In her famous book "The Death and Life of Great American Cities", Jane Jacobs defined vitality as continuous pedestrian activity in a city. She argued that it is a crucial indicator of thriving cities. In their paper "Jane Jacobs in the sky: predicting urban vitality using open satellite data", Bell Labs researchers have developed a methodology to infer vitality of cities using openly available satellite data, and systematically show that the inferred vitality indeed maps onto particularly relevant urban features on the ground. They validate their method using satellite images from 6 culturally and geographically diverse Italian cities, which for the first time, validates Jane Jacob's theory at such scales.
To do so, they first built a deep learning framework (based on convolutional neural networks, and convolutional auto encoders) that is able to extract relevant features from satellite views of urban districts (from imagelets cropped from large Sentinel-2 images) and infer which urban elements make a particular urban environment more or less vital. Their findings are aligned with Jane Jacob’s theories establishing a link between vitality and the diversity of land use, small block sizes, a mix of economic activities, and the concentration of people. By juxtaposing their model's inference with the points of interest (PoI) on the ground, they validated that the model indeed learns latent features about presence of certain vital points of interest, such as transportation hubs, entertainment and sustenance joints.
By basing this work on openly available satellite data, such as Sentinel, they show that we can systematically derive insights about the built environment (such as vitality) with a performance at par with methods based on other access controlled and expensive datasets. This is notwithstanding the differences in spatial resolution between Sentinel and other datasets. These results could further empower the community to perform follow-up studies towards better urban planning and development, particularly around the developing parts of the world, which often lack other types of data.
