Habitat fragmentation reduces grassland connectivity for both short-distance and long-distance wind-dispersed forbs

Authors

  • M. B. SOONS,

    Corresponding author
    1. Plant Ecology Group, Department of Plant Biology, Utrecht University, Sorbonnelaan 16, NL-3584, CA Utrecht, The Netherlands, and
    • Present address and correspondence: M. B. Soons, Landscape Ecology Group, Department of Geobiology, Utrecht University, Sorbonnelaan 16, NL-3584 CA Utrecht, The Netherlands (e-mail: m.b.soons@bio.uu.nl).

    Search for more papers by this author
  • J. H. MESSELINK,

    1. Plant Ecology Group, Department of Plant Biology, Utrecht University, Sorbonnelaan 16, NL-3584, CA Utrecht, The Netherlands, and
    Search for more papers by this author
  • E. JONGEJANS,

    1. Nature Conservation and Plant Ecology Group, Wageningen University, Bornsesteeg 69, NL-6708 PD Wageningen, The Netherlands, and Experimental Plant Ecology Section, and Radboud University Nijmegen, Toernooiveld 1, NL-6525 ED Nijmegen, The Netherlands
    Search for more papers by this author
    • Present address: Department of Biology, The Pennsylvania State University, 208 Mueller, Laboratory, University Park, PA 16802, USA.

  • G. W. HEIL

    1. Plant Ecology Group, Department of Plant Biology, Utrecht University, Sorbonnelaan 16, NL-3584, CA Utrecht, The Netherlands, and
    Search for more papers by this author

Summary

  • 1Although habitat loss and fragmentation are assumed to threaten the regional survival of plant species, their effects on regional species dynamics via seed dispersal and colonization have rarely been quantified.
  • 2We assessed the impact of habitat loss and fragmentation on the connectivity, and hence regional survival, of wind-dispersed plant species of nutrient-poor semi-natural grasslands. We did this using a new approach to relate quantified habitat loss and fragmentation to quantified colonization capacity.
  • 3We quantified loss and fragmentation during the 20th century of moist, nutrient-poor semi-natural grasslands in study areas in the Netherlands, as well as their current distribution. After testing how well the habitat distribution matches species distributions of two wind-dispersed grassland forbs (Cirsium dissectum, representative of species with long-distance wind dispersal, and Succisa pratensis, representative of species with short-distance wind dispersal), we combined the habitat distribution data with simulated seed dispersal kernels in order to quantify the impact on connectivity.
  • 4Habitat loss and fragmentation has dramatically reduced both the area (by 99.8%) and the connectivity of the grasslands. The remaining grasslands are practically isolated for seeds dispersed by wind, even for species with high wind dispersal ability (for which, interestingly, connectivity by wind dispersal decreased most). Linear landscape elements hardly contribute to connectivity by wind dispersal. Regional survival of the studied species has become completely dependent on the survival of a few large populations in nature reserves. Other remaining populations are decreasing in number and size and have low colonization capacity.
  • 5Habitat loss and fragmentation have drastically changed the regional species dynamics of wind-dispersed plant species, indicating that it is of utmost importance to preserve remaining populations in nature reserves and that the probability of colonization of new or restored sites is very low, unless the sites are adjacent to occupied sites or dispersal is artificially assisted.

Introduction

Reallocation of land for agricultural, industrial and other purposes continues to result in loss and fragmentation of (semi-)natural areas all over the world, threatening the regional survival of plant and animal species dependent on these habitats (Saunders et al. 1991; Tilman et al. 1994; Vitousek et al. 1997; Hanski & Ovaskainen 2000). Reduced habitat area for species in a region directly influences the species’ survival, but many effects are indirect: (i) remaining habitat patches are more vulnerable to influences from the surroundings, which can reduce habitat quality; (ii) they may contain smaller populations, which have higher extinction probabilities; and (iii) they may be at greater distance from each other, which reduces colonization of unoccupied habitat patches and gene flow between occupied patches, and hence may increase extinction probabilities of local populations and regional population assemblages.

The regional survival of individual species depends not only on the rate and spatial pattern of loss and fragmentation of their habitat, but also on the population dynamics and colonization capacity of the species, as these factors may determine the species’ vulnerability to small habitat patch areas and large inter-patch distances (Keymer et al. 2000; Tischendorf & Fahrig 2000; Vos et al. 2001b; Moilanen & Nieminen 2002). Therefore, assessment of the threat of habitat loss and fragmentation for species survival requires quantification of the process of habitat loss and fragmentation in relation to species dynamics and colonization capacity. However, such quantifications have only rarely been carried out for plant species. Quantifying habitat loss and fragmentation is difficult, largely because of the problems of mapping the habitat of plant species, especially at sites where the species is not present and at various times in the past. It is also difficult to relate habitat loss and fragmentation to the relevant plant population processes, unless relationships between habitat distribution and local population dynamics and colonization capacity of the plant species can be established. Quantification of the colonization capacity of plant species includes long-distance seed dispersal, a crucial component of plant colonization capacity that is notoriously difficult to quantify (Cain et al. 2000; Nathan 2001; Nathan et al. 2003). Finally, the relationships need to be quantified in an ecologically meaningful way at spatial and temporal scales relevant to the species.

