Understanding how energy is consumed at the scale of individual buildings is fundamental to urban decarbonization and heat mitigation. Although building energy consumption is individually metered, citywide building-resolved datasets are typically unavailable due to privacy protections and the lack of centralized, research-ready data infrastructures. Consequently, cities lack the fine-scale information needed for spatial energy analysis. This study reconstructs missing building-level energy consumption using a machine-learning framework that combines limited available consumption data with rich physical, socioeconomic, meteorological, and human-activity inputs. We develop a machine learning model trained on more than 3.5 million samples across Seoul from 2020 to 2024. The model achieves strong predictive performance (R2 = 0.71) despite the substantial heterogeneity of building types and occupant behaviors. Diurnal population ratios reveal that human activity in Seoul is strongly concentrated during early afternoon hours, with weekday-dominant buildings—such as offices, educational facilities, and Type-1 neighborhood services—exhibiting pronounced daytime intensity. Weekend-dominant buildings show a more balanced day–night pattern, illustrating how differing behavioral schedules shape energy demand. Spatial analysis highlights distinct energy hotspots associated with mixed-use corridors, dense commercial districts, and winter-sensitive residential areas located near mountainous terrain. By combining population dynamics with machine-learning-based energy estimation, this study provides the first citywide, parcel-level electricity dataset for Seoul alongside a behavioral interpretation of why energy use varies across the urban fabric. The results offer a scalable pathway for urban decarbonization planning, enabling more accurate assessments of energy demand, waste heat emissions, and human-environment interactions at unprecedented spatial detail.
Lee Yeonsu is a PhD candidate in Environmental Science at the Ulsan National Institute of Science and Technology (UNIST). Her research focuses on urban AI, remote sensing, and urban climate sustainability. She is currently completing her doctoral thesis and has published various research papers, including urban tree carbon storage, evaluating the cooling effect of urban canopy, and designing decarbonization pathways for the transport sector. She has been awarded a doctoral research scholarship from the National Research Foundation of Korea.
Professor Jungho Im is a full professor at the Ulsan National Institute of Science and Technology (UNIST) in South Korea, where he leads the Intelligent Remote Sensing and Geospatial Information Science (IRIS) Lab and focuses on environmental science, remote sensing, and artificial intelligence. He also holds editor-in-chief of GIScience and Remote Sensing and associate editor for the ISPRS Journal of Photogrammetry and Remote Sensing.