Volume 61

Solar PV Installation Potential Assessment on Building Façades Based on Vision and Language Foundation Models Ruyu Liu 1* , Dongxu Zhuang 2* , Xiufeng Liu 1✉, Arega Getaneh Abate 1 , Per Sieverts Nielsen 1 , Jianhua Zhang

Abstract

The assessmen of solar photovoltaic (PV) potential on
urban building façades is pivotal for sustainable urban
planning, yet is often constrained by manual, timeintensive
methods. This study introduces a novel end-to- end framework, The Semantic Façade Solar-PV Assesse
-ment (SF-SPA) , that pioneers a paradigm shift by lev- eraging generalist foundation models for this task. Our
pipeline integrates vision foundation models (VFMs) and
large language models (LLMs) for rapid, automated, and accurate façade-based PV assessment from single
2D street-view images. The framework was validated
on a diverse dataset of 80 buildings from four cities ac- ross different climates and architectural styles. Re- sults show high accuracy, with an average area estim- ation error of 6.2% against expert-defined ground truth, and exce-ptional efficiency at approximately 100 sec- onds per bui-lding. This work demonstrates a scalable
and data effici- ent alternative to traditional methods
that rely on 3D data or specialized trained models, paving the way for large scale urban energy analysis.

Keywords Solar PV potential, Large Language model, Semantic segmentation, Urban Building façades