High-resolution Coastal Change Detection using AI and HPC
Please login to view abstract download link
Rapid climate change is leading to an increase in extreme weather events in coastal regions, posing significant risks to infrastructure, ecosystems and local communities. This work focuses on detecting changes in high-resolution spaceborne radar data, acquired every 6 days, along the entire North Sea coast of Germany between 2016 and 2023. For large-scale analysis, we apply an anomaly detection algorithm pixel by pixel, i.e. millions of times, to the entire coastline. The computational complexity of pixel-by-pixel anomaly detection over such a large spatio-temporal dataset is immense, requiring significant computing power and memory. To address this problem, we use Heat ("Helmholtz Analytics Toolkit") [1], an HPC-optimised framework for large-scale machine learning and array computing developed jointly by DLR, Jülich Research Centre (FZJ) and Karlsruhe Institute of Technology (KIT). Our implementation was run on the terrabyte HPC cluster operated by DLR and the Leibniz Supercomputing Centre (LRZ) [2]. The results, which will be presented in the talk, clearly highlight significant weather events, such as the storm Malik on January 29-30, 2022, and identify vulnerable coastal regions along the entire German North Sea coast. [1] F. Hoppe, J. Muriedas, M. Tarnawa, P. Knechtges, B. Hagemeier, K. Krajsek, A. Rüttgers, M. Götz and C. Comito. Engineering a large-scale data analytics and array computing library for research: Heat. Accepted for publication in the post-proceedings of the deRSE conference in ECEASST, 2024. [2] F. Hoppe, W. Koslow, K. Rack and A. Rüttgers. Exploring and processing large data sets in earth observation on HPC-systems with Heat. 75th International Astronautical Congress, Milan, Italy, 14-18 October 2024.
