04/2022 – 04/2025


Detecting landslides and illegal waste dumps showcase a chance for Earth observation missions. This can be done most effectively with Artificial Intelligence methods that are trained to detect spatio-temporal surface anomalies – a task that also brings challenges (see the figure).

(1) A large-scale application is possible given the full archives of global EO data with numerous revisits of areas of interest. Yet, the labeled portion of these data, needed to train AI methods, is sparse.

And (2), existing models were trained in specific environments and need to be transferred and generalized, i.e., re-trained, in order to work on larger spatial scales.

(3) The described use cases require the combined use of multiple data sources, i.e., different sensors. We therefore need to extend the capabilities of AI models accordingly. As part of this project, we will tackle these challenges together with our colleagues from GFZ.