Discovering inequality in green space exposure from street view images: a case study in 25 small- and medium-sized cities in the U.S.

Keywords: Green space exposure, Inequality, Street view imagery, Deep learning, Small- and medium-sized cities
Abstract Type: Paper Abstract

Abstract

While green space exposure has been considered essential for both physical and mental health of people, not every population is experiencing the same level of exposure in their daily lives. With the increase in the availability of various spatial data, green space exposure in the urban environment has been extensively explored in respect to the socioeconomic background of cities. Yet, we find research gap in that small- and medium-sized cities have been understudied as data in these sites remain relatively scarce. In this research, we take 25 small- and medium-sized U.S. cities as study sites, from which Google Street View (GSV) images were collected along 45,188 walk-based commute trajectories retrieved from Census Transportation Planning Products (CTPP) data. We applied deep-learning based semantic segmentation technique to GSV images in order to measure the level of green exposure along individual commute trips. As a result, our correlation results showed positive and negative associations with the income level and non-white population percentage of each site, respectively, which reveal the inequality in green space exposure based on socioeconomic status. Furthermore, the contrast between the distribution of home- and mobility-based green exposure levels provide evidence of the "neighborhood effect averaging" problem, which states that individual exposure level tends to converge toward the average level of the population when human mobility is considered. This study contributes to demonstrating the research potential of street-view imagery analytics in small- and medium-sized cities and raising awareness to the unequal provision of green spaces in daily trips in these areas.