Generalized linear models to study spatial distribution of tree species in Argentinean arid Chaco

This work adapts some generalized linear models in order to study the spatial pattern of an important tree species. The classical multivariate Ising model, which incorporates the dependence on neighbour individuals in a regular lattice, was adapted by setting a Poisson regression with an extra variation parameter to fit over-dispersion. Because the spatial pattern is only evident to a special reference scale, plots were sampled at two different scales. Two individual presence-absence matrices were analysed for each case through over-dispersion Poisson regression and log-linear models, including binary indicators for a neighbour in the four directions in the linear predictor. The results showed that the species, in the adult stage, has a spatial distribution in patches having no more than two adult individuals.