An integrated analysis of the effects of past land use on forest herb colonization at the landscape scale

Authors


Kris Verheyen, Laboratory for Forest, Nature and Landscape Research, University of Leuven, V. Decosterstraat 102, B-3000 Leuven, Belgium (e-mail kris.verheyen@agr.kuleuven.ac.be).

Summary

  • 1A framework that summarizes the direct and indirect effects of past land use on forest herb recolonization is proposed, and used to analyse the colonization patterns of forest understorey herbaceous species in a 360-ha mixed forest, grassland and arable landscape in the Dijle river valley (central Belgium).
  • 2Fine-scale distribution maps were constructed for 14 species. The species were mapped in 15 946 forest plots and outside forests (along parcel margins) in 5188 plots. Forest stands varied in age between 1 and more than 224 years. Detailed land-use history data were combined with the species distribution maps to identify species-specific colonization sources and to calculate colonization distances.
  • 3The six most frequent species were selected for more detailed statistical analysis.
  • 4Logistic regression models indicated that species frequency in forest parcels was a function of secondary forest age, distance from the nearest colonization source and their interaction. Similar age and distance effects were found within hedgerows.
  • 5In 199 forest stands, data about soils, canopy structure and the cover of competitive species were collected. The relative importance of habitat quality and spatio-temporal isolation for the colonization of the forest herb species was quantified using structural equation modelling (SEM), within the framework proposed for the effects of past land use.
  • 6The results of the SEM indicate that, except for the better colonizing species, the measured habitat quality variables are of minor importance in explaining colonization patterns, compared with the combination of secondary forest age and distance from colonization sources.
  • 7Our results suggest the existence of a two-stage colonization process in which diaspore availability determines the initial pattern, which is affected by environmental sorting at later stages.

Introduction

A better understanding of colonization processes is critical to understanding the general patterns and processes in plant communities. Grubb (1977) was among the first to challenge the traditional view of niche-assembled plant communities. By elaborating on the different processes involved in colonization (the production of viable diaspores, dispersal in space and time, germination requirements, establishment of seedlings and the further development of the immature plant), Grubb stated that, at the local scale, differential colonization capacities of plant species are of crucial importance for the coexistence of large numbers of species with similar resource requirements. Twenty years later, a growing body of literature confirms Grubb's thesis. It has been demonstrated that dispersal is a key factor in community assembly. Evidence comes from observational studies (e.g. Hubbell et al. 1999), experimental studies (e.g. Turnbull et al. 2000) and mathematical models (e.g. Tilman 1994).

Most data concern established communities perpetuated by small-scale, low-intensity disturbances (i.e. gap-phase dynamics). However, there is no reason to believe that fundamentally different processes operate when larger spatial scales and more intensive disturbances are involved (Bazzaz 1996). Although our understanding of the theoretical framework of succession has been significantly improved (e.g. Connel & Slatyer 1977; Huston & Smith 1987), the possibility that a successional sere could be generated by species having different dispersal capacities remains largely unexplored. Recently, however, a simulation study by Hovestadt et al. (2000) demonstrated that typical successional patterns can be generated by differential dispersal alone.

Forest herb communities develop slowly in secondary forest established on former agricultural land. It has been demonstrated that some species do not colonize secondary forests even after tens to hundreds of years, while others are able to recolonize quickly (e.g. Peterken & Game 1984; Motzkin et al. 1996; Wulf 1997; for recent overviews see Hermy et al. 1999 and Verheyen et al. 2003). A niche-based approach would suggest that these highly different colonization capacities are mainly a consequence of recruitment being hampered by the poor habitat quality in young forests. The relationship observed between habitat quality and past land use depends both on the nature and duration of the former land use and on processes related to aggrading forest ecosystems. For instance, following arable use, levels of mineral nutrients, especially phosphate, generally increase (Koerner et al. 1997; Honnay et al. 1999). In conjunction with a still open canopy, these promote the vigorous growth of competitive species (sensuGrime et al. 1988) such as Urtica (Pigott 1971), which may, in turn, hamper the recruitment of forest species. More recently, however, both observational studies (Matlack 1994; Grashof-Bokdam 1997; Brunet & von Oheimb 1998; Bossuyt et al. 1999; Butaye et al. 2001; Dzwonko 2001; Singleton et al. 2001; Verheyen & Hermy 2001a,b) and experimental studies (Ehrlén & Eriksson 2000) have demonstrated that some form of dispersal limitation can also occur in secondary forest succession.

