Super-Resolution of Sentinel-2 Imagery Using Latent Diffusion Models for Photovoltaic Site Assessment

The growing demand for renewable energy emphasizes the critical need for detailed geospatial information in photovoltaic (PV) site assessment and planning. While Sentinel-2 imagery provides a valuable resource, its native 10-meter spatial resolution limits the identification of small urban structures, such as individual rooftops and narrow roads, thereby constraining accurate solar suitability analyses. To overcome this limitation, this paper presents a comprehensive PV assessment and optimization framework integrating a resolution enhancement module based on latent diffusion models. Operating in the latent space, this module utilizes an iterative diffusion process to accurately reconstruct fine urban structures. Cloud-filtered Sentinel-2 L2A scenes are processed to produce enhanced imagery with an effective 2.5-meter resolution. Pretrained on cross-sensor datasets, the model realistically recovers critical small features while maintaining spectral coherence. This enhanced imagery enables precise rooftop segmentation, which drives a robust PV potential assessment. The subsequent installation optimization maximizes energy generation by integrating solar radiation, shading analysis, rooftop orientation, tilt angles, and panel layout efficiency, alongside technical and economic constraints. Qualitative evaluations demonstrate high-quality visual enhancement, confirming the relevance of this resolution-enhancement step for real-world PV site suitability analysis and solar deployment optimization.