Faculty
Science
Supervisor Name
Dr. Douglas Woolford; Dr. Kevin Granville
Keywords
Wildland Fire Science, human-caused wildland fire, Poisson models, Negative Binomial models, Cross-Validation
Description
Wildland Fire Science has become an increasingly hot topic in recent years. The goal of this report is to investigate human-caused wildland fire occurrence prediction. The two main predictors of interest are the mean value of the Fine Fuel Moisture Code (FFMC) and the month when a fire ignites. An Exploratory Data Analysis is presented first, after which we fit models to predict daily fire counts. We first consider Poisson models to fit the count data, but also attempt to fit Negative Binomial models to deal with overdispersion. We compare these models in the following ways: plotting the difference in observed and predicted values, making a standardized residual plot, comparing the Akaike Information Criterion, and conducting Cross-Validation to approximate predictive accuracy. Our results show that the addition of the month covariate improves model fit, and that the Negative Binomial models are very similar to the corresponding Poisson models in performance.
Acknowledgements
We acknowledge the support of the Natural Sciences and Engineering Research Council of Canada, the Ontario Ministry of Natural Resources and Forestry, and the University of Western Ontario’s USRI program.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Document Type
Paper
Included in
Exploring Human-Caused Fire Occurrence Prediction
Wildland Fire Science has become an increasingly hot topic in recent years. The goal of this report is to investigate human-caused wildland fire occurrence prediction. The two main predictors of interest are the mean value of the Fine Fuel Moisture Code (FFMC) and the month when a fire ignites. An Exploratory Data Analysis is presented first, after which we fit models to predict daily fire counts. We first consider Poisson models to fit the count data, but also attempt to fit Negative Binomial models to deal with overdispersion. We compare these models in the following ways: plotting the difference in observed and predicted values, making a standardized residual plot, comparing the Akaike Information Criterion, and conducting Cross-Validation to approximate predictive accuracy. Our results show that the addition of the month covariate improves model fit, and that the Negative Binomial models are very similar to the corresponding Poisson models in performance.