Plant composition and diversity at edges in a semi-natural forest–grassland mosaic

As key components of landscapes, edges have received considerable scientific attention in anthropogenic ecosystems. However, edges in natural and semi-natural forest–grassland mosaics have received less attention, despite the fact that they cover a considerable proportion of these mosaic ecosystems. We studied forest edges in a semi-natural forest–grassland mosaic ecosystem of the Samobor Mountains (Croatia). Our aim was to compare the species composition, diversity and ecological indicator values of forest edges to those of the interior parts of the adjacent forest and grassland habitats. The vegetation was studied in 80 plots established in forest patch interiors, north-facing forest edges, south-facing forest edges and grassland interiors. We found that edges had a unique species composition, containing species from both the forest and the grassland interiors plus their own edge-related species (i.e. species that significantly preferred the edge habitat). These local edge-related species did not correspond to regionally-identified edge-related species. Compared to the forest and the grassland interiors, we revealed increased species richness in north-facing edges but not in south-facing edges. The mean light availability and nutrient supply indicator values of the edges were intermediate between those of the forest interiors and the grasslands. The mean soil moisture indicator values of the edges were similar to those of the grasslands. Our results show that edges form a unique component of forest–grassland mosaics, and they contribute considerably to landscape complexity, which should be taken into account during conservation decisions and habitat management.


Introduction
Edges have received considerable attention in the ecological literature, as they are key structural and functional components of landscapes (Risser 1995;Cadenasso et al. 2003a;Ries et al. 2004Ries et al. , 2017Yarrow and Marín 2007;Hufkens et al. 2009;Kolasa 2014). Edges influence the flow of organisms, materials and energy (Cadenasso et al. 2003b); influence population interactions (Fagan et al. 1999) and may serve as habitats or conduits for many species (Forman and Moore 1992).
Edges situated between forest and grassland ecosystems belong to the most conspicuous edge types (Forman and Moore 1992;Risser 1995;Cadenasso et al. 2003b). With accelerating forest fragmentation and an associated increase in edge proportion, forest edges have received considerable scientific attention in anthropogenic ecosystems (Merriam and Wegner 1992;Harper et al. 2005;Peters et al. 2006;Tokuoka et al. 2011;Dodonov et al. 2013;Haddad et al. 2015). For example, forest edges adjacent to clear-cuts (e.g. Chen et al. 1992;Euskirchen et al. 2001;Burton 2002;Harper and Macdonald 2002) or arable fields (e.g. Fraver 1994;Honnay et al. 2002;Devlaeminck et al. 2005) have been in the focus of ecological research. However, forest edges also play an important role in natural and semi-natural ecosystems, yet edges in these systems have received less attention in previous studies (but see Müller et al. 2012;Ibanez et al. 2013;Dislich and Mantovani 2016;Harper et al. 2018).
In Central and Southeast Europe, mosaic habitats consisting of alternating forest and grassland patches are an important component of landscapes, especially under relatively harsh conditions such as sand dunes and south-facing rocky slopes (Horvat et al. 1974;Ö llerer 2014;Erd} os et al. 2018a). A considerable proportion of these mosaic ecosystems, including extensively used pastures and pastures that have been abandoned recently, are semi-natural, i.e. modified by human activity but still dominated by native species that establish and reproduce spontaneously (Sjörs 1986).
Semi-natural landscapes in general and extensive pastures in particular have an outstanding conservation importance as they contain a large diversity of plants, including several rare species (Horvat et al. 1974;Ellenberg 1988;Bergmeier et al. 2010). It seems highly likely that the notable species diversity and natural value of forest-grassland mosaics are strongly connected to their high habitat heterogeneity, i.e. the presence of structurally very different patches in small proximity (e.g. Erd} os et al. 2018b;Tölgyesi et al. 2018). Currently, however, the habitat heterogeneity of semi-natural forest-grassland mosaics is rapidly diminishing through different forms of homogenisation: overgrazing and intensification result in the disappearance of the forest component, while afforestation, the spread of invasive trees and the cessation of grazing threaten the survival of the grassland component (Bergmeier et al. 2010;Erd} os et al. 2018b). It seems certain that a better understanding of the importance of habitat heterogeneity could contribute to a more efficient conservation of these valuable ecosystems.
