Predicting local extinctions of Amazonian vertebrates in forest islands created by a mega dam

Hydropower projects are rapidly expanding across lowland Amazonia, driving the conversion of large tracts of once-continuous forests into archipelagos embedded within a vast open-water matrix. Forest vertebrate populations thus become stranded in habitat islands, with their persistence governed by a combination of species life-history traits, habitat patch, and landscape context. We investigate the patterns of species extinction of 34 arboreal and terrestrial vertebrate species within three continuous forest sites and 37 land-bridge islands within one of the largest South American hydroelectric reservoirs, based on a combination of camera trapping, line-transect censuses, sign surveys, and armadillo burrow counts. Forest area was the best predictor of species persistence, so we classified all species into three levels of vulnerability to habitat insularization, with most species defined as 'area-sensitive'. However, island occupancy was decisively determined by individual species traits, with wide-ranging species and poor dispersers showing high local extinction rates. We detected higher island occupancy rates of large vertebrate species compared to other Neotropical fragmented forest landscapes, suggesting that this is critically attributed to the absence of hunting pressure at Balbina. Nevertheless, most terrestrial and arboreal species have been driven to local extinction within the vast majority of islands, which have been largely defaunated. We predicted species composition across all 3546 islands within the reservoir, indicating that only <= 2% of all islands continue to harbour at least 75% of all species. To minimise loss of vertebrate diversity, future hydroelectric dam projects in lowland tropical forests, if unavoidable, should consider their geographic location and landscape structure to maximise both island size and landscape connectivity, and set aside strictly protected reserves within reservoir areas. (C) 2015 Elsevier Ltd. All rights reserved.