The effects of drainage and restoration of pine mires on habitat structure, vegetation and ants

Habitat loss and degradation are the main threats to biodiversity worldwide. For example, nearly 80% of peatlands in southern Finland have been drained. There is thus a need to safeguard the remaining pristine mires and to restore degraded ones. Ants play a pivotal role in many ecosystems and like many keystone plant species, shape ecosystem conditions for other biota. The effects of mire restoration and subsequent vegetation succession on ants, however, are poorly understood. We inventoried tree stands, vegetation, water-table level

peatlands have already been restored and about 18 000 ha still need to be restored to safeguard the ecological value of the protected areas (Similä et al. 2014).
The effects of peatland restoration on the biota need further investigation, especially on habitat specialists (Rydin and Jeglum 2006). Ants provide valuable information when evaluating land-management actions and assessing long-term ecosystem changes because they are sensitive to changes in habitat quality and play a pivotal role in many ecosystems (Andersen and Majer 2004;Underwood and Fisher 2006;Fagan et al. 2010). They, like many keystone plant species, shape ecosystems for other organisms. The activities of ants and their nest and trail constructions may affect other biota considerably (Hölldobler and Wilson 1990;Punttila and Kilpeläinen 2009;Finér et al. 2013, and references therein).
Mires comprise the primary or secondary habitat for at least a third of the 55 native ant species of Finland . In boreal areas, a few studies have shown that drainage of mires affects mire-ant species negatively (Krogerus 1960;Collingwood 1963;Vepsäläinen et al. 2000;Punttila and Kilpeläinen 2009), but knowledge on mire ant communities and their assembly processes are scarce globally (Sveum 1978;Vepsäläinen et al. 2000;Dlussky 2001;Ellison et al. 2002;Gotelli and Ellison 2002a;Mabelis and Chardon 2005;Ratchford et al. 2005;Sanders et al. 2007;Bujan et al. 2010). The effects of mire restoration and subsequent vegetation succession on ants and other insect fauna are poorly understood (Laiho et al. 2001;van Duinen et al. 2003;Watts et al. 2008;Elo et al. 2015;Noreika et al. 2015).
Our primary aim is to characterize differences in mire habitats, ant assemblages and the occurrence of individual ant species among pristine, drained and recently restored mires. We focused on three questions in the short term: (1) What are the most essential differences in vegetation between pristine and drained mires, and how do restoration affect these?; (2) Does the ant-assemblage structure differ between pristine and drained mires and how does restoration affect ant assemblages?; (3) How are the ant species distributed among pristine, drained and restored mires, and what are the most important mire characteristics affecting the ants? Further, our study serves as a baseline for future monitoring of the studied mires when longer-term restoration success is evaluated.
Generally, we test if restoring the water-table level by ditch filling and reconstructing naturally sparse and low pine stands by heavy thinning and partial clear-cutting, will restore habitats to allow recolonization and recovery of the vegetation and ants of pristine mire habitats.

Study design
Metsähallitus Parks & Wildlife Finland manages the Finnish conservation area network and aims to restore mire ecosystems that have been drained for forestry prior to the establishment of conservation areas. Restoration success is also monitored (Aapala et al. 2012). We utilized a sampling-plot network established to monitor restoration success of drained pine mires with the goal of returning the drained mires to their natural state.
The nine study mires (Table 1) are located in two regions in Finland, Northern Karelia and Central Finland, along the border between the southern and middle boreal forest vegetation zones (Ahti et al. 1968), mostly in the eccentric bog zone (Sphagnum fuscum raised bogs) and partly (Kiemanneva and Väljänneva in Table 1) in the southern aapa mire zone (Ruuhijärvi 1983). The mires are mainly ombrotrophic, though oligotrophy is also wide-ranging and some mire parts are mesotrophic. All mires belong partly to the Natura 2000 network of nature protection areas but parts of the mires had been drained in 1960s-1970s prior to the establishment of nature conservation areas (Uusitalo et al. 2006). Restoration of parts of all the studied mires started in 2003-2005 by filling the ditches and by harvesting a varying proportion of trees in [2003][2004][2005][2006]. During harvesting, timber and pulpwood were removed, but all tree stumps and varying amounts of logging residues (branches and tree tops) were left behind.

