Macroclimate and local hydrological regime as drivers of fen vegetation patterns in Tierra del Fuego (Argentina)

In southern South America, a sharp macroclimatic gradient is found, triggered by the Pacific oceanic influence through the Andes. Mires are substantial landscape units along the gradient, where they play varying roles through the area and include varying plant communities. In this study, we analysed the relationship between vegetation and main abiotic factors in the minerotrophic mires (fens) at two different scales.

Cordillera experiences water surplus, and the balance is roughly zero in the ecotone zone. These climatic gradients entail contrasting shifts in the structure and composition of mires and in the vegetation matrix (Collantes et al., 2009;Grootjans, Iturraspe, Lanting, Fritz, & Joosten, 2010). In addition to water-related factors, growing thermal continentality north-eastwards also affects the vegetation patterns (Kleinebecker, Hoelzel, & Vogel, 2007).
Mires are substantial landscape units all over Tierra del Fuego, although they play varying roles through the area. At the rainiest south areas, mires and forests are the two main landscape units (Blanco & de la Balze, 2004;Pisano, 1977), whereas in the Magellanic Steppe, mires cover much smaller areas, inside grassland landscapes (Collantes et al., 2009). Within the mire systems, ombrotrophic Sphagnum bogs only develop where precipitation exceeds evapotranspiration. Thus, they thrive in the Andean Cordillera, vanish through the ecotone zone, and do not occur in the Steppe, where mires are entirely minerotrophic (Blanco & de la Balze, 2004;Collantes & Faggi, 1999;Grootjans et al., 2010). This vegetation gradient strongly drives the distribution of mire biota from landscape to regional scales, given the strong differences in structure and functioning between minerotrophic and ombrotrophic vegetation units (Bragazza, Rydin, & Gerdol, 2005;Malmer, 1986;Sjörs & Gunnarsson, 2002).
Studies aimed at mire vegetation in Tierra del Fuego are scarce.
However, the studies made in Tierra del Fuego do not include information on the fen vegetation of the Cordillera region and very rarely of the ecotone zone (from where a single report from Roig, 2001, is available). These studies (Grootjans et al., 2010;Grootjans, Iturraspe, Fritz, Moen, & Joosten, 2014) report on the fens occurring abundantly in slope springs and at the river banks. They evaluate the relationship between environmental variables and mire vegetation at the region scale (over tens of kilometres) but do not include accurate analyses at the mire scale. Thus, they did not analyse possible shifts in the relationship between environmental variables and vegetation along the pronounced bioclimatic gradients found in Tierra del Fuego. Such shifts have been evidenced in the European and North American mires, where the relevance of the environmental variables driving vegetation patterns varies from one biogeographical region to other and at different scales (Belland & Vitt, 1995;Bragazza et al., 2005;Økland et al., 2001).
In this study we aimed (a) to describe the main ecological features driving the variation in theTierra del Fuego fens and to evaluate the relationship between fen vegetation and main abiotic factors and (b) to understand the patterns observed in the distribution, diversity, and singularity of vascular plants at two complementary scales, that is, at landscape and regional levels, namely, within fen mosaics and between biogeographical regions. To do so, we characterized the visually distinct habitats by means of monitoring the hydrological regime through the growing season and analysing the table water, the peat substrata, and the species composition of the vascular vegetation at the optimal development period. This was done in three study sites set along a sharp climatic gradient in Isla Grande, Tierra del Fuego, where the fen vegetation plays varying ecological roles. Based on previous studies, we expected macroclimatic differences to be the main regional driver of vegetation shifts and the poor-rich gradient to drive floristic change at local scale.

| Study sites
We chose three medium-sized wetland areas along the climatic gradient going from wet oceanic to dry continental. These study sites were good examples of the two contrasting biogeographical regions recognized and their ecotone ( Figure 1) and were characterized in ecohydrological terms by Grootjans et al. (2010). The sites are as follows: • Ushuaia: The site lies in Andorra Valley (54°45′40″S, 68°18′8″W; 190-200 m a.s.l.), a RAMSAR site in a suburban area of the city.

| Chemical analyses
At early February 2005, we measured the pH and electric conductivity of the water in each pipe, with a portable WTW TetraCon 325 sensor.
At the same time, we took water and peat samples in a wide selection of the sampling points, stratified according to their situation in the water table gradient. This resulted in 22 water samples from Ushuaia, 20 from Tolhuin, and nine from Río Grande and in 10 peat samples from Ushuaia, 14 from Tolhuin, and 10 from Río Grande. Water samples, taken from the pipes, were filtered (through 0.20-μm pore filters) and acidulated, to be analysed for their total relevant ionic contents (see the ions analysed in Table 1) by means of inductively coupled plasma mass spectrometry. Peat samples were the uppermost 15 cm of cores obtained with a soil drill. They were analysed in terms of pH (in dilution 1:2.5 between soil and distilled water), total carbon percentage, total nitrogen percentage (through the Kjeldahl method), and the concentration of Ca, Mg, K, and Na (by means of an acid digestion, followed by inductively coupled plasma mass spectrometry).

