Estimating the solar rooftop potential of buildings' rooftops at a large scale is a fundamental step for every country to utilize its solar power efficiently. However, such estimation becomes time-consuming and costly if done through on-site measurements. This paper uses deep learning-based multi-class instance segmentation to extract buildings' footprints from satellite images. Hence, we introduce Lebanon's first complete and comprehensive buildings' footprints map. Furthermore, we propose a photovoltaic panels placement algorithm to estimate the solar potential of every rooftop, which results in Lebanon's first buildings' solar rooftop potential map too. Finally, we report total and average solar rooftop potential per district and localize regions corresponding to the highest solar rooftop potential yield. Conducted analysis reveal solar rooftop potential urban patterns and provide policymakers and key stakeholders with tangible insights.
View on arXiv