Disentangling stand and environmental correlates of aboveground biomass in Amazonian forests

Tropical forests contain an important proportion of the carbon stored in terrestrial vegetation, but estimated aboveground biomass (AGB) in tropical forests varies two‐fold, with little consensus on the relative importance of climate, soil and forest structure in explaining spatial patterns. Here, we present analyses from a plot network designed to examine differences among contrasting forest habitats (terra firme, seasonally flooded, and white‐sand forests) that span the gradient of climate and soil conditions of the Amazon basin. We installed 0.5‐ha plots in 74 sites representing the three lowland forest habitats in both Loreto, Peru and French Guiana, and we integrated data describing climate, soil physical and chemical characteristics and stand variables, including local measures of wood specific gravity (WSG). We use a hierarchical model to separate the contributions of stand variables from climate and soil variables in explaining spatial variation in AGB. AGB differed among both habitats and regions, varying from 78 Mg ha^?1 in white‐sand forest in Peru to 605 Mg ha^?1 in terra firme clay forest of French Guiana. Stand variables including tree size and basal area, and to a lesser extent WSG, were strong predictors of spatial variation in AGB. In contrast, soil and climate variables explained little overall variation in AGB, though they did co‐vary to a limited extent with stand parameters that explained AGB. Our results suggest that positive feedbacks in forest structure and turnover control AGB in Amazonian forests, with richer soils (Peruvian terra firme and all seasonally flooded habitats) supporting smaller trees with lower wood density and moderate soils (French Guianan terra firme) supporting many larger trees with high wood density. The weak direct relationships we observed between soil and climate variables and AGB suggest that the most appropriate approaches to landscape scale modeling of AGB in the Amazon would be based on remote sensing methods to map stand structure.