RESIKOAST

The RESIKOAST project develops strategies for long-term adaptation to changing climate conditions and tools for early detection of emerging risks. The project involves ten DLR institutes working with model regions in coastal areas and with national agencies and institutions in the fields of infrastructure, climate protection and weather forecasting.

The Institute of Software Technology develops and applies AI methods in RESIKOAST to enable early detection of risks in coastal areas of the North Sea and Baltic Sea. For this purpose, large earth observation data sets are analysed on mainframes. The challenges we face are the size of the data sets to be considered and the risk assessment of the anomalies detected. There are many natural changes in the coastal area, but they do not pose a threat to the local population.

We use two different approaches to detect anomalies. One is classic density-based anomaly detection techniques, and the other is modern deep neural networks based on autoencoders or vision transformers. The combined use of different methods can improve the robustness and reliability of the prediction and avoid false positives. The highly parallel Helmholtz Analytics Toolkit (Heat), co-developed by the institute, is used to process the huge amounts of data quickly and efficiently.