Abstract
Fire severity mapping is conventionally accomplished through the interpretation of aerial photography or the analysis of moderate- to coarse-spatial-resolution pre- and post-fire satellite imagery. Although these methods are well established, there is a demand from both forest managers and fire scientists for higher-spatial-resolution fire severity maps. This study examines the utility of high-spatial-resolution post-fire imagery and digital aerial photogrammetric point clouds acquired from an unmanned aerial vehicle (UAV) to produce integrated fire severity–land cover maps. To accomplish this, a suite of spectral, structural and textural variables was extracted from the UAV-acquired data. Correlation-based feature selection was used to select subsets of variables to be included in random forest classifiers. These classifiers were then used to produce disturbance-based land cover maps at 5- and 1-m spatial resolutions. By analysing maps produced using different variables, the highest-performing spectral, structural and textural variables were identified. The maps were produced with high overall accuracies (5m, 89.5±1.4%; 1m, 85.4±1.5%), with the 1-m classification produced at slightly lower accuracies. This reduction was attributed to the inclusion of four additional classes, which increased the thematic detail enough to outweigh the differences in accuracy.
Cited by
11 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献