The study was carried out in a rural area of southeastern France, in the Mediterranean Basin region (43°44′N, 5°18′E) (Fig. 1). The site of Lauris is located in the Natural Regional Park of Luberon, bordered to the north by the watershed of the Petit Luberon and to the south by the Durance River. The entire study area was 2181 ha in size and essentially made up of 53% of woodland, 25% of agricultural areas, 10% of fallow lands and 12% of urban areas; it is under the influence of the meso-Mediterranean bioclimate. It is in the zone of influence of two big cities, Aix-en-Provence and, to a lesser extent, Marseille, where urbanization has been spreading to the surrounding agricultural and natural countryside since 1975. In 30 years, the population of Lauris has doubled, rising from 1620 inhabitants in 1975 to 3143 inhabitants in 2005.

Three different housing densities were defined by locally estimating the built-up site density (proportion of built surface per unit area) with Arc View GIS (r. 3.2 software) (Fig. 2). The three types corresponded to the different phases of the village urbanization and were related to housing and garden types [38]:

• – type 1 corresponds to a high housing density (built-up area >20%) and particularly to the old centre of Lauris. It is composed of small houses (two or more adjoining dwellings) built during the 12th and 13th centuries. They have front gardens of 20 m² on average;
• – type 2 corresponds to a medium housing density (built-up area between 10 and 20%). It concerns two residential areas close to the village centre, built between 1965 and 1975. Each area is composed of detached or semi-detached estates, incorporating private gardens of 600 m² on average;
• – type 3a corresponds to a low housing density, characterised by villas scattered in pine forest (built-up area <10%). These were built from 1975 to 1995 when, in an attempt to preserve agricultural activities, natural and semi-natural forests composed of Pinus halepensis (Mill.) were targeted by urbanization [41]. The garden size of these dwellings ranged from 2000 to 10,000 m²;
• – type 3b corresponds to a low housing density (built-up area <10%), characterised by modern villas encroaching on fallow land following the cessation of agricultural activity. The garden size of these dwellings ranged from 2000 to 10,000 m².

The size of each housing density area is 9 ha for type 1, 34 ha for type 2 and 216 ha for type 3.

In order to provide a homogeneous distribution of gardens, 30 houses from five main streets within each housing density type were chosen for survey. Each street was then exhaustively visited so that the entire length and both sides of each street were examined and each house visited. After requesting permission to undertake the survey on the resident's property, we sampled immediately the garden. In order to reduce refusal or absence of homeowners, a publicity campaign was carried out using local papers and the local council, prior to visits. Given that socio-economics status is an important driver of floral patterns in urban areas [39,42], each area was systematically visited between April and July 2005 during the week and the weekend, in order to sample the gardens of people working and of people who have a second home. Moreover, less than 10% of homeowners refused the floral sampling of their garden, so that the study did not concern an only social class. The combined area of all the sampled gardens was 215,000 m².

In each garden, native and alien cultivated plants (excluding the constituent species – grasses – in lawns) were recorded during an exhaustive survey of the garden. The inventory was drawn up distinguishing 13 landcovers of the garden: lawn, gravelled path, flower bed, pot and tub, hedge, wall, borders of swimming pool, playing field, pine forest, oak grove, orchard, vegetable garden, and olive grove. Taxonomic identification was carried out on the basis of the Universal Encyclopaedia of 15,000 garden plants and flowers [43]. Plants that could not be identified were labelled as ‘unknown’ or ‘Genus sp.’ when the genus was identifiable or ‘Genus group’ when the group of cultivar was identifiable. The intergeneric hybrids were recorded as ‘x Genus species’, interspecific hybrids ‘Genus x species’. Some horticultural plants involve species that are difficult to distinguish from varieties, such as taxa of Arum, Aubrieta, Chrysanthemum, Dahlia, Dianthus, Iris, Gazania, Gladiolus, Heuchera, Lilium, Mandevilla, Narcissus, Ostheospermum, Paeonia, Petunia, Pyracantha, Tulipa, Rhododendron and Rosa; these were identified only to the genus level.

