Lunch Talk: Machine Learning in Environmental Seismology: Linking Signals to Processes

Research talk with Dr. Josefine Umlauft on machine learning for environmental seismology, heterogeneous sensor data, and reproducible time-series analysis.

The next BIFOLD Lunch Talk features Dr. Josefine Umlauft, who will present her research on “Machine Learning in Environmental Seismology: Linking Signals to Processes.” The talk looks at environmental seismology as an application field for time-series machine learning and addresses the challenges of working with weak, overlapping, non-stationary, and strongly autocorrelated signals.

Using seismic forest monitoring as a case study, the session explores how seismic data can be combined with meteorological observations and physiological measurements to investigate tree and forest dynamics. Topics include the harmonization of heterogeneous sensor data, temporal dependence in model training and evaluation, information leakage, and the transferability of learned relationships across sensors and environmental conditions.

The talk also discusses approaches ranging from signal-derived features and dimensionality reduction to predictive modelling, as well as what is needed to make environmental ML analyses more reproducible and transferable.

Date: October 28, 2026
Time: 12:00–1:00 PM
Location: BIFOLD, FR 701, 7th floor, Franklinstraße 28/29, 10587 Berlin + online
Organizer: BIFOLD
Speaker: Dr. Josefine Umlauft
Role: Head of the Earth and Environmental Sciences research group at ScaDS.AI
Admission: Registration / further information via BIFOLD
Audience: BIFOLD members, fellows, external partners, and researchers interested in machine learning and Big Data
Focus: Environmental seismology, time-series machine learning, heterogeneous sensor data, seismic forest monitoring, information leakage, transferability, reproducibility
Case study: Combining seismic, meteorological, and physiological data to study forest and tree dynamics