Vegetation, drainage and ant data collection
At each sampling location, the mire-site type was recorded prior to restoration measures in 2003 (Table 1), and several sets of data were collected: (1) tree-stand characteristics were recorded from a circular 100 m 2 tree-sampling plot, in the middle of which (2) tree-sapling and microsite-type data were recorded from a 25 m 2 sapling square. In each corner of the 25 m 2 sapling square, (3) the covers of vascular plant, moss and lichen species, and litter and surface water were estimated within 1 m 2 vegetation squares (four squares per sapling square). The coverages were estimated on a scale 0. 1, 0.5, 1, 3, 5, 7, 10, 15, 20, 30, 40, 50, 60, 70, 80, 85, 90, 95, 97 Three circular tree-sampling plots (radius 5.64 m = 100 m 2 ) were evenly placed along each transect, approximately 62.5 m apart. From each tree-sampling plot, tree (height > 150 cm) species and their stem numbers were recorded. The trees were classified into size classes according to their diameter at breast height (DBH, d 1.3 m classes < 7, 7-20 and > 20 cm) and height (classes 1.5-3, 3-8 and > 8 m). Dead standing trees (snags) and fallen dead trees (logs) were recorded in a similar way as the living trees (at 1.3 m DBH from the ground for snags and 1.3 m from the butt end for logs).
Tree-sapling and bush (height 50-150 cm) species and their numbers were recorded from the sapling square (25 m 2 ) within the tree-sampling plot. Within the sapling square the percentage covers of mire surface topography types, i.e. microsite types (hummock, lawn and flark) were estimated to the nearest 10 percent.
Based on a-priori knowledge of the general importance of different environmental characteristics on the occurrence of ants, we selected a set of potential explanatory variables in the analyses on ants. Variables selected relate to the degree of shading of canopy layers of different height and vegetation type, to the occurrence of potential nesting sites, and to the occurrence of potential prey and aphid colonies. Also, the most important predictor variables indicating successful restoration were included. The variables used included (1) treatment (a three-level factor; pristine, drained, restored), and several variables from (2) the 100 m 2 tree-sampling plots (number of low trees [1.5-3 m], tall trees [> 3m] and dead trees), (3) the 25 m 2 sapling squares (number of tree saplings and proportions of the microsite types hummock, lawn and flark), (4) the 1 m 2 vegetation squares (percentage covers of surface water and litter, the pooled cover of Sphagnum spp., the pooled cover of other mosses, the pooled cover of herbs, sedges and grasses (i.e. non-woody annual plants), the pooled cover of low (< 20 cm) dwarf shrubs, and the pooled cover of tall (> 20 cm) dwarf shrubs and shrubs (for details, see below); average cover of the four 1 m 2 vegetation squares was used) and (5) water-table depth (average of the two wells per sapling square).
Ants were sampled using pitfall traps. One trap (a 1 dl plastic jar with inner diameter of 56 mm and a depth of 70 mm) was set in a representative location within each 100 m 2 tree-sampling plot, which yielded 6 traps per treatment per mire, i.e. 162 traps in total. Trapping was continuous for six weeks, with catches collected and traps reset every second week, between May and July 2007. To remove the surface tension and preserve the arthropods caught, 5 cl of a 10% NaCl solution with a few drops of detergent was added to the traps. Ants were identified with the keys of Collingwood (1979), Seifert (2000;2007) and Czechowski et al. (2002). On the basis of published data (Krogerus 1960;Vepsäläinen et al. 2000;Punttila and Kilpeläinen 2009) and our own field experience, all the frequently collected ant species were ranked according to their affinity for mire habitats, response to drainage and affinity for pine heath forests on a scale from 1 (strongest mire affinity) to 6 (strongest pine-forest affinity). Additionally, when analysing the effects of drainage and restoration on the number of mire-ant species, we included only those species assessed in the latest red-list work of Finland by the expert group for Hymenoptera (Rassi et al. 2010) to live primarily in mires: Formica exsecta Nylander, 1846, F. picea Nylander, 1846, F. uralensis Ruzsky, 1895 and Myrmica scabrinodis Nylander, 1846 in our data.

Statistical analyses
First, we used log-likelihood ratio tests and Kruskal-Wallis non-parametric median tests to explore variation in the tree-stand variables and other variables characterizing mire habitats, and in ant occurrence among the treatments. To control for false discovery rate in multiple testing, the original p-values were adjusted with the method in Benjamini and Yekutieli (2001) using the function p.adjust (method = "BY") in the package stats in R version 3.0.1 (R Core Team 2013). Additionally, we analysed ant occurrence among different mire types with Kruskal-Wallis non-parametric median tests to increase our limited knowledge of habitat associations of mire ants. We excluded all sampling locations on restored mires and all mire types with less than four sampling locations, which resulted in 89 sampling locations in these analyses.
Second, we applied non-metric multidimensional scaling (NMDS) to characterize differences in tree stands, floristic composition and ant-assemblage composition among the treatments (pristine, drained and restored). The three NMDS ordinations were performed using the vegan community ecology package, version 2.0-8 (Oksanen et al. 2013) in R (R Core Team 2013). We used the vegan function metaMDS with monoMDS to produce two-dimensional NMDS ordinations with several hundred random starts to find stable solutions. The ordination dimensions ("axes") were scaled to half-change units. We fitted a set of environmental variables (see above) into ordinations using the vegan function envfit and tested their fit with permutation tests.
For the NMDS of the tree-stand characteristics, the Gower dissimilarity measure was used; the data were the number of sapling and tree individuals per species, the number of sapling and tree species, the number of trees in three diameter and three height classes, the number of snags in two diameter classes (< 7 cm, 7-20 cm), and the number of logs in three diameter classes in the 100 m 2 tree-sampling plots and 25 m 2 sapling squares. We excluded variables with less than 8 positive values out of 162 data entries.
For the NMDS of the floristic data, the Bray-Curtis dissimilarity measure, square root transformation and Wisconsin double standardization were used.
For the NMDS of the ant data, the Raup dissimilarity measure with ant presence-absence data in the pitfall traps was used, because conspecific individuals of social insects captured in a pitfall trap are not statistically independent units and do not directly reflect assemblage structure (Melbourne 1999;Vepsäläinen et al. 2000;Gotelli et al. 2011;Higgins and Lindgren 2012). Only worker-ant data were included in the NMDS and other multivariate analyses because only workers indicate established colonies.
Third, we used generalized linear mixed models (GLMM, glmer function in the lme4 package, Bates et al. 2013) in R (R Core Team 2013) to evaluate the effects of treatment and a set of environmental variables on the occurrence of ant species. Ant presence-absence data were modeled following a binomial error distribution. Mires nested within study regions were included as a random factor. Nine species occurring in > 10% of the traps were analysed individually.
GLMM analyses were subject to model selection, but treatment, Sphagnum cover, hummock cover and number of tall trees (> 3 m) not, owing to their a-priori expected importance for the individual species analysed. The following predictor variables were, however, subject to model selection: number of dead trees (logs and snags pooled), litter cover, covers of low (< 20 cm) and tall (> 20 cm) dwarf shrubs and the pooled cover of herbs, sedges and grasses. These variables were removed, one at a time, if their p-values exceeded 0.1 and if AIC values (see Burnham and Anderson 2002) decreased after excluding the particular variable. All predictor variables (except treatment) were standardized to zero mean and unit variance (Schielzeth 2010) to make them comparable. Mires from which no individuals of a particular species were caught were not included in the GLMM analyses since they may be unsuitable for that species for reasons that are not considered in this study.

