A crowdsource approach to documenting users' preferences for landscape attributes in the proposed Appalachian Geopark Project in West Virginia, United States

Abstract

This study demonstrated the potential utility of crowdsource photographs to investigate users' (residents and visitors) preferences for landscape features in the proposed Appalachian Geopark Project in southern West Virginia, United States, assuming that a photographic choice implies a preference. The study used photographs from Flickr crowdsource. Our method combined existing technologies of crowdsourcing, geographic information system (GIS), machine learning (ML) and added a new metrics generation capability to provide a novel approach to classifying users' preferences. First, spatial distribution of the photographs was assessed. Second, the amount of area in pixels covered by each feature was calculated to quantify the different landscape features contained in the photographs. The results revealed that the photographs were congregated in specific locations of the study region showing clustered patterns of distribution. The content analysis revealed that the forest was the predominant landscape feature, followed by rock, antrhopogenic (anthro), sky, water, grass and road that were captured in the photographs. This approach is an indirect approach that help understand what landscape features are captured in the photographs by the users. The ability to provide statistics for pixel-by-pixel classification in the ML output represents a new functionality that can be useful in other studies such as in a single landscape type (e.g., urban, agriculture, etc.) and temporal data (e.g., date taken).