INTRODUCTION

Peatlands are limited to areas with humid climate or local water excess (Joosten et al. 2017). At various stages during the development of peatlands, their vegetation communities undergo successional changes which are partly controlled by autogenic processes related to environmental changes caused by the increasing thickness of the peat layer. While multiple potential trajectories for succession are recognised (Walker et al. 2010), the typical successional trajectory of northern peatlands is from minerotrophic fen towards ombrotrophic bog (Klinger & Short 1996, Hughes & Dumayne-Peaty 2002). Any specific peatland trajectory arises from external forcing factors (e.g. climate), boundary conditions (e.g. drainage basin characteristics) and internal or autogenic processes (Charman 2002). However, successional change seems to be extremely slow or stagnant in some regions; for example, in the aapa mire zone of northern Eurasia (Väliranta et al. 2017) which is characterised by high spring-season water excess due to a combination of snowpack melt, catchment size and high effective precipitation minus evapotranspiration (P -ET) in spring/early summer (Ruuhijärvi 1982). Solantie (1986) places the southern limit of aapa mire occurrence at the latitude where June ET exceeds precipitation.

Ecotones are transition zones between ecosystem types and are considered to be highly vulnerable to climate change (Allen & Breshears 1998, Neilson 1993). Within the ecotone between aapa mires and raised bogs the persistence of aapa mires relies on local water excess; in other words, the ratio of catchment to peatland area needs to exceed a certain threshold. Accordingly, when a connection delivering catchment runoff to an aapa mire located within the southern ecotone of the aapa mire zone was disrupted, a vegetation transition towards ombrotrophication occurred in the mire within a few decades (Tahvanainen 2011). Peatland regime shifts or state changes other than fen-bog transitions are also associated with changes in water table; for example, the transition from non-forested to forested peatland (Laine et al. 1995). However, little is known about how the nature, magnitude and/or duration of a hydrological shift relates to a regime shift.

In a few parts of the world, peatland initiation via primary paludification still occurs, and small altitudinal and hydrological differences on newly exposed land determine whether the succession will proceed towards peatland or forest (Laine et al. 2016). Areas that become peatlands (Tuittila et al. 2013) are termed primary mires (Joosten et al. 2017), and their peat layers are not yet thick enough to allow autogenic hydrological regulation (e.g. Leppälä et al. 2011). On the Finnish coast of the Gulf of Bothnia, where postglacial rebound provides new land for ecosystem development, newly exposed wet parts of the landscape have been intentionally drained since the 1960s-1980s in order to direct succession towards forest instead of peatland. These areas are naturally poorly drained due to low, flat relief and relict shoreline ridges, but their sandy soils are prone to percolation and may therefore readily dry out, i.e. the amplitude of water table fluctuations may be large (Rehell & Heikkilä 2009, Laine et al. 2016). The low (2 m a.s.l.) flat terrain gives a poor prognosis for drainage success although, on the other hand, the poor hydrological self-regulation of the thin organic layers of primary mires means they are responsive even slight forcing of their hydrology, which is completely governed by basin characteristics (including any ditches). Therefore, these sites are particularly sensitive to alterations of drainage. Within this coastal landscape various stages of drainage can be found: undrained mires in early phases of development towards peatlands, drained sites with developing tree stands, and unsuccessfully drained sites not supporting tree encroachment. We have previously studied both the natural successional development of peatlands in this setting along the 10 km spatial continuum from the seashore to a 3000year-old bog, and the effects of forestry drainage and restoration on vegetation structure and greenhouse gas emissions in the youngest sites. We found that the current vegetation assemblages formed a continuum from minerotrophic to ombrotrophic plant communities, consistent with the past succession revealed by vertical peat sequences at the older sites (Tuittila et al. 2013); and that vegetation structure and greenhouse gas emissions were rather resilient to environmental changes, i.e. restoration rather quickly reinstated functions typical of undrained conditions (Laine et al. 2016(Laine et al. , 2019)).

Here we analyse long-term water table data in conjunction with plant community data collected from two undrained and two drained primary mires on the Finnish coast of the Gulf of Bothnia (Figure 1). Because the sites are located within the aapa mire region and in a natural environment it is likely that, if their central parts stay wet, (ombrotrophic) bog vegetation will develop only at their edges as mire succession proceeds, similar to 'true well-developed' aapa mires. We aim to study the mechanism of water table change required to divert the successional trajectory from open mire to forested habitats. The specific hydrological regime of the sites makes it possible to quantify the threshold of water-level drawdown, as well as the seasonality of drainage effectiveness, that is most likely to causally connect to drainage succession.

