Leveraging AI and satellite imagery to delineate floods and support the calibration of hydraulic models in extreme conditions
Summary of the presentation
The delineation of flooded areas relies largely on field observations collected during floods, but these observations become rare once conditions become extreme. Satellite imagery offers a valuable alternative, as it provides reliable information on the extent of flooding and can support the calibration of hydraulic models when major floods occur.
This presentation will describe the development of a flood mapping approach based on radar and optical satellite imagery, deep learning that is transferable to the Quebec context, and a hydraulic regulation method applied to the target inference region (Quebec).
It will briefly cover the method, and will mainly discuss the results obtained, the potential of the approach, and the challenges of generalization with AI.
The project was carried out as part of a postdoctoral fellowship funded following a joint call for proposals by Ouranos and the RIISQ.
Learning objectives:
• Become familiar with the basic principles of the deep learning approach and how it can be used to leverage satellite imagery to delineate flooded areas
• Understand the complementary role of remote sensing in the calibration of hydraulic models in Quebec
• Understand the challenges associated with the generalization of AI in hydrology
Register for the webinar to get the ZOOM link.
* Presentation in french
Speaker(s)
Claudie Ratté-Fortin is a research associate at INRS and co-founder of Clean Nature. She specializes in artificial intelligence applied to hydrology and hydrological modeling. She is interested in applying the principles of responsible AI in the design and deployment of systems where the integrity of results is critical.
