Wildland fuel type mapping in Attica using Sentinel-2 time-series

Wildland fuels are a key component in wildfire behavior modelling and wildfire prevention actions and their accurate mapping is becoming more critical, on account of the increased pressure emanating from human activities and the effects of climate change. The work presented in this manuscript aims at mapping fuels at a regional scale using a newly proposed fuel type classification system. The region of Attica in Greece was selected as the study site with an area of approximately 2,900 km
²
. Two fuel type maps were generated for the years 2020 and 2022, each based on a timeseries of six Sentinel-2 images with minimum cloud cover, selected in two-month intervals. The methodology proposed separates the study area into 3 land cover types, namely agricultural, urban and vegetation, and performs a classification of the area into the respective fuel types employing machine learning algorithms along with information derived from official thematic maps. The final fuel map is generated with the integration of vegetation height and density information derived from official products. To assess the accuracy of the produced map for the year 2020 the Land Use and Coverage Area frame Survey (LUCAS 2018) dataset was employed. The results display a promising accuracy of 72%, considering the first level of the hierarchical classification scheme. Our work demonstrates that accurate and adaptable maps of wildland fuels can be generated for Mediterranean complex ecosystems using a novel classification approach. Updated fuel type maps will be generated in the future, with enhanced information regarding height information in grasslands and shrublands and understory structure by means of expert knowledge or field observations.