Deutsch Intern
    Data Science Chair

    Praktikum: Machine Learning for Time Series Analysis

    Deep Learning for Climate Modelling

    Supervisors: Dr.-Ing. Anna KrauseFlorian Gallusser

    Scope: 5 ECTS

    Course ID: 10-I=PRJAK

    WueCampus course: Link 

    Kickoff date: 17th April 2024, 16:00-17:00, G50 (Emil-Fischer-Straße 50), Seminar room 1

    Registration: via WueCampus course with the key "Pr_DL4Clim"


    In this Master's practical course, the students will deal with Deep Learning (DL) methods for spatio-temporal data analysis. The students will individually work with a meteorological dataset and develop different approaches for solving a given problem. To do this, students draw upon current approaches from the scientific literature. At the end of the semester, the different approaches are presented to the entire group and an evaluation on a common held-out test dataset is performed.

    The focus of this course is on the theoretical development, practical implementation and scientifically sound evaluation of new Deep Learning approaches for spatio-temporal data.


    In this semester, students will work with the meteorological reanalysis dataset WeatherBench2. This contains global reanalyzed weather data on a two-dimensional latitude-longitude grid of different physical variables, e.g. temperature, wind speed, geopotential, relative humidity, etc., on multiple vertical levels. The aim of the practical course is to train a supervised DL model for global medium-range weather forecasting with a time horizon of up to 14 days.


    Registration for topics opens 10th April 2024 14:00.

    After registration for the course, all participants decide on one of the predefined methods published in the scientific literature for working on the given problem. In the first two weeks of the course, students present their chosen method to the group. In the following weeks, the students implement their respective approach independently. During regular meetings with their supervisor, the students have the opportunity to discuss encountered problems and questions.

    At the end of the semester, each student should be able to present a functioning forecasting model, which is evaluated on a held-out test data set. If the system works well, there is the possibility of a scientific publication and a presentation at a scientific workshop.



    At the end of the semester, all students present their work in a 15-30 minute talk. In addition, a report of 10-15 pages must be submitted.