Satellite based land use and landscape complexity indices as predictors for regional plant species diversity

Landscape and Urban Planning 63 (2003) 241–250

Satellite based land use and landscape complexity indices as predictors for regional plant species diversity O. Honnay a,∗ , K. Piessens a , W. Van Landuyt b , M. Hermy a , H. Gulinck a a

Laboratory for Forest, Nature and Landscape Research, University of Leuven, Vital Decosterstraat 102, Leuven B-3000, Belgium b Institute for Nature Conservation, Kliniekstraat 25, Brussels B-1070, Belgium Received 9 April 2002; accepted 9 April 2002

Abstract It is known that habitats composed of spatially heterogeneous abiotic conditions provide a great diversity of potentially suitable niches for plant species. The scientific premises of landscape ecology suggest that, at a higher spatial level, also the composition and structure of the landscape mosaic, influences biotic processes and hence species richness. In this exploratory study we investigated if plant species diversity could be correlated with landscape structure and complexity indices which were based on Landsat Thematic Mapper satellite imagery. Plant species data were derived from the 4 km × 4 km resolution Flora Database of Flanders (i.e. northern Belgium). Plant species number within the 4 km × 4 km grid cells was positively correlated with most of the landscape diversity indices whereas landscape fragmentation indices only affected the group of the threatened species. We found a gradient of increasing species richness beginning from the rural areas of Flanders over the suburban towards the urban areas. This gradient was mostly due to the higher number of alien plant species, warmth indicators and threatened species in urbanised areas. We conclude that, at least in the studied region, the effects of landscape changes on plant species diversity can be monitored and predicted on a large scale and over long periods of time using land cover data. Bottleneck in this kind of analyses remains the reliability of the land cover data and the availability and reliability of the biological data. © 2002 Elsevier Science B.V. All rights reserved. Keywords: Landscape indices; Satellite images; Plant diversity; Alien species

1. Introduction It is known from basic ecology that habitats composed of spatially heterogeneous abiotic conditions provide a greater diversity of potentially suitable niches for plant species than habitats with homogeneous characteristics. Variation in physical structure (e.g. aspect, soil drainage and soil texture) within a habitat has proven to be a suitable predictor for plant ∗ Corresponding author. Tel.: +32-16-32-97-73; fax: +32-16-32-97-60. E-mail address: [email protected] (O. Honnay).

0169-2046/02/$20.00 © 2002 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 9 - 2 0 4 6 ( 0 2 ) 0 0 1 9 4 - 9

species richness in various ecological studies (e.g. Hobbs, 1988; Lapin and Barnes, 1995; Burnett et al., 1998; Nichols et al., 1998; Honnay et al., 1999). In fragmented landscapes not only habitat heterogeneity but also the distance (i.e. isolation) between suitable habitats determines plant species composition and species richness of the patches (Grashof-Bokdam, 1997; Butaye et al., 2001). Less isolated habitats are generally more species rich because they can easily become colonised and a constant inflow of new individuals prevents local extinction due to demographic and environmental stochasticities (Shaffer, 1981). Finally also the shape of a habitat patch may

242

O. Honnay et al. / Landscape and Urban Planning 63 (2003) 241–250

influence species diversity. Within certain margins, irregularly shaped habitat patches contain generally more plant species because of their higher number of environmental gradients (Honnay et al., 1999). All of the former abiotic factors are related to the species diversity of a single habitat patch. The question can be raised if the positive correlation between abiotic heterogeneity and habitat connectivity on the one hand and plant diversity on the other is also valid at a higher spatial level (i.e. the landscape matrix level). The scientific premises of landscape ecology suggest that also the landscape mosaic-as a mixture of different sized and shaped natural and human-managed patches of homogeneous land use-influences biotic processes and hence plant species richness (Forman and Godron, 1986; Turner, 1989). It is clear that at the landscape scale other measures for abiotic heterogeneity than the typical ecological ones have to be applied. This implies that the direct relation between the autecological characteristics of a plant species-like e.g. dispersal ability and degree of habitat specialisation and its presence in a certain landscape becomes much less obvious and that there is some shift towards a black-box approach of biotic diversity. However, this approach allows large scaled analyses of human intervention in the natural landscape and the monitoring of its consequences for plant diversity in general, and for the diversity of specific species subgroups like e.g. threatened species in specific. This may be an advantage if one is required to monitor the state of the environment at a large scale. A very large set of measures for landscape structure analysis (landscape indices) has been developed during the past two decades (e.g. Forman and Godron, 1986; O’Neill et al., 1988; Baker and Cai, 1992; McGarigal and Marks, 1995; Frohn, 1998) and their response to the spatial structure and scale of the landscape has been described and quantified (e.g. Rescia et al., 1997; Hargis et al., 1998). The relation between these landscape indices on the one side and biotic diversity in general, and plant diversity more in specific, on the other side, however, has received remarkably less attention (but see e.g. Roy et al., 1999; Collingham et al., 2000). This is very likely due to the fact that the use of GIS techniques and landscape data sources, such as satellite imagery, which has facilitated large scale landscape pattern analysis, has not kept pace with the availability of reliable biological data at the