We propose a model to explain the effects of past land use on both environmental and canopy variables and forest herb colonization (Fig. 1). Past land use can have direct effects on herb colonization (from the physical removal of the plant species and the subsequent spatial and temporal isolation from their respective colonization sources) as well as indirect effects caused by its impact on habitat quality variables (e.g. on the cover of competitive species or on soil nutrient levels). In this study, structural equation modelling (SEM) is introduced to quantify these direct and indirect effects, and thus to perform a novel integrated analysis of the colonization process. Furthermore, our study extends work at the local scale (i.e. one forest patch), to the landscape or land mosaic scale (i.e. forest patches embedded in a non-forested, agricultural matrix).

Figure 1.

Model hypothesizing the effects of past land use on forest herb species colonization. COLONIZATION (the frequency of colonizing forest herbs) is a latent variable, while the variables in octagons are statistical composites: LAND USE = land-use history and spatial isolation, SOIL = soil conditions and CANOPY = canopy structure. Variables shown in boxes are the measured variables: ‘forest age’= secondary forest age; ‘distance’= species-specific distances from colonization sources; ‘texture’= soil texture; ‘moisture’= soil moisture; ‘pH’= soil pH; ‘shrub BA’ = log-transformed basal area of the shrubs; ‘cover competitives’ = the cover of competitive herbs and ‘colonization frequency’ = ln-transformed frequency of the colonizing forest herbs. Double-headed arrows represent correlations between independent variables, while single-headed arrows describe assumed causal relationships.

Three specific questions are addressed. (i) Is spatio-temporal isolation from diaspore sources a limiting factor for forest herb colonization at a landscape scale? (ii) What are the direct and indirect effects of past agricultural land use on present-day habitat quality of forests? (iii) What is the relative importance of these two factors (isolation and quality) for landscape scale colonization?

Materials and methods

study area

The study was carried out in the 360 ha ‘Doode Bemde’ nature reserve (Fig. 2), 20 km east of Brussels in central Belgium. The whole study area is located in the valley bottom of the Dijle river, a medium-sized lowland river, that has cut deep (c. 40 m) into the surrounding plateau during past glaciations. Large-scale deforestation and several thousand years of human settlement have led to river discharge being increasingly influenced by surface run-off. The gradual deposition of eroded silt sediments resulted in relatively elevated levees with alluvial floodplain depressions. The soils in the depressions are fine textured, base rich, poorly drained Fluvisols, while levee soils are dryer and relatively coarse textured Fluvisols (De Becker et al. 1999).

Figure 2.

Map of the study area (the 360-ha ‘Doode Bemde’ nature reserve) showing the 1999 land use, the location of the hedgerows (i.e. parcel margins composed of trees and/or shrubs indicated by bold lines) and the location of sampling points for environmental and canopy variables (circles). The scale is approximately 1 : 17000.

At present, almost half (150 ha) of the study area is forested and Populus x canadensis (Moench.) is the dominant canopy species in almost all stands. From the 1950s onwards, the tree layer in most of the existing forests was cut and replanted with poplars, which were also widely planted on abandoned agricultural land. However, the wide (10 m × 10 m) spacing used allowed the spontaneous establishment of other tree and shrub species and hence many forests have a well-developed subcanopy characterized by Alnus glutinosa, Coryllus avellena, Fraxinus excelsior and Quercus robur. The non-forested parts of the valley are almost all grasslands and arable land is relatively uncommon (Fig. 2). The forests on the better drained sites are rich in early flowering forest species characteristic of mesic, rich sites, while alder carr develops on the wettest sites. Nomenclature follows Lambinon et al. (1998).

