Disturbance alters relationships between soil carbon pools and aboveground vegetation attributes in an anthropogenic peatland in Patagonia

Abstract Anthropogenic‐based disturbances may alter peatland soil–plant causal associations and their ability to sequester carbon. Likewise, it is unclear how the vegetation attributes are linked with different soil C decomposition‐based pools (i.e., live moss, debris, and poorly‐ to highly‐decomposed peat) under grassing and harvesting conditions. Therefore, we aimed to assess the relationships between aboveground vegetation attributes and belowground C pools in a Northern Patagonian peatland of Sphagnum magellanicum with disturbed and undisturbed areas. We used ordination to depict the main C pool and floristic gradients and structural equation modeling (SEM) to explore the direct and indirect relationships among these variables. In addition, we evaluated whether attributes derived from plant functional types (PFTs) are better suited to predict soil C pools than attributes derived from species gradients. We found that the floristic composition of the peatland can be classified into three categories that follow the C pool gradient. These categories correspond to (1) woody species, such as Baccharis patagonica, (2) water‐logged species like Juncus procerus, and (3) grasslands. We depicted that these classes are reliable indicators of soil C decomposition stages. However, the relationships change between management. We found a clear statistical trend showing a decrease of live moss, debris, and poorly‐decomposed C pools in the disturbed area. We also depicted that plant diversity, plant height, and PFT composition were reliable indicators of C decomposition only under undisturbed conditions, while the species‐based attributes consistently yielded better overall results predicting soil C pools than PFT‐based attributes. Our results imply that managed peatlands of Northern Patagonia with active grassing and harvesting activities, even if small‐scaled, will significantly alter their future C sequestration capacities by decreasing their live and poorly‐decomposed components. Finally, aboveground vegetation attributes cannot be used as proxies of soil C decomposition in disturbed peatlands as they no longer relate to decomposition stages.


| INTRODUC TI ON
Peatlands account for one-third of the world's soil carbon (C) and are essential to regulate global climate (Joosten & Couwenberg, 2008).
They have a high C sink capacity and store ~95% of their C pools belowground (Smith et al., 2004). The amount of C sequestered depends on the balance between gross primary production (from photosynthesis), respiration, and decomposition of plant materials. Therefore, changes in species composition may significantly affect C sequestration as species differ considerably in productivity and decomposability (Cornwell et al., 2008). Likewise, human activities, such as peat harvesting and cattle grazing, are significant factors accounting for changes in peatland floristic composition (Domínguez et al., 2012), affecting the rates of organic matter decomposition, reducing the soil carbon stock (Valdés-Barrera et al., 2019), and releasing CH 4 and CO 2 into the atmosphere (Loise et al., 2020;Miettinen et al., 2017;Walker et al., 2016). Thus, accelerating the greenhouse gas effect and contributing to global warming (Phillips & Beeri, 2008;Schaepman-Strub et al., 2008).
In contrast to the extensively studied peatlands of the Northern Hemisphere, research on Patagonian peatlands is scarce (León et al., 2018). Joosten and Clarke (2002) estimated the surface of peatlands in Chilean Patagonia to be near 10,470 km 2 , corresponding to 1.4% of the national territory (Iturraspe, 2016). On Chiloé Island, Northern Patagonia, vast expanses of forest areas were burned or cleared for timber or to prepare the land for agriculture or livestock due to accelerated human colonization. Nahuelhual et al. (2013) stated that the northern part of the island suffered a loss of 38% of the oldgrowth forest area from 1976 to 2007 (equivalent to 39,013 ha). As a result, poor drainage areas were colonized by Sphagnum mosses (mainly S. magellanicum) or successional shrublands . These peatlands are called "anthropogenic peatlands" (Díaz et al., 2008;León et al., 2018).
Given the recent nature of their formation, anthropogenic peatlands differ substantially from natural peatlands in flora and levels of carbon storage (Díaz et al., 2008). For instance, the average C accumulation rate in anthropogenic peatlands of Northern Patagonia is higher (107.3 g C m −2 yr −1 ) than natural Patagonian peatland (78.3 g C m −2 yr −1 ) (León et al., 2018). Moreover, the C accumulation rate in anthropogenic peatland was significantly higher in young peat deposits near the surface (507.52 g C m −2 yr −1 ) in comparison with natural peatland (335.58 g C m −2 yr −1 ) (León et al., 2018). These differences may be attributed to the unsaturated conditions of young peat deposits near the surface, leading to a greater rate of C accumulation than old peat, and potentially due to the differences in their floristic composition. Anthropogenic peatlands of Northern Patagonia are commonly harvested for the superficial layer of S. magellanicum moss for horticultural purposes and used for cattle grazing (Cabezas et al., 2015;Díaz et al., 2008).
These activities have been shown to increase the species richness, facilitate the arrival of exotic species, and reduce their soil carbon sink capacity (Cabezas et al., 2015;Domínguez et al., 2012;Valdés-Barrera et al., 2019).
Anthropogenic perturbations on peatlands often have a cumulative effect: when combined, the impacts are more significant than those expected from each disturbance operating separately . Nevertheless, it is still unclear how these disturbances further alter the interactions between belowground C pools, including live mosses, debris, and different peat layers, and aboveground vegetation attributes, like vegetation height, biomass, and community composition (Hapsari et al., 2018;Yang et al., 2017).
Likewise, dominant vegetation with different plant functional types (PFTs) or growth forms provides a wide variety of food sources with varying quality for belowground decomposers, affecting carbon turnover (e.g., Chen et al., 2016). For example, peatland mosses, shrubs, and sedges produce a litter with different lignin and polyphenol contents, resulting in varying effects on soil respiration and organic matter decomposition rates (Ward et al., 2009). Likewise, Dorrepaal (2007) found that differences in leaf litter decomposability are large between broad PFT groups (vascular and non-vascular plants) but small and environment-dependent among narrow vascular growth forms (Sphagnum, evergreen-shrubs, deciduous-shrubs, forbs, and graminoids). However, it is still unclear if these PFTs are reliable indicators of decomposition stages in anthropogenic peatlands under grazing and harvesting. Hence, accounting for C stored in peatland and its decomposition stages are essential for conservation and monitoring purposes.
This study aims to assess the effects of anthropogenic pressure on peatland structure and functions and its associated C dynamics.
We hypothesized that sustained grazing and moss harvesting leads to differences in the relationships between aboveground vegetation attributes and belowground C pools. The study site corresponds to an anthropogenic peatland in Northern Patagonia with two different managements: disturbed and undisturbed. We further asked the following questions: 1. Which components of the soil C pool continuum differ significantly between the undisturbed and disturbed peatland? 2. How do the main floristic gradients relate to soil C decomposition?
3. Which relationships between aboveground vegetation attributes and belowground C pools differ significantly between the undisturbed and disturbed peatland?
growth forms, management, plant functional types, PLS path modeling, structural equation modeling

