Effects of persistent water-table level drawdown on ne-root biomass and production in different site types of a boreal forested peatland

Purpose In peatlands, ne-root biomass (FRB) and production (FRP) and their depth distribution and plant functional type (PFT) composition are poorly known. We studied the effects of persistent water-table level drawdown on these characteristics in four forested boreal peatland site types that varied in soil nutrient and WTL regimes, ground vegetation and tree stand characteristics. Methods FRB was estimated by separating and visually identifying roots from soil cores extending down to 50-cm depth. FRP was estimated using ingrowth cores covering the same depth and the separated roots were identied using infrared spectroscopy. several root one we combine depth segments from the same core. We will thus present the PFT data at the level of the ingrowth core, without depth information. The calibration models were created using calibration data consisting of known identity root samples. The calibration samples included 22 species common in Finnish peatlands (trees: Pinus sylvestris, Picea abies, Betula pubescens; shrubs: Andromeda polifolia, Betula nana, Calluna vulgaris, Empetrum nigrum, Ledum palustre, Vaccinium myrtillus, Vaccinium oxycoccos/microcarpum, Vaccinium uliginosum, Vaccinium vitis-idaea; graminoids: Carex lasiocarpa, C. rostrata, Deschampsia exuosa, Eriophorum vaginatum, Trichophorum cespitosum; forbs: Dryopteris carthusiana, Epilobium angustifolium, Equisetum uviatile, Menyanthes trifoliata, Rubus chamaemorus). The calibration models were designed to quantify the mass proportions of PFTs i) conifers, ii) shrubs and birch, iii) graminoids and iv) forbs. These PFTs were chosen since their ne roots could be distinguished from each other chemically and quantied in composite samples with RMSE generally < 10% The forb Rubus to be the PFT similar to The


Introduction
Peatlands are a great storage of carbon (C) and peat accumulation is one of the major natural pathways to sequester C from the atmosphere to long-term organic deposits (e.g., Page et al. 2011;Yu 2012). Peatlands are a diverse group of habitats that share just one common feature, an organic soil, peat, or Histosol. Otherwise, they range, in the boreal region, from the extremely low-nutrient conditions of some Sphagnum-dominated bogs to the high-nutrient conditions of eutrophic fens, and from the soaking wet ark fens to peatland forests, where the soil water-table level (WTL) can be down to 30 cm below the soil wetter sites would give way to trees and shrubs that are more abundant in drier sites; and iii) depth distribution of both FRB and FRP would be more super cial in drained sites, as the deep-rooted graminoids and forbs of undrained sites would be replaced by more shallow-rooted trees and shrubs.

Study sites and their vegetation characteristics
This study was conducted in Lakkasuo and its neighboring Hanhisuo peatlands (61°48' N, 24°19' E; Fig.   SI1) in the municipality of Orivesi in southern Finland. Lakkasuo is a partly drained eccentric boreal raised bog complex with a large lagg area and a wide variety of different site types ). The rst drainage operations took place in the western part of Lakkasuo in 1928, and in 1961 the whole eastern part of Lakkasuo was drained, making it possible to study changes caused by drainage on several site types (e.g., Minkkinen et al. 1999). For this study, we chose four pairs of undrained and drained sites (Fig. SI1, Fig. SI2) situated in the close vicinity of long-term monitoring plots (see Laine et al. 2004, Sarkkola and Päivänen 2016). Two site pairs were pine-dominated, nutrient-poor sites, and two were spruce-dominated, nutrient-rich sites. The tree stands in the drained sites had developed from the natural tree stands that had grown at the sites already before drainage.
