Determining post-fire information is crucial for post-firemanagement activities and rehabilitation treatments. Detailed and currentinformation concerning the location and extent of the burned areas isrequired to assess economic losses and ecological impacts. Satellite datahave been used extensively for many years for the detection and mappingof fire-affected areas and represent nowadays a standard and reliablesource of information. Moreover, satellite data become more and moreeasily available and one can download very high (spatial) resolution data,such as Sentinel-2 and Landsat 8, with no charge. Several robust andadvanced approaches can be found in the literature to determine fireseverity and thoroughly analyze post-fire rehabilitation period. However,the methods used for information extraction and thematic mapping shouldbe reconsidered in order to process this very high spatial resolution data.Indeed, the spatial resolution of satellite data together with landscapeconfiguration guide the decision for the classification approach to beapplied, according to the resolution of the model of the scene. Object-basedimage analysis (OBIA) is a powerful approach that has been successfullyapplied in many research problems in remote sensing. The strength ofobject-based analysis lies on the fact that individual pixels, composing realworld objects, can be identified in the satellite imagery and classified usingsemantic and heuristic knowledge. This study was carried out after a largewildfire occurred on the Mount Somma and Vesuvio Volcano near Naples(Southern Italy) on July 2017. In this paper, we discuss the strengths andthe weaknesses of the object-based classification approach and compare it with classical pixel-based for burned area mapping. Furthermore, wecompare the performances of widely-used burned area related spectralindices in identifying burned, slightly burned, water and non-burned areasfrom each other, and to delineate the boundaries of burned area. In thiscontext, spectral indices of Normalized Burn Ratio (NBR), NormalizedVegetation Index (NDVI), Burned Area Index (BAI) derived from thesatellite image were employed. Pre- and post-fire Sentinel-2 and Landsat 8images were used to identify the extent of forest fire within the region. Theimages were orthorectified and classified using pixel and object-basedclassification algorithms to accurately delineate the boundaries of burnedareas. In order to aggregate pixels into objects, we applied amultiresolution segmentation algorithm called Edge Mark and Fill (EMF),which carries out a marker-controlled watershed segmentation andproduces a segmentation map with a good trade-off between accuracy andnumber of segments. Then a fuzzy membership function classifier wereapplied to the combinations of the selected indices (NDVI, BAI-NBR,NDVI-NBR) to discriminate burned, slightly-burned and non-burned areasfrom each other. Results showed that all combinations constructed in thisstudy produced satisfactory results in terms of classification accuracy. Theresults of this work have been made available for the government agenciesto delineate fire perimeter and determine risk of wildfire for post-firedamage management.
|Titolo:||Post-fire assessment of burned areas with very high resolution Sentinel-2 and Landsat-8 images|
|Data di pubblicazione:||2018|
|Appare nelle tipologie:||4.1 Contributo in Atti di convegno|