Thus, although regional dynamics of plant species under changing landscape conditions have received the attention of ecologists (e.g. Freckleton & Watkinson 2002; Ehrlén & Eriksson 2003; Ouborg & Eriksson 2004), very few studies have quantified them. However, recent digitization of detailed historical and current vegetation data for the Netherlands (Hennekens et al. 2001) may facilitate habitat classification and reconstruction of habitat loss and fragmentation. Ongoing development in mechanistic seed dispersal models (e.g. Okubo & Levin 1989; Greene & Johnson 1995; Jongejans & Schippers 1999) has recently resulted in models that can estimate long-distance seed dispersal and colonization of wind-dispersed plant species, based on characteristics of the plants and their environment (Nathan et al. 2002, 2005; Soons et al. 2004a, 2004b; Nathan et al. 2005), and thus allows us to relate habitat loss and fragmentation to colonization. GIS tools can combine spatial data from many sources to create habitat maps and to calculate habitat distribution statistics and connectivity measures.

We combined these tools to study the loss and fragmentation of nutrient-poor semi-natural grasslands, a typical habitat for wind-dispersed plant species, in the Netherlands. Our aim was to assess the impact of habitat loss and fragmentation on the connectivity, and hence regional survival, of wind-dispersed grassland forbs. Our approach involved three steps: (i) quantification of habitat loss and fragmentation, (ii) quantification of species seed dispersal kernels and (iii) combining these to quantify the impact of habitat loss and fragmentation on connectivity in an ecologically meaningful way.

Materials and methods

the study system

Human population growth, increasing industrialization and intensification of agriculture have resulted in large-scale, well-documented landscape changes in the Netherlands during the 20th century. Nutrient-poor semi-natural grasslands, which were once widespread, have become highly fragmented. For an accurate and practical definition of the habitat, we restricted this study to Pleistocene soil areas (Fig. 1), selecting a set of floristically closely related vegetation communities of moist, nutrient-poor semi-natural grasslands (Table 1).

Figure 1.

Location of the three study areas in the Pleistocene soil areas (darker shade) of the Netherlands. Location of Pleistocene soil and flora areas adapted from Weeda (1990).

Table 1. The vegetation communities that define the studied grassland habitat. These communities often occur adjacent to each other and show floristic gradients in contact zones. A few closely related communities that are very rare or not present in the study areas are not listed. Detailed information on the communities is given by Schaminée et al. (1999)
Vegetation communityDescriptionHabitat class
Cirsio dissectiMolinietumSemi-natural, relatively nutrient-poor grassland communities at moist sites (commonly inundated during winter)1
Gentiano pneumonanthes–Nardetum Extensively managed, nutrient-poor grassland communities at moist sites, often adjacent to moist heathland1
Campylio–Caricetum dioicae Nutrient-poor wetland communities, commonly inundated during winter1
Carici curtae–Agrostietum caninae Extensively managed (relatively) nutrient-poor wetland communities, e.g. in the lagg zone of bogs1
Ericetum tetralicis (esp. orchietosum)Semi-natural, nutrient-poor heathland communities at moist to wet sites2
Basal/derivative communitiesCommunities that resemble the communities listed above but lack (several of) the characteristic species2

We selected three study areas that represent different land-use histories (Fig. 1). The Achterhoek (AH) area (11 × 10 km, lower-left coordinate 51°59′20″ N, 6°25′15″ E) contained large wetlands until the end of the 19th century. Nutrient-poor moist grasslands were abundant and were mostly used for cattle grazing. Land drainage had already started to reduce the area of the moist grasslands by 1900 and this process accelerated early in the 20th century, when artificial fertilizers became available. However, land use remained relatively small-scale until the 1970s, when a major land reallotment procedure resulted in the loss of small-scale patchiness and the disappearance of almost all remaining wet or moist (semi-)natural areas. The Gelderse Vallei (GV) area (10 × 10 km, lower-left coordinate 51°58′31″ N, 5°31′58″ E) was intensively cultivated much earlier than the AH area. Although originally also a wet area, large parts of the GV area were drained and turned into arable land before 1900. The wettest areas, around brooks and rivers, remained in use as grazing lands until the early decades of the 20th century. From the first half of the 20th century onwards the area has been a highly productive agricultural area. The Veluwe (VE) area (10 × 10 km, lower-left coordinate 52°13′36″ N, 5°39′02″ E) is the driest of the three study areas, with only local wetlands. Most of the area was in use as common grazing ground for sheep until the early decades of the 20th century. Since then, moist and fertile areas have been drained and reclaimed for agriculture. We documented habitat loss and fragmentation during the 20th century in the AH area using three time steps (1900, 1950 and 2000) but analysed only the habitat distribution in 2000 at the other sites.

