Multidimensional Site Description of Peatlands Drained for Forestry

Sites (n = 119) on drained mires located in the southern aapa mire zone in Finland were analysed by multivariate techniques. The compositional trends of the understorey vegetation were analysed by means of hybrid multidimensional scaling (HMDS). In addition to field classification, two-way indicator species analysis (TWINSPAN) and flexible unweighted paired group arithmetic average (FUPGMA) classifications were used. The 1st HMDS dimension primarily reflected variation along a gradient from spruce mire influence to hummock-level bog influence. Variation in nutrient status was also connected to this gradient. Factors underlying the 2nd dimension were variation in nutrient status and drainage succession (moisture). Some sample plots representing herb-rich or Molinia-rich types were separated along the 3rd dimension. The variation in understorey vegetation (i.e. the ordination space) showed high maximum correlation with stand volume r = 0.81, mean annual stand volume increment r = 0.76, and post-drainage dominant height r = 0.75. The covariation between the vegetation and peat bulk density in both the 0–10 and 10–20 cm peat layers was also strong: r = 0.55 and r = 0.80. The correlations for Hv.Post were 0.64 and 0.81, respectively. Of the total macronutrient concentrations, phosphorus (r = 0.73, r = 0.75) and nitrogen (r = 0.59, r = 0.64) were the most strongly correlated with species composition. The environmental sample variables were also presented by the vegetation units of numerical classification. Most of the recorded variables, including nutrient amounts (kg ha–1), were examined in site quality (fertility) classes by succession phases as well. Border variants or transitional forms of the site types were common. Additional vegetation criteria (e.g. surface-water influence) more closely defined the ecology of the site. In addition to the site quality classes, a considerable amount of information about the tree stands, vegetation diversity and peat properties was associated with the separation of the succession phases, i.e. in this study transforming (phase II) vs. transformed (final phase III) sites. In conclusion, the actual vegetation appeared to well reflect various aspects of the ecological conditions, even in labile communities of commercial forests on drained peatlands.


Introduction
In Finland, the underlying principles used in the classification of drained mires originate from Cajanderʼs (1911,1913,1949) doctrine.As a result, the classification has primarily been based on the understorey vegetation (Keltikangas 1945 and references therein, Lukkala and Kotilainen 1951, Huikari 1952, Sarasto 1961a,b, Huikari et al. 1964, Heikurainen 1986, Laine and Vasander 1990).The post-drainage succession, initiated by the lowering of the water level, has been divided into three phases (Sarasto 1961a,b, Heikurainen andPakarinen 1982).The classification of recently drained mires and transforming drained mires (phases I and II; Finnish abbreviations oj and mu, respectively) is based on the original mire site types.The relatively stable climax plant communities that develop on older drainage areas are traditionally classified into four (to six) transformed site types (phase III; tkg; Lukkala and Kotilainen 1951, Sarasto 1961a,b, Huikari et al. 1964, Heikurainen and Pakarinen 1982).About 60% of the ca 5 million ha of peatland that have been drained for forestry in Finland represent the second phase, and 27% represent the third phase (Metsätilastollinen vuosikirja 2001, cf. Hökkä et al. 2002).
Determination of the original mire site types on drained peatlands is regarded as problematic, especially in the later phases of secondary succession (Laine 1989, Hotanen andNousiainen 1990).The borders between the successional phases are diffuse (Pienimäki 1982, Reinikainen 1988, Laine 1989, Saarinen and Hotanen 2000).According to Laine (1989), the classification system could be made more operational by allocating the drained mire sites already in the successional phases to the corresponding drained peatland forest site types.Seven peatland forest site types have been distinguished by Laine, and adopted for use by e.g.Laine and Vasander (1990) and Paavilainen and Päivänen (1995).
The drainage of pristine mires ceased in Finland during the late 1990ʼs and the ditched area is no longer increasing.As a result, we have to deal with vegetation that has been affected by drainage for an ever-increasing period of time, and which displays additional variation in the peatland environment -caused by the above-mentioned factors.Considering the great variation in the ecological conditions on drained peatlands derived from the numerous mire site types, and additional variation in different climatic areas and variation in silvicultural history, it is clear that our current knowledge of drained mire vegetation and its relation to different external factors is insufficient.
The site classification used in forestry is primarily based on determination of the potential capacity of the site to produce stemwood (Huikari 1952, Heikurainen 1973, 1979).Information about the different factors affecting stand yield capacity is also required -especially from the peatland classification (Reinikainen 1983(Reinikainen , 1984a(Reinikainen , 1989)).This is related to the fact that the variation on peatlands occurs along a larger number of ecological gradients than that in mineral soil forests, and therefore a large set of control measures of the growth factors are needed.
The increasing importance of forest planning has resulted in a need to improve the integration of peatland classification with growth and yield models (e.g.Gustavsen et al. 1998, Hökkä andPenttilä 1999).Developing a system that describes site quality by means of continuous variables instead of categorical ones would permit more efficient use of site quality in growth modelling (Ojansuu 1996, Hökkä 1997, Hökkä and Penttilä 1999).However, the use of site index methods on peatlands is problematic because a wide range of labile variables affect the state of the generally uneven-aged and heterogeneous tree stand (cf.Penttilä 1984, Gustavsen 1996, Hökkä and Penttilä 1999).Determination of the taxatorical importance of different site and vegetation classes requires further examination (Laine and Starr 1979, Hökkä 1994, Gustavsen et al. 1998).
The development of multivariate methods has enabled effective analysis of compositional (vegetational) and ecological gradients in connection with more detailed analyses of the relationships between vegetation and environmental factors on drained mires (e.g.Mannerkoski 1979, Westman 1987, Reinikainen 1988, Nieminen and Pätilä 1990, Laine and Vanha-Majamaa 1992, Eurola et al. 1995, Laine et al. 1995, Vasander et al. 1997, Hotanen et al. 1999).The vegetation gradients are also important from the point of view of the mire site typology used in forestry, because they are connected to the site productivity (Heikurainen 1979, Reinikainen 1989).
Among the factors determining vegetationbased peatland site type classification, the nutrient reserves in the substrate are crucial for peatland forestry (Kaunisto andPaavilainen 1988, Paavilainen andPäivänen 1995).The possibility of using peat and nutrient variables in the classification of drained mires has been discussed (e.g.Vahtera 1955, Kaunisto 1983, Westman 1987, Kaunisto and Paavilainen 1988, Laiho 1994).However, more surveys of the nutrient status and reserves of drainage areas of varying age, different mire site type and phytogeographical region are needed before a consistent classification of this kind is available (Kaunisto and Paavilainen 1988, Laiho and Laine 1994, Westman 1994).As a basis for this, there is an urgent need for confirmation of the ecological background for the classification (Reinikainen 1988, Laine andVanha-Majamaa 1992).
The aim of this study is to describe the site in a multi-dimensional manner, i.e. 1) to analyse the main compositional gradients, 2) to elucidate the relationships between the vegetation and environment, 3) to classify the sites in different ways, 4) to examine the variation in both vegetation and environmental sample variables between (and within) vegetation units in a systematic sample taken from a random area entity, and 5) to com-pile additional information on the average nutrient reserves in the surface peat in drained peatland forests in eastern Finland.This study, primarily an explorative one, is also intended as an ecological background analysis for follow-up studies.

