Automated geometric decomposition and semantic enrichment of shared wall boundary surfaces in LoD 2 CityGML models for urban building energy modelling

Abstract. Urban Building Energy Modelling (UBEM) can be adapted to assess thermal comfort and climate resilience in dense urban cities. However, common modelling and simulation practices treat shared wall boundaries between adjacent buildings as adiabatic, de facto ignoring inter-building heat transfer, which can lead to deviations in energy and temperature simulations. To resolve this challenge, an automated framework is developed that identifies, geometrically decomposes, and semantically enriches shared wall boundaries within complex Level-of-Detail (LoD) 2 CityGML models. Application of the framework to a 1,429-building dataset in Antwerp, Belgium, successfully identified and decomposed 6,667 shared wall surfaces, maintaining topological validity for 1,404 of the 1,425 (98.53%) originally valid buildings prior to any external healing. For comparative analysis, both the original and the geometrically decomposed CityGML models were processed through a computational pipeline to generate simulation-ready epJSON files for EnergyPlus. This translation highlights a fundamental friction between the inherent geometric imperfections of sourced real-world, surveyed geospatial data in CityGML and the idealized geometries expected by EnergyPlus. Comparative urban simulations confirm that the modelling of explicit shared surface heat transfer, compared to adiabatic assumption, yields systematically higher heating demand projection (mean bias error (MBE) = +23.67 kWh/m², mean absolute error (MAE) = 24.14 kWh/m², root mean square deviation (RMSD) = 31.99 kWh/m²) and lower indoor operative temperatures (MBE = −0.216 °C, MAE = 0.235 °C, RMSD = 0.366 °C). These findings emphasize the necessity of explicit shared-boundary modelling for accurate, actionable district-level energy and indoor thermal comfort assessments.