landscape scale or is simply not sufficiently applied as a toolbox for landscape ecologists (Gulinck et al., 2000). The availability of systematically recorded biological data at the landscape scale remains a serious bottleneck. In this exploratory study we linked the grid-based Flora Database of Flanders (N-Belgium) with satellite derived landscape indices and land cover data. The main research question was if total plant species diversity and the diversity of various species sub-groups can be derived from landscape structure and complexity indices. In other words, can the basic ecological relation between plant species richness and habitat heterogeneity be extended to a higher spatial level? More generally we wanted to investigate in what degree landscape structure indices have a biological meaning, and vice versa which indices should be used in ecological monitoring.

2. Material and methods Plant species data were derived from the Flora Database of Flanders (Van Landuyt et al., 2000). This is a relational database that relates the distribution of all vascular plant species growing in Flanders to their autecological data. The basic spatial units are the 1 km × 1 km grid cell and the 4 km × 4 km grid cell, where the presence or absence of a plant species is determined by different field surveys using pre-printed species lists. Since 1972 the grid cells are systematically surveyed by volunteers and the collected floristic data is centralised and filed electronically. In this exploratory analysis we used the 4 km × 4 km grid cells. For each 4 × 4 grid cell the total number of plant species was calculated, together with the abundance of a number of species categories, referring to the status of threat (Cosyns et al., 1994), nativeness (Cosyns et al., 1994) and their Ellenberg indicator values for soil reaction, temperature and nitrogen (Ellenberg et al., 1990) (see Appendix C for an overview). Mean species richness of the grid cells is summarised in Fig. 1. Nomenclature follows Delanghe et al. (1988). Landscape data were ultimately derived from satellite images of the Landsat Thematic Mapper (1995). These images were originally classified using a maximum likelihood methodology and complemented with

O. Honnay et al. / Landscape and Urban Planning 63 (2003) 241–250

243

Fig. 1. Mean species number per 4 km×4 km grid cell for each species group. The error bar represents one standard deviation from the mean.

external information layers (essentially concerning built-up classes) into 27 land use or pixel categories. Finally they were resampled from the original 30 m image resolution to a 20 m grid (Dufourmont et al., 1996). For this study, a generalisation to 20 relevant land use categories was made (see Appendix A for an overview of the categories). The maps were further analysed with the ArcView 3.2 GRID module and the Patch Analyst 2.1 with an interface to FRAGSTATS (McGarigal and Marks, 1995; Rempel et al., 1999; McGarigal, 2000). For each 4 km × 4 km flora grid cell, 20 landscape indices were calculated, based on the 20 m × 20 m land use pixels within each grid cell. The landscape indices were classified as: (1) patch diversity and distribution indices; (2) patch shape indices; (3) isolation indices and (4) land use indices (see Appendix B for an overview). A North–South transect through Flanders was selected as study area. For this transect the Flora Database of Flanders is very reliable (i.e. having a high and homogene degree of inventory effort) and the transect represents a variety of land use classes along a rural–urban gradient with the city of Ghent

