Fighting Future Flames: Modelling Forest Fire Vulnerability in Nova Scotia, Canada

Abstract

Historically, fire has been, and continues to be, a natural driver of forest renewal and regrowth, shaping Earth's landscapes into what we see today. The complex relationship between changing climate patterns, fuel types, and human activity has contributed to an increase in the frequency of forest fires. An effective method to quantify and monitor the changes to a forest ecosystem is the use of integrated remote sensing and spatial analysis techniques. In the summer of 2023, Nova Scotia experienced their most devastating fire season with 220 fires burning 25,093 hectares of land, highlighting the growing importance of monitoring forest fire vulnerability. The goal of this study is to develop a suite of indicators that, when considered together, identify areas at potential high-risk of forest fires in Nova Scotia. Two study areas were considered: Upper Tantallon and Barrington fire locations from the summer of 2023 in Nova Scotia, Canada. An ISODATA unsupervised classification was performed to identify patterns of similar spectral characteristics among biophysical variables that was used to create a map of forest fire vulnerability using an ordinal scale. The input variables include spectral indices like Normalized Difference Vegetation Index, Normalized Difference Moisture Index, slope and proximity to human-built areas, as identified across several previous studies. The vulnerability scale was tested against a high accuracy burned area classification that was generated through band differencing Sentinel-2 derived NBR (kappa 0.905). In this validation there was a high level of agreement between burned and vulnerable areas in both the reference and the map at both locations. Therefore, a significant number of areas classified as vulnerable did burn in the resulting fire. There was low agreement between the not burned and not vulnerable areas in both the reference and the map at both locations; however, this could have been caused by fire control efforts in those areas. The results of this study will help improve future wildfire science towards forest fire prediction to increase disaster preparedness and decrease the damage of forest fires. Keywords: Forest fire; Remote sensing; Vulnerability; Unsupervised classification; Nova Scotia