T A X O N O M Y C L A S S I F I C A T I O N
Applied ecology 4. Do PFT-based outperform species-based attributes to predict soil C pools in the study area?

| Study area
The study site is an anthropogenic peatland located at the FLUXNET site Senda Darwin Peatland (CL-SDP; 41°52′S, 73°40′W), north of Chiloé Island, Los Lagos region of Chile (Figure 1a). North patagonia presents many similar isolated peatlands, which, despite their relatively low individual area, sum up to considerable amounts of total C stocks (León et al., 2018).
The peatland originated from a forest fire that occurred >50 years ago (Perez-Quezada et al., 2021). The peatland is about 16 ha, and it is divided into two areas ( Figure 1b): an undisturbed area (~5.5 ha) that has been protected and used for scientific investigations during the last 26 years and a disturbed area (~10.5 ha) that has been used used for artisanal extraction of Sphagnum mosses for commercial purposes (~10 kg of dry moss per month during >50 years) and grazing of four oxen. Both grazing and artisanal extractions are common practices in the anthropogenic peatlands of Chiloé, which cover about 8% of the landscape. Nowadays, the undisturbed area behaves like a small C sink, while the disturbed area is very close to C neutrality but could turn into a C source if degradation continues where the dominant species are Carex distenta, Danthonia chilensis, and exotic species (e.g., Leontodon saxatilis). Generally, classes one and three are located in higher hummocks, while class two (waterlogged species) are located in low hummocks. Sphagnum magellanicum is present in the three classes but more abundant in the first class (complete list of species in Appendix S1 of Cabezas et al., 2015).

| Field data
We used the data collected by Cabezas et al. (2015), including a veg- The underground C pools were extracted using cylindrical samples (52 mm diameter) at the plot's central point using a peat profile sampler (Eijkelkamp, Giesbeek, Nether-lands). We measured until the end of each plot's soil profile, so the general sampling soil depth was variable. We categorized the decomposition stages using the methodology proposed by Von Post (1924). Each peat core was subdivided into three classes (i.e., R1, R2, and R3; Table 1) (Cabezas et al., 2015;Monsalve et al., 2021;Savvas & Passam, 2002). We further kept the live moss and fine and coarse debris as separate pool classes (Cabezas et al., 2015). See Table 1 for reviewing the F I G U R E 1 (a) Location of Senda Darwin peatland; (b) High-resolution RGB composite image of the study area with plot colors scaled with the measured belowground C stocks (sum of C pools) and the Isopam classes. The dashed line shows the border between the disturbed area (upper) and undisturbed (bottom) areas; (c) detailed sampling design belowground pools. Then, we estimated the soil layer densities of the live moss, fine and coarse debris, R1, R2, and R3 C pools from samples of known volume. We then dried all samples at 70°C for 72 h and weighed all C pools separately. Finally, we depicted five random sub-samples of 10 g from each C pool (Cabezas et al., 2015). These sub-samples were ground and later analyzed with an elemental analyzer to characterize live moss, fine and coarse debris, R1, R2 C and biomass stocks independently (NA2500, Carlo Erba, Milan, Italy).