central in the transect (Fig. 2). The selected transect consisted of 89 complete 4 km × 4 km flora grid cells. Border cells only partially occurring in the Flanders region were omitted. First, we applied an exploratory principal component analysis (PCA) on all landscape variables in order to derive patterns of independent variation. The PCA factor scores of all grid cells were then correlated with the species data using a Spearman rank correlation coefficient. Based on the PCA factor scores of the grid cells we pre-selected relatively uncorrellated and representative landscape indices from each of the four categories and used these variables as predictors for species richness in a stepwise regression model. We used a simple linear least square model for all species groups with the exception of the threatened species where we used a poisson model because of the skewed frequency distribution. We applied SPSS 10.0 and SAS 8.01 for all statistical analysis. Because of possible autocorrelation, the sample units (grid cells) may be not independent, resulting in false rejections of the H0 and hence claiming significant results where there are none (Manly, 1997a). To handle this problem we recalculated and adjusted

244

O. Honnay et al. / Landscape and Urban Planning 63 (2003) 241–250

Fig. 2. Selected N-S transect through the Flanders region of Belgium (grey strip) overlayed with the Belgian Flora Database 4 km × 4 km grid cells.

all significance values using RT 2.1 (Manly, 1997b). This software uses randomisations in order to calculate significance levels. We used simple and multiple regression algorithms with randomisation of the regression residuals.

3. Results Four PCA axes with an eigenvalue greater than 1 were derived, explaining 85% of the variance (Table 1). PCA 1 correlates with landscape diversity

Table 1 Principal component analysis factor scores on the four derived PC axes after a VARIMAX rotation Component

SDI SEI MSIDI MSIEI % Agriculture SIDI SIEI IJI % Built-up area NUMP PR LPI AWMSI MPI MNN AWMPFD TE MPS MPFD MSI % Forest cover

1

2

3

4

0.902 0.897 0.894 0.883 −0.882 0.876 0.870 0.827 0.777 0.664 0.578 −8.162E−02 −0.354 −0.414 −0.232 −0.531 0.462 −0.634 −7.096E−02 −0.264 0.365

−0.358 −0.335 −0.334 −0.356 0.125 −0.417 −0.420 −0.266 −0.149 −0.597 −0.250 0.873 0.841 0.790 0.777 0.749 −0.711 0.661 −0.105 −8.690E−02 −0.170

−6.450E−02 −6.158E−02 −0.184 5.954E−02 0.246 −5.157E−02 7.568E−03 −0.298 −0.192 −0.309 −0.418 −0.193 −0.125 9.244E−02 −9.301E−03 0.107 0.127 0.285 0.943 0.917 −0.158

0.167 0.215 0.139 0.221 −0.170 0.178 0.215 0.180 −0.169 −0.164 −8.495E−02 −0.235 −0.304 −0.289 0.140 −0.235 −0.331 0.153 −5.681E−02 −0.109 0.708

O. Honnay et al. / Landscape and Urban Planning 63 (2003) 241–250

245

Table 2 Spearman rank correlation coefficients between the PCA factor scores and the abundance of the different species groups PCA 1 Species number Threatened species Native species Neophytes Nitrophobe Nitrophileous Acidophileous Acidophobe Thermophobe Thermophyleous

0.62 0.35 0.58 0.73 0.56 0.60 0.42 0.60 0.33 0.66

PCA 2 ∗∗∗ ∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗ ∗∗∗

−0.31 −0.27 −0.32 −0.19 −0.14 −0.26 −0.06 −0.32 −0.25 −0.23

PCA 3 ∗ ∗ ∗

NS NS NS NS ∗

NS NS

−0.09 −0.16 −0.13 0.05 0.17 −0.13 −0.26 −0.21 0.24 −0.08

PCA 4 NSa NS NS NS NS NS NS NS NS NS

0.15 0.07 0.18 0.06 0.33 0.08 0.34 0.08 0.22 0.11

NS NS NS NS ∗∗

NS ∗∗

NS ∗

NS

a

NS: not significant. 0.01 ≤ P < 0.05. ∗∗ 0.001 ≤ P < 0.01. ∗∗∗ P < 0.001. ∗

measures and with the degree of urbanisation. It is negatively correlated with the percentage agricultural land use. This means that the landscape diversity and complexity increases with urbanisation. The rural agricultural landscape is characterised by low patch

diversity while urban and semi-urban landscapes are associated with a high number of other land use classes (orchards, forest fragments, . . . ) which increases the landscape diversity indices. PCA 2 correlates with the degree of landscape fragmentation. It correlates

Fig. 3. Relation between the total number of plant species in a 4 km × 4 km grid cell and the Shannon Diversity Index of the landscape within the grid cell. The % built-up area, divided in quartiles, is indicated for each grid cell.

