Urban Recreational Ecosystem Services Investigation based on Social Media Images

Abstract

Recreational ecosystem services (RES) are understood as the benefits that people derive from landscapes and natural environments through recreational activities. The growing social media datasets have contributed to overcoming limitations of spatial and temporal coverage for RES studies that traditional survey-based approaches have. Related RES research using social media such as photo-sharing platforms has primarily focused on natural and ecological areas outside cities at regional or national scale and utilized geotagged photographs as reliable proxies for empirical access rates. The urban dimension of RES is under-explored, and potential information about the environmental composition and user preferences in photos is overlooked. Using data retrieved from the photo-sharing platform Flickr, we explore the potential role of computer vision (CV) in understanding RES related to environmental composition and human activities. After that, we assess RES for the urban outdoor environment of Ann Arbor. Specifically, by manual validation of recognition results for 1,500 Flickr photographs, we evaluate whether scene recognition algorithms and models pre-trained with three different labeling systems on a standard CV dataset can be applied to tackle complex visual tasks in realistic urban scenarios. Contrary to consistent outstanding performance on standard CV datasets, we find substantial changes in the performance of recognizing physical environmental composition and human activities depending on the semantic scale the model uses for labeling. Via recognition results, we further study people’s preferences for environmental composition and outdoor activities and their associations, then detect popular RES places for different

recreational usages in Ann Arbor with a high spatial resolution. This article concludes with the feasibility of applying pre-trained CV models for urban RES studying. Time and resource permitting, future studies should consider combining information from multiple sources for a more accurate evaluation of RES characteristics, thus can be integrated with decision making, planning, and management to enhance city planning and human well-being.