Agriculture and development transform forest ecosystems to human‐modified landscapes. Decades of research in ecology have generated myriad concepts for the appropriate management of these landscapes. Yet, these concepts are often contradictory and apply at different spatial scales, making the design of biodiversity‐friendly landscapes challenging. Here, we combine concepts with empirical support to design optimal landscape scenarios for forest‐dwelling species. The supported concepts indicate that appropriately sized landscapes should contain ≥ 40% forest cover, although higher percentages are likely needed in the tropics. Forest cover should be configured with c. 10% in a very large forest patch, and the remaining 30% in many evenly dispersed smaller patches and semi‐natural treed elements (e.g. vegetation corridors). Importantly, the patches should be embedded in a high‐quality matrix. The proposed landscape scenarios represent an optimal compromise between delivery of goods and services to humans and preserving most forest wildlife, and can therefore guide forest preservation and restoration strategies.

Global biodiversity is increasingly threatened by land‐use change, but the direct and indirect drivers of species diversity in human‐modified tropical landscapes are poorly known. Forest‐dependent species are expected to be particularly sensitive to changes in landscape composition (e.g., forest loss) and configuration (e.g., increase of forest edges), both directly and indirectly through cascading landscape effects on local patterns of forest structure and resource availability. In contrast, non‐forest‐dependent species are probably more strongly related to landscape changes than to local forest patterns, as these species are able to use resources not only from the forest, but also from other landscape elements over larger spatial scales. We tested these hypotheses using structural equation modeling. In particular, we sampled 20 landscapes (115 ha each) from the Brazilian Atlantic rainforest to assess the effect of landscape‐scale forest cover and amount of forest edges on the diversity of frugivorous birds, both directly and indirectly through the effect that these landscape variables may have on vegetation complexity and fruit biomass. We separately assessed the response of forest‐dependent and non‐forest‐dependent frugivores to infer potential mechanisms underlying bird assemblages in fragmented landscapes. The diversity of forest‐dependent birds mainly decreased with the simplification of vegetation complexity in more deforested landscapes, but increased with increasing fruit biomass in more forested landscapes (indirect effects). Both patterns were significant, thus supporting a strong bottom‐up control, i.e., local habitat simplification and resource scarcity in highly deforested landscapes limits the maintenance of forest‐dependent birds. Conversely, but as expected, non‐forest‐dependent birds were more strongly and directly related to landscape‐scale patterns. In particular, landscapes with higher forest edge amount showed higher bird species diversity, probably because the increasing length of ecotones and interspersion/juxtaposition of different habitat types in landscapes with more forest edges can increase resource availability and foraging efficiency of non‐forest‐dependent birds. As the seed dispersal services offered by forest‐dependent species cannot be ecologically compensated for by the proliferation of non‐forest‐dependent species, preventing forest loss is imperative to maintain forest‐dependent birds and forest regeneration in this vanishing biodiversity hotspot.