Teacher: Prof. Giovanna Venuti
Supervisor: Dr. Domenico Grandoni E-GEOS
Description: Physics-based Large-Eddy Simulation (LES) models such as PALM require detailed, spatially distributed thermo-radiative surface parameters — wall and roof materials, pavement and soil types, albedo, emissivity — encoded in the PALM Input Data Standard (PIDS) static driver. Current workflows rely heavily on manual photointerpretation and generic look-up tables, introducing spatial inconsistencies and significant human effort. This project develops a GeoAI pipeline that automatically classifies building facade and ground-surface materials from open Mapillary street-level imagery, leveraging deep learning architectures (CNN/ViT, semantic segmentation, Vision-Language Models) and geospatial linking with 3D city models (CityGML LOD2, OSM footprints).