​

​Past Seminars​​​

​

Jurriaan van 't Hoff (Netherlands Institute for Space Research SRON, TU Delft) – Data-driven discovery and model reduction methods for atmospheric chemistry transport model data (13 March 2026)

​​

Abstract: Chemistry transport models play a crucial role in the evaluation of anthropogenic emissions on the composition of our atmosphere and global radiative forcing. These models also come with a very high computational costs and require specialized compute clusters and know-how to use, while also producing a large volume of output data. These properties can make it difficult to use CTMs to support processes such as multidisciplinary design optimization, or regulatory and operational decision making. We explore the use of data-driven model discovery methods such as dynamic mode decomposition (DMD) and proper orthogonal decomposition coupled with the sparse identification of non-linear dynamics (POD-SINDy) to alleviate some of these challenges. Applying these methods to multiannual datasets describing the tropo- and stratospheric ozone response to supersonic aircraft fleets, we show that these methods can produce efficient reduced order models for the reconstruction of the original data, and in some cases even forecast future trends. 

​

​

Maurice Schmeits (KNMI) – Machine Learning Weather Prediction development at KNMI​ (12 March 2026)

​

Abstract: At KNMI we are developing so-called stretched-grid models (SGMs), building on work of ECMWF and Met Norway. First, a deterministic base SGM was trained on ERA5 and CERRA reanalysis (5.5-km resolution) data, after which a finetuned SGM was developed using ECMWF IFS and UWC-West Harmonie-Arome analysis (2-km resolution) data. Preliminary verification results are shown for this SGM. As a next step we are developing an ensemble version of an SGM, because weather forecasts are inherently uncertain. Then I will shortly talk about 2 other projects in which KNMI is involved, and I will finish by giving an outlook.
 

​

Felix Dahle (TU Delft) – Antarctic Time Machine 3D Reconstruction of Glaciers in the Antarctic Peninsula using Historical Structure-from-Motion (11 March 2026) ​

 

Abstract: I will be presenting a methodology to reconstruct the historical topography of the Antarctic Peninsula using a vast archive of aerial photography from the mid-20th century. By converting these historical 2D images into 3D models, the research aims to visualize decades of hidden climate history and track long-term glacier retreat. The approach automates the Structure-from-Motion pipeline through four key steps: semantic segmentation to filter unusable data, metadata extraction to retrieve camera geometry, geo-referencing to align historical photos with modern satellite imagery, and finally, 3D reconstruction. This automated process allows for the creation of historical elevation models that can be compared against modern data to accurately measure ice loss and improve future sea-level rise predictions.​

​​

​​

Luca Lanzilao (BFH) – Intraday solar energy forecasting at national scale using satellite-based solar forecast models (29 January 2026) [Abstract]

​​​

​

Ana Marza (UNIL / BFH) – Unraveling the sources of subseasonal predictability (9 December 2025) [Abstract]

​

​​