Mire-site types
Prior to starting restoration in 2003, 21.6% of the 162 sample locations represented transformed drained mires where the identification of the original mire type was impossible owing to successional changes. For the 34.0% and 43.8% of the sample locations established in pristine or transforming mire types, respectively (Table 1), pine mires and pine bogs were dominating (57.4% of all 162 locations), and pine fens and rich pine fens made up 18.5%. The most common pristine types were Sphagnum fuscum bogs (RaR, 13.0% of all sample locations, Table 1) and low-sedge pine fens (LkR, 9.9%), and the most common transforming types were transforming dwarf shrub pine bogs (muIR, 25.9%) and transforming Sphagnum fuscum bogs (muRaR, 10.5%).

The effects of drainage and restoration on tree stand and vegetation
Drainage led to a four-fold increase in the number of tree stems relative to pristine mires, and the increase in birch reduced the dominance of pines and led to mixed pine-birch stands in the drained mires (Table 2). Tree growth and thus growing stock increased, as reflected especially by the greater abundance of larger trees (d 1.3 > 7 cm or h > 3 m) in drained than in pristine mires. Restoration harvesting successfully converted stand structure (in terms of stem number, treesize distribution and tree-species composition) closer to pristine conditions (Table 2). Similarly, there were more birch saplings in the drained than pristine mires, whereas restored mires had intermediate numbers ( Table 2). The amount of dead trees was generally low and rather similar among the treatments, but the amount of small-sized logs was higher in drained and restored than pristine mires ( Table 2).
The microsite-type distribution of drained and pristine mires differed considerably from one another: the surface of drained mires was almost completely (97%, Table 2) covered by hummock, whereas more than half of the surface of pristine mires was covered by lawn (38%) and flark (16%). The microsite-type distribution in restored mires resembled that of pristine mires, but the share of hummock was larger (67%) than in pristine mires (46%). Table 2. Mean (x̄ ± SE) tree-stand and sapling characteristics, microsite-type coverage, mire-surface coverage and water-table levels of the sampling locations in pristine, drained and restored mires. Differences in variables among the treatments were tested with Kruskal-Wallis rank sum test (H, df = 2; treatments not sharing the same letter differed significantly according to a posteriori test with critical α = 0.05). We adjusted the original p-values (p adj. in the Table) to control false discovery rate in multiple testing using the method in Benjamini and Yekutieli (2001). h = height, d 1.3 = diameter at breast height (1.3 m from the mire-surface level or 1.3 m from the butt end of logs), and N = number of sampling locations. Data on living and dead trees are from the 100 m 2 tree-sampling plots, data on tree saplings and microsite types are from the 25 m 2 sapling squares, data on mire-surface coverage are from the 1 m 2 vegetation squares, and data on water-table depth are from the ground-water wells.   Similarly, the surface cover of drained mires was dramatically different from pristine mires: the moss layer was dominated by Sphagnum mosses (90%, Table 2) in pristine mires, in drained mires 31%, with a higher cover of other mosses (38%). Differences in the herb layer were clear: the pooled cover of herbs, sedges and grasses was lower and the cover of tall dwarf shrubs higher in drained than pristine mires (Table 2). Pristine mires had higher cover of Sphagnum mosses but lower cover of other mosses, litter and tall dwarf shrubs than restored mires ( Table 2).
The water-table level was lower in the drained (at a depth of 38 cm, Table 2) than in the pristine (15 cm) and restored (16 cm) mires.
Tree-stand and sapling characteristics of the drained mires differed clearly from those of the pristine and restored mires in the NMDS ordination (treatment, r 2 = 0.205, p < 0.001): sampling locations of the drained mires were to the right in the ordination and displayed large scatter, whereas locations of the pristine and restored mires were to the left with small scatter (Fig. 1 A). Drained mires were characterized by a birch mixture and an abundance of large trees, but pristine and restored mires by pines and smaller trees (Fig. 1 B). The scatter of the restored sampling locations, however, was larger than that of the pristine ones ( Fig. 1 A). Five of the 15 a-priori selected and fitted environmental variables showed significant correlations (p < 0.001) with the NMDSordination space: the number of tall trees, water-table depth and the cover of hummock microsite type had higher values in the drained mires than elsewhere ( Fig. 1 A). The values of pooled cover of Sphagnum mosses and those of herbs, sedges and grasses were higher in the pristine and restored mires than in the drained ones ( Fig. 1 A).
The floristic composition of the drained mires differed clearly from both the pristine and restored mires in the NMDS ordination (treatment, r 2 = 0.352, p < 0.001, Fig. 2 A). Sampling locations of the drained mires were mostly to the right in the ordination, whereas those of the pristine mires were mostly to the left, and those of the restored mires mostly in between (Fig. 2 A). Sampling locations to the right, the drained mires and many of the restored ones had higher cover and occurrence rate of forest species such as Vaccinium myrtillus, V. vitis-idaea, Pleurozium schreberi (Willd. ex Brid.) Mitt. and Dicranum polysetum Sw. ex anon., whereas species characterizing pristine mires such as Carex limosa L., C. pauciflora Lightf., Drosera rotundifolia L., Sphagnum balticum (Russow) C.E.O.Jensen, S. fallax (H.Klinggr.) H.Klinggr., S. fuscum (Schimp.) H.Klinggr., S. papillosum Lindb., S. rubellum Wilson and V. microcarpum were to the left in ordination (Fig. 2 B, Appendix 1, available at http:// dx.doi.org/10.14214/sf.1462). Restoration seemed to have increased the covers and occurrence rates of mire species such as Sphagnum russowii Warnst., S. fallax and Polytrichum commune Hedw. in the short term as seen by their locations in the top half of the ordination (Fig. 2 B, Appendix 1).
A number of species located in the top half of the ordination have benefitted from recent disturbances during restoration, e.g. Carex globularis L., Polytrichum commune, Sphagnum squarrosum Crome and Straminergon stramineum (Dicks. ex Brid.) Hedenäs (Fig. 2 B, Appendix 1). Additionally, the cover of especially Eriophorum vaginatum L. seemed to have increased considerably in restored mires relative to both drained and pristine mires although there were no great differences in its occurrence rate among the treatments (Appendix 1).
The same five a-priori selected environmental variables as in the tree-stand NMDS showed significant correlations (p < 0.001) here and correlated similarly with the vegetation ordination space as above, but also five additional variables correlated significantly with the ordination (Fig. 2 A): low dwarf shrubs and litter tended to increase in cover towards the drained mires, other mosses than Sphagnum spp. tended to increase in cover towards the drained and restored mires, and the cover of the microsite type flark increased towards the pristine mires and that of lawn towards both pristine and restored mires. Generally, the ordination revealed a gradient from more wet and fertile conditions characterized by a high cover of lawn and flark microsites to drier and poorer conditions with a high cover of hummock microsites (Fig. 2 A).   Table 3. Mean (x̄, range 0-1) and total (fr) occurrence rates of ant workers and queens, and the mean number (x̄ ± SE, standard error given in column fr) of mire ant species and all ant species in the sampling locations representing different mire types in pristine and drained mires (mire types with less than four sampling locations in pristine and drained mires, and all sampling locations in restored mires were omitted). In the table, mire types within both pristine and the transforming and transformed mires are ordered by increasing growing stock from left to right (mire types to the left tend to have lower growing stock than types to the right). Differences in the occurrence rate (fr) of ant species among the mire types were tested with log-likelihood ratio test (G 2 , df = 6), and differences in the numbers of species were tested with Kruskal-Wallis rank sum test (H, df = 6; a posteriori test was unable to locate differences among mire types at α = 0.05 for the variable "number of mire ant species"). We adjusted the original p-values (p adj. in the Table) to control false discovery rate in multiple testing using the method in Benjamini and Yekutieli (2001). N = number of sampling locations (total N = 89).

