The effect of drainage ditches on vegetation diversity and CO2 fluxes in a Molinia caerulea‐dominated peatland

Peatlands are recognized as important carbon stores; despite this, many have been drained for agricultural improvement. Drainage has been shown to lower water tables and alter vegetation composition, modifying primary productivity and decomposition, potentially initiating peat loss. To quantify CO2 fluxes across whole landscapes, it is vital to understand how vegetation composition and CO2 fluxes vary spatially in response to the pattern of drainage features. However, Molinia caerulea‐dominated peatlands are poorly understood despite their widespread extent.


INTRODUCTION
A small imbalance between primary productivity and decay within peatlands has led to the accumulation of large carbon stores (Yu et al., 2010). However, many peatlands are subject to damaging land management practices, principally drainage for agriculture and forestry (Joosten and Clarke, 2002), which alter the balance between primary productivity and decomposition shifting peatlands towards CO 2 release (Gorham, 1991). Due to the recent addition of peatland restoration into the Kyoto Protocol, carbon markets may provide funding for peatland restoration (Bonn et al., 2014); however, there are currently no appropriate default international emission factors for drained Molinia caerulea-dominated peatlands (Alm et al., 1999b). Therefore, quantification of emissions from a wider range of vegetation and management types is required (Evans et al., 2011).
Drainage ditches are frequently the main spatial feature within managed blanket bogs. They have been shown to lower the mean water table (Coulson et al., 1990, Wilson et al., 2010, which in turn reduces species richness (Bellamy et al., 2012), particularly affecting species dependent on high water levels including Sphagnum. M. caerulea thrives where water table depths fluctuate (Jefferies, 1915) and has encroached on many drained upland areas (Bunce and Barr, 1988). Although grasses have been shown to produce more biomass and uptake more carbon annually (Berendse, 1998, Otieno et al., 2009, Ward et al., 2009 than dwarf shrubs (e.g. Erica tetralix and Calluna vulgaris) and Sphagnum, the biomass produced is more labile and readily decomposed Butterfield, 1978, Berendse, 1998). Understanding how vegetation composition and CO 2 fluxes vary spatially in response to these features is therefore vital in upscaling CO 2 fluxes across the whole landscape through models. Using a combination of gas flux chambers and soil collars enables the measurement of both ecosystem and partitioned below-ground fluxes at discrete distances from drainage features. Such monitoring facilitates understanding of the variables driving spatial variation in CO 2 fluxes assisting upscaling (Laine et al., 2006).
Most studies investigating the effect of drainage have compared pristine to drained areas of northern peatlands (Silvola et al., 1996, Alm et al., 1999a, Straková et al., 2011b as they sought to understand the broad effect of drainage in these ecosystems. Methane emissions (Minkkinen et al., 1997) and nitrogen mineralization (Tarvainen et al., 2013) have been shown to be greater near or in a drainage ditch compared with half-way between ditches. Studies on CO 2 exchange have compared microforms within drained peatlands (Komulainen et al., 1999) or microforms along natural hydrological gradients  rather than investigate the explicit role of the drainage features.
It is hypothesized that proximity to drainage ditches will influence water table depths, which will affect vegetation composition and primary productivity and consequently CO 2 fluxes (ecosystem respiration; photosynthesis; and total, heterotrophic and autotrophic below-ground respiration) in a drained M. caerulea-dominated peatland. The following paper tests this hypothesis in two drained, temperate, maritime blanket bogs in the UK.
Within each catchment, six pairs of sites were chosen to encompass the expected variation in altitude, aspect, slope, peat depth and ditch dimensions ( Table I). As the ditches are unevenly spaced, plots were located on transects perpendicular to the ditch at ⅛, ¼ and ½ of the distance between the ditch being monitored and the adjacent ditch. Proportional distances from the ditch were chosen to test whether CO 2 fluxes could be upscaled for the whole peatland rather than discrete bands either side of a drainage ditch despite the known variations between sites. Locations are shown in Figure 1c and d (n = 36).

