Using social media photos and computer vision to assess cultural ecosystem services and landscape features in urban parks

Abstract

Urban parks are important public places that provide an opportunity for city dwellers to interact with nature. In recent years, social media data have become a promising data source for the assessment of cultural ecosystem services (CES) and landscape features in urban parks. However, it is a challenging task to identify and classify the CES and landscape features from social media photos by manual content analysis. In addition, relatively few studies focused on the differences in landscape preferences between tourists and locals in urban parks. In this study, we used geotagged social media photos from Flickr and computer vision methods (scene recognition, image clustering and image labeling) based on the convolutional neural networks (CNN) and the Google Cloud Vision platform to assess the spatial preferences and landscape preferences (cultural ecosystem services and landscape features) of tourists and locals in the urban parks of Brussels. The spatial analysis results showed that the tourists’ photos were spatially concentrated on well-known parks located in the city center while the locals’ photos were rather spatially dispersed across all parks of the city. We identified 10 main landscape themes (corresponding to 4 CES categories and 10 landscape feature categories) from 20 image clusters by automated image analysis on social media photos. We also noticed that tourists paid more attention to the place identity featured by symbolic sculptures and buildings, while locals showed more interest in local species of plants, flowers, insects, birds, and animals. This research contributes to social media-based user preferences analysis and CES assessment, which could provide insights for urban park planning and tourism management.