We selected two wind-dispersed grassland forbs that co-occur in moist, nutrient-poor semi-natural grasslands and which represent two common seed dispersal strategies. Like their habitat, the species were once common in the Pleistocene soil areas of the Netherlands, but are now relatively rare. Cirsium dissectum (L.) Hill has plumed diaspores that are adapted to long-distance dispersal by wind, whereas the diaspores of Succisa pratensis Moench (which have a persistent calyx that is dry and hairy and increases the surface area of the diaspores without adding much weight; Bouman et al. 2000) are also dispersed by wind, but only over short distances. Both species are hemicryptophytes that overwinter as leaf rosettes. They propagate sexually via seeds and asexually by side-rosettes and clonal extension (Hartemink et al. 2004; Jongejans et al. 2005), and have transient seed banks (Thompson et al. 1997). More details on the species are given by Soons & Heil (2002) and Jongejans & de Kroon (2005).

quantification of habitat loss and fragmentation

We created a time series of maps (years 1900, 1950 and 2000) of the studied grassland habitat in the AH area and mapped its current distribution for the GV and VE areas. For a practical and objective approach to mapping the grassland habitat we defined habitat classification criteria and made a set of basic assumptions, which are listed in Appendix S1 (available as supplementary material online). In brief, we defined the studied grassland habitat using a set of floristically closely related grassland communities in which C. dissectum and S. pratensis co-occur (Table 1). Within the study areas the species co-occur almost exclusively in these grassland communities, although with different probability of occurrence for each community. Therefore, we assigned the communities to different habitat classes: class 1 has a very high probability of occurrence of C. dissectum and S. pratensis, whereas class 2 has a lower probability (Table 1). For the years 1900 and 1950, topographic maps showed potential habitat patches for which no further data were available, and because they could not be assigned to class 1 or 2 but there was still a probability that they were habitat patches they were designated as habitat class 3.

The habitat classification was based on vegetation communities and the presence of indicator species. This method was used mainly because the use of vegetation compositions (rather than only the presence or absence of C. dissectum and S. pratensis) allows more accurate classification of patches in which C. dissectum or S. pratensis are not present. Habitat maps for 2000 were based on field surveys, whereas historical vegetation and species distribution data from floristic databases, historical descriptions of the nature reserves, and interviews with farmers and other local people (Messelink 2001) were used for the maps for 1950 and, to a more limited extent, for 1900. For 1900 we also used detailed historical landscape and vegetation descriptions and information from historical topographic maps, which contain more detail on land use and soil moisture than later maps (Messelink 2001).

We applied two tests to assess whether our approach to grassland habitat mapping was effective for mapping the habitat of C. dissectum and S. pratensis. First, we compared the distributions of C. dissectum and S. pratensis in 2000 with the distribution of habitat in 2000 to see whether the species occur in sites that were not mapped as habitat (e.g. linear landscape elements such as road verges). To map the distributions of C. dissectum and S. pratensis we searched all semi-natural areas in the study areas, plus linear landscape elements in the AH area (the only area where road verges have purely natural plant communities). We searched linear landscape elements on a random basis and where there was information suggesting that C. dissectum or S. pratensis may previously have been present in the surroundings. In total, about 50% of all linear landscape elements were investigated.

Second, we carried out a seed addition experiment for C. dissectum and S. pratensis to compare seedling establishment between sites classified as habitat and non-habitat. We selected 10 nature reserves (to ensure that each habitat class was represented) and 10 road verges (to represent the full variation in road verges) in the AH area and included three recently (winter 1999/2000) restored sites where the topsoil had been removed to eliminate nutrients and unwanted species. We measured vegetation productivity and openness to quantify site conditions. We clipped above-ground biomass in three vegetation plots (20 × 20 cm) per site. Biomass samples were weighed stove-dry and averaged per site. We estimated vegetation height and percentage cover at peak standing crop. At each site, seeds of the selected species were sown in eight plots (5 × 50 cm) per species after the vegetation was mown in October–November 2000. The plots were randomly located along a 20-m-long transect and mixed with 16 control plots in which no seeds were sown. In each plot we sowed 70 C. dissectum (560 per site) and 100 S. pratensis (800 per site) seeds. All seeds were collected from several large populations in summer 2000, mixed and stored in the dark at room temperature until the start of the experiment. Establishment from the added seeds was calculated as the number of newly established individuals in October–November 2001 in the seed addition plots minus the number in the control plots (Jongejans et al. 2005). All seedlings of the selected species in the plots were removed at the end of the experiment.