Net ecosystem exchange measurements
Net ecosystem exchange (NEE) was measured from three pairs of plots located ⅛, ¼ and ½ distance from the ditch at each site (n = 6)a total of 36 plots. A 55 × 55 × 25 cm Perspex gas flux chamber was rested on permanently installed 50 cm tall legs with a plastic skirt weighted down by a heavy chain to form an airtight seal with the soil surface [following Shaver et al. (2007) and Street et al. (2007)]. An EGM-4 infrared gas analyser (PP Systems, Hitchin, UK) measured CO 2 concentration every 10 s for 2 min concurrently with chamber temperature and photosynthetic active radiation (PAR) (Skye Instruments, Llandrindod Wells, UK). CO 2 flux measurements were taken at full light, full dark, and~60%,~40% and~10% light levels using a combination of shade cloths. Chamber air temperature was not directly controlled. Measurements were alternated between brighter and darker light levels to minimize any heating/shading effects. Neither condensation in the chamber, which reduces transparency, nor high humidity, which alters gas diffusion from the leaves, was noted over the sample period due to the abnormally cool weather conditions. The chamber was removed between measurements to restore ambient conditions. The net CO 2 exchange at each light level was calculated from the linear change in CO 2 concentration in the chamber (Pumpanen et al. 2004). The headspace volume was estimated by measuring the height from the ground to the base in a grid of nine points added to the chamber volume. The water table depth below the peat surface and soil temperature at 5 cm were also measured at every plot (n = 36).

Vegetation greenness
Vegetation colour, in particular the ratio of green to red and/or blue, has been shown to vary seasonally and be useful as a proxy for vegetation phenology (Richardson et al., 2007, Migliavacca et al., 2011 and health (Mizunuma et al., 2013). Downward facing true colour photographs of the vegetated NEE plots (n = 36) were collected on ten occasions between 20/06/2012 and 25/10/2012 (n = 282). Due to equipment bulk, images were not collected on the same day as CO 2 flux measurements. Between 22 and 36 plots were photographed on each occasion, except 10/8/2012 when only eight images were collected. Images were taken 116 cm above the ground using a Canon EOS-10D with a 28 mm fixed lens set to autofocus and fully automatic aperture and shutter speed. The mean red (DN Red) and green (DN Green ) colour values for the images collected were determined using MATLAB R2011b 408 N. GATIS et al (Mathworks Inc, Natick, MA, USA). The greenness excess index (GEI) has been shown to be useful as an indicator of spring green up (Richardson et al., 2007, Migliavacca et al., 2011. It was calculated for each image where GEI = (DN Green À DN Red ) / (DN Green + DN Red ). A daily GEI timeseries was modelled for each catchment using a third order Fourier series.
Net ecosystem exchange modelling NEE measurements were collected approximately monthly over the growing season from 16/05/2012 to 19/09/2012 (n = 163 sets). A complete set of measurements (n = 36, a 'sampling round') took between 5 and 14 days dependent on weather conditions; to remove this temporal variability, photosynthesis and ecosystem respiration were modelled [Equations (1) and (2)] using all data collected for each plot (n = 36). R Eco was assumed to be equivalent to NEE under dark conditions, and gross photosynthesis (P G ) was calculated as the difference between average R Eco (two measurements taken in dark conditions) and NEE measured at different light levels.
