Our study site is located at Lauris (44°44′N; 5°18′E), a village in southeastern France sited on a calcareous plateau on the banks of the Durance River. The region has a Mediterranean climate characterized by cool winters, irregular precipitation (between 700 and 1000 mm per year) and pronounced summer drought that limits the growing season to brief periods in spring and autumn [22].

In this study site, as in the Provencal rural hinterland, the abandonment of cultivated areas and pasture has been responsible since the 1950s for the development of fallow lands and forest fragments (23% more wooded area in France between 1965 and 1985 [23]). The evergreen brushwood of Quercus ilex (Linnaeus), formerly maintained by forest exploitation (pasture, coppicing, litter-collection, etc.), is now densifying. Forest fragments are also expanding, colonized by Pinus halepensis (Miller). The remnant cultivated areas consist in vineyards, olive and fruit trees and occasionally cereals. But vineyards, which contribute to the regional identity now that the “Côtes du Luberon” wines have been listed as an AOC (registered designation of origin), are in intensive competition with other crops. Since the 1970s, fallow lands (consisting here in dry calcareous grasslands) resulting from the abandonment of cultivated land, have been in decline in the face of pressure from construction. This urbanization process has induced major changes in the landscape organization patterns and the artificialization of the local flora, both associated with the development of private gardens in the landscape mosaic. Human pressures on this landscape mosaic have consequently evolved in both form and intensity during the last half century.

Today, the landscape mosaic of Lauris is composed of four main land-uses: 10% fallow lands, 12% built areas, 25% agricultural areas and 53% forest [11]. This landscape mosaic offers a variety of landscape contexts from the village centre to the forested hillsides surrounding the village.

Four land-uses have been investigated in this study: fallow land, private garden, vineyard and forest. These habitats, described above, also constitute different levels of anthropogenic pressure on habitats. Forests would appear to be the least disturbed habitat, whereas vineyards are a highly altered one. Like vineyards, gardens have resulted from high anthropogenic pressure. Finally, although they are agricultural areas that are left untilled, fallow lands integrate the effects of past disturbance and thus constitute an intermediate disturbance level.

Butterfly surveys were conducted at eight different sites, all located outside the village centre in patchy urban extensions (Appendix 1). These eight sites were chosen for gathering the four studied land-use types in a reduced area. In each of these eight sites, four sampling locations of approximately 500 m² were sampled, corresponding to the four land-use types: fallow land, private garden, vineyard and forest fragment. These 32 sampling locations were visited (replicated) three times between May and July 2005, resulting in 96 sampling units. The size of the sampled locations (500 m²) has been chosen according to the size of the smallest studied private garden.

In order to standardise the climatic conditions, plots were sampled between 9 AM and 5 PM on days when the temperature was above 20 °C, cloud cover was less than 50% and wind speed was less than 20 km/h. During these surveys, the recorder covered the sampling location in 30 minutes catching every encountered individual with a butterfly net. Thirty minutes of recording seems to be a good compromise between realistic species richness assessment and reasonable cost in terms of time, since Steffan-Dewenter and Tscharntke [24] have shown that butterfly species accumulation curves increase rapidly to reach a plateau within the first 10 minutes, whatever the surface area of the site studied.

To characterize the landscape context around each of the eight sites, six landscape descriptors have been calculated in a 400 meters buffer [25] based on a land-use map (Appendix 1): percentage of ground occupied by dwellings, roads, forest, orchards, agricultural land and fallow lands. A Principal Component Analysis (PCA) was performed to assess differences in these landscape descriptors between the eight sites. This has allowed us to create two groups of landscape contexts. The discrimination of these two groups has been tested using a paired Hotelling's T² test based on percentage of ground occupied by dwellings and forest.

Kruskall-Wallis tests were used to test for differences between land-use types in butterfly species richness, abundance, diversity (Shannon index) and evenness (Pielou index), based on the 96 sampling units. Where significant differences occurred, a Mann-Whitney post-hoc test was used to specify the nature of the differences.

In the same way, we used a Mann-Whitney test to test for differences in butterfly species richness, abundance, diversity and evenness between the two landscape contexts highlighted with PCA and Hotelling's T² test.

In order to assess specific similarity between habitat types, the Jaccard index was calculated for each possible pair of samples based on the 32 sampling locations. In order to compare the importance of “landscape effect” and “land-use effect” on sample similarity, “within-land-use similarity” and “within-site similarity” have been compared. Mann-Whitney tests were used to assess whether there were significant differences between mean within-land-use similarity and mean within-site similarity for each land-use type.

To assess the influence of land-use type and landscape descriptors on functional composition of the community, a RLQ analysis has been performed (library ade4 in R, [26]). RLQ analysis is an extension of co-inertia analysis that performs a double inertia analysis of two arrays (R and Q) with a link expressed by a contingency table (L). Our analysis is based on the specific composition of the 32 sampling locations (Table L on which a correspondence analysis has been performed). Each sampling location has been characterized according to its land-use type and landscape descriptors (Table R on which a Hill-smith analysis has been performed). Each species has been described according to the following functional attributes (Table Q on which a multiple correspondence analysis has been performed).

