Fire Detection with Multitemporal Multispectral Data and a Probabilistic Unsupervised Technique

The frequency of forest fires has increased signifi- cantly in recent years across the planet. Events of this nature motivate the development of automated methodologies aimed at mapping areas affected by fire. In this context, we propose a method capable of accurately mapping areas affected by fire using time series of remotely sensed multispectral images by statistical modeling and classification. In order to evaluate the introduced proposal, we carry out a case study on a region in Brazil with recurrent history of forest fires. Furthermore, images obtained by the Landsat-8 satellite are used in this case study. Comparisons with an alternative method are included in this analysis.