Anomaly detection based method for spatio-temporal dynamics mapping in dam mining regions

Abstract

Remote Sensing technologies and Machine Learning methods rise as a potential combination to assemble new environmental monitoring applications. In this context, the presented work proposes a new method that exploits anomaly detection models applied to Remote Sensing imagery to identify the spatio-temporal changes over the Earth's surface. The potential of the introduced approach is shown in a study case concerning the analysis of the landscape changes using One-Class SVM and Isolation Forest methods in Landsat and Sentinel images for Brumadinho and Mariana regions, Brazil, after its re¬ cent dam collapses.