Voltinism:

• • univoltine species (one generation per year);
• • bivoltine species (two generations per year) and;
• • multivoltine species (three or more generations per year).

Larval diet:

• • monophagous species (reported to feed on one host-plant species);
• • “strongly oligophagous species” (reported to feed on host-plants of one genus);
• • oligophagous species (species reported to feed on plant species belonging to one taxonomic family) and;
• • polyphagous species (species reported to feed on a variety of host-plants belonging to two or more taxonomic families) [17,24].

Over-wintering stage:

• • species that over-winter at egg or larval stage (wintering_st1) and;
• • species that over-winter at pupal or adult stage (wintering_st2).

Mediterranean affiliation has been also reported on the RLQ plot [27].

All statistical tests and analysis have been performed using R software [28] and PAST software [29].

On the basis of the PCA plot, landscape descriptors discriminate two landscapes contexts. These two groups of sites are well individualised on the first principal component which represents approximately 72% of inertia (Fig. 1). Sites E, F, G and H are regrouped on the left of the plot, associated with high forest density and high orchards density. Inversely, sites A, B, C and D are associated with high density of dwellings and fallow lands. Density of roads and agricultural lands do not really participate to discrimination. Values of environmental variables for each sampled site are shown in Appendix 2. The Hotelling's T² test performed on percentage of forest and percentage of dwellings confirms the discrimination of the two groups of landscape context (T² = 212; p = 0.014).

One thousand three hundred and seventy eight individuals of 54 species were recorded over the three sampling periods (Appendix 3).

3.2.1 Landscape effect

3.2.2 Land-use effect

3.2.3 Specific similarity

The first two axes of RLQ account for 94% of the total inertia (78 and 16%, respectively). The Monte-Carlo test indicates that the co-structure between R and Q is significant (p < 0.0001, based on 999 permutations). The first two axes of RLQ analysis accounted for most of the variance of the corresponding axes in the separate analyses of environmental descriptors (91% for Hill-Smith analysis of Table R) and species traits (85% for multiple correspondence analysis [MCA] of Table Q), which testifies to the strength of the link between Tables R and Q (Fig. 2).

Correlations between environmental descriptors and the axes of RLQ are related in Table 3. The first axis of RLQ corresponds to an urbanization gradient from the most forested sites (right of the plot) to the most urbanized sites (left of the plot), while the second axis discriminates sampling location according to land-use type (Table 3 and Fig. 2). The high percentage of inertia related to the first axis of the RLQ analysis underlines that landscape context is more discriminating than land-use type.

Correlations between species traits and the axes of RLQ are related in Table 4. The species traits that correlate best with the first axis are voltinism and larval diet. On the second axis, the best correlated traits are over-wintering stage and voltinism. These functional attributes discriminate generalist species (polyphagous, multivoltine species that over-winter at adult or pupal stage) found in the left top of the plot from more specialist species (oligophagous, monovoltine species that over-winter as egg or larvae) found in the right bottom of the plot (Figs. 2 and 3). Monophagous species have been pooled with strongly oligophagous species because of their weak number of individuals.

The separate MCA of the Table Q highlights three groups of species according to their trait assemblage (Fig. 3, plot on the left): specialist species (in black and bold), generalist species (in black and italic) and intermediate species (in grey). This discrimination has been reported on the RLQ plot and is consistent with the correlations between species traits and the two first axes of the RLQ analysis (Table 4). All the species classified as specialist in the MCA and 13 of the 17 Mediterranean species appear in the right bottom of the RLQ plot, and all the species classified as generalist species appear in the left top of the plot.

Our results highlight that abundance of butterflies, species richness, diversity and evenness differ significantly between the four land-use types, even if they do not allow us to highlight significant differences due to landscape context.

Fallow lands constitute the richest habitat type in both species and individuals, directly followed by gardens. Vineyards and forests host poorer communities, which differ from one another in individual abundance (the number of recorded individuals in forests is twice as high as in vineyards).

These differences could be linked to assumed levels of anthropogenic pressure on habitats. Actually, Connell suggested the strong diversity of intermediately disturbed habitats is due to intermediate frequency of disturbance that reduces the pressure of dominant species on others and allows less competitive species to establish themselves [30]. In a same way, Niell et al. observed peaks of richness of butterfly species at intermediate levels of residential development in an area historically dominated by oak woodland. Blair and Launer also demonstrate that this idea of maximal diversity at intermediate levels of disturbance can be extended to include anthropogenic pressure on habitats [15]. Intermediate levels of human development could constitute less constrained habitats that can harbour richer communities than natural areas (strong biotic limitations) or highly human development impacted areas (strong physical limitations) [31]. Thus, the high species richness of fallow lands, compared to forests and vineyards, could be explained by the fact that fallow lands represent an intermediate habitat between natural and highly impacted habitats. Fallow lands could thus attract individuals from these two communities (natural and highly developed areas) that would find favourable conditions to achieve their development. This idea supports the central position of fallow lands on the RLQ plot (Fig. 2). They appear as an important diversity reservoir in the landscape mosaic.

