Factors that Influence Daily Human-Caused Forest Fires in Alberta

• Main Author: Morrison, Kimberly H.
• Other Authors: Wotton, Mike (University of Toronto), Flannigan, Mike (Renewable Resources), Beverly, Jen (Renewable Resources)
• Format: Thesis
• Language: English
• Published: University of Alberta. Department of Renewable Resources. 2016
• Subjects: envir; anthro-bio; Flat Top; Fort McMurray
• Online Access: http://hdl.handle.net/10402/era.43417

Specialization: Forest Biology and Management Degree: Master of Science Abstract: Humans are the major cause of forest fires in the spring in Alberta, and have resulted in major property damage in both the Flat Top Complex fires in 2011 and the Fort McMurray fire in 2016. Fire occurrence prediction (FOP) models can help predict when and where fires can be expected in order to help fire managers manage resources. In Alberta, these FOP models need to be improved, especially in regards to spring fire starts and the timing of the end of the spring fire season as most human-started forest fires in Alberta occur in the spring. Candidate models were created to explore which independent variables best predict human-caused fire starts in Alberta. The independent variables are separated into four groups: spatial distribution, Fire Weather Index System codes and indices, human influences, and seasonality. Finally, several model forms were explored to determine the best model, as determined by best fit to the data and/or best predicted fire occurrence. These were: Generalized Linear Model (GLM), Hurdle Model and Zero-Inflated Model, each with a Poisson and negative binomial link. A GLM with a negative binomial link and the following variables predicted the number of human-caused fire starts the best: FFMC, FWI, ECOREGION, FFMC X ECOREGON, and SEASON3 (a three level variable with a transition season between spring and summer). This model had a RMSE of 0.697 when tested on a bootstrapped set of test data. This model could be used as the basis of future FOP model research in Alberta, and management of wildfire fighting resources in conjunction with other fire activity prediction methods.