High‐resolution peat volume change in a northern peatland: Spatial variability, main drivers, and impact on ecohydrology

The depth of the groundwater table below the surface and its spatiotemporal variability are major controls on all major biogeophysical processes in northern peatlands, including ecohydrology, carbon balance, and greenhouse gas exchange. In these ecosystems, water table fluctuations are buffered by compression and expansion of peat. Controls on peat volume change and its spatial variability, however, remain elusive, hampering accurate assessment of climate change impact on functioning of peatlands. We therefore (1) analysed patterning of seasonal surface elevation change at high spatial resolution (0.5 m); (2) assessed its relationship with vegetation, geohydrology, and position within the peatland; and (3) quantified the consequences for peatland surface topography and ecohydrology. Changes in surface elevation were monitored using digital close‐range photogrammetry along a transect in a northern peatland from after snowmelt up to midgrowing season (May–July). Surface elevation change was substantial and varied spatially from −0.062 to +0.012 m over the measurement period. Spatial patterns of peat volume change were correlated up to 40.8 m. Spatial variation of peat volume change was mainly controlled by changes in water table, and to a lesser extent to vegetation, with peat volume change magnitude increasing from lawn < hollow < flark. Our observations suggest that patchiness and vertical variability of peatland surface topography are a function of the groundwater table. In dry conditions, the variability of surface elevation increases and more localized groundwater flows may develop. Consequently, spatially variable peat volume change may enhance peatland water retention and thereby sustain carbon uptake during drought.

The distance between the groundwater table and peat surface (relative groundwater table; GWT R ) is controlled by multiple ecohydrological feedbacks, which together result in a stabilization of GWT R and surface wetness (Belyea & Baird, 2006;Waddington et al., 2015). One of the feedbacks essential in stabilizing GWT R in northern peatlands is peat volume change (Kennedy & Price, 2005;Nijp et al., 2017;Price, 2003). Peat in northern peatlands generally consists of partially decomposed peatmosses (Sphagnum; Rydin & Jeglum, 2013). The open pore structure and high compressibility of this fibrous material (Price, Cagampan, & Kellner, 2005;Waddington, Kellner, Strack, & Price, 2010) enable expansion of the saturated peat matrix in wet periods and compression during dry spells (referred to as peat volume change). The stabilizing effect of peat volume change on GWT R can be considerable (~0.01 to 0.28 m reduction of temporal groundwater fluctuations; Almendinger, Almendinger, & Glaser, 1986;Baden & Eggelsmann, 1964;Fritz, Campbell, & Schipper, 2008;Roulet, 1991) and effectively reduces peatmoss moisture stress . From classical soil mechanics it follows that the peat volume change potential increases with larger peat compressibility and peat thickness (Almendinger et al., 1986;Schlotzhauer & Price, 1999) and that actual peat volume change is controlled by changes in mechanical effective stress (Terzaghi, 1943).
The overall development of northern peatlands is controlled by ecohydrological interactions operating at various spatiotemporal scales (Ivanov, Thomson, & Ingram, 1981;Waddington et al., 2015). Therefore, alterations in local processes (e.g., plant community) induced by changes in external factors may affect larger scale (e.g., whole peatland) processes and vice versa (Belyea, 2009;Belyea & Baird, 2006). Such cross-scale feedbacks may result in self-organized spatial structures of vegetation, mechanical and hydrophysical properties, and of peat accumulation rates (Belyea & Baird, 2006;Belyea & Clymo, 2001;Eppinga, de Ruiter, Wassen, & Rietkerk, 2009;Rydin, 1986). As direct controls on peat volume change (compressibility, peat thickness, and changes in water table depth) are indirectly affected by feedbacks at various spatiotemporal scales, it seems likely that also peat volume change is spatially structured at multiple spatial scales.
Modifications in the spatial arrangement of spatial structures within northern peatlands caused by changes in external drivers may thus alter the hydrological and biogeochemical functioning (Baird, Belyea, & Morris, 2013), which may nonlinearly feed back to the global climate (Belyea, 2009;Belyea & Baird, 2006). Monitoring modifications of such self-organized patterns may assist detecting and preventing undesired sudden and irreversible regime shifts in ecosystem functioning (Kéfi et al., 2014;Nijp et al., 2019).