Extensive and recently abandoned pastures typically have a fine-scale mosaic, that is, both the forest and the grassland patches are small (Horvat et al. 1974;Bergmeier et al. 2010;Erd} os et al. 2011Erd} os et al. , 2018bBorhidi et al. 2012). Consequently, the proportion of edge habitats is considerable, and they may have a disproportionately high conservation importance (Kent et al. 1997).
Edges have been proposed to have their own characteristic species composition, supporting species from both habitat interiors plus so-called edge-related species (i.e. species that tend to be concentrated within habitat edges) (Odum 1971;di Castri and Hansen 1992;Risser 1995;Kent et al. 1997). This has important conservation implications, as edge-related species, if they exist, would undoubtedly increase species richness at the landscape and regional scales (Naiman et al. 1988). Unfortunately, field studies are scarce, and the majority of them did not use any significance test to identify edge-related species (Lloyd et al. 2000;Baker et al. 2002).
Edges with different orientations tend to differ regarding environmental conditions (Chen et al. 1995;Gehlhausen et al. 2000;Ries et al. 2004;Heithecker and Halpern 2007), and consequently, regarding species composition (Dierschke 1974), which may contribute to a further increase in landscape-or regional-scale diversity. However, these differences remain poorly understood in semi-natural forestgrassland mosaics.
From a nature conservation perspective, edges may also be extremely important components of landscapes because their habitat-scale diversity is sometimes expected to be greater than that of either of the two adjacent habitat interiors (Odum 1971;Pianka 1983;Risser 1995;Kent et al. 1997). However, Ries et al. (2017) pointed out that this should not be considered a general phenomenon. Van der Maarel (1990) suggested that only blurred edges with stable environmental conditions have higher diversity than patch interiors, while abrupt edges with fluctuating environmental conditions (i.e. strong microclimatic variations in time) have lower diversity. Some field evidence shows that edge diversity may be intermediate between the diversities of the two habitat interiors (Walker et al. 2003;Erd} os et al. 2011). The results of Łuczaj and Sadowska (1997) emphasise that edge diversity may vary considerably among different taxonomic groups. So far, generalisations have been rather difficult because of the limited number of case studies, especially for semi-natural systems (Harper et al. 2005;Kark and van Rensburg 2006).
In spite of the important role edges presumably play in these complex semi-natural mosaic ecosystems, their properties have been addressed by a surprisingly low number of studies. Our aim was to investigate forest edges in a semi-natural mosaic ecosystem with no current management activity, where small forest patches are embedded in a grassland matrix. We studied north-and south-facing forest edges in relation to the neighbouring forest and grassland habitats. Our specific questions were as follows: (1) Do edges have a specific species composition that differs from the habitat interiors? (2) Do edges possess their own edgerelated species that significantly prefer edge habitats? (3) Do edges have larger per plot species richness and Shannon diversity than forest and grassland interiors? (4) Is the proportion of phytosociological preference groups different between habitat interiors and edges? (5) Are the mean ecological indicator values of edges and habitat interiors different? (6) Are north-facing and south-facing edges different regarding the above characteristics?

Study area
Our study was conducted in the Samobor Mountains (northwest Croatia), which form a transition between the Alps, the Dinarides and the Carpathian Basin (Trinajstić 1995). We chose a south-facing slope (N45°48 0 02 00 , E15°38 0 30 00 ) west of the town of Samobor. The elevation is 370-410 m asl, the bedrock is dolomite and the soil is rendzina (Mayer and Vrbek 1995;Trinajstić 1995). The mean annual temperature in Samobor is 11°C, and the mean annual precipitation is 1015 mm, most of which falls in June and October-November (Mayer and Vrbek 1995). The natural vegetation of the study site consists of xeric forests. As a result of human impact (grazing and mowing), these forests have developed into a mosaic of xeric forest patches and dry grasslands (Horvat et al. 1974) (Fig. 1). The size of the forest patches usually varies between ca. 0.02 and 0.2 ha. Due to the high number of protected and rare species, these mosaics have extreme conservation importance but are diminishing as pastures and meadows are abandoned.
The study site was used as an extensively managed pasture, but grazing stopped in the 1980s. Currently there is no land-use or management activity.
Species names are used according to The Plant List (www.theplantlist.org), and plant community names follow the nomenclature of Trinajstić (2008).

Field work
Twenty forest patches were selected for the study. For all patches, four 2 m 9 1 m plots were established in the following arrangement, corresponding to four different habitats: one plot in the forest patch interior, one plot in the north-facing forest edge, one plot in the south-facing forest edge and one plot in the neighbouring grassland. We thus used a total of 80 plots (20 patches 9 4 habitats). The minimum distance between neighbouring patches was 75 m, while the distance between the neighbouring plots in the four habitats belonging to the same patch was 10-15 m. An edge was defined as the zone outside of the outermost tree trunks but still under the canopy. Forest edges in similar ecosystems are usually very narrow (Jakucs 1972;Erd} os et al. 2011Erd} os et al. , 2014; thus, using small and elongated plots ensured that the plots fit into the edges. The cover of all vascular plant species of all vegetation layers was visually estimated in May 2017. As the canopy layer was low (typically 4-5 m, sometimes even less), and it merged with the shrub layer, the shrub and the canopy layers were treated jointly in this work.

Data analysis
To study the compositional differences among the four habitats, detrended correspondence analysis (DCA) (Hill and Gauch 1980) was performed on square roottransformed cover scores. The analysis was carried out in the R environment (R Core Team 2018) using the 'decorana' function of the vegan package (Oksanen et al. 2018).
We prepared a Venn diagram to show the number of species that are restricted to a single habitat and the number of species that are present in two or more habitats. We used the online Venn diagram generator of the Bioinformatics and Systems Biology Group of the Department of Plant Systems Biology, Ghent University (https://bioinformatics.psb.ugent.be/ webtools/Venn/).
We also statistically identified diagnostic species, i.e. species that preferentially occur in certain habitats and are absent or rare in the other habitats (Barkman 1989). For this purpose, we used the phi coefficient, which has been shown to be an appropriate indicator of species' concentrations in certain habitats (Chytrý  Tichý and Chytrý 2006). The phi coefficient compares the observed frequencies of a species within a given community with frequencies that would be expected if the species was randomly distributed. The coefficient varies between -1 and ? 1; higher values reflect higher diagnostic values. Significant diagnostic species were identified with Fisher's exact test. We used JUICE 7.0.45 software (Tichý 2002) for the calculations.
Species number and Shannon diversity were computed for each plot. To examine whether there were any significant differences between the habitats, we applied the Friedman test using the 'friedman.test' function of the stats package (R Core Team 2018). The individual patches were used as blocking factor in the analyses. For the post-hoc pairwise comparisons of the habitats, the Nemenyi test was used with the 'posthoc.friedman.nemenyi.test' function of the PMCMRv4.3 package (Pohlert 2014).
All species were classified into phytosociological preference groups according to Borhidi (1995) and the Flora Croatica Database (https://hirc.botanic.hr/fcd/). Frequency distributions were calculated for each habitat, which were then compared using Pearson's chi-square test with the 'chisq.test' function of the stats package (R Core Team 2018). For the post-hoc pairwise comparisons of the frequency distributions of the habitats, we used the 'pairwiseNominalIndependence' function of the rcompanion package (Mangiafico 2018).
We also calculated the mean ecological indicator values for soil moisture, light availability and nutrient supply for each plot. We used the indicator values of Pignatti (2005), which are based on the values of Ellenberg et al. (1992) but extended for southern Europe. Earlier field measurements have shown that ecological indicator values are able to provide reliable estimates of site conditions (e.g. Schaffers and Sýkora 2000;Dzwonko 2001;Tölgyesi et al. 2014). It has been shown that mean ecological indicator values perform well and have a solid theoretical basis (ter Braak and Gremmen 1987;Diekmann 2003). The Friedman test was used to examine differences among the habitats, using the 'friedman.test' function of the stats package (R Core Team 2018), while the Nemenyi test was used for post-hoc comparisons, using the 'posthoc.friedman.nemenyi.test' function of the PMCMRv4.3 package (Pohlert 2014).

Results
We found a total of 131 plant species in the 80 plots (species cover values for all plots can be found in the Online Resource 1). North-facing edges had 93 species, south-facing edges had 88 species, while 88 species occurred in the forests, and 61 species in the grasslands.
According to the DCA ordination, forest plots and grassland plots formed two well-distinguishable groups (Fig. 2). Edge plots were situated in an intermediate position. North-facing edges and southfacing edges overlapped considerably in the ordination space.
A large number of species occurred in all four studied habitats (39 species) (Fig. 3). Somewhat fewer species were shared among forests, north-facing edges and south-facing edges (16 species) or between forests and north-facing edges (10 species). The number of species restricted to north-facing edges (14 species) or forests (12 species) was also considerable.
Forests had 16 diagnostic species, while grasslands had 11 diagnostic species (Table 1). The number of diagnostic species in north-facing and south-facing edges was 10 and 5, respectively. Notably, the diagnostic species of edges had rather low fidelity values. Among the significant diagnostic species of north-facing edges, there was only one species (Peucedanum cervaria) that is regionally regarded as edge-related. The situation was similar for southfacing edges, since Peucedanum oreoselinum was the only diagnostic species known for its regional affinity to edges. However, in our study, this species was also diagnostic for grasslands.
Habitat type had a significant influence on per plot species number according to the Friedman test (v 2 = 13.338, df = 3 , p \ 0.01). As shown by the post-hoc tests, north-facing edges were the most species rich, while forests and grasslands had significantly lower per plot species numbers (Fig. 4a). South-facing edges did not differ significantly from any other habitat, although they seemed to be more species rich than habitat interiors.
Habitat type significantly influenced Shannon diversity, as indicated by the Friedman test (v 2 = 14.460, df = 3, p \ 0.01). The post-hoc comparisons showed that forests had the lowest Shannon diversity values, north-facing edges and grasslands were significantly more diverse, while south-facing edges were intermediate (Fig. 4b).
There were significant differences among the frequency distributions of the phytosociological preference groups in the four habitat types, as shown by Pearson's chi-square test (v 2 = 209.43, df = 18, p \ 0.001). The post-hoc tests revealed no significant differences between north-facing edges and southfacing edges, while the other habitats differed significantly from one another (Fig. 5). The forest habitat was dominated by species of mesic and xeric forests and scrubs, while species of mesic and xeric grasslands were more typical than other types of species in the grassland habitat. Edges were generally intermediate: the proportion of species of mesic and xeric forests and scrubs was lower than in the forest interior but higher than in the grassland interior, while the reverse pattern was true for the species of mesic and xeric grasslands.
Mean ecological indicator values differed significantly among the four habitats, as shown by the Friedman test (soil moisture: v 2 = 33.603, df = 3, p \ 0.001, light availability: v 2 = 48.780, df = 3, p \ 0.001, nutrient supply: v 2 = 39.780, df = 3, p \ 0.001). According to the post-hoc tests, forests  Venn diagram of all species found in the study plots, according to their habitats. F forest, NE north-facing edge, SE south-facing edge, G grassland had significantly higher moisture values than the other three habitats (Fig. 6a). Forests had the lowest and grassland the highest light indicator values, while edges were intermediate (Fig. 6b). Forests proved to have the highest nutrient values, while grasslands were nutrient-poor, edges being intermediate (Fig. 6c).   berg et al. 2005) and Kazakh forest-steppes . Thus, based on species composition, it seems justifiable to treat edges as separate communities in all the abovementioned study areas. However, different patterns also exist. For example, the species composition of edges may be very similar to that of forest interiors, as was the case in the xeric scrub of the Brazilian Caatinga (Santos and Santos 2008). Alternatively, edge composition may be similar to the composition of grassland interiors, as was shown by Erd} os et al. (2011) in a xeric rocky scrubland in the Carpathian Basin. Edges in our study area hosted both forest-related and grassland-related species. This result, however, was also obtained for the forest and grassland habitats (i.e. grassland-related species occurred in the forests, and some forest-related species were found in the grasslands). This fact may be explained by the finescale mosaic pattern of the study area (Horvat et al. 1974;Vukelić 2012): forest patches are so small that their total area is affected by the neighbouring grasslands; because the forest patches have a relatively dry and warm microclimate, colonisation by grassland species can occur. Similarly, the small grassland patches are probably influenced by the canopy of the nearby trees, and therefore, some forest species can easily extend into grasslands (Baker et al. 2013).  As predicted by the edge effect theory (e.g. Risser 1995;Kent et al. 1997), edges had their own species, i.e. species that were significantly concentrated in edges. Interestingly, species that are regarded as edge related in regional phytosociological databases were under-represented among the significant edge diagnostic species identified in our study area. This result indicates that local edge species do not necessarily correspond to regional edge species (Lloyd et al. 2000;Erd} os et al. 2013). Some species that are common outside of edges in a given region may be restricted to edges locally, provided that only edges have an appropriate combination of environmental factors in that specific location.
