Introduction

Allowing species to re-assemble spontaneously into natural communities (‘self-design’) has become an important principle in restoration ecology, lowering project costs by avoiding the need for expensive interventions such as planting and landscaping (Odum 1989; Mitsch et al. 1998). The rationale of using self-design to guide restoration is that target species composition can be achieved by manipulation of one or more environmental parameters, thereby modifying the environment to favour desirable over undesirable species. The target species of such manipulations in freshwater palustrine wetlands are most often vascular plants that dominate cover and provide the habitat for species at higher trophic levels. As hydrology and nutrient availability are the two most important factors controlling vegetation composition in wetlands (Bridgham and Richardson 1993; Rheinhardt and Faser 2001), restoration usually involves attempting to recreate pre-disturbance hydrological regimes and nutrient supply (David 1999; Fennessy and Mitsch 2001; Ross et al. 2003). The development of a natural vegetation assemblage is therefore a common restoration goal, but species composition in created and restored sites often fails to match that of comparable reference sites (Zedler 2000; Campbell et al. 2002).

When water tables have been lowered by drainage, loss of wetland character is typified by an increase in the number and cover of upland species at the expense of facultative and obligate wetland species (Wilcox 1995). Reduction in soil water content increases oxygen penetration into soils, resulting in a shift from anaerobic microbial processes such as methanogenesis and denitrification to aerobic respiration, and an increase in soil redox potential and oxygen concentrations (Laanbroek 1990). Upland species, which require external soil oxygen for root respiration (Armstrong and Beckett 1985), are then favoured over wetland species that can survive in permanently flooded soils because of their capacity for root aeration by internal oxygen transport (Armstrong and Beckett 1987; Sorrell et al. 2000). When soils are re-flooded, it is expected that soil biogeochemistry will follow a predictable series of redox transformations that re-establish the anoxic conditions, which favour wetland plants and exclude upland species. Whether re-flooding is sufficient to achieve this is likely to be an important limitation to successful restoration, but the relationships between hydrological regime, soil biogeochemical processes and vegetation composition remain poorly understood.

The aim of this study was to use a field restoration experiment to understand the responses of soil biogeochemistry and floristic composition to hydrological manipulations in a New Zealand fen with a long history of drainage. Hydrological manipulation is being widely applied in New Zealand as elsewhere as a tool to reverse the country’s well-documented loss of wetland resources (Mitsch and Gosselink 2000), but there is still little understanding of the effectiveness of the technique or how it changes soil and florisitic patterns. We initially used a 2-year hydrological record to characterise the effects of drains in the wetland on soil water regime, and related this to gradients in biogeochemical processes and community composition relative to drains. We then tested the effect of hydrological restoration on water regime, soil biogeochemistry and vegetation.

Materials and methods

Study site and experimental design

The Bullock Creek Polje is an enclosed depression in a steep karst landscape with a catchment of indigenous forest cover in the Paparoa Ranges, South Island, New Zealand. The study area is a remnant of a palustrine fen (sensu Cowardin et al. (1979)) within the polje that drains northwards to a narrow stream, which flows into the sinkhole to the northwest (Fig. 1). The waters then travel through complex underground passages to emerge as springs in a number of downstream rivers. During periods of high rainfall, the sinkhole is unable to cope with the volume of floodwater and the water flow reverses, flooding the wetland and discharging southwards into Bullock Creek. Water input is therefore a combination of rainfall and overland floods from the catchment, including the backflow from the submergence. Further descriptions of the hydrology and of the site are given elsewhere (Williams 2004).

Fig. 1
figure 1

Location of sampling sites relative to network of drains in the fen at Bullock Creek polje, South Island, New Zealand. Areas of wetland demarcated by drains are labelled A–H, 13 sites sampled for soil and environmental data are marked 1–13, and normal directions of water flow in non-flooding periods are shown by arrows. Transects of capacitance probes and piezometers (5, 15, 30 and 60 m from drains) are shown by dashed lines marked ‘tr’ in blocks D and E; additional capacitance probes were placed at sites 1, 2 and 8. Dam installed on junction of drain W3 and central drain on 28 March 2001

Full size image

Human disturbance in the polje began in 1867 with the logging of forest, harvesting of New Zealand flax (Phormium tenax) for its fibre, and conversion of most of the area south of Bullock Creek to farmland. The drain network shown in Fig. 1 was developed in stages during the twentieth century in an attempt to convert the northern area for grazing, but the site remained uneconomic and the entire polje was returned to public ownership for conservation in 1986. A main central drain runs from south to north with side branches to the west (labelled W1–W4) and east (labelled E1, E2, NE). The drains are unlined and mostly steep-sided with higher spoil banks on one side created during construction. The drains are acidic (pH 5.2–6.2, conductivity 50–100 μS cm−1) throughout, except the northeastern drain (“NE”, Fig. 1), which has pH 7.2–7.4 and conductivity 190–260 μS cm−1. Inorganic nitrogen in the drains is solely NH +4 (20–50 μg l−1), and orthophosphate is undetectable.