METHODS

The study was carried out on the Finnish land-uplift coast of the Gulf of Bothnia in Tauvo, Siikajoki (64° 48′ N, 24° 38′ E, Figure 1). The 30-year average precipitation and mean annual temperature are 539 mm and 2.6 °C, respectively, and the length of growing season is 150 days (Revonlahti, Siikajoki, 64° 41′ N, 25° 05′ E, 48 m a.s.l.; Finnish Meteorological Institute). We selected four primary mires, two undrained (Undrained mire, tree stand volume < 1 m 3 ha -1 ) and two that have been drained for forestry purposes since the 1960s. At one site, drainage has led to tree stand formation (Drained forest, tree stand volume 106 m 3 ha -1 ); at the other drained site, drainage has been much less successful in terms of forestry (Poorly forested drained mire, tree stand volume 37 m 3 ha -1 ). Before drainage, the sites were all coastal primary mires in small (~0.5-3 ha) depressions between sand dunes, formed on nutrient-poor sandy soil 100-200 years ago via primary paludification of newly exposed uplifted land. The sites are now 1.5-2 m a.s.l. and have a 5-10 cm thick peat layer. The field layer vegetation of the Undrained mires is composed of graminoids (Carex nigra, C. canescens) and forbs (Comarum palustre, Equisetum fluviatile, Peucedanum palustre), while the moss layer is composed mainly of Warnstorfia exannulata and Calliergon species, with scattered patches of Sphagnum mosses (e.g. S. squarrosum, S. fimbriatum). At the drained sites, the field layer vegetation is dominated by shrubs (Vaccinium uliginosum, V. vitis-idaea, Salix repens) and the moss layer by feather mosses (Pleurozium shreberii, Polytrichum commune), although at the Poorly forested drained mire sedges (Carex nigra, C. canescens) are also abundant. More detailed site descriptions are given in Laine et al. (2016Laine et al. ( , 2019)).

In the drained sites, the vegetation composition of field and ground layers (not including trees) was surveyed at 13-14 points using 1 × 1 m plots along transects covering the sites during the years 2006, 2009 and 2013. In undrained sites vegetation was surveyed in 2007, 2010 and 2013 at six 60 × 60 cm plots. Projection cover (%) of vascular plants and moss species was quantified visually. Water table depth (WT) was measured in May, August, and October, as the distance between water table and moss carpet surface or soil surface where mosses did not occur. At each site WT was measured in 6-14 dipwells (perforated plastic tubes ~ 3 cm i. value of WT recorded at the site during that measurement round (> 60 cm below surface) was adopted in place of any missing values. Multivariate analyses, carried out with Canoco 5.02, were used to analyse the vegetation composition data; analysis details are reported in the Figure legends.

RESULTS

The sample plots from the Poorly forested drained mire scattered widely over the ordination space but could be divided into three groups along CCA Axis 1 that separated Undrained and Drained forest sites as constrained by canonical function of drainage state (0/1) (Figure 2). The vegetation composition of Group A of the Poorly forested drained mire grouped closely with the Undrained mire plots, indicating that vegetation had not gone through a significant successional change despite the long-term ditching. Therefore, this group is regarded as a "Failed drainage" group. The vegetation composition of Group C was closely related to the Drained forest plots and had experienced the intended regime shift towards a forest ecosystem. It is therefore classified as a "Successful drainage" group. The plots of Group B did not align within either the Drained forest or Undrained plots and we refer to these as Transitional plots.

During the study period WT varied from 68 cm below to 31 cm above the moss/soil surface (Figure 3). The deepest WT values were measured during August and at the Drained forest site, while May and the Undrained sites were generally wettest. The vegetation differences of the Failed drainage and Successful drainage groups within the Poorly forested drainage site prompted us to further investigate the differences in their WT patterns. The largest differences in WT between the failed and successful drainage groups were observed during May, when the Failed drainage plots were always inundated (Figure 3a). During August and October, the differences were less distinct (Figures 3b,3c). During the study period, the relative seasonal WT differences between Successful drainage and Failed drainage were on average 12, 6 and 11 cm in May (Figure 3a), August (Figure 3b) and October, respectively (Figure 2c) and these differences did not change markedly over the years.

