Published works

Recent increases in the frequency and intensity of active fire has heightened the importance of remote sensing as a source of early warning information for fire incidents. The launches of new geostationary sensors, such as Himawari-8 over the Asia-Pacific, have vastly increased the information available with which to detect and attribute these incidents, with observations every 10 minutes possible over key parts of the electromagnetic spectrum (3.8 -- 4µm). Remotely sensed fire products such as Sentinel Hotspots and the MODIS active fire product have focussed upon use of contextually derived background temperatures for isolating hotspots, dictated by the low temporal frequency of available images. This research proposes a new paradigm in fire detection, which utilises the increased temporal resolutions of geostationary sensor imagery to provide a baseline dataset for land surface temperature estimation based upon location and time of day. To achieve this, a multi-temporal diurnal characterisation of temperature is calculated for each pixel based upon a large area latitudinal transect. Hot spot anomalies are then identified based upon the deviation of the location’s temperature from the expected diurnal cycle. Validation of the fire detection algorithm has focussed upon case study fires from the 2016/17 fire season, by way of inter-comparison with commonly used MODIS and VIIRS active fire products, and a burned area product. Results show increased capability for early fire detection using the new algorithm in comparison to traditional single image contextual algorithms employed for polar orbiting systems. Other advantages include notable resilience to sources of occlusion such as cloud and smoke. Further research will focus on the wider application of this method across the Australian continent and methods for countering more challenging detection conditions.