The vegetation of northern peatlands regularly forms easily recognizable units in vegetation composition and function (microforms) and
is commonly classified in elevated dry hummocks, wet hollows, and lawns located in between (Andrus, Wagner, & Titus, 1983;Rydin & Jeglum, 2013). As peatland microforms may persist for centuries or even millennia (Hughes, Mauquoy, Barber, & Langdon, 2000;Karofeld, 1998;Nungesser, 2003;van der Linden, Barke, Vickery, Charman, & van Geel, 2008), the composition of plant litter added to the peat surface will determine the hydrophysical characteristics of the peat matrix later formed (Belyea & Clymo, 2001). Previous studies suggest that peat volume change and compressibility are related to microform and its associated characteristics such as bulk density, and degree of decomposition Waddington et al., 2010). Positive feedbacks between compressibility and plant species composition seem to occur, so that spatial variation of peat volume change could be related to present microform distribution (Waddington et al., 2010;Whittington et al., 2007).
Spatially structured peat volume change patterns may furthermore arise from processes that affect the peat thickness and fluctuations in water table. The local peat thickness is related to the alignment of the peatland in the landscape and peat development over larger spatiotemporal scales (Belyea & Baird, 2006). Besides rainfall and evapotranspiration, lateral flow constitutes a major part of the peatland water balance and therefore impacts water table fluctuations and peat volume change (Kellner & Halldin, 2002;Peichl et al., 2013). Lateral water transport is controlled by the hydraulic gradient, hydraulic conductivity, and aquifer thickness (Hillel, 2004;Ivanov et al., 1981). In turn, these properties are a function of larger-scale positional variables describing the regional flow and position of the peatland in the landscape (Grootjans, van Wirdum, Kemmers, & van Diggelen, 1996;Ivanov et al., 1981;Kemmers, 1986).
In summary, many processes operating at various spatiotemporal scales may affect the magnitude and spatial organization of peat volume change. Yet, little is known about the spatial structure of peat volume change and its spatial scale of patterning. Moreover, studies establishing links between peat volume change in northern peatlands and potential drivers are rare (Waddington et al., 2010).
This study aims to (1) explore the fine-scale (0.5 m resolution) spatial structure of peat volume change patterns; (2) determine to what extent peat volume change is affected by (a) vegetation composition,  (Belyea & Baird, 2006;Waddington et al., 2010), we hypothesize that vegetation composition is a good predictor of peat volume change magnitude and its spatial variability.
Using photogrammetric processing of digital images taken along a transect in a natural peatland at multiple times throughout the growing season, we obtained highly accurate, high-resolution (<0.5 m) digital terrain models (DTMs). These DTMs allowed geostatistical quantification of the spatial structure of peat surface elevation through time, and relating it to vegetation composition, positional, and geohydrological site factors.

| Site description and climate
This research was performed in Degerö Stormyr, a natural, minerogenic, oligotrophic mixed peatland complex located in northern Sweden (64°N°19E°; Figure 1). The peat is underlain by undulating acidic gneissic bedrock and glacial till material (Malmström, 1923), and ranges in thickness from 3 to 4 m, locally reaching up to 8 m.
2.2 | Data collection 2.2.1 | Spatially explicit peat volume change at high resolution Within the Degerö Stormyr peatland, we used a 571-m-long and~6m-wide transect from the south-western peatland margin to the north-eastern margin along an existing boardwalk (Figure 1). Peatland vegetation and surface elevation are highly sensitive to (compaction by) treading, and the boardwalk provided an essential means to prevent disturbance of the peat surface. To estimate temporal peat volume change, we created a time series of DTMs of the boardwalk transect at five points in time (see Appendix 2 in the supporting information for dates and more details), in the period right after the snowmelt until the middle of the growing season (7 May to 12 July 2013).
We assume that the growing season started with a saturated peat matrix, supported by the shallow groundwater tables (<0.05 m below peat surface) in May as observed in a 12-year record of groundwater (Peichl et al., 2014) and of diluted TOC concentrations measured in peatland peak discharge at the onset of the growing season (Leach,FIGURE 1 Overview of the Degerö Stormyr study site and transect of interest. The definitions of distance to nearest peat-forest border (DB) and south-western border (DB SW) from point × are illustrated with arrows. The red rectangle shows the extent of Figure 3. Aerial imagery (0.5-m resolution) was provided by © Lantmäteriet Larsson, Wallin, Nilsson, & Laudon, 2016). As such, winter conditions reset potential water deficits accumulated in previous year's fall.
DTMs were constructed from high-resolution top-view photographs taken~6 m above the peat surface with digital close-range photogrammetry (see section 2.3). A Canon EOS 450D camera (4,272 · 2,848 pixels) with a Canon EFS 18-55 mm F/3.5-5.6 lens was attached on a 7 m-long rod. During the installation of ground control points, we also verified the presence and thickness of ice. Our survey showed that only in some isolated, well-developed hummocks along the transect some small ice lenses (<0.05 m) occurred in the first half of May. Given the very low areal contribution of hummocks along the transect and the thinness of the ice lenses observed, we assumed that the effect of ice lenses on the peat surface elevation changes observed in our study is minimal and not comparable to that of the palsa peatlands in the more northern discontinuous permafrost zone (Petrone et al., 2008). The filter of the piezometers was constructed by covering a perforated section (30% perforated area; hole diameter of 0.008 m) with a filter cloth to prevent clogging with peat debris. Groundwater table with respect to the peatmoss surface (GWT R ; m) was measured at the five dates coinciding with image collection. A quick computation based on Darcy's law (q = −Ks dH/dz) suggests that the use of piezometers as groundwater wells is acceptable in used filter setting. With a minimum saturated hydraulic conductivity (K S ) of 5 m/day (see Appendix 1 in the supporting information), a maximum distance between groundwater table and bottom filter of 0.5 m (dz), and a flux of 0.01 m/day (q), the expected hydraulic head difference (dH) between groundwater table and bottom of filter is 0.001 m. This is equal to the precision of the groundwater pressure sensors and is thus considered negligible.