Much of the vegetation in this northern area is still dominated by indigenous wetland plants, but alien pasture species dominate drier areas, especially close to drains. The predominant native species is the tussock-and sward-forming sedge Carex sinclairii, but there are also discrete patches dominated by the taller, dense sedge Baumea rubiginosa, and the peat-forming moss Sphagnum cristatum. The management goal for the site is to return the ecosystem to a more indigenous condition, reducing the presence and cover of pastoral alien plants.

Relationships between hydrology, edaphic parameters, nutrients and community composition were established between November 1998 and March 2001. Sampling locations were selected to focus on gradients in relation to presence of drains, with continuous hydrological records obtained from permanently installed piezometers and water level recorders. Soil, soil water and physico-chemical measurements were taken in November 1999 and March 2000 at the 13 sites shown in Fig. 1. These included five paired sites in blocks B, C, D, and E located 5 and 30 m from drains E1, E2, W3 and W4. Three additional sites were placed to increase replication of major vegetation types poorly represented in the paired sites, including the alien-dominated pasture (site 8), Baumea rubiginosa sedgeland (site 2), and Sphagnum mossland (site 9).

On 28 March 2001, we constructed a 1.2 m-high compacted soil dam across the W3 drain at its confluence with the main central drain, creating a pond in the W3 drain and visibly raising the water table in all areas west of the main drain. Hydrological measurements continued following the manipulation. Edaphic measurements and vegetation descriptions were repeated in March 2002 and October 2002. We then applied BACI (Before-After-Control-Impact) designs (Underwood 1993) to test for effects of flooding on soil parameters and vegetation composition. Sites 1–9, east of the main central drain, were treated as control sites, and sites 10–13, west of the main drain, were treated as impact sites, as blocking drain W3 increased water levels at all these sites.

Hydrological regime

The hydrological regime of the site and effects of the drains were characterised by transects of automatic water level recorders and tensiometers perpendicular to drains W3 and W4 (Fig. 1), and additional recorders placed in other vegetation types at sites 2, 3, and 8. We used capacitance type water level recorders to monitor water levels, capable of detecting changes of water level up to ±8 mm accuracy with a 3 mm resolution. They were programmed to average 30 readings in 30 min intervals, and record water levels in 1-h time intervals for up to 16 months. Readings were corrected for a slight diurnal temperature effect. Recorders were installed in boreholes of 0.08 m diameter with a PVC casing, with recorded values corrected to local water levels from positions relative to local ground level. Tensiometers were differential pressure transducers connected to standalone dataloggers, and allowed water level to be tracked below the lower limit of the capacitance probes. Throughout the study, a Texas Instruments rain gauge with a tip size of 0.1 mm was used to record rainfall as 10 min totals. A nest of six time domain reflectometers and eight tensiometers in Block D allowed water level-water content data to be converted to soil moisture content. Water temperature, velocity and depth exiting the polje in the main central drain were recorded continuously with a Starflow ultrasonic Doppler recorder (Unidata, WA, Australia).

Vegetation sampling

Initial vegetation sampling for community characterisation was carried out in late summer of both 1999 and 2000 from 481 plots (2 m × 2 m), located on transects covering all blocks (A to H) in Fig. 1. Most plots were placed at 10 m intervals on transects, with additional plots concentrated close to drains where there were narrow bands of distinct vegetation types. Species were classified according to functional groups (Tiner 1999) and life history strategies (Grime 2001). At each plot, species cover was assessed using a modified Braun–Blanquet cover scale (Mueller-Dombois and Ellenberg 1974):

1: <1%, 2: 1–5%, 3: 5–25%, 4: 25–50%, 5: 50–75%, 6: 75–100%.

Cover score midpoints (%) for all species recorded in the 481 plots were subjected to classification (cluster analysis) and ordination techniques to define vegetation types and examine ecological gradients. The programs used were FUSE (Agglomerative Hierarchical Fusion) for Cluster Analysis, and SSH (Semi-Strong-Hybrid Multidimensional Scaling) for Ordination, within the PATN multivariate analysis package (Belbin 1995).

For determining environmental correlations with vegetation classes and for monitoring responses to drain closure, we used a sub-sample of 52 of the 481 plots, covering the 13 experimental sites (n = 4 per site) and permanently located by GPS. These were sampled in 2000–2001 for edaphic and nutrient parameters identified after pilot sampling in 1999 (see subsequently), and vegetation was re-surveyed shortly before the drain was closed in March 2001, and again in March 2002 and November 2002.

