We further applied the sampling results (i.e., Fig. 4)
to develop more sophisticated models using logistic
regression, discriminant function analysis, GARP, fuzzy
logic and artificial neural network using a 1 × 1 Km grid
dataset. Instead of treating all explanatory variables as
equal, these new models tried to weight the variables
differently. Our results indicated that overall accuracies
ranged from 63% to 85% and the GARP model had the
highest predictive power (Lee et al., 2006). With the
ensemble forecasting approach (Araújo and New,
2007), we further developed a robust predictive model
for Fairy Pitta. The combined model yielded a narrower
predictive range than the rule-based model in our study,
though the overall pattern was similar.