The site with lowest soil nutrient levels represented a Dwarf-shrub Pine bog (DP; Finnish name isovarpuräme, classi cation by Laine et al. 2012, site types described also in Laine et al. 2004) and its 1961 drained counterpart DPdr. DP sites are ombrotrophic, receiving water and nutrients solely from precipitation, and relatively dry even without drainage. Peat depth in both sites was approximately 2 m ). In the DP site, the tree stand consisted solely of Scots pine (Pinus sylvestris; plant names follow WFO 2021 unless otherwise speci ed) and the ground vegetation was dominated by shrubs such as Ledum palustre, Vaccinium uliginosum and Empetrum nigrum, with some graminoid Eriophorum vaginatum and forb Rubus chamaemorus. Moss cover was continuous and consisted of species with low nutrient demand such as Sphagnum fuscum, S. medium, S. angustifolium and Pleurozium schreberi. In the DPdr site, the tree stand had bene ted from the drier conditions and was greater by basal area and volume in comparison to DP. Differences in ground vegetation in comparison to DP were rather minor, with less Rubus chamaemorus and Sphagnum and more forest mosses such as Pleurozium schreberi and Dicranum sp. Table 1 Tree stand basal area, tree stand stem volume, ground vegetation projection cover, mean soil water-table level during the study period (2016)(2017), and peat properties in the surface peat (0 -20 cm) in the sites DP, Dwarf-shrub Pine bog; TP, Tall-sedge Pine fen; VS, Vaccinium myrtillus Spruce swamp; HS, Herb-rich hardwood-Spruce swamp; su x dr indicates drained site The second pair of sites was Tall-sedge Pine fen (TP; varsinainen sararäme) and its 1961 drained counterpart TPdr. These were minerotrophic oligotrophic sites, receiving ground-water ow from the surroundings. The peat depth in both sites was approximately 2 m ). The undrained TP had a sparse mixed tree stand with pine as the dominant species with some Norway spruce (Picea abies) and downy birch (Betula pubescens) (  Fig. SI2). In the HS site the peat layer was approximately 1 m thick , and the site received constant strong ground-water ow from the nearby spring area in the esker Vatiharju (Fig. SI1). The ground was very uneven with trees growing on drier hummocks with wet depressions in between. The vegetation in the hummocks consisted of species such as Vaccinium myrtillus, V. vitis-idaea and Trientalis europaea, whereas forbs such as Calla palustris and Lysimachia thyrsi ora characterized the depressions. Moss cover was patchy and consisted of forest mosses such as Dicranum sp. and Pleurozium schreberi in the hummocks and Sphagnum centrale, S. girgensohnii, S. squarrosum and several species of the order Bryales in the depressions. The HSdr tree stand had the highest volume of all the eight sites and was dominated by spruce with a notable mixture of birch and alder (mostly Alnus glutinosa, with some Alnus incana (

Fine-root biomass (FRB)
We focused on ne root biomass (FRB), ne roots generally de ned here as roots with diameter ≤ 0.5 mm. All the graminoid, herb and dwarf-shrub ne roots tted into this category except for the graminoid Eriophorum vaginatum, which typically has somewhat thicker roots (e.g., Proctor and He 2019), but was here included in FRB. To cover the varying diameter de nitions applied for ne roots (Addo-Danso et al. 2016 and references therein), we also studied diameter classes 0.5-1 mm and 1-2 mm for tree roots (< lower limit, ≤ higher limit). The rhizomes of the dwarf-shrubs were not included. Our methods did not facilitate studying intact root systems, which is why we were not taking advantage of the recent To estimate the living ne root biomass (FRB) in situ in the eight sites, we took peat cores from soil surface down to 50 cm depth in late August, 2016, one core from each subplot, altogether nine cores per site and 72 cores in total. We used a box-shaped 65 mm x 37 mm peat corer, except in the wet TP site where we used a 60 mm x 60 mm corer because of di culties in successfully recovering the wet brous peat with the smaller corer. The cores were divided into ve 10 cm long segments and stored in freezer for further treatment.
In laboratory, the segments were taken to defrost in refrigerator for approx. 12 hours. We separated the living roots from the peat and dead roots and root sections with tweezers under a magnifying glass. The living roots were divided to the different diameter classes, and to species or species group based on visual identi cation (see speci c section below). The distinction between living and dead roots was done based on the color and elasticity of the root by pulling them with the tweezers. The living roots were dried in 40°C for at least 72 hours and weighed. The values were transformed to per m 2 using the surface area of the cores.
Fine-root production (FRP) To study ne-root production (FRP), we used the ingrowth core method (e.g., Milchunas 2009; 2012). We used ingrowth cores that were 50 cm long and approximately 3 cm in diameter, following Laiho et al. (2014) and Bhuiyan et al. (2017). The cores were made of polyester fabric with mesh size approximately 1 mm x 1 mm. We lled the cores with root-free peat matching the peat of the study sites: non-fertilized horticultural Sphagnum peat for the nutrient-poor DP sites (bogs), typical sedge peat collected from a site representing the same site type for TP sites (fens), and dark more decomposed swamp peat for the VS and HS sites (swamps).
In each subplot we installed two ingrowth cores, altogether 18 ingrowth cores per site and 144 in total.
The cores were installed with a two-piece corer-installer described in Laiho et al. (2014) and expected to reduce the disturbance in the ambient soil as compared to traditional coring. The cores were incubated for two years, as recommended by Bhuiyan  In laboratory, the ingrowth cores were taken to defrost in refrigerator for approx. 12 hours, and then cut into 10 cm segments. The diameter of each segment was measured from two angles from both ends, and the average was used for calculating the surface area. Roots were separated from peat with tweezers under a magnifying glass. All roots found in the cores were < 0.5 mm in diameter. There were few dead roots as short segments attached to living roots, and they were not separated since all roots inside the cores represent production during the incubation period. The roots were dried in 40°C for at least 72 hours and weighed. To obtain the annual FRP (g m −2 year −1 ) for each core, we divided the dry mass by two (years; incubation time) and by the surface area of the cores.

