Mapping Savanna Wildfires in Southern Belize using Sentinel-1 SAR and Object Based Imagery Analysis.
A new approach for extracting spatially explicit estimates of burned areas is developed, using a time series of pre-processed Sentinel-1 SAR imagery and object based image analysis. By using scientific analysis-ready and easily downloadable Sentinel 1 data from both ascending and descending orbital paths, we are able to increase the observational frequency of wildfires to several scenes per month, enabling more precise detection of fire events and mapping their progression from week to week during a fire season. By using object based segmentation and fuzzy logic threshold classification of temporal backscatter coefficient indices, we are able to improve the mapping of the overall extents of burns, and improve on the timings of when burns occur. We apply the method to Sentinel-1 data spanning two fire seasons in 2018 and 2019 in lowland pine savannas of Belize, Central America, to produce new enhanced information products for land and fire managers. The new mapping contains more detailed and precise dates and extents for wildfires than has previously been possible to produce using passive multispectral platforms. We discuss the scope and challenges of extending the method for further years, and to other savanna areas, and present some practical advice from our experience as to the more influential channels and temporal indices for detection of burnt area in this landscape.