Ant occurrence among mire-site types
The ant data comprised 20 species: 10 579 workers of 17 species, 169 queens of 13 species and two males of one species ( The species ranked a-priori to have the strongest pine-forest affinities, C. herculeanus and M. ruginodis, were more frequent in transforming and transformed than in pristine mire types (worker data, Table 3). Species ranked to have the strongest mire affinities showed more variable pattern: F. picea occurred almost exclusively in pristine mire types, whereas F. uralensis and M. scabrinodis were found frequently in both pristine and transforming and transformed mire types ( Table 3). The rest of the species seemed to occur more evenly among the mire types (Table 3). Queens of the forest species M. ruginodis were more frequent in the transforming and transformed than in the pristine mire types, whereas queens of M. scabrinodis showed the opposite pattern (Table 3). The number of all ant species did not differ among mire types, whereas the number of mire ant species was higher in pristine mire types than in transforming and transformed types, and seemed to decrease with increasing growing stock (Table 3). Table 4. Mean (x̄, range 0-1) and total (fr) occurrence rates of ant workers and queens, and the mean number (x̄ ± SE, standard error given in the column fr) of mire ant species and all ant species in pristine, drained and restored mires. Differences in the occurrence rate (fr) of ant species among the treatments were tested with log-likelihood ratio test (G 2 , df = 2), and differences in the number of species were tested with Kruskal-Wallis rank sum test (H, df = 2; treatments not sharing the same letter differed significantly from each other according to a posteriori test with critical α = 0.05). We adjusted the original p-values (p adj. in the Table) to control false discovery rate in multiple testing using the method in Benjamini and Yekutieli (2001). N = number of sampling locations (total N = 162).