Soil and plant sampling and data analysis

We described general features of the soil (soil textures, peat depth, colour and organic matter content) from cores collected in a D-corer at all 13 experimental sites. To determine the most suitable environmental parameters for vegetation correlations, a pilot comparison was made of a wide range of soil (TC, inorganic content, bulk density, pH, conductivity, TN, TP, Olsen-P, H2SO4-P, available anaerobic-N, KCl-extractable NH +4 and NO −3 ) and interstitial water (TDN, TDP, DRP, NH +4 and NO −3 ) parameters collected at 0–0.1 m depth in areas dominated by Carex sinclairii, Baumea rubiginosa, Sphagnum cristatum and the pastoral aliens. Parameters that were both present at measurable levels and differed between these four vegetation types (TC, soil pH, conductivity and bulk density, TN, TP, available N and H2SO4-P) were selected for more detailed sampling at the 52 plots and correlation with communities identified from cluster analysis of these plots. These data, together with two soil oxidation parameters (redox potential (EH) and oxidation depth) and integrated soil water content data, were analysed using a vector-fitting approach to examine species-environment responses. We used Principal Axis Correlation within PATN to identify how well the environmental factors fitted into the ordination space, and plotted the vectors on a two-dimensional plot ordination to indicate the direction of best fit for each of the environmental variables and the correlation in that direction (Clarkson et al. 2004a, b).

To test gradients in nutrient limitation suggested by the species–environment responses, we collected representative samples of vascular plant material in each of the 52 plots for N and P analysis. To document any seasonal variation during growth cycles, we sampled both in spring (October 1999) and at peak autumn biomass (March 2000). To test gradients in species distribution relative to water regime suggested by the species–environment models, we collected live roots of the dominant species from oxidised sites and anoxic sites, as identified from EH data, washed from soil at 0.15–0.25 m depth where most live roots occur, for examination of anatomical properties known to confer flooding tolerance.

Differences in individual soil parameters between the 13 sites and before and after closing drains were analysed with two-way ANOVA (site × time) in Systat 10 after transforming (ln(1+x)) to normalise data, which showed high degrees of heteroskedasticity. When log-transformation failed to normalise data, we ranked the data before using ANOVA. Post-hoc comparison of means was carried out using Bonferroni tests. Differences in vegetation N, P and N:P ratio were also analysed by two-way ANOVA (community × season) after normalisation by log-transformation. Root anatomical properties were compared by two-way ANOVA with species and soil type (oxic vs anoxic) as the factors.

Analytical methods—soils

Interstitial water chemistry. We sampled soil water at a depth of 0.1–0.2 m in the root zone using a suction device as described by McKee et al. (1988). Conductivity and pH were measured in situ using a portable pH meter and solid-state probe (Sentron Model 1000, Roden, The Netherlands) and a portable conductivity meter (Hanna Instruments Model HI98114, Woonsocket, RI, USA). Temperature was measured simultaneously with a thermistor for corrections. Separate water samples for nutrient (200 mL) analysis were stored in acid-washed polyethylene bottles. Water samples were frozen in the field until they could be returned to the laboratory for filtering and analysis. Dissolved nutrients (nitrate, ammonium, orthophosphate, total P and total N) were determined by autoanalyser.

Soil chemistry. Volumetric soil samples were collected from the surface layer (0–0.07 m) in a 0.1 m diameter stainless steel ring for determination of bulk density and soil chemistry. All determinations were made after air drying samples. Bulk density was measured gravimetrically after drying the core in a forced air convection drier at 35°C. As there were significant differences in bulk density between sites, all results are presented on a volumetric basis. Soil pH and conductivity were determined on 10 g samples by standard methods (Blakemore et al. 1987). Organic matter content was determined as Total C with samples heated in a stream of high purity oxygen in a Leco furnace to produce CO2 and measured with an infrared detector.

Total N and Total P were measured following Kjeldahl digestion of soil and colorimetric analysis for ammonium and orthophosphate on a Lachat flow injection analyser. An index of available N was assessed using the anaerobic incubation method developed by Keeney (1982), which measures a pool of N rather than a mineralisation rate. Available pools of phosphorus were determined from 0.5 g sub-samples of dried soil extracted with either (1) an Olsen-P 0.5 M bicarbonate-extract, or (2) 0.5 M H2SO4 (100 ml) for 16 h, and analysed colorimetrically for orthophosphate.

Soil oxidation. Platinum wire electrodes for EH were constructed using 18-gauge platinum wire as described by Faulkner et al. (1989) and deployed in the field for 4 weeks prior to measurement. Duplicate electrodes were placed at 0.1 m depth in the root zone. Redox was measured using a portable millivoltmeter and an Ag/AgCl reference electrode, with 199 mV added to readings for conversion to EH values and no correction for temperature or pH as variation in these parameters was small. Soil oxidation depths were measured using the steel rod oxidation method, following the methods and precautions of Bridgham et al. (1991). We used 1 m long uncoated steel rods inserted 0.5 m into the soil for 4 weeks prior to measurements. Oxidation depths were recorded as the deepest point at which distinct rust bands could be observed. Soil oxidation measurements in the post-flooding data were supported by an additional technique, oxygen diffusion rates (ODR). ODR were determined using a commercially available meter and platinum electrodes (Jensen Instruments Model E, Tacoma, WA, USA).

Analytical methods—plant material

Tissue nutrients. Aggregated plant samples from each plot were dried at 70°C for 24 h and treated with salicylic acid in sulphuric acid followed by Kjeldahl digestion, with ammonium and orthophosphate determined colorimetrically on a flow injection analyser. We interpreted vegetation N and P concentrations for indicating nutrient limitation according to Güssewell and Koerselman (2002), i.e. that gravimetric tissue N:P ratios  <13:1 indicate N limitation and ratios  >16:1 indicate P limitation. Values between 13:1 and 16:1 indicate that either nutrient may be limiting or that there is co-limitation.

