Michael Maier-Gerber (IMK-TRO):
Statistical-dynamical forecasts of tropical cyclone activity
Nicole Bäuerle (STOCH) & Uwe Ehret (IWG):
HYbrid, NETwork-based modelling of HYdrological systems (HY-NET)
Sebastian Lerch (STOCH), Julian Quinting & Christian Grams (both IMK-TRO):
Probabilistic weather regime prediction: Combining physical models and generative machine learning
When: 5th October, 12 - 2 PM
Where: Online in Zoom
Dial In Information: Zoom Link
Meeting-ID: 629 5966 8343
Kenncode: 377076Zoom Link
Method Area 4
The research activities in MB 4 focus on applications to forecasting, with preexisting, strong ties between mathematical scientists and meteorologists within Transregional Collaborative Research Center “Waves to Weather” (CRC TRR 165), and leading to the recent creation of a junior research group on "Artificial Intelligence for Probabilistic Weather Forecasting". We continue to develop methods for the generation and evaluation of weather forecasts, with the aim of developing the best forecasting systems possible.
Conversely, the encountered applied challenges stimulate and inform theoretically and methodologically oriented research on the foundations of predictive science, and particularly on the evaluation of predictive performance, with close ties to current developments in machine learning.
Similarly, in a Helmholtz Association supported project on "Scalable and Interpretable Models for Complex and Structured Data" (SIMCARD) we tackle data science challenges in meteorology and epidemiology. In view of new societal challenges, attention on this project has shifted to the scientific response to the coronavirus pandemic, including the development of the German-Polish COVID-19 Forecast Hub (https://kitmetricslab.github.io/forecasthub/forecast) and associated methods for the generation and evaluation of epidemiological ensemble forecasts.
Statisticians from HITS, KIT, and Heidelberg University introduce the new "CORP" approach to better determine the reliability of forecasting methods. CORP also has an impact on machine-learning methods in general. The paper was published in the Proceedings of the National Academy of Sciences (PNAS).