Abstract:

In this study, based on data of the forest fire occurrence and meteorological variables in spring fireproofing period in Qiannan area, Poisson regression model, negative binomial model, zero-inflated negative binomial model and Hurdle model were respectively employed to predict the forest fires under fire-danger climate, and those models were compared with each other based on the prediction. The results showed that: Poisson regression model did not fit well into the over-dispersion data. Negative binomial distribution fitted better into the data than Poisson distribution. But both of them were not suitable for simulating zero drived dispersion data. Zero-inflated negative binomial regression model and Hurdle model were useful methods for such data. Zero inflated negative binomial regression model and Hurdle model performed better than other two models in predicting forest fires. Moreover, Hurdle model was even superior to zero-inflated negative binomial model.

Key words: forest fire, fire-danger weather, Poisson regression model, negative binomial model, zero-inflated negative binomial model, Hurdle model