Results of total likelihood and spatial likelihood (Table 1) indicate also a gain of the EAST ∗EEPAS model with respect to the EEPAS model.
When Maxent was used to evaluate the correlation between observed deforestation and selected explanatory variables for the years 2000-2006, andessibility andistancece to previous deforestation emerged as the most important variables, positively correlating with deforestation likelihood (Table 1).
Indeed, the full model, i.e. asymmetric dispersal between all populations both within landscapes and between different landscapes, had the highest likelihood (Table S5).
However, 74% of all respondents indicated the training had 'definitely increased' (48%) or 'mostly increased' (26%) the likelihood (Table 6).
Another striking feature was that once parameters for modeling rate variation had been incorporated into the model, unlinking topologies did not seem to have as pronounced effect on the model likelihood (Table 2), compared to the models without rate variation.
The Bayes factor tests showed extensive incongruence between partitions, at least in the sense that relaxing the assumption of a common topology parameter always gave a better model likelihood (Table 1).
However, the ML tree (summarized in Figure 4) and several topologies with alternative branching patterns among at the basal-most nodes are statistically indistinguishable under maximum likelihood (Table 4).
When Maxent was used to evaluate the correlation between observed deforestation and selected explanatory variables for the years 2000-2006, andessibility andistancece to previous deforestation emerged as the most important variables, positively correlating with deforestation likelihood (Table 1).