| Explanatory variables
We estimated the saturated horizontal hydraulic conductivity (K S ) of the 0.20-to 0.40-m-deep peat layer using in situ slug tests as described in van Dijk, Nijp, Metselaar, Lamers, and Smolders (2017). In slug tests, a volume of water (slug) is quickly poured into the piezometers, which generates a hydraulic head difference and induces flow from the piezometer into the peat matrix. K S was estimated using the rate of change in the water level in the piezometer over time and piezometer flow system characteristics (Zlotnik, Goss, & Duffield, 2010). The water level in the piezometers was measured using groundwater-level loggers (Diver type DI220, Van Essen Instruments, Delft, The Netherlands). As water chemistry affects saturated hydraulic conductivity estimates (Kettridge & Binley, 2010), the slug-water (100 mL) was collected from a pool in the peatland to match local water quality. Repeated slug tests (n = 3) within 1 day on the same filter (n = 7) indicate that the mean measurement error of K S is 5.4% of the mean observed value. This confirms the high precision that can be obtained with slug tests, in agreement with van Dijk et al. (2017).
Peat thickness (PT; m), here defined as the distance between the peatmoss surface and the peat to mineral interface at saturation (i.e., start of the growing season), was measured by pushing an extendable metal rod into the peat. The water table relative to a fixed datum, referred to as the absolute groundwater table (GWT A ), was measured in the first week of May at all piezometer locations. Changes in absolute groundwater table (ΔGWT A ) were calculated from changes in GWT R and surface elevation inferred from DTMs.

Ecohydrological landscape position factors
The ecohydrological landscape position within a peatland may be an overall descriptor of peat thickness and water flow (Ingram, 1982;Ivanov et al., 1981) and therefore control the magnitude of peat volume change. We quantified this position with the Euclidian distance to the closest peat-forest border, to the southwest peat-forest border, and with the initial peat surface elevation above sea level, which were all derived from the ground survey data.

Vegetation
The abundance of vascular plant and bryophyte species was observed in 28 vegetation plots (0.5 · 0.5 m) centred around each of the 28 piezometers along the transect in the end of May 2013. The cover of dead plant material, bare peat, and liverwort cover (classified in living and dead) was recorded (see Appendix 4 in the supporting information for data on plant composition).

