Evaluation of image spatial resolution for machine learning mapping of wildland fire effects

Abstract

Wildfires burns 4–10 million acres across the United States with suppression costs approaching $2 billion, annually. High intensity wildfires contribute to post fire erosion, flooding and loss of timber resources. Accurate assessment of the effects of wildland fire on the environment is critical to improving the management of wildland fire as a tool for restoring ecosystem resilience. Sensor miniaturization and small unmanned aircraft systems (sUAS) offer a new paradigm, providing affordable, on-demand monitoring of wildland fire effects at a much finer spatial resolution than is possible with satellite or manned aircraft, providing finer detail at a much lower cost. This project examined the effect hyperspatial imagery acquired with a sUAS has on improving the extraction of post-fire effects knowledge from imagery. Support vector machines were shown to map post-fire effects land cover classes more accurately using hyperspatial color imagery than 30 m color imagery.