mapping of the plant species

During April and May 1999, the distribution of 14 typical forest understorey herbs was mapped. Verheyen & Hermy (2001a,b) provide more details about the selection of fast and slow colonizing species. The latter ‘ancient forest species’ recolonize forests after agricultural abandonment at very slow rates (cf. Hermy et al. 1999). In the forest parcels, the species were fully mapped by walking along parallel transects approximately 10 m apart and scoring every 10 m the presence/absence of the species in a circular plot of 5 m radius. A compass was used where the regular planting pattern of the poplars did not allow alignment of transects. Outside the forests, only the parcel margins were mapped, to include both hedgerows (i.e. margins composed of trees and/or shrubs) and non-woody margins (fences, ditches, roadsides). Again, the presence/absence of the species was scored every 10 m. This resulted in 21 134 plots, of which 15 946 were situated in forest parcels and 5188 along parcel margins. The location of all transects was marked on large scale (1 : 2500) land register maps in the field and all plots were later digitized as points using ArcView 3.1 (ESRI 1998). This method allowed the distribution of plant species in a 360-ha landscape to be mapped relatively quickly and accurately.

ecological factors: historical land use

Land use was reconstructed using historical maps (1775, 1850, 1865, 1891, 1936) and aerial photographs (1947, 1969, 1980, 1990, 1995), with scales varying between 1 : 12500 and 1 : 40000, which are all available from the National Geographic Institute of Belgium. In addition, the primitive land register (from 1834 onwards) and an historical work (Martens 1994) were consulted. Large-scale maps of 1665, 1759 and 1769 that covered only parts of the study area were used to verify the 1775 map. Land use was categorized as forest or non-forest at each date. Throughout the historical period, non-forested land was predominantly grassland. Land-use changes between 1775 and 1999 are shown in Fig. 3. Forested area was very low, with a minimum in the middle of the 19th century, until a sharp increase following World War II.

Figure 3.

Land use changes in the Dijle valley study area between 1775 and 1999. Areas are expressed in ha. Diamonds indicate forest, squares ancient forest and triangles non-forest.

We then delineated parcels within which there was a common land use at each survey date. For each, we calculated the number of years since reforestation, assuming that any land-use changes occurred midway between the dates of two maps (cf. Verheyen et al. 1999). Five classes were distinguished, with ages of 1–24 years, 41 years, 58 years, 86–195 years, and more than 224 years. At present, the extent of the oldest class is only 15 ha (Figs 3 and 4), but as such stands are probably much older than 224 years, they are referred to as ancient forest. To distinguish possible relict populations from newly colonized populations of forest plant species along woody parcel margins, we identified ancient parcel margins. These were defined as margins that (i) were present as woody margins on the 1775 map, (ii) were mapped as boundaries on the 1834 land register, and (iii) still exist today (Fig. 4). This definition allowed us to account for ancient parcel margins that are actually part of a forest. Next, we tried to assess the age of the hedgerows present in 1999. However, except for 1775, it was very hard to distinguish hedgerows on the historical maps. Therefore, only the five aerial photographs were used to determine the hedgerow age, classifying them with ages of 1–24 years, 41 years and more than 58 years.

Figure 4.

Map showing the distribution of forest age classes and the ancient parcel margins, based on historical maps and aerial photographs. The scale is approximately 1 : 17000.

ecological factors: spatial isolation

The combination of land-use history data and species distribution maps allowed us to identify species-specific colonization sources (cf. Peterken & Game 1981). These were defined as plots in which a particular species was present and that were situated either in ancient forest parcels or within 10 m of an ancient parcel margin. We refer to Verheyen & Hermy (2001a) for a more detailed description of the method and the assumptions involved. Hence, 14 sets of species-specific colonization sources were obtained (see example in Fig. 5).

Figure 5.

Distribution map of Adoxa moschatellina, indicating the species-specific colonization sources, the colonized points and the non-colonized points. The scale is approximately 1 : 17000.

Using the ArcView NearFeat extension (Jennes 2000), the distance from all points (occupied and non-occupied) to the five nearest colonization sources was calculated for each species. The average of the five distances was calculated and grouped into six consecutive classes: < 100 m, 100–199 m, 200–299 m, 300–399 m, 400–499 m and > 500 m.

ecological factors: environmental and canopy variables

The environmental and canopy data were collected during July and August 1999 at 503 sampling points distributed over 199 forest stands at a density of approximately four points/ha (Fig. 2). Forest stands were defined as parcels with both a homogeneous land-use history and canopy structure. The average stand area was 0.6 ha, with a minimum of 0.1 ha and a maximum of 5 ha. At each sampling point, data on soil, the canopy structure and the competitive species were collected.