246

O. Honnay et al. / Landscape and Urban Planning 63 (2003) 241–250

with the degree of isolation of the different patches (MPI, MNN, IJI) and with landscape measures which reflect the distribution of patch size (MPS, LPI, TE) and area weighted-shape (AWMSI, AWMPFD). PCA 3 correlates with shape indices (MSI, MPFD). The two area-weighted shape indices correlate with the second axis and not with the third because the shape aspect of these measures is strongly affected by patch size, and hence they are more related with the distribution of patch size. PCA 4 correlates with % forest cover in each grid cell. All derived species groups and especially the number of alien plant species are strongly positively correlated with PCA I (landscape diversity/urbanisation) (Table 2). The relation between total species richness, SDI and % built-up area is illustrated in Fig. 3. PCA 2 (degree of fragmentation) negatively correlates with the total number of species, the native species and the threatened species. There are no significant correlations with PCA 3 (patch shape) while there is a positive correlation between PCA 4 (% forest area) and the number of acidity indicators and nitrophobe species. The multiple regression results are presented in Table 3. SDI, MPI, MSI and percentage forest area were selected as independent model variables because they exhibited the highest factor scores on each of the four different PCA axes. Hence, they are uncorrellated and represent a good estimate of the total variation in the data set. Percent built-up area was added to this pool of variables (although it correlates with SDI) because this variable has different physical meaning than the patch diversity and patch distribution measures. Introduction of this variable in the model did not cause problems with multicollinearity and associated variance inflation. The abundance of some specialist species groups mainly depends on the presence of a certain habitat type. This is true for warmth indicators, nitrophobic species (correlated with % built-up area), and cold indicators, acidity indicators and nitrophobic species (correlated with % forest area). The abundance of the derived other plant species groups depends mainly on the landscape structure. The fragmentation measure MPI seems only relevant as a predictor of the threatened species while the patch diversity index SDI appears to be the most important landscape index. Also MSI plays an important

Table 3 Multiple regression results. Dependent variables (Y) are the abundance of the different species groups B

t-value

Significant values

Y: total species number; F = 3.72; (P < 0.001); R2 = 0.54 Constant −1036.87 −2.95 0.004 SDI 211.27 4.89 <0.001 % Built-up 5.81 3.53 0.001 MSI 751.02 3.15 0.002 Y: alien species; F = 82.77; (P < 0.001); R2 = 0.66 Constant −16.18 −2.98 0.004 SDI 18.42 5.37 <0.001 % Built-up 0.64 4.747 <0.001 Y: native species; F= 34.88; (P < 0.001); R2 = 0.45 Constant −580.59 −2.84 0.006 SDI 160.70 8.35 <0.001 MSI 411.04 2.97 0.004 B χ2 Significant values Y: threatened species; χ 2 = 181.20; (P < 0.001); (poisson regression) Constant 0.95 19.35 <0.001 % Built-up 0.03 28.68 <0.001 MPI −0.005 14.31 <0.001 B t-value Significant values Y: cold indicators; F = 17.98; (P < 0.001); R2 = 0.17 Constant 2.68 8.54 <0.001 % Forest 0.16 4.24 <0.001 Y: warmth indicators; F = 67.97; (P < 0.001); R2 = 0.61 Constant −15.25 −2.33 0.022 % Built-up 0.81 5.04 <0.001 SDI 17.03 4.12 <0.001 Y: acidity indicators; F = 68.84; (P < 0.001); R2 = 0.62 Constant 85.07 4.18 <0.001 % Forest 1.17 9.88 <0.001 MSI −54.24 −3.66 <0.001 Y: basophilic species; F = 36.56; (P < 0.001); R2 = 0.56 Constant −526.75 −4.64 <0.001 SDI 73.86 5.29 <0.001 MSI 359.64 4.66 <0.001 % Built-up 1.86 3.49 0.001 Y: nitrophobic species; F = 76.28; (P < 0.001); R2 = 0.64 Constant 11.99 4.95 <0.001 % Forest 1.92 8.71 <0.001 % Built-up 1.03 6.89 <0.001 Y: nitrogen indicators; F = 38.02; (P < 0.001); R2 = 0.47 Constant −233.49 −3.04 0.003 SDI 63.01 8.72 <0.001 MSI 172.35 3.32 0.001 Independent variables (X) are the landscape diversity and land use variables.