Root anatomy. Root segments were infiltrated with water under vacuum to remove air from the tissues, and fresh hand sections were prepared and stained with Toluidine Blue (0.05 w/v) and examined under the microscope. All sections were taken at 50 mm from the root tip. Four parameters known to be important for flooding tolerance—root diameter, area of the stele relative to the whole root, root porosity, and staining patterns (McDonald et al. 2002) were determined for all sections.

Results

Hydrology

Time courses of capacitance probe data showed patterns of sudden increases in water level in response to rainfall events, with rapid draining of standing water after floods, followed by more gradual decreases in soil water level between floods (Fig. 2). Periods of standing water were therefore relatively brief, in comparison with periods of drainage between floods, and water levels were below the soil surface for more than 90% of the long-term data records at all capacitance probe sites.

Fig. 2
figure 2

(a) Example of changes in water level relative to soil surface over time in relation to rainfall events (1 December 2000–7 February 2001), from capacitance probe data at 15 m from drain W4 on transect in block D (Fig. 1), compared to water level at outflow of main central drain. Dotted line shows soil surface. (b) Hourly rainfall totals and main central drain flow velocity over the same period. Note reversal of flow velocity from ‘normal’ northward to southward flow during the larger flood events

Full size image

The transects adjacent to the W3 and W4 drains and their responses to the damming of W3 revealed different effects of these two drains on soil water level (Fig. 3). Before and after comparison of data from the larger W3 drain shows that damming the drain raised water level at distances of 5 to 15 m from the drain, but there was little change in water level at 30 and 60 m from the drain, suggesting it had little effect at these distances. Lower water levels relative to the surface at 60 m are possibly due to this being a Sphagnum site with peat accumulation. The smaller W4 drain appeared to be relatively ineffective in drainage, as closing of W3 clearly increased flooding throughout blocks D and E, even though W4 was itself not closed.

Fig. 3
figure 3

Change in water level along transects from W3 drain and W4 drain after experimental flooding by closing W3 drain. Data are from two periods of similar rainfall and evaporation conditions before and after damming of W3 drain. Median values and interquartile ranges are shown (n = 400 daily averages of water level before and after damming)

Full size image

Median water table depth relative to the soil surface differed significantly among the four vegetation types where capacitance probes were installed (Fig. 4). In particular, the median value for areas dominated by pastoral aliens was considerably lower than for the other communities. There was also a greater interquartile range for the alien community, revealing a more variable water table with wider fluctuations, in comparison with relatively narrow ranges nearer the surface for the sites dominated by native species. Sphagnum areas were characterised by lower and more constant water tables than in the native Carex and Baumea vegetation, reflecting the higher elevation where they occurred in Blocks A and D, and during the driest periods they experienced less drawdown than the other vegetation. The wide range and number of outliers in Fig. 4 demonstrate the highly dynamic hydrology of the fen, with all the non-Sphagnum vegetation experiencing brief periods of both deep flooding and extreme drying. Figure 4 also shows that minimum and maximum average daily water tables were not useful discriminants of flood-tolerant and flood-intolerant vegetation at this site.

Fig. 4
figure 4

Box and whisker plots showing water depth relative to soil surface (daily averages) for areas dominated by Baumea rubiginosa, Carex sinclairii, alien pasture species, and Sphagnum cristatum prior to flooding. Boxes represent median values and upper and lower quartiles, whiskers are 90% intervals, with points showing outliers beyond 90%. Data are from 400 daily averages at each capacitance probe site. See text for explanation

Full size image

General soil properties

Coring revealed that the dominant soil was a recent peaty gley to at least 3 m soil depth. The subsurface material was poorly stratified, with a high content of organic material, and a layer of coarse sand at 0.6–1.5 m depth. The surface layer of the soil (to a depth of 0.07 m, as sampled using stainless steel rings) was also a peaty silt loam, with TC  ≤  30% in all sites and as low as 3% in some of the high spoil banks. Peat depth ranged from 0 m (no peat formation) at sites 1, 6 and 8 to a maximum of 0.9 m in the Sphagnum-dominated site 9. It was 0.2–0.3 m at most other sites, even when Sphagnum was present, although depths up to 0.7 m were found in Baumea sites.

Table 1 shows the values and differences of the physico-chemicals parameters and soil nutrients used in the pilot comparisons of soils. Those that were present in measurable levels and exhibited significant differences between the four vegetation types were used for vector fitting in the Principal Axis Correlation. KCl-extractable N pools and Olsen-P were either close to detection limits or did not differ between sites, and interstitial nutrient concentrations were also very low and did not differ between sites, so neither was used. Interstitial pH was significantly lower in Sphagnum areas than in other areas, but was not included in subsequent analyses as it did not discriminate amongst the other vegetation types, and because the low interstitial conductivity indicated low buffering capacity of the interstitial water and probable high temporal variability in interstitial pH.