| Statistical analyses
We performed a classification of the 75 vegetation relevés based on their species composition, to evaluate their floristic relationships. To do so, we previously transformed the Domin values to cover percent values and then these to their logarithms. We then set a Hellinger distance matrix to produce a beta-flexible hierarchical classification (with β = −0.25, according to Borcard, Gillet, & Legendre, 2011). We pruned the classification tree progressively to obtain clustering levels going from two to 15 groups (or vegetation types). We followed the Mantel criterion (Borcard et al., 2011) and chose the clustering level that kept the best distance relationships of the original matrix. However, we rejected those classifications with vegetation groups with less than three relevés in order to have enough samples per group. For each of the resulting group and for each site, we calculated the phi fidelity index of all plant species (Chytrý, Tichý, Holt, & Botta-Dukát, 2002).
Then, to compare the distribution of environmental variables between Note. The number of species in 16 relevés was calculated by means of rarefaction, and β diversity was evaluated as the average dispersion of the samples from the centroid of the study site; different letters in the same row indicate significant differences in the Kruskal-Wallis test. Phi fidelity values are given for the ecologically most significant species (φ > 0.3 and p < .01).
vegetation groups, we performed pairwise multiple comparisons through Kruskal-Wallis tests.
To obtain the main floristic patterns, we ordinated the vegetation relevés through a principal component analysis (PCA) based on the species matrix, where species cover values were standardized emulating Hellinger distances (Legendre & Gallagher, 2001). After that, we produced three detrended canonical analyses (DCAs; Borcard et al., 2011), each for each study site, to reveal their particular floristic patterns and their main axis of variation. The relationships between the vegetation ordination and the habitat conditions were analysed by means of vector fitting the environmental variables (Borcard et al., 2011). We also calculated the Pearson correlation coefficient between the environmental variables and between them and the first axis of the DCA.
We obtained and compared the species accumulation function at the different study sites according to the proposal of Cayuela, Gotelli, and Colwell (2015). The accumulation curves (based on the vegetation plots) were created through rarefaction and were used to contrast two null hypotheses: ecological (H0eco) and biogeographical (H0biog). Agreement with H0eco would mean that the vegetation samples corresponded to a sole plant community, thus that any difference between them (concerning species richness, composition, or relative abundance) was due to sampling bias. Contrarily, agreement with H0biog would mean that the samples corresponded to distinct plant communities differing in their species composition, even if sharing similar species richness and species abundance distribution. Moreover, we evaluated the species β diversity for each study site. We used the Hellinger distance matrix of the relevés to calculate, for each relevé group, the average distance between each relevé and the centroid of the group, which is a measure of its β diversity (Anderson, Ellingsen, & McArdle, 2006).

| Vegetation types and study sites
Following the Mantel criterion, we chose the classification of the 75 vegetation relevés into six vegetation types ( Figure A1). Both six-and seven-cluster classifications had equivalent Pearson r correlation (.640 and .642, respectively), but the seven-cluster classification created a group with only two samples, which we considered not acceptable.
The types resulted equally distributed among the three sites, namely, two types per site ( Figure 2). Concretely, Types 1 and 2 were found in Ushuaia (except for one relevé of Type 1 taken in Tolhuin); Types 3 and 4 corresponded to Tolhuin (but one relevé came from Río Grande); and Types 5 and 6 corresponded exclusively to Río Grande. All vegetation types corresponded to minerotrophic mires, where dominant plants were generally different Cyperaceae (or Poaceae) species (Table A1) and locally the moss S. magellanicum (Type 4, at Tolhuin).
Correspondingly, each study site was characterized by a fair number of faithful species (with phi fidelity values higher than 0.3 and p value lower than .01) and exclusive species (only found at a given study site; Table 2). However, the three study sites gave very similar figures in terms of total species richness and faithful and exclusive species. Indeed, the diversity patterns were rather uniform between sites, as revealed by almost coincident numbers of species in 16 plots (calculated through cumulative curves after rarefaction of vegetation plots) and species density per area ( Figure 3). Only the β diversity gave significant differences between Tolhuin-with higher floristic dispersion among plots-and the other two sites. In relation to biodiversity structure, we rejected the null ecological hypothesis (p = .002), whereas we accepted the null biogeographical hypothesis (p = .387).