We specified the most frequently recorded families, genus and species by giving frequency values (120gardens=100% frequency). The abundance of each species was also given, except those of Iris sp. and Hypericum calcynum, which have high rates of vegetative reproduction. A frequency distribution of species abundance was presented. The proportion of growth forms (annual, biennial, perennial) and Raunkiaer's form was given [44]. Each species was also assigned to one of three status defined by Aboucaya (1998) [45]: “known to be invasive”, “potential invasive”, “on waiting list” in France, in order to assess the number of invasive species that are introduced to gardens. To detect any significant difference in the proportion of invasive species among housing density types, χ-square tests were performed.

A species-accumulation curve was plotted to assess completeness of sampling effort at the combined area and at each housing density type. The cumulative number of species is plotted against some measure of the effort expended to find them. Species accumulation curves based on ‘proxy’ units such as area of quadrats, trap-hours or hours of observation represent a uniform process; this is the reason why the sampling effort must be standardized [46]. The order in which samples are added to the species accumulation curve affects the shape of the curve produced; the use of randomization procedures is useful to overcome this problem [47]. Contrary to Thompson et al. [36] and Loram et al. [48], who used a standardized measure of sampling effort in each garden (quadrats), we used in our study a non-uniform sampling process. As we wanted to sample exhaustively native and alien cultivated species of all garden landcovers on the total area of each garden, our sampling unit was the total garden area, which ranged from 2 to 11,000 m². The use of a randomization procedure in this case leads to an overestimation of species richness due to the heterogeneity of sampling units. Therefore, we could not use the traditional method of plotting a species accumulation curve. In order to plot a species accumulation curve that corresponds to a plot of the cumulative number of species discovered, within a defined area, as a function of some measure of the increasing effort expended to find them, we had to classify in our case the studied gardens in increasing order of size, interpreting an increase in sampling effort, and for each cumulative garden area, we added the number of new species. The 30 gardens of each housing density area were used to make up each cumulative area. This methodology, used to plot the species accumulation curve, does not permit to determine the minimum sampling effort required for estimating the species richness in urban areas, but it is useful to assess the completeness of the sampling effort in our particular case.

A principal correspondence analysis (PCA) was also performed in order to identify the most widely represented species in gardens of the different housing density types. This analysis was conducted with a species frequency matrix (62 rows corresponding to taxa and four columns corresponding to urban areas) and only with taxa whose frequency was greater than 20% (see Appendix A). We clarified then in which landcovers the most frequent species were present. A simple correspondence analysis (CA) was also performed with the same species frequency matrix in order to identify whether or not specific groups fit with housing density types. A Monte Carlo permutation test based on 1000 random permutations was used to test the null hypothesis, i.e. species were unrelated to the housing density type.

In order to assess the ratio of regional (Mediterranean Basin) and exotic species, information on the origin of the species was taken from Brickell and Mioulane (2004) and classified according to 12 origins [43]: Europe, America, Asia, Africa, Oceania, Mediterranean Basin, Tropical, Eurasia, Eurafrica, Northern Hemisphere, mixed (species coming from more than three different continents) and horticultural (species resulting from artificial hybrid selection). To detect any significant differences in the proportion of species origins among housing density, analysis of variance (ANOVA) tests were performed.

All statistical analyses were run with Minitab (r.14 software).

Nine hundred and seventy-three horticultural plants were collected in the combined area of all gardens (215,000 m²). Out of the 114 plant families recorded, those with the highest number of taxa were Asteraceae (7.2%), Rosaceae (6.6%), Liliaceae (4.8%), Crassulaceae (4.6%), Lamiaceae (4.2%) and Caprifoliaceae (3.4%). All other plant families had a frequency of less than 3.4%. Taxa were also divided into 376 genera, 80% of which were represented by only one species. Ninety-one percent (519 species) were much less frequent than 20% (Appendix A); 82% of species were recorded fewer than 50 times (Fig. 3). The most abundant species were x Cupressocyparis leylandii (1548), Cupressus arizonica (1243), Pyracantha sp. (1104), Prunus laurocerasus (1011), which were all planted in hedges, and Rosa sp. (1034).

