High-Resolution Solar Energy Estimation in Urban Areas: Integrating LiDAR Data and Deep Learning for Sustainable Planning

Accurate high-resolution solar energy potential maps are essential for optimal rooftop solar panel placement in urban areas. Physics-based models such as the Area Solar Radiation (ASR) tool produce reliable estimates but require prohibitive computation times when applied to high-resolution LiDAR Digital Surface Models (DSMs) at the city scale. While deep learning models have been proposed as faster alternatives, their practical feasibility for end-use applications such as rooftop panel siting remains unexamined. This study introduces a two-phase framework to address this gap. In Phase ۱, three U-Net-based architectures (U-Net, Attention U-Net, and U-Net ۳+) are trained to estimate annual solar energy potential maps (ASMs) from ۰.۵ m LiDAR DSM patches, using ASR-generated ASMs as reference. In Phase ۲, the best-performing model is evaluated for the practical task of rooftop solar panel placement under real-world constraints, including minimum roof area, maximum slope, and aspect. Results show that U-Net ۳+ achieves the highest accuracy (RMSE = ۹۴.۳۵۳ kWh/m², R² = ۰.۹۱) while reducing computation time from over ۱۴ hours (ASR) to approximately ۲۵ seconds. When applied to rooftop panel siting, the predicted ASM yields an R² of ۰.۹۷ relative to building-level reference values. These findings demonstrate that deep learning models can serve as computationally efficient and practically accurate alternatives to physics-based solar radiation models, enabling rapid city-scale solar potential mapping for sustainable urban planning.

کلیدواژه ها:

Solar Energy Potential Map ، Solar Panel Placement ، Deep Learning ، U-Net Architecture ، LiDAR DSM