Drivers of plant species assemblages in forest patches among contrasted dynamic agricultural landscapes

1. Although habitat fragmentation is recognized as a major threat to biodiversity, few studies have examined the relative importance of local, landscape and historical factors in controlling local species assemblages, and how these factors interact, in patchy ecosystems. We quantified the direct and indirect effects of patch size, patch heterogeneity, agricultural intensity and patch age on plant species richness and composition of forest patches embedded in agricultural landscapes. 2. In six 5 x 5 km-sampling landscape windows, we surveyed each forest patch for vascular plant species and collected three sets of independent variables, describing patch size and heterogeneity, landscape composition and history. The six windows were arranged along a gradient of agriculture intensity in rural landscapes of the Picardy region (N France). 3. We used non-metric multidimensional scaling (NMS) to detect major environmental gradients underlying variation in species composition among patches. We then constructed structural equation models (SEM) to quantify the direct and indirect effects of the three sets of variables on local plant diversity, which was successively incorporated as patch scores along the first three NMS axes, woody species richness, forest herb species richness, and non-forest herb species richness. 4. A major influence of the landscape matrix on local species composition was revealed by NMS and subsequent SEM, mainly through non-forest herb species, which explained most of the between patch floristic dissimilarity. Species richness increased with patch heterogeneity, whereas patch area never had a direct effect. Forest herbs were more responsive to patch age and connectivity than other species, whereas non-forest and woody species were more influenced by agriculture intensity in the surrounding matrix. 5. Synthesis. We used one of the largest data sets ever collected in temperate fragmented forests to build, for the first time, a structural model incorporating all suspected drivers of local plant communities. We showed that the number and identity of local species coexisting in successional fragments of a forest metacommunity at a given time is controlled by a unique combination of interacting local, landscape and historical factors. Preserving the largest, oldest fragments and favouring species movements in the surrounding matrix is the best way to conserve forest specialists in changing rural landscapes.