Table 1 Pilot comparison of edaphic and nutrient parameters (±1 SD) for Bullock Creek soils in four main vegetation types
Full size table

Species composition and vegetation classification

A total of 52 species were recorded in the full dataset from the 481 plots, including 22 alien species and 30 native species (Appendix A). The native species included a wide range of growth forms, ranging from woody shrubs and tall emergent species to low-growing grasses, creeping herbs, ferns and mosses. Four key native species were both widespread and dominant, structuring the vegetation—B. rubiginosa, C. sinclairii, P. tenax, and S. cristatum. The alien species included one woody shrub (Ulex europaeus), rushes, grasses and herbs. Most alien species were restricted to pasture vegetation dominated by the grasses Holcus lanatus and Agrostis stolonifera and the herbs Ranunculus repens and Lotus pedunculatus. The alien rush Juncus canadensis was present throughout the wetland sites, often comprising the only non-indigenous species in these plots.

Cluster analysis of the 481 plots, based on the 1999–2000 data prior to hydrological manipulation, produced 12 distinct community groups (Appendix B). Ordination revealed an outlier group of sites dominated by native aquatic species that were restricted to the drains (Eleocharis acuta, Isolepis prolifer, Potamogeton cheesemanii, P. suboblongus), and this community was excluded from subsequent analysis. The groups showed an ordination gradient from sites dominated by native wetland species (groups 1–6) to those dominated by pasture species (groups 10–12), with groups 7–9 representing intermediate mixtures of native and alien species. As well as forming this native to alien sequence, the groups also loosely represented the hydrological gradient discussed earlier, with groups 1–6 generally found remote from drains in wetter areas, and groups 10–12 found close to drains in elevated sites.

Effects of drains were also evident from gradients in composition along transects located perpendicular to drains. Species density increased with distance from drains (Fig. 5a), reflecting the transition from a less species-rich community of competitive, fast-growing species close to the drains, to more species-rich communities of less competitive species away from the drains (Fig. 5b). C-S-D strategy classification (Appendix A) confirmed that alien taxa dominating close to the drains were mainly competitive C and C-D strategists, whereas the native species, which were the dominating species away from the drains were predominantly S and C-S strategists. Both alien and native groups included large numbers of facultative wetland species, but the alien group had many upland and facultative upland species. Cover of these aliens was higher on the sides of drains with high spoil banks than on lower drain banks.

Fig. 5
figure 5

Gradients in species density (±1 SD) with distance from drains for all plant species (a) and comparing alien versus native species (b)

Full size image

Figure 6 shows the ordination based on the subset of 52 permanent plots, with vectors summarising the relationships between species abundances and significant environmental variables. Seven vegetation classes derived from cluster analysis of the 52 plots have been superimposed on the ordination. The sequence of plots on axis 2, from those dominated by Agrostis stolonifera and H. lanatus, through mixtures of C. sinclairii with other taxa, to Sphagnum–Carex associations, are predominantly on a hydrological gradient. This axis has highly correlated vectors of increasing water content and decreasing soil redox potential and steel rod oxidation depth. There is also a soil nutrient gradient on this axis, from the more organic, peaty soil in the wetter sites to the more mineral, drier soils with higher total P and available P. The plots dominated by Baumea rubiginosa that separate with strongly negative loadings on axis 1 represent the highly discrete patches of this species, which had a dense growth form and excluded most other species. Soil conductivity and available N correlated most strongly with the Baumea plots.

Fig. 6
figure 6

Two-dimensional ordination of 52 plots based on canopy cover of species in Appendix A, and significant (P < 0.05) fitted vectors for environmental variables. The seven dominant community groups have been superimposed. The length of vector equals degree of correlation. AvN, Available N; BD, dry bulk density; Cond., soil conductivity; O depth, steel rod oxidation depth; TC, Total C; TN, Total N; TP, Total P

Full size image

Environmental data for the seven vegetation classes are shown in Fig. 7, to clarify gradients in properties that may drive composition within these associations. The greatest differences were in soil moisture and oxidation properties, which were strongly correlated with each other (Table 2). Deep soil oxidation was a particularly distinctive feature of associations with high representation of the alien pastoral species and was not seen in Baumea, Carex–Juncus and Sphagnum–Carex associations, which also had higher soil moisture content. There were also strong positive correlations between soil bulk density and both TP and available P content, and negative correlations between these parameters and TC. These correlations reflected the gradient from drier, higher-P soil close to the drain to wetter, more organic and infertile soil away from the drains.