| Ecological variation at regional scale
The sites studied exhibited noticeable differences in most of the environmental variables (Table 3). Soil variables contributed the less to distinguish the study sites, compared with water chemistry and hydrological regime. Almost all hydrological variables were significantly different between sites, and most of them were strongly correlated (Tables A2 and A3). The two vegetation types within each site differed in less environmental variables, that is, seven in Ushuaia and eight in Río Grande and Tolhuin (Table 4). Globally, the hydrological variables gave the highest differences between vegetation types (and lower p values) in the Kruskal-Wallis test (Table 4). Only three variables gave significant difference neither between sites nor between types: flood, the carbon nitrogen ratio (pCOt/pNt) of the peat, and the peat pH (ppH).
Referring to the hydrological regime, the water table depth was clearly different between sites (cf. maximum, minimum, and mean depths in Table 4), although differences within sites were also notice-

| Ecological variation at local scale
The main floristic gradient in each study site was evidenced as the first axe in each of the corresponding DCA ( Figure 4). In Ushuaia, six variables proved to be significant in the vector fitting (Table A4) (Table A5). So, we plotted in Figure 4 only the one most correlated to the first axis. Following the same criterion that in Ushuaia, the number of significant variables (Tables A6 and A8) (Kleinebecker et al., 2007;Vitt, 2006). As a result, the fen of the Steppe is more contrasting through the local landscape, because the plant communities bordering springs and rivulets were kept flooded or nearly to through summer.
The pH values and the Ca contents measured in peat and water correspond to fens from slightly acidic to slightly alkaline (Sjörs & Gunnarsson, 2002;Wheeler & Proctor, 2000). Our sampling points  (Chee & Vitt, 1989;Sjörs, 1952;Tahvanainen, 2004). This classification is based on the intermediate levels of Ca-ranging from 18 to 28 ppm, with no significant differences between sites-and the moderately acidic pH values-from 5.6 to 6.2-through the study sites.
However, the values indicated more acidity in Río Grande than in Ushuaia and Tolhuin. The higher levels of sulfur in Río Grande might be the cause of this moderate decrease in pH (Gunnarson, 2000).
The high values of electric conductivity-from 197 to 290 μS/cm 2are evidence of the strong geogenous influence in all three fens. This also includes the sphagnum low hummocks (Vegetation Type 4), which are clearly nonombrotrophic. Fertility was higher in Río Grande than in the other systems, as indicated by the soil and water contents of P and K-but not by the total N. This regional difference in fertility is likely related to higher grazing use at the Río Grande steppe than in the other two systems. Moreover, the Na content increases northwards along the regional gradient. Beyond the distance to the see, this may be partly due to an increasing influence of air-borne marine NaCl deposited by the prevailing western winds. In addition, the fen of Río

| Flora and vegetation types, and the singularity of the sphagnum low hummocks
The six vegetation types found evidenced varying similarity levels among them, as shows Figure 2 (see also Figure A1 and  (Bragazza et al., 2005;Grootjans et al., 2014;Tahvanainen, 2004). Moreover, the PCA ordination of vegetation types and sites do not follow the macroclimatic gradient. In addition to the singularity of the sphagnum low hummocks, this should be mostly due to the similarity between the fens of Type 1 (Ushuaia) and Type 5 (Río Grande).
The ecological singularity of the sphagnum low hummocks also affects their local diversity (Table 4), which results into lower α and γ diversities (total and rarefied values) with respect to the other vegetation types. This comes from the stressing conditions for vascular plants enhanced by the sphagnum hummocks (Kleinebecker et al., 2010), as is also known from the mires in the Northern Hemisphere (Hájková & Hájek, 2004;Malmer, 1986).