O. Honnay et al. / Landscape and Urban Planning 63 (2003) 241–250

role in the total species richness and native species richness.

4. Discussion As could be intuitively expected the total species number within a grid cell is positively correlated with landscape diversity (SDI, MSI). A high patch richness, high patch evenness (both reflected by high SDI) and irregularly shaped patches (high MSI) enhance landscape diversity and hence plant diversity. This is just a spatial extention of the ecological axiom that high heterogeneity within a habitat patch causes high species diversity. Also the positive correlation between % built-up area and species richness was expected. In Central Europe, other authors already found that plant diversity in urban and suburban areas was much higher than in the surrounding rural area (Kowarik, 1990; Pysek, 1993; Kowarik, 1995; but see Roy et al., 1999). Urban areas should contain more species due to better immigration possibilities by antropochory and the presence of more different habitats (Kowarik, 1995). Moreover, our results show that the number of alien plant species is also positively correlated with % built-up area. Alien plant species are better adapted to man-made perturbations, typical for urban and suburban areas (Kowarik, 1990, 1995). On a disturbance scale from no disturbance (1) to extreme disturbance (9) (the so called ‘hemeroby scale’, (Sukkop, 1990)) Kowarik (1990) found the highest neophyte species richness at a disturbance level of 7, while the number of native species peaked at level 3. A lot of the occurring alien plant species are warmth indicators (e.g. Buddleja davidii Franch., Ficus carica L.), which is supported by the fact that % built-up area is the most important predictor for this specialist species group, supporting the idea that cities act like heath islands (Gilbert, 1989; Rebele, 1994). A more surprising outcome of our analyses is that the number of threatened species is positively correlated with % built-up area within a grid cell. There can be several not mutually exclusive explanations for this counterintuitive relationship. Firstly, as already mentioned above, suburban and urban areas contain a high range of different land use classes, all with their specific degree of disturbance and nutrient availability. Hence a high number of potentially suitable niches is

247

available. In terms of Grime’s C-S-R model (Grime, 1979), almost the whole triangular spectrum of habitats for pure competitive species over stress tolerators to ruderals can be found (Gilbert, 1989; Roy et al., 1999). This increases the opportunities for threatened and rare plant species to find a suitable habitat. Secondly, the remaining agricultural land in the suburban area is usually not so intensively managed as in the surrounding rural areas. Less application of agrochemicals and the remaining of hedge relics have a positive effect on the number of threatened species which have less survival chances in the surrounding more intensively managed agricultural part of the rural area. Thirdly, an important number of threatened species (e.g. Ceterach officinarum Willd., Asplenium adiantum-nigrum L.) are species which associated with ancient walls, which are evidently more abundant in the urban and suburban areas while the natural equivalent of this habitat (natural rocky cliffs) is lacking almost entirely in Flanders. Finally, the degree of inventory (inventory effort, measured as the number of species lists made up within a grid cell) of the suburban area may be greater. Since the number of species lists (i.e. the survey effort) made up in a grid cell is positively correlated with the number of inventoried species (Fig. 4), the positive relation of the threatened species with % built-up area may be partly an artefact. In this exploratory study we did not attempt to correct for this possible artefact statistically because we initially selected the most intensively surveyed grid cells of Flanders, we expect no serious bias. The negative correlation between the number of threatened species and the MPI indicates that fragmentation negatively affects the number of species of this category but that this effect is amply compensated by the effect of habitat diversity (SDI). This is a phenomenon that has been described in ecological literature. The effect of habitat diversity is generally much greater than the effect of fragmentation on plant species richness, as has been documented extensively in the so called single large or several small (SLOSS) discussion. Several small, geographically spread habitat patches contain usually more plant species than a few large habitat patches, indicating that the effect of patch area per se is less important than covering as much as possible different habitats. (Peterken and Game, 1984; Game and Peterken, 1984; Dzwonko and Loster, 1989; Margules et al., 1994; Honnay et al., 1999).

