The Research
Our methodology was developed at the University of Warwick and the Alan Turing Institute, then published in the peer-reviewed journal Royal Society Open Science in 2017. A related study, "Happiness is Greater in More Scenic Locations," was published in Scientific Reports (Nature Publishing Group) in 2019.

The original research analysed over 200,000 images of Great Britain, rated for scenicness 1.5 million times by tens of thousands of people through the online game Scenic-Or-Not.

How It Works
Recognising the potential of Convolutional Neural Networks (CNNs) in image tasks, we employed them to evaluate the beauty of various locations. Our goal was to have the CNN evaluate the aesthetic appeal of the environment captured in the photos, not just the photos themselves.

To enhance our model's performance, we utilised transfer learning to modify an established model, the MIT Places365 CNN, to better understand and rate scenic beauty.

What We Discovered
Our research revealed that beauty is more nuanced than "natural is beautiful." As well as natural features like coastlines, mountains, and canals, man-made structures such as castles, towers, viaducts, and historic cottages lead to places being considered more scenic. Importantly, bland green features like flat grass and athletic fields scored lower than characterful architecture.

In practical tests on London imagery, our model identified beauty in natural settings like parks and was equally adept at recognising the splendour of architectural landmarks.

Continuous Improvement
Building on this foundation, we continue to refine and strengthen our algorithm, with particular focus on enhancing its accuracy across diverse international landscapes and complex urban environments. Every rating contributed helps improve the model further.

For more information on this methodology, refer to our study: Using Deep Learning to Quantify the Beauty of Outdoor Places.