Grid-Based Estimation of Building Structure and Construction Year by Integrating Multispectral Satellite Data and Building Footprints

Building location and attribute data are essential for urban planning and disaster mitigation. However, their availability remains limited, and current estimation methods suffer from limited accuracy and applicability. We previously developed a building footprint detection model using open-access satellite imagery. In this study, to assign attributes to individual footprints, we estimated building structure type and construction year for each cell of 500 m grid. We constructed multiple grid-based models using Sentinel-2 imagery and publicly available building footprints, and compared their performances under different input configurations. The multimodal model that integrates Sentinel-2 and footprint images consistently outperformed the single-source models across all categories. Additional analysis showed that the multimodal model improved performance in densely built-up residential areas, indicating that building footprints complements spectral features. These findings demonstrate the feasibility of inferring building attributes at the 500 m grid scale using only open-access data.