Identi cation of the species and species groups
The identi cation of the plant species of the roots in the FRB samples was done by comparing them with several plant specimens collected from the study sites and stored in alcohol. In most cases there were also some plant remains and living plants with roots still attached on the surface of the uppermost sample (0-10 cm from the peat surface) that helped the identi cation. In some cases, identi cation was not possible to species level but only to species group level. This was the case especially in the smallest diameter group (< 0.5 mm) of conifers, certain shrubs (Vaccinium sp., Ledum palustre and Andromeda polifolia, Carex sp. and ferns. In any case, species level data were also pooled into plant functional types (PFTs) for easier presentation of the results and comparison with the FRP data. The PFTs were: i) conifers, ii) birch, iii) alder, iv) shrubs, v) graminoids and vi) forbs.
For the FRP samples, we used Fourier transform infrared spectroscopy (FTIR) to estimate the relative abundance of the PFTs using calibration models developed by Straková et al. (2020). For that, the separated roots were powdered with an oscillating ball-mill MM400 (Retsch, Germany) and FTIR spectra were measured with a Bruker VERTEX 70 FTIR spectrometer (Bruker Optics, Germany) with a horizontal diamond ATR sampling accessory. In several cases the amount of root material in one segment was so low that we had to combine depth segments from the same core. We will thus present the PFT data at the level of the ingrowth core, without depth information. The calibration models were created using calibration data consisting of known identity root samples. During the incubation period (Nov. 2015-Nov. 2017) the mean air temperature was + 4.3°C and the mean annual precipitation approximately 690 mm. Even though the differences in the annual temperature and precipitation between the two years were small, the growing seasons 2016 and 2017 were far from similar: in 2016 the temperature sum was 1280 d.d. (degree days, >5°C) whereas in 2017 it was only 977 d.d. This difference was also re ected in soil temperatures and soil temperature sums that were lower in 2017 than in 2016 (Fig. SI3). Differences in soil temperature sums between the sites were relatively small. Soil temperature sum in the 5 cm depth was highest in the poorest sites and declined towards the richer sites.
WTL was generally less variable and closer to peat surface in the undrained sites, especially in the TP, VS and HS sites (Fig. 1, Table 1). WTL loggers were inserted to locations that represented the soil surface elevation of the trees. In DP, TP, and VS sites the variation in topography was small and the WTL data is representative for the whole sites. The HS site, instead, had a higher topographical variation from hummocks with trees to pools with grasses, forbs, and sedges. Thus, as the loggers were located in the hummocks, the WTL data at the HS site represents about 15 cm higher ground (and 15 cm deeper WTL) than that of the pools. The poorest DP site was drier than the other undrained sites and had relatively low mean WTL and variation in the WTL, resembling the drained sites. In the drained sites the WTL was more variable than in the undrained sites throughout the measurement period.

Peat properties
Peat properties of the sites (Table 1) were determined from peat cores taken with similar corers as the FRB samples. One core was taken for this purpose from each plot in each site. The cores were divided into 10-cm segments from soil surface to 50 cm depth. Peat pH of each segment was measured with a Radiometer PHM 82 electrode from fresh subsamples mixed with water to 1:3 ratio suspensions. The segments were then dried at 70°C and weighed. Dry matter content of the segments was measured from subsamples dried at 105°C, and the peat mass were corrected accordingly (also taking into account the pH subsample) for calculation of bulk density as dry mass per fresh volume. For nutrient analyses, the segments representing the same depths from the 3 plots per site were combined and homogenized by milling. Carbon (C) and nitrogen (N) concentrations were determined with combustion method (Vario MAX CN). After dry ashing and HCl digestion, phosphorus (P) was determined spectrophotometrically with the molybdenum blue method (Shimadzu UV-2401 PC) and potassium (K), calcium (Ca), magnesium (Mg) and iron (Fe) with atom absorption spectrometer (Varian AA 240 FS).
In peat properties, the differences between the sites were clear (Table 1). Bulk density increased towards the more nutrient-rich sites and was higher in the drained sites compared to their undrained counterparts, except in the VS pair where the mean BDs did not differ signi cantly. Soil pH ranged from 4.0 in the DPdr to 5.5 in the HS. In each pair, the undrained site had higher pH than its drained counterpart. Ash percentage increased from the nutrient-poor sites towards the nutrient-rich sites, and the drained sites had a higher ash % than their undrained counterparts, except for VS. Concentrations of mineral nutrients generally increased from the poor sites towards the rich sites. C/N ratio was highest in the nutrient-poor DP sites and in each pair higher in the undrained site than in the drained site.