248

O. Honnay et al. / Landscape and Urban Planning 63 (2003) 241–250

Fig. 4. Relation between the inventory effort (number of species lists per 4 km × 4 km grid cell) and the recorded total number of species within the cell. Data for the whole of the Flanders region.

The negative effect of habitat fragmentation appears also in the correlation analysis between PCA factor scores and species groups (Table 2) for the total species richness and the native species richness but this effect seems not to be significant in the regression analysis. The identity of the predictors for the abundance of the more specialist species subgroups is more or less straightforward. Forest habitat contains more cold indicators due to the specific micro climate of the forest ecosystem (e.g. Chen et al., 1995). Also the number of acidity indicators and nitrophobes is associated with forest habitat. This is probably due to the fact that in the studied region most forests are situated on sandy, nutrient-poor soils. Basophilic species are associated with the % built-up area due to the high pH value of concrete.

factorily. Landscape indices seem to be useful to predict biotic processes and they appear to have an ecological meaning. Very likely these predictors are region dependent. Verification of the models in other regions of Belgium and/or Europe is necessary, but lies beyond the scope of this exploratory study. At least in the studied region, the effects of landscape changes on plant species diversity can be monitored and predicted at a large scale and over long periods of time using satellite derived landscape data. Bottleneck in this kind of analysis remains the availability and reliability of the biological data. The fact that some areas are better inventoried than others may lead to artefacts in the statistical relations and points at a fundamental problem with area-covering grid based floristic data. Acknowledgements

5. Conclusion Based on relatively simple landscape indices, regional plant species richness can be predicted satis-

Thanks to Hans Jacquemyn for help with the poisson regression model in SAS. This paper is partially based of the MSc. Thesis of K. Piessens.

O. Honnay et al. / Landscape and Urban Planning 63 (2003) 241–250

Appendix A. Land use categories (based on classified Landsat satellite image, 20 m × 20 m pixels) Land use categories marked with (B) are integrated in % built-up area (see Appendix B) Core city (B) Built-up area (B) Open built-up area (B) Reclamation Airport (B) Harbour (B) Infrastructure Watercourse Highway (B) Main road (B)

Arable land Grassland Wet grassland Orchard Broad-leaved forest Conifer forest Mixed forest Heath Mud flat Sweet water

Appendix B. Landscape structure measures (see McGarigal and Marks (1995) for the exact definition) (calculated for 4 km × 4 km grid cells) Patch diversity and distribution measures Shannon diversity index (SDI) Simpson diversity index (SIDI) modified Simpson diversity index (MSIDI) Shannon evenness index (SEI) Simpson evenness index (SIEI) Modified Simpson evenness index (MSEI) Patch richness (PR) Largest patch index (LPI) Patch Shape indices Mean shape index (MSI) Area-weighted mean shape index (AWMSI) Mean fractal dimension (MFD) Area-weighted mean patch fractal dimension (AWMPFD) Fragmentation and Isolation indices Mean nearest neighbour distance (MNND) Mean proximity index (MPI) Interspersion and juxtaposition index (IJI) Mean patch size (MPS) Total edge (TE)

249

Appendix B (Continued ) Land use indices % Cover forest % Cover agriculture % Cover built-up area