Study Area and Sampling
The study area, comprising the former Nurmes and Lieksa forestry management districts of the Forest and Park Service, lies in the south-eastern part of the southern aapa mire zone in Finland (Fig. 1).The mean altitude in the area is about 180 m a.s.l.(Sevola 1983), annual mean precipitation 550-600 mm, length of the thermal growing period 145-150 days, and effective temperature sum about 1000-1100 d.d.(threshold +5 °C) between eccentric bogs and the southern aapa mire zone (Atlas of Finland 1988).
The permanent sample plots of the study (n = 119) were located in 40 systematically distributed units (tracts) of the 7th National Forest Inventory (Valtakunnan metsien... 1977, Sevola 1983).The distance between individual tracts was 4 km, and between individual plots (three in each tract) was 40 m.The first three numbers in the plot code refer to tract (Table 1).The plots were established in 1981 according to the guidelines of Gustavsen et al. (1988).In 1991 the plots were linked to the SINKA system (Permanent Forest Inventory Growth plots for peatlands; Penttilä and Honkanen 1986).The size of the circular plots was adjusted according to the stand density in order to have at least (about) 35 trees on each plot.
The successional phases were classified in the field according to Sarasto (1961a,b) and Heikurainen and Pakarinen (1982).In 1991, all the plots represented either transforming (phase II; mu) or transformed drained mires (phase III; tkg).If regressive development had occurred in the succession, this was expressed using the symbol r (mur, tkgr; Table 1).The original mire site type in the mu phase was determined according to Laine and Vasander (1990).The tkg phases were allocated to drained peatland forest site types in accordance with Laine and Vasander (1990).Border variants or transitional types (more fertile + / infertile -) of the site types were identified on the basis of the site fertility characteristics of the vegetation.The additional qualifiers, Kn = spruce mire influence, Lu = surface-water influence, Mol = abundant occurrence of Molinia caerulea, Ram = rahka-hummocky (less than 75% coverage of Sphagnum fuscum), and Ri = flark-rich (Huikari 1952, Huikari et al. 1964, Eurola et al. 1984), were used to describe the site in more detail (Table 1).Furthermore, the site quality class (Huikari et al. 1964, Huikari 1974) of the plots was determined according to the actual vegetation.
Information about ditching and fertilization was obtained from the Nurmes and Lieksa offices of the Forest and Park Service.The (improvement) ditching and fertilization years of some of the plots were uncertain and labelled with question marks (Table 1).

Vegetation Analysis
The vegetation descriptions were made in June-August, 1991.The coverage of the field and bottom layer species was estimated in six quadrats (2 m 2 ) located systematically on the plots (Fig. 2) using the scale 0.1, 0.2, 0.5, 1, 2, 3, 5, 7, 10, 15, 20, 25...90, 93, 95, 97, 98, 99 and 100%.If an obstacle (stump, large tree, ditch, ditch spoil) occurred in a quadrat, the quadrat was moved to the nearest normal mire surface and a note of this made on the cover drawing of the plot.Shrubs and tree seedlings less than 0.5 m high were included in the field layer.Other shrubs and tree seedlings up to 1.5 m high (genuine shrub species without an upper limit) were included in the shrub layer and estimated for the whole sample plot.Plotwise mean coverages of the individual plant species were calculated.If a species occurred in both the field and shrub layer, the two coverage values were summed.The nomenclature follows Hämet-Ahti et al. (1998) for vascular plants, Eurola et al. (1994) for bryophytes, and Vitikainen et al. (1997) for lichens.

Recording of Environmental Variables
The tree stands on the plots were measured in 1981, 1986 and 1991 according to the guidelines of Gustavsen et al. (1988) and Penttilä and Honkanen (1986).The basic stand characteristics were calculated using the KPL programme package (Heinonen 1994).A site index, H40dr (Gustavsen 1996), was also calculated.H40dr is the post-drainage dominant height value 40 years after drainage.The post-drainage dominant height is defined as the difference between the current dominant height and the dominant height at the time of drainage.The latter was estimated using Equation 4of Gustavsen (1996).
The peat depth was measured down to a maximum depth of 1.5 m at a point next to each vegetation quadrat.The mean value was calculated for the plot.Peat was systematically sampled from twelve points adjacent to each vegetation quadrat (Fig. 2), using an auger with cutting edge 4.0 × 6.2 cm.Separate samples were collected for the 0-10 and 10-20 cm depth intervals; the latter omitted if the peat was less than 20 cm thick.If a loose raw humus layer, consisting of litter and especially Pleurozium schreberi, was present on the peat surface, it was removed before measuring and sampling the peat layers (Kaunisto and Paavilainen 1988).The sub-samples from the twelve points were combined by layers and kept frozen until analysis.
The peat type was determined macroscopically from the thawed peat samples (Laine and Vasander 1986) (Table 1).The humification degree was assessed by the method of von Post (1922).These determinations were made by the author on anonymised samples and controlled by a second person.
The peat samples were dried at 40 °C to constant weight and weighed.pH was determined from a soil-water suspension (1:2 volume ratio).The peat samples were then dried at 105 °C for dry mass determination and weighed for bulk density calculation.Total nitrogen was analyzed by the Kjeldahl method, and phosphorus spectrophotometrically after dry digestion (550 °C) and extraction with HCl (Halonen et al. 1983).The concentrations of the other macronutrients were determined by inductively couple plasma atomic emission spectrometry (ICP/AES) after wet digestion with HNO 3 /H 2 O 2 .The quality of the analyses was checked by performing parallel determinations on a subset of the samples.The nutrient stores (kg ha -1 ) were calculated for the 0-10 and 10-20 cm peat layers.

Data Analysis
The compositional gradients of the vegetation were identified by hybrid multidimensional scaling (HMDS) using DECODA software (version 2.04; Minchin 1991).HMDS was applied to a matrix of the Bray-Curtis coefficient (Faith et al. 1987).This matrix of dissimilarities between sample plots was calculated from log 10 transformed abundance values of the species.Default options were used throughout.Thus, the dissimilarity threshold for hybrid scaling was 0.8.In HMDS ordination, solutions of 1-4 dimensions were calculated and ten starting configurations were used in each number of dimensions.The HMDS dimensions were scaled to half-change units.All possible pairs of ordination configurations in the same number of dimensions were compared using the method of Procrustean analysis (Schönemann and Carrol 1970).
In HMDS, three dimensions were needed for proper interpretation of the variation in vegetation.The minimum stress values were 1D = 0.272, 2D = 0.149, 3D = 0.109 (4D = 0.091).The minimum stress was effectively achieved from all 10 starts in the three-dimensional solution.In Procrustean analysis these 10 minimum stress configurations were all identical.Interpretation was determined to this solution.
In DECODA, vector fitting option was applied.This option calculates, for each variable in the second set, a vector or direction through the configuration, along which the scores of the sites (or species) have maximum correlation with that variable (Minchin 1991).Vector fitting allows the identification of variables which display significant monotonic trends across an ordination.A Monte Carlo approach (in DECODA) was used to test the significance of the maximum correlation for environmental variables through the configuration.
TWINSPAN (two-way indicator species analysis ;Hill 1979) and FUPGMA (flexible unweighted paired group arithmetic average; Belbin 1994) classifications were carried out on the material.Octave scaling was used as the abundance threshold values of pseudo-species in TWINSPAN (e.g. van der Maarel 1979).Four maximum levels of division were applied, and default options were used in all other cases.The recommended Bray-Curtis coefficient was used as association measure in FUPGMA (Belbin 1994).All species were included in the numerical analyses.
The number of species (species richness), and Shannonʼs (Hʼ) and Pielouʼs (Jʼ) diversity indices (Pielou 1966, Peet 1974) were calculated for the individual plots in order to describe different features of the alpha diversity.Hʼ contains elements both of species richness and evenness, and Jʼ is actually Hʼ based evenness.One-way analysis of variance (ONEWAY) was performed to investigate the variation of the environmental sample variables between numerical vegetation units.Two-way analysis (MANOVA) was applied to test the variables between site quality classes by succession phases (SPSS-X Userʼs Guide 1988).Since the dependent variables were characterized by high variation within the groups, and the variances were in general not homogeneous, a log 10 transformation was performed (except for pH) prior to the analyses.

Major Compositional Gradients
In the 1st dimension the spruce mires were separated from the pine mires; the plots of MkKmu and RhKmu, as well as the transformed sites of spruce mire origin (in this case MtkgI and Rhtkg-), obtained low scores, whereas the plots of ombrotrophic pine bogs, e.g.RamTRmu, obtained high scores along HMDS1 (Fig. 3).The pine mires of spruce mire influence (KgRmu, KRmu, PsRmu) were located in the middle parts of this dimension.Also the fertility series (Mtkg-Ptkg-Vatkg) of transformed sites (except Rhtkg) appeared to run parallel with the 1st axis.The factors most strongly correlated with the 2nd HMDS dimension were variation in drainage succession phase (moisture) and nutrient status (cf.Fig. 4).For instance, the regressive sample plots (mur, tkgr), and the sample plots of surfacewater influence (Lu) and the flark-rich (Ri) plots, were separated on the 2nd axis (Fig. 3).
The RhKmu, Rhtkg-and regressive MkK sites shared a considerable part of the species pool, and were not separated from each other until the 3rd dimension (Fig. 3).On the RhK plots there were also slight fen-like features (of herb-rich sedge hardwood-spruce fen, RhSK), e.g.Menyanthes trifoliata, Potentilla palustris and Carex rostrata (App.1).One of the plots with an abundant occurrence of Molinia caerulea was also discerned along this dimension, but in the opposite direction due to its special species composition.2).Above: Axes 1 and 2. Below: Axes 1 and 3.
Table 2. Maximum correlations between the sample variables and the three-dimensional HMDS ordination space.Numbers 1 (0-10 cm) and 2 (10-20 cm) connected with the variable name refer to the peat layer.Nutrients as concentrations (mg g -1 ) (p<0.001 denoted as 0.000).

