Study region
Located in south-western France, the Landes de Gascogne Forest covers approximately 1.16 million hectares and is predominantly composed of pure stands of maritime pine (P. pinaster) (Barbaro et al., 2005). These plantations are intensively managed, with soil preparation and thinning operations before clear-cut harvesting at 30 to 40 years, creating temporary open areas across the landscape. Within this homogeneous landscape matrix, a limited number of unmanaged riparian forests (dominated by Alnus glutinosa L. and Quercus robur L.) as well as low-intensity managed stands of native oaks (primarily Q. robur and Q. pyrenaica Willd.) persist (Mora et al., 2012).
Experimental design
For the purpose of the study, the Living Lab “Forest Bocage” (https://www.plantedforests.org/fr/infrastructures/superb-bocage-forestier/) was created in the Gironde district, in the Landes de Gascogne forest (50,000 ha area, barycenter coordinates: X: −0.776865; Y: 44.560623). In the Living Lab area, we used high-resolution infrared colour orthophotographs (IGN; 20 cm pixel resolution, https://geoservices.ign.fr/documentation/donnees/ortho/bdortho ) analysed at a 1:2500 scale in a GIS environment to identify (1) broadleaved hedgerows and (2) broadleaved and mixed forest stands. Then, we selected 36 study sites according to factorial design with two factors following the recommendation from Fahrig, 2013. Factors included (1) the type of habitat with three modalities (pine plantation edge, connected hedgerow or isolated hedgerow), (2) landscape composition with two modalities (low or high amount of broadleaved stands in the landscape). This resulted in 6 combinations with 6 replicates (Fig. 1).
Sampled hedgerows and pine edges were 100 to 325 m in length. Pine edges were at least 6 m in height and adjacent to a forest road, representing the control treatment. Hedgerows were defined based on field observations as linear formations of broadleaved trees at least 8 meters in height, typically one or two trees wide, with continuous canopies, and dominated by native oak species (Q. robur and Q. pyrenaica). Only hedgerows in between two pine plantations were sampled, with an average distance of 26 m between the two pine stands. Connected hedgerows were directly adjacent at one end to a broadleaved stand (i.e., distance to the nearest natural habitat patch = 0 m). Isolated hedgerows and pine edges with no hedgerow were at least 100 m distant from a broadleaved stand (i.e., distance to the nearest natural habitat patch > 100 m) and 50 m from other hedgerows. These distance thresholds were selected to assess the influence of habitat connectivity on multi-taxonomic biodiversity, particularly for organisms with limited dispersal abilities such as ground beetles and spiders. However, they may represent relatively short distances for taxa with greater dispersal capacity, such as birds (Fahrig, 2013; Slagsvold et al., 2013). Nevertheless, these distances were chosen as a compromise, given the difficulty in identifying truly isolated hedgerows in landscapes with high broadleaved cover due to (1) the high density of hedgerows, and (2) the limited availability of landscapes with high broadleaved cover. The percentage of broadleaved forest cover in the surrounding landscape (hereafter “broadleaved cover") was quantified within a 500 m buffer around the sampled hedgerows or pine stand edges, at each study site. Two classes of broadleaved cover were selected: low (0–6%; n = 18) and high (14–36%; n = 18).
The 36 sampled sites were evenly distributed across the study area, with an average inter-site distance of 1.7 km (range: 0.7–3.9 km) to minimize spatial autocorrelation (see the map of the Living Lab: Fig.S1).
Multi-taxonomic biodiversity survey
We selected six different taxonomic groups, ranging from primary producers to secondary predators, in order to cover a wide range of dispersal abilities and biotic interactions. Additionally, all the groups studied are known to be sensitive either to forest tree composition, landscape composition or to hedgerows in agricultural landscapes (Ampoorter et al., 2020; Montgomery et al., 2020).
