Aim The spatiotemporal connectivity of forest patches in lowland agricultural landscapes and their age matter to explain current biodiversity patterns across regional as well as biogeographical extents, to the point that it exceeds the contribution of macroclimate for plant diversity in the understory of temperate forests. Whether this holds true for other taxonomic groups remains largely unknown. Yet, it has important consequences for ecosystem functioning and the delivery of ecosystem services. Focusing on carabid beetle assemblages, we assessed the relative importance of macroclimatic, landscape, and patch attributes on driving local species richness (-diversity) and species dissimilarity between patches (-diversity).

Location Deciduous forest patches in seven regions along a 2,100-km long latitudinal gradient across the European temperate forest biome, from southern France to central Sweden.

Methods We sampled 221 forest patches in two 55-km landscape windows with contrasting management intensities. Carabid beetles were classified into four habitat-preference guilds: forest-specialist, forest-generalist, eurytopic, and open-habitat species. We quantified the multi-level environmental influence using mixed-effects models and variation partitioning analysis.

Results We found that both - and -diversity were primarily determined by macroclimate, acting as a large-scale ecological filter on carabid assemblages among regions. Forest-patch conditions, including biotic and abiotic heterogeneity as well as patch age (but not patch size), increased -diversity of forest species. Landscape management intensity weakly influenced -diversity of forest species, but increased the number of non-forest species in forest patches. Beta-diversity of non-forest species increased with patch heterogeneity and decreased with landscape management intensity.

Main conclusions Our results highlight the leading role of broad macroclimatic gradients over local and landscape factors in determining the composition of local carabid communities, thereby shedding light on macroecological patterns of arthropod assemblages. This study emphasizes the urgent need for preserving ancient forest patches embedded in agricultural landscapes, even the small and weakly connected ones.

All data collection information may be found in the published article related to this dataset and its appendices.

Study sites We collected data across a total of 221 deciduous forest patches, distributed among seven European regions. Regions were distributed along a south-west-to-north-east gradient of ca. 2,100 km across the European temperate forest biome (Fig. 1). In each region, we selected two 5 km 5 km landscape windows differing by their degree of landscape permeability (see Valds et al., 2015 for more details). The first window was characterized by forest patches embedded in an intensively cultivated open field landscape (hereafter open field). The second window contained forest patches more or less connected by woody corridors (e.g., hedgerows) within a less intensively managed landscape, dominated by grasslands and small crop fields (hereafter bocage). A detailed and updated list of landscape and macroclimatic variables associated with each landscape window can be found in Vanneste et al. (2019). For each landscape window, we computed area, perimeter, and age of all forest patches using digitized maps (one contemporary map at a scale of 1:25,000 and historical maps from the 17th, 18th, 19th, and 20th centuries) in a Geographic Information System (GIS; ArcGIS 9.3, ESRI). Patches were subsequently distributed among four classes according to their area and age: small ( median patch area value for the focal window) and recent ( 150 years); large ( median patch area value for the focal window) and recent; small and ancient ( 150 years); and large and ancient.