The effects of drainage and restoration on ants
The mean number of mire-ant species was highest in pristine mires, but the number of all ant species was highest in drained mires (Table 4). The mire species Formica picea occurred almost exclusively in the pristine mires, and the queens of another mire species, Myrmica scabrinodis, were most frequent in pristine mires, but this pattern was not observed for its workers (Table 4).
The workers of the forest species Camponotus herculeanus and M. ruginodis -and queens of the latter -were most frequent in drained mires and common also in restored mires. Occurrence rate peaked in restored mires only for C. herculeanus queens ( Table 4). The occurrence rates of the rest of the species did not differ statistically significantly among the treatments (Table 4).
Of the nine species analysed individually with GLMM, three were ranked a priori to have the strongest mire affinities (rank 1: F. picea, F. uralensis and M. scabrinodis), two were ranked a priori to have the strongest pine-forest affinities (rank 6: M. ruginodis and C. herculeanus), and the remaining four species to be more generalists in their habitat affinities (rank 2: L. platythorax, F. sanguinea and rank 3: L. acervorum, species that occur both in mires and forests but disappear with tree-canopy closure during forest succession, and rank 4: M. rubra (Linnaeus, 1758), which is known to tolerate tree-canopy shading). Below, we focus on the overall trends of occurrence of species along the mire-forest continuum, rather than statistically significant differences in the occurrence rate of individual species. Fig. 3. Statistical responses of individual ant species to mire treatment (A = mire specialist species, B = generalist species, C = forest species). * = statistically significant (p < 0.05) responses (see Table 5).
Mire specialist species showed variable responses to treatment (Fig. 3 A, Table 5). Formica picea responded statistically significantly and negatively to restored sites, and the trend for F. uralensis suggests negative association with restored mires. Interestingly, M. scabrinodis and F. uralensis tended to associate with drained sites. The fairly generalist species, L. platythorax responded statistically significantly and positively to restored sites, and the trend for F. sanguinea suggests that it associates with restored sites (Fig. 3 B, Table 5). Myrmica rubra behaved like the forest-associated species, whereas L. acervorum responded more as a mire specialist (Fig. 3 B).
As expected, the two forest-associated species, C. herculeanus and M. ruginodis, occurred statistically significantly more often in drained sites, less in restored sites and least frequently in pristine sites (Fig. 3 C).  Ants responded statistically significantly to a number of variables tested, although the species seemed to differ in sensitivity (Table 5). Only the mire specialist M. scabrinodis responded significantly and positively to Sphagnum moss cover, although there is a suggestive trend of mire specialist species being positively associated with Sphagnum cover and forest species the opposite way (Fig. 4 A). The number of tall trees (> 3 m) seemed to affect the majority of species as expected a priori: the probability of occurrence of mire specialist species tended to decrease and that of forest species to increase with an increase in the number of tall trees (Fig. 4 B). Two species responded statistically significantly to the number of tall trees: as predicted, the mire specialist M. scabrinodis negatively and the forest species C. herculeanus positively. None of the species responded statistically significantly to hummock cover, although the mire specialist F. picea and the generalist L. platythorax appeared to respond positively and the forest species M. ruginodis negatively to hummock cover (Fig. 4 C, Table 5). Additionally, a few other variables remained in the final models, but for only two species the responses were statistically significant: C. herculeanus showed a negative response to dead trees, and M. ruginodis responded positively to low and tall dwarf shrubs and annual plants ( Table 5).
The ant-assemblage structure of the drained mires differed from the pristine mires in the NMDS ordination (treatment, r 2 = 0.221, p < 0.001), but the scatter of the sampling locations was large in each of the three treatments, and the locations of the restored mires overlapped with those of both pristine and drained ones (Fig. 5 A). There seems to be a naturalness gradient from the sampling locations of pristine mires (to the right) towards those of drained mires (to the left), with the locations of restored mires in-between (Fig. 5 A). Seven of the 15 fitted environmental variables correlated significantly (p < 0.001) with the NMDS-ordination space: the cover of Sphagnum mosses and that of the microsite type flark increased towards the pristine locations, whereas the rest of the statistically significant variables were related to drainage and correlated approximately in the opposite direction, towards the drained and restored mires (Fig. 5 A). Ant assemblages differed from each other such that the more forested mires seemed to be located to the left-hand side (correlation with tall trees) and the more open mires with abundant lower trees more to the top of the ordination.
Differences in ant assemblages among the non-pristine mires were expressed such that the shade-tolerant forest species C. herculeanus and M. ruginodis occurred in the more heavily forested sampling locations at the bottom left of the ordination, whereas species of open-canopy forests and mires -such as L. platythorax, Leptothorax acervorum, Formica sanguinea and F. exsectacharacterized the less densely forested sampling locations to the top right (Fig. 5 B). Open pristine mires were characterized by the occurrence of the mire species F. picea and M. scabrinodis, at the bottom right of the ordination (Fig. 5 B). Another mire species, F. uralensis, however, had benefitted from drainage at least temporarily, and is located closer to the forest species C. herculeanus and M. ruginodis in ordination (Fig. 5 B). Indeed, half of the occurrences of F. uralensis were from drained mires (Table 4).

Discussion
In general, our results showed predictable responses in both habitat structure -including vegetation -and ant assemblages to mire drainage and restoration. This conclusion is similar to the results in a mire restoration study dealing with a number of solitary invertebrate groups (Noreika et al. 2015). Taken together, our results are consistent with the Field of Dreams hypothesis (Palmer et al. 1997), which suggests that if you rebuild the habitat structure, the flora and fauna will return; here mire restoration resulted in rapid recovery of mire vegetation and ant assemblages (see also  Noreika et al. 2015). However, for mire-specialist ant species the results were variable and longerterm monitoring is needed to evaluate the success of restoration. Considerable time lag -relative to solitary insects with shorter generation times -is expected in the response of ants, which form long-lived perennial colonies with long colony cycles.

Effects of drainage and restoration on mire-habitat structure
Restoration measures -the filling of ditches, heavy thinning and partial clear-cutting of tree stands -were successful in that both the water-table level and stand structure of restored mires matched those of pristine mires, although heavier cuttings of larger trees would have been required to accord better with pristine mires in terms of tree-canopy shading. The amount of dead wood was generally low in all mires, and the amount of small logs was only slightly higher in drained and restored mires than in pristine ones indicating that tree mortality due to competition had not yet started in drained mires, and that coarser logging residues (felled trees) were not left behind in restored mires in large quantities. We showed that tree stand structure was more variable in the drained than in the pristine or restored areas ( Fig. 1 A), which followed probably both from drainage and forest management activities. In the long run, forest management tends to homogenize stand structure by, e.g. decreasing variation in stem size distribution, whereas forest stands in pristine mires exhibit multi-modal structure (Uuttera et al. 1997).
The pristine microsite-type distribution was re-established a few years after restoration: the mire surface was almost completely covered by the hummock microsite type in drained mires, but was reduced to two thirds in restored mires, whereas it was 46% in pristine mires. Although habitat structure of the restored mires had clearly changed towards pristine conditions in a few years after restoration, the cover of Sphagnum mosses was still lower (similarly as 10 years after restoration in Haapalehto et al. 2011), and the cover of e.g. tall dwarf shrubs was still higher in restored than pristine mires (see Fig. 3 in Laine et al. 1995a).
Drainage seemed to have led to similar successional changes of mire vegetation as in earlier studies -increased occurrence and cover of forest species and the disappearance or decline of mire species (Laine et al. 1995a;Jauhiainen et al. 2002;Haapalehto et al. 2011;Haapalehto 2014). Similarly, a rapid change towards pristine floristic composition in restored mires has been observed in earlier studies. The strength of the response depends on the severity of degradation of the drained mires, and on mire fertility. Rapid responses have been restricted to less heavily degraded sites within restored mires, whereas no or only minor changes have been reported from heavily degraded sites (Komulainen et al. 1999;Jauhiainen et al. 2002;Haapalehto et al. 2011;Haapalehto 2014). In our data, the rapid recovery may have been enhanced by a lack of dispersal barriers, as the restored mires were parts of larger mires with varying proportions of unditched areas. These likely provide refugia for mire species in close proximity of the drained and restored mires, and thus they also serve as potential areas for recolonization of restored mires.
Similarly to earlier studies, some species, e.g. Eriophorum vaginatum, responded strongly by increasing in cover (Komulainen et al. 1999;Jauhiainen et al. 2002;Haapalehto et al. 2011). On the other hand, some bog species of wet hollows, e.g. Carex limosa and Scheuchzeria palustris L., seemed to be missing from restored mires in the short term (similarly as 10 years after restoration in Haapalehto et al. 2011), and in another study these species seemed not to have colonized restored mires even after half a century, despite the existence of nearby populations (Soro et al. 1999). Laine et al. (2011) monitored the effects of restoration on vegetation of oligotrophic pine fens in a more northern area in Kainuu, eastern Finland for two years. They showed that the vegetation of drained mires had not changed much from pristine conditions during 30 years of drainage: only the typical hollow species Sphagnum majus (Russow) C.E.O.Jensen, S. balticum and Scheuchzeria palustris were missing. Laine et al. (2011), however, also found that there were only minor differences in the floristic composition before vs. after restoration in the two-year monitoring period: the only statistically significant restoration-related change was that the cover of sedges increased in restored and decreased in pristine fens.

