Diversidade beta na floresta ombrófila densa da Amazônia Brasileira

Abstract

The drivers of the tropical forest beta diversity are still a matter of intense debate. No consensus has been reached; different studies have led to different conclusions. Nonetheless, mapped forest types are frequently used by conservation planners to represent change in floristic composition. The object of this work is to evaluate this practice, and to verify how the composition of Dense Lowland Forest (FTB) changes through geographic space and along environmental gradients of climate seasonality, soil fertility and altitude. Floristic composition variation of FTB at the genus level for canopy trees (> 31,2 cm DAP) was examined using the RADAMBRASIL Project´s inventory plots, grouped into 49 objects with ~260 trees each. After the introduction of objects from a different forest type (Open Forest with Palms), no significant compositional difference between these two vegetation types was by ANOVA tested in a 2-dimensional NMDS ordination space. For canopy trees at the genus level, these two forest types are therefore poor proxies for beta diversity. The environmental and geographic variables were then used in regression trees applied to FTB data only, that indicated the existence of a compositional separation on the east-west axis, and groups determined by altitude and seasonality. The contributions of geographical and environmental distances were examined through regressions between matrices, that consisted of all objects pairs. Two sets of points were examined a complete one and another without 14 objects that exceeded the limit of altitude and climate defined for FTB. For the quantitative floristic distance and full set of points, environmental and geographical distances together explained 38% of the total variation in composition. The simple effect of environment distance (35%) was larger than that of geographic distance alone (26%). When the number of objects was reduced, both distances explained together only 22% on the variation, overlap was much less (4%) and the simple effect of geographic distance exceeded the simple effect of environmental distance. Analyses with qualitative data (presence/absence) led to similar results, for both the complete and the reduced sets of objects. Generalized Dissimilarity Modelling (GDM), a non linear approach to matrix regression, gave results very different from those obtained with linear techniques. This could be a consequence of the GDM´s capacitity for isolating the correlations between variables, but given the limitations of the model version presently available, the results should be interpreted with caution.