As an introduction, Caroline Arnold explained background and best practice applications of machine learning and reviewed Python basics that are essential for their use. Christopher Kadow presented examples of the application of ML methods in climate science and showed that all modern ML techniques can be applied here.

In the main part of the course, Étienne Plésiat demonstrated the common ML framework PyTorch. In interactive Jupyter notebooks, the participants learned how to define a neural network and a data set themselves. Together they trained a classification algorithm that could be used to predict the month of data observation for temperature data from the ERA5 data set.

Finally, Johannes Meuer presented the use of ML on the GPU partition of Levante. The participants were able to train a prepared network that fills gaps in radar images.

Despite the wide range of ML online courses, the demand for a special workshop for climate scientists was very high. Since the course was overbooked many times over within just a few hours and the evaluation was also very positive, the DKRZ plans to offer it again in autumn.

The training material is not available online, as the hands-on course content in particular was first developed together with the participants. The following online courses can be recommended for self-study: