Latent variable models for multi-species counts modeling in ecology

Herliansyah R, Fitria I. 2018. Latent variable models for multi-species counts modeling in ecology. Biodiversitas 19: 1871-1876. High-dimensional multi-species counts are often collected in ecology to understand the spatial distribution over different locations and to study effects of environmental changes. Modeling multivariate abundance is challenging as we need to consider the possibility of interactions across species. Latent variable models are the recent popular approaches in statistical ecology to address such issue that has a similar framework to ordinary regression models. In this paper, we employed the poisson distribution for modeling count responses and a negative binomial distribution for more frequent zeros in observations. The implementation of a latent variable model, Generalized Linear Latent Variable Models (GLLVMs), was demonstrated on multi-species counts of endemic bird species collected in 37 different sites in Central Kalimantan, Indonesia. The main objectives were to study the effect of logging activities on abundance of endemic species and their interactions and to observe the habitat preference of certain species. Our study found that out of four endemic species, Alophoixus bres and Eurylaimus javanicus species were significantly affected by logging activities. The sign of parameters was negative indicating the logging activities in 1989 and 1993 bring significantly negative impacts on those species. The interaction created among species was strongly negative for major endemic species especially Alophoixus bres and Eurylaimus javanicus that prefer living in primary forest than in logging areas