With regards to the growth pattern of species, 92% were perennials, 7% annuals and 1% biannuals. An overwhelming number of taxa were trees and shrubs, divided as follows: phanerophytes 55.1%, chamaephytes 21.4%, geophytes 12.4%, hemicryptophytes 6.3% and therophytes 4.8%.

The species accumulation curve for all the studied gardens (120 gardens) had a logistical shape that tends toward an asymptote (Fig. 4a) and suggests that our sampling effort provided a good representation of horticultural flora. Curves vary with housing density types with that for species of high housing density, rapidly reaching an asymptote after 600 m² (Fig. 4b), with 217 plants collected. This accumulation curve was also steeper near the origin, reflecting the high local richness of cultivated species communities. A larger sample area would thus provide very little additional information. Rather, greater numbers of plants were recorded in gardens of medium- and low-density housing types because of the larger cumulative garden areas. A total of 272 horticultural plants were collected on 19,000 m² in the gardens of medium housing density (Fig. 4c), while in gardens of low housing density, 412 and 310 species were respectively collected on 79,000 m² in an agricultural environment and 116,000 m² in a forest environment (Fig. 4d).

The PCA performed on all species gave no relevant result, because in total area, there were numerous species with a frequency lower than 20%. Fig. 5 shows results of PCA for types of housing densities with taxa whose frequency was greater than 20%. The results were presented only for the first two axes, which represent 86.1% of the variance (Fig. 5a and b). The first axis of PCA accounted for 63.10% of the variance, while the second axis accounted for an additional 23%. The first axis clearly sorted taxa by frequency, whilst the second separated gardens by housing density type. The most frequent species in gardens of type 1 were Rosa sp., Pelargonium ‘zonal group’, Petunia sp. The gardens of types 2 and 3a showed similarities in the most frequent species and were characterised by Syringa vulgaris, Rosmarinus officinalis, Olea europea, Cupressus sempervirens, Iris sp. and Forsythia x intermedia. In the gardens of type 3b, Pyracantha sp., Viburnum tinus, x Cupressocyparis leylandii, Althaea sp., Prunus cerastifera, Laurus nobilis, and Ficus carica were the most frequent species. Two species, Nerium oleander and Lavandula angustifolia, were common in all gardens, and consequently did not belong to a specific housing density type.

Considering all gardens, Rosa sp. was the most frequent species in ‘flower bed’ and ‘wall’ landcovers, with respectively 66% and 38% frequency, while the Pelargonium ‘zonal group’ was the most frequent species in ‘pot and tub’ landcover (31%). The most commonly and frequently found in ‘flower bed’ landcover were Iris sp. (53%), Lavandula angustifolia (49%) and Nerium oleander (43%), in ‘lawn’, Olea europea (64%), Syringa vulgaris (45%) and Prunus cerasifera (41%), and in ‘hedge’ Pyracantha sp. (48%) and x Cupressocyparis leylandii (41%).

Species were related to the housing density type (Monte Carlo test, χ2=338, ddl=183, p<0.001). Results of CA on housing density types are presented only for the first two axes, which represent 86.75% of the total inertia, respectively 61.20% and 25.55% (Fig. 6). Begonia ‘semperflorens group’, Dianthus sp., Lonicera japonica, Delosperma cooperi, and Pelargonium ‘zonal group’, which contributed to the first axis, were related to gardens of type 1. Prunus dulcis, Vitis sp. and Prunus armeniaca were the main contributors to the second axis and characterised gardens of type 3, while Morus kagayamae, which also contributed to the second axis, was the typical garden taxa of type 2.