Ant occurrence in mire-site types
Knowledge on the ant fauna of mire types is limited (Oinonen 1956;Krogerus 1960). We showed the importance of tree cover on the occurrence of ant species among mire types: the occurrence rates of mire species tended to decrease with increasing growing stock, whereas forest species showed the opposite pattern (Table 3).
A NMDS analysis of the data of Krogerus (1960) revealed a clear gradient in the composition of ant assemblages from open fens and rich fens through pine mires and pine bogs to spruce mires . Dlussky (2001) found a similar openness gradient within an oligotrophic bog in the Novgorod district, Russia. The ant assemblages of the open central parts of the bog were characterized by true ("endemic") mire species (F. picea, F. uralensis and F. forsslundi Lohmander, 1949) and some "forest species" (L. platythorax, M. rubra, M. scabrinodis and L. acervorum), whereas pine forests of the mire edges were occupied by forest species (C. herculeanus, L. platythorax, M. rubra and F. polyctena Förster, 1850). This openness gradient resembles our NMDS-ordination results (Fig. 5 B), which showed that assemblages at the open end of the gradient were characterized by the mire species F. picea, M. scabrinodis and F. uralensis. On the other hand, the assemblages in sparsely forested areas with low trees were distinguished by species that colonize recently disturbed or naturally open forest areas (e.g. L. platythorax, L. acervorum, F. sanguinea, F. exsecta). Finally, the more heavily forested areas with tall trees were characterized by forest species that are capable of persisting through the whole cycle of forest succession, i.e. C. herculeanus and M. ruginodis (Niemelä et al. 1996).

Drainage and ants
The succession of tree stands on drained mires with increased growing stock of large trees and increased deciduous mixture of birch with pine result in shaded conditions. This change decimated colonies of mire-ant species requiring sun exposure, which prevails in pristine conditions of pine mires whereas shade-tolerant forest species, such as Camponotus herculeanus and Myrmica ruginodis, benefitted similarly as in earliers studies and thus, the ant assemblages of drained mires differed clearly from those of pristine mires (Krogerus 1960;Collingwood 1963;Vepsäläinen et al. 2000;Punttila and Kilpeläinen 2009). The mire specialist Formica picea in particular suffered from drainage and this is also expected in the long term for F. uralensis (which seems to benefit from the early stages of drainage) and M. scabrinodis (Krogerus 1960;Vepsäläinen et al. 2000). Krogerus (1960) found that F. picea inhabited the moistest parts of pine mires but F. uralensis the driest sites. Punttila and Kilpeläinen (2009) found that the occurrence rates of the three moundbuilding species restricted to mires in their data are highest in undrained pristine mires, lower in transforming mires and lowest in transformed ones (F. uralensis), or the species are missing from transformed mires (F. fennica Seifert, 2000 and F. forsslundi), whereas the most common species of forests on mineral soils, the wood ant F. aquilonia Yarrow, 1955, showed the opposite pattern. Similar results have been obtained in southern Sweden where the slow drying-out of bog areas has decimated the prevalence of the mire species F. uralensis, F. forsslundi and F. picea in only one decade (Collingwood 1999).
In our data, ant species of drier open habitats (e.g. open early successional forests after disturbances) and generalist ant species benefitted from the early phases of mire drainage. Most of these species disappeared with tree-canopy closure also during forest succession on mineral soils (Punttila et al. 1991), but in the early stages of drainage they seem to replace mire species. For instance, M. ruginodis replaced M. scabrinodis (Vepsäläinen et al. 2000) although M. scabrinodis appeared to be more resilient than other mire species (see also Collingwood 1999), and especially Lasius platythorax (Collingwood 1999;Dlussky 2001) but possibly also Formica fusca Linnaeus, 1758 andF. lemani Bondroit, 1917 seemed to replace F. picea, and in the later successional stages, F. aquilonia replaced territorial mound-building mire species (Vepsäläinen et al. 2000;Punttila and Kilpeläinen 2009). Open treeless mires or sparsely forested, drained open mires do not necessarily provide any habitat for wood ant species of the Formica rufa group (F. aquilonia, F. lugubris Zetterstedt, 1838, F. rufa Linnaeus, 1758, F. polyctena and F. pratensis Retzius, 1783 because arboreal aphid colonies providing nourishment for wood ants are scarce, and the foraging activity of wood ants may cease for long periods of time during high day-time temperatures in open mires and, finally, wood ants cannot find suitable below-ground overwintering sites deep enough because of the high water-table level (Vepsäläinen et al. 2000). In wooded and drained mires the situation is different which enables colonization of wood ants: when mire drainage proceeds, territorial wood-ant species of mineral soils also start to dominate transformed mires (Punttila and Kilpeläinen 2009). In our data, the wood ants F. lugubris and F. pratensis were missing from pristine mires, but these species occurred in seven (15%) and two (4%) sampling locations in drained, and in three (5%) and one (2%) locations in restored mires, respectively (Appendix 2).

