Just Accepted

This article has been peer reviewed and accepted for publication. It is in production and has not been edited, so may differ from the final published form.

A comparative analysis of wildfire initial attack containment objectives and modelling strategies in Ontario, Canada

Kennedy Korkola, Melanie Wheatley, Jennifer Beverly , Patrick James, Mike Wotton

Abstract

Background: Fire management agencies use the proportion of fires classified as initial attack (IA) success as a suppression performance metric, making IA success a common indicator of suppression effectiveness in research. The criteria and definition for IA success varies based on operational objectives, making comparisons across studies difficult. Aims: To examine the sensitivity of different time and size-based IA success definitions on model predictive accuracy and compare different modelling approaches. Methods: Using 30 years of historical fire report data from Ontario, Canada (n = 26,171), we developed logistic regression models, bagged classification trees and random forest models to predict IA success for eight different definitions. Model predictive accuracy, sensitivity, and specificity were assessed on an independent validation dataset. Key results: The eight definitions classified between 79% to 98% of fires as IA successes. There was no clear pattern between model strength across prediction metrics and IA success definition. Logistic regression generally outperformed machine learning methods in classifying IA escapes. Conclusions and implications: The definition of IA success does not greatly impact model performance across the modelling techniques used. Models of IA success and suppression-system performance metrics should be set with specific research or operational objectives in mind.

WF24104  Accepted 19 November 2024

© CSIRO 2024