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

A one-hour research talk at BIFOLD on what environmental seismology does to time-series machine learning. Dr. Josefine Umlauft works with seismic data that is weak, overlapping and autocorrelated, and usually has to be fused with other environmental sensors before a model can say anything useful. Using seismic forest monitoring as the case study, she covers data harmonisation, temporal dependence, information leakage and transferability, and what reproducible analysis actually requires in this setting. Part of the BIFOLD Lunch Talk series, held on the 7th floor at Franklinstraße 28/29 and also streamed. Why go: a concrete hour on the problems that break time-series models in the field, from someone whose data will not behave. Best for ML engineers and researchers working on sensor, signal or environmental data.