Built Environment Features Associated with Street Scape Attractiveness in 56 Cities across 28 Countries

Abstract

The built environment influences human health through several pathways, including environmental exposures and physical activity. Psychological pathway are also important but are not well understood given the difficulty in measuring related constructs (e.g. built environment attractiveness) and the specific components of built environments that contribute to these constructs (e.g. trees, density, mixed land use, etc.).We leveraged a novel crowdsourced data set (MIT Pulse Place) that included over 1.5 million pairwise comparisons of 424,929 Google street view images from 56 Cities across 28 Countries where individuals selected the image that was more beautiful. From these ranking we calculated a q-score for each image that ranged from 0 (lowest attractiveness) to 10 (highest attractiveness). We then derived multiple estimates of urban composition for each image to identify specific components of the built environment that are associated with streetscape attractiveness. We utilized a feature extraction deep learning algorithm to derive estimates of visible percent grass, trees, cars, sidewalks, roads, bushes, persons, houses, and sky from each image. Satellite-based estimates of road density, annual maximum NDVI, percent tree cover, population density, impervious surface area, and annual mean NO2 and PM2.5 were also derived from 100 and 250 meter radius buffers around the street view image locations.Preliminary results indicate that urban green space, especially tree canopy cover, is a particularly important driver of streetscape attractiveness. We are currently utilizing correlation globes, lasso variable selection, and regression modeling to further disentangle the relationships between built environment features, environmental exposures and attractiveness perceptions. A better understanding of these relationships will be used to inform future built environment health studies as well as policy and planning principals.