| Time series of peat volume change, meteorology, and groundwater table
To verify seasonal peat volume change patterns detected with photogrammetry, we monitored the temporal variation of peat surface elevation at 30-min intervals at the centre of the transect ( Figure 1) with a magnetostrictive linear position sensor during the whole period (MTS Linear Position Sensor Type CM250AVH2, MTS Sensor Technologie, Lüdenscheid, Germany). The position sensor was mounted on a frame between two iron tubes that were installed into the underlying mineral soil, providing a fixed height reference to which peat surface elevation changes could be referenced (see Nijp et al., 2017, for technical details).
Rainfall was measured with a tipping bucket rain gauge (ARG 100, Campbell, Scientific, Logan, Utah, USA, 0.0002-m resolution) near the peat volume change measurements ( Figure 1). The tipping bucket records were corrected for a 10% systematic underestimation of rain amount following Eriksson (1983) in Nilsson et al. (2008). Hourly potential evapotranspiration rates were estimated from air temperature, wind speed, net radiation, and relative humidity following Food and Agriculture Organization standards (Allen et al., 2006;Allen, Raes, Pereira, & Smith, 1998). To determine the drought status from a meteorological perspective, we calculated the cumulative potential rainfall surplus subtracting daily potential evapotranspiration from daily rainfall (rainfall surplus), and adding this value to the rainfall surplus of the previous day.
A half-hourly time series of GWT R was obtained with a float-and counterweight system attached to a potentiometer (Roulet, 1991) in a lawn microform at about 200 m distance from the centre of the transect.

Constructing DTMs
DTMs were created for all measurement dates using a Structure-from-Motion algorithm (Jebara, Azarbayejani, & Pentland, 1999) implemented in Photoscan v1.1.2 Professional Edition (Agisoft LCC, St. Petersburg, Russia). First, a sparse point cloud was constructed by aligning the images with high accuracy, generic point selection, keypoint limit at 40,000, and tie point limit at 1,000. Misaligned images were manually realigned or removed if realignment was not successful. Next, to georeference the point cloud and remove nonlinear distortions in the constructed peat surface, camera calibration parameters and point coordinates were optimized by comparing observed coordinates of the ground control tubes obtained with the ground survey with simulated coordinates. See Appendix 2 in the supporting information for detailed information on the constructed sparse point clouds.
Points not representing the peat surface (e.g., vascular vegetation) were removed using a variant of the triangular irregular network FIGURE 2 Time series of (a) peat surface position at the peatland centre, (b) water table relative to peat surface, (c) cumulative rainfall surplus, and (d) rainfall and potential evapotranspiration (PET) during the measurement period. Peat surface position is measured with a linear position sensor in the centre of the peatland; the y-axis represents the surface position relative to the first observation. Dates of transect measurements are presented with points in 1a. Rainfall, potential evapotranspiration, and cumulative rainfall surplus are daily sums. Potential evapotranspiration is calculated following Food and Agriculture Organization standards  refinement algorithm (Axelsson, 2000) implemented in LAS Tools (version 150406, Rapidlasso GmbH, Gilching, Germany). To further prevent vascular plants from biasing peat surface estimates, the lowest surface point in a regular 0.5 · 0.5-m grid was used as peat surface proxy. All data within 1 m of human influence (e.g. boardwalks) were discarded. A high-resolution gridded dataset of peat volume change was constructed by subtracting the DTM at the wettest point in time (13 May) from the driest (12 July) to obtain the maximum seasonal peat volume change.

DTM validation
The quality of the generated DTMs was assessed in three ways. A first estimate of the quality was obtained by comparing the difference between the projected and measured coordinates at the ground control points (i.e., top of tubes installed in mineral soil) after the photogrammetric processing. The mean error in vertical position at the ground control points was 0.0065 m, averaged over the five points in time. Second, manual peat surface position measurements at the ground control points were compared with photogrammetrically produced DTM values at the closest grid cells. The peat surface position was on average underestimated by 0.0026 m (median vertical error), and the mismatch with observations ranged between −0.024 and +0.003 m (25th-75th percentiles; Appendix 2 in the supporting information). The root-mean-square error between manual and photogrammetric peat elevation estimates was 0.022 m. This difference, however, also includes the manual measurement error of the distance between peat surface and top of ground control tube. Third, to determine how the quality of the DTMs is affected by the proximity to ground control points, we sequentially removed ground control points . This test demonstrated that at the mean separation distance from a ground control point along the transect (35 m), the vertical error was maximally 0.007 m. The distance between ground control points was thus appropriately chosen.

