This project is a collaboration with the European Space Agency (ESA) which will investigate the implementation of an “AI-driven, global, near real-time, active fire detection method using remote sensing data”.  The project will aim to improve the accuracy and timeliness of wildfire detection using deep learning techniques explored within Andrianirina Rakotoharisoa’s PhD.

Research Activities

Over the 3-month collaboration period, the main research areas of the placement will be:

1. Designing and training a super-resolution deep learning model to downscale data from the Sea and Land Surface Temperature Radiometer (SLSTR) onboard Sentinel-3, using historical data to train the model.

1. Performing anomaly detection with a diffusion model on the downscaled data to identify active fire pixels and newly burnt areas, while also integrating auxiliary physical variables (e.g., vegetation, wind) to ensure that the model’s outputs are physically consistent.

Possible Results

We anticipate the following outcomes:

1. The release of a publicly available dataset derived from downscaled SLSTR data to support future research in active fire detection.
2. A physics-informed deep learning anomaly detection model for near real-time active fire detection with global applicability.
3. A journal paper – draft title: “Towards AI-driven global near real-time active fires detection”.
4. Groundwork for potential future works at ESA on wildfire monitoring.

Feature Image: Contains modified Copernicus Sentinel data (2023), processed by ESA