The rapid deployment of decentralized power generation in urban areas, through solar photovoltaic (PV) technology on building rooftops, calls for a comprehensive database with locations and sizes of existing solar installations in urban areas. Machine Learning approaches based on Convolutional Neural Networks combined with satellite and aerial imagery can be used to establish an up to date spatial mapping of the existing PV installations and their sizes. Local environmental and socio-economic factors can then be used as features to model and to predict future solar technology adoption in urban areas.