Publication Date:
2021-10-18
Description:
Canada experiences a relatively large number of tornadoes, which can cause a significant amount of damage and fatalities. In the present study, a preferred prediction model for the spatially varying tornado occurrence rate is developed for Canada. The development takes into account the most commonly used spatial stochastic models and the underreporting due to low population density. It incorporates the annual average cloud-to-ground lightning flash (ACGLF) density and annual average thunderstorm days (ATD) as covariates in the prediction model. The model parameters estimation is carried out by using both the maximum likelihood method and the Bayesian inference.The analysis results indicate that the negative binomial model is preferable to the zero-inflated Poisson model and the Poison model. The results show that the tornado occurrence in Canada is associated with large overdispersion. Also, the statistical analysis indicates that the prediction model for the tornado occurrence rate developed based on Bayesian inference is relatively insensitive to the assumed “non-informative” prior distributions. A prediction model is suggested for the spatially varying tornado occurrence rate based on the negative binomial model with the ACGLF density and ATD as covariates.
Print ISSN:
1558-8424
Electronic ISSN:
1558-8432
Topics:
Geography
,
Physics