Data analyses
To estimate the standard error of the mean FRB and FRP for each site, taking into account the clustered sampling, we used linear mixed models. In these models, the plots were accounted for in the random part of the model in the following way: FRB ij or FRP ij = β 0 + u j + ε ij where β 0 is the intercept or mean of FRB or FRP, u j is a random variable describing the cluster level (=plot) variation that follows the normal distribution ~ N (0, σ 2 plot ) and ε ij is the residual ~ N (0, σ 2 ).
We explored the overall differences in the FRB and FRP between the eight sites with one-way ANOVA. In case of a signi cant site effect, we tested the differences between the undrained and drained sites of each site pair with independent samples t-tests.
For studying the effects of site characteristics and environmental variables on the FRB and FRP, we used linear mixed models. Similar models were tested for total FRB, FRB of tree species and PFTs, total FRP and FRP of PFTs. The tested models were 3-level models, where FRB or FRP from each core was set as a response variable that was nested within a plot, nested within a site. Subplot level was not included in the models since it was not available in the FRB and was available only in part of the FRP plots. FRB ijk or FRP ijk = β 0 + β 1 *env k + …. β n *env k + β 2 *env jk + …. β n *env jk + u k + u j|k + ε ijk where β 0 is the overall intercept of the FRB or FRP; β 1 *env k + …. β n *env k represent the xed effects associated with the environmental variables in the site level k; β 2 *env jk + …. β n *env jk represent the xed effects in the plot level j; the u k is a random effect associated with the intercept for site k that follows the normal distribution ~ N (0, σ 2 intercept:site ); u j|k ~ N (0, σ 2 intercept:plot ) is a random effect associated with a plot j nested within site k; and ε ijk ~ N (0, σ 2 ) represents the residual variance associated with the core level (ingrowth core or peat core used for biomass determination). The model formulation and selection of the best model was done by following the procedure described by West et al. (2014) for three-level models for clustered data, where we used the -2 restricted log-likelihood scores to select the best models with maximum likelihood estimation (ML). After the best model for a selected response variable was found, we used restricted maximum likelihood estimation (REML) to derive the nal parameter values.
All statistical analyses were performed by using the IBM SPSS Statistics version 25.

Fine root biomass (FRB)
Variation in FRB (diameter < 0.5 mm) between individual cores was high, with the lowest FRB in one core found in HS (33 g m −2 ) and the highest in HSdr (785 g m −2 ). The lowest mean FRB of the eight sites was 121 g m −2 in the TPdr site and the highest 342 g m −2 in the HSdr site (Fig. 2). In the pine-dominated, nutrient-poor sites (DP and TP), the drained sites had on average lower total FRB than the undrained sites, while the opposite was true for the spruce-dominated, nutrient-rich sites (VS and HS) (Fig. 2). Pairwise differences were signi cant only for TP and HS (Table SI1a).
The species composition of FRB varied considerably among the sites (Fig. 2). Tree roots dominated in FRB at all but the TP site, where most of the FRB were of sedges (Carex sp.). Also, forbs such as Menyanthes trifoliata were abundant at TP. This dominance of sedges and forbs in TP changed to dominance of arboreal roots in the drained TPdr. In DP, DPdr, HS and VS sites, shrubs formed a notable part of the FRB. In the nutrient-rich drained sites VSdr and HSdr, the biomass of conifer roots was signi cantly higher than in their undrained counterparts.
In general, the FRB of the different tree species re ected the basal areas of the same species in the sites (Fig. 2, Table 1). There were, however, some minor inconsistencies. For example, there was a small amount of spruce in TPdr (basal area 2 m 2 ha −1 ), but no spruce roots were found in the biomass cores. In HS on the other hand, no pines occurred in the stand measurement plots, but some roots were identi ed as pine roots. There can be some occasional small pines in this site type, thus this inconsistency can be coincidental.
Examining the relation of the total FRB with environmental variables gave meager results (Table SI2) Including roots up to diameter 2 mm changed the picture, especially for the TP pair (Fig. 3, Table SI1b). TP had very little other than the smallest diameter class roots in comparison to the drained site where bigger-diameter tree roots were abundant, so that the total FRB of roots ≤ 2 mm were quite similar in TP and TPdr. In the other pairs, the differences between the undrained and drained sites regarding diameter classes were smaller.