After removing the litter, 10 × 5 cm deep soil samples were taken within a 5-m radius around each sampling point and bulked per stand. In the laboratory, the pH (KCl) 1 N, 1 : 5 suspension was determined after the samples had been air-dried. The canopy was described using a modified point-centred quarter method (Bonham 1989). The tree layer was subdivided in four diameter classes: < 15 cm, 15–30 cm, 30–45 cm, and > 45 cm. Four 90° sectors around each sampling point were established. In each sector and for every diameter class, the distance to the nearest tree or shrub was measured. The average diameter of each class (7.5 cm, 22.5 cm, 37.5 cm and 52.5 cm, respectively), was used to calculate its basal area (m2 ha−1). The omnipresent homogeneous poplar canopy meant that only the smallest diameter class showed significant variation and therefore only this class (the shrubs) was used for further analysis. Finally, the combined percentage cover of tall competitive herbs (sensuGrime et al. 1988) was estimated within a radius of 5 m around the sampling point using the following cover classes: 0%, < 5%, 6–10%, 11–15%, 16–20%, 21–40%, 41–60%, 61–80%, and > 81%. Urtica dioica was by far the most dominant competitive herbaceous species, but Angelica sylvestris, Galium aparine, Filipendula ulmaria, Phragmites australis and Symphytum officinale also occurred frequently.

For all environmental and canopy variables, average values at the forest stand level were calculated. Information about the soil texture and soil moisture regime was extracted from the digital Belgian Soil Map (1 : 20 000) for each stand. Soil texture and moisture were assigned scores on an ordinal scale and an area-weighted texture and moisture score was calculated for each stand. Scores for soil texture ranged from 1 to 3 (1 = loamy sand, 2 = loam, 3 = clay) and from 1 to 4 for soil moisture regime (1 = relatively well drained, 2 = imperfectly drained, 3 = poorly drained, 4 = very poorly drained). An overview of the environmental and canopy variables is given in Table 1.

Table 1.  Overview of the environmental and canopy variables. See text for more details of measurement
VariableMeanRangeTransformation
Soil moisture2.79  1–4None
Soil texture2.08  1–3None
Soil pH4.973.3–7.2None
Basal area shrubs (m2 ha−1)3.04  0–35Log (x + 0.1)
Cover competitive herbs (%)62.95  3–90None

statistical analyses

A necessary first step of the statistical analysis was the assessment of the species’ association with forests, as all the subsequent analyses rely on the assumption that the studied species occur predominantly in forest habitat. Therefore, the fraction of species occurrences in non-woody parcel margins vs. the occurrences in forest parcels and hedgerows was calculated. Statistical significance was assessed with a χ2 statistic and species not exhibiting a significant positive association with forests and woody parcel margins were excluded.

For the remainder of the statistical analysis, plots located in ancient forest parcels and along ancient parcel margins were omitted, as they do not provide any information about the colonization process. In the first part of the analyses data at the plot level are used, whereas in the second part data at the forest stand level are used.

  • • To explore the spatio-temporal colonization patterns, a multiple logistic regression of the individual species presence/absence scores against secondary forest age, average distance from the colonization source and the age–distance interaction was performed (cf. Verheyen & Hermy 2001a). Inclusion of the interaction allowed us to assess whether the importance of distance from the colonization source decreases with increasing secondary forest age. The resulting ‘odds ratios’ indicate how much more or less likely it is that a species is present when the respective variables are increased by one class. Statistical analyses were done separately for the plots in the forest parcels and along the hedgerows. Plots along non-woody parcel margins, where colonization by forest herbs is unlikely, were excluded. All analyses were performed with SPSS 10.0®.
  • • Structural equation modelling (SEM) (e.g. Maruyama 1998) was used to quantify the relative importance of spatio-temporal isolation and habitat quality for the colonization of the different forest species. SEM compares observed covariances among variables with those expected from hypothesized relationships. Model fit to the data is assessed with a χ2 test between the expected and observed covariance matrices. Previous studies of colonization processes led to our hypothesized model (Fig. 1), which includes observed ‘indicator variables’ (boxes), ‘composite variables’ (octagons) and a ‘latent variable’ (ellipse). Statistical composites (Fornell & Cha 1994; Grace & Jutila 1999; Grace et al. 2000) were constructed in order to examine the collective effects of landscape, soil and canopy variables, using a multistage least squares approach. Each composite was constructed individually so as to maximize the variance explained for colonization frequency. The SAS® Proc Reg procedure was used to generate a predictive equation and predicted scores were used for the composite values. LAND USE represents the composite effects of the secondary forest age and the distance to the species-specific colonization sources; SOIL represents the composite effects of soil moisture, soil moisture and soil pH and CANOPY represents the composite effects of the basal area of the shrubs and the cover of competitive herbs.

