Speaker: Prof. Dr. Alexander Brenning, Friedrich Schiller University Jena, Germany Date: October 15, 2026, 10:00 CEST Abstract Spatial prediction models are commonly validated under...
Show moreShow less
Speaker: Prof. Dr. Alexander Brenning, Friedrich Schiller University Jena, Germany
Date: October 15, 2026, 10:00 CEST
Abstract
Spatial prediction models are commonly validated under conditions that differ from those encountered at deployment, potentially resulting in misleading performance assessments. This talk presents three complementary approaches for aligning validation with its intended use: target-weighted cross-validation for distribution shift, decision-aligned performance measures for bias correction in regulatory decision-making, and support adjustment for mismatches between observation and prediction support size. Case studies from environmental pollution modelling illustrate how these adjustments can improve the relevance and interpretability of validation results.
Short bio
As a professor of geographic information science in the Department of Geography of Friedrich Schiller University Jena, Germany, Alexander Brenning’s research focuses on modeling Earth surface processes using machine-learning and geostatistical tools as well as process-based models. Application domains include but are not limited to landslides, mountain permafrost, and environmental remote sensing, and current methodological interests include model interpretation, hybrid modeling, and the quantification of model uncertainties.
He joined the University of Jena in 2015 as a full professor after holding a faculty position at the University of Waterloo, Ontario, Canada since 2007. Alex holds a Ph.D. in Geography from Humboldt-Universität zu Berlin and graduated in Applied Mathematics at Technical University of Freiberg, Germany. He has visited the University of Heidelberg, Germany as a Humboldt Research Fellow and the Pontifical Catholic University of Chile as a Distinguished Visiting Professor.