FRB declined in all sites from the 0-10 cm peat layer downwards (Fig. 4). In all DP and VS sites, as well as TPdr, FRB was concentrated in the topmost 10 cm layer with almost no roots below 20 cm depth. In TP and both HS sites, FRB declined more gradually from the surface towards deeper layers so that there were still notable amounts of roots in the 30 cm depth.
The FRB depth distribution of species and PFTs varied between sites (Lampela et al. 202x data). In the DP and DPdr sites, shrubs and the graminoid Eriophorum vaginatum were the only species reaching the deepest, 40-50 cm, layer studied. In TP, there were both graminoids and forbs in the deepest layer, but in TPdr no roots at all in the deepest layer. In VS some forb and graminoid roots were found in the deepest layer of one subplot, whereas the roots of pine and forbs reached to the deepest layer in VSdr. In both HS and HSdr there were alder and pine roots in the deepest layer, whereas spruce and birch roots were found in the 30-40 cm depth only in HSdr.

Fine root production (FRP)
Variation in FRP between individual cores was high, one core at VS lacked roots altogether whereas the highest FRP in one core was 300.4 g m −2 year −1 at DP. Mean FRP varied from 30.7 g m −2 year −1 at HSdr to 119.9 g m −2 year −1 at TPdr (Fig. 5). There were no signi cant differences in total FRP between the undrained and drained sites in any of the site pairs (Table SI1c).
Tree and shrub roots dominated in the FRP, except in TP where graminoids had the biggest share (Fig. 5). FRP of forbs was minor at all sites. When comparing the undrained and drained sites, TP had signi cantly less FRP of conifers, shrubs and broadleaf trees, and more FRP of graminoids than TPdr (Fig. 5). In HS, shrubs and broadleaf trees had more FRP than in HSdr, whereas DP and VS showed no differences between the undrained and drained sites. Rubus chamaemorus was grouped together with shrubs in FTIR analyses, thus there appears to be no forbs at DP sites where this was the only forb species present.
Of the tested environmental variables, only two had signi cant correlation with the FRP (Table S2): Pine basal area correlated positively with total FRP, FRP of conifers, and FRP of shrubs and broadleaf trees, and -self-evidently -the coverage of graminoids had a positive correlation with FRP of graminoids and forbs.
FRP declined from the topmost 0-10 cm layer towards deeper layers at all sites except TP where the highest mean FRP was found in the 10-20 cm depth (Fig. 6). The TP sites showed higher FRP in deeper layers than the other sites, and a notable amount of roots even in the 40-50 cm depth. At all VS and HS sites, FRP was concentrated in the topmost 20 cm with sharp decline towards the deeper layers.

Comparisons between FRB and FRP
Generally, variation was high in both FRB and FRP and few signi cant patterns emerged. The most evident difference appears to be that FRP was generally lower in the nutrient-rich VS and HS sites whereas FRB showed no such general pattern (Figs. 2 and 5). Intriguingly, the highest mean FRB was found in the drained nutrient-rich HSdr site and lowest in the drained nutrient-poor TPdr site, while in the FRP the pattern was exactly the opposite. The PFT distribution generally followed similar patterns in both FRB and FRP. The most notable exception was the proportion of graminoids that was generally higher in FRP, except in TP, where it was lower in FRP.
The share of conifers was somewhat higher in FRB in all sites but TP and HS. In HS, the share of shrubs and broadleaf trees was higher in FRB. The depth distribution of FRB was more super cial than that of FRP, except for the HS sites (Fig. 7).

Discussion
As far as we know, this was the rst comprehensive survey of the effects of WTL drawdown on the patterns of ne-root biomass (FRB) and ne-root production (FRP) covering several peatland site types. In both FRB and FRP, the differences between undrained and drained sites were surprisingly small in the forested peatland types that we studied. In FRB (diameter ≤ 0.5 mm), the clearest changes were seen in the two wettest sites, Tall-sedge Pine fen (TP) and Herb-rich hardwood-Spruce swamp (HS), where also the contributions of different plant functional types (PFTs) changed the most. In the nutrient-rich HS, the increase in tree FRB following WTL drawdown was more than enough to compensate for the reduction of FRB in species adapted to wet conditions, while in the nutrient-poor TP, this was not the case. In the two other, drier, site types, trees dominated FRB already in the undrained sites, with only small changes in the amount and PFT composition of FRB.
In FRP, the clearest pattern was the generally higher production in the nutrient-poor, pine-dominated sites. Our results thus suggest that in nutrient-rich peatlands, less FRP may be needed for nutrient acquisition than in nutrient-poor sites. The number of sites is limited, however, so results from a larger sample of sites are needed for stronger conclusions.