Equation (1) where P 1 is the maximum rate of photosynthesis (μgC m À2 s À1 ), GEI is the modelled greenness excess index, I is the incident PAR (μmol photons m À2 s À1 ), k 1 is the halfsaturation coefficient (μmol photons m À2 s À1 ), T 5 is the soil temperature at a depth of 5 cm (°C), and a (dimensionless) and b (°C À1 ) are empirically derived coefficients describing an Arrhenius (Arrhenius 1898) response to temperature. Equation (2): ecosystem respiration model R Eco ¼ c: exp À d= T 5 þ e: exp À0:5 where T 5 , T Opt and T Tol are the measured, optimum and maximum tolerable soil temperature at a depth of 5 cm (°C); and c (dimensionless), d (°C À1 ) and e (dimensionless) are empirically derived coefficients describing an Arrhenius (Arrhenius, 1898) response to temperature; and e is an empirically derived coefficient describing a Gaussian temperature response. R Eco and P G600 were then calculated for a soil temperature of 12°C, GEI of 60 and PAR of 600 μmol photons m À2 s À1 using the empirically derived parameters (Supplementary Material Table 1). A PAR of 600 μmol photons m À2 s À1 was selected as it lay within the range of PAR observed, and most plots were light saturated at this PAR, enabling photosynthetic efficiency to be compared between locations.

Soil CO 2 efflux measurements
At each plot, four polyvinyl chloride (PVC) collars (16 cm diameter, 8 cm high) were placed on and sealed to the surface of the peat using non-setting putty (Evo-Stik 'Plumbers Mait'). Collars, installed in March 2012, were located between 0·5 and 2 m downgradient of the NEE plots. Above-ground vegetation was removed by regular clipping from all PVC collars, enabling the measurement of below-ground fluxes only. In addition, circular 20 cm deep trenches (56 cm diameter) were cut around half the collars to sever live roots, allowing the below-ground heterotrophic component to be measured. The collars with only above-ground vegetation removed were used to measure total below-ground respiration. Trenches 20 cm deep were considered sufficient as although cord roots are 15-45 cm long (Jefferies, 1915), most of the root biomass is concentrated nearer the surface (Taylor et al., 2001). For each plot, the two replicates of each treatment were averaged to produce a single value. Autotrophic respiration (including root respiration and microbial respiration of root exudates) was calculated from the difference between average total (n ≤ 2) and average heterotrophic below-ground respiration (n ≤ 2) measured at each location for each sampling round (complete set of measurements from 144 collars).
CO 2 measurements (n = 222) were taken in a semirandomized pattern approximately every 3 weeks from 16/04/2012 to 26/10/2012. Data collected between 16/04/2012 and 25/05/2012 were excluded due to obvious treatment effects. CO 2 flux was measured over 2 min using an EGM-4 infrared gas analyser and a CPY-4 canopy assimilation chamber (PP Systems, Hitchin, UK). At the same time as CO 2 flux measurements were made, the depth of the water table below the peat surface and soil temperature 5 cm were measured at each plot (n = 36).
As below-ground respiration has been shown to be strongly controlled by soil temperature (Lloyd and Taylor, 1994), which varies diurnally, adjusting soil respiration to a fixed temperature removes this temporal variability. The By using the Q 10 values calculated by Equation (3) for total, heterotrophic and autotrophic below-ground respiration for each site (refer to Table 2 in the supporting information), all respiration rates were normalized to 10°C (r 10 ) [Equation (4)].
Equation (4) r 10 ¼ R t : Q 10 10 Àt= 10 (4) where r 10 is the temperature-adjusted respiration at 10°C for a measured respiration rate (μmol m À2 s À1 ) (R t ) at a given location (n = 1) at temperature t°C and Q 10 as mentioned previously.

Vegetation composition and primary productivity
Annual net primary productivity (ANPP) was measured in late August by destructive harvest of a 55 × 55 cm plot (n = 36) approximately 2 m downgradient of the NEE plot. Vegetation composition of vascular plants and bryophytes (% cover) of the NEE plot was estimated by visual inspection in August. As 14/18 of the species present were observed in less than six locations, the total percentage cover of non-Molinia species was calculated. The number of species present at each location was counted to derive the species richness. The inverse Simpson diversity index [Equation (5)] was also calculated. This determines the probability that two individuals randomly selected from a sample will be of the same species. D increases from 1 to n as diversity increases. Equation (5): inverse Simpson diversity index Published Ellenberg's moisture indicator values (Hill et al., 1999) were assigned to all species identified where these species have been classified. These ranged from 6 (moist to damp, e.g. Vaccinium myrtillus and Gallium saxatile) to 9 (wet, e.g. Narthecium ossifragum and Sphagnum fallax); M. caerulea has a value of 8 (damp to wet). Ellenberg's moisture indicator values for the classified species present at each location were averaged to give the Ellenberg's moisture indicator value.

