaked in July and August (max monthly medians: 3.87 and 2.79 µmol CO₂ m⁻² s⁻¹, respectively; Fig. B1). Concurrently, CH₄ fluxes at the partly and fully thawed stages were slightly higher during nighttime (Fig. B1). Both of these GPP and CH₄ flux trends could further indicate seasonally (via higher herbaceous plant biomass) and diurnally- (via stomatal opening and closure) varying plant-mediated CH₄ transport . However, as GPP and CH₄ flux peaks did not coincide in the same hourly time periods, further research into diel CH₄ fluxes and plant-mediated CH₄ transport is needed (Fig. B1) (Knox et al., 2021). As Carex, Eriophorum and Equisetum rhizomes are perennial (Husby, 2013; Shaver, 1976), the pre-existing rhizomes may have contributed to CH₄ transport particularly in the beginning of the productive season with active root growth (esp. E. angustifolium; Shaver and Billings, 1975), resulting in strong rhizome-CH₄ flux relationships (Figs. 4, B7).

Our exploratory analyses suggested stronger shrub carbon substrate provision (SRL and diameter) at peak CH₄ flux period and across the season (season median; Fig. 3). While uncertain due to the static trait values, these results could be supported by the greater GPP at the intact and partly thawed stages (Łakomiec et al., 2021) in July and August (Fig. B1), as GPP can also be a proxy for root exudation and correlates with increased wetland CH₄ emissions (Knox et al., 2021). In addition, in a separate study within the same productive season, shrubs and graminoids were found to release more root-derived carbon at the end of the season (Mollenkopf et al., 2026). However, the biomass-weighted SRL trends were negative, which could result from species-specific SRL variation and the heterogeneous herbaceous root system (Gao et al., 2024; Picon-Cochard et al., 2012). The stronger negative trends in the beginning of the productive season may arise from increased plant carbon investment into herbaceous root growth in comparison with shrubs (Billings et al., 1978; Kummerow and Russell, 1980; Wang et al., 2016). However, within the intact and partly thawed stages and in a mixture of herbaceous and shrub plants, the SRL-CH₄ flux trends could in fact be positive (Fig. 3), but inferential statistics with larger sample sizes per thaw stage are required to investigate this further. As temporal variation in SRL and its relationship with root exudation can be negligible throughout growing season (Gao et al., 2024) and studies in (sub)arctic peatlands are lacking (Iversen et al., 2015), the negative trends between biomass-weighted SRL and CH₄ fluxes across the thaw gradient remain uncertain.

Taken together, the temporal variation in root and rhizome traits may have contributed to CH₄ fluxes but belowground trait sampling with larger sample sizes throughout the productive season is needed to investigate these relationships further. Given the uncertain herbaceous SRL and root diameter relationships, future studies could benefit from separating the roots into coarse and fine fractions, measuring the associated root exudation rates, and utilizing minirhizotrons (Iversen et al., 2012; Phillips et al., 2008) to disentangle the temporal variation in root carbon substrate provision and CH₄ transport across the permafrost thaw gradient.