Speaker: Prof. Dr. Alexander Brenning (Friedrich Schiller University Jena)
Whether groundwater contamination, soil properties, or natural hazards in a changing climate, geostatistics and spatial machine learning now enable high-resolution, data-driven predictions. But how well do these models generalize to new geographic areas and environmental conditions that are barely represented in the training data? Using examples from nitrate contamination modelling, digital soil mapping of desert pavements, and landslide hazard assessment, the lecture explores the challenges of extrapolation, nonstationarity, and model validation. It also shows how process knowledge and data-driven approaches can be combined through process-informed AI to enable more robust predictions under changing environmental conditions.