The possible intercorrelation between LAND USE and SOIL is represented in Fig. 1 by double-headed arrows. The use of the latent variable COLONIZATION is based on the premise that the observed frequency of forest herbs in the secondary forest parcels is measured with error and provides an imperfect estimate of the underlying variable of conceptual interest. Following the convention of SEM, latent variables are presumed to be the underlying causal influence on observed values, which is the reason that the arrow points from the latent variable to the indicator. All SEM analyses were done using LISREL 8.30 (Jöreskog & Sörbom 2000).

Results

The typical distribution pattern of the forest understorey herbaceous species is represented by Adoxa moschatellina (Fig. 5). Three species (Aegopodium podagraria, Ajuga reptans and Stellaria holostea) were not significantly associated with forests and woody parcel margins and thus were excluded from further analysis (Fig. 6a). A large number of the species occurrences were relict populations (i.e. situated in ancient forest or along ancient parcel margins) and only six species were frequent enough in recent forests and three species in recent margins to use for the remainder of the statistical analyses (Fig. 6b).

Figure 6.

(a) Percentage of the species occurrences in forests and hedgerows vs. non-woody parcel margins. The species marked with an asterisk are not significantly associated with woody habitats. (b) The proportion of species occurrences situated in ancient forests and ancient parcel margins (i.e. the colonization sources). Only the species marked with an asterisk were frequent enough in recent forests to use for the remainder of the statistical analyses. Values adjacent to the bars indicate the absolute number of occurrences.

All logistic regression models were highly significant, but the explained variation varied considerably between species (Table 2). Both age and distance from the species-specific colonization sources were important factors explaining the colonization patterns (Table 2 and Fig. 7). In the forest stands, the frequency of all species, except Geum urbanum, consistently increased with increasing secondary forest age and decreased with increasing distance from the colonization source (Table 2a and Fig. 7a). Nevertheless, it also appears that there is a large variation in colonization potential among these species. Some species (e.g. Lamium galeobdolon) only colonized older forest parcels lying close to the colonization sources, while others (e.g. Ranunculus ficaria) also colonized points in the youngest forest age class and the furthest distance class. Furthermore, most age–distance interactions were significant (Table 2a), indicating that the distance effect is not the same for all the age classes. For species with an interaction odds ratio larger than one (A. moschatellina, L. galeobdolon and, to a lesser extent, G. urbanum), the distance effects disappeared as the secondary forest age increased, suggesting a gradual completion of the colonization process (Fig. 7a). The opposite pattern, i.e. a species exhibiting only distance effects in the oldest age classes, was found for R. ficaria, Primula elatior and, although not significant, Arum maculatum (Table 2a and Fig. 7a). Somewhat similar spatio-temporal colonization patterns are observed in the hedgerows (Tables 2b and Fig. 7b). For the few species that colonized hedgerows, frequency also increased with increasing hedgerow age and decreased with increasing distance from the colonization sources, except for A. moschatellina.

Table 2.  Multiple logistic regression of the species presence/absence scores on secondary forest age, average distance from the colonization source and the interaction between these variables. Analyses were done separately for plots located in forests (a) and for plots along hedgerows (b)
SpeciesnNagelkerke R2*Odds ratios
AgeDistanceDistance × age
  • *

    χ2 test statistic;

  • †Wald test statistic;

  • ****

    P≤ 0.0001;

  • ***

    P≤ 0.001;

  • **

    P≤ 0.01; *P≤ 0.05; (*)P≤ 0.1; NS = not significant.

(a) Forests
Adoxa moschatellina14 2200.42****1.18*0.09****1.75****
Arum maculatum14 3050.23****2.30****0.48***0.96 NS
Geum urbanum14 3230.09****1.50****0.73****1.07**
Lamium galeobdolon14 3660.52****2.90****0.21****1.27**
Primula elatior14 3230.21****2.09****0.61**0.86*
Ranunculus ficaria13 8400.05****1.96****0.83***0.95*
(b) Hedgerows
Adoxa moschatellina10280.14****0.45 NS0.09**2.64***
Geum urbanum10230.10****3.70 NS0.68 NS0.90 NS
Ranunculus ficaria10060.08****2.90**1.28 NS0.66*
Figure 7.