Habitat loss and fragmentation were quantified from the grassland habitat maps using GIS and four general and easily interpreted quantitative habitat descriptors: total habitat area, number of patches, average (and range in) patch size, and average (and range in) distance of patches to their nearest neighbours. Because the three habitat classes have different probabilities of containing actual habitat for C. dissectum and S. pratensis, habitat loss and fragmentation we calculated three scenarios. In the ‘worst-case’ scenario, only patches assigned to class 1 are assumed to contain habitat for the selected species, whereas in the ‘intermediate’ scenario patches assigned to either class 1 or class 2 are assumed to contain habitat, and in the ‘best-case’ scenario patches assigned to classes 1, 2 and 3 are all assumed to contain habitat.

quantification of species seed dispersal kernels

To assess the connectivity of habitat patches for the selected species, both the spatial configuration of the species’ habitat in the landscape and the dispersal ability of the species need to be quantified. The dispersal ability of most plant species is described by their seed dispersal kernel. For species with long-distance dispersal, dispersal kernels are, however, notoriously difficult to measure (Cain et al. 2000; Nathan et al. 2002, 2003). Therefore, we used a mechanistic simulation model that is able to simulate long-distance seed dispersal to calculate complete dispersal kernels for C. dissectum and S. pratensis (see Nathan et al. 2002 and Soons et al. 2004a, 2004b– the Markov chain Synthetic Turbulence Generation model – for details of the model and its reliability). We calculated seed dispersal kernels for Cdissectum and Spratensis by simulating seed dispersal trajectories of 10 000 seeds, using average values (including standard deviations) of the dispersal parameters (seed terminal velocity, release height and vegetation height) measured for each species under natural conditions (Soons & Heil 2002). We simulated dispersal for the average vegetation conditions (Soons & Heil 2002) and the average wind velocity distribution (data from the Royal Netherlands Meteorological Institute) during the dispersal season (June–October) in the interior of the Netherlands.

quantification of the impact of habitat loss and fragmentation on connectivity

Quantitative analyses of the effects of habitat loss and fragmentation on connectivity were carried out as scenario analyses, because of the three habitat classes involved (see Quantification of habitat loss and fragmentation section). Habitat connectivity is more difficult to quantify than simple quantitative habitat descriptors such as total habitat area. Functional habitat connectivity (sensuTischendorf & Fahrig 2000) reflects the probability of colonization of unoccupied habitat patches and gene flow between occupied habitat patches. Many different connectivity measures have been used (Schumaker 1996; Tischendorf & Fahrig 2000; Vos et al. 2001b; Moilanen & Nieminen 2002), but most of them fail to quantify connectivity in a realistic way for plant species. This is primarily because they do not include realistic seed dispersal distributions, and especially long-distance dispersal. In addition, many measures of connectivity do not meet the requirements (cf. Tischendorf & Fahrig 2000; Moilanen & Nieminen 2002) that (i) all habitat patches within species’ colonization and gene flow distances are considered, (ii) relative contributions to colonization and gene flow by habitat patches at different distances and with populations of different size and/or colonization capacity are considered, (iii) the effects of the landscape matrix on dispersal are considered, (iv) both within-patch and between-patch dispersal are considered, and (v) connectivity is expressed per equal-sized unit of habitat. We focus here on gene flow by seed dispersal only.

Our habitat maps and dispersal kernel data allow spatial calculations using GIS and make it possible to use a spatially explicit connectivity measure that meets the above requirements. We used a new measure of connectivity, the species-specific probability Cij(r) that seeds dispersing from a habitat patch over a distance r reach another habitat patch, to combine the spatial data on habitat distribution and seed dispersal in an ecologically meaningful way. This measure is the result of the probability that a seed that disperses over distance r lands at a habitat site [Oij(r), cf. Wiegand et al. (1999) and the probability that a seed disperses over a distance rDi(r), the distribution of seed dispersal distances].

O ij (r) is the probability that a seed that disperses over distance r from a habitat unit of class i lands at a habitat unit of class j. For each i:

image

where h = 1, … , n are the habitat units of class i, Nhj(r) is the number of units of habitat class j at distance r from habitat unit h, and Nh(r) is the total number of landscape units at distance r from habitat unit h. We calculated Nhj(r) using a raster GIS (PCRaster, Van Deursen 1995; http://www.pcraster.nl) and digitized habitat maps with a resolution of 4 × 4 m (i.e. unit size 4 × 4 m).

D i (r) is the probability that a seed that disperses from a habitat unit of class i travels a distance r. We calculated Di(r) for each species from the simulated seed dispersal kernels. To make Di(r) compatible with Oij(r) we grouped the seed dispersal distances into intervals of 4 m (the resolution of the habitat maps). For simplicity we assumed Di(r) to be equal for all three habitat classes.

