Promises and pitfalls of using computer vision to make inferences about landscape preferences: Evidence from an urban-proximate park system

Abstract

The ubiquitous use of the internet and social media has provided social and spatial scientists with a wealth of data from which inferences about landscape preferences can be gained. These data are increasingly being used as an alternative to data collected from surveys of recreationists. While the rapidly growing body of research using social media is impressive, little work has been done to compare the image content of social media to preferences elucidated via more traditional methods. We compare the landscape features derived through a computer vision algorithm used to analyze social media photographs with preferences derived through a traditional on-site intercept survey. We found that landscape features identified through the computer vision algorithm were, by and large, significantly different compared to landscape features that park users said improved their recreational experiences. Additionally, we did not find substantial differences in landscape preferences between visitors who share photographs of their park visit on social media and those who do not. We suggest a diversity of data sources and analytical methods should be used in a complementary and comparative way. Our analysis here suggests both surveys and social media images can provide important insights about landscape preferences, but neither in isolation is perfect.