Modeling Probable Wild Fire Types Using Multinomial Logistic Regression Model in North Eastern Alberta

Author: Fahim Hossain, Cullen Mulroy and Hon Cooper.

Project Summary: 

Wildfire is very common in Northern America and causes great economic loss. Different researchers have used different models for predicting wildfire incidences. In this study, a multinomial logit model has been used to predict different fire size occurrences from a dataset that incorporates anthropogenic, environmental and meteorological variables. The study area extent for this study includes Fort McMurray and Edmonton regions. The results show that those severe fires are more likely in the eastern portion of the study region. The majority of the study region is predicted to have class 1 (least severe) fires. This study, however, lacks an accuracy assessment which is suggested for further study.






Predictive mapping of wildfire types is a useful tool to resource managers in Alberta. It can provide information that can be utilized in recourse allocation planning as well as understanding environmental processes. The data available for this study limited the statistical models that could be used to predict fire types. In the end, the multinomial logit model was used to predict which fire class is the most likely across the study region. This model yielded improved results over previous spatial autoregressive models. The model incorporated 15 variables because the variables were accounting for different types of fire classes. Future studies could focus on creating a fire likelihood map that predicts the probability of a fire occurring throughout the study region.