Understorey vegetation
All vascular plant species were recorded along a 50 × 1.5 m transect positioned within the hedgerow or the first line of trees for pine edges. All species present in the transect were classified into two vertical strata: the herbaceous layer, comprising herbaceous species and ligneous species under 0.5 m in height, and the shrub layer, including only ligneous species and climbers (vines) between 0.5 and 5 m in height (Canullo et al., 2011). To estimate plant species cover, three 5 × 1.5 m quadrats were placed within this the transect. Two pairs of trained observers (JD/TB and IVH/NP) independently assessed species cover within each stratum using the Domin scale (Rich et al., 2005). Understorey vegetation was assessed in June 2023 to capture nearly the entire local species pool. Species identification was conducted using a taxonomic guide (Rameau et al., 2018), and complex identifications were verified in the laboratory under a binocular magnifier. Following Corcket et al. (2020), Domin scale values were converted to mean percentage cover, and these values were then averaged across the three quadrats to obtain a cover value per species per site. For ligneous species present in both strata, cover values were summed. Species observed within the transect but absent from the quadrats were assigned the lowest mean cover value (0.033%). Seedlings and saplings from Q.robur, Q.pyrenaica and P.pinaster were removed from analysis as study sites were selected to be dominated by trees of these species.
Butterflies
Butterflies were surveyed using the line-transect method (Pollard & Yates, 1993). At each site, one trained observer (IVH) walked along a 100 m transect (either along the hedgerow or the pine edge), recording all butterflies observed within 2.5 m on either side of the transect line and up to 5 m ahead. Individuals were identified visually or, when necessary, captured with a net and released after identification. The transect was walked in both directions, but only the maximum number of individuals per species observed over one 100 m transect was retained for analysis to avoid double counting. Each of the 36 study sites was visited three times, once a month, between May and July 2023, which allowed observing different individuals and species. Surveys were carried out between 10:00 and 18:00, and only under suitable weather conditions (temperature > 20°C, low cloud cover, and wind speed < 30 km/h). The order of site visits was randomized across the three survey periods. For each site, the total number of individuals per species was summed across the three visits and used in statistical analyses.
Carabid beetles and ground dwelling spiders
Carabid beetles (here after “carabids”) and ground-dwelling spiders (hereafter “spiders”) were sampled using pitfall traps (Brown & Matthews, 2016). Each trap consisted of a glass jar (90 mm in diameter, 100 mm in height, 445 mL volume) filled with a mixture of propylene glycol and water, and covered with a plastic rain guard. At each site, three traps were placed along the hedgerow: one at the center and two others located 25 m away on either side. Traps were active during three sampling periods: April, June, and July 2023, each lasting two weeks, for a total of 42 trapping days. Collected arthropods were stored in a deep freezer prior to identification.
Carabids were identified to the species level by an expert (SJ), using external morphological discriminating characters as well as genitalia and aedeagus examinations if necessary (Janovska et al., 2013). Reference books were those for the French carabids fauna (Coulon et al., 2011; Jeannel, 1941), and the European fauna (Hurka, 1996; Trautner & Geigenmüller, 1987). Spiders were identified to the species level by two experts (OB and SJ), using one taxonomic book (Roberts, 1985) and two taxonomic websites for spiders for France and Belgium (https://arachno.piwigo.com/) and for Europe (https://araneae.nmbe.ch/). Juvenile spiders were excluded from the analysis due to identification difficulties. Captures from the three seasons were pooled to obtain a species richness value per site. Although the number of individuals captured per species represents an activity density metric, it was used as an acceptable proxy of abundance (Chaladze, 2020) for codominance analysis.
Birds
Bird communities were surveyed using passive acoustic monitoring, combined with expert identification of species-specific calls and songs (Schillé et al., 2024). In April 2023, one Song Meter Mini Bat recorder (Wildlife Acoustics), equipped with an omnidirectional microphone, was installed in each of the 36 study sites. Devices were installed at a height of 2 meters on pine trees located along the pine edge, either adjacent or not to hedgerows. Recorders operated for five consecutive days, recording one-minute audio file every three minutes from sunrise to sunset. Within a period of two days with no rain and low wind speed (< 30 km/h), a subset of five one-minute recordings was extracted between 08:30 and 08:45 am. This time period was selected as it corresponds to 1.5 hour after sunrise, matching the peak bird vocal activity. This resulted in a total of 360 audio files of one-minute (5 recordings × 2 days × 36 sites), corresponding to six hours of acoustic data. Expert ornithologists (YC, MS, and IGC) listened at each recording to identify all bird species present based on their vocal activity. For each site, the number of files in which a given species was detected (ranging from 0 to 10) was used as a proxy for bird vocal activity (Schillé et al., 2024). Although individual birds may have been recorded multiple times, vocal activity was considered a reliable indicator of relative abundance and used for subsequent codominance analyses.
