PV potential analysis through deep learning and remote sensing-based urban land classification

Urban land utilization for commerce, residence, grassland, and other administrative subdivisions will affect the available area for renewable infrastructure setup, such as photovoltaic (PV) panels. Incorporating land use types into PV potential assessments is essential for optimizing space allocation, aligning with energy demand centers, and enhancing efficiency. To address the limitations of previous studies that overlook urban land use, this study introduces a framework leveraging remote sensing data and deep learning methods to achieve eight fine-grained and three coarse-grained land use classifications. The framework calculates the PV installation area for each land use type and evaluates their power generation potential based on the yearly average solar irradiance in 2023. Case studies demonstrate that Germany Heilbronn land is suitable for ground PV installations, with a power generation of 5333.85 GWh/year, and rooftop PV installations are the most productive for electricity generation in New Zealand Christchurch, with 3290.08 GWh/year. Unutilized land in Heilbronn and Commercial land in Christchurch is estimated to be the most productive per unit area. Finally, the uncertainty of the PV installation ratio by adopting σ i and the confidence interval of potential estimation is discussed. This work experiments with the framework successfully and highlights the effects of the PV installation ratio on the power generation of each land use, providing valuable instructions for urban land utilization and PV installation. • A framework is proposed to evaluate the impact of urban land on PV installation availability and power generation. • Classifying eight fina-granularity land use types by utilizing the deep learning method and Million-AID dataset. • Considering land use types and PV installable area could avoid an overestimation of 76 % on potential. • The study provides insights into suitable land use types for PV installation to different stockholders.