A Toolkit and Framework for Continuous Large-Scale Quality Measurement of Open Geodata with Efficient AI and Multimodal Remote Sensing (AI4OpenGeodataQuality)

• Applicant: Professor Dr. Martin Werner
• Subject Area: Geodesy, Photogrammetry, Remote Sensing, Geoinformatics, Cartography
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 563303373

This project aims at developing quality engineering technology for volunteered geographic data including the definition of quality measures, their semi-automated extraction with AI, and their dynamic through time under both live contribution and historical VGI databases. This project can have a comparably large impact, as the valuable data source of volunteered geographic information cannot be used for tasks where data integrity is crucial at the moment due to the lack of consistent quality attribution. The overall goal of AI4OpenGeoData project is to acquire reliable and up-to-date insights into the open geodata quality – especially for OpenStreetMap and GDI-DE – through the combination of efficient AI models and multimodal Remote Sensing data, which helps to establish continuous and accurate quality awareness at scale. The major innovations of the project are: 1) the integration of the temporal dimension into geospatial data quality measurement, 2) the usage of multimodal RS data to enable transferable and scalable data assessment, 3) the development of quality-aware GeoAI models to automatically derive and use quality indicators, which will advance and complement the state-of-the-art intrinsic and extrinsic data quality assessment approaches. With respect to integrating the temporal dimension, we develop a spatiotemporal data container providing a very high flexibility in spatial and temporal aggregation supporting complex and computationally expensive quality indicators. This approach differs from data cubes, because neither a spatial nor temporal aggregation scale has to be known beforehand, but can be part of the quality metric itself. In addition, we use remote sensing data from multiple sources to reliably correlate unbiased observations with the changing VGI data and develop novel techniques to directly relate observation, geodatabase, and quality indication. Finally, with respect to GeoAI, we want to ensure that the quality indicators are available to artificially intelligent systems in order to improve training and inference through quality awareness. Beyond our main interest in cross-modal quality assessment, we want to contribute to a wider adoption of geospatial technology through supporting the FAIR principles making all our data and source code findable, accessible, interoperable and re-usable.

• DFG Programme: Research Grants
• Co-Investigator: Dr. Hao Li