Automated training sample definition for seasonal burned area mapping

摘要: Abstract Monitoring of environmental change can benefit from the increasing availability multitemporal satellite imagery, and efficient effective analysis tools are needed to generate relevant spatio-temporal land cover datasets. We present a data driven approach for automatic training sample selection support supervised mapping seasonally burned areas in semi-arid savannas Southern Africa. Our leveraged distinctive spectral-temporal trajectories associated with on landscape at different times or remaining unburned over time. Using fuzzy c-means clustering, we extracted mid-infrared burn index (MIRBI) derived Landsat characterized them based empirically developed labeling rules. The selected captured both condition (burned unburned) if burned, timeframe event. assessed by Random Forests model using 2500 automatically validated against ground truth years 2009 2014. Based 1000 validation points each year, obtained overall accuracies above 90% showing reliable consistent were supplied our approach. method provides which reduce time-consuming expensive task, enabling quicker generation area information that fire monitoring programs climate research.