Spatial reach of visual information: Optimizing buffer distances for computer vision–based measures of microscale walkability and visual quality

An increasing number of studies use computer vision and street view images to evaluate how favorable a location is for pedestrians. Yet the spatial reach of visual information that informs pedestrian perceptions has received little systematic investigation, resulting in widely varying buffer distances across walkability studies. Furthermore, studies using these image-based metrics often overlook the possibility that optimal buffer distances may vary across different trip purposes and pedestrian characteristics (e.g., older adults), further exacerbating the uncertain geographic context problem of visual information. This paper examines the spatial reach of image-based walkability indicators by identifying the buffer distances at which image-based microscale and visual quality measures best explain walking environment satisfaction in Seoul, Korea. Walking environment satisfaction was measured through intercept surveys in which pedestrians rated their satisfaction with the surrounding environment for walking on a five-point scale. Findings show that optimal buffer distances for image-based microscale features generally range between 100m and 400m. For visual quality indicators, optimal buffers are 200m or less for most groups, which is significantly smaller than 1000-1400m distances found for GIS-based macroscale factors. Trip purposes and familiarity of the area result in distinctive patterns in contributing walkability factors and their optimal buffer distances. Visual quality measures offer substantial improvements to model fits if appropriate buffer distance is used. A higher macroscale index, greater building-to-street ratio, and safer-looking locations are associated with higher walking environment satisfaction. Livelier-looking locations showed negative relationship. Implications for policy and research are discussed.