Predicting walking-to-work using street-level imagery and deep learning in seven Canadian cities.

New 'big data' streams such as street-level imagery are offering unprecedented possibilities for developing health-relevant data on the urban environment. Urban environmental features derived from street-level imagery have been used to assess pedestrian-friendly neighbourhood design and to predict active commuting, but few such studies have been conducted in Canada. Using 1.15 million Google Street View (GSV) images in seven Canadian cities, we applied image segmentation and object detection computer vision methods to extract data on persons, bicycles, buildings, sidewalks, open sky (without trees or buildings), and vegetation at postal codes. The associations between urban features and walk-to-work rates obtained from the Canadian Census were assessed. We also assessed how GSV-derived urban features perform in predicting walk-to-work rates relative to more widely used walkability measures. Results showed that features derived from street-level images are better able to predict the percent of people walking to work as their primary mode of transportation compared to data derived from traditional walkability metrics. Given the increasing coverage of street-level imagery around the world, there is considerable potential for machine learning and computer vision to help researchers study patterns of active transportation and other health-related behaviours and exposures.

Comparing the Normalized Difference Vegetation Index with the Google Street View Measure of Vegetation to Assess Associations between Greenness, Walkability, Recreational Physical Activity, and Health in Ottawa, Canada.

The manner in which features of the built environment, such as walkability and greenness, impact participation in recreational activities and health are complex. We analyzed survey data provided by 282 Ottawa adults in 2016. The survey collected information on participation in recreational physical activities by season, and whether these activities were performed within participants' neighbourhoods. The SF-12 instrument was used to characterize their overall mental and physical health. Measures of active living environment, and the satellite derived Normalized Difference Vegetation Index (NDVI) and Google Street View (GSV) greenness indices were assigned to participants' residential addresses. Logistic regression and least squares regression were used to characterize associations between these measures and recreational physical activity, and self-reported health. The NDVI was not associated with participation in recreational activities in either the winter or summer, or physical or mental health. In contrast, the GSV was positively associated with participation in recreational activities during the summer. Specifically, those in the highest quartile spent, on average, 5.4 more hours weekly on recreational physical activities relative to those in the lowest quartile (p = 0.01). Active living environments were associated with increased utilitarian walking, and reduced reliance on use of motor vehicles. Our findings provide support for the hypothesis that neighbourhood greenness may play an important role in promoting participation in recreational physical activity during the summer.

The Canadian Urban Environmental Health Research Consortium - a protocol for building a national environmental exposure data platform for integrated analyses of urban form and health.

BACKGROUND: Multiple external environmental exposures related to residential location and urban form including, air pollutants, noise, greenness, and walkability have been linked to health impacts or benefits. The Canadian Urban Environmental Health Research Consortium (CANUE) was established to facilitate the linkage of extensive geospatial exposure data to existing Canadian cohorts and administrative health data holdings. We hypothesize that this linkage will enable investigators to test a variety of their own hypotheses related to the interdependent associations of built environment features with diverse health outcomes encompassed by the cohorts and administrative data. METHODS: We developed a protocol for compiling measures of built environment features that quantify exposure; vary spatially on the urban and suburban scale; and can be modified through changes in policy or individual behaviour to benefit health. These measures fall into six domains: air quality, noise, greenness, weather/climate, and transportation and neighbourhood factors; and will be indexed to six-digit postal codes to facilitate merging with health databases. Initial efforts focus on existing data and include estimates of air pollutants, greenness, temperature extremes, and neighbourhood walkability and socioeconomic characteristics. Key gaps will be addressed for noise exposure, with a new national model being developed, and for transportation-related exposures, with detailed estimates of truck volumes and diesel emissions now underway in selected cities. Improvements to existing exposure estimates are planned, primarily by increasing temporal and/or spatial resolution given new satellite-based sensors and more detailed national air quality modelling. Novel metrics are also planned for walkability and food environments, green space access and function and life-long climate-related exposures based on local climate zones. Critical challenges exist, for example, the quantity and quality of input data to many of the models and metrics has changed over time, making it difficult to develop and validate historical exposures. DISCUSSION: CANUE represents a unique effort to coordinate and leverage substantial research investments and will enable a more focused effort on filling gaps in exposure information, improving the range of exposures quantified, their precision and mechanistic relevance to health. Epidemiological studies may be better able to explore the common theme of urban form and health in an integrated manner, ultimately contributing new knowledge informing policies that enhance healthy urban living.