Decomposing biodiversity data using the Latent Dirichlet Allocation model, a probabilistic multivariate statistical method.

We propose a novel multivariate method to analyse biodiversity data based on the Latent Dirichlet Allocation (LDA) model. LDA, a probabilistic model, reduces assemblages to sets of distinct component communities. It produces easily interpretable results, can represent abrupt and gradual changes in composition, accommodates missing data and allows for coherent estimates of uncertainty. We illustrate our method using tree data for the eastern United States and from a tropical successional chronosequence. The model is able to detect pervasive declines in the oak community in Minnesota and Indiana, potentially due to fire suppression, increased growing season precipitation and herbivory. The chronosequence analysis is able to delineate clear successional trends in species composition, while also revealing that site-specific factors significantly impact these successional trajectories. The proposed method provides a means to decompose and track the dynamics of species assemblages along temporal and spatial gradients, including effects of global change and forest disturbances.

Assessment of forest fuel loadings in Puerto Rico and the US Virgin Islands.

Quantification of the downed woody materials that comprise forest fuels has gained importance in Caribbean forest ecosystems due to the increasing incidence and severity of wildfires on island ecosystems. Because large-scale assessments of forest fuels have rarely been conducted for these ecosystems, forest fuels were assessed at 121 US Department of Agriculture forest service inventory plots on Puerto Rico, Vieques, and the US Virgin Islands. Results indicated that fuel loadings averaged 24.05 Mg ha(-1) in 2004-2006. Forest litter decreased from wetter to drier forest life zones. These island forests showed a paucity of coarse woody fuels (CWD) (2.91 Mg ha(-1)) and relatively greater quantities of smaller-sized fine woody fuels (FWD) (10.18 Mg ha(-1) for FWD and 10.82 Mg ha(-1) for duff/litter) when compared to continental tropical forests. Between 2001 and 2006, CWD fuel loads decreased, while fine fuels and litter increased, such that total fuel loads remained constant on a subset of plots on Puerto Rico. This trend indicates that continued decomposition of CWD deposited by the last severe hurricane is balanced by increasing inputs of FWD from recovering and maturing secondary forests. Forest disturbance cycles and successional development must be taken into account by agencies charged with fire protection and risk assessment.