A MACHINE LEARNING SOLUTION FOR OPERATIONAL REMOTE SENSING OF ACTIVE WILDFIRES

Quantitative wildfire behavior data is invaluable for model development and real-time situational awareness during a fire emergency. However, there exists at present no scalable solution to acquire consistent information about active forest fires that is both spatially and temporally explicit. Spaceborne sensors must meet exigent tradeoffs between spatial and temporal resolution, and there is no single platform that allows detailed measurement of fire behaviour from space. To overcome this limitation, we developed a machine learning solution designed to leverage the complementary features of various remote sensing platforms. Our system relies on a machine learning algorithm to statistically downscale Geostationary (GEO) satellite imagery and continuously monitor active fire location with the spatial resolution typical of Low Earth Orbit (LEO) sensors. In order to achieve this, the model trains on LEO imagery, land use information, vegetation properties, and terrain data. This paper describes the system architecture and demonstrates its performance in two case studies. Results presented here prove the viability of the proposed strategy and encourage further development.