Appendix C. Derived species groups Total species number Alien species (i.e. only naturalised species) (Cosyns et al., 1994) Threatened species∗ (Cosyns et al., 1994) Nitrophobic species (Ellenberg N Figs. 1-3, Ellenberg et al., 1990) Nitrogen indicators (Ellenberg N Figs. 6–8) Basophilic species (Ellenberg R Figs. 1-3) Acidity indicators (Ellenberg R Figs. 6–8) Warmth indicators (Ellenberg T Figs. 6–8) Cold indicators (Ellenberg T Figs. 1-3) ∗ Based

on the temporal evolution of the status of the species in the 4 km × 4 km grid before and after 1972 (Van Landuyt et al., 2000). References Baker, W.L., Cai, Y., 1992. The r.le programs for multiscale analysis of landscape structure using the GRASS GIS. Landscape Ecol. 7, 291–302. Burnett, M.R., August, P.V., Brown, J.H., Killingbeck, K.T., 1998. The influence of geomorphological heterogeneity on biodiversity. Part I. A patch-scale perspective. Conservation Biol. 12, 363–370. Butaye, J., Jacquemyn, H., Hermy, M., 2001. Differential colonisation causing non-random forest plant community structure in a fragmented agricultural landscape. Ecography 24, 369–380. Chen, J., Franklin, J.F., Spies, T.A., 1995. Growing-season microclimatic gradients from clear cut edges into old-growth Douglasfir forests. Ecol. Appl. 5, 74–86. Collingham, Y.C., Wadsworth, R.A., Huntley, B., Hulme, P.E., 2000. Predicting the spatial distribution of non-indigenous riparian weeds: issues of spatial scale and extent. J. Appl. Ecol. 37, 13–27. Cosyns, E., Leten, M., Hermy, M., Triest, L., 1994. Een statistiek van de wilde flora van Vlaanderen. Vrije Universiteit Brussel, Brussels. Delanghe, J.E., Delvosalle, L., Duvigneaud, J., Lambinon, J. Vanden Berghen, C., 1988. Flora van België, het Groothertogdom Luxemburg, Noord-Frankrijk en de aangrenzende

250

O. Honnay et al. / Landscape and Urban Planning 63 (2003) 241–250

gebieden (pteridofyten en spermatofyten). Nationale Plantentuin van België, Meise. Dufourmont, H., Gulinck, H., Janssens, K., 1996. Ontwikkeling van een gebiedsdekkende informatielaag afgeleid uit satellietbeelden als basis voor monitoring en structuurkartering van het landelijk gebied in Vlaanderen. Eindeverslag Project Telsat 3/01/40, Nationaal Onderzoeksprogramma inzake Teledetectie per Satelliet. Leuven. Dzwonko, Z., Loster, S., 1989. Distribution of vascular plant species in small woodlands on the Western-Carpathian foothills. Oikos 56, 77–86. Ellenberg, H., Weber, H.E., Düll, R., Wirth, V., Werner, W., Paulissen, D., 1990. Zeigerwerte von Pflanzen in Mitteleuropa. Scripta Geobotanica 18, 1–248. Forman, R.T.T. en Godron, M., 1986. Landscape ecology. Wiley, New York. Frohn, R.C., 1998. Remote sensing for Landscap e ecology. New metric indicators for montoring, modeling, and assessment of ecosystems. Lewis Publishers, USA. Game, M., Peterken, G.F., 1984. Nature reserve selection strategies in the woodlands of Central Lincolnshire, England. Biol. Conserv. 29, 157–181. Gilbert, O.L., 1989. The Ecology of Urban Habitats. Chapman & Hall, London. Grime, J.P., 1979. Plant strategies and vegetation processes. Wiley, New York. Grashof-Bokdam, C.J., 1997. Forest plants in an agricultural landscape in the Netherlands: effects of habitat fragmentation. J. Vegetation Sci. 8, 21–28. Gulinck, H., Dufourmont, H., Coppin, M., Hermy, M., 2000. Landscape research, landscape policy and earth observation. Int. J. Remote Sensing 21, 2541–2554. Hargis, C.D., Bissonette, J.A., David, J.L., 1998. The behaviour of landscape metrics commonly used in the study of habitat fragmentation. Landscape Ecol. 13, 167–186. Hobbs, E.R., 1988. Species richness of urban forest patches and implications for urban landscape diversity. Landscape Ecol. 1, 141–152. Honnay, O., Hermy, M., Coppin, P., 1999. Effects of area, age and diversity of forest patches in Belgium on plant species richness, and implications for conservation and reforestation, and implications for conservation and reforestation. Biol. Conserv. 87, 73–84. Kowarik, I., 1990. Some responses of flora and vegetation to urbanisation in Central Europe. In: Sukopp, H., Hejny, S., Kowarik, I. (Eds.), 1990. Plants and Plant Communities in Urban Environments. SPB Academic Publishing, Den Haag, The Netherlands, pp. 45–74. Kowarik, I., 1995. On the role of alien species in urban flora and vegetation. In: Pysek, P., Prach, K., Rejmanek, M., Wade, M., 1995. Plant invasions–General Aspects and Special Problems. SPB Academic Publishing, Amsterdam, The Netherlands, pp. 85–103. Lapin, M., Barnes, B.V., 1995. Using the landscape ecosystem approach to assess species and ecosystem diversity. Conserv. Biol. 9, 1148–1158. Manly, B.F.J., 1997a. Randomization, bootstrap and Monte-Carlo Methods in biology. Chapman & Hall, New York.