| Modeling approach
The proposed modeling approach can be summarized into four main steps ( Figure 2

| Model variables and ordination transformations
We selected five types of aboveground vegetation attributes linked with belowground C pools: (i) vegetation height, (ii) plant diversity, (iii) aboveground biomass, (iv) the assemblages of vascular species communities, and (v) the assemblages of plant functional types (PFTs). These variables are directly or indirectly related to the decrease of water-logging conditions and peat mineralization in disturbed and undisturbed peatlands, causing the decline of Sphagnum cover and promoting colonization by vascular plants. We considered the PFTs as the bryophytes, graminoids, forbs, and shrubs' growth forms, as this classification showed strong relationships with carbon cycling feedbacks (Dorrepaal, 2007 (Shepard, 1962), which was the case of these three data types.
Finally, we used one-sided ANOVA tests to check for significant differences in the C pool components between the undisturbed and the disturbed area. We used the R-package "vegan" (Oksanen et al., 2016) (scripts available, see Data availability statement). We used multiple ordinary least square regressions to select the species-and PFT-based components included during modeling (e.g., Grace et al., 2016). We trained and compared two models (Figure 2c):

| Modeling variable causal effects via structural equation modeling
1. Models using only data at the undisturbed area 2. Models using only data at the disturbed area We repeated this analysis for the selected NMDS-based C pool components (Figure 2a). We used Moran's I index to check for spatial autocorrelation on the residuals of the LVs (Lopatin et al., 2019;Perez-Quezada et al., 2021). Finally, we used the R-package "plspm" (Sanchez et al., 2017) for the analysis (scripts available, see Data availability statement).

| Differences between species-and PFTbased information
Finally, we tested if the effects of the species-based community composition on the models statistically differ from the effects of the PFT-based gradients. For this test, we compared linear regressions models predicting the C pool components using all PLS-PM-based vegetation components with models first excluding the speciesbased compositions and then the PFT-based gradients. We used the Akaike Information Criterion (AIC) to depict the best models and a one-sided ANOVA test to assess the significant levels. We tested normality in the model's residuals using the Shapiro-Wilk normality test (Royston, 1982).

| Which components of the soil C pool continuum differ significantly between the undisturbed and disturbed peatland?
The distribution of the six belowground C pools in the feature space reflects two main gradients, which result in similar eigenvalues and axis limits (Figure 3a-b). However, the second axis (C Comp. 2) separates classes and perturbation site conditions. The first C axis  Figure 3d). Figure 4 presents the species-based floristic composition of the 49 peatland species, and Table S1 summarizes all derived vegetation attributes. We selected two components after the multiple regression analysis (Table S2 in Appendix S1). The most common species are shown in Figure 4b,c. The first NMDS species-based axis (Spp.

| Which relationships between aboveground vegetation attributes and belowground C pools differ significantly between the undisturbed and disturbed peatland?
The PLS-PM models yielded accurate overall goodness-of-fit, rang-  Figure S2). In contrast, the disturbed area depicted only clear significant relations with the floristic composition. The PLS-PM model for the overall data (disturbed and undisturbed together) is presented in Figure S2 (Appendix S1). Finally, we observed spatial autocorrelation problems in the residuals only in the predictions of the C Comp. 1 of the disturbed area (see Table S3 for the detailed Morans' I and p-values).

| Do PFT-based outperform species-based attributes to predict soil C pools in the study area?
Finally, Table 2 shows the pairwise tests between the effects of the species-and the PFT-based community compositions on the PLS-PM belowground C pools predictions. Species-based ordination systematically explained more belowground C pools variance than PFT-based ordination (i.e., lower AIC values). However, these differences were only significant (α < 0.001) in the undisturbed models for the moderate to completely decomposed C components (R2→R3; C Comp. 1), and in the disturbed model for fresh and poorly-decomposed C pools (Moss→R1; C Comp. 2; alpha = 0.01).