| Spatial analysis of peat volume change patterns
Variograms were calculated to quantify spatial heterogeneity of peat volume change in terms of patchiness, variability, and random versus spatially structured components. Briefly, a variogram assesses how spatial autocorrelation changes as a function of distance (Journel & Huijbregts, 1978;Webster & Oliver, 2007). The distance up to which a variable is spatially autocorrelated is referred to as correlation range and represents the characteristic patch size (Webster & Oliver, 2007).
Consistent with geostatistical terminology we refer to sill (C+C 0 ) as the asymptotic value of the variance, which provides a measure of variability (range of values) encountered along the transect. Due to random noise, measurement error, and variability occurring at distances smaller than the separation distance, a random component may be added to the variogram, referred to as nugget variance (C 0 ). The spatially structured component of the sill is referred to as the partial sill (C; see Appendix 3 in the supporting information for details). To identify whether peat volume change was predominantly spatially structured or randomly distributed, we calculated the relative structural variance, expressed as (C/(C +C 0 ); Cirkel, Witte, van Bodegom, Nijp, & van der Zee, 2014).
As variogram calculations are sensitive to outliers (Rossi, Mulla, Journel, & Franz, 1992;Webster & Oliver, 2007), we removed surface elevation outliers (defined as values smaller than Q 25 -1.5 · interquartile range (IQR) or larger than Q 75 + 1.5 · IQR, where Q 25 and Q 75 are the 25th and 75th quantiles and IQR the interquartile range sensu Hoaglin, Iglewicz, and Tukey (1986)). To meet the requirement of second order stationarity (i.e., mean and variance do not change in space), we removed the systematic trend with a second-order polynomial trend surface. Spherical, exponential, and Gaussian theoretical variogram models were fitted through the empirical variogram based on trendremoved residuals using a weighed nonlinear least squares algorithm with the gstat package (Pebesma, 2004) in R v3.1.0 (R Core Team, 2018).
Linear combinations of up to three variograms models were also tested because empirical variograms indicated the presence of multiple spatial scales (see Appendix 3 in the supporting information for details). The Akaike Information Criterion (Akaike, 1973) was used to select the model describing the spatial structure best with least parameters. To determine how patterns in surface elevation changed with drier conditions, we also constructed variograms of peat surface elevation for all dates.

| Vegetation: Correspondence analysis and clustering
Correspondence analyses were used to determine whether the magni- To group the vegetation relevés to coarser-scale microforms, we used hierarchical agglomerative clustering with Euclidian distance as distance measure and Ward's minimum variance method as clustering algorithm using the pvclust package in R (Suzuki & Shimodaira, 2006). A multiscale nonparametric bootstrapping resampling procedure (10,000 bootstrap replications) was performed to estimate the uncertainty of the hierarchical clustering and determine the number of significantly distinct clusters (Suzuki & Shimodaira, 2006). was representative for the longer term average (Nijp et al., 2015). Peat volume change was clearly spatially structured (Figure 3d and Appendix A2.3 in the supporting information). Indeed, the high relative structural variance (see section 2.3.2) of 74% indicates that most of the high-resolution (0.5 m) spatial variation in peat volume change can be attributed to spatially structured properties rather than to random properties. Two spatial scales of patterning were present, as suggested by the combined exponential-Gaussian variogram model describing the spatial structure of peat volume change best with least parameters (Appendix 3 in the supporting informationx). The two spatial scales of patterning were 3.4 ± 0.8 m (short-range; mean correlation range ± standard error) and 40.8 ± 0.6 m (long-range). Even though the combined variogram model had the best fit, the shortrange pattern was not significant (p = .16) and the long-range (Gaussian) model component explained most (78%) of the spatially structured variation in peat volume change. Spatial patterns, that is, characteristic patch size of peat volume change, were thus best described at a spatial scale of 40.8 m.   (Table 1 and Figure 6). Peat thickness, hydraulic conductivity (see Appendix 1 in the supporting information), and distance to the (closest) peat-forest margin were not related to changes in peat surface elevation. The average hydraulic conductivity along the transect was 22.9 m/day at 21 May and did not significantly decrease when the growing season progressed (Appendix 1 in the supporting information).

| Spatiotemporal changes in surface topography
Peat volume change stabilized the distance between the peat surface and GWT R , buffering on average 26% and maximally 84% of the GWT A decline of 0.09 m between 13 May and 12 July. Spatial variation in decline of GWT A through time was unrelated to the position in peatland, as expressed by the global surface elevation and distance from the peat-forest margin (Table 1).
Locations in the peatland with the largest compression corresponded to sites with shallow relative water tables during the peak of the growing season (July GWT R ; Table 1). The decline of GWT R over the growing season was largest for sites with deep July GWT R (Table 1), illustrating that dry sites became drier. This is consistent with the analyses on the spatial structure of topography, which indicated that patchiness and variability of peat surface elevation increased with deeper absolute water tables ( Figure 5).   The magnitude of peat volume change differed among the four identified microforms, with flarks having 90% larger shrinkage than the lawn microforms ( Figure 7a). Although substantial, these differences were not statistically significant (analysis of variance [ANOVA]; F = 0.71, P = 0.55). The four microforms as established by the cluster analysis were mainly separated along the water table gradient (Fig. 8),

| Vegetation in relation to peat volume change
with hollows and flarks prevailing at sites with shallower GWT R (ANOVA; F = 6.1, p < .001; Fig. 7b).

