Oceanographic Data Services as a Foundation for Machine-Learning Applications and Digital Twins
Please login to view abstract download link
Contemporary digital twins for oceanic and atmospheric systems, alongside advanced machine learning methods, fundamentally depend on high-quality data produced by established model systems. As an example of operational model integration within digital twins, this work describes the linkage between the European Digital Twin Ocean (EDITO) and the Copernicus Marine Services and presents one underlying coupled ocean-ice-biogeochemical model system employing data assimilation. The modeling system is built upon the open-source frameworks NEMO [1], ERGOM [2] and PDAF [3] and the applied Ensemble Optimal Interpolation (EnOI) data assimilation method will be presented in detail using the example of biogeochemical profile data assimilation. Additionally, this work presents a second multi-fidelity model system: an atmosphere-ocean-ice coupled climate model used for regional climate projections within the EURO-CORDEX framework and applied in German national core climate services DAS-Basisdienst. The model system developed for the North and Baltic Sea, cf. [4], consists of the atmospheric model ICON- CLM [5] coupled to the ocean core model NEMO using a flux-based OASIS coupling approach [6] and can be used as training data for machine-learning applications of ocean-atmosphere coupling dynamics. Finally, initial developments of reduced-order emulators derived from these operational models are examined, with emphasis on computing performance and predictive skill.
