Assessment of rooftop solar photovoltaic potential through spatial analysis and implementation of machine learning techniques

The necessity to reduce greenhouse gas emissions to meet the global warming regulations has increased the demand for renewable energy sources; unlike power generation from fossil fuels, renewables have relatively low geographic density and they’re unpredictable. Solar energy deployment is gaining greater attention and it’s a technically and economically feasible solution as a sustainable alternative that might alleviate aspects of the current climate crisis, especially in cities, where it is crucial to promote the use of solar technologies. The roof surfaces within urban areas are constantly attracting interest as they supply huge potentials for the mitigation strategy to minimize the environmental impact by achieving the sustainable development goals. The local generation of renewable electricity through roof-mounted photovoltaic (PV) systems on buildings in urban areas can play a significant role within the transition to a low-carbon energy system as the resulting large-scale deployment is quite straightforward once the methodology of the solar energy potential assessment has been developed. Unfortunately, some roof surfaces are unsuitable for installing photovoltaic systems, in fact one of the major challenges today is to evaluate the suitability of PV systems’ installations on buildings’ roofs, as they are generally lowered by superstructure. To date, the lack of high-resolution data and also the large uncertainties related to existing processing methods impede the accurate estimation for measuring the rooftop solar energy potential over a heterogeneous urban environment containing flat and pitched roof surfaces at different slopes and directions; the idea is to rate the roof surfaces based on their solar potential and suitability for the installation of photovoltaic systems. In this thesis work, we will try to address these issues and therefore the gap between existing methodology and technological development through remote-sensing data and the implementation of Machine Learning (ML) algorithms to simplify the intricate topography of cities. This technique is based on aerial images that are analyzed using image recognition, Geographic Information Systems to estimate the rooftop photovoltaic potential of buildings in an urban environment, the city of Aosta, Italy. ML techniques, combined with satellite images, allow to overcome the constraints of lack of information in providing this mapping at large scale. The scalability of the trained model allows to predict the existing solar panels deployment at the Italian national scale which might be able to extract predictive models in urban areas and may be a valuable input for policymakers and for investing in distributed energy infrastructures. The methodology, however, is generalizable to any region where similar data is available and could therefore be useful for researchers, energy service companies and municipalities to assess the rooftop PV capacity of the region.