Young Investigator Group Artificial Intelligence for Probabilistic Weather Forecasting
The young investigator group is funded by the Vector Stiftung within the framework Nachwuchsgruppe MINT für die Umwelt (YIG STEM for the environment) and started in April 2021.
We will develop statistical and machine learning methods for probabilistic weather forecasting. Leveraging advances in deep learning, we aim to improve ensemble post-processing methods by incorporating spatial and temporal dependencies, flexible non-parametric models for complex response distributions and the development of novel analog forecasting methods.
Project scientists:
Sebastian Lerch (group leader)
Jieyu Chen (PhD student)
Nina Horat (PhD student)
SPAtial ensemble post-pRoCessing using generative Machine Learning (SPARC-ML)
The SPARC-ML project is funded by the Deutscher Wetterdienst (German Weather Service) with the framework Extramurale Forschung IV (Extramural Research Programme) and started in January 2024.
With Co-PI Peter Knippertz, our aim is to develop spatial verification methods and generative machine learning approaches for spatial post-processing of wind gust forecasts over central Europe.
Project scientists:
NN (PhD student)
Waves to Weather (W2W)
W2W is a trans-regional collaborative research center funded by Deutsche Forschungsgemeinschaft (DFG). Combining expertise in meteorology, mathematics and computer science, W2W aims to identify limits of weather predictability and improve forecasts.
With Peter Knippertz, I am Co-PI for project C5 - Dynamical feature-based ensemble post-processing of wind gusts within European winter storms where we will develop physically-based ensemble post-processing methods for individual dynamical features within winter storms over central Europe.
Project scientists:
Lea Eisenstein (PhD student)
Benedikt Schulz (PhD student)
Development of a deep learning prototype for operational probabilistic wind gust forecasting
This transfer project connected to W2W aims at transferring research on neural network-based post-processing methods to operational weather prediction at national meteorological services, in collaboration with KNMI, the Dutch meteorological service. The project started in July 2023.
Project scientists:
Benedikt Schulz (Postdoc)
Jieyu Chen (PhD student)
TEEMLEAP - A new TEstbed for Exploring Machine LEarning in Atmospheric Prediction
TEEMLEAP is an interdisciplinary KIT Future Fields Pilot Project and started in December 2021. Lead by Peter Knippertz, I am Co-PI with Uwe Ehret, Jörg Meyer, Julian Quinting and Barbara Verfürth.
TEEMLEAP is a collaboration between scientists from the KIT Center for Climate and Environment and MathSEE aiming at establishing an idealized testbed for exploring machine learning in weather forecasting. We will evaluate the application possibilities and benefits of machine learning along the entire process chain of weather forecasting in this testbed.
Project scientists:
Thomas Muschinski (Postdoc)
Jannik Wilhelm (Postdoc)
MathSEE Bridge PhD: Probabilistic weather regime prediction
The interdisciplinary Bridge PhD project Probabilistic weather regime prediction is funded by the KIT-Center MathSEE and started in May 2022. I am Co-PI with Julian Quinting and Christian Grams.
In the Bridge PhD project, we will combine physical models and generative machine learning approaches to provide skillful multivariate probabilistic predictions of weather regimes on sub-seasonal to seasonal time scales. A specific focus will be on incorporating domain knowledge from the atmospheric sciences into deep learning models.
Project scientists:
Fabian Mockert (PhD student)