You are here: Home / Communication / News archive / Machine Learning for Earth system sciences – the Helmholtz AI Support team AIM at DKRZ

Machine Learning for Earth system sciences – the Helmholtz AI Support team AIM at DKRZ

Nov 24, 2020

The AIM support team at DKRZ is funded by the Helmholtz Association and applies methods of Machine Learning (ML) and Artificial Intelligence (AI) in the field of climate and earth system research. As part of Helmholtz AI, a comprehensive application-oriented initiative, the AIM team supports scientists from various Helmholtz research centers in implementing and evaluating ML/AI technologies in practice.

More specifically, scientists from the Helmholtz centers active in the research field “Earth and Environment” can approach the AIM team with support requests. These formal user requests are collected, assessed and usually processed as part of a support project via a ticketing system. The complexity and the according effort required to address such projects can vary substantially: ranging from providing answers to simple technical and methodological questions all the way to planning, implementing and executing entire complex ML workflows.

One example for such a support project is a request by GFZ, where the goal is to leverage artificial neural networks (ANNs) to determine wind speeds over oceans based on measurements of signals from global navigation satellite systems that are reflected at sea surfaces. In this case, the challenges to overcome with ML methods stem from the substantial fraction of noise in the raw data and manifold statistical biases in them.

Another example: KIT requested support to determine how far an ML-based approach can emulate the predictions of an atmospheric chemistry simulation model in order to potentially perform future simulations with less computing effort. To address these questions, the AIM team employed methods to process the typically bulky input data at scale, used established software libraries such as pytorch and nni and developed and optimized the required ML models. This also included activities to parallelize the data processing and model optimization workflows, so that the large data volumes could be efficiently processed by multiple GPU computing nodes.

These support services are provided by the DKRZ free of charge for users from Helmholtz centers due to the funding by Helmholtz AI, but DKRZ also offers general support and advice for users from other institutions.

If you are interested, please feel free to contact us via [Email protection active, please enable JavaScript.].

Further information about the AIM support team:

Support requests to Helmholtz AI can be directly submitted via the voucher system:

Document Actions