Carabid sampling and habitat preference Whenever possible, we selected four forest patches (i.e., four repetitions) per level of patch size-age combination (n = 4) and per landscape window (n = 14). A perfect balanced design was achieved in five out of seven regions, to finally include a total of 221 forest patches to trap carabid beetles (i.e., 16 forest patches in each window, except 14 in the openfield window in eastern Germany and 15 in the openfield window in southern Sweden). For this purpose, we used 10-cm diameter pitfall traps installed for 14 consecutive days in both spring (ca. April) and summer (ca. August) 2013. Traps were filled with 200 mL of a 50 % conservative solution of ethylene-glycol and a few drops of detergent, and protected from litter and rain fall by aluminum roofs. A total of four pairs of traps were disposed in each patch as follows. A first pair, consisting of two traps separated by a plastic barrier (100-cm long, 18-cm high), was disposed into the inner part of a south-facing edge (or, when not possible, first a west-, and then an east-facing edge was chosen). This setup was replicated 5-m apart along the same edge. A third pair was installed at the barycenter of the forest patch (except in eastern Germany, where all traps were located in the edge), and similarly replicated (fourth pair). The plastic barrier was always parallel to the selected forest edge. To make data comparable among the seven studied regions, and because of the latitudinal climatic gradient covered by our study, the two sampling sessions carried out in each region started when local growing degree hours (GDH ; Graae et al., 2012) reached values of ca. 10,000C h and 20,000C h, respectively. Following trap collection, carabid beetles were sorted in the lab in a 70% ethanol solution and identified to the species level following Jeannel (1941, 1942). Species names follow Fauna Europaea (de Jong et al., 2014). Data from all pitfall traps of a given patch and from the two trapping sessions were pooled in all subsequent statistical analyses. Species were distributed among four guilds, according to their habitat preference and using knowledge from the scientific literature (Hůrka, 1996; Sadler et al., 2006; Gaublomme et al., 2008; Brunicke Trautner, 2009): forest-specialist species, limited to stable, mature forest stands; forest-generalist species, occurring in any type of forest stand, in ancient as well as recent forest; open-habitat species, associated with non-forest habitats such as grasslands and arable lands; and eurytopic species, occurring in open habitats and tolerating transiently forest habitats. We could not assign only one species (Oodes helopioides, n = 1 individual) to any group because of a lack of information in the literature.

Environmental variables Three groups of explanatory variables (patch, landscape, and macroclimatic variables) were derived from field observations, historical archives, or global climatic layers, for and around each forest patch using a GIS.

Macroclimatic variables To assess the influence of macroclimate on species diversity, we extracted 10 candidate bioclimatic variables from the WorldClim global database (1-km resolution, http://www.worldclim.org), and averaged each variable for each forest patch using all 1-km2 pixels intersecting it. Four macroclimatic variables were retained for further analyses, based on a principal component analysis (PCA; see Valds et al., 2015), namely maximum temperature of the warmest month (MaTWm; BIO5); minimum temperature of the coldest month (MiTCm; BIO6); precipitation of the wettest month (PWm; BIO13); and precipitation of the driest month (PDm; BIO14). The selection was made in a way to minimize the correlation between variables and to maximize the correlation with the PCA axes. Our variables were correlated (Pearsons r) as follow with the PC1 (67.0 % of explained variance) and PC2 (17.2 %): BIO5 (PC1: r = -0.64; PC2: r = -0.12), BIO6 (PC1: -0.93; PC2: 0.34), BIO13 (PC1: -0.52; PC2: -0.85), and BIO14 (PC1: -0.94; PC2: -0.04).

Landscape variables Landscape variables were computed for five concentric doughnut-like buffers of increasing width around each forest patch: 50; 100; 250; 500; and 1,000 m width. We used Corine Land Cover 2006 (Bttner Kosztra, 2007) to map the distribution of woodland, cropland, and grassland. We digitized hedgerows from aerial photographs. As proposed by Martin Fahrig (2012) and Fahrig (2013), we considered composition-based measurements of spatial isolation for each forest patch, by calculating the proportion of each cover type and the hedgerow density within each buffer.

Patch-scale attributes We included patch area and patch age as covariates in our analyses, to account for both the species-area and species-time relationship (Rosenzweig, 1995). We took the coefficient of variation in elevation values (CVe) within a given forest patch using the ASTER Global Digital Elevation Map at 30-m resolution (see Valds et al., 2015 for further details), as a proxy for heterogeneity of abiotic conditions (incl. microclimate, soil conditions, and light availability) (Lenoir et al., 2017; Graae et al., 2018). Finally, we computed a dissimilarity index in understory plant species composition within each forest patch (i.e., intra-patch b-diversity; see Valds et al., 2015 for details on computation), separately for forest plant specialists and generalists following distinction criteria as in Valds et al. (2015).

Exact spatial coordinates of sampling points have been removed from the public dataset. They may be available upon direct request to the corresponding author.