Plant community composition level
Peat volume change explained only 7.3% (R 2 ) of the total variation in plant species composition, and its effect was not significant (Monte Carlo permutation test; F = 2.0, p = .13). The groundwater table relative to the peat surface (GWT R ) in July explained 30.5% ( F = 11.4; p < .01) of the total variation in plant species composition and was the main explanatory factor for spatial variation in vegetation composition along the transect (Figure 8). However, due to the collinearity of GWT R in July with ΔGWT R , peat volume change, distance from the peat-forest border, and surface elevation, its effect on plant species composition cannot be separated from these factors (Table 1)  The magnitude of peat volume change observed in the studied peatland (up to 0.062 m, from start-middle growing season) is in the range found for Sphagnum-dominated ombrotrophic peatland systems in the boreal and temperate zone, generally ranging between 0.02 and 0.11 m (Almendinger et al., 1986;Nijp et al., 2017;Price, 2003;Schlotzhauer & Price, 1999;Uhden, 1967;Whittington et al., 2007).
The result that spatial peat volume change patterns emerge at a spatial scale of about 40 m indicates that peatland simulation models based on horizontally uniform peat volume change can represent a limited area only. Instead, peatland hydrological models need to accommodate for spatial differences of peat volume change. Berne, Delrieu, Creutin, and Obled (2004) and Schilling (1991) suggest the measurement resolution for autocorrelated data to be 3-4 times smaller than the scale of pattern formation to adequately capture the spatial process. Accordingly, a grid resolution of about 12-15 m would be required in spatially explicit peatland hydrology models to capture peat volume change (Kennedy & Price, 2004;Nijp et al., 2017) for peatland systems comparable to Degerö Stormyr, that is, mixed mires or aapa mires. Given the strong correlation between peat volume change and groundwater dynamics (Table 1), this recommendation is likely also applicable for groundwater flow simulations in peatlands in general.

| Relations between peat volume change and environmental factors
In accordance with studies indicating that water table changes are an important control on temporal surface elevation changes (Fritz et al., 2008;Price, 2003;Roulet, 1991), we found that spatial variation in the change of absolute water table through time was the key factor related to spatial variability of peat volume change ( Figure 6 and Table 1). Spatial variability of changes in absolute groundwater table (ΔGWT A ) may arise from spatial differences in local water balance components, that is, rainfall, evapotranspiration, and lateral groundwater flow. Spatial coverage of rain events is larger than square kilometres in flat areas such as peatlands (Ciach & Krajewski, 2006;Rakovec, Hazenberg, Torfs, Weerts, & Uijlenhoet, 2012) and can thus be assumed uniform along the transect. Evapotranspiration rates may differ among vegetation types (Heijmans, Arp, & Berendse, 2001;Limpens et al., 2014), but ΔGWT A was not found to be related to vegetation composition (Figure 8) nor to microform distribution (ANOVA; F = 0.229, df = 3, p = .87; R 2 = .035). It therefore seems unlikely that spatial variability in ΔGWT A was caused by spatially variable evapotranspiration rates. Spatial variability in ΔGWT A thus mainly resulted from lateral redistribution of groundwater, corroborating peatland model studies stressing the need to account for lateral flow Qiu et al., 2018).
In contrast to the observation that the hydraulic conductivity was lower at peat margins of blanket bogs (Lapen, Price, & Gilbert, 2005) and a raised bog (Baird, Eades, & Surridge, 2008), suggested to maintain wet conditions in central bog areas, our correlation analysis suggested that hydraulic conductivity was unaffected by distance to the peat margin (Table 1). Nevertheless, closer inspection reveals that K S actually is lower at peat borders but also other positions with thinner peat thickness ( Figure A2 and Table 1). As sites with larger peat thickness provide a larger expandable medium, we expected that peat volume change was positively correlated with peat thickness as was also hypothesized by Almendinger et al. (1986). However, our results indicate that peat thickness was uncorrelated with peat volume change (Table 1), which is in accordance with observations by Fritz et al. (2008). Likely, due to the increased bulk density and decomposition with depth (Boelter, 1969;Loisel et al., 2014;Päivänen, 1982), most peat volume change is restricted to the upper, less decomposed strata (~top 0.50 m;Price, 2003;Waddington et al., 2010). The saturated hydraulic conductivity (K S ) was poorly and insignificantly related to peat volume change, implying that no autogenic peat formation process at coarse spatiotemporal scales relates these two factors.
The absence of relations between peat volume change and geohydrological and positional factors could be a consequence of the transect being placed predominantly perpendicular to the dominant flow direction (Malmström, 1923). This positioning may have led to an underestimation of the effect of geohydrological and

