GeoAI-based AugmenTation of muLti-source urbAn GIS – ATLAS

Taking full advantage of the wealth of geospatial data available nowadays is a major scientific and technological challenge, with significant societal and economic impacts.
The multidisciplinary ATLAS project is set in this context and aims to augment the expressiveness and quality of geographic information systems (GIS) by integrating data from multiple sources of different nature and quality, with urban flooding as a common thread application. Meeting a complex challenge such as this requires a diverse range of expertise, including GIS, artificial intelligence and machine learning, image analysis, statistics, geographic alignment and many others. Bearing this in mind, this consortium was created to design innovative and practical solutions for collecting, organising, extracting, selecting, transforming, combining and integrating multi-source geospatial data, of various nature and quality. This will enrich and augment GIS in various ways.

This project aims to achieve four major scientific objectives, all focused on strengthening GIS, each represented by a separate work package.
The first objective encompasses the collection of data to complement the existing datasets already collected by the consortium, organise them and complete missing data if necessary. It also tackles the difficult task of selecting and geographically aligning GIS data with external sources having complex data structures.

The second objective of the project is to combine and integrate external data into GIS. This first calls to define a common language and representation format that will enable all available information to be commensurable and easily integrated. Next, the project explores different integration approaches, each serving a different purpose with a particular focus on i) integrating external data to revise, complete and expand existing GIS data and ii) enriching the expressiveness of GIS; in particular by integrating 3D data associated with the data.

The third objective of this project concerns the difficult problem of integrating external data tainted by imperfection or uncertainty. This concerns, for example, the integration of amateur videos or scanned maps where it is difficult (or even impossible) to identify geographic coordinates. We propose uncertainty management mechanisms adapted to our context in order enhance GIS data quality by reducing imprecision and uncertainty. We also plan to provide a representation based on spatial relationship graphs which would allow the integration process to continue even if the geographical positions are not precisely known. Finally, we address the issue of conflicting data that are inevitable when processing multi-source geospatial data.

The final scientific objective is to create sustainable open access datasets, design prototypes and proofs of concept integrating efficient algorithms for the different tasks described in the three first objectives of this project. We also aim to illustrate our generic methods, based on GIS augmentation, through a concrete application to urban flooding. We will use at least one case study from the city of Montpellier. Thanks to close collaborations between the members of this consortium and public and private stakeholders in water management, rich and representative data has been collected to illustrate and validate the project's objectives. It will continue to be collected throughout the project.