Introduction to Deep Learning for Climate Scientists

22 Sept 2026, 10:30 → 24 Sept 2026, 21:30 Europe/Berlin

Room 034 (DKRZ Main Building)

Alexander Fischer (DKRZ), Caroline Arnold (HEREON), Etienne Plésiat (DKRZ), Johannes Meuer (DKRZ), Maximilian Witte, Paul Keil (HEREON)

Target group: beginner in Deep Learning with some Python experience

Deadline for registration: July 24 18:00 CET

Machine learning, and more particularly deep learning, is gaining popularity among climate scientists. While many tutorials and courses are available, researchers often face challenges in applying the concepts to climate data and adapting them to address the specific needs of their research.

Therefore, we are offering an Introduction to Deep Learning for Climate Scientists at DKRZ. The course will be held in person, and the number of participants is limited to 25. 

The course will be held from September 22 10:30 to September 24 12:30 at DKRZ seminar room 034. Coffee breaks will be provided during the workshop.

Participants are required to have a working knowledge of Python. Some experience with Machine Learning or Deep Learning is welcome but not necessary.

 

Tuesday 22 September

10:30 → 17:30 Day 1

10:30 Introduction to Machine Learning | Lecture 🎓

• A comprehensive overview of concepts & principles behind Machine Learning
• Exploration of real-world applications of Machine Learning
• Differentiating different ML types
• Introducing popular Machine Learning tools and frameworks

Speaker: Alexander Fischer (DKRZ)

12:30 Lunch Break

14:00 Architectures and Applications | Lecture 🎓

• An overview of state of the art Deep Learning Methods
• Examples from weather, climate and beyond
• Explainable AI

Speaker: Paul Keil (HEREON)

15:45 Coffee break

16:00 Introduction to PyTorch: Core Concepts | Hands-on 💻

• Preparation of the DKRZ accounts for the Pytorch tutorials
• Brief theoretical introduction to PyTorch
• Hands-on illustrating the core concepts

Speaker: Etienne Plésiat (DKRZ)

Wednesday 23 September

09:00 → 17:30 Day 2

09:00 Introduction to PyTorch: Application to climate data | Hands-on 💻

• Definition of the task
• Creation of the training, validation and test datasets
• Building the model
• Training the model
• Testing the model

Speaker: Etienne Plésiat (DKRZ)

10:45 Coffee Break

11:00 Introduction to PyTorch: Application to climate data | Hands-on 💻

• Improving the model

Speaker: Etienne Plésiat (DKRZ)

12:00 Lunch break

13:30 Introduction to Pytorch Lightning | Hands-on 💻

• Learn the basics of the deep learning framework Pytorch Lightning
• Write deep learning code that is flexible and performant

Speaker: Caroline Arnold (HEREON)

14:30 Advanced Deep Learning use-case: Reconstructing missing climate data | Hands-on 💻

• Design and implement inpainting CNN for reconstructing climate data

Speaker: Johannes Meuer (DKRZ)

15:15 Coffee break

15:30 Advanced Deep Learning use-case: Reconstructing missing climate data | Hands-on 💻

• Train the model with different configurations
• Validate the model on test data

Speaker: Johannes Meuer (DKRZ)

16:30 Advanced Deep Learning Use Case: AI Weather Prediction | Lecture 🎓

• Learn about AI weather models that can take it up with numerical weather prediction

Speakers: Caroline Arnold (Hereon), Paul Keil (DKRZ / HEREON)

Thursday 24 September

09:00 → 12:30 Day 3

09:00 Beyond CNNs: Transformers & Attention | Lecture 🎓

• Limitations of CNNs
• Introduction of transformer architecture
• Alternative approaches for predicting continuous data with ML

Speaker: Alexander Fischer (DKRZ)

09:45 Probabilistic Deep Learning | Lecture 🎓

• Overview of Probabilistic Deep Learning
• Introduction to Diffusion Models
• Examples from Weather and Climate Research

Speaker: Paul Keil (HEREON)

10:30 Coffee break

10:45 Attention & Transformer | Hands-on 💻

• Hands-on downscaling of multiple variables using a transformer architecture on the HEALPix sphere.

Speakers: Alexander Fischer (DKRZ – DA Group), Johannes Meuer (DKRZ), Maximilian Witte

10:45 Probabilistic Deep Learning | Hands-on 💻

• Extension of Reconstructing missing climate data exercise
• Implementation of a simple Flow Matching Model
• Evaluation of ensemble predictions and model uncertainty

Speaker: Paul Keil (HEREON)

10:45 Q&A | Recap💬

• Recap session for deep learning basics covered in the course
• Open Q&A

Speaker: Caroline Arnold (Hereon)

12:00 Closing remarks | Discussion 💬

Speaker: Paul Keil (HEREON)