Vineyards appear as the most diverse habitat, although the abundance of butterflies is the lowest. They constitute a heterogeneous habitat gathering an important number of butterfly species despite a low abundance of individuals. These results lead us to conclude that vineyards are likely to play the role of crossing areas, this assumption being confirmed by the high evenness showing an absence of dominating species in vineyard samples.

Forests appear as the least diverse habitat with only 20 species. Although they accumulate fewer butterflies than fallow lands and gardens, the number of individuals is twice as high as in vineyards. Moreover, some species such as Satyrium esculi (Hübner) dominate the community, indicating a more specialized community than in other habitats. This is supported by the high discrimination of forests in RLQ analysis, and by the homogeneity of forest samples which shows a certain stability of this habitat compared to other land-uses.

Thus, the pattern of increased richness in fallow lands compared to vineyards and forests supports the idea of increased species richness at intermediate levels of anthropogenic pressure. Recent studies on butterfly communities also show this pattern of increased species richness at intermediate levels of an anthropogenic disturbance gradient [17,32]. But the high species richness of gardens is confusing: this man-made habitat supports a species richness equivalent to that of fallow lands. Actually, gardens gather as many species as fallow lands with 21% fewer individuals. Their samples also appear as more heterogeneous than those of fallow land which contributes to linking gardens to vineyards in terms of functioning. Gardens could thus constitute an intermediate habitat between fallow lands (“diversity reservoir”) and vineyards (“crossing area”) which is congruent with their position on the RLQ plot (Fig. 2). This idea of crossing area is consistent with Young's observations in a residential garden where the duration of butterfly visits was characteristically short, with a mean visit time of nine seconds [33]. Finally, it is worth noticing that exchanges between gardens and fallow lands seem to be particularly important highlighted by the high specific similarity between these two land-uses. Actually for Vickery, gardens of rural and suburban areas almost always have nearby habitats where butterflies can spend a large part of their life cycle only visiting gardens for nectar plants they contain [34].

Butterfly assemblages of our study site show a consistent pattern in the distribution of functional attributes of species on the main gradients of landscape descriptors. Composition of the landscape is directly related to species traits that discriminate generalist species (polyphagous, multivoltine species that over-winter at adult or pupal stage) associated with urbanized sites from more specialist species (oligophagous, monovoltine species that over-winter as egg or larvae) associated with forested sites. Moreover, this landscape context interacts with land-use types to influence functional composition of the community. Based on these species traits, two communities can be discriminated.

In vineyards and gardens of the most urbanized sites, polyphagous, multivoltine and bivoltine species are over-represented. This particular trait assemblage corresponds to generalist characteristics attributed to disturbed habitats in many studies [13,35–37]: good dispersing abilities and weak host-plant specialization. This community also exhibits an over-representation of species over-wintering as adult or pupa. This strategy can represent an advantageous foraging behaviour at the favourable season. The latter species are able to move to find resources, whereas species over-wintering as egg or larvae depend on the presence of their host-plants where they over-winter. Moreover, gardens bring into the landscape new exploitable resources for butterflies, in particular ornamental and/or exotic plants, water sources which extend nectaring plant availability, particular configurations of vegetation such as hedgerows or flower beds [13,34]. These management practices, and especially watering practices, may explain multivoltine species’ over-representation in gardens in our study.

Inversely, in fallow lands and especially forests of the most forested sites, Mediterranean species and species with specialist traits, such as Pyronia cecilia (Vallantin) or Maniola jurtina (Linné), are over-represented.

This variation of species traits according to land-use types and landscape contexts is also congruent with Connell's reasoning who stated that too infrequent disturbance regime would favour good competitors, resulting in the dominance of few specialist species in the community, whereas too frequent or too severe disturbance would favour good dispersing and/or quickly maturing species [30].

The representation of the generalist characteristics increases with landscape context from forested sites to urbanized sites, and with land-use type from forests to gardens and vineyards with fallow lands in intermediate position. So the number of generalist butterflies in our results would appear to reflect the level of disturbance at both landscape scale and habitat scale [24,38–40]. This result is consistent with observations by Niell et al. along an anthropogenic disturbance gradient consisting in forest fragment decline with rural residential development [17]. They also suggest that an increase in the number of generalist butterflies is a more sensitive indicator of changes implied by anthropogenic disturbance than a decline in the number of specialists. Vulnerability to landscape degradation (fragmentation, disturbance and habitat loss) is known to be greater for specialist species than for generalist species [38–40] which may benefit from reduced competition with specialists [41]. That's why the very weak representation of the specialist species in our urbanized sites underlines a risk of functional homogenization associated with urbanization processes [40,42]. In a same way, Kuussaari et al. show that increasing species in Finland are more mobile, use a wider range of host-plants and live in more eutrophic habitats than declining species [39] which can be worrying as this overall trend in butterfly communities seems to reflect a more general phenomenon in Europe.