Paul Keil

• Machine Learning in Weather and Climate
I am interested in developing diverse machine learning solutions on all scales in weather and climate, from small models that predict Arctic surface fluxes or urban air quality to large models trained on vast amounts of patio-temporal data, for example to forecast soil moisture.

• Hybrid Modelling:
Machine learning in climate science faces unique challenges due to the out-of-training-distribution application for a future or past climates, as well as alternative scenarios. I believe that a hybrid approach that couples numerical methods with novel machine learning methods offers the best path forwards for researchers to tackle these challenges.

• Explainable und Probabilistic Machine Learning
Both researchers and stakeholder need to present and justify weather and climate forecasts to the public. Therefore, machine learning needs to explainable and must include an uncertainty estimate to increase trust and provide accountability.