Detecting changes in spatial characteristics of Colorado human-caused wildfires using APRIORI-based frequent itemset mining

Abstract

The phenomenon of human-caused wildfire presents significant dilemmas for states like Colorado, given a growing population and large portions of forested mountain ecoregions. This study seeks to understand the changes in characteristics of Western Colorado human-caused wildfire (HCF) occurrences from 1992 to 2015 while considering their locations and spatial context. Fires are grouped into three seven-year study periods, 1992–1999, 2000–2007, and 2008–2015. First, we constructed an extensive spatial database documenting the spatial characteristics of these fires. The spatial characteristics represent the spatial relationships and interactivity of fire occurrence locations with various environmental and anthropogenic entities. Second, an unsupervised data mining approach (APRIORI-based itemset mining) is used to extract the most frequent characteristics of the fire occurrences during these periods. Finally, we present a comprehensive visual graph-based evaluation approach to evaluate the mined results effectively. Our findings indicate explicit and fresh insights into changes in human-caused wildfire characteristics for the study area. The method presented and enhanced here is proven robust and very effective at extracting an extensive and comprehensive set of characteristics of HCFs. The operative spatial predication schemas and the result evaluation approach based on visual analytics are this study's successful and novel aspects.