Advancing urban solar assessment: A deep learning and atmospheric modelling framework for quantifying PV yield and carbon reduction

Abstract

This study investigates the potential of rooftop photovoltaic (PV) systems to contribute to sustainable urban energy transitions in Lahore, Pakistan. While solar energy offers a significant opportunity to reduce urban carbon emissions, accurately estimating rooftop PV potential is challenged by complex atmospheric conditions and limitations in existing solar irradiance models. To address these challenges, this study proposes a novel framework combining atmospheric radiative transfer modelling with deep learning. High-resolution satellite imagery and optimized deep learning models are leveraged to precisely delineate existing solar panel installations across Lahore. Solar insolation on existing and potential rooftop PV installations is estimated by incorporating diurnal cloud fraction, cloud optical depth, and aerosol optical depth data to refine transmittance and diffuse fraction calculations, moving beyond simplified atmospheric parameterizations. Our analysis reveals that existing PV systems in Lahore generated 373 GWh of electricity in 2023, mitigating 0.23 Mt CO2e. However, by harnessing the full rooftop PV potential, estimated at 10,877 GWh, Lahore could reduce its carbon emissions by 6.74 Mt CO2e annually. This study identifies key areas within Lahore with the highest potential for rooftop PV development, providing valuable insights for urban planners and policymakers seeking to integrate solar energy into building infrastructure and advance towards net-zero goals.