For each species, Cij(r) was calculated by multiplying Oij(r) by Di(r). The resulting probability distribution Cij(r) is a species-specific measure of functional habitat connectivity: it is the probability that a seed dispersing from a habitat unit of class i disperses to a habitat unit of class j at dispersal distance r. The probability of patch colonization can be assessed for each population by multiplying Cij(r) by the number of germinable seeds produced. Seed production and germination ability were as determined previously (Soons & Heil 2002).

Results

habitat maps

Habitat maps of the AH area in 1900, 1950 and 2000 show how much habitat has been lost and how fragmented the remaining patches have become (Fig. 2). By 2000, only eight small nature reserves that lie scattered throughout the landscape matched the habitat criteria (Fig. 2c), with one species-rich road verge also classified as habitat class 2. The recently restored nature reserves in the AH area (mentioned in the Seed addition experiment section) were not mapped as habitat because they were hardly vegetated. In the GV area, only three patches in nature reserves matched the habitat criteria (all class 1). The VE area had only one nature reserve containing an area of class 1 habitat and one species-rich road verge of class 2 habitat.

Figure 2.

Habitat map of the AH area in (a) 1900, (b) 1950 and (c) 2000. Black: habitat class 1, dark grey: habitat class 2, light grey: habitat class 3. The area is 11 × 10 km; coordinates of the lower-left corner are 51°59′20″ N, 6°25′15″ E.

tests of the habitat mapping approach

Six of the eight habitat patches in all areas that were classified as class 1 in 2000 contained both selected species. The distribution of C. dissectum matched the habitat distribution. C. dissectum is rarer than S. pratensis and occurs exclusively in patches classified as habitat (Table 2). Populations of S. pratensis are larger and more numerous (Table 2). There are more populations of S. pratensis in sites classified as non-habitat (all of these in linear landscape elements) than in patches classified as habitat. However, the populations in non-habitat are generally much smaller, with four of the S. pratensis populations in linear landscape elements just consisting of only one or two rosettes.

Table 2. Distributions of Cirsium dissectum and Succisa pratensis in all study areas. Number of populations and range in population sizes (measured as number of flowering rosettes) for the habitat classes and the linear landscape elements (LLEs). The two habitat patches in LLEs are listed with the habitat patches (class 2). All populations found at sites that were classified as non-habitat were in LLEs
Year 2000 Class 1 (total: 8 patches)Class 2 (total: 6 patches)LLE (AH only)
Cirsium dissectum No. of populations  6 0 0
Population size range  4–11 000 – –
Succisa pratensis No. of populations  8 315
Population size range150– > 100 00075–2200 1–100

The seedling establishment experiment showed that establishment from added seeds was higher in patches classified as habitat than in patches classified as non-habitat (Fig. 3). Establishment was higher in habitat class 1 than in class 2, but this difference was not significant. High establishment was also found in the recently restored nature reserves where the topsoil had been removed, indicating that restoration improved conditions for seedling establishment of C. dissectum and S. pratensis. Low productivity (measured as above-ground biomass) and high vegetation openness (measured as percentage cover) increased seedling establishment significantly in both habitat and non-habitat sites (logistic regression, P < 0.001 and P < 0.004, respectively). This relationship partly explains the low suitability of road verges for establishment, although establishment was higher in nature reserves than in road verges even when these had similar productivity, and vegetation openness and other factors must therefore also play a role.

Figure 3.

Seedling establishment at sites of different habitat classes 1 year after sowing. TSR = topsoil removed site (no habitat class assigned; n = 3), habitat class 1 (n = 2), habitat class 2 (n = 2), non-habitat (includes all road verges; n = 13). Different letters indicate statistically significant differences for the sum of the seedlings of both species (Kruskall–Wallis test with Dunn's post-hoc test, α = 0.05).

habitat loss, fragmentation and connectivity

Habitat destruction in the AH area has drastically reduced the total habitat area and the number and size of habitat patches since 1900 and increased the distance between patches (Fig. 2, Table 3), leading to decreased habitat connectivity (Fig. 4a). Seed dispersal distributions are shown in Fig. 5. For the species with plumed seeds, C. dissectum, habitat connectivity was reduced drastically by habitat fragmentation (Fig. 6). In particular, the probability of seed dispersal by wind to habitat sites at a distance > 50 m was greatly reduced (by > 90%). Habitat connectivity for S. pratensis was hardly reduced, because its seeds are generally dispersed by wind over such short distances (< 4 m) that almost all seeds land in the habitat patch in which they are released, in 1900 as well as in 2000.

