Advances in active fire detection using a multi-temporal method for next-generation geostationary satellite data

Abstract

A vital component of fire detection from remote sensors is the accurate estimation of the background temperature of an area in fire's absence, assisting in identification and attribution of fire activity. New geostationary sensors increase the data available to describe background temperature in the temporal domain. Broad area methods to extract the expected diurnal cycle of a pixel using this temporally rich data have shown potential for use in fire detection. This paper describes an application of a method for priming diurnal temperature fitting of imagery from the Advanced Himawari Imager. The BAT method is used to provide training data for temperature fitting of target pixels, to which thresholds are applied to detect thermal anomalies in 4 μm imagery over part of Australia. Results show the method detects positive thermal anomalies with respect to the diurnal model in up to 99% of cases where fires are also detected by Low Earth Orbiting (LEO) satellite active fire products. In absence of LEO active fire detection, but where a burned area product recorded fire-induced change, this method also detected anomalous activity in up to 75% of cases. Potential improvements in detection time of up to 6 h over LEO products are also demonstrated.

Keywords

• fire detection
• diurnal variation
• geostationary sensors
• broad area training
• advanced himawari imager