Per plot species richness was highest in northfacing edges, followed by south-facing edges; however, south-facing edges were not significantly different from the forest and grassland habitats. The Shannon diversity of north-facing edges was higher than that of forests but did not differ significantly from that of grasslands, while the diversity of south-facing edges did not differ significantly from any of the other studied habitats. Increased species richness has also been found in similar xeric forest-grassland mosaics in Eastern Europe (Molnár 1998;Erd} os et al. 2013Erd} os et al. , 2014, Asia  and South America (de Casenave et al. 1995).
The total (i.e. pooled) species number was highest in north-facing edges and slightly lower in southfacing edges and forests, while it was lowest in grasslands (Fig. 7). In an earlier study conducted in the Carpathian Basin, Erd} os et al. (2013) also found that the total species number was highest in the edge habitat. However, grasslands in that study were almost as species rich as edges, while forests were particularly species poor, which is in contrast to the species richness patterns of the Samobor Mountains revealed in our present study. One possible explanation for these differences may be found in biogeographic patterns. The Samobor Mountains belong to the zone of deciduous forests, where grasslands were formed by human activity in historical times (Ellenberg 1988). This pattern explains why the species pool of forests exceeds that of grasslands. In contrast, substantial parts of the Carpathian Basin belong to the foreststeppe belt (Magyari et al. 2010), where grasslands are natural and have a much longer history, resulting in a considerably larger species pool.
According to the ecological indicator values, edges had mostly intermediate environmental conditions between the forest and the grassland habitats (Fig. 7), which is in line with earlier studies based on direct We found moderate differences between northfacing and south-facing edges regarding species composition, while no significant differences were revealed regarding species richness, Shannon diversity, phytosociological preference groups and mean ecological indicator values. Some earlier studies suggested that there may be considerable differences between differently exposed edges in terms of abiotic factors (Ries et al. 2004;Wicklein et al. 2012), species richness (Fraver 1994;Erd} os et al. 2013Erd} os et al. , 2018aErd} os et al. , 2018b and species composition (Brothers and Spingarn 1992;Fraver 1994). However, the study of Erd} os et al. (2011), conducted in a recently abandoned pasture with a fine-scale forest-grassland mosaic found no significant differences between differently exposed slopes, which is in line with our current results.
In sum, we found that edges have a unique species composition, supporting species from both habitat interiors plus their own edge-related species. These edge-related species did not correspond to regionallyidentified edge-related species. We found evidence for increased species richness in north-facing edges (Fig. 7), while this was not true for south-facing edges. Our findings support the notion that edges should be recognised as a special component of forestgrassland mosaics, which has important conservation implications. In many forest-grassland mosaics of Europe, land abandonment results in succession and gradual development into forest or shrubland (Ellenberg 1988). This process is considered undesirable, as grasslands represent high conservation value (Dengler et al. 2014;Valkó et al. 2018). It is clear, however, that if the mosaic character is lost with ongoing succession, not only the grassland but also the edge component will disappear. As shown by our results, edges contribute considerably to the compositional and structural complexity of the landscape. Thus, forestgrassland mosaics should be preserved not only because of grasslands but also because of edges.
The re-establishment of traditional low-intensity, extensive agricultural practises has been suggested as an appropriate management tool to preserve seminatural grasslands in many European landscapes (Ostermann 1998). The historical land-use of our study area in the Samobor Mts is grazing. It has been shown that in such cases grazing is the best option, as many species are adapted to the specific disturbance dynamics of grazing (Römermann et al. 2009). Unfortunately, grazing may not be an economically viable solution any more. While mowing and mulching may be less favourable from a nature conservation perspective, they are usually considered acceptable alternatives in calcareous grasslands, as they are easier to implement and are able to maintain the mosaic  Moog et al. 2002;Wallis de Vries et al. 2002).