A Spatio-temporal Data Mining Approach to Global scale Burned Area Monitoring

Abstract:

We present a novel technique for burned area mapping in forests using the Enhanced Vegetation Index (EVI) from the MODIS 16-day Level 3 1km Vegetation Indices (MOD13A2) and the Active Fire (AF) from the MODIS 8-day Level 3 1km Thermal Anomalies and Fire products (MOD14A2). The proposed method leverages the spatial and temporal co-occurrence of thermal anomalies and vegetation loss caused due to forest fires to detect burned areas. Our approach derives features from Enhanced Vegetation Index that target locations which show an abrupt change in their vegetation time series that take at least several months to recover. One unique aspect of our approach is that it uses data from multiple months around the fire event and is therefore more robust to issues in data quality. Comparison with other burned area products show that our approach detects several large previously undetected burned areas across multiple geographical regions. In particular, we found that our approach detects several large burned regions in the tropical forests of Indonesia and South America that had been missed by the state-of-arts burned area approaches. For example, using our approach in Indonesia we discovered that the state-of-the-art MODIS Burned area product had missed around 20,000 sq. km. of burned area (nearly as much burned area as it has reported). We show that all these previously unreported burned areas detected by our approach are actually significant fires which suffered a large, abrupt loss in their vegetation at the time of the fire event and take at least several months to recover back to their normal vegetation. To evaluate these burned areas we compared the Landsat-based composites before and after the date of the event. Our Landsat analysis shows that the burned areas detected by the proposed approach are true burns with a very small error of commission. We believe our work has the potential to provide a scalable approach to global forest monitoring as well as reduce the uncertainty in quantifying the carbon emissions from forests due to fire activity.