Detecting large-scale diversity patterns in tropical trees: Can we trust commercial forest inventories?

In this paper we seek to identify the floristic determination biases contained in large-scale commercial inventories conducted by logging companies and to determine whether this impacts on the observed patterns of alpha and beta diversity. The study focused on floristic data recently collected by industrial timber companies in the tropical forests of the Central African Republic (28,229 0.5-ha plots spread over 14,000km2). A subset of these plots (n = 1107) was later re-sampled for controlling purposes by experienced botanists. The proportion of agreement between the two samplings was assessed for each species and independently for small and large trees, and at genus and family resolutions. Unsurprisingly, large trees and common species were more accurately identified than small trees and rare species. We found that the quality of the floristic determination increased slightly from species to families. We also detected a significant variation between concessions in the quality of the floristic determination that was more dependent on working conditions during forest inventories than on field workers. Contrary to a widespread belief, we did not find a strong bias toward commercial species, showing that commercial inventory data could also be valid for non-commercial species in ecological studies. Finally, we found that both alpha and beta diversity patterns in commercial inventories were highly consistent with those of the re-sampled inventory. This latter result shows that commercial inventories are well suited to detect large-scale patterns of floristic variation. Large-scale commercial inventories could thus play an important role in the identification of large-scale patterns in tropical tree diversity. This could enhance our ability to manage tropical forests by designing representative reserve networks and developing management plans that integrate diversity patterns at the landscape scale. (Résumé d'auteur