Statistical analysis
To test for spatial variation, a two-way analysis of variance (ANOVA) was carried out on seasonal mean water table depths, measured vegetation indices, P G600 and R Eco with site, proportional distance from the ditch and proportional distance from the ditch nested within site as between-subject effects. A repeated-measures ANOVA was carried out on below-ground respiration rates, with site or proportional distance from the ditch (plot) as between-subject factors and sampling round (time) as a within-subject factor. Post hoc least squares difference tests were carried out to identify statistically different groups. Linear and quadratic regressions were carried out to test for a relationship between distance from the ditch and water table depth, measured vegetation indices, P G600 , R Eco and below-ground respiration. The most significant relationships are reported.
It was expected that spatial variation in CO 2 fluxes would be driven primarily by variation in water table depth and/or vegetation composition. Water table depth and vegetation indices (percentage cover of Molinia, leaf litter and non-Molinia species, ANPP, species richness, inverse Simpson diversity index and Ellenberg's moisture indicator values) were regressed against CO 2 fluxes (modelled ecosystem respiration, modelled photosynthesis, seasonal mean total, heterotrophic and autotrophic below-ground respiration at 10°C for each location). All statistical analyses were performed with SPSS 19 (SPSS Inc., Chicago, Illinois, USA).

Spatial variation with distance from a drainage ditch
Water table depth was deepest closest to the ditch (⅛ distance) and became shallower at ½ distance ( Figure 2a), but the difference was not significant (p = 0·197, Table III). Although vegetation properties showed variation (Table II;  Table 3 in the supporting information), analysis of variance indicated none varied significantly with proportional distance from the ditch (Table III). Both percentage coverage of non-Molinia species and the Simpson diversity index were lower at ⅛ distance than at ½ distance ( Figure  2d and e). However, the difference between proportional distances from the ditch (Table III) was not significant (p = 0·083 and p = 0·076 for non-Molinia and the Simpson diversity index, respectively). Species richness (Figure 2f) also showed a non-significant increase at greater proportional distance from the ditch but no significant relationship with absolute distance from the ditch (p = 0·135). ANPP, percentage cover of leaf litter and Molinia were lowest at ¼ distance (Figure 2g, h and i), but showed no significant differences between proportional distances from the ditch (Table III) or from the absolute distances from the ditch (Table V).  Average Ellenberg's moisture indicator values were greater at ⅛ distance than at ½ distance (Figure 2b), indicating a drier plant community further away from the ditch, which contrasts with the measured water table depths (Figure 2a). Ellenberg's moisture indicator values ranged from 6·6 to 8·5 with a mean of 7·5 (constantly moist or damp but not wet) and showed a non-significant positive relationship (r 2 = 0·08, p = 0·105) to water table depth, indicating drier conditions (lower Ellenberg's moisture indicator values) occurring where the water table was closer to the soil surface. Ellenberg's moisture indicator values could not be used as indicators of moisture conditions and were therefore excluded from further investigation.
Neither modelled P G600 nor R Eco varied significantly with proportional distance from the ditch (Table III) or absolute distance from the ditch (Table V). P G600 was greatest furthest from the ditch and least at ¼ distance ( Figure 3a) whilst R Eco increased non-significantly between closest to the ditch and ¼ distance (Figure 3a). The interaction term between site and proportional distance from the ditch was not significant for either P G600 or R Eco (Table III), indicating that the effect of proportional distance from the ditch did not depend on which site was being analysed. No below-ground respiration source varied significantly with proportional distance from the ditch (Table IV) or absolute distance from the ditch (Table V). However, they were all greatest at ¼ distance with ½ distance smallest (Figure 4a).

