Pixel-based and object-based change detection methods for assessing fuel break maintenance

Abstract

This last decade, large wildfires have increased in number, size and consequent damages in various countries worldwide. Since 2017, the large fire hazard is a major concern for Portugal. An important fuel break (FB) network is currently implemented in strategic areas by the Portuguese Institute of Nature and Forest Conservation (ICNF). The objective of reducing fuel loads on those thin strips is to reduce fire propagation and to improve firefighting conditions. The efficiency of FB depends on its periodic maintenance. The increasing quality and frequency of Earth Observation Satellite imagery nowadays allow the implementation of change detection methods to identify the occurrence of FB maintenance operations and help their necessary management. This article presents two approaches, a pixel-based and an object-based semi-automated supervised classification using monthly composites of Sentinel-2 imagery to achieve this detection. The pixel-based approach resource to the Maximum Entropy classifier while the object-based to an Artificial Neural Network. The overall accuracies range from 96.5% to 97.5%, which are promising results. Both methods can be combined for optimal detection over the whole territory.