| DISCUSS ION
Our results emphasize the apparent management-based differences in the linkages between soil C pools and vegetation attributes and the importance of plant community composition in their causal associations. The floristic composition is strongly affected by management and the site's microtopography, including water-logged (hollows) and dryer-high areas (Mathijssen et al., 2019). Similarly, our study site showed significant differences in floristic compositions and plant diversity between the undisturbed and disturbed areas.
The undisturbed area consisted of species-poor communities with vascular shrubs dominated by Baccharis patagonica, Sticherus cryptocarpus, and Myrteola nummularia. These are the most common peatland species in the Chiloé island (Díaz et al., 2008) and often indicate an advanced vascular successional state (Castillo-Riffart et al., 2017;. However, these dense shrub covers could restrain the peatlands' ecological succession as shrubs intercept less precipitation than trees, causing flooding conditions that prevent tree species from colonizing the peatland Díaz et al., 2008). These naturalized ecosystems are small C sinks, influences belowground C pools during humification ( Figure 5). In general, the soil C pools were larger in dryer shrub-dominant communities than in Juncus' water-logged communities. As water-logged plant communities often have anaerobic conditions that inhibit C decomposition (Beer et al., 2008), we found that Juncus areas have high live (moss) and intermediate decomposition C components.
However, many Juncus areas, especially in the disturbed peatland, did not show highly-decomposed peat, indicating either a low decomposition rate or due to flooding, which encourages dissolved C transport (Mulholland, 2003 (Dorrepaal et al., 2005;e.g., Ward et al., 2009), the pairwise test of significance showed consistently better results for the models considering the species-based community composition information over the PFT-based information (Table 3). However, we found clear statistical differences between the species-and PFT-based composition to predict R2-R3 stages in undisturbed peatlands and debris allocation in disturbed peatlands (Table 3) We found that vascular plant diversity was negatively associated with peatland's fine and coarse debris and positively with poorlydecomposed C components in the undisturbed area ( Figure 5a). The negative influence of vascular plant diversity in soil C reservoirs has been associated with increased soil oxygenation, microbial activity, and the organic matter decomposition rate. Vascular plant litter is less recalcitrant than Sphagnum litter and decomposes faster and showed higher productivity over peat mosses Scheffer et al., 2001;Thormann et al., 1999). These interactions eventually facilitate the release of ancient carbon as CO 2 into the atmosphere (Walker et al., 2016). This process may be enhanced under current climate change scenarios, as increasing temperatures promote drought and accelerate colonization by vascular plants, increasing the microbial activity (Fenner & Freeman, 2011). However, aboveground biomass was only significant for moderate to highlydecomposed soil C in the overall model ( Figure S2). Hence, a larger biomass gradient could be needed to assess the importance of biomass in C decomposition accurately (Lopatin et al., 2019).
It is worth noticing that the statistical linkages between aboveground vegetation attributes and soil pools may be due to a legacy of past vegetation types. For example, after the burn of the original temperate rainforest, succession, and colonization of Sphagnum species took place . Hence, the current state of the soil C pools could reflect, at least to some degree, this successional process instead of the actual community characteristics. Unfortunately, we did not characterize plant macrofossil remains in the soil cores to assess past vegetation successional stages (e.g., Mathijssen et al., 2019). This information could have given us an insightful view of how much the community compositions have changed over the last decades, which is essential to assess how much of the current plant composition relates to actual C pools. Note: Significant codes: ***0.001; **0.01; *0.05; 0.1 are shown in bold numbers.

TA B L E 2
Pairwise tests of differences between the effects of the species-and the PFT-based community compositions on the PLS-PM belowground C pools predictions for monitoring C degradation using in-situ measurements alone or in combination with remote sensing data (Cabezas et al., 2015;Castillo-Riffart et al., 2017;Harris et al., 2015;Jaenicke et al., 2011; e.g., Lopatin et al., 2019). The latter could provide the means to periodically assess the ecological status of peatlands and its relation to the C cycle at large scales (Ciais et al., 2014;Cole et al., 2013;O'Rourke et al., 2015). Nevertheless, as the human impact on natural ecosystems still increases, it is imperative to find solutions to accurately monitor the C cycle components in semi-natural and degraded ecosystems. the Open Access publication of the manuscript.

CO N FLI C T O F I NTE R E S T
The authors declare that there are no conflicts of interest.

O PE N R E S E A RCH BA D G E S
This article has been awarded Open Materials, Open Data Badges. All materials and data are publicly accessible via the Open Science Framework at https://github.com/JavierLopatin/ Peatland-Disturbance-C-Pools.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data and R scripts used in this investigation are available from the GitHub repository https://github.com/Javie rLopa tin/Peatl and-Distu rbanc e-C-Pools.