Overall, FRB and FRP were not always fully coupled, but could respond in opposite ways to WTL drawdown. This is, as such, not odd, as FRB may better re ect occupation of available space while FRP may better re ect investment into nutrient acquisition. Depth distribution of FRB and FRP also differed to some extent, suggesting that explorative roots may be produced deeper down than where conditions for longer-term root survival are generally found. In all sites with predominantly tree and shrub roots (i.e., all but TP), roots were more super cially distributed than the average for boreal forests (Schenk and Jackson 2002), re ecting the general importance of the WTL in controlling rooting depth in peatland forests.
The study sites varied widely in the measured vegetation parameters and environmental variables, as is typical for boreal forested peatlands (e.g., Laine et al. 2004). They varied from sparse to full tree canopy cover, and ground vegetation was equally versatile ranging from dense shrub or sedge cover to patchy cover of mosses under the densest tree stand. Also, the WTL and nutrient regimes varied. There are thus numerous factors potentially affecting the FRB and FRP in these sites. Accordingly, the variation in all features of FRB and FRP was high. Thus, statistically signi cant patterns were scarce and the con dence intervals of the estimates wide. High spatial variability in root biomass and production is common in different environments and planning a sample size big enough to gain reliable estimates can be challenging (Berhongaray et al. 2013).
The overall level of FRB was comparable to previous results from both drained peatlands and mineral soil forests (Vogt et al. 1995 pine-dominated nutrient-rich site is tenfold higher than our estimates that ranged 30-70 g m −2 . FRP estimates vary depending on the estimation method, ingrowth cores generally yielding lower FRP estimates than, e.g., minirhizotrons (e.g., Hendricks et al. 2006;Milchunas 2012); however, this has not been con rmed for total FRP (trees + ground vegetation) in forests (Finér et al. 2011).
We had only one site representing one site type and drainage status, and thus can draw limited inference on site-type related phenomena. However, the data revealed several interesting patterns, but also indicated some potential problems in our methodology. Responses -or the lack of those -of both FRB and FRP to the WTL drawdown varied among site types and re ected the original (undrained status) WTL and nutrient regimes. In peatlands, these two regimes largely determine the PFT composition that is the Under the wet conditions of TP, both FRB and FRP were dominated by graminoids. Trees and shrubs in this site type are usually found on slightly elevated hummocks that offer better soil aeration. For trees, the conditions are in any case challenging, as shown in low tree basal area. The graminoids in TP were largely sedge species (Lampela et al. 202x data) that tolerate anoxic soils thanks to their aerenchymous root systems (e.g., Visser et al. 2000). These root systems allow oxygen transport to roots as deep as two meters below the soil surface (Saarinen 1996). In TP, the share of conifers was much higher in FRP than in FRB, unlike in any other site, which could indicate that these presumably highly stressed plants had to renew their ne roots faster than the plants better adapted to the site conditions. Another explanation would be that tree roots would have colonized the ingrowth cores faster than sedge roots, but this is unlikely since the opposite has been observed earlier (Finér and Laine, 2000).
At TPdr, all sedge and forb species adapted to wet conditions were completely lacking, as is commonly the case following persistent WTL drawdown (e.g., Laine et al. 1995). Interestingly, this affected overall FRB but not FRP. The lower graminoid FRP was compensated for by higher FRP of trees and shrubs. Also, the FRB of trees and shrubs was higher, but not as much as the FRB of graminoids was lower than in TP.
Yet, when including ne roots up to the diameter of 2 mm, also FRB was similar to that of TP. This investment in longer-living roots in the drained conditions probably partly just re ects the overall larger tree stand, but may also be linked to the need to expand the rooting system to support the increased aboveground biomass in the TP deep-peat conditions. Earlier observations of relatively higher coarse root biomass in peatland versus mineral-soil pine stands may support this assumption (Laiho and Finér 1996).
In the Dwarf-shrub Pine bog (DP) site pair, in contrast, there was no difference between the undrained and drained site in either FRB or FRP. This site type is relatively dry to begin with and has low peat pH and nutrient concentrations. Consequently, there is little chance for a major vegetation change, since the plant species that are competitive in these conditions are limited and they are already present in the undrained site. Generally, the response of vegetation composition to persistent WTL drawdown is the faster and more extensive the wetter and more nutrient-rich the site initially is (Laine et al. 1995). The clearly larger tree stand of the DPdr was not re ected in either FRB or FRP. The biomass of tree roots in diameter class 0.5-1 mm was actually lower under the drained condition; we can think of no explanation for this. Similarly to TPdr, conifers showed smaller share in FRP than FRB, indicating slower turnover than shrub and, especially, graminoid roots. The only graminoid found in this site was Eriophorum vaginatum, which renews its roots annually (Chapin et al. 1979).
