Object-Based Validation of a Sentinel-2 Burned Area Product Using Ground-Based Burn Polygons

An object-based validation of a country (Italy) level burned area product is proposed, to assess not only the thematic accuracy, but also the geometric accuracy. The product was derived from Sentinel-2 data by using an algorithm, already published in the literature, which applies image segmentation. Ground-based burn polygons, representing the official burned area data for Italy, were used, as reference, to assess the product. The validation was performed by implementing a set of accuracy metrics to quantify the similarity between reference objects and mapped objects, the object-based thematic accuracy, and the global quality of the segmentation. The challenges in performing the object-based validation, mainly related to the absence of a one-to-one relationship between a reference object and a mapped one, are discussed and the strengths and weaknesses of the considered burned area product are highlighted. Results showed that the product was characterized by a good geometric accuracy even if the number of false positives represented a critical point. Although a specific product was used, the proposed validation method can be in principle applied to any burned area data derived from satellite images. To point out this characteristic, an experiment on another product was performed too.