Spatial variation between sites
Water table depth, percentage coverage of leaf litter and peat depth all varied significantly between sites (Table III). Site A2 was drier and site S1 wetter than all the other sites (Table II). Peat depth was also significantly greater at site S2 than at all the other sites (Table I). The sites could be divided into two groups based on the percentage coverage Table III. Two-way analysis of variance for mean water table depth, vegetation indices, peat depth and modelled ecosystem respiration (R Eco ) and photosynthesis at 600 μmol photons m À2 s À1 (P G600 ) (μgC m À2 s À1 ) at all locations (n = 36) with site, proportional distance from the ditch (plot) and proportional distance from the ditch nested within site as between-subject variables.
Values where p < 0·050 are shaded dark grey. SS, sum of squares; df, degrees of freedom; MS, mean sum of squares; F, F ratio; p, significance. 413 DRAINAGE DITCHES, PEATLAND VEGETATION DIVERSITY AND CO 2 FLUXES of leaf litter, those with ≥85% coverage (A1, A2, A3 and S2) and those with ≤85% coverage (A3, S1 and S3) ( Table  II). The interaction term between site and proportional distance from the ditch was not significant for any of the spatial variables tested (Table III), indicating that, for example, the effect of proportional distance from the ditch on water table depth did not depend on which site was being analysed. P G600 varied significantly between sites (Figure 3b), with photosynthesis at site A2 greater than at sites S1 and S2 with the other sites intermediate. Sites A2 and S1 were also the driest and wettest, respectively (Table II). Ecosystem respiration was also greatest at site A2 but did not vary significantly between sites (Table III). Total and heterotrophic below-ground respiration showed similar spatial variation between sites with respiration greatest at S1 and least at A2, whereas autotrophic respiration was greatest at S3 and least at A3. Only heterotrophic respiration varied significantly between sites (Table IV) with site S1 having significantly greater respiration than all the other sites (Figure 4b). The significant interaction term between site and sampling round (time) for heterotrophic respiration (Table IV) indicates that heterotrophic respiration varied differently over time dependent on which site was being analysed.

Drivers of spatial variability
Percentage cover of non-Molinia species and the inverse Simpson diversity index both showed a significant negative covariance with water table depth (Table V). Greater diversity and more non-Molinia species occurred where the water table depth was closer to the surface.
Water table depth showed a significant positive covariance (Table V) with P G600 ; greater photosynthesis occurred where the water table was deeper. No other CO 2 fluxes co-varied with water table depth. P G600 showed no significant relationships with ANPP or vegetation composition indices (Molinia, non-Molinia, leaf litter, species richness or inverse Simpson diversity index) ( Table V).   (Table V). However, autotrophic respiration showed some level of co-variation (r 2 = 0·09, p = 0·077) with percentage cover of Molinia; autotrophic respiration was greater where there was more Molinia coverage. Heterotrophic respiration significantly co-varied (r 2 = 0·11, p = 0·046) with peat thickness with greater respiration where the peat was thinner (Table V) whilst total soil respiration was significantly greater (r 2 = 0·15, p = 0·021) where there was less coverage of leaf litter.

Drainage ditches, water table depths and vegetation
The drainage features on Exmoor although small (typically <0·5 m wide and <0·5 m deep) penetrate deep into the shallow peat [Exmoor average 0·33 m (Bowes, 2006)] and are regularly spaced (approximately 20 m) (Figure 1), making them important spatial features governing ecohydrological processes in these uplands (Grand-Clement et al., 2013). The mean water table was deeper closer to the ditch than at ½ distance (Figure 2a), but the significant variability between sites (Table III) was such that the difference between proportional distances from the ditch was not significant. Differences between sites may have been due to a combination of different site    Coulson et al. (1990) found no significant difference in water table depths, possibly due to high rainfall conditions. This non-significant variation in water table depth may explain why there was no significant (Table III) variation in percentage coverage of non-Molinia species and the inverse Simpson diversity index. Coulson et al. (1990) found coverage of C. vulgaris to increase away from the ditch with a concurrent decrease in grass species in two low-altitude British blanket bogs where they observed difference in water table depths, but in two higher altitude, higher rainfall bogs where no variation in water table depth was observed, there was no significant change in vegetation composition.