Correlation between Vegetation and Sample Variables
The variation in vegetation showed high maximum correlations with the stand characteristics (Table 2).The vectors expressing e.g.stand volume (V), stand basal area (G) and site index (H40dr), mainly ran from pine mires towards spruce mires (Fig. 4).The mean annual stand volume increment for the period 1981-91 (Iv10) showed somewhat higher covariation with the vegetation than the mean increment for the five-year-period before sampling (Iv5).The variables describing the quantity of deciduous trees pointed towards MkK and herb-rich types (cf.Table 1), whereas the dominance of Pinus sylvestris increased in almost the opposite direction (Fig. 4).
The alpha diversity variables increased principally along the second gradient (Fig. 4).However, the correlation of the species number within the ordination space was not the same order of magnitude as the correlations of Hʼ and Jʼ (Table 2).
Drainage age was weakly correlated with variation in the ordination space, partly because the material only consisted of the transforming (phase II; mu) and transformed sites (final phase III; tkg), and even regressive stages of these forms (Table 2).The peat thickness increased from spruce mires to pine bogs.
The correlations between the vegetation and humification degrees of the peat (particularly of the deeper 10-20 cm peat layer) in terms of the von Post value or bulk density were strong (Table 2).These two measures behaved in almost the same way (Fig. 4); the Pearson correlation coefficients between these two variables were as high as r = 0.78, p < 0.001 (0-10 cm) and r = 0.80, p < 0.001 (10-20 cm).
Of the total macronutrient concentrations in the peat (mg g -1 ), phosphorus (P) was the most strongly correlated with the variation in the vegetation (Table 2).The covariation between vegetation and nitrogen (N) was stronger for the 10-20 cm peat layer than for the 0-10 cm layer.The correlation between pH and the vegetation was not very strong but consistent.
The correlation between the ordination space and the potassium (K) concentration of the deeper peat layer was very weak.In five cases the K concentration of the deeper peat layer was below the analytical determination limit, 0.08 mg g -1 .The trend for calcium (Ca) was also weak.The organic matter content (%) of the peat increased in an opposite direction to the most important nutrient vectors.
It appears from the nutrient vectors that both the 1st and 2nd dimensions were connected to fertility.The 3rd compositional dimension was indeterminate in terms of the environmental variables measured.As presented above, however, this dimension arose as a result of distinct variation in vegetation between some of the sample plots.

Numerical Units and Site Types
The sample plots classified as the same type in the field were sometimes placed in different clusters by the numerical techniques.Conversely, there was a range of different original mire site types and actual types in the same numerical clusters (Table 1, Figs 5 and 6).Border variants or transitional forms of the site types were common (Table 1, Fig. 3).
TWINSPAN Groups 1 and 2 mainly contained ombrotrophic types of Site quality class V. Three oligotrophic (IV) sample plots of the composite site types with Sphagnum fuscum were also included in Group 2 (Table 1, App. 1).Minerotrophic vegetation was only scantily present on these plots.Group 7 in the FUPGMA classification corresponded to TWINSPAN Groups 1 and 2 (Fig. 6).Picea abies saplings occurred on almost all of the plots in the present study, however, the lowest number was present in infertile TWIN-SPAN Groups 1 and 2 (App.1).
TWINSPAN Groups 3 and 4 mainly contained peatlands of Site quality class IV.In Group 3, however, there were nine plots determined as Vatkg (Class V) in the field (Table 1).Group 3 mainly consisted of transformed (tkg) sites, whereas Group 4 consisted of transforming (mu) sites.Mire species were the most important indicator species for Group 4, separating it from Group 3 (Fig. 5).In the FUPGMA classification, the same plots were found in Groups 1 and 4, with some exceptions (Fig. 6).The relatively homogeneous FUPGMA Class 4 almost exclusively contained tkg sites, as well as both Vatkg (V) and Ptkg (IV).
It appears from the sorted TWINSPAN table (App. 1) that S. fuscum still occurred, although relatively sparsely, on most of the plots classified as Vatkg in Group 3. Hylocomium splendens occurred on most of the remaining plots classified as Ptkg (or sites developing most likely into Ptkg).There were also differences in the abundance relationships of some common species, e.g.Eriophorum vaginatum, Cladina arbuscula, C. rangiferina, between the two subroups of Group 3 (App.1).The differences in the abundances of Vaccinium myrtillus and Carex globularis were relatively small between these two subgroups.
Sites classified to PtkgII in the field were not distinguished in the numerical analyses from PtkgI, or from the sites most likely developing into PtkgI.The Carex peat constituent of the deeper peat layer in the PtkgII plots indicated origin from the minerotrophic composite type.Possibly some other sample plots (e.g.3442) would have been PtkgII rather than PtkgI on the basis of the composition of their surface peat (Table 1).
The plots in Group 7 represented typical V. myrtillus tkg sites (MtkgI).Thus, there were differences in light conditions, in the hummock-level bog -spruce mire influence and in the moisture conditions between these groups.FUPGMA Group 3, which however contained only tkg sites, mainly corresponded to these groups.The only MtkgII plot in this material was located in TWINSPAN Group 6 (and in FUPGMA 3).TWINSPAN Groups 8 and 9 (FUPGMA Groups 5 and 6) were fairly luxuriant, with many herb and grass species (App.1).Most of the plots in TWINSPAN Group 8 were regressive, i.e. re-paludified.

Numerical Units and Sample Variables
TWINSPAN Groups 1 and 2 were characterized by low stand volume increment (Iv10), ranging from 1.8 to 2.1 m 3 ha -1 a -1 on the average, and consequently a low H40dr of about 11 m (Fig. 7) (cf.also FUPGMA 7).Iv10 varied more clearly than H40dr between TWINSPAN Clusters 3-5.Iv10 was larger in Group 3 than in Groups 4 and 5, which contained transforming and regressive transforming sites.The values of H40dr were relatively equal in TWINSPAN Groups 6 and 7.The variation in Iv10 was considerable in Group 7. Regressive development on most of the plots in Group 8 was reflected e.g. in the site indices (Iv10, H40dr).
The alpha diversity in TWINSPAN Groups 1 and 2, which mainly contained ombrotrophic sites, was moderate (Fig. 8).There were various ecological influences on the sites of TWINSPAN Group 5, which was also reflected as a relatively high number of species.The species number was the highest in herb-rich types of TWINSPAN Group 9.
According to the F-values, Hʼ and Jʼ clearly differed between the FUPGMA classes (Fig. 8).The indices were low, indicating increasing species dominance, in Classes 3 and 4, which contained transformed sites.TWINSPAN Groups 1 and 2 were also characterized by the clearly thickest peat layer, although it was measured down to a maximum depth of 1.5 m (Fig. 9).The humification degree of the peat was low in these Groups 1 and 2.
The organic matter content was low in the surface peat in Group 5 (Fig. 9), which presumably reflected the additions of suspended mineral colloids with moving water.The high bulk density value (and low organic matter content) of the deeper peat layer in TWINSPAN Group 8 most probably indicated the same: there was surface-water influence on many of the plots also in Group 8 (Table 1); as indicated, e.g., by Salix phylicifolia, Calamagrostis purpurea and Carex canescens (App.1).
Most of the nutrient concentrations were low in TWINSPAN Groups 1 and 2 (Fig. 10).For example, the average N concentration of the surface peat in Group 1 was about 1% (10 mg g -1 ).The average N concentration was highest in Group 5, where it was as much as 2.54% in the deeper peat layer.The P and K concentrations of the peat were lower in Group 3 than in Groups 4 and 5.The values of most of the peat and nutrient variables, were relatively equal in TWINSPAN Groups 6 and 7.The P concentration was high in clusters 8 and 9, but e.g. the K concentration varied considerably (Fig. 10).
Degree of humification, pH value and the N and S concentrations were generally higher in the deeper peat layer, whereas the concentrations of P, K, Ca and Mg were usually higher in the surface layer (Figs 9 and 10).