Table 3. Habitat area and configuration at different times in the AH area and in the VE and GV areas in 2000. Total area of the AH area is 11 000 ha; those of the GV and VE areas are both 10 000 ha
Habitat classAH 1900AH 1950AH 2000VE 2000GV 2000
11 & 21 & 2 & 311 & 21 & 2 & 311 & 211 & 21
Total habitat area (ha)931313237048.7424.445.92.404.112.062.9919.1
No. of patches872633358227449123
Average patch size (ha)10.711.9111.11.091.110.620.600.462.061.496.35
Range in patch sizes (ha)0.11–1790.08–3840.03–3840.31–2.930.03–4.690.01–4.690.25–0.950.03–0.950.92–2.061.62–12.9
Average inter-patch distance (km)0.240.140.111.40.270.401.61.81.22.2
Range in inter-patch distances (km)0.01–1.20.01–0.810.01–0.320.09–5.90.01–1.30.01–2.41.1–2.10.22–4.70.67–5.3
Figure 4.

O ij (r), the probability that seeds that disperse over distance r from one habitat unit end up at another habitat unit (here, a habitat unit of the same class). (a) The time series in the AH area; (b) the three study areas in 2000.

Figure 5.

D i (r), the probability that seeds disperse over distance r from their release point. (a) C. dissectum (seed terminal velocity 0.38 ± 0.03 m s−1, seed release height 0.50 ± 0.14 m, vegetation height 0.35 m); (b) S. pratensis (2.14 ± 0.20 m s−1, 0.56 ± 0.14 m, 0.35 m). The resolution of r is 4 m to match Oij(r), which is calculated from the 4 × 4 m resolution habitat maps. Zero values are not shown.

Figure 6.

Habitat connectivity measure Cij(r), the probability that seeds that disperse from one habitat unit end up at another habitat unit (here, a habitat unit of the same class) at distance r from their release point. (a) C. dissectum; (b) S. pratensis; all parameters as in Fig. 5. The y-axis on a logarithmic scale. Zero values are not shown.

The largest reductions in habitat area and connectivity took place between 1900 and 1950, with 922 ha of habitat lost in the scenario where only class 1 is habitat and 3658 ha lost in the scenario where classes 1, 2 and 3 are all habitat. This is equivalent to 18 ha year−1 and 73 ha year−1, respectively. The average patch size was reduced by 10 ha (0.2 ha year−1). These rates were much lower during the second half of the 20th century: a reduction in habitat area of 6 ha vs. 42 ha, respectively, in the two scenarios (equivalent to 0.1 ha year−1 and 0.8 ha year−1) and a reduction in average patch size of 0.5 ha vs. 0.2 ha (0.01 ha year−1 and 0.004 ha year−1). Only the increase in average inter-patch distance was similar for the two time periods: 1.2 km vs. 0.3 km (24 m year−1 vs. 6 m year−1) for 1900–1950 and 0.2 km vs. 1.4 km (4 m year−1 vs. 28 m year−1) for 1950–2000. In total, 929 ha and 3700 ha of habitat was lost during the 20th century, total habitat area decreased by 99.7% and 99.9%, and the number of patches decreased by 95% and 97%, respectively.

By 2000 very little grassland habitat area was left (Table 3) and the few remaining habitat patches were very small and the distances between them large (Table 3). The values of Oij(r) show that, as expected, large patches at short distances from each other provide the highest connectivity (Fig. 4). The GV area contained three relatively large patches and the VE area two smaller patches that were located relatively close to each other (Table 3). The AH area contained most patches, but these were the smallest and most scattered throughout the area (Table 3) and therefore had lower connectivity (Fig. 4).

In 2000 the habitat connectivity Cij(r) was very low for S. pratensis (Fig. 6), owing to its low seed dispersal ability (< 4 m) relative to inter-patch distances (≥ 220 m). For C. dissectum the habitat connectivity was higher (Fig. 6), because its seeds can be dispersed over much greater distances by wind (Fig. 5). However, the habitat connectivity for C. dissectum is much reduced by the low availability of habitat, small size of habitat patches and large distance between habitat patches relative to the species’ seed dispersal distances. The probability that seeds disperse over 220 m (the minimum patch separation) and land at a habitat site is < 0.0001 for C. dissectum and 0 for S. pratensis. The largest populations of C. dissectum in the study areas (Table 2) produce approximately 14 × 104 germinable seeds. Thus, only about 14 seeds year−1 of C. dissectum (and none of S. pratensis) could reach a patch 220 m away. Given that the two patches that are separated by 220 m are both unoccupied by the selected species and that the minimum distance between two occupied patches is 670 m, the probability that seeds of the selected species travel between these patches is almost zero.

Discussion

the habitat maps

Our habitat classification system served its purpose well. For the year 2000, patches classified as habitat class 1 are highly likely to contain actual habitat for C. dissectum and S. pratensis and are more likely to do so than patches classified as class 2. All large populations of C. dissectum and S. pratensis occur in patches classified as habitat and more and larger populations occur in patches classified as habitat class 1 than in class 2. Seedling establishment from experimentally added seeds was much higher in patches classified as habitat than in patches classified as non-habitat, and was higher in habitat class 1 than in class 2.

