Inferring long distance seed dispersal from seedling count data: a hierarchical Bayesian approach

Background/Question/Methods

Long distance seed dispersal (LDD) can have major consequences for plant population and community dynamics.  Because rates of LDD can enable range expansion of invasive species, plant migration in response to climate change and species persistence in fragmented landscapes, LDD also has relevance for biodiversity conservation.  Understanding LDD requires quantifying rates of seed arrival at long distances from the source population.  However, the large spatial scale of LDD, the rarity of LDD events, and the difficulty of finding dispersed seeds, render direct measurements of LDD logistically difficult.   Established seedlings can often be found and counted much more efficiently than seeds, however, seedling abundance represents both seed arrival and establishment of seeds into seedlings.  Here, we present a method for estimating seed arrival from seedling count data and seed addition experiments, using a hierarchical Bayesian approach.  We illustrate this method using a canopy tree species in a patchy landscape in Western Thailand, where LDD is likely to play a crucial role in tree species distributions.  The tree species, Miliusa horsfieldii (Annonaceae), has seeds which are dispersed by a diverse assemblage of wide-ranging mammals, including bears, civets and primates.  We quantified rates of seed arrival for this species across a 5 km transect, from a source patch where the tree is common to habitat where the tree is rare.

Results/Conclusions

 Our method enabled estimation of a rare event, seed arrival at long distances (> 2 km) from the source patch, and provided insight into landscape-level patterns of seed arrival. Across the entire landscape, including the source patch, predicted seed arrival was very heterogeneous, with some areas predicted to receive many seeds and most receiving close to zero, possibly reflecting patterns of animal movement.  Because we simultaneously estimated seed arrival from seedling count data and seed addition experiment, our predictions for seed arrival propagate uncertainty in these data sources.  Samples from the joint posterior distribution for the seed arrival model can be used as input into spatially-explicit demographic or distribution models, which often require estimates of dispersal.  Estimating seed arrival from data on seedling abundance and establishment using hierarchical Bayesian models provides one solution to the difficult challenge of how to quantify rare LDD events.