Sample Variables by Site Quality Class and Succession Phase
In general, there were clear differences in the average values of the mean annual stand volume increment (Iv10), the post-drainage dominant height (H40dr) and the stand volume (V) between the site quality (fertility) classes.Both Iv10 and H40dr in Classes II and III were relatively close to each other (there were few observations from Class II).The values were higher on transformed (tkg) sites than on transforming (mu) sites (Fig. 11, Table 5).
The alpha diversity was the highest on herb-rich  (II) sites while the differences between the other classes were small (Fig. 12).Higher index values were generally observed for the mu sites than for the tkg sites.However, the average number of species on the oligotrophic sites (IV) was almost the same between the succession phases, and at the infertile end (V) the value for the tkg sites even exceeded that of the mu sites.The clearest differences in the average values of the bulk density and the von Post humification degree with respect to adjacent fertility classes were between the oligotrophic (IV) and ombro- trophic (V) sites (Table 3).The values appeared to be generally slightly higher on the tkg sites than on the mu sites (Table 3), even though the differences were not in general statistically significant (Table 5).Among the mu sites in Site quality class III, there were sample plots with surface-water influence, and the bulk density of the deeper peat layer was high.In the deeper layer the peat pH was somewhat lower on the tkg sites than on the mu sites (Tables 4 and 5).
Nutrient amounts combine the ʻcharacteristicsʼ of bulk density and nutrient concentrations.The average amount of N was clearly the lowest in the ombrotrophic site types (V), about 1950-2550 kg ha -1 for the peat layers combined (0-20 cm) (there were 17 thin-peated cases from which only surface-layer samples could be taken).The corresponding values for herb-rich sites (II) were 5050-6450 kg ha -1 .The differences between the succession phases were small.On nutrient-poor sites there was slightly more N in the tkg phase than in the mu phase (Fig. 13, Table 4).
The average amount of P in the 0-20 cm layer varied between 95-470 kg ha -1 among site quality classes.The P concentration was to some extent lower on the tkg sites than on mu sites (Tables 4  and 5).No clear differences in the amounts of P were found between succession phases.In particular, this was the case on ombro-oligotrophic sites (Fig. 13).
The largest amounts of K were found in Site quality class III, at an average of about 110-180 kg ha -1 .The amount of 180 kg of K obtained in the mu phase was due to only a few high values, and the within-site variation was large.The average K amount in Class V varied about 50-70 kg ha -1 .There was slightly less K on the tkg sites than on the mu sites.This appeared more clearly when expressed as concentrations than expressed as amounts (Fig. 13, Tables 4 and 5).
The average amount of Ca varied from 400-550 kg on the ombrotrophic sites to 730-900 kg ha -1 on the herb-rich sites (Fig. 13).In Site quality classes IV-V there seemed to be slightly more Ca in the tkg phase than in the mu phase, while on the more fertile sites the situation was reversed by relationships in the 10-20 cm peat layer (Fig. 13, Table 4).However, the within-class variation was large.
The amount of Mg was the highest (even about 250 kg ha -1 ) on the mu sites in Fertility class III, but this was primarily due to a few high values.On the other sites the average amount of Mg varied between 65 and 125 kg ha -1 .There seemed to be slightly more Mg on the mu sites than on the tkg sites, while on the infertile sites the values were of the same order of magnitude (Fig. 13, Table 4).
The average amount of S varied from 240-290 kg ha -1 in Class V to 800-950 kg ha -1 in Class II (Fig. 13).In site quality classes III-IV there was almost the same amount of S, about 470-500 kg ha -1 .No significant differences were found between the succession phases (Fig. 13, Tables 4 and 5).

Methodological Aspects
The problem with the use of ordination and classification methods in ecology is the selection of appropriate strategies from an increasing range of choices (e.g.Belbin and McDonald 1993, Belbin    1994, Økland, R. 1996).The parallel use of different multivariate approaches is generally useful because the results are frequently complementary (e.g.Wiegleb 1989, Økland 1990a, Belbin and McDonald 1993, Økland et al. 2001).
On the basis of Minchinʼs (1987) simulations, nonmetric multidimensional scaling (NMDS) is recommended as a robust technique for indirect ordination.According to Minchin (1991), hybrid MDS (HMDS), which was applied in the present study, performs even somewhat better than global (G) or local (L)NMDS.Hybrid MDS, which unites the techniques of metric and nonmetric scaling, is therefore preferable (Belbin 1994).
For most purposes the results from e.g.GNMDS or LNMDS are adequate (Minchin 1991(Minchin , Økland, T. 1996; see also Hotanen 1994 for preanalyses of the present material).
According to the ordination diagrams and FUPGMA classification, there appeared to be mode structures (Dale 1995) in the material.However, the clusters are likely to have been the result of sampling effects (Belbin andMcDonald 1993, Dale 1995).Random or systematic sam-pling over an area produces a material in which the combinations of environmental variables with greatest areal importance recur as clusters in the ordination space (cf.Økland 1990a).However, this could be the very feature that researchers are frequently seeking, i.e. frequent plant communities, and the kind of variation in the environmental variables which they contain.
The low number of observations in some categories, and the possible occurrence of spatial autocorrelation (sample plots that are close together tend to be more similar than plots further apart) resulting from systematic cluster sampling (Legendre and Legendre 1998) precluded further testing of relationships among clusters.The variation in the individual tracts, however, might have been large.For instance, the succession phase varied within 12 tracts and the site quality class in 9 tracts out of 40 (at least one plot differed from the two others), irrespective of the site type, additional criteria or border variants.
The TWINSPAN and FUPGMA classifications showed obvious similarities, thus verifying the results.Both methods appeared to classify relatively well the different succession phases; FUPGMA perhaps better than TWINSPAN.In the comparisons made by Belbin and McDonald (1993), FUPGMA provided a better recovery of true cluster structure than TWINSPAN.This is expected, because TWINSPAN does not strictly seek a cluster form.For example, FUPGMA pays better attention to a (similar) abundance distribution of species than TWINSPAN.This was reflected e.g. in the high F values of indices Hʼ and Jʼ, which are determined by both species number and abundance relationships, between the FUPGMA clusters.According to Dale (1995), TWINSPAN may better reflect what ecologists are intuitively seeking than alternative variance minimization methods, because it is primarily based on a method that identifies gradients (ordination).Furthermore, TWINSPAN incorporates characterization through indicator species.We are often more interested in these differentiating species than in the existence of clusters per se (Dale 1995).The table sorting procedure in TWINSPAN is useful, as well as informative (cf.App. 1).
The numerical methods were used in this study as tools for summarising relationships.An element of subjectivity -in addition to the selection of method -is invoked by setting the programme options, as well as in the final compilation of the results (Oksanen 1984, Kuusipalo 1985).As pointed out by Oksanen (1984), the classifier must be responsible for his results, he cannot hide behind ʻobjectiveʼ methods.The optimal level of classification, the number of clusters and the details affecting the desirable locations of boundaries, are then controlled by the ecologist (Kuusipalo 1985, Jeglum 1991).

