From landscapes to cityscapes : quantifying the connection between scenic beauty and human wellbeing

Intuitively, we often seek out beautiful scenery when we want a respite from our busy lives, but do such settings actually help to boost our wellbeing? While architects, urban planners and policymakers have puzzled over this question for centuries, quantitative analyses have been held back by a lack of data on the beauty of our environment. However, the vast volumes of geotagged images readily shared on the Internet, alongside developments in computer vision and deep learning, are opening up opportunities to quantify aspects of the visual environment that were previously hard to measure. In the research reported here, we ask: might the beauty of outdoor environments have a quantifiable association with increased wellbeing? This thesis explores the following related strands of work: (1) How accurately can we automatically predict the beauty of scenes for which we do not have survey or crowdsourced scenicness data? (2) Is there a quantifiable connection between the beauty of the environment, as measured by scenicness, and people’s wellbeing? (3) Can we develop a broader understanding of what beautiful outdoor spaces are composed of? In the first strand, we investigate whether a deep learning model can be trained to automatically infer the scenicness of images. We find that a retrained convolutional neural network performs remarkably well, and that this network highlights not only natural but also built-up locations as being scenic. In the second strand, we explore the connection between beautiful scenery and different types of wellbeing: happiness, mental distress and life satisfaction. We find that individuals experience more happiness when visiting more scenic locations, even when we account for a range of factors such as weather conditions and the income of local inhabitants. However, in terms of mental distress and life satisfaction, we do not find evidence that individuals who live in more scenic locations report higher levels of wellbeing. In the third and final strand, we analyse crowdsourced data and discover that beautiful places are composed of natural features such as ‘Coast’, ‘Mountain’ and ‘Canal Natural’ as well as man-made structures such as ‘Tower’, ‘Castle’ and ‘Viaduct’. Importantly, while scenes containing ‘Trees’ tend to rate highly, places containing more bland natural green features such as ‘Grass’ and ‘Athletic Fields’ are considered less scenic. The research reported in this thesis takes an important step towards providing evidence that the beauty of the environment, and therefore decisions made about the design of environments, might have a crucial impact on people's everyday wellbeing. Our results also demonstrate that online data combined with neural networks can provide a deeper understanding of which environments humans might find beautiful