Details of the talk:
- Date: July 25, 2024
- Time: 11:00 a.m. - 12:00 p.m.
- Location: Ground-floor lecture hall, Tübingen AI Center (Maria-von-Linden-Str. 6, 72076 Tübingen)
Talk title: 'Leveraging graph neural networks in data-driven weather forecasts at an operational weather centre'
Abstract: Machine learning, especially Graph Neural Networks (GNNs), is revolutionizing real-world applications, and numerical weather prediction is no exception. Weather forecasting models are traditionally built on physical simulations and complex mathematical formulations. At ECMWF we run these models operationally 24/7 on four super-computers, to provide a reliable, accurate and high-resolution global forecast twice a day that is provided to our member states and used around the globe. However, recent advancements in GNNs present an exciting opportunity to capture the intricate spatial and temporal relationships in weather data. In this talk, we will explore the role of machine learning in improving the accuracy and efficiency of weather forecasts. We will delve into how GNNs can model the interconnectedness of atmospheric variables across different locations and scales, offering a powerful tool to complement and potentially surpass traditional methods. We’ll touch upon how we do this safely and reliably, considering the weather impacts us all and a missed forecast means more than just a misplaced advertisement on a website.
Bio: Jesper Dramsch works at the intersection of machine learning and physical, real-world data. Currently, they're working as a scientist for machine learning in numerical weather prediction at the coordinated organisation ECMWF.
Jesper is a 2022 fellow of the Software Sustainability Institute, creating awareness and educational resources, like https://ml.recipes around the reproducibility of machine learning results in applied science. Before, they have worked on applied exploratory machine learning problems, e.g. satellites and Lidar imaging on trains, and defended a PhD in machine learning for geoscience. During the PhD, Jesper wrote multiple publications and often presented at workshops and conferences, eventually holding keynote presentations on the future of machine learning in geoscience.
Moreover, they create educational content in the form of notebooks on Kaggle applying ML to different domains, reaching rank 81 worldwide out of over 100,000 participants. Their video courses on Skillshare have been watched over 5000 hours by over 9,000 students. Jesper was invited into the Youtube Partner for their videos around programming, machine learning, and tech and they write a weekly newsletter to over 1,111 subscribers about non-hype AI at https://late.email.