Fig. 7
figure 7

Box-and-whisker plot summary showing medians, upper and lower quartiles (boxes), and 90% intervals (whiskers) for environmental data in each of seven vegetation types determined from the ordination. Dashed line on oxidation depth plot indicates that data are constrained by the maximum steel rod depth of 500 mm. Bau, Baumea rubiginosa; Car−Hol, Carex sinclairii-Holcus lanatus; Jun−Car, Juncus canadensis-Carex sinclairii; Car, Carex sinclairii; Sph−Car, Sphagnum cristatum-Carex sinclairii; Hol−Ran, Holcus lanatus-Ranunculus repens; Agr−Hol, Agrostis stolonifera-Holcus lanatus

Full size image
Table 2 Pearson correlations between environmental variables in 52 permanent plots using a Bonferroni adjustment for multiple comparisons
Full size table

Figure 8 presents tissue nutrient concentration data. The N concentration was significantly higher in spring than in autumn in all associations except Baumea, consistent with seasonal growth dilution of a limiting nutrient, whereas tissue P concentration showed less seasonal variation. Hence, N:P ratios decreased by the end of the growing season to values strongly indicative of N-limitation in most associations. Only Baumea-dominated vegetation had N:P ratios high enough to indicate P-limitation or N and P co-limitation.

Fig. 8
figure 8

Vegetation N and P concentrations and N:P ratios in spring and autumn for the seven vegetation types determined from the ordination. Dashed lines demarcate the N:P = 13:1 and N:P = 16:1 lines that are thought to discriminate N-limited and P-limited vegetation (Güssewell and Koerselman 2002). Baumea, Baumea rubiginosa; Car−Hol, Carex sinclairii-Holcus lanatus; Jun−Car, Juncus canadensis-Carex sinclairii; Car, Carex sinclairii; Sph−Car, Sphagnum cristatum-Carex sinclairii; Hol−Ran, Holcus lanatus-Ranunculus repens; Agr−Hol−Car, Agrostis stolonifera-Holcus lanatus-Carex sinclairii

Full size image

Differences in root anatomy amongst the dominant species at the site (Table 3) included a trend to greater root diameter, greater porosity and lower relative contributions of the stele to root area in species characteristic of anoxic, flooded areas such as C. sinclairii, B. rubiginosa and J. canadensis. Species of the pastoral alien association had higher root porosity in the flooded environments, but lower porosity than the wetland species, which either had highly porous roots in all environments, or had much higher porosity in flooded environments. Table 3 also shows that other key anatomical properties conferring flooding tolerance, such as higher root diameter and lower fraction of the root occupied by the stele, were also expressed further in the wetland species than in the pastoral alien species.

Table 3 Anatomical properties of roots of two native and four alien species dominating vegetation cover, collected from sites with oxic (EH >400 mV) and anoxic (EH < 250 mV) soil
Full size table

Soil and vegetation responses to flooding

Comparison of soil oxidation and nutrient parameters before and after closing the W3 drain showed large changes in soil oxidation parameters but no significant change in soil nutrient concentrations before and after flooding (Table 4). There were large differences in soil nutrient availability amongst sites both before and after flooding, but the patterns shown in Fig. 7 remained unchanged after the flooding manipulation. ANOVA revealed significant differences in redox potential, oxidation depth and water content between sites both before and after the flooding manipulation, but with a strong interaction term (Table 4) that reflected a change in these parameters at some sites only. Figure 9 compares these three parameters amongst sites before and after flooding, showing no change in the control area (sites 1–9) and large changes in the impact area (sites 10–13). Increases in soil water content occurred at all four sites, including sites 12 and 13 at 5 and 30 m from the unblocked W4 drain, consistent with the capacitance probe data. Decreases in redox potential occurred in sites 10 and 12, the sites that were close to drains and well-aerated (EH >300 mV) prior to flooding, accompanied by decreases in soil oxidation depth from >500 to <50 mm depth. However, there was no significant decrease in EH in the sites where soil was already de-oxygenated before flooding (sites 11 and 13, 30 m from drains), although soil oxidation depth decreased significantly at site 13. Oxygen diffusion rates, although only available after flooding, supported the differences between sites from other soil oxidation measures. Dry sites (1, 4, 6, 8) had ODR rates from 0.11–0.15 μg cm−2 min−1, whereas the other sites, all flooded at the time of sampling, had ODR rates from 0.03–0.06 μg cm−2 min−1.

Table 4 F ratios for two-way ANOVA comparison of soil oxidation and nutrient parameters before and after flooding in the 13 sampling sites
Full size table
Fig. 9
figure 9

Comparison of redox potential, oxidation depth and water content at the 13 sites (Fig. 1) before and after drain closure in March 2001. Oxidation depth data are constrained by the maximum steel rod depth of 500 mm

Full size image

Decreases in soil oxidation status in sites 10–13 were accompanied by decreases in presence and cover of the less flood-tolerant alien species (Table 5). Species such as H. lanatus and L. pedunculatus decreased in cover at all four impacted sites, including those 30 m from the drain as well as the sites 5 m from the drain, where they had high cover in 1999 and 2000. The least flood-tolerant species, Ranunculus repens and A. stolonifera, disappeared from site 10, 5 m from the blocked drain, where they had been common in 1999 and 2000. Of the alien species, only the flood-tolerant J. canadensis was unaffected by flooding and remained locally common with low cover (<40% in all plots). Cover of C. sinclairii, the dominant native wetland species, also did not change at these sites. Hence, total plant cover in 2002 was lower at sites 10, 12 and 13 than in the reference sites, due to the loss of the upland species. Site 11 was the only impact site to retain high total cover in 2002, due to its low cover of alien species in 2000. Cover remained high and composition did not change at the nine reference sites.