S. pratensis also occurs in linear landscape elements that were not classified as habitat, but these populations are predominantly relict populations reflecting past site suitability. In the AH area in 2000, most rosettes in road verge populations were large and relatively old, and rosette leaves were often overshaded by higher growing grasses and nitrophilous species. During the landscape survey we found no seedlings of S. pratensis in road verge populations, and only a few clonally produced offspring. Seedling establishment from the experimentally added seeds was very low, probably because of the high productivity and vegetation cover in most road verges. In the AH area the populations of S. pratensis in linear landscape elements are declining in number and size. During previous vegetation surveys in 1988–95, 38 populations were recorded in linear landscape elements (Messelink 2001), but these decreased by 61% to 15 populations in 2000. We did not discover new populations in the AH area during our mapping survey in 2000.

The habitat classification for 1950 and 1900 involved more uncertainty than for 2000. The main uncertainties are demonstrated by our scenario analyses, which show the estimated extremes of the habitat fragmentation process. The absolute differences between the fragmentation scenarios are large, but the relative changes in habitat area and configuration are similar.

the fragmentation and connectivity measures

We used simple and straightforward measures to quantify habitat loss and fragmentation, so that our results can be interpreted easily. For a realistic quantification of connectivity it is important to relate habitat area and configuration measures to species-specific dispersal ability and colonization capacity. Our measure of connectivity, Cij(r), is a highly realistic measure, provided that long-distance seed dispersal is quantified realistically. Because this requires seed dispersal kernels to include long-distance dispersal and such kernels are difficult to measure (Cain et al. 2000; Nathan 2001; Nathan et al. 2003), use of mechanistic simulation models that can simulate long-distance dispersal is currently the best option available. The model used in this study simulates long-distance dispersal in a realistic way (Nathan et al. 2002; Soons et al. 2004a, 2004b), allowing estimation of dispersal kernels under average conditions during a dispersal season. Extreme conditions reducing (e.g. large barriers to wind flow in the landscape) or increasing dispersal distances (e.g. extreme wind events) were not simulated to ensure generality of the dispersal kernels.

C ij (r) differs from previous measures of habitat connectivity mainly because it results in figures that are directly and easily interpretable as it represents the probability that seeds that disperse from one habitat site end up at another habitat site. It is also spatially explicit, indicating the probability that seeds disperse from one habitat site to another habitat site over any given distance. Our measure improves on previous measures of habitat connectivity (Tischendorf & Fahrig 2000; Moilanen & Nieminen 2002) because (i) it considers all habitat within the seed dispersal range of a species, instead of only habitat within a predefined area; (2) contributions to colonization are realistically weighted by inter-patch distances, and can easily also be weighted by population size, seed production and seed germinability in habitat patches; (3) it uses a pixel-based approach (instead of a patch-based approach) that allows us to express both within-patch and inter-patch seed dispersal; and (4) it uses a pixel-based approach (instead of a patch-based approach) to express connectivity per equal-sized unit of habitat. Expression of connectivity per equal-sized unit of habitat is more realistic for plant species than expression of connectivity per habitat patch, because plants do not move around within their habitat patches as (most) animals do.

We did not take into account connectivity through time, because the selected species have only short-lived seed banks, or the presence of dispersal barriers in the landscape matrix, either of which would reduce the estimates of habitat connectivity further. The qualitative results of our analyses would, however, not be affected much by this inclusion, because habitat patches in 2000 are already practically isolated with respect to seed dispersal by wind.

Other measures of connectivity can be derived from genetic differences between populations (Ouborg et al. 1999; Vos et al. 2001a). For instance, AFLP markers have been used to assess the genetic difference between some populations of C. dissectum and S. pratensis (Smulders et al. 2000). The results from application of our measure of connectivity match those from genetic analyses, which show that populations of S. pratensis in the AH area are almost completely isolated (Mix et al. unpublished data) and that the smaller populations have high inbreeding coefficients (Vergeer et al. 2003).

habitat loss, fragmentation and connectivity

We quantified the impact of the loss and fragmentation of moist, nutrient-poor semi-natural grasslands during the 20th century on the habitat connectivity for wind-dispersed grassland forbs. Loss and fragmentation of the studied grasslands in the AH area occurred predominantly during the first half of the 20th century, indicating that most of the remaining habitat patches have been small and isolated for at least 50 years. The currently remaining habitat patches in the three large study areas are all located in nature reserves, except for two on road verges, which are classified as habitat class 2. Both are low-productivity, species-rich road verges that are probably remnants of former nutrient-poor, species-rich grasslands. The road verge in the AH area is located at a site that was mapped partially as habitat class 1 and partially as class 2 for 1900.

