The unified neutral theory of biodiversity and biogeography provides a promising framework that can be used to integrate stochastic and ecological processes operating in ecological communities. Based on a mechanistic non-neutral model that incorporates density-dependent mortality, we evaluated the deviation from a neutral pattern in tree species abundance distributions and explored the signatures of historical and ecological processes that have shaped forest biomes. We compiled a dataset documenting species abundance distributions in 1168 plots encompassing 16 973 tree species across tropical, temperate, and boreal forests. We tested whether deviations from neutrality of species abundance distributions vary with climatic and historical conditions, and whether these patterns differ among regions. Non-neutrality in species abundance distributions was ubiquitous in tropical, temperate, and boreal forests, and regional differences in patterns of non-neutrality were significant between biomes. Species abundance evenness/unevenness caused by negative density-dependent or abiotic filtering effects had no clear macro-scale climatic drivers, although temperature was non-linearly correlated with species abundance unevenness on a global scale. These findings were not significantly biased by heterogeneity of plot data (the differences of plot area, measurement size, species richness, and the number of individuals sampled). Therefore, our results suggest that environmental filtering is not universally increasing from warm tropical to cold boreal forests, but might affect differently tree species assembly between and within biomes. Ecological processes generating particularly dominant species in local communities might be idiosyncratic or region-specific and may be associated with geography and climate. Our study illustrates that stochastic dynamical models enable the analysis of the interplay of historical and ecological processes that influence community assemblies and the dynamics of biodiversity.