Restoration and ants
We are not aware of any other study on the effects of mire restoration on ants, but with the results of the present study and the results of earlier studies concerning drained mires, together with information about the autecology, competitive capacity and dispersal of different ant species, we can outline some general principles to be taken into account when planning mire restoration.
The rewetting of mires and removal or strong thinning of the growing tree stock are key components in successful restoration of mire ants. A high water-table level prevents the overwintering of top-competitors of the territorial Formica ants of heath forests in mire habitats and thus leaves the terrain free for typical mire-ant assemblages to develop with their own top-competitors of other territorial Formica ants, such as F. uralensis, F. exsecta and F. forsslundi (see Punttila and Kilpeläinen 2009).
Another important issue in mire restoration for ants is the treatment of logging waste, i.e. tree stems, tops, branches and cut stumps after logging in naturally treeless or sparsely forested mires: whether these are left behind or harvested .
Despite the fact that the amount of dead trees in our study was generally low in all treatments, the amount of small logs was somewhat higher in drained and restored mires than in pristine ones, thus providing more nesting sites for forest ants there. Most of the boreal forest ant species can nest in or even require dead wood for establishing new colonies, and for many species in many habitat types a large share of ant nests are located in dead wood (e.g. Oinonen and Wuorenrinne 1976;Franch and Espadaler 1988;Włodarczyk et al. 2009;Persson et al. 2013). Thus, leaving stumps, felled trees or logging residues in restored mires may enhance colonization of non-desired forest species in the restored mire habitats. Dead wood provides elevated microhabitats that may offer the only sun-exposed and warm above-ground nesting sites for forest ants. Such conditions are found in habitats where dense, shading herb-layer vegetation prevents sun exposure of the ground. In forested habitats, this may happen shortly after the crea-tion of tree-canopy openings by wind throws, fire or cuttings, but this may also take place after disturbances -e.g. ditching or restoration -in mires.
Although the rapidly recovering moss layer of the restored mires may overgrow logging residues and cut stumps within a few years following restoration, they nevertheless provide luxuriant nesting substrates for a large variety of forest and generalist ant species. These newcomers may later prevent or slow down the colony establishment of mire specialist ant species by predation pressure on the dispersing mire-ant queens attempting colony founding, and through nest-site competition and pre-emption of nesting sites -such priority effects are exemplified, e.g. in the colonization by ants of small land uplift islands of the Baltic Sea (see Vepsäläinen and Pisarski 1982). These types of nesting substrates are missing almost entirely from pristine open mires and thus, their creation should be minimized in the restoration process. Ant species of open successional forests may also be capable of replacing the few remaining but weakened mire ant populations of restored mires. These forest species are also able to inhabit logging residues in the middle of larger restored mires, as the dispersing females of some of these species are rather strong fliers (Vepsäläinen and Pisarski 1982;Czechowski et al. 2013). The colonization of many aggressive and territorial Formica species is dependent on the occurrence of species in whose nests they start new colonies by temporary parasitism (Gösswald 1951;Kutter 1969;Wilson 1971;Collingwood 1979;Seifert 2000;2007;Czechowski et al. 2012). These species can often utilize both mire specialist species remaining in the area, and generalist species of mineral soils, which have colonized the drained mire. Thus, their colonization may be enhanced in areas where both mire and forest species occur together (Collingwood 1979;Seifert 2007). Another option for logging-residue harvesting is to girdle the trees and let them die standing, which reduces the amount of potential dead-wood nesting sites for forest ant species by spreading the increase of logs on the ground over a longer time span.
In restoration planning, the surrounding terrestrial environment should also be taken into account to inhibit possible colonization of forest ants in restored mires. Restoration efforts can be timed to years when there are no larger regeneration cuttings or young sapling stands in close proximity to the restored areas. For instance, in southern Sweden, the eradication of Formica picea from slowly drying bogs was accelerated by colonization pressure and subsequent competitive exclusion by Lasius platythorax, which immigrated from the surrounding fresh thinning cuttings performed a few years earlier. In the cut area, the forest floor was covered by logging residues, fallen trees and sun-exposed cut stumps. Nearly all the cut stumps were occupied by flourishing L. platythorax colonies, and these ants were also seen in large quantities in bog tussocks in the mires (Collingwood 1999).
Our knowledge on the dispersal capacity and competitive ability of mire ants is scanty (but see Collingwood 1979;Vepsäläinen et al. 2000). Thus, it is difficult to predict the recolonization of restored mires by mire ants. Apparently dispersal among different habitat patches by winged females is rather limited at least in many polygynous ant species and populations (Sundström et al. 2005;Seppä 2008), but generally the dispersal capacity seems to vary widely from very strong dispersers (e.g. L. niger, C. herculeanus) to seemingly poor ones (see Vepsäläinen and Pisarski 1982). There is evidence that the dispersal of the mire ant Formica picea is weak, not only between different mires but also within mires (Mabelis and Chardon 2005;Rees et al. 2010). In our data, this species seemed to be almost exclusively confined to pristine mires. The recolonization of F. picea in restored mires probably depends heavily on the closeness of healthy populations. This may be critical for restoration because successful colonization of the rewetted mires by F. picea is a prerequisite for the subsequent colonization of the other true mire species, F. forsslundi and F. uralensis, which establish their new colonies through temporary parasitism in the nests of F. picea.