Manly, B.F.J., 1997b. RT a program for randomization testing vs. 2.1. Western EcoSystems Technology Inc., Cheyenne, WY. Margules, C.R., Nicholls, A.O., Usher, M.B., 1994. Apparent species turnover, probability of extinction and the selection of nature reserves: a case study of the Ingleborough limestone pavements. Conserv. Biol. 8, 398–409. McGarigal, K., 2000. Fragstats Arc: spatial pattern analysis program for quantifying landscape structure: manual. http://blaze. innovativegis.com/products/fragstatsarc/manual/. McGarigal, K., Marks, B., 1995. Fragstats: spatial pattern analysis program for quantifying landscape structure. USDA Forest Service, Pacific Northwest Research Station. Nichols, W.F., Killingbeck, K.T., August, P.V., 1998. The influence of geomorphological heterogeneity on biodiversity. Part II. A landscape perspective. Conserv. Biol. 12, 371–379. O’Neill, R.V., Krummel, J.R., Gardner, R.H., Sugihara, G., Jackson, B., DeAngelis, D.L., Milne, B.T., Turner, M.G., Zygmunt, B., Christensen, S.W., Dale, V.H., Graham, R.L., 1988. Indices of landscape pattern. Landscape Ecol. 1 (3), 153– 162. Peterken, G.F., Game, M., 1984. Historical factors affecting the number and distribution of vascular plant species in the woodlands of central Lincolnshire. J. Ecol. 72, 155–182. Pysek, P., 1993. Factors affecting the diversity of flora and vegetation in central European settlements. Vegetation 106, 89–100. Rebele, F., 1994. Urban ecology and special features of urban ecosystems. Global Ecol. Biogeography Lett. 4, 173–187. Rempel, R.S., Carr, A., Elkie, P., 1999. Patch analyst and patch analyst (grid) function reference. Centre for Northern Forest Ecosystems Research, Ontario Ministry of Natural Resources, Lakehead University, Thunder Bay, Ontario. Rescia, A.J., Schmitz, M.F., Martin de Agar, P., de Pablo, C.L., Pineda, F.D., 1997. A fragmented landscape in northern Spain analysed at different spatial scales: Implications for management. J. Vegetation Sci. 8, 343–352. Roy, D.B., Hill, M.O., Rothery, P., 1999. Effects of urban land cover on the local species pool in Britain. Ecography 22, 507– 515. Shaffer, M.L., 1981. Minimum population sizes for species conservation. BioScience 31, 131–134. Turner, M.G., 1989. Landscape ecology: the effect of pattern on process, Annual Review of Ecology. Annual Review of Ecology and Systematics 20, 171–197. Van Landuyt, W., Heylen, O., Vanhecke, L., Van den Bremt, P., Baeté, H., 2000. Verspreiding en evolutie van de botanische kwaliteit van ecotopen gebaseerd op combinaties van indicatorsoorten uit Florabank. Flo.Wer, Institute of Nature Conservation, Brussels. O. Honnay is post-doctoral research fellow with the Flanders Fund for Scientific Research. He obtained his PhD from the Faculty of Applied Biological Sciences of the University of Leuven, Belgium, in 2000 with work on the spatial and temporal distribution of forest plant species in a fragmented landscape. He currently, studies the effects of habitat fragmentation on the demographic and genetic characteristics of forest plant species populations.