A SEMI-AUTOMATIC METHOD FOR DETECTING BURNED AREAS AND DATING HISTORICAL FIRES USING LANDSAT DATA Ana Carvalho
Edith Cowan University
Abstract
Accurate historical fire records are crucial for understanding fire impacts and guiding forest management However many fire history databases suffer from incomplete or inaccurate fire boundaries and dates limiting their usefulness Landsat with its 30 meter spatial resolution can detect fires as small as 009 hectares and has the potential to fill data gaps and improve historical fire records However its use has been limited by labor intensive processing especially when fire dates are unknown requiring advanced thresholding and machine learning classification methods. This study presents a methodology using Landsat imagery to detect burned and unburned areas within registered fire perimeters and estimate accurate fire start and end dates from 1990 to 2021 The fire history database FHD shows signs of stabilization in the number and size of polygons and satellite derived burned map areas However data agreement between the FHD and LST mapped burned areas decreases with polygon size. The LST method identified dates for 308 of 340 fires increasing recorded dates by 44 percent MEF values for start and end dates were 098 of agreement with the FHD with a statistically significant post 2000 improvement The MDS method identified dates for 68 percent of 253 fires while LST identified 87 percent Post 2000 both methods perfectly aligned with FHD records MDS struggled to detect fires smaller than 150 hectares whereas LST successfully captured both small and large fires. LST showed superior performance in enhancing fire records especially for detecting small fires and multiple date ranges Automating burned area detection and fire date estimation using Landsat and API integration significantly improves analysis efficiency and accuracy
Keywords: Historical fire records, Landsat imagery, Burned area mapping, Fire date estimation, Machine learning classification