Vegetation Gradients, Diversity
The main compositional gradient primarily displayed variation in spruce mire influence -hummock-level bog influence (Eurola et al. 1984), reflecting supplementary nutrient effects (external supply of nutrients; mire margin of Sjörs 1948) against mire inherent (centre) effects (mire expanse of Sjörs 1948;Eurola et al. 1984), and simultaneously, the variation in fertility.
The main compositional gradients of drained boreal mires generally reflect spruce mire -hummock-level bog influences, the variation in nutrient (trophic) status and the variation in moisture / secondary succession (Reinikainen 1988, Hotanen and Nousiainen 1990, Hotanen and Vasander 1992, Eurola et al. 1995, Paalamo 1996, Hotanen et al. 1999).In general, the gradients are intermixed with each other.Especially the trophic gradient is complex and divergent (see also Pakarinen andRuuhijärvi 1978, Økland et al. 2001 for pristine mires).In material confined to certain mire types, for example pine mires, the gradient structure may be simpler (e.g.Laine et al. 1995).
Forest drainage, because it results in decreased variation in soil moisture and trophic status (e.g.Vahtera 1955, Eurola and Holappa 1984, Laiho and Laine 1994), emphasizes the importance of spruce mire and hummock-level bog influences in controlling the structure of the plant communities on drained peatlands (Eurola et al. 1995, Hotanen et al. 1999).These influences become stronger against the other gradients (e.g.rich fen, poor fen, surface-water) because they also include, on the one hand, the vegetation of spruce-dominated mineral soil forests and, on the other hand, the vegetation of more infertile pine-dominated mineral soil forests (Eurola et al. 1984(Eurola et al. , 1994)).
In general, effective drainage disconnects the surface-water influence.However, especially in material collected in the aapa mire zone, the surface-water influence is frequently indicated by the vegetation (Pienimäki 1982, Hotanen et al. 1999).As in the present material, there are swamp vascular plant species in the vegetation tables of old drained sites in southern Finland, too (Sarasto 1961a, Laine and Vanha-Majamaa 1992, Laine et al. 1995).Mosses typical of swamps and mires are more sensitive to the water level drawdown.The flows of surface-water have a long-lasting effect on the peat by the mineral material and nutrients that they have transported (Eurola andHolappa 1985, Eurola andHuttunen 1990).
The alpha diversity increases temporarily during secondary succession (Sarasto 1961a, Reinikainen 1984b, Vasander et al. 1997, Hotanen et al. 1999).This is attributed to the simultaneous occurrence of mire species, colonizing pioneer species and forest species.The alpha diversity is, generally speaking, the highest on transforming sites (Sarasto 1961a, Reinikainen 1984b).Since the present material included no undrained or recently drained mires, and there was a combined moisture -fertility gradient that mainly ran in the same direction, the alpha diversity distinctly increased along this dimension.
On infertile sites, however, the differences in species number between transforming and transformed phases are not clear (Sarasto 1961a, Reinikainen 1984b).This was also evident in this study in Site quality classes IV and V. On these sites the values of Hʼ and Jʼ showed increasing dominance and a clearer shift towards forest vegetation than the number of species.

Comparison among Classifications
The dispersion of the sample plots, classified as the same type in the field, into several numerical clusters may result for the following reasons: a numerical method and the human eye do not analyse the vegetation in the same way (Jongman et al. 1987), the same original mire site type may diverge into a different result after drainage (Hotanen et al. 1999), and misclassifications.The range of original mire types represented in the same clusters immediately seems to accord with the hypothesis that the post-drainage succession of many mire types leads to relatively uniform types of peatland forest (Sarasto 1961a, Hotanen andVasander 1992).
Border variants or transitional forms between site types commonly occur on drained mires (also Hotanen and Nousiainen 1990).The understorey vegetation composition overlaps more with neighbouring mire site types on successional sites than on stable ones (Cajander 1949, also Keltikangas 1945).The relationship between vegetation and site factors in the succession communities of drained mires should also be weaker compared to ʻclimaxʼ forests (e.g.Keltikangas 1945, p. 33-45, 91-107).
Some oligotrophic sites (with Sphagnum fuscum) in the transforming phase were not distinguished from ombrotrophic sites in the numerical analyses.The mixing of ombrotrophic and minerotrophic mire vegetation in this area, which is located in the intermediate zone between the middle-boreal (aapa mires) and southern vegetation (eccentric bogs), is evident (Tolonen 1967).The peat surface is often stratified in this border zone.Especially minerotrophic vascular plants, e.g.sedges, may occur on infertile sites if their roots extend into a deeper, more fertile layer.The results of Pakarinen (1976) and Pakarinen and Ruuhijärvi (1978) also suggest that there is frequently no sharp limit between ombrotrophic and minerotrophic vegetation.This limit is supposed to become blurred as a result of factors ranging from small-scale variation, via betweensite variation, to regional variation of the peatland environment (cf.Pakarinen 1979Pakarinen , Økland 1989)).Moreover, in boreal conditions, the composition of the vegetation of minerotrophic composite pine mires (as in this study) becomes poorer after drainage, at least temporarily (Sarasto 1961a, Eurola et al. 1984, Laine et al. 1995, Hotanen et al. 1999).In general, ombrotrophy becomes more apparent in the soil water chemistry than in the vegetation or peat chemistry (Tolonen 1974, Eurola and Holappa 1984, cf. Damman 1990).
Furthermore, the data set contains many sample plots that originally belonged to the body of pine mires with spruce mire influence, as well as closely related oligotrophic spruce mires.This part of the mire type system is heterogeneous with respect to classification (Hotanen 1989, Nieminen and Pätilä 1990, Korpela 1999).The result of drainage of such sites may vary from poor Ptkg (or even luxuriant Vatkg) to Mtkg (Sarasto 1961a,b, Laine and Vasander 1990, Hotanen and Vasander 1992).Additional impacts such as fertilization further adds complexity to the variation, and the presence of different variants ranging from Vatkg to Mtkg is readily understandable.The under-growth of Picea abies on dwarf-shrub pine bogs (IR) (Heikurainen 1959) may sometimes result in their being classified as spruce-pine swamp (KR).However, saplings of P. abies are assumed to grow clearly better in KR than in IR.
The results show that distinguishing the PtkgII type from the PtkgI type in the transformed phase (Sarasto 1961a,b) purely on the basis of vegetation is difficult.Information about the peat type and sometimes also about the structure of the tree stand is needed for proper classification (Laine 1989, Laine andVasander 1990).
The only sample plot classified as a S. fuscum pine bog (RaRmur, Site quality class VI) was located among other infertile, often rahka-hummocky cottongrass pine bogs (RamTR, V) in the numerical analyses.RamTR changes gradually into RaR; in fact the former is a precursor of the latter (Tolonen 1967, cf. also ʻBunte Reisermooreʼ in Eurola 1962).

Species Appraisal
Being a hummock-level species, S. fuscum survives for a long time on well-lit drained pine mires, but in the transformed (tkg) phase it is sparse (cf.e.g.TWINSPAN Group 3) because it does not withstand shading (Sarasto 1961a, Hotanen 2000a).Since old drainage areas on more fertile pine mires with S. fuscum on hummocks most probably have a higher stocking and are more shaded than the poorer ones, the differences in the abundance of S. fuscum between ombrotrophic and minerotrophic sites especially may, however, persist for a considerable period of time (Sarasto 1961a, Laine et al. 1995).
There are no unambiguous indicator species constant enough in the understorey vegetation between Vatkg and Ptkg.In many cases, as in the present material, it is a question of the abundance relationships of certain common species, e.g.Eriophorum vaginatum (Hotanen 2000b), Cladina spp., Cetraria spp.(Nousiainen 2000a,b) and dwarf shrubs, e.g.Calluna vulgaris (Salemaa 2000), between these sites.However, C. vulgaris and other pine mire dwarf shrubs may be abundant in Ptkg of KgR origin (Laine and Vasander 1990).There might be also regional differences in both Ptkg and Vatkg with respect to the coverages of Empetrum and Calluna.Also the ordination diagrams of Laine (1989) and Laine and Vanha-Majamaa (1992) show that the differences between Vatkg and PtkgII of c. 25-year-old drainage areas are diffuse.
Being a hummock-level species, E. vaginatum (Eurola et al. 1994) is frequent, although not abundant, even in the tkg phase (Sarasto 1961a, Hotanen 2000b,).Ombrotrophic E. vaginatum becomes more abundant on regressive sites of many minerotrophic types, too, changing their overall appearance in a less fertile direction (Kuusipalo and Vuorinen 1981).Thus, E. vaginatum is problematic in nutritional site type classification because, in addition to the nutrient status, the amount of light and especially the depth of the water table level strongly affect its abundance (Laine andVanha-Majamaa 1992, Laine et al. 1995).In any case its high coverage on drained mires is a negative sign in the sense of wood productivity.
Carex globularis occurred on only one plot in TWINSPAN Group 1, and on clearly less than half of the plots in Group 2. In these infertile groups the average bulk density of the surface peat (0-10 cm) was below 0.08 g cm -3 (cf.also the transforming (mu) sites in Class V; Table 3).In the material presented by Laine and Vanha-Majamaa (1992), C. globularis did not occur on drained pine mires where the bulk density of the surface peat was lower than ca.0.08 g cm -3 .C. globularis is scarce on ombrotrophic peatlands (Hotanen 2000c), but is assumed to increase slightly on those sites with a secondary succession caused by humification and an associated fertilization.Being a hummock-level species, C. globularis, which also indicates spruce mire influence (Eurola et al. 1984), may temporarily increase in abundance after ditching and subsequently change the overall vegetation appearance of oligo-mesotrophic peatlands (Class III) towards Class IV (Sarasto 1961a,b).On old drainage areas of fen and composite type origin (e.g.PtkgII, MtkgII), C. globularis seems so far to be scanty (Sarasto 1961a, Reinikainen 1984b, Laine et al. 1995).
Care should be taken when applying Vaccinium myrtillus in the classification of drained mires if the material contains both treeless/composite types and genuine forested types.V. myrtillus colonises and increases slowly on former drained sites (Sarasto 1961a, Laine et al. 1995, Laiho 1996).However, on genuine transformed sites its coverage generally increases on moving from pine bogs, through spruce-pine swamps, to oligomesotrophic spruce swamps (Sarasto 1961a).
Hylocomium splendens was abundant in MtkgI, and it also occurred in Ptkg although rather scattered (also Reinikainen 1984b, Laine et al. 1995).When the fen influence increases, H. splendens appears to decrease; for example, it did not occur on Sarastoʼs (1961a) Ptkg sites of tall-sedge fen (VSN) origin.H. splendens did not grow on Vatkg sites of the present study but it may grow, however, scantily in Vatkg of especially genuine forested pine mires (Sarasto 1961a, Reinikainen 1984b).
In addition to the site fertility, the phase in secondary succession also affects the abundance of Dicranum majus (cf.TWINSPAN Groups 6 and 7).D. majus is a species of the final phase of drainage succession, and its optimal sites appear to be transformed sites of Site quality class III of spruce-mire origin (Sarasto 1961a, cf. also App. 1).In Ptkg it may occur sparsely, but is very scanty in PtkgII especially; in MtkgII, too, it is not as abundant as in MtkgI (Sarasto 1961a, Laine et al. 1995).Thus, the origin of the site (genuine forested vs. treeless or composite) also affects its abundance.When moderately abundant it is a useful species for distinguishing Site quality class III from Class IV on tkg sites (cf.also Mäkipää 2000 for mineral soils).
Equisetum sylvaticum is most abundant on sites of classes III and II of spruce swamp origin (Sarasto 1961a, Korpela 2000).In the present material it was common (on some plots rather abundant) in moist TWINSPAN Groups 6 and 8, but less abundant in tkg Group 7. It was also abundant on both herb-rich plots (mu and tkg) in Group 9. Generally speaking, the coverage of this species has been found to decrease in the later succession phases of spruce swamps (Korpela 2000).However, after drainage it may temporarily increase in V. vitis-idaea spruce swamps (PK), making the vegetation appearance of the swamp more luxuriant (Sarasto 1961a,b).It also benefits from logging (Korpela 2000).Thus, in the classification of drained mires this species is to some extent problematic.In data from southern Finland, E. sylvaticum proved to be the most effective differential species for spruce mires in association analysis (Pakarinen 1982).