Habitat disturbance and ant colonization
Habitat disturbance often opens up new colonization possibilities for boreal ants in forested areas. Disturbances create favourable climatic conditions for species that require warm and well-lit conditions at least during the colonization stage. Large disturbances in mature forests also open competition-free terrain by often wiping out the territorial wood-ant colonies, whereas these colonies seem to tolerate smaller disturbances (Punttila et al. 1991;Punttila et al. 1994;Punttila 1996;Sorvari and Hakkarainen 2005;2007;Kilpeläinen et al. 2008;Punttila and Kilpeläinen 2009;Sorvari et al. 2011). In our data > 60% of Camponotus herculeanus queens were found in the samples of restored mires. Restored mires, however, should be inferior habitat for this wood-inhabiting species, because most of the trees have been removed from there, and the cut stumps are probably rather quickly buried under the growing mire mosses. Such colonization attempts may, however, be explained by possible attractiveness of recently disturbed open areas following restoration measures. Flying Camponotus queens might orientate towards light reflecting from disturbed habitat patches similarly as many other ant species do (Brian 1952;Pontin 1960;Brian et al. 1966;Wilson and Hunt 1966;Fowler 1987;Tschinkel 1987). In forested areas, recently disturbed areas devoid of territorial wood ants (Formica rufa group) are presumably optimal habitats for colony founding, and high queen numbers have indeed been observed in fresh clear-cuts following cutting and prescribed burning (Punttila et al. 1991;. Also Lasius platythorax probably benefitted from recent disturbances in restored mires as suggested by our data. Disturbed forested boreal areas are quickly colonized by many ant species requiring open conditions in at least the colony-founding stage, and thus the habitat becomes crowded such that the only options remaining for colony founding are nest parasitism and -for queens in established colonies -colony budding (Punttila et al. 1991). Such crowded conditions, however, seem not to be reached often in open pristine mires although these are long-lived habitats. Our own experience is that open pristine mires are often sparsely inhabited by ant colonies. The reason why apparently suitable nesting sites often remain uninhabited is not known, but presumably any larger ant-free terrain can be created by extreme events such as long-term flooding that drowns the ant colonies, or severe winters with thin snow cover lowering the overwintering survival of colonies, or wildfires in exceptionally dry years. The commonness of such events might ultimately determine the nest density of the mire. If this were the case, polygyny and subsequent budding, leading to multinest colonies, would increase the resilience of mire ant colonies in the long run, as the risk of extirpation of an extensive nest network should be smaller than that of a single nest (Mabelis and Chardon 2005).
On the other hand, ant assemblages in drained mires in our data had become more crowded than in pristine mires, as they housed populations of both forest and mire species. Thus, in both pristine and drained mires, nest budding (leading to multinest colonies) and nest parasitism are most likely as important modes of nest-founding as in successional boreal forests. In our data, the high frequency of Myrmica queens -M. scabrinodis in pristine and M. ruginodis in drained mires -indicated the occurrence of polygynous colonies, where the queens are moving or transferred among the nests of multinest colonies (Punttila et al. 1991;Seppä et al. 1995). Our data were, unfortunately, insufficient for making any observations on nest parasites because our sampling window was outside the dispersal period of most species, which is in the late season.

Conclusions
Generally, we tested whether restoring simple but crucial habitat characteristics, the water-table level by ditch filling and the naturally sparse and low pine stand by thinning and clear-cutting, would lead to recolonization and recovery of the characteristic vegetation and ant assemblages of pristine mire habitats. Here we cannot yet evaluate long-term restoration success, but our short-term results are consistent with predictions of the Field of Dreams hypothesis (Palmer et al. 1997), which does not necessarily apply to more heavily degraded mire ecosystems. The recovery of our study mires was fast because the drained mires still had an acrotelm (the surface layer of mire soil) and catotelm (core of peat), and mire organisms could recolonize the restored mires either from local refugia or from populations of nearby mires (see the discussion in Komulainen et al. 1999;Vasander et al. 2003). The connectedness of mire habitats, however, has been severely reduced especially in southern Finland; consequently, recolonization has become unlikely for many mire specialist species, which have become locally threatened (Rassi et al. 2010). Such demanding species in boreal areas may include the workerless social parasite Myrmica karavajevi (Arnoldi, 1930) and the temporary parasite M. vandeli Bondroit, 1920. Both species require dense colonies of their host ant Myrmica scabrinodis for which, especially within the boreal mainland areas of Finland, the most important habitat is mires (Vepsäläinen et al. 2000;Punttila et al. 2013). Degradation of mire habitats and their fragmentation may lead to the disappearance of these highly specialized rare social parasites and other mire specialist species from large areas. In Finland, the threat is most acute in the south, where almost 80% of mires have been drained and the distances between restored mires and pristine mires housing potential source populations are increasing with decreasing connectedness, which weakens the recolonization prospects of specialist species once they have disappeared (Vepsäläinen et al. 2000;Auvinen et al. 2007;Punttila et al. 2013). Atte Komonen of the University of Joensuu for participating in the planning of the study and for organizing many practicalities starting from study-area selection. We thank Liisa Karhu, Kari Lahtinen, Lauri Mikonranta and Tiina Virta for participation in the field work, and Simo Väänänen for identification of the ant samples and participation in the early stages of data processing and literature searches. We also thank Kaisu Aapala, Tuomas Lahti, Krister Karttunen, Tapani Sallantaus, Jari Teeriaho, Seppo Tuominen and Satu Turtiainen of the Finnish Environment Institute for indispensable scientific advice and practical help during the course of our study. Data collecting and ant identification were funded by Metsähallitus Parks & Wildlife Finland, and the studied mires belong to two LIFE Nature projects participated by Metsähallitus (Karelian mires and virgin forests -pearls in the chain of geohistory -2007, and Protection of valuable bird-rich wetlands in Central Finland 2001.