One reason for the small or non-existing differences in FRB and FRP in the ombrotrophic DP site may be that the drained site has become signi cantly wetter during the last years due to the gradual blocking of the border ditch by mire vegetation. In 1991, the difference in median WTL was clear: 21 cm in the undrained vs. 34 cm in the drained (Minkkinen et al. 1999). Now the difference had almost disappeared: 22 vs. 23 cm in the undrained vs. drained, respectively (Table 1). Similar development has taken place with the TP and TPdr: in 1991 the WTLs were 10 and 41 cm, whereas now they were 13 vs 20 cm in the undrained and drained sites, respectively.
The spruce-dominated Vaccinium myrtillus Spruce swamp (VS) and HS site pairs showed many similarities but also some differences. Here the differences in vegetation, tree stand, and peat properties were relatively small between the undrained and drained sites. FRB was predominantly arboreal, and mostly of conifers, except for HS that had a considerable share of alder in the tree stand. The overall shallow ne-root distribution is explained by the lack of deep-rooted graminoids and forbs in these sites.
The extremely low FRP observed for the HSdr site, which showed the highest FRB of all the sites, is surprising. Overall, the FRP in the HS sites was clearly lower than we expected. There may be three explanations for this: i) in these nutrient-rich sites vegetation can forage enough of nutrients even with low FRP; ii) FRP is in fact higher but our method did not capture it in these sites, or iii) our method fails altogether to capture FRP in peat soils. Option iii) seems unwarranted since the FRP estimated for the nutrient-poor sites appeared realistic. Option i) is not in line with the few earlier results from nutrient-rich sites (Bhuiyan et al. 2017) but requires data from a larger set of sites for proper evaluation. For forests on mineral soils, different hypotheses have been presented on the relationship between FRP and soil resource availability (e.g., Hendricks et al. 2006). However, there seems to be no overall consensus of any consistent relationship between FRP and soil nutrient status based on eld studies (Hendricks et al. 2006, Yuan andChen Han 2012). In any case, FRP is also constrained by the tree stand basal area, and FRB (e.g., Finér et al. 2011).
Concerning option ii) there are (at least) three ways in which our measurements could have gone wrong. Even though the installation of the cores is in general very easy with our corer-installer (Laiho et al. 2014), in sites with high biomass of coarse roots, especially, the "root mat" is not readily penetrated by the corer and thus, the cores could end up being installed in locations with less tree root biomass and potentially also less FRP. Even though the installation was planned as random, we cannot completely overrule this possibility. This could also add to the ground vegetation contribution being higher in FRP than FRB, apart from differences in turnover rates since ground vegetation generally thrives in the more open areas between trees.
Also, the HS sites had the densest peat and the shallowest peat layer, and the drained site was drier than the other sites. Contact between dry peat and ingrowth cores may remain poor, which would hamper root ingrowth. This is one reason why we recommend always installing the cores in the previous autumn, to ensure the formation of seamless contact between the cores and ambient soil (Laiho et al. 2014). Yet, we do not believe that peat contact was the problem here, since recovery of the cores indicated the opposite -it was "hard enough", while poorly connected cores would be easy to pull out. However, since all cores were prepared to the same length, in these sites a higher proportion of peat-lled core remained on top of the peat surface. That could potentially result in capillary drying of the core peat during the summer, resulting in smaller-than-ambient root growth.
Yet another potential factor is the quality, especially nutrient content of the peat used as substrate in the cores. We did not use peat from the sites but tried to nd standard peat types that would match the peat of the sites. Unfortunately, we did not measure the peat quality extensively before use. In retrospect, we can conclude that the success varied. The clearest failure was the swamp peat used for the nutrient-rich spruce sites (VS and HS pairs). This peat turned out to have much lower phosphorus (P) concentration than the peat at the sites ( Supplementary Fig. SI4.). This may have led to lower-than-real FRP estimates, since P is often the limiting nutrient in forested peatlands, alone or together with potassium (K) (e.g., Moilanen et al. 2010, Hill et al. 2014). K concentration is thus also important, and challenging to manage, since the peat K is mostly soluble and in deep-peated sites shows a sharp decline from the topmost peat downwards (e.g., Laiho et al. 1999). This is impossible to reproduce with standard peat, and challenging also with peat from the site. This is another reason why we recommend always installing the cores in the previous autumn, so that biological processes could start to shape the K pro le.
In the recovered cores, the K pro le had to some extent developed towards the natural shape. Yet, the cores still had lower K concentrations than the ambient peat in the topmost layer in all sites, and higher in the three lowest layers in the DP and TP pairs. This may in principle have reduced FRP in the topmost layer(s) but enhanced FRP in the deeper layers and contributed to the pattern of FRB being more super cial than FRP. Still, all FRP in the deeper layers must have come from live root biomass in those layers since no vertical root growth inside the cores was observed, which speaks against a K-induced artefact there.
