Field evaluation of two image-based wildland fire detection systems

Rapid detection of wildfire outbreaks is a critical component of fire management because suppression activities are most effective when fires are small. One method of fire detection and location is computer analysis of images from sensors mounted on towers. In this paper we report on a trial of two image-based detection systems under operational conditions in forests and pasture in south-eastern Australia. The systems were deployed for 3 months in autumn, 2010, during which time a total of 12 experimental fires, 31 planned fires lit by public land management agencies, approximately 250 planned fires lit by private individuals, and 1 unplanned fire were recorded. Both image-based systems were able to detect and locate fires. They performed well for larger planned fires but poorly for small fires (<1ha area) at moderate distances (10–20 km). System performance was compared to a human observer for a subset of fires. For these fires the human observer had a higher detection rate and shorter reporting time than the image-based systems. All methods of detection had a similar level of error for locating fires in the landscape once they had been detected. Operator skill was an important factor in the performance of all systems. Image-based fire detection could be a useful complement to other detection methods already in use in Australia.

► Two image-based wildfire detection systems were tested in the field.
► Detection performance was compared with a human observer.
► Both systems could detect fires during the day and night-time.
► The human observer outperformed the image-based systems.
► Image-based detection could complement existing fire detection methods.