Assessing recreationists’ preferences of the landscape and species using crowdsourced images and machine learning

Abstract

Crowdsourced data are now well-established for assessing cultural ecosystem services (CES). In rural areas, understanding which land covers people prefer to recreate in, and how these land covers provide different CES, is necessary to support sustainable use. In this study, we aim to assess recreationists’ revealed preferences of landscape aesthetics and species observation as two CES, considering multiple land cover types in a rural area in France. This assessment was carried out using georeferenced images from two crowdsourced sources (Flickr and Wikiloc) and by analyzing their content using a machine-learning algorithm. We further developed a framework to classify images based on their content into CES-related images (those depicting landscapes or species) and non-CES-related images. Finally, we assessed how images depicting landscape aesthetics and species observation are distributed across the land covers visited by recreationists, and which species groups are the most photographed. Our results showed the dominance of images of open landscape views over close-up species images, and that grasslands are the primary providers of open views. In addition, we found that forests also provide open landscape views, suggesting that forests with gaps in canopy cover and viewpoints can be as important as grasslands in providing aesthetic views, especially in hilly landscapes. For species, the category “plants and flowers“ was the most photographed, followed by invertebrates and birds on Flickr, and domestic livestock on Wikiloc. This study provides insights into the importance of using multiple crowdsourced sources in CES assessment, providing critical insights for both landscape managers and conservationists.