Near real-time radar imaging using AI (Artificial Intelligence) to publish lake ice phenology

The recent surge in the availability of space data sources has massively accelerated the growth of the downstream space sector. However, most methods developed to analyze the available space data are outdated and platform dependent. Indeed, most space data products leverage outdated threshold-based algorithms (e.g. NASA’s FIRMS) and the methods used are typically highly tailored to the platform’s specifications (i.e. altitude, spectrum, resolution, etc).

The proposed project is a platform-agnostic wildfire management system developed by Lux Aerobot, using requirements from SOPFEU and NRCan to accomplish near-real-time processing to provide a high-resolution thermal mapping of active wildfires within less than 15 minutes of the image being acquired. Using pseudo-datasets of images generated from an INO-led satellite wildfire image simulator, Lux Aerobot will be able to train platform-agnostic deep learning computer vision models designed specifically to extract hotspots on a pixel level. To address potential telecommunications bandwidth limitation, the model will also be further adapted to be able to operate directly on-board satellite platforms, subject to the various constraints of space such as finite power supply, thermal energy dissipation and limited computing power. The system will also include a wildfire simulation sub-system to generate near-real-time probabilistic predictions of the wildfire’s growth. 

The project will showcase cutting-edge Canadian innovation in space-derived products to the benefit of Canadian society. It is particularly aligned with Canada's ambition to contribute to fighting climate change on a global level.