Author(s): Shrijana Poudel, University of Leicester (First Author, Presenting Author) Robert Parker, University of Leicester, National Centre for Earth Observation Heiko Balzter, Centre for Landscape and Climate Research (CLCR), National Centre for Earth Observation (NCEO) Tristan Quaife, University of Reading Douglas Kelley, UK Centre for Ecology and Hydrology
Tropical forests are vital carbon sink but are increasingly threatened by unusual fire events. Using a machine learning method called Isolation Forest, we detected these atypical fires such as those occurring in unexpected times or places by testing both synthetic and real satellite data. Including seasonal information greatly improved detection accuracy, even for small anomalies. Our approach successfully identified overlooked fire disturbances in tropical regions, supporting forest monitoring and conservation.
