GeoAI-based augmentation of multi-source urban GIS

This presentation addresses a major challenge: fully leveraging the potential of geospatial data to improve Geographic Information Systems (GIS). Using urban flooding as a case study, it aims to integrate heterogeneous data sources of varying nature and quality levels in order to enhance both the expressiveness and reliability of GIS.

 

This work presents ongoing and planned research activities within the ATLAS CHIST-ERA project, which is entirely dedicated to this objective through a multidisciplinary approach. The project mobilizes complementary expertise in GIS, artificial intelligence, machine learning, computer vision and 2D/3D image analysis and object detection, statistics, urban network mapping, as well as geoalignment techniques.

 

The presentation is structured around two main objectives, both oriented toward GIS enrichment, with direct applications for flood risk management.

 

The first objective consists of combining and integrating external data within GIS. This approach enables seamless data integration and facilitates the revision, completion, and enrichment of existing datasets, while improving their expressiveness, particularly through the introduction of 3D representations. Such enriched representations are essential for accurately modeling surface runoff, flow paths, and hydraulic connectivity in urban environments subject to flooding.

 

The second objective focuses on integrating imperfect or uncertain data, such as amateur videos, crowdsourced observations, or data lacking precise georeferencing. To address these limitations, the project relies notably on the use of variational autoencoders for processing imprecise data, and proposes uncertainty and imprecision management mechanisms aimed at improving data quality by reducing inaccuracies and explicitly modeling confidence levels.

 

Acknowledgments :
This work was supported by the CHIST-ERA project ATLAS "GeoAI-based augmentation of multi-source urban GIS" under grant numbers CHIST-ERA-23-MultiGIS-02 and ANR-24-CHR4-0005 (French National Research Agency).