GEE and Machine Learning for mapping burnt areas from ESA’s Sentinel-2 demonstrated in a Greek setting

Abstract

<p>Climate change has resulted to an increase in the occurrence and frequency of natural disasters worldwide. An increased concern today is wildfire incidents, which constitute one of the greatest problems due to the ecological, economical and social impacts. Thus, it is very important to obtain accurately and robustly information on burned area cartography. Recent advances in the field of geoinformation have allowed the development of cloud- based platforms for EO data processing such as Google Earth Engine (GEE). The latter allows rapid processing of large amount of data in an efficient way, saving costs and time since there is also no need to locally download and process the EO datasets in specialized software packages committing also own computing resources. In the present study, a GEE-based approach that exploits machine learning (ML) techniques is developed with the purpose of automating the mapping of burnt areas from ESA’s Sentinel-2 imagery. To demonstrate the developed scheme, as a case study is used one of the largest wildfire events occurred in the summer of 2021 in the outskirts of Athens, Greece. A Sentinel-2 image, obtained from GEE immediately after the fire event, was combined with ML classifiers for the purpose of mapping the burnt area at the fire-affected site. Accuracy  assessment was conducted on the basis of both the error matrix approach and the Copernicus Rapid Mapping operational product specific to this fire event. All the geospatial analysis was conducted in a GIS environment. Our results evidenced the ability of the synergistic use of Sentinel-2 imagery with ML to map accurately and robustly the burnt area in the studied region. This information can provide valuable help towards prioritization of activities relevant to the rehabilitation of the fire-affected areas and post fire management activities. Last but not least, this study provides further evidence of the unique advantages of GEE towards a potential an automation of burnt area delineation over large scales.</p><p>KEYWORDS: GEE, Machine Learning, Sentinel-2, Burnt area mapping, Copernicus</p>