STACKING-BASED UNCERTAINTY MODELLING OF STATISTICAL AND MACHINE LEARNING METHODS FOR RESIDENTIAL PROPERTY VALUATION

Glaciers are melting worldwide, but direct measurements are spatially sparse. We developed a new model that combines mechanistic physical rules with machine learning to better estimate yearly gains and losses of ice mass across Europe. By learning local corrections to a physical melt model, the machine learning component improves accuracy and works well on glaciers not used for training. This will help produce more reliable glacier assessments and support future climate studies.