Twelve percent of the horticultural taxa originated from the Mediterranean Basin. The alien cultivated taxa (88%) came mainly from Asia (23%), America (19%), Europe (11%), and Africa (7%). Among American and African taxa, 12% originated from the southwest of North America and South Africa.

Fig. 7 shows the spectrum of origin of the species that occur in gardens in each housing density type. Gardens of type 1 showed a significantly different proportion (ANOVA, F3,116=12.95; p<0.001) of African taxa such as Pelargonium ‘zonal group’, Pelargonium ‘ivy group’ and Delosperma cooperi as mixed taxa such as Dianthus sp. and Hydrangea macrophylla. The highest percentage of taxa from the Mediterranean Basin were recorded for gardens of types 2 and 3a, with respectively on average 25% and 21% (ANOVA, F3,116=14.31, p<0.001). The most strongly represented Mediterranean taxa were Cupressus sempervirens, Lavandula angustifolia, Nerium oleander, Olea europea, Rosmarinus officinalis, Viburnum tinus.

Twenty-one out of the 573 taxa were registered on the three lists of invasive species established by Aboucaya [45] (Table 1), seven being notoriously invasive in the Mediterranean area. Their frequencies in gardens were highly variable, ranging from 1 to 48% of 120 gardens. The proportion of invasive taxa were not significantly different between housing density types (χ2 test, χ2=0.21, ddl=3, p=0.975). The abundance of notorious invaders in gardens was low. The most abundant species were hedges species such as Pyracantha sp., Pittosporum tobira, and Elaeagnus x ebbingei.

Our study confirms that private gardens contain a large cultivated biodiversity, strongly dominated by perennial and woody alien species. With 573 taxa recorded from a combined area of 215,000 m², private gardens present a high level of horticultural floral richness in urbanizing rural areas. In comparison, the level of introduced plant richness in gardens of rural areas is nevertheless lower than in city gardens. A study of 400 domestic gardens in Mexico City revealed that 525 out of 750 species recorded were aliens in a total garden area of 55,000 m² [49]. Similarly, Smith et al. [3] inventoried 798 alien species on 12,700 m² of gardens in Sheffield. This difference of level of introduced plant richness between rural and urban areas can be explained by the differences in garden size, rural gardens being much bigger than those of cities and, therefore, the species density of cultivated plants lower. Moreover, the concentration of houses in cities also induces a stronger need to be in contact with nature, unlike in rural environment close to natural areas. In the artificial environment of urban areas, introductions of horticultural plants are more important for the physical and mental well-being of the human population that lives in urban areas [4,5]. If we consider the rapid development in housing density in the years to come, especially from low to medium housing density types, the flora of rural areas will undergo the same evolution as that seen in cities, particularly the increase in alien species.

Our study also shows that the accumulation of new cultivated species is variable under different urbanization pressures, especially in high housing density type, where it is greater at the local scale. These differences may be related to the garden's structure in each housing density types [36], and more particularly to the diversity of garden landcovers.

Although the horticultural flora of rural gardens is rich, it is also very heterogeneous, like in urban gardens. In the 120 gardens of Lauris, most of the taxa have low frequency values, apart from a few very popular species. The study of the species abundance showed also that the number of introduced individuals was low, except for species commonly planted in hedges. The majority of the species occurred as solitary specimens; therefore, the opportunities for the large majority of horticultural plants to colonise outside gardens may also be relatively few. These differences in frequency values have already been evaluated by studies dealing with ornamental woody species in urban gardens. Acar et al. [29] mentioned that in different residential gardens of Trabzon city in Turkey, out of 274 woody species, 231 were much less frequent than 15%. Similarly, in studying patterns of trees in a residential neighbourhood of a low-density housing estate in Hong Kong, Jim [28] highlights two groups of tree species, one composed of a few common species and a second of many uncommon species. In a recent study of Sheffield garden flora, Smith et al. [3] show that a large proportion of alien species were recorded only once, as well. The similarities between the spatial floral patterns of ornamental trees in Turkey and Hong Kong, of the Sheffield garden flora and of the ornamental flora in Lauris may reflect the operation of two opposite social attitudes, namely conformity versus individualism [28]. On the one hand, conformity, generally defined as the tendency to act or think like other members of a group, could be expressed by spatial similarities through the most popular species. On the other hand, an individualism phenomenon could be expressed by the heterogeneity of horticultural flora.