The species’ frequency (%) in each age and distance class (grey circles) and in each age-distance class combination (open circles). Colonization patterns (a) in forest stands and (b) in hedgerows.

Most of the environmental and canopy variables had a relatively wide range (Table 1). Only soil texture and, to a lesser extent, soil drainage exhibited less variation: most soils have a loamy texture and are relatively poorly drained. The results of the SEM analyses are given in Figs 8 and 9. For each species modelled, non-significant P-values associated with the χ2 tests indicate that the covariance structure in the data fitted well to the hypothesized covariance structure (Fig. 8). The effect of LAND USE on CANOPY was consistently strong for all species, while the effect of SOIL on CANOPY was weaker or absent (Fig. 8). LAND USE had a strong effect on the frequency of the colonizing forest herbs in secondary forest parcels (Figs 8 and 9). However, this effect was less pronounced for the relatively good colonizing species G. urbanum and R. ficaria (cf. Figs 6b and 7) and SOIL appeared to be equally important for the latter species. CANOPY had minor to no effect on colonizing forest herbs.

Figure 8.

Models for the six study species showing χ2 model fit statistics and standardized partial correlation and regression coefficients. See Fig. 1 for symbols. Dashed lines refer to relationships that were non-significant. R2-values associated with COLONIZATION describe the total amount of variance explained.

Figure 9.

Total effects of the composite variables on the frequency of colonizing forest herbs. The total effect of a variable is the sum of all direct and indirect pathways (see Fig. 8), standardized with a maximum possible of 1.0.

Discussion

We attempted to perform an integrated study of the factors controlling forest herb colonization during secondary forest succession in a recently reforested valley landscape dominated by relatively simple structured poplar plantations. It has long been recognized that past agricultural land use can have multiple, but very often correlated, effects on the recolonization of forest herbs. However, for the first time, a formalized framework is proposed that quantifies the relative import-ance of its direct and indirect effects.

spatio-temporal distribution patterns

Verheyen & Hermy (2001a) discuss the assumptions made in delimiting species-specific colonization sources. However, as most of the forests in the Dijle valley are situated on former grasslands, the assumption that none of the studied species survived in the pre-forest grasslands needs further consideration. At present, no populations of the studied species occur in the few relicts of semi-natural grasslands in the Dijle valley. Furthermore, 164 of the 180 recent forest stands were established after World War II, when intensive farming practices (e.g. the application of chemical fertilizers) had already been introduced. Such practices severely reduce the survival chances of forest herbs in grasslands. Finally, none of the studied species forms a persistent seed bank (e.g. Bossuyt & Hermy 2001).

The high number of species occurrences in ancient forests and ancient margins compared with recent forests (Fig. 6b) indicates the low colonizing capacities of most species. Large variation in the size of the relict populations further suggests large differences in colonization source strength among the different species. Interestingly, some of the source populations occur in ancient margins, suggesting that they have persisted for at least 200 years. Since 1775, these populations have thus survived long-term adjacent agricultural use (predominantly as grassland), margin management (e.g. pollarding, coppicing) and, in some cases, even complete hedgerow clearance. These results are consistent with the findings of Pollard (1973) and Peterken & Game (1981) in England, who demonstrated long-term survival of Mercurialis perennis in hedgerows. In North America, relict populations of forest species also have been observed in hedgerows (Fritz & Merriam 1993). Survival chances would probably have been much smaller if intensively farmed arable fields rather than grasslands had bordered the margins (Boutin & Jobin 1998). Indirectly, our results support the assumption made by Honnay et al. (1999) and Verheyen & Hermy (2001a), who deduced from the present-day forest plant species distributions in two forests, that forest plant species had survived in (woody) margins during periods of past agricultural use.

