Skip to main content
eScholarship
Open Access Publications from the University of California

Evaluating the Use of Various Distance Metrics for Assessing a Model's Wilfire Prediction Performance

  • Author(s): Chao, Jeffrey
  • Advisor(s): Schoenberg, Frederic P
  • et al.
Abstract

While being able to predict wildfires is crucial in being able to safely control them, also important are valid methods and metrics to evaluate the effectiveness and performance of these predictions. Thus, in this paper we adapt fourteen existing distance metrics for use in assessing the accuracy of wildfire predictions. These fourteen distance metrics include: Simple Matching, Simpson, Kulczynski, Jaccard, Yule, and Phi. The theoretical implications of each distance metric are examined, and these metrics are applied to comparisons of actual wildfire perimeter data from ten wildfires and their predicted perimeters generated by the FARSITE fire prediction software. By doing so, we are able to illuminate the characteristics of each of the distance metrics (takes rotations into account, punish “overburn”/”underburn”, etc.), as well as illustrate how these distance metrics can be easily applied and chosen from based on one's requirements of a prediction model.

Main Content
Current View