When we further compared average May WT with the vegetation composition of the sample plots (Figure 4), we observed a clear division; in Failed drainage plots the May water table was above the soil surface (WT = 8 to 13 cm), giving similar readings to the Undrained plots (WT = 0 to 20 cm), whereas in Successful drainage plots it was mostly below the soil surface (WT = 3 to -13 cm) as in Drained Forest plots (WT = 1 to -27 cm) (Figure 4). During August and October the water table in Failed drainage plots was lower than in Undrained plots but remained shallower than in Successful drainage plots (results not shown).

DISCUSSION

The data collected over the course of ten years, although limited to four sites, appear to support the idea that small seasonal changes in water table depth may initiate a regime shift in peatland vegetation. Spring flooding rather than summer water table drawdown appeared to be crucial for controlling vegetation state shifts in primary mire.

It is well established the peatland drainage leads to a transition to forest vegetation (Laine et al. 1995, Minkkinen et al. 1999, Talbot et al. 2010). At our study sites, drainage ditches were installed ~ 50 years before we started our monitoring. The vegetation of Drained forest and Successful drainage sites clearly differed from that of Undrained sites, and the vegetation states at all sites remained stable during our ten-year monitoring period. The tree cover of the drained sites was established during the first decades following ditching, probably due to lowered water table, as has been observed in peatlands with thicker peat layers (Laiho & Laine 1997). At the Failed drainage plots, where vegetation remained similar to that of Undrained sites after 60 years of drainage, there were no signs that they had been forested at any time prior to or since drainage ditches were installed.

In our study, the water table in the Failed drainage plots consistently drops well below the soil surface in summer (August), as it does in all other plots except Undrained ones. Yet the Failed drainage plots are not transitioning to forest. Interannual variability in WT is most pronounced in August, particularly for the Failed drainage and Successful drainage plots, which show similar patterns (Figure 3). Since Failed drainage and Successful drainage plots had different drainage outcomes, we conclude that interannual variability did not play a major role in the effects of WT on the system. It appears that low summer WT alone is not a sufficient driving force for secondary forest succession if spring flooding continues. This is consistent with a palaeological study in a northern aapa mire by Väliranta et al. (2017), who concluded that the reason why the dry mid-Holocene climate did not trigger bog development in a northern Finnish fen, but instead produced a dramatic slowing of peat accumulation (Mathijssen et al. 2014), was that the spring flood patterns did not change in this region. Spring season flooding seems to play a stabilising role.

We propose that the difference in WT observed in May, from 5 cm (inundated) to -5 cm (shallow), has a greater effect on vegetation, particularly for tree stand establishment, than the equal difference from -45 cm to -55 cm observed in August, as most of the local tree species (e.g. Picea abies and Pinus sylvestris) have low tolerance to flooding, particularly in late spring, when their seasonal growth has just started (Glenz et al. 2006, Repo et al. 2016, Wang et al. 2016). For trees such as Pinus sylvestris that have deep rooting systems, extended flooding is likely to be more disruptive than extended periods of deep water table during summer. In our sites WT was seldom lower than -60 cm (our measurement limit due to the lengths of the dipwells), and in the humid climate of Finland forestry drainage that commonly leads to deeper water table has never been shown to be too powerful for forest growth (Paavilainen & Päivänen 1995). October (Autumn), typically outside the growing season in this region, is ecophysiologically less important to vegetation stability and state transitions than May (Spring) or August (Summer). Overall, we propose that spring flooding is a crucial modifier of vegetation transition for these ecosystems or, to be more precise, a factor that prevents regime shift.

Climate change scenarios predict changes in the hydrological cycle driven by increased temperature and evapotranspiration and changes in atmospheric circulation (IPCC 2014). In many northern areas like Finland, winter precipitation is expected to increase, while summer conditions are likely to be dryer in the future (Pal et al. 2004, Lehtonen et al. 2014). Earlier predictions suggested an 8-22 cm drop in water tables for northern peatlands (Gorham 1991, Roulet et al. 1992) but these estimates did not consider changes in the seasonality of precipitation. The timing of increased precipitation is important as it affects the intensity and duration of spring flooding. Thus, a focus on growing season conditions (only) may be misplaced.

Our data introduce a likely mechanism for the control of vegetation regime shifts in northern peatlands by water table, with time of year being as important a factor as the magnitude of change. Because climate change effects on peatlands are most likely to be mediated via changes in hydrology and water table level (e.g. Munir et al. 2015, Buttler et al. 2015, Mäkiranta et al. 2018), our study highlights the need for a more thorough investigation of seasonal variability in controlling factors. Northern boreal and subarctic fens are widespread ecosystems covering large areas in Canada, Fennoscandia and Russia (Joosten et al. 2017), where hydrological changes have the capacity to alter the ecosystem regime.