Like other authors (e.g. Matlack 1994; Brunet & von Oheimb 1998; Bossuyt et al. 1999; Singleton et al. 2001), we assume that the observed decline in frequency of species with increasing distance from the colonization source can be attributed to distance-restricted seed dispersal. For the forest parcels, it appears that every species exhibits an age and distance effect (Tables 2a and 7a). However, distance effects for species such as G. urbanum and R. ficaria are only present in the youngest forest parcels and therefore dispersal cannot be regarded as a factor limiting colonization. In contrast, limited dispersal is an important factor in shaping the distribution of the other species throughout the landscape. In a comparable study in a 10 × 10 km landscape elsewhere in central Belgium, Butaye et al. (2001) found similar dispersal limitation in A. moschatellina, L. galeobdolon and P. elatior. As in the present study, no significant distance effects were found for R. ficaria but, in contrast, the distance-effect was also significant for G. urbanum. Distance-dependent colonization at the landscape scale was also found in the Netherlands by Grashof-Bokdam (1997).

The similar distance-dependent decline of species frequency in hedgerows suggests that they have a conduit corridor function (sensuCorbit et al. 1999) in our study area. However, as only a few species have significantly colonized the margins, we can assume that the woody parcel margins act as a filter that discriminates among the species that move along them. These findings support those of McCollin et al. (2000), who found that some forest species are less frequent in hedgerows than in forests. This is not surprising, as hedgerows represent a rather different habitat (e.g. Fritz & Merriam 1993, 1996). Furthermore, more or less the same age-distance effects are observed as with colonization in forest stands. Corbit et al. (1999) also observed a distance effect in hedgerows, implying colonization from an adjacent attached stand. However, unlike water dispersal of riparian flora, for example (e.g. Honnay et al. 2001), it remains to be demonstrated whether hedgerows are functional corridors for forest plant species.

Finally, it is noteworthy that the spatio-temporal colonization patterns presented here are very similar to the patterns found in a methodologically identical study covering the same time-scale but a smaller spatial scale (34 ha; Verheyen & Hermy 2001a). Differences were mainly found for moderately good colonizing species like A. moschatellina that did not exhibit distance effects at the local scale, but strong distance effects at the landscape scale. Hence, it can be deduced that in this case similar diffusion-like processes dominate the colonization process both at the local and at the landscape scale. However, in less forested landscapes than the Dijle valley (forest covers 42% of the area), this may not be the case and rare long-distance dispersal events will probably play a more important role in the colonization process.

relative importance of spatio-temporal isolation and habitat quality

It is interesting to note that the joint age-distance effect is by far the most important factor that explains the species colonization patterns (Fig. 9). Our results suggest that dispersal limits the forest plant species distribution at a landscape scale, which confirms the findings of Ehrlén & Eriksson (2000) and Butaye et al. (2001). As our analysis accounts both for spatial isolation and for indirect effects of secondary forest age, an effect of time per se may be to cause slow colonization processes. Hence, it is likely that the delayed colonization of forest plants is not only caused by poor diaspore dispersal, but also by factors such as limited and/or delayed diaspore production or diaspore predation (cf. Verheyen et al. 2003). Indeed, the colonization patterns of the study species cannot be explained by taking into account only their dispersal strategy. For instance, the diaspores of the good colonizing R. ficaria lack adaptations for long-distance dispersal, while the slow colonizing A. maculatum produces bird-dispersed fleshy fruits. However, A. maculatum only starts producing seeds after 6 years (Sowter 1949), and the seeds do not germinate easily (personal observation). By contrast, R. ficaria has already started producing easily germinating, vegetative bulbils by the second year (Taylor & Markham 1978).

In conclusion, we think that the importance of the measured environmental and canopy variables in the SEM models for G. urbanum and R. ficaria (i.e. species that exhibit relatively good colonizing abilities), suggests that colonization is a two-stage process in which diaspore availability determines the initial colonization pattern and in which the effects of environmental sorting become apparent at later stages. In areas with more pronounced environmental gradients than in our study area, the sorting effects will probably become apparent earlier. To strengthen these ideas, future models should include other key habitat quality variables, such as soil nutrient content, canopy composition and litter depth. For instance, it has already been demonstrated that litter accumulation can hamper recruitment in non-mull ecosystems (Staaf 1992; Eriksson 1995).

Acknowledgements

The authors thank Eric van Beek for assistance with the field work and Bea Bossuyt, Hans Jacquemyn, George Peterken, Jim Grace and three anonymous reviewers for their useful comments. The paper was written while the first author held a grant from the Flemish Institute for the encouragement of Scientific and Technological Research (I.W.T.).

Ancillary