Site, water table depths and vegetation
Given the range of average water table depths for each site in this study (11-34 cm) (Table II), and previous work, a greater range of vegetation communities across the sites studied would be expected. For example, a 22 to À2 cm range in mean water table depth affected a change in vegetation from E. vaginatum to Scheuchzeria palustris in an undisturbed Finnish fen (Riutta et al., 2007). In this study, M. caerulea (86 ± 3%) dominated with minimal non-Molinia species present (9 ± 2%). This vegetation composition reflects deeper water table depths under wet conditions, the competitive nature of Molinia, its ability to flourish where water table depths fluctuate (Jefferies, 1915) and tendency to exclude other plants (Taylor et al., 2001). Minimal variability in vegetation composition would be expected to limit the magnitude of variation possible in CO 2 both between sites and with distance from the ditch.
Across all locations, more non-Molinia species (r 2 = 0·15, p = 0·021), greater species richness (r 2 = 0·16, p = 0·087) and higher diversity (r 2 = 0·14, p = 0·024) occurred where the water table depth was closer to the surface (Table V), indicating that Molinia may be less dominant where water tables are shallower enabling other species to grow. This finding is similar to that of Laine et al. (2007) who found species richness to decrease as water table depths dropped below approximately 10 cm in an undisturbed Irish blanket bog.
Other studies in pristine peatlands , Maanavilja et al., 2011 found vascular green area to increase as water tables fell and vegetation composition changed. In this study, ANPP was not affected by water table depth (Table V). Again, this most likely reflects the greater vegetation diversity and wetter conditions within these studies compared with that observed on Exmoor. Rutter (1955) found mean water table depth to determine the shape of the Molinia tussock and the vegetation composition present in a wet heath. It is known that vegetation structure varies with wetness in these catchments . However, in this study, mean water table depth did not relate to percentage coverage of Molinia (Table V). It may be that the vegetation survey failed to capture structural variation as the long spreading leaves covered most of the plot, resulting in limited variation in Molinia cover between locations (86 ± 3 %).
Bellamy et al. (2012) found a wet vegetation index (based on vegetation with an Ellenberg's moisture indicator value of 8-10) to be lowest 0·5 m from the ditch in a blanket bog and increase with distance from the ditch. Conversely, an index of drier vegetation (Ellenberg's moisture indicator values of 4-7) was highest close to the ditch and decreased with distance. In the current study, Ellenberg's moisture indicator values decreased (nonsignificantly) with increased distance from the ditch, indicating wetter conditions nearer the ditch (Figure 2b). They also increased where water table depths were closer to the surface; contrary to expectations given, Ellenberg's moisture indicator values range from 1 (extreme dryness) to 12 (submerged plants) (Hill et al., 1999), demonstrating that Ellenberg's moisture indicator values are not appropriate as a proxy for wetness in this relatively dry and low diversity environment where only N. ossifragum (9) and S. fallax (9) had higher indicator values than M. caerulea (8).
Spatial variability of CO 2 fluxes CO 2 fluxes from the Molinia-covered peatland did not vary significantly with proportional distance from the ditch (Tables III, IV), arguably due to limited variation in either water table depth (Figure 2a) or vegetation composition (Figure 2d-g). In other studies where spatial features such as ditches have been explicitly monitored, clear differences in functional responses have been measurede.g. ephemeral erosional gullies in a British blanket bog have been shown to have significantly higher ecosystem respiration (McNamara et al., 2008, Clay et al., 2012 and photosynthesis (Clay et al., 2012) than the surrounding blanket bog. These gullies were 416 N. GATIS et al deeper (up to 3 m) and wider (5-9 m) and had a greater effect on both water table depth and vegetation community than the smaller drainage ditches of Exmoor. It is likely that CO 2 fluxes varied across ephemeral erosional gullies due to the spatial variation in vegetation and vegetation cover with the greatest rates of photosynthesis and ecosystem respiration from Eriophorum communities and lowest fluxes from bare peat.