Tree Stand Parameters
The nine TWINSPAN groups were roughly divided into four levels with respect to the site index (H40dr) values.The situation with respect to the stand volume increment (Iv10) was more variable, and the variation within the groups was also greater.The increment during the last ten growing periods (Iv10) seems more sensitive to e.g.variations in moisture than H40dr, which takes into account the whole development of the standing stock after drainage.The changes in moisture, e.g.regressive development, are then reflected in different classifications of the actual vegetation.In the short term, the stand volume increment reacts faster to environmental changes than the other vegetation.For example, the covariation between the vegetation and the last five increment periods (Iv5) was weaker than the covariation between the vegetation and Iv10.This is expected as the ten-year data are likely to show a clearer relationship just because more of the noise (random variation among years) has been averaged out when the longer time period is concerned.
In several studies the post-drainage stand increment has been found to be related to the original site type (e.g.Laine and Starr 1979, Keltikangas et al. 1986, Hökkä 1997, Gustavsen et al. 1998, Hökkä and Penttilä 1999 and references therein).The increment also appears to correlate strongly with the actual vegetation in ordinary forests on drained peatlands (also Laine and Vanha-Majamaa 1992).However, large variation in different site types is typical of the tree stand characteristics of peatlands (e.g.Hökkä 1997).In addition to the stand development phase (e.g.Keltikangas et al. 1986, Laine 1989), the drainage phase also causes variation in tree growth, as was seen in the present study.The differences in standing volume explain a high proportion of the differences in the increment values, since there is a strong positive correlation between stand increment and volume (e.g.Laine and Starr 1979).In the data of Hökkä and Penttilä (1999) the relative growth rate (RGR) index means also increased over time, which was partly interpreted as an effect of a true increase in the wood production potential of the sites.It will be clear from the foregoing that, as far as stand yield characteristics are concerned, only relatively rough comparisons can be made between the present study and other studies.
The stand volume increments of the different site quality classes corresponded rather well to the values of the comparable peatland forest types in southern Finland, except for the herb-rich type (Rhtkg) (Laine 1989).The increment 7.4 m 3 ha -1 a -1 reported by Laine for Rhtkg was clearly higher than that obtained in this study for Site quality class II.However, this class contained only a few observations.The volume increments in Class III were approximately equivalent to those for the corresponding spruce mire group reported by Gustavsen et al. (1998), and for MK (III) obtained by Laine and Starr (1979).The values in Vatkg (V) were almost as high as the values for old drainage plots of IR and TR ombrotrophic sites given by Gustavsen et al (1998).According to their study, the growth values of infertile LkN (low-sedge fen) and RaR were even lower than those for the transforming (mu) sites in Class V in this study.The growth values for the mu sites of Class V were equivalent to the figures published by Laine and Starr (1979) for 20-to 30-year-old drained and unfertilized IR.
The average range of H40dr in Site quality class V varied from 11 to almost 13, while in the ombrotrophic group presented by Gustavsen (1996) it was 10-12.The ombrotrophic group of Gustavsen (1996) contains a wider range of mire site types, also including the infertile LkN and RaR.The highest H40dr values presented by Gustavsen (1996) exceeded those of the present material due to the inclusion of fertile mire site types, e.g.LhK (eutrophic paludified hardwoodspruce forest).Otherwise, the values in the comparable groups in these two studies were close to each other.However, the fact that the individual yield groups of Gustavsen contained types of various site quality classes makes comparison difficult.The materials of Gustavsen (1996) and of Gustavsen et al. (1998) were to some extent representative of older drainage areas and more southerly regions than the material of the present study, but they contained no fertilized sites.

Bulk Density and Degree of Humification
The bulk density (D b ) and the degree of humification (H v.Post ) followed almost the same pattern.D b correlates strongly with H v.Post , especially in Sphagnum and woody peat samples regardless of drainage (e.g.Päivänen 1969, Tolonen andSaarenmaa 1979).A strong correlation between vegetation and D b (Laine and Vanha-Majamaa 1992), and between vegetation and H v.Post , has also been reported (Hotanen et al. 1999).These variables increase along both the secondary succession (Laiho and Laine 1994) and along the fertility gradient (Laine andVanha-Majamaa 1992, Minkkinen andLaine 1998).The degree of humification also increases, as in the present material, along the complex gradient of hummock-level bog -spruce mire influence and fertility (Hotanen et al. 1999).D b and H v.Post increased on moving from transforming sites to transformed sites, and from surface peat (0-10 cm) to deeper peat (10-20 cm).Similar results have been found by e.g.Laiho and Laine (1994).The water level drawdown after ditching results in subsidence and compaction of the peat surface, which increases the bulk density of the peat.Subsequently, the accelerated rate of organic matter decomposition and increasing weight of the tree stand further compacts the peat (Minkkinen andLaine 1998, Laiho et al. 1999).The downward compaction of peat is not restricted to drained sites, as described in detail e.g. by Clymo (1983) and Økland and Ohlson (1998).
There are many potential reasons for the stronger covariation between the vegetation and both D b and H v.Post in the deeper peat layer.After drainage the Sphagnum peat of infertile ombrotrophic sites decomposes at a slower rate than the peat of more fertile sites (Vahtera 1955, Laiho et al. 1999).This is clearly shown in TWINSPAN Groups 1-2, where the differences in both D b and H v.Post between the different peat layers were smaller than those on the more fertile sites (Fig. 9).The thickness of the peat layer may also have an effect; in the study of Minkkinen and Laine (1998), the relative increase in D b after drainage was greatest in the shallow peat layers (cf.Fig. 9).The total variation in species composition between different site types, i.e. the beta diversity, decreases along with the post-drainage succession (e.g.Vasander et al. 1997).Thus, both the litter and the topmost peat surface are also assumed to be more similar between site types after drainage.

