Quantifying urban building shading effect for data-driven building energy modeling

Urban shading is an urban phenomenon driven by geometric interactions between buildings, exerting a significant influence on heating and cooling loads. Nevertheless, current urban-scale building energy modeling remains heavily reliant on building attributes, weather data, and historical energy consumption data. Although incorporating historical energy consumption as an input feature can enhance model prediction accuracy, it may limit the contribution of other input features and reduce the model's practical utility—particularly when models are intended not only for predicting energy consumption but also for evaluating design alternatives, such as building energy retrofit decisions. Under these circumstances, augmenting urban layout information—such as shading—as additional input features can maximize the utilization of available data. However, discussions regarding the effectiveness of such augmentation and how these features should be structured as model inputs remain largely unexplored to date. In this context, this study investigates how shading-related input features can be effectively structured and incorporated into data-driven models to predict monthly building energy consumption without relying on historical energy data. Leveraging an urban-scale virtual experiment framework based on Korean GIS and EnergyPlus, a building surface-level shading database—including shading ratio, area, and orientation for each building surface—was developed for over 250,000 buildings in Seoul. From this database, candidate building-level shading indices were derived and quantitatively evaluated regarding their impact on prediction performance. Results from the case study indicated that even a simplified shading index, averaged across all wall orientations and temporal averaging intervals, significantly enhanced prediction accuracy compared to a baseline model relying solely on building and weather data. Moreover, the analysis revealed correlations between shading indices and urban layout parameters, particularly building density within urban blocks. The findings underscore the value of incorporating shading information into building energy prediction models, demonstrating that such input features remain effective and practical even at reduced dimensionality. These insights offer practical guidance for developing robust urban-scale energy models and have implications for urban planning and energy retrofit decision-making.