AutoBEM-SkelZone: automated thermal zoning for Urban building energy modelling

Urban building energy modelling (UBEM) requires thermal-zoning strategies that balance physical detail, computational cost, and limited availability of interior building data. Conventional one-zone-per-floor representations reduce runtime but aggregate perimeter and core conditions, potentially masking spatial differences in envelope exposure and HVAC loads. This study presents AutoBEM-SkelZone, an automated zoning workflow that applies core–perimeter partitioning to convex footprints and straight-skeleton partitioning to non-convex geometries. The method is integrated into AutoBEM and automates geometry validation, surface matching, construction assignment, and HVAC propagation for simulation-ready EnergyPlus models. AutoBEM-SkelZone was evaluated using 143 archetypes representative of the New York metropolitan area. Of the 143 cases, 139 (97.2%) completed within the predefined batch runtime limit, while the remaining four generated valid models and completed after the limit was relaxed. Relative to the one-zone-per-floor baseline, the multi-zone models predicted 4.9% higher aggregate electricity use and 6.2% higher aggregate whole-building natural-gas use. These results demonstrate that zoning resolution materially affects simulated energy consumption. Representative time-series comparisons showed similar temporal patterns, with differences concentrated in seasonal heating and cooling periods. Both AutoBEM variants were also compared with 16 detailed DOE Standard 90.1–2016 prototype models. End-use-specific analysis showed that aggregate natural-gas space-heating bias changed from −12.7% for the one-zone-per-floor models to +3.5% for the multi-zone models, while service-water-heating consumption remained effectively identical. However, the multi-zone models were closer in only 5 of the 14 prototypes with positive reference gas heating, and building-level space-heating errors increased. For electricity, the multi-zone models were closer in 6 of 16 cases and reduced mean absolute percentage error from 30.9% to 29.2%, although median absolute percentage error and RMSE increased. These findings show that automated zoning materially changes predicted loads but does not uniformly improve prototype-level agreement. Although AutoBEM-SkelZone entails greater model complexity and longer runtimes, its integration within an automated GIS-native workflow provides a practical basis for studying zoning effects in stock-scale UBEM. The method supports reproducible perimeter–core load differentiation while making the associated accuracy–runtime trade-offs explicit.