Table 5 Comparison of percent cover (median values) for those species contributing 5% or more cover at the 13 sites in 2000 (before flooding = B) and 2002 (after flooding = A)
Full size table

Discussion

This study characterised the hydrologic and edaphic conditions of the dominant plant communities that occurred along moisture gradients in a modified fen, and how the zonation of the dominant communities were related to water level fluctuations and associated soil oxidation, both before and in response to hydrological manipulation. In the wetland communities remote from the drains, the dominant Carex and Baumea species were tolerant of shallow water table depths and anoxic soil conditions, which apparently limited the distribution of the alien species. Strong correlations between species composition, soil moisture content, EH and soil oxidation depth revealed that the species composition corresponded to the dominant environmental gradient identified by the ordination.

Damming or filling in ditches usually rapidly re-establishes wetland hydrology (Bruland et al. 2003). Re-establishment of wetland soil properties can require much longer time periods, especially if past drainage caused extensive soil carbon loss (Stolt et al. 2000), but soil carbon loss at Bullock Creek appeared to be localised close to drains. Soil oxidation parameters in these areas nevertheless returned to similar values to flooded control areas within the 18-month timeframe of the study. This supports previous assertions that it is highly feasible to re-establish physico-chemical conditions and ecological values in partially degraded minerotrophic wetlands (Mitsch and Gosselink 2000; Weinstein et al. 2001), as opposed to synthesising them in wetland creation projects (Zedler 2000).

The absence of any changes in nutrient availability after flooding in our study allows the changes in species composition we observed to be attributed solely to the increase in water level. Flooding can cause changes in several soil properties affecting plant distribution, including porosity, structure and conductivity, and particularly decreases in soil oxidation. Lower oxygen penetration causes root mortality in less flood-tolerant species (Armstrong and Beckett 1987), and hence loss of such species and their eventual replacement by more flood-tolerant taxa. Our study suggests that restoring higher water tables is likely to be a successful management approach in sites where the undesirable alien species are less flood-tolerant than the desired wetland species, and this has also been trialled successfully elsewhere (David 1999).

Re-flooding may nevertheless fail to re-establish desired communities for a number of reasons. Soil oxidation during the drainage period may have depleted organic matter and nutrients, and caused soil subsidence (Armentano and Menges 1986), and these changes may be difficult to reverse (Stolt et al. 2000). The new water regime may lack the appropriate seasonal or short-term periodicity to create the spatial and temporal balance of aerobic and anaerobic metabolism for the desired vegetation. Standing water may persist for long periods, inhibiting internal aeration or photosynthetic carbon assimilation by the plants (Visser et al. 2000). In addition, the upland species, once established, may persist and compete well even when re-flooded, or there may not be an adequate seed bank of the desired wetland species for re-colonisation (Wetzel et al. 2001), and large fauna may disturb re-growing vegetation or introduce alien species.

Differences in degree of adaptation for flooding tolerance can explain most of the distribution patterns. Sedges such as Carexand Baumea form well-developed aerenchyma in their roots, ventilating respiring tissues and equipping them to survive flooding, allowing extensive belowground growth in anoxic environments (Moog 1998; Visser et al. 2000). Although the dominant pastoral alien species at Bullock Creek are flood-tolerant to some extent (Misra and Tyler 2000; He et al. 1999), the lower development of some key anatomical characteristics for flood tolerance in their roots—porosity, total diameter, relative stelar dimensions, and development of exodermal lignification (Sorrell et al. 2000; McDonald et al. 2002)—confers a lower degree of flood tolerance than in the sedges and Juncus species. Some of these semi-flood tolerant species (e.g. H. lanatus and L. pedunculatus) were common minor components of the vegetation remote from the drains, but were less competitive there than in the drier soil of the drain banks. It is also reflected in a decrease in percent cover at the sites impacted by flooding (sites 10–13) following the hydrological manipulation, as shown in Table 5. In contrast, Carex sinclairii, although dominant in wetter, anoxic soil, was still present in low cover in the drier sites. Carex species vary considerably in flood tolerance, and C. sinclairii has similar root anatomy to European species that, although characteristic of moderately reduced soils, can still survive in aerobic soil, such as C. limosa and C. davalliana (Visser et al. 2000). The EH values in the flooded sites at Bullock Creek are within the range normally considered to be anoxic but only moderately reducing, consistent with the relatively oligotrophic fen environment and flushing by frequent rainfall events that would favour such species.

The study also supports the importance of hydrologically controlled soil oxidation variables as determinants of species composition in fen and marsh environments (Bridgham and Richardson 1993). Gradients in vegetation composition, especially of dominant plant species, have previously been associated with soil oxidation depth, as revealed by the steel rod method (Bridgham et al. 1991) and by EH (Armstrong et al. 1985; Dwire et al. 2004). The rapid decrease in dominance of alien species with distance from drains, and their higher cover on the high spoil banks of drains than on the lower side, demonstrate that small differences in elevation relative to water table had large effects on composition. It is also illustrated by the decrease of alien species at the sites impacted by flooding, which was related to a decrease in redox potential and oxidation depth at these sites (Fig. 9, Table 5).