In addition to the habitat patches being small, the remaining populations of the selected species are small, especially at sites classified as non-habitat or habitat class 2, but also in some class 1 patches. Two of the six C. dissectum populations consist of between 4 and 17 flowering rosettes and 4 of the 11 S. pratensis populations in habitat patches consist of < 200 flowering rosettes (range 75–182). Such small populations have an increased probability of extinction and a reduced colonization capacity, as the few seeds they produce may have low germination ability (Fischer & Matthies 1998; Kéry et al. 2000; Soons & Heil 2002; Vergeer et al. 2003). Continued survival of small populations in small habitat patches is probably due to the longevity of individuals of S. pratensis (Jongejans & de Kroon 2005) and high clonal propagation in C. dissectum (Jongejans et al. 2005). The current small populations, especially at sites that were classified as non-habitat and on road verges in the AH area, probably reflect the past habitat distribution rather than the current one (as also found in Swedish semi-natural grasslands, Lindborg & Eriksson 2004).

The high seedling establishment from experimentally added seeds at recently restored sites, however, suggests that restoration measures may be successful for the selected species. The low productivity and vegetation cover at these sites probably contributed to the high seedling establishment. Restoration thus might increase the habitat area available to the grassland forbs. If we add the recently restored sites in the AH area to the class 1 habitat patches, the total area of habitat class 1 in 2000 would increase by about 16 ha (almost a factor of eight increase).

Habitat connectivity has greatly decreased during the past century. Surprisingly, habitat connectivity decreased more for C. dissectum, the species with seeds adapted to long-distance dispersal by wind, because for this species habitat connectivity was higher in the past. Habitat connectivity is now very low for both C. dissectum and S. pratensis in all three study areas. Linear landscape elements such as road verges do not increase habitat connectivity for these species. First, most linear landscape elements are currently unsuitable for seedling establishment, probably because of their high productivity and vegetation cover. Second, the area and shape of habitat patches in linear landscape elements are such that seeds dispersing from them have an extremely low probability of ending up at another habitat site. This is in agreement with the low functionality of linear landscape elements as corridors for migration of wind-dispersed plants (Van Dorp et al. 1997). Third, the populations in linear landscape elements have a low colonization capacity because they produce only a few seeds and their wind dispersal ability is low as a result of high and dense vegetation (Soons & Heil 2002; Soons et al. 2004b). The remaining habitat patches and populations of both species are practically isolated with respect to seed dispersal by wind.

Dispersal between the remaining habitat patches and populations by other mechanisms is unlikely. The changes in land use that accompanied the habitat fragmentation have interrupted most previously existing seed dispersal mechanisms (Poschlod & Bonn 1998). The only mammals left in the study areas that are large enough to disperse seeds between the grassland habitat patches are roe deer, whose home ranges of 2–24 ha (Tufto et al. 1996; SanJosé & Lovari 1998) are not large enough to include more than one habitat patch. Flooding nowadays occurs only locally within nature reserves and seed transport by water is thus limited to within patches. Even dispersal of seeds by mowing machinery (Strykstra et al. 1997) is unlikely, because most habitat patches are too distant to be mown by the same farmer or piece of machinery (J. Wensink, personal communication). Spreading out of mown material from species-rich reserves in nearby restored sites is currently the only management tool applied to increase seed dispersal between habitat patches (J. Wensink, personal communication). Most nature reserves in the study areas are not open to the general public. The most likely vector of long-distance seed dispersal may be attachment to biologists visiting one nature reserve after another.

As a consequence of the very low habitat connectivity in the study areas, there is a very low probability of natural colonization of recently restored (or created) patches, even for species that are adapted to long-distance seed dispersal by wind. For the two selected species, re-colonization from the seed bank is also unlikely, because the species have transient seed banks (Thompson et al. 1997). The regional survival of the selected species in the study areas is now completely dependent on the survival of the few remaining large populations in habitat patches in nature reserves. For their conservation it is therefore of utmost importance that (i) management of the remaining habitat patches is aimed at the conservation of the remaining populations, (ii) new habitat patches are restored or created and (iii) these new patches are located in close proximity, preferably adjacent, to the remaining large populations in habitat patches, or that seed dispersal to the new patches is actively artificially assisted.

Acknowledgements

This research was supported by the Research Council for Earth and Life Sciences (ALW), Netherlands Organization for Scientific Research (NWO). We thank J. Wensink and F. van Wijngeeren (Staatsbosbeheer) for permission to carry out experiments in nature reserves and for information from the Staatsbosbeheer archives, J. H. J. Schaminée and S. M. Hennekens (ALTERRA Dutch Vegetation Database), M. Rijken (Provincie Gelderland) and C. L. G. Groen (FLORBASE) for giving us vegetation and species data, L. Brouwer for help with the habitat mapping, W. P. A. van Deursen for help with PCRaster calculations and M. J. A. Werger, J. M. van Groenendael, F. Knauer, L. Haddon and two anonymous referees for comments on earlier versions of the manuscript.

Ancillary