Photosynthesis weakly (r 2 = 0·13, p = 0·034) co-varied with water table depth; greater photosynthesis occurred where the water table was deeper. This may have occurred as drier conditions encourage greater above-ground biomass (Murphy and Moore, 2010) or promote increased coverage by graminoids, which have been shown to have higher NEE rates than mosses (Otieno et al., 2009) and C. vulgaris (Aerts, 1990). Although the cover of non-Molinia species was less where the water table was deeper (Table  V), there was no significant relationship between P G600 and non-Molinia directly. In addition, P G600 showed no significant relationships with either ANPP or percentage coverage of Molinia (Table V). It is possible that despite adaptations to live in conditions of fluctuating water table depths (Taylor et al., 2001), the conditions for M. caerulea were sub-optimum at the wetter locations (e.g. site S1), thus reducing photosynthesis.
Where water table depth has been found to control spatial variation in photosynthesis, fluxes have also been greater in drier microforms (Laine et al., 2006, Maanavilja et al., 2011, Schneider et al., 2012. In such studies, there was clear differentiation in water table depths and vegetation community between microforms so it is less clear if variation in photosynthesis was due to water table depth, vegetation community or both. Conversely, Bubier et al. (2003) found different microforms to have similar rates of photosynthesis, despite variation in vegetation composition and water table depth, due to similar leaf biomass.
Where variation in photosynthetic rates have been assessed within microforms in a pristine Finnish fen, hummocks and E. vaginatum lawns and hollows were found to respond to water table depths but not Carex lawns (Riutta et al., 2007). However, in a pristine Russian boreal peatland, Carex lawns showed the greatest within microform variability in photosynthesis driven by variation in vegetation composition and water table depth (Schneider et al., 2012), indicating the uncertainty in assessing controls on photosynthesis within a microform.
Where ecosystem respiration has been found to vary between microforms in pristine peatlands (Bubier et al., 2003, Laine et al., 2006, Maanavilja et al., 2011, Juszczak et al., 2013, wetter areas had distinct vegetation communities and lower respiration rates. Again, the minimal variation in vegetation composition observed in this study may explain why there was no statistically significant spatial variation in ecosystem respiration (Table III). This finding suggests that photosynthesis is the main control on the spatial distribution of NEE. Riutta et al. (2007) also found photosynthesis to vary more between communities than ecosystem respiration in a Finnish fen.
Peatland restoration programmes (Grand-Clement et al., ) typically aim to raise water tables and re-establish the ecohydrological structure and functionality of peatlands. Re-colonization by peat-forming Sphagnum-rich vegetation communities has been identified as particularly important to promote carbon sequestration (Lunt et al., 2010). Raising mean water table depths may have no effect on heterotrophic respiration of the peat store but decrease photosynthesis (Table V) shifting the ecosystem towards a greater CO 2 source unless change in water table depth is sufficient to alter the vegetation composition (and leaf litter quantity and quality) beyond that observed in this study.
As below-ground autotrophic respiration is strongly dependent on photosynthesis (Metcalfe et al., 2011), it would be expected that autotrophic (root) respiration would mirror photosynthesis and be dependent on water table depth. Instead, photosynthesis ( Figure 3) and autotrophic respiration (Figure 4) showed dissimilar spatial patterns and varied with different spatial variables ( Table V), suggesting that autotrophic respiration was controlled by factors additional to photosynthetic activity, such as morphological differences in root biomass (Heinemeyer et al., 2012), variation in the allocation of carbon between growth and maintenance (Bond-Lamberty et al., 2004) and moisture and nutrient availability (Chapman and Thurlow, 1998). Autotrophic respiration showed the strongest (r 2 = 0·09) and most significant (p = 0·077) co-variation with percentage cover of Molinia, with greater autotrophic respiration where there was more Molinia coverage. As above-ground and below-ground biomass have been shown to be linked (Murphy and Moore, 2010), it may be that where there is greater Molinia coverage, there is greater root biomass, resulting in increased root respiration and microbial respiration of root exudates.
