KIT-Bibliothek
FAQ Audio-/Videodatei publizieren

Project 04: Replacement of chemistry simulation by machine learning in an Earth system model (doctoral position, employment at KIT)

Autor

oliver kirner

Herausgeber

Martin Frank, Uwe Ehret, Angela Hühnerfuß

Beteiligtes Institut

KIT-Zentrum Mathematik in den Natur-, Ingenieur- und Wirtschaftswissenschaften (KIT-Zentrum MathSEE)
Institut für Wasser und Gewässerentwicklung (IWG)
Scientific Computing Center (SCC)

Genre

Sonstiges

Beschreibung

Project 04: Replacement of chemistry simulation by machine learning in an Earth system model (doctoral position, employment at KIT)
Project description

Earth system modeling plays a central role in climate change research. Earth system models (ESMs) describe processes such as the change in atmospheric dynamics and temperatures and the distribution of aerosols and chemical substances and their interaction with climate change. One of the most computationally intensive processes in ESMs is the calculation of chemistry.

The goal of this doctoral research is to show that the simulation of chemistry within an ESM can be replaced by an AI model. By combining domain knowledge from earth sciences with state-of-the-art ML techniques, the project seeks to develop novel hybrid ESM models that incorporate physical laws and KI driven approaches. Classical climate models use complex partial differential equations (as e.g. Navier-Stokes equations, continuity equations, radiative transfer equations, thermodynamic energy equations, or tracer transport and diffusion equations) to simulate the state of the atmosphere and ocean. More recent ESMs integrate additional processes to the classical climate models, as for example the simulation of chemistry, which is based on a stiff set of ODEs. We aim to replace these difficult and expensive (in terms of computing time) to solve equations by neural networks that conserve the physical constrains of the system. Thereby we aim to train all chemical tracers of the ESM with the help of an AI model and integrate afterwards a neural network into the ESM.

It will be investigated and evaluated if this enables a realistic simulation and thus improves the performance of the ESM.
The work involves an interdisciplinary approach and includes mathematics, atmospheric science and computer science. Corresponding contact persons are available.

The doctorate will be in Mathematics.

Tasks of the thesis include:

- Performance analysis of the Earth System model code at different resolutions on a High Performance Computing (HPC) system
- Creation of a concept for replacing the chemistry simulation with a suitable AI model (such as Transformer, LSTM, CNN) including the
identification of suitable features dimension reduction, and hyperparameter tuning
- Implementation and evaluation of the procedure and investigation of suitable metrics
- Evaluation of the AI model integrated into the Earth System Model (including parallelization)

Necessary requirements:

- Completed studies (master) in mathematics
- Knowledge of current deep learning frameworks (e.g., PyTorch or Tensorflow)
- Programming skills (e.g. Fortran, C++, Python)
- Ability to work and publish in a targeted and scientific manner.
- Good communication and presentation skills and willingness and ability to work a team
- Good writing and oral communication skills in English

Optional requirements:

- Experience in working with climate/earth system models

Laufzeit (hh:mm:ss)

00:09:19

Serie

KCDS Virtual Open House 2025

Publiziert am

27.02.2025

Fachgebiet

Mathematik

Lizenz

Creative Commons Namensnennung – Weitergabe unter gleichen Bedingungen 4.0 International

Auflösung 1280 x 720 Pixel
Seitenverhältnis 16:9
Audiobitrate 65322 bps
Audio Kanäle 1
Audio Codec aac
Audio Abtastrate 48000 Hz
Gesamtbitrate 643633 bps
Container mov,mp4,m4a,3gp,3g2,mj2
Dauer 558.600000 s
Dateiname DIVA-2025-51_mp4.mp4
Dateigröße 44.941.732 byte
Bildwiederholfrequenz 25
Videobitrate 572207 bps
Video Codec h264

Mediathek-URL

Embed-Code