DEEP LEARNING-ENABLED ASSESSMENT OF ROOFTOP SOLAR POTENTIAL: INSIGHTS FOR SUSTAINABLE URBAN ENERGY PLANNING IN PAKISTAN

This study presents a novel deep learning and remote sensing framework to accurately evaluate both current and prospective rooftop solar photovoltaic (PV) potential in urban settings, with a focus on Islamabad and Lahore, Pakistan. Utilizing high-resolution Google Earth satellite imagery, we employ optimized deep learning models-Mask R-CNN and DeepLab variants-trained on custom datasets to precisely detect existing solar panels and identify suitable rooftops. Our results indicate over 85,000 solar arrays currently producing approximately 140 GWh in Islamabad and 370 GWh in Lahore annually, while untapped rooftop potential could yield 5,502 GWh and 10,815 GWh, respectively, revealing substantial scope for expansion. This dual-purpose approach, tailored to Pakistan’s unique urban morphology, offers policymakers and investors critical insights for strategic planning and investment in solar infrastructure, advancing sustainable energy development.