全球变化知识资源中心

Aim Mangrove forests are among the most threatened and rapidly vanishing, but poorly understood ecosystems. We aim to uncover the variables driving mangrove biodiversity and produce baseline biodiversity maps for the Sundarbans world heritage site-the Earth's largest contiguous mangrove ecosystem. Location The Bangladesh Sundarbans, South Asia. Methods We collected species abundance, environmental and disturbance data from 110 permanent sample plots (PSPs) covering the entire Bangladesh Sundarbans (6,017 km(2)). We applied generalized additive models to determine the key variables shaping the spatial distributions of mangrove diversity and community composition. Biodiversity maps were constructed using covariate-driven habitat models, and their predictive performances were compared with covariate-free (i.e., direct interpolation) approaches to see whether the inclusion of habitat variables bolster spatial predictions of biodiversity or whether we can rely on direct interpolation approaches when environmental data are not available. Results Historical forest exploitation, disease, siltation and soil alkalinity were the key stressors causing loss of alpha and gamma diversity in mangrove communities. Both alpha and gamma diversity increased along the downstream-to-upstream and riverbank-to-forest interior gradients. Mangrove communities subjected to intensive past tree harvesting, disease outbreaks and siltation were more homogeneous in species composition (beta diversity). In contrast, heterogeneity in species composition increased along decreasing salinity and downstream-to-upstream gradients. We find that the surviving biodiversity hotspots (comprising many globally endangered tree species) are located outside the established protected area network and hence open to human exploitation. We therefore suggest bringing them immediately under protected area management. Main conclusions We provide the first habitat-based modelling and mapping of alpha, beta and gamma diversity in threatened mangrove communities. In general, habitat-based models showed better predictive ability than the covariate-free approach. Nevertheless, the small margin of differences between the approaches demonstrates the utility of direct interpolation approaches when environmental data are unavailable.