pH and Macronutrients
pH.Although the covariation between pH and the vegetation was not very strong, it was consistent.The pH value generally increases when moving down the peat profile (e.g.Vahtera 1955).On the other hand, pH usually decreases with increasing drainage efficiency and during secondary succession (Lukkala 1929, Vahtera 1955, Laine et al. 1995).This was evident in the present material, especially for the 10-20 cm layer between succession phases.The decrease in the pH of the uppermost peat layer resulting from drainage is also greater in the more fertile site types than in infertile types (Lukkala 1929, Laine et al. 1995).Consequently, the covariation was slightly weaker for the surface peat layer.According to Vahtera (1955), the original acidity ranking of the sites is, however, usually retained during secondary succession (cf.Laine et al. 1995).
Nitrogen and Phosphorus.Past fertilizations may have to some extent influenced the correlations between the vegetation and nutrient concentrations in the present study.However, the fertilized plots were distributed relatively evenly between the different vegetation classes (Table 1).It is also likely that fertilization has an effect on the amounts of nutrient in the surface peat layers, despite leaching and nutrient uptake by the trees.The objective just was to analyse the prevailing conditions in ordinary, commercial forests on drained peatlands, which often contain fertilizations.
Regardless of the fact that a data set from drained mires does not necessarily contain an unambiguous trophic gradient (e.g.Westman 1987), the correlation between the actual vegetation and certain nutrients, e.g.N and P, may be strong, as was observed in the present material.The differences in the N concentration of the peat between the site types was more apparent in the 10-20 cm peat layer than in the surface layer.This is most probably due to the increasing amount of litter derived from the developing tree stand, which decreases the N concentration of the uppermost peat in Carex-dominated fertile peat at a faster rate than in Sphagnum-dominated infertile peat (Vahtera 1955).Also Laiho et al. (1999) found that the N concentration from the surface layer down to the 10-20 cm layer seemed to increase faster in the meso-oligotrophic sites than in the oligo-ombrotrophic sites.The N concentration generally increases on moving downwards in peat profiles (e.g.Vahtera 1955, Holmen 1964, Laiho et al. 1999), but there are also some opposite observations between e.g. the 0-10 and the 10-20 cm layers (Kaunisto and Paavilainen 1988).
The average amounts of total N in the 0-20 cm peat layer were slightly lower or at the same level as those presented by Kaunisto and Paavilainen (1988) for old transformed areas.In their material the average range for ombrotrophic to herb-rich sites varied between 3000 and 7000 kg ha -1 .The value (4300 kg) for the transformed V. vitis-idaea spruce swamp (PK) reported by Kaunisto and Paavilainen was equivalent to that of Site quality class IV in the present study.Conversely, the amounts were slightly larger (or at the same level) as those reported by Nieminen and Pätilä (1990) and Laine et al. (1995) for comparable sites.The value (1912 kg) for the dwarf-shrub transforming type (IRmu) in eastern Finland presented by Finér (1991) is almost the same as that obtained here for the mu phase of Class V.In an earlier material from the same research area (Hotanen 1991), the average amounts of N in the sites developing into Vatkg and Ptkg were slightly lower than the values obtained for corresponding sites in this study.The average values for the sites developing into Mtkg were similar in both of these studies.
The observed correlation between the vegetation and P concentration was slightly stronger for the deeper peat layer, which indicates that the intensity of relative surface enrichment (or impoverishment of the deeper layer) of P to some extent differed between the sites (also Laiho et al. 1999).P generally accumulates slightly in the surface peat (e.g.Kaunisto and Paavilainen 1988).In some materials the gravimetric P concentration seems to be relatively even from the 10-20 cm layer to the topmost layer (Laiho et al. 1999).
The average amounts of total P (0-20 cm) in Fertility class II (405-470 kg ha -1 ) and in the transforming sites of Class III (330 kg) were larger than those reported by Kaunisto and Paavilainen (1988) for comparable fertility classes.In other respects the amounts of P in corresponding classes were close to each other in both of these studies.The value for IRmu reported by Finér (1991) was 105 kg ha -1 , while in this study it was 95-102 kg for Class V.In the study of Nieminen and Pätilä (1990), the average P range for ombrotrophic to oligotrophic sites was between 110-220 kg ha -1 .In the material of Hotanen (1991), the amount of P was about 80 kg ha -1 in Class V, 180 kg in Class IV (here 180-205 kg) and 285 kg in Class III, while in this study it was 210 kg ha -1 in the transformed (tkg) Class III on the average.
The nutrient stores in the peat also vary with the secondary succession phase and drainage age (also Laiho and Laine 1994, Westman 1994, Laine et al. 1995).Especially the variations in the amounts of N and P, which are mainly organically bound, are related to the bulk density (Westman 1994).Therefore the amounts of these nutrients may increase along with secondary succession, at least in ombro-oligotrophic sites (cf.Fig. 13, Laiho andLaine 1994, Westman 1994).In the more fertile sites the N and P stores may remain unchanged, even though large amount are bound in the increasing tree stand biomass (Laiho and Laine 1994).This is caused by post-drainage subsidence of the mire surface and consequent compaction, which increases the bulk density and brings new stores from the deeper peat layers (Laiho and Laine 1994).According to Westman (1994), the amounts of N and P in MtkgI at first increased, but subsequently turned to a decrease along with drainage age class.In the study of Laine et al. (1995) this was also the case for N in meso-oligotrophic drained pine mires (RhSR, VSR).
Potassium.The correlation between the vegetation and K concentration of the surface peat (0-10 cm) was also relatively strong.The K concentration of the deeper layer was considerably lower, and there was no correlation with the vegetation.K is a very mobile nutrient (Damman 1978, Kaunisto and Paavilainen 1988, Laiho et al. 1999).The surface enrichment of K is due to effective biological cycling by the developing vegetation after drainage (Laiho 1997).In the drained pine mire material of Laine and Vanha-Majamaa (1992), the correlation between the vegetation and K concentration of the peat (0-20 cm) was at almost same level as the correlations for N and P. Nieminen and Pätilä (1990) found no close connections between the nutrient concentrations and vegetation, possibly due to the small differences between the trophy of the sites and to the fact that the material consisted of all succession phases from pristine pine mires to transformed sites.
The average amounts of K were larger than those presented for old drainage areas by Kaunisto and Paavilainen (1988), or those presented earlier for the same area (Hotanen 1991).Especially the value for the tarnsforming sites in Site quality class III was high (even 180 kg ha -1 ).This, however, was due to a few high K concentration and bulk density values, most probably caused by the presence of mineral material.The amount of K in the drained pine bog reported by Finér (1991) was 64 kg ha -1 , while in Class V in this study it was 50-70 kg.The average range between the different site types reported by Nieminen and Pätilä (1990) was 60-80 kg ha -1 .
In general, the concentrations and amounts of K are affected by the secondary succession phase and drainage age, as was evident in the present study: K is assumed to decline when virgin peatland becomes transformed (Kaunisto and Paavilainen 1988).In the study of Laiho et al. (1999), the gravimetric K concentrations dropped soon after drainage, but later on they remained relatively unchanged in spite of increased uptake by the tree stand.The decrease might be due to disturbances in the biological cycle and consequently increased leaching, while the K input in dry and wet deposition was probably the main factor explaining the lack of a reduction in soil K concentrations in older drained sites (Laiho et al. 1999).According to Westman (1994), the amount of K may even turn to a slight increase on some old transformed sites.This might be partly caused by the allocation of K into the fine roots of the developing stands, which is then included in the results of analyses of the surface peat (Westman 1994).
Calcium and Magnesium.There were no clear relationships between the vegetation and the concentrations of Ca and Mg.Kaunisto and Paavilainen (1988) reported a tentative positive trend with the site index in some drained spruce mires, but not in pine mires (cf.however Laine et al. 1995).In the data of Laine and Vanha-Majamaa (1992), the covariation between the vegetation and Ca concentration of the peat was somewhat lower than that for N, P and K.In a material from an old drainage area of spruce swamp (RhK) origin, the Ca concentration was the peat parameter that differed the most significantly between the vegetation groups (Mannerkoski 1979).Generally the concentrations of Ca and Mg appear to decrease from the surface layer down to the 10-20 cm layer (also Laiho et al. 1999), but the situation may also sometimes be the reverse (Kaunisto and Paavilainen 1988).According to Laiho et al. (1999), the concentrations may again increase below the rooting zone (>30 cm).
The differences between transforming and transformed sites in the concentrations and amounts of Ca and Mg were varied.This was also the case for Ca according to Laine et al. (1995).According to Laiho et al. (1999), the gravimetric Ca and Mg concentrations decrease in the rooting zone after drainage.In the present material this was not found between succession phases in nutrient-poor sites.In old drainage areas a considerable amount of Ca and Mg may have been taken up by the tree stand or leached out of the peat (Laiho 1997).
The average amount of total Ca varied less between the site types compared to the data of Kaunisto and Paavilainen (1988).The values for Rhtkg and Mtkg from Vesijako (1500-1800 kg ha -1 ) clearly exceeded the values of the present material.Otherwise, the amounts in corresponding site types were at almost the same level.In the material of Laine et al. (1995) the average amount of Ca varied between 298 and 1220 kg ha -1 .The average range between site types according to Nieminen and Pätilä (1990) was 500-750 kg ha -1 , and the amount in the IRmu reported by Finér (1991) was 392 kg ha -1 on the average.In the study of Hotanen (1991), the amounts of Ca varied only slightly, between 450 and 520 kg ha -1 on the average.However, the within-type variation was large.
The amounts of total Mg in certain Rhtkg and Mtkg reported by Kaunisto and Paavilainen (1988) exceeded the amounts of Mg in the present study (except the transforming III sites).Otherwise, the differences between corresponding site types were small.In the data of Nieminen and Pätilä (1990), the average amounts of Mg varied between 90 and 110 kg ha -1 , while in that of Hotanen (1991) between 60 and 80 kg ha -1 .Finér (1991) presented a value of 69 kg ha -1 for IRmu.On the basis of different reference materials (also e.g.Holmen 1964), there seem to be large regional differences in the amounts of Ca and Mg on minerotrophic mire sites especially, because these nutrients are mainly derived from the surrounding mineral soil or the mineral soil underlying the peat layer.
Sulphur.The total S concentration correlated consistently with the vegetation (also Pätilä and Nieminen 1990), and also increased from the surface layer down to the deeper layer.Similar results have been reported for drained pine bogs and oligotrophic pine mires (Pätilä andNieminen 1990, Braekke andFinér 1991).According to Finér (1991), the average amount of S in the 0-20 cm peat layer in IRmu in eastern Finland was 258 kg ha -1 , which is almost the same value as for the transforming (mu) sites of Site quality class V in this study.The values for the ombrotrophic group in the study of Pätilä and Nieminen (1990) were slightly higher, and corresponded well to the transformed sites of Class V in the present material.The average amount of S in the lowsedge pine mires of Pätilä and Nieminen was 429 kg ha -1 , while in Class IV of this study about 470-500 kg ha -1 .