Neither total nor heterotrophic below-ground respiration varied with water table depth (Table V). Jaatinen et al. (2008) found long-term (45 years) water table drawdown of a fen to increase total soil respiration rates in the driest areas; however, the decomposition potential of the substrate remained greater in the wetter areas. As the drainage ditches on Exmoor are over 150 years old, much of the labile organic matter will have already degraded (Bridgham and Richardson, 1992). This will have left a humified peat that, although potentially vulnerable to priming (Freeman et al., 2004, Fontaine et al., 2007, is less responsive to variation in water table depth than recently drained peat. Grand-Clement et al. (2014) found consistently low humic to fulvic acid ratios for dissolved organic carbon from these catchments, indicative of more humified peats. 417 DRAINAGE DITCHES, PEATLAND VEGETATION DIVERSITY AND CO 2 FLUXES It was expected that heterotrophic respiration would vary with changes in leaf litter quality and quantity (Straková et al., 2011a). Heterotrophic respiration showed no significant co-variation with percentage coverage of non-Molinia or inverse Simpson diversity index (Table V), variables influencing litter quality or ANPP, a measure of litter quantity. This may be due to the limited variation in vegetation composition and ANPP observed (Table II). Instead, heterotrophic respiration significantly co-varied with peat thickness (Table V) with greater respiration occurring where peat was thinner. This finding is contrary to those of other studies where thinner peats were found to have lower heterotrophic respiration rates in a C. vulgaris blanket bog (Hardie et al., 2009) and also where peat thickness was found to have little effect on heterotrophic respiration as most respiration occurred near the surface of the peat (Blodau et al., 2007) primarily due to a lack of oxygen below the water table. There is no obvious explanation for this relationship.
Where microforms or spatial features were found to have distinct CO 2 fluxes, they have been mapped and used to upscale CO 2 fluxes across a landscape (Laine et al., 2006, Riutta et al., 2007, McNamara et al., 2008. In these Molinia-dominated peatlands, there was significant spatial variation (Table III) in P G600 . However, this was not directly associated with proportional distance from drainage ditches (Table III), so mapping these features cannot be used directly to upscale CO 2 fluxes in this landscape. Given the sparse vegetation in the ditches, it is unlikely that these would have large CO 2 fluxes; however, as these were not measured, it is currently unknown if these are important when estimating landscape scale fluxes.

CONCLUSION
Modelled CO 2 fluxes (photosynthesis and ecosystem respiration, total, heterotrophic and autotrophic belowground respiration) showed no significant spatial distribution in response to drainage ditches, arguably due to a lack of significant spatial distribution in water table depths and minimal variation in vegetation composition (percentage cover of non-Molinia species and inverse Simpson diversity index).
Across all locations (n = 36) where the average water table depth was closer to the surface, more non-Molinia species coverage, increased vegetation diversity and reduced P G600 occurred, indicating wetter conditions may be sub-optimum for M. caerulea. Our data emphasize that substantial reductions in heterotrophic respiration may not always occur following restoration, unless water tables rise to be consistently very close to the soil surface. As a consequence, raising mean water table depths through ecohydrological restoration may shift the ecosystem towards greater CO 2 release unless the vegetation composition alters beyond that observed in this study.
Modelled P G600 showed significant spatial variation between sites and significantly co-varied with water table depth. This offers a potential means to estimate CO 2 fluxes at a landscape scale. Although water table depth showed variation between proportional distances from the ditch, the uncertainty is such that it should not be assumed that water table depth is distributed according to proportional distance from a drainage ditch. Therefore, other methods of determining the spatial distribution of water table depth, which may be only partially explained by ditch density, such as vegetation structure (Rutter, 1955) and thermal emissivity, ) should be explored.