Remote Sensing Applications for Mapping Large Wildfires Based on Machine Learning and Time Series in Northwestern Portugal

Mapping large wildfires (LW) is essential for environmental applications and enhances the understanding of dynamics affected areas. Remote sensing techniques supported by machine learning time series have been increasingly used in studies addressing this issue shown potential type analysis. The main aim article to develop a methodology mapping LW northwestern Portugal using algorithm from Landsat images. For burnt area classification, we initially Fourier harmonic model define outliers that represented pixels possible areas and, then, applied random forest classifier classification. results indicate analysis provided estimates with actual observed values NBR index; thus, classified were only those masked, collaborated processing, reduced spectral confusion between targets similar behaviour. maps revealed ~23.5% territory was at least once 2001 2020. temporal variability indicated that, on average, 6.504 hectares within 20 years. annual varied over years, minimum detected 2014 (679.5 hectares) maximum mapped 2005 (73,025.1 hectares). We concluded process defining mask considerably universe be each image, which leaves training focused separating set into two groups very characteristics, thus contributing so separation behaviour performed automatically without great sampling effort. method showed satisfactory accuracy little omission