Reptiles
Reptiles were surveyed using a combination of active visual search and artificial refuge methods to optimize detection efficiency (Michael et al., 2012; Pottier, 2023). In January 2023, four artificial refuges (asphalt roofing sheets) were installed at 20 m intervals along a 100 m transect, placed either along the hedgerow or the pine edge. Two pairs of trained observers (JBR/TB and OB/NP) conducted visual surveys along the transect, recording all reptiles observed within the hedgerow or pine edge, including individuals located on or beneath the artificial refuges. Each of the 36 study sites was surveyed five times, once per month from April to October 2023, excluding July and August due to excessively high temperatures. Surveys were conducted between 09:00 and 13:00, which corresponds to the period during which reptiles remain in the sun for thermoregulation. Rainy days and those with wind speed exceeding 30 km/h were avoided. The order of site visits was randomized across the five survey periods. As individuals were not marked, for each site, the maximum number of individuals per species observed during one visit was used as an abundance metric in codominance analyses.
Dominant and rare species evaluation
Numerous metrics have been developed to identify dominant and rare species, but most were designed for plant communities (Avolio et al., 2019; Dee et al., 2019). To our knowledge, only two studies have investigated species dominance and rarity using a multi-taxonomic framework (Allan et al., 2014; Soliveres et al., 2016). However, both relied on arbitrary thresholds to define dominant (e.g., top 10% most abundant) and rare species (e.g., bottom 50–90% less abundant). In contrast, we applied the codominance approach proposed by Gray et al. (2021), which (1) more accurately captures dominance patterns by identifying, for each taxon and community, the optimal number of codominant species based on their relative abundances, and (2) can be applied across taxa regardless of differences in abundance metrics. The approach selects subsets of the most abundant species and calculates a harmonic mean of their relative abundances, favouring subsets where species have similar abundances. This shared abundance is then contrasted with that of the next most abundant species to compute a codominance index. The optimal codominant subset is the one that maximizes both internal similarity and contrast with subordinate species, ultimately yielding the number of codominant species in the community (see Gray et al., 2021 for details). This method also allowed to detect mono-dominated communities, when the ratio between the first and second species in terms of abundance is higher than 3. The codominance metric was calculated for communities of each taxonomic group at two spatial scales: (i) at the local level, for each of the 36 study sites independently, and (ii) at the regional level, by aggregating species abundances across all 36 sites. This method allowed us to classify species within each taxonomic group into 3 categories: (1) locally dominant species, defined as codominant in at least one site, (2) regionally dominant species, codominant in the entire dataset and (3) rare species, never identified as codominant at the local or regional level. We assessed local dominance only in communities with more than 20 individuals. Consequently, no reptile species were classified as locally abundant. At the local level, in a few cases, no clear pattern of codominance emerged; in these cases, no species were classified as locally dominant. The complete classification of species according to dominance status is provided in Table S2. Notably, between 1 and 5 species per group were classified as regionally dominant, and these species were always locally dominant in at least one site. This approach also allowed the classification of between 5 and 16 species as locally dominant.
Forest specialist species identification
For each taxonomic group, we identified species considered as forest specialists at the French or European scale. As classification methods differ among groups, we applied group-specific criteria designed to consistently exclude generalist species and retain only those with a strong forest affinity (hereafter referred to as “forest species”).
For understorey vegetation, we used the European Forest Plant Species List (Heinken et al., 2022). Using data specific to the Atlantic region of France and following the same methodology as Litza et al. (2022), we selected species classified in the first two categories: (1.1) taxa primarily found in closed forest habitats, and (1.2) taxa typically associated with forest edges and open forest conditions. Forest butterflies were identified using the European classification of butterfly species by biotope (van Swaay et al., 2006). For carabid beetles, we adopted the classification method used by Jouveau et al. (2022) in the same study region, classifying species into three habitat classes: forest specialists, generalists, and open-habitat specialists. For species not covered in Jouveau et al. (2022), we supplemented the habitat information using the publications of Murdoch (1967) and Jaskula & Soszyńska-Maj (2011). Spiders were classified using the World Spider Trait database (Pekár et al., 2021), focusing on the "Light 2" trait, which categorizes species based on their light preference (Buchar et al., 2020). Values for this trait range from 1 (species of open habitats) to 5 (species of dark habitats). Species with a score > 3.0, corresponding to a preference for shaded or dark environments, were classified as forest dwellers. When a species had multiple habitat associations in the database, we used the mean score above 3 as threshold for classifying it as forest-associated. Forest bird species were identified based on the STOC report (Fontaine et al., 2019). This French national monitoring program classifies birds as forest species when they occur significantly more often in 1 km² areas dominated by forest cover compared to areas dominated by agricultural or urban land uses, based on 30 years of national survey data. Reptiles were excluded from the analysis, as only five species were present in the dataset, all being classified as generalist species (INPN; https://inpn.mnhn.fr).