| The macroclimatic or regional gradient
The chief floristic differences found between vegetation types were due to the site factor. A fair percentage (around 25%; Table 2) of the flora in each fen was exclusive of that site, a notable finding given the proximity between the sites in geographical terms. This local singularity is evidenced in the PCA ordination, where the sites appeared as three clouds, and in the classification of relevés, because the vegetation types obtained where consistent with the sites-two types from each site-with the only exception of two of the 75 relevés.
This regional pattern might respond to physical and chemical variables or to the continentality and moisture gradients. The measured environmental variables are known to promote or to hamper distinct mire species. This refers mostly to the acidity (pH), the electric conductivity (and Ca and base content), the fertility, and the water table depth (Filipová et al., 2013;Økland et al., 2001;Wheeler & Proctor, 2000).
Among these variables, the most influencing to the plant species variation in our study were the mean water table depth and the variation in the fertility indicators. The site Río Grande stands out in both variables, namely, for lower water table (Table 3) and for the highest fertility (K, pK, P, and pNt variables), most probably related to persistent sheep grazing, which do not occur in the other sites. In this regard, the short pH gradient is likely to be related to a rather homogenous bedrock composition.
The other variables are fair less determinant to mire vegetation according to most studies, except for the Na content. However, Na is a major vegetation driver at higher concentrations than those found in our study, even at Río Grande (cf. Collantes et al., 2009;Filipová et al., 2013).
The continentality and moisture gradients affect the regional distribution of most ecosystems in Tierra del Fuego through macroclimatic shifts, particularly forests and grasslands (Collantes & Faggi, 1999;Frederiksen, 1988;Pisano, 1977;Roig et al., 1985). According to that, the harsher bioclimate in the Magellanic Steppe would hamper the occurrence of several plant species, including most apparently trees.
The main bioclimatic factor driving the vegetation patterns is the relationship between potential evapotranspiration and precipitation, according to Grootjans et al. (2010). In the case of mires, the fundamental border between water surplus and deficit is the key for understanding the distribution of ombrotrophic peatlands and may be also relevant in the particular patterns found within minerotrophic systems. Thus, in fens under drier bioclimates, the seasonal drying out of the superficial peat enhances its oxygenation and mineralization, which strongly influences soil chemistry. Therefore, drier bioclimatic conditions promote both the seasonal variation in the water table depth and, within a given fen, higher contrast among different fen points. This may be the case in our steppic fen system, where the soil in marginal areas dries out through summer. However, local factors, such as belowground drainage, can alter hydrological regimes elsewhere and may be also contributing to water drawdown in the Río Grande fen.
In addition to ecological factors, the historical influence of strong bioclimatic differences between cold steppic landscapes in the north and more temperate woodlands in the south has shaped the evolution, diversification, and establishment of particular flora and vegetation in the two biogeographical regions (Collantes et al., 2009;Pisano, 1977;Roig et al., 1985).  (Cayuela et al., 2015). Tierra del Fuego may have been place for this process, because the different biogeographical areas harbour different regional floristic pools, and even the corresponding mire systems harbour fair distinct local floristic pools.

| The landscape (local) gradients
The DCA ordinations of the plant relevés express that most of the species variation respond to the varying water table depth and to its dynamics across study sites (Figure 4). This is also expressed by the cluster distribution of two plant community types in each fen, from which one thrives in habitats with significantly deeper water table than the other does. Therefore, the hydrological regime stands out as the first ecological gradient at the fen scale. As far as we know, our data are the first evidence of the protruding role played by the water table depth and its dynamics in the distribution patterns of vascular plants in the fens of Tierra del Fuego (Collantes et al., 2009;Filipová et al., 2013). In the Northern Hemisphere, some studies proved the water table gradient to be the main driver of fen vegetation (Pérez-Haase & Ninot, 2017;Schenková et al., 2014). The relative weight of the hydrological variables seems to vary depending of the space scale considered in different studies, and this may be related to the controversy generated when different areas have been compared (Økland et al., 2001;Wheeler & Proctor, 2000).
Beyond the preponderance of the hydrology, the ecological gradients in Tolhuin and Río Grande are more complex than in Ushuaia.
The occurrence of the sphagnum hummocks in Tolhuin is cause for lengthening the pH, pNt, and maximum water level depth gradients ( Figure 4). Once initiated, sometimes by colonization of locally thicker peat, sphagnum hummocks have a strong influence on habitat characteristics, which result in more acidity, deeper water table, and lower fertility (Sjörs & Gunnarsson, 2002). In Río Grande, the soil contents of carbon and sodium are the most correlated variables with Axis 1 of the DCA, together with the hydrological variables. Given the relatively lower content of sodium in our samples, the low ecological relevance of soil carbon, and the importance of the water table depth in the species distribution in mires (Kleinebecker et al., 2007;Økland, 1990;Vitt, 2006), the latter factor stands out as the main driver of species variation also in the Río Grande fen. Indeed, the total percentage of carbon in the peat is most probably driven by the dynamics of the water table, which strongly influences the organic matter mineralization (Moore & Basiliko, 2006;Rydin & Jeglum, 2006).

| Concluding remarks
Our data point that macroclimate variation at regional scale leads to complex responses in mires, whereas variations in the water table depth are the main drivers of vegetation patterns at the landscape scale. The poor-rich gradient, which many authors found to be the most important driver of vegetation, is here correlated to the macroclimatic gradient, and against our expectation, it is a secondary driver at local scale. The short pH gradient measured in our three fens may have contributed to the leading role of water table dynamics.
On the one hand, the macroclimatic gradient driving the zonal vegetation in Tierra del Fuego influences the vegetation patterns of the fens-even if this vegetation category is chiefly azonal. From the last Quaternary glaciation onwards, the macroclimatic gradient has shaped the regional floristic pools and hence the local pools. The gradient of increasing water deficit northwards would filter the plant species able to settle on the different mires (Boelcke, Moore, & Roig, 1985;Collantes et al., 2009;Grootjans et al., 2010).

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are openly available in   Note. Grey shadows indicate species fidelity (phi) index higher than 0.3 to a given vegetation type.   Note. Their correlation to DCA Axis 1 is given in the last column.