Climate-HPC-MLClimate research is characterized by the investigation of manifold characteristics of our climate system, which is already supported by a large number of computer-aided analyses in numerics and statistics at DKRZ. Through ML technologies, both new and existing research approaches can now be investigated with appropriate hardware at DKRZ. DKRZ already offers research and support in the area and will further explore the topic with appropriate software and hardware.



ML Research Group

ml-reconstructing-climateModern ML methods for climate research on the DKRZ high-performance systems will be researched, transferred, further developed and made available to the climate community. One focus is on the fusion with Earth system modeling, for example to improve climate predictions, but also to reconstruct missing climate records.

With the research group "Climate Informatics and Technologies", a special interface between climate and AI/ML research will be established at DKRZ.

  • Interface between ML and climate science
  • ML for the DKRZ high-performance computing infrastructure
  • Knowledge transfer and methodological research for the climate community
  • Provisioning of latest ML technologies for climate scientists


Helmholtz AI support team AIM

nn-example-architectureThe Helmholtz AI Support Team AIM at DKRZ, funded by the Helmholtz Association, works on and with Machine Learning (ML) and Artificial Intelligence (AI) methods in the field of climate and environmental research. As part of Helmholtz AI, a broad application-driven initiative, the AIM team supports researchers from different centers of the Helmholtz Association in the introduction, evaluation and practical use of ML/AI technologies.

  • Support for implementation of ML methods for use cases in the thematic area Earth & Environment
  • Performance optimization and assistance in the use of GPU computing nodes by ML procedures in Python
  • Optimization of data to be used for training ML applications
  • Guidance on methods and requirements of ML procedures

GPU server for ML applications

DKRZ's GPU partition is particularly well suited for computing ML algorithms with NVidia GPUs and large main memory. The direct connection to the parallel file system also saves time-consuming file transfers. By using Jupiter notebooks, the GPU partition can be used in a user-friendly and interactive manner.

Further information pages

Information concerning the practical use of ML is available in the user portal.

Points of contact

In case of questions, please also do not hesitate to contact the DKRZ user support.