GeoAI-Driven mapping of urban rooftop Photovoltaics: A sustainable energy framework for Karachi, Pakistan

• GeoAI framework for RPV mapping and potential estimation. • Weighted majority ensemble enhanced RPV detection reliability over individual DL models. • Spatiotemporal weather data integration for precise RPV yield estimation. • Karachi’s RPV potential: 602 GWh annually, reducing 0.37 Mt CO2. • RPV economic viability confirmed; scalable framework for urban energy planning. Rooftop Photovoltaics (RPV) offer a sustainable solution for renewable energy integration, leveraging urban landscapes to address environmental challenges such as carbon emissions and resource management. This study introduces a novel methodology for mapping and estimating RPV potential in Karachi, Pakistan, using earth observation data derived from satellite imagery. We developed a custom dataset of high-resolution satellite images capturing diverse solar panel configurations and employed an ensemble deep learning approach—combining UNet-ResNet50, DeepLabv3-ResNet50, Mask2Former-SwinTransformer, SamLoRA-vit_b, and PSPNet-ResNet50—to detect RPV installations with high precision. Through weighted majority voting, the ensemble model achieved superior accuracy, precision, recall, and F1-score compared to individual models, enhancing the reliability of geospatial mapping under variable urban and weather conditions. To estimate photovoltaic yield, we integrated spatiotemporal solar irradiance data with surface meteorological measurements, including ambient temperature, wind speed, and humidity, yielding an annual RPV potential of 602.83 GWh for Karachi, with a potential reduction of 0.37 megatons of CO 2 equivalent. This GeoAI-driven framework serves as a foundational layer for an Urban Digital Twin, enabling dynamic modelling of RPV potential to support sustainable urban energy transitions. Additionally, we evaluated the economic feasibility of RPV deployment using levelized cost of electricity (LCOE) analysis. By advancing image processing and big spatiotemporal data analytics with GeoAI, this study provides a scalable framework for inventorying urban renewable energy resources, offering actionable insights for governance and sustainable land use planning.