Well-drained sites often have greater species richness than adjacent wetland sites, because there are fewer flood-tolerant than flood-intolerant vascular species (Visser et al. 2000; Dwire et al. 2004). Wet fen meadows are often dominated by a few clonal species such as Carex spp., whereas drier sites can support a more diverse array of growth forms, such as forbs and graminoids (Dwire et al. 2004). However, although C. sinclairii often dominated cover in the wetter sites in Bullock Creek, its tussock growth form allowed other native species to co-exist at low cover, many of which were ‘interstitials’—weak competitors and S-strategists with limited ability for clonal spread (Boutin and Keddy 1993; Lenssen et al. 1999). Hence, in contrast to these studies, we found lower rather than greater species diversity in the drier soils, which can be attributed to the inability of the native species to compete with the predominantly C and C-D aliens when the latter are released from flooding stress. This supports the principle that abiotic factors are the main determinant of composition and species richness in flooded habitats (Gough et al. 1994; Lenssen et al. 1999), and that competition becomes more important in drier sites where stress-tolerators can be out-competed. Hydrological restoration at Bullock Creek is therefore unlikely to eliminate the alien plants from the system, but can reduce their cover to minor contributions within the native vegetation. Only one alien species at the site, Juncus canadensis, is flood-tolerant to the extent that it is likely to be encouraged by restored hydroperiod, but it rarely dominated cover, even in the control sites.

Low-nutrient sites such as Bullock Creek are also more likely to be structured by abiotic factors than high-nutrient sites, where greater productivity would increase plant interactions. The ordination revealed a strong nutrient gradient correlated with the moisture gradient at our site, which would also favour the competitive aliens. Our soil cores revealed decreasing peat depth and soil organic matter from south to north, reflected in the shift from high Sphagnum cover in Blocks A and D to increasing Carex sinclairii dominance at the northern end of the site. Overlain on this was the development of higher bulk density and nutrient availability in the driest soils on the spoil banks and deforested areas of Block A, where pastoral aliens were dominant, which we can attribute to the history of clearing, draining and grazing activity (Bruland et al. 2003). Further evidence of strong nutrient limitation was the very low proportion of available N and P relative to total N and P in soil pools, and the lack of any change in available N in the experimental area after flooding, which indicates that most of the organic N present was relatively refractory and unavailable for release to denitrifying bacteria.

Nutrients also provided a significant environmental discriminant between the Carex and Baumea communities, which occupied similar hydrological niches, but with deeper peat in Baumea sites. Large seasonal fluctuations in tissue N, and N:P ratios well below 13, indicated N-limitation in Carex and pastoral alien associations, whereas higher soil available-N and tissue N:P ratio indicated a trend to P limitation or N and P co-limitation in Baumea-dominated vegetation (Güssewell and Koerselman 2002). Vascular plant composition in peatlands is often primarily associated with N and P availability rather than pH and base cation levels (Vitt and Chee 1990; Verhoeven et al. 1996), and this apparently holds for Bullock Creek. Although the polje is in a karst landscape, most of the site has pH  <5.5 and low Ca and Mg concentrations, with only a very small area of circumneutral pH and elevated base cation concentrations at the northern margin, associated with community 8 of the classification (Appendix B). Disturbance is also likely to be an important factor in these communities; the poor stratification and sand layers in our soil cores reveal a recent (<1,000 years) history of frequent, large flood disturbances from Bullock Creek. The few and highly discrete patches of Baumea suggest that this species may be a geologically recent arrival, as is Sphagnum, given its shallower peat depths at Bullock Creek than other Sphagnum wetlands (Clymo 1984). These characteristics are very similar to the minerotrophic, low-cation fen peatlands of North Carolina described by Walbridge (1991) and Bridgham and Richardson (1993).

As our study site is unique, being the only known polje wetland in New Zealand, use of the classical reference system approach to establish functional equivalency (Zedler 2000) has not been useful for setting restoration targets, because of the lack of equivalent systems to compare against. Instead, the current management plan for the site has specified more practical targets, such as reduction of impact of alien species. In this paper, we have followed the reduction in cover of alien species close to drains after re-wetting; whether desirable species show a corresponding increase in cover in these areas requires further monitoring. Rather than trying to recover some entirely pre-human condition, other wetland values are also being considered. For example, our use of earth dams rather than in-filling of drains has resulted in unnatural lentic ponding of water that, in addition to raising water tables, is also providing habitat for waterfowl at the site. The relatively limited impact of past agricultural practices at the site and the rapid recovery of desired soil properties in our experiment supports the concept of re-establishing natural character by principles of self-design—i.e. minimal intervention and allowing natural processes time to organise species composition (Mitsch et al. 1998). Success using this approach is likely to be enhanced by the location within a largely pristine forest catchment, and the hydrologically unmodified river and karst system. Further research into the hydrological and nutrient responses is necessary to fine-tune ecological requirements to provide enough flooding to discourage alien species, but not to stress the dominant sedges, and to allow them to re-establish natural fen communities and habitat.