Citywide Parcel-Level Electricity Use Estimation from Building GIS, Microclimate, and Human Activity Data in Seoul

Despite being a key determinant of urban energy demand, anthropogenic heat emissions, and indirect carbon dioxide emissions, precise observational data for entire cities remains scarce. This study develops a data-driven framework that reconstructs monthly electricity consumption for individual parcels across Seoul by integrating building characteristics, microclimate, and human activity. We collected millions of monthly observation records from 2020 to 2024 and converted billed electricity quantities into electricity use intensity (EUI, kWh m⁻²) using building floor areas. These records were linked with parcel-level attributes (e.g., land use, land price, gross floor area, construction year), local climate zones, socioeconomic indicators, high-density Smart Seoul City Data of Things (S-DoT) meteorological observations, and hourly living population data. Random Forest and LightGBM models were trained and evaluated using 5-fold cross-validation. LightGBM demonstrated the best performance across all parcels, achieving a Mean Absolute Error (MAE) of 6,712 kWh, a Weighted Mean Absolute Percentage Error of 39.6%, and an R² of 0.709. SHAP (SHapley Additive exPlanations) analysis revealed urban land price, building size, construction year, and income as key determinants of EUI. Concurrently, the living population and microclimate variables exerted nonlinear additional effects, particularly in high-activity commercial districts. High-density, high-rise business centers exhibited high power intensity despite relatively mild outdoor maximum temperatures, suggesting a decoupling between indoor cooling demand and the surrounding thermal environment. The estimated dataset for building-specific electricity consumption across the entire city provides essential data for artificial heat estimation, energy planning, and future urban climate and emissions modeling.