MapInWild: Map​ping and ​ In​terpreting ​ Wild​erness from Space

• Applicants: Professorin Dr.-Ing. Ribana Roscher; Professor Dr.-Ing. Michael Schmitt
• Subject Area: Geodesy, Photogrammetry, Remote Sensing, Geoinformatics, Cartography
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 458156377

Against the background of socially highly relevant topics such as the designation of nature reserves or the prevention of pandemics through nature conservation, a combination of AI- and remote sensing based methods is a promising approach to objectively and efficiently quantify the naturalness of landscape features and to map wilderness areas. In the project MapInWild we will develop deep learning methods for mapping wilderness areas using satellite image data. Important methodological components in the project are contributions to weakly supervised machine learning, in order to be able to generate robust and supra-regionally applicable models even with comparatively poor training data, and contributions to explainable machine learning. By developing and applying complementary tools, the transparency and interpretability of deep neural networks is increased in various ways. Thus, information is given about why the learned wilderness mapping models lead to specific decisions. In combination with known prior knowledge about wilderness, the obtained results are explained in the context of the application and thus lead to an improved definition of the term wilderness. The wilderness mapping and the investigations on explainability are closely related and are continuously improved by feedback loops. Thus, the project contributes to the methodical development of transferable procedures of machine learning with scarce and error-prone training data and their interpretability. Furthermore, progress is expected in the field of automation of remote sensing.

• DFG Programme: Research Grants