Our study shows that similarities in species composition can be observed between gardens of a same housing density type. Zmyslony and Gagnon [31] found the same result with the floral composition of front yards in a residential street of Hochelaga–Maisonneuve District, Montreal. They suggested that it was related to a conformity phenomenon through mimicry activities at a local scale. Residents in the street section are influenced by the shape, colour and location of the vegetation they see in gardens of their nearest neighbours. Our study also shows that the composition of popular species observed within a housing density area was not the same as the species composition of another housing density area. This result could suggest that the conformity phenomenon is not expressed in the same way through different housing density types.

The results concerning the floral composition of Mediterranean private gardens show us that the influence of regional factors on plant introductions must be taken into account. Firstly, taxa such as Prunus cerasifera, Prunus armeniaca and Vitis sp. are remainders in the gardens of past agricultural activities (cherry orchards, olive groves, vineyards). The history of land use appears, consequently, determining the distribution and composition of horticultural flora, as in the Mediterranean flora [50]. Secondly, private gardens were characterised by a strong proportion of species specifically adapted to the Mediterranean climate. A quarter of the horticultural taxa cultivated in Mediterranean gardens (24%) were either from the Mediterranean Basin or from climatic regions under Mediterranean influence. This trend has also been observed in other Mediterranean gardens, like in Trabzon city in Turkey, where 25% of taxa originated from Turkey and Mediterranean regions. In northern temperate gardens throughout the world, Mediterranean taxa are sometimes present, but by no means common. These results show that the choice of homeowners could be influenced by regional climatic conditions, but also by the regional market availability of Mediterranean taxa. In order to assess the relative importance of social and economic factors on gardening practices, a socio-economic study should be conducted.

Rural gardens also constitute sources of many notorious invasive species. More than half of the notorious invasive species in the Mediterranean region are cultivated in rural gardens. Also, the probability of colonisation from a garden to a neighbouring habitat might be stronger in urbanizing rural areas than in cities if we take the highest proportion of interfaces between gardens and uncultivated habitats in urbanizing rural areas into account. These houses create more urban/wildland interfaces. The French Mediterranean countryside could be threatened by biological invasions. It would be interesting to monitor areas adjoining gardens in order to identify which species had escaped from gardens and in which part of the landscape mosaic they grow. Field observations suggest that recent fallow, very frequent in rural areas, may be favourable sites for invasive species to seep into the landscape.

The seasonally xeric nature of the Mediterranean climate is often a barrier to the establishment of introduced species. Our study showed that 12% of horticultural taxa cultivated in Mediterranean gardens come from climatic regions under the Mediterranean influence like the southwest of North America and of South Africa. The proportion of water-stress-tolerant species (South-African taxa, often planted in pots and tubs) is higher in gardens of high housing density, which are exposed to a more xeric environment than the surrounding area [51]. Thus garden owners, planting species adapted to the Mediterranean climate, reinforce the risk of alien species invading the Mediterranean countryside [52].

Moreover, 12% of cultivated taxa are native. More than half of them are among the most common horticultural taxa: Cupressus sempervirens, Ficus carica, Lavandula angustifolia, Laurus nobilis, Nerium oleander, Olea europea, Rosmarinus officinalis and Viburnum tinus. They can potentially contribute to overall genetic diversity and maintain gene flow between remnant populations, thus buffering small and otherwise isolated populations from extinction. However they can also threaten the genetic integrity of natural populations: gene flow from such plantings can limit local adaptation and lead to population decline and local extinction if foreign genes disrupt locally adapted gene complexes [20].