Starting point, Contents and goals
Until now, all large-scale land use and land cover (LULC) products are generated by human operators in massive manual labelling processes.
AI4LULC will therefore strive to contribute to innovative and automated LULC analysis methods, especially for urban areas, by combining the latest advancements of Earth Observation with the power and scalability of state-of-the-art Artificial Intelligence for semantic geospatial analysis.
The project will build on effective methods for machine and deep learning for LULC mapping and focus on meaningful indicators for cities and their surroundings. Three use cases on three geographic scales have been identified that will showcase the high potential and effectiveness for AI-based automated feature extraction for urban LULC mapping tasks mainly using Copernicus data:
- Global Scale: mapping of megacities outside Europe based on Sentinel-1 and 2 data.
- Pan-European Scale: mapping of European cities to support Copernicus HR Urban Atlas by means of Copernicus VHR Image Mosaic (2.,5 m resolution) and Sentinel-1 and 2 data for selected cities in Europe.
- National Scale (Austria): Investigate to what extent the AI-based methods are suitable for automated mapping of urban LULC on national scale based on VHR data (aerial images, VHR satellite data, DSM) in combination with Sentinel-1 and 2 data.
To avoid manual labelling for AI-based training, all reference information is gathered from existing data stored as vector information and automatically optimally discretized to perfectly assist the training step.
Blackshark.ai (a deep-tech SME) and Joanneum Research (a leading EO research organisation) planned the project in an agile manner that also reflects the nature of the applied scientific methods. Three iterative loops of AI-training and validation of their results are scheduled to enable the project team to reach optimal results for the three use cases.
The project’s outcomes will translate into scientific innovation and economic exploitation. The results will foremost boost core products of the Copernicus initiative by enabling the AI-based automation of laborious LULC mapping and Urban Atlas classification tasks.
Furthermore, the project will contribute to provide geospatial information for future Digital Twin Earth applications and other geospatial services that process semantic inputs in large quantities.
Ms. Andishe Zhand
Am Eisernen Tor 1/3