| Vegetation composition as predictor of peat volume change?
Previous studies indicate that the compressibility and magnitude of peat volume change may differ among microforms (Roulet, 1991;Waddington et al., 2010;Whittington & Price, 2006). Along the studied transect, flarks and hollows seemed to represent peat volume change 'hotspots' (Figure 7a) at the microform level, although differences in peat volume change were not statistically significant among microforms. In contrast to our hypothesis, plant species abundance was not related (R 2 = .073; p < .05) to peat volume change in the studied peatland. This seems to suggest that the positive feedback between peat volume change and microform distribution in the studied peatland is weak. However, the lack of relationship may be attrib- An indirect positive feedback between vegetation, peat volume change, and water table depth may, however, still arise as follows.
Despite the large lowering in absolute groundwater level at dry microforms, the observed surface elevation change was small as compared with wet microforms (Table 1 and Figure 7b). This suggests that dry microforms have a lower compressibility, as also reported by Waddington et al. (2010). The low compressibility at dry microforms reduces the peat volume change magnitude, leading to deep relative water table upon drying, as supported by observations in the midgrowing season (GWT R July; Table 1). Deeper water tables promote the establishment of species adapted to deep water tables (Luken, 1985;Rydin, 1986), which, due to their low compressibility, may further decrease peat volume change magnitude. These positive feedbacks hint towards a coevolution between vegetation and peat compressibility.
At a species level, Drosera rotundifolia and Trichophorum cespitosum may be indicative for peat volume change "hotspots." Yet, as D. Stormyr that the vegetation remained unchanged. With this assumption, the correlation suggests that plant litter or roots of these plant species increased the rigidity of the peat matrix and thereby reduce its compressibility. The low sample size makes these results speculative, and using these species as indicators for peat volume change "hot spots" requires confirmation with other peatland sites before application.

| Effects of peat volume change on surface topography
The surface topography of the studied peatland was spatially structured, with 73% of the surface patterns originating at a spatial scale of 90-140 m and 20% at about 4 m. In a UK raised bog, microtopographical hummock-hollow patterns emerged at a spatial scale of 1-4 m (Anderson, Bennie, & Wetherelt, 2010), similar to the short-range pattern scale found in our study. Given that the shortrange patch size remained constant over the growing season, it seems likely that the short-range pattern is related to microform identity.
Our study suggests that reduced overall peatland wetness over time reduced the patch size and increase vertical variability of peat surface elevation within the studied peatland ( Figure 5). Spatial variation in peat volume change within the peatland was mainly related to spatially variable changes in absolute water  Figure 9). As a result, topographical differences became more pronounced and more patchy, and also the variance of water table depth relative to the peat surface (GWT R ) will increase ( Figure 9). Such increased habitat variability may increase species richness and thereby enhance peatland carbon sink stability and resilience to environmental extremes, such as droughts (Korrensalo et al., 2017;Loreau et al., 2001;Peterson, Allen, & Holling, 1998;Scherrer & Körner, 2011).
Additionally, the increased topographical variability with increased peatland drought status may affect lateral groundwater flow direction and rate. In unconfined aquifers, the water table is frequently assumed to follow topography (Desbarats, Logan, Hinton, & Sharpe, 2002;Hoeksema et al., 1989;Sonnentag, Chen, Roulet, Ju, & Govind, 2008), which is confirmed for an ombrotrophic temperate bog (Malhotra, Roulet, Wilson, Giroux-Bougard, & Harris, 2016). This would imply that with increased vertical variability and horizontal patchiness of surface elevation, the patchiness of GWT R may also increase, resulting in temporarily steeper and more local hydraulic gradients. This may potentially lead to localized groundwater flows from elevated sites to low sites during drought, enhancing the ecohydrological differentiation between wet and dry sites.
Oosterwoud, Ploeg, Schaaf, and van der Zee (2017) and Quinton and Roulet (1998) demonstrated the importance of microform variation in regulating the connectivity of overland flow, which decreased with deeper water tables. We hypothesize that also more localized groundwater flows induced by spatially variable peat volume change may reduce hydrological connectivity. Such reduced connectivity would lead to more tortuous flowpaths and increase peatland water retention. As such, spatially different peat volume change yields another feedback mechanism to sustain peatland carbon uptake under drought, complementary to reduced hydraulic conductivity at greater depth as a result of compression (Whittington & Price, 2006). Moreover, the retention of local water may impede the inflow of minerogenic water and sustain or enhance ombrotrophy in peatlands with minerogenic water input.
The increased lateral discharge from elevated sites to lower depressions may thus amplify a positive feedback between microform compressibility and species composition and promote the spatial selforganization of vegetation patterns. As locations with deeper July GWT R have a larger decline in GWT R during the growing season (Table 1), our data seem to support this hypothesis. However, this suggested impact of peat volume change on redistribution of ground- water flow needs to be tested in future studies using a denser piezometer network. Moreover, although the assumption that the water table follows peatland topography (Haitjema & Mitchel Bruker, 2005) at a spatial scale from 90 to 140 m seems reasonable (Fraser, Roulet, & Lafleur, 2001;Malhotra et al., 2016), it should be verified under different geohydroclimatic conditions.