Red List species identification
The conservation status at the European level was checked for every species according to the IUCN Red list (https://www.iucnredlist.org/, 04-2025). 47 understorey plant species were classified as NA as they were either data deficient species (n = 39), invasive species (n = 5) or species identified at genus level only (n = 3). As spiders are not included in the IUCN European Red list, we used the red List of French spiders (https://uicn.fr/wp-content/uploads/2023/03/liste-rouge-araignees-de-france-metropolitaine.pdf). There was no red list available for carabids at the European or French level.
Data analysis
To integrate the responses of all taxonomic groups into a single biodiversity metric, we computed a multidiversity index following the method proposed by Allan et al. (2014). This index is used to weigh taxa so they have the same importance. It was calculated by scaling the species richness of each group by its maximum observed value across the 36 study sites. The standardized richness values were then averaged across all six taxonomic groups to obtain one multidiversity value per site.
To evaluate the effect of broadleaved cover on biodiversity, since the spatial scale at which landscape composition influences the presence of species can vary depending on both species' movement ranges and the ecological context (Fahrig, 2013), we assessed the relevance of the 500 m buffer size through a model-based approach. Specifically, we constructed 10 linear models using multidiversity as a composite biodiversity response variable, with broadleaved cover measured at buffer distances ranging from 100 m to 1000 m in 100 m increments. We then plotted the R² values of these models against buffer size. The highest R² values were observed for buffers between 400 m and 1000 m. Moreover, broadleaved cover estimates were highly correlated across this range, supporting the choice of a 500 m buffer as a representative and ecologically meaningful scale (Fig S2, Table S1). In all analyses and in line with our study design, broadleaved cover was treated as a quantitative variable with two classes (low or high), rather than as a continuous gradient, because its range in our study area was limited (0–36%) and values were not evenly distributed along the gradient.
We used the following statistical models to test our hypotheses:
H1: To assess the effects of habitat type and broadleaved cover on multi-taxonomic biodiversity, we first used the species richness of each taxonomic group separately using Generalized Linear Models (GLMs) with a Poisson distribution, appropriate for count data. Second, multidiversity was used as a response variable in a Linear Model (LM).
H2: To examine the influence of habitat type and broadleaved cover on community composition of each taxonomic group, we performed Permutational Multivariate Analysis of Variance (PerMANOVA) using a Bray–Curtis dissimilarity matrix, appropriate for ecological community data. Species abundance data were Hellinger-transformed to reduce the disproportionate influence of dominant species (Legendre & Gallagher, 2001). PerMANOVAs were performed on the full dataset for each taxonomic group to assess differences between landscapes with low and high broadleaved cover categories, and on sub-datasets including only two of the three habitat types (pine edge, connected hedgerows and isolated hedgerows), to compare community composition between each pair of habitat types. Principal Coordinate Analysis (PCoA) was used to visualize patterns in species composition. For reptiles, the results must be interpreted with caution due to the low number of species and individuals.
H3: To test the effect of habitat type and broadleaved cover on dominant species (local or regional level), rare species, and forest species, we calculated additional multidiversity indexes based on different subsets of the dataset (i.e. multidiversity of local dominance, multidiversity of regional dominance, multidiversity of rare species and multidiversity of forest species). Reptiles were excluded from both the local dominance and forest analyses, as no species in this group met the respective criteria. Multidiversity indexes served as response variables in separate LMs.
For GLMs and LMs, the interaction between habitat and broadleaved cover was tested and retained only when statistically significant. Model assumptions of residual normality and homoscedasticity were assessed graphically. We checked for spatial autocorrelation of the residuals of each model using Moran’s I (4-nearest neighbours) across the 36 sites and found no significant effect. Post hoc comparisons between habitat types were conducted using Tukey’s HSD tests. All statistical analyses were conducted in R version 4.4.0 (R Core Team, 2016), using the following packages: lmerTest for GLMs and LMs, vegan (function adonis2) for PerMANOVA, emmeans for post hoc comparisons and spdep for Moran’s I values.