Abstract: The Poisson’s and Zero Inflated Poisson (ZIP) models that meet the data structure of forest fire occurrence were used to explore the relationships between the forest fire occurrence and climate factors in Daxing’an Mountains in 1980-2005. Compared with the ordinary least squares (OLS) model which often produced poor fitting results (R²=0.215), the Poisson’s and ZIP models operated better, and had better prediction ability on the forest fire occurrence. The AIC and Vuong tests further indicated that ZIP model produced better fitting results, and thus, had better prediction ability than Poisson model.

Key words: OLS model, Poisson model, ZIP model, forest fire, weather factor, southeast Pacific, longline, bycatch, vertical distribution.