The success in matching nutrient concentrations further varied between undrained and drained sites, the concentration patterns being better matched in the drained sites. This implies that different peat should be used for undrained and drained sites in studies including both. Naturally, the best option would be to collect peat for the ingrowth cores (layer-wise) from the very sites at which the cores will be incubated. How much the nutrient contents in the ingrowth cores actually affected root growth and the FRP estimates is di cult to evaluate without cores with various peats installed at the same site. The withinsite variation in peat nutrient concentrations is high , and our data for both the ambient and ingrowth-core peats is limited. Also, the availability of K, especially, that is mostly soluble (Starr and Westman 1978), moves readily with water, including e.g., throughfall (), and is tightly cycled where scarce (e.g., Stone and Kszystyniak 1979), may not be perfectly described by one-time sampling. There are earlier observations that higher nutrient concentrations inside ingrowth cores may, to a varying extent, increase root ingrowth (Raich et al. 1994;Steingrobe et al. 2000), and it thus seems well possible that lower concentrations would reduce ingrowth. The depth distribution of FRB and FRP was mainly related to the species composition and the WTL in the site. In the undrained sites with high WTL, there were species adapted to anoxic conditions, such as sedges and the forb Menyanthes trifoliata, that are known to have deep roots (Sjörs 1991;Kohzu et al. 2003). Of trees, alder, pine and birch had some roots in the deeper layers also in the undrained sites, which was somewhat unexpected, whereas spruce roots were concentrated in the topmost layers. When peatland is drained, the oxic layer in peat extends deeper and creates space for aerobic processes. This did not, however, lead to deeper rooting depth in this study. Also, in more extensive studies on the effect of peatland forest drainage on rooting depth, it has been found that the impact is minor (Paavilainen 1966), about 1 cm difference of rooting depth in 10 cm difference in WTL (Heikurainen 1955). It should be noted here that the differences in the WTL between the undrained and drained sites were overall rather modest. Relatively high WTLs are common in boreal extensively managed drained sites (e.g., Ojanen et al. 2010 Fig. 6). In deeply drained sites, tree roots can extend deeper than in our sites (Holmen 1964).
The interactions of the variables studied are complex, and thus chances for successful modeling of FRB or FRP in a dataset as varied as ours are quite limited. For instance, differences in the soil temperature between our sites were relatively small and generally related to the shading and cooling effect of the tree stand, so that the highest soil temperatures were found in the nutrient-poor DP sites with the smallest tree stands. The temperature-related higher biological activity that could enhance root growth (Alvarez-Uria and Körner 2007; Malhotra et al. 2020) was thus in our data set probably overridden by other tree stand effects. Further, the WTL that is a critical constraint for most features of peatland ecosystem structure and function, including root systems (e.g., Lieffers and Rothwell 1987;Murphy et al. 2009), simultaneously affects so many parameters that extracting its pure impact in a data set as varied as ours is not in practice feasible. That the environmental variables had little effect in our models agrees with ndings for forests on mineral soils (e.g., Finér et al. 2011).
Our results revealed what probably could be expected -that ne-root biomass and production vary widely both within and among different types of boreal forested peatlands. They further highlight some challenges in using peat-lled ingrowth cores. Even though it is common these days to request straightforward conclusions from all individual studies, we do not despair in front of the diversity of the results: we believe that they re ect quite well the inherent diversity and heterogeneity of peatlands, which we also wish to emphasize. It is critical to note that a pattern nicely observed in one peatland site and one method is not necessarily repeated elsewhere, and no rm conclusions should be based on results from a limited number of sites. Even though root studies are laborious and methodologically challenging, the research community should aim to gradually build up a data base for forming a realistic bigger picture of the rooting patterns and the related carbon uxes in peatlands.

Data Availability
The data used in this article have been submitted to PANGAEA and are currently being processed. A link will be added to the manuscript after it has been accepted for publication.      Mean total ne-root production (FRP, g m -2 year -1 ) and its plant functional type (PFT) composition in the sites. Error bars represent SE. A star inside a stack indicates a signi cant difference within that PFT for the undrained and drained sites of a site pair. PFT identi cation was based on FTIR spectral analyses. Rubus chamaemorus was grouped together with shrubs in FTIR analyses. DP, Dwarf-shrub Pine bog; TP, Tall-sedge Pine fen; VS, Vaccinium myrtillus Spruce swamp; HS, Herb-rich hardwood-Spruce swamp. Su x dr indicates drained site.