摘要：
Aims Relationships between local habitat heterogeneity and tree communities in miombo woodlands have been very little studied. While some studies have addressed this topic at broad scales and based on few environmental parameters, this study aims at (i) detecting fine-scale habitats (≤10 ha) on the basis of a detailed characterisation of soil explicitly considering past anthropogenic disturbances, and an exhaustive census of the tree community, and at (ii) searching for indicator tree species corresponding to the resulting habitats.
Methods The study was carried out in the miombo woodland of Mikembo Forest Reserve, Upper Katanga, The Democratic Republic of the Congo. A complete census of the tree community was conducted in a 10-ha forest dynamics plot comprising 160 adjacent quadrats of 25 × 25 m, with a total of 4604 trees (diameter at breast height> 10 cm). Thirty-six physicochemical soil parameters were measured. Studying the frequency distribution of soil charcoal content allowed identifying local signature of past human agriculture in the soil. Two strategies were used to define habitats: (i) a combination of principal component analysis (PCA) on soil variables and Ward clustering and (ii) multivariate regression trees (MRT) to search for key soil parameters allowing the best prediction of species composition. Tree-habitat associations were tested by means of a robust statistical framework combining the IndVal index and torus randomisations.
Important findings The forest contained 82 tree species and a significant proportion of wet miombo species (e.g. Marquesia macroura). We detected a strong east–west edaphic gradient driven by soil texture; most chemical soil parameters followed this pattern. Five habitats were identified based on soil factors and floristic composition. Nine indicator species of these habitats were found. The key soil factors discriminating habitats were total calcium, available forms of phosphorus and clay content. Even though past agricultural practices were successfully detected in soils, they did not display any significant influence neither on habitat differentiation nor on the associated tree communities. Based on an unprecedented large number of soil parameters, fine-scale soil heterogeneity and niche partitioning were shown to contribute to the variability of the floristic composition in this forest. Our results indicated that considering the most variable environmental parameters, as in PCA, is a poor manner for defining habitats. In contrast, combining MRT with the IndVal index and torus randomisation has proved to be a much more robust and sensitive approach to highlight tree-habitat associations at this scale. The common dichotomous viewpoint of considering deterministic and neutral effects as acting at broad and fine scales, respectively, is not confirmed when measuring suitable environmental variables, even in a case where the physical environment does not exhibit strong heterogeneity.
