Spatial analysis of tree species occurrence using generalized linear model and bayesian approach: a case study of Mt. Kenya region

Abstract

The development of predictive distribution models has become important in making predictions about occurrence of species based on variables derived from remote sensing or Geographical Information System (GIS). This project investigates the hypothesis that environmental variables can be used to predict the occurrence of species. Generalized linear and Bayesian models were developed to predict the occurrence of Grevillea robusta, Croton megalocarpus and Carica papaya species on Mt.Kenya region. The environmental (independent) variables used included rainfall, altitude, agroecological zones and vegetation class. The models were fitted on species presence/absence data sampled in a 265 plots vegetation survey carried out by ICRAF between 1999-2004. For mapping a 25344 grid data set was used. The GLM results showed that the model for vegetation class and agroecological zones predicted the occurrence of Grevillea robusta with a very small error. Although ", most levels for vegetation and agroecological zones were not significant, they explained most deviance for the three species. Altitude gave a good prediction for both Carica papaya and Croton megalocarpus, while rainfall predicted well the occurrence of Croton. The Bayesian results showed rainfall and altitude as good predictor for the three speciesmodeled .