| CONCLUSION AND IMPLICATIONS
Using a novel approach to construct spatially continuous highresolution (0.5 m or higher) DTMs, our results clearly show that peat volume change is spatially structured for the studied fen, with a characteristic length scale of 40.8 ± 0.6 m (± SE). In contrast to previous research (Almendinger et al., 1986;Waddington et al., 2010;Whittington & Price, 2006), we found only a weak direct relation between peat volume change and microform or ecohydrological position in the peatland. Instead, spatially variable peat volume change was mainly related to changes in absolute water table, likely originating from lateral groundwater redistribution. Surface topography had a patch size of about 140 m at the start of the growing season but decreased to 90 m in the middle of the growing season, when deeper water tables occurred. In addition, the vertical variability increased with deeper water tables. We hypothesize that the increased topographical heterogeneity with deeper water tables may result in a more local redistribution of groundwater from elevated sites to depressions.
This may further enhance microform patterning and the contrast between wet and dry sites during droughts and boosts peatland water retention. This hypothesis needs to be tested in other peatlands and various hydro-climatological settings. Here, the Structure-from-Motion technique was successfully applied to capture temporal changes in peat surface topography with a median vertical difference of 0.0026 m. The spatial scale of peat volume change identified in this study has direct implications for the spatial representativeness of spatially implicit simulation models that include peat volume change (Kennedy & Price, 2004;Nijp et al., 2017).
Like other self-regulating ecosystems, northern peatlands may respond nonlinearly to external perturbations and rapidly shift to a virtually irreversible, undesired state (Hilbert, Roulet, & Moore, 2000;Rietkerk, Dekker, de Ruiter, & van de Koppel, 2004;Scheffer, Carpenter, Foley, Folke, & Walker, 2001). This would imply that peatlands may potentially shift from a net carbon sink to source, feeding back to the global climate by altering radiative forcing (Bridgham, Pastor, Dewey, Weltzin, & Updegraff, 2008;Frolking & Roulet, 2007;Frolking, Roulet, & Fuglestvedt, 2006). Based on observations, our FIGURE 9 (a-c) Hypothetical conceptual model of peat volume change impact on topography and hydrology during progressive drying. The horizontal axis represents a horizontal transect through a peatland; the vertical axis represents the (exaggerated) position of peat surface elevation (solid black lines) and water table (dashed black lines). Groundwater flow paths are indicated with arrows and peat surface and groundwater positions in previous conditions in grey. Considering compression during the growing season, the peat elevation decreases at all positions through space upon drying (b and c). Due to spatial variability in lateral groundwater flow and peat substrate compressibility along the transect, some places start to compress earlier and more than others, enhancing the vertical and horizontal topographical heterogeneity. As a result of the increased vertical range of topography, hydraulic gradients may hypothetically become steeper and groundwater flow more local, as is illustrated with the flow lines study illustrates that spatial feedbacks between peat volume change, vegetation, and hydrology may enhance the resilience of peatlands to climate change.