Concluding Remarks and Prospects for Further Research
On the basis of classifications made in the field, numerical community analysis, parallel stand measurements, peat chemical and physical properties, it is concluded that the actual understorey vegetation well reflects various aspects of the ecological conditions, even in labile and disturbed communities of drained peatland forests.
In a representative (systematic or random) sample of drained peatlands, the compositional and ecological gradient structure may be complex due to differences in origin of the site types, successions and other impacts.The trophic gradient is particularly complex, consisting of several single variables that do not vary completely in parallel.In ordinations, the individual nutrient vectors are usually located at angles to each other and to the main compositional gradients.Thus, there is not necessarily an unambiguous trophic gradient.
Border variants or transitional forms of site type are common in the successional communities of drained mires.Consequently, the classification is fine-grained and demanding, often based on the differences in abundance of common species.According to the numerical and field classifications, the borders between transformed (tkg) phases are also often diffuse.However, classification is useful for complementing the site description made by ordinations.
The additional criteria indicated by the vegetation, e.g.surface-water influence and Sphagnum fuscum-hummocky, more closely define the ecology of the site.These criteria are also useful in the classification of drained peatlands.In addition, the degree of humification (H v.Post ) (especially of the 10-20 cm peat layer) seems to provide a simple and usable auxiliary variable in the ecological description and classification of drained mires.
There are clear differences in many peat and stand yield characteristics, particularly between ombrotrophic and oligominerotrophic sites, i.e. between site quality classes V and IV.Although the division between Vatkg and Ptkg based on the vegetation is diffuse, there are differences in environmental variables between these transformed sites.For example, the mean annual stand volume increment, degree of humification, concentrations and amounts of N, P, K and S were higher in Ptkg (IV) than in Vatkg (V).However, there are in fact no differences in the alpha diversity between these classes.But, as expected, the alpha diversity on herb-rich sites (II) is high compared to that on the other site types.
In general, the average amounts of nutrients in surface peat (0-20 cm) are at about the same level as those reported earlier for corresponding sites.On fertile sites there is somewhat more P than the values presented in other studies.High amounts of K and Mg in some categories are caused by a few high values.Almost all the sites were fertilized, as is the case for a considerable part of drained mires in Finland.The large within-type variation in nutrient concentrations and amounts is also typical of peatlands.
In addition to the site quality classes, a considerable amount of information about the tree stands, peat properties and vegetational diversity was connected to the separation of the succession phases, in this study transforming (mu) vs transformed (tkg) sites.Thus, distinguishing the succession phases from each other appears to be reasonable by their characteristics.It should also be borne in mind that the possible different assemblage of original mire site types in different succession phases -even of the same quality class -may also have had an effect on the observed differences in environmental sample variables.However, the differences between successional phases reflect the actual ecological situations prevailing in the field.Time since drainage does not well indicate the structure of the understorey vegetation due to variations in drainage efficiency (e.g.regressive succession), and to the fact that the rate of secondary succession is strongly affected by site type fertility, together with the rate of post-drainage stand increment.
On the basis of comparisons between numerical clustering and precise field classification, it appears that the (more or less) subjective field classification might be unnecessarily detailed for classifying drained mires: most probably we can disregard the original mire site types in the transforming phase.Thus, the drained peatland forest types (Laine and Vasander 1990) or the site quality classes (Huikari et al. 1964) with the secondary succession phases, as well as additional criteria, are needed.The peatland forest types and site quality classes with succession phases are usable as a framework for generalization, e.g. when comparing the results between different studies.It is more difficult to compare numerical units directly with the results of other studies.However, the numerical methods bring supplementary objectivity to the classification analyses and to the presentation of the results.They can be used as tools, e.g. in order to pay more attention to certain species and abundance criteria when specifying and identifying more traditional vegetation types.
A need for more accurate descriptions and instructions for the classification of transformed sites is apparent.The vegetation descriptions, notably for the aapa mire zone, are so far insufficient to canonize properly differential plant species for tkg types (cf.also Reinikainen 1994).More attention has to be paid to the indicator value, e.g.abundance criteria, of differentiating species.This especially concerns the indicator species of TWINSPAN: the site discriminants are often constant or dominant species.Some species of this kind (e.g.Eriophorum vaginatum, Carex globularis, Vaccinium myrtillus, Hylocomium splendens, Dicranum majus) were discussed with respect to suitability for classification.In the future, more emphasis should be paid to the nature of the species response to environmental factor gradients (e.g.Mäkipää 1999).
More information is needed about the consequences of subsidence, compaction and mineralization of peat due to long-lasting drainage on different sites and in different climatological areas.To what extent the original mire site type, together with additional qualifiers, predicts the long-term timber productivity is still a current issue (Hökkä and Penttilä 1999).However, it is obvious that the importance attributed to the original mire site type should be reduced while characteristics of the actual vegetation should be more strongly emphasized in the classification of drained mires.Our knowledge of the relationships between the actual understorey vegetation and site characteristics on old drainage areas is still insufficient.What are the effects of different methods of timber harvesting on the understorey vegetation?The effects of competition between vegetation and tree seedlings after harvesting and stand regeneration are also unknown.

Appendix 1.
Vegetation on the sample plots by TWINSPAN group.Species abundance is according to octave scaling (see Fig. 5).

Fig. 1 .
Fig. 1.Location of the study area and the border (B)between eccentric bogs and the southern aapa mire zone(Atlas of Finland 1988).

Fig. 2 .
Fig. 2. Location of the vegetation quadrats and the peat sampling points on the sample plots.

Table 3 .
The mean values (and standard deviations) of bulk density (D b ) and degree of humification (H v.Post ) in the 0-10 and 10-20 cm peat layers by site quality class and succession phase.

Table 4 .
The mean values (and standard deviations) of pH and macronutrient concentrations (mg g -1 ) in the 0-10 and 10-20 cm peat layers by site quality class and succession phase.