Why do we remember summers of our youth as warmer, and the skies bluer? Despite our nostalgic view of the past, climate is getting warmer not cooler. Old photos hold clues to past climate in the clouds and the blue tones of the sky. In 2014 people uploaded an average of 1.8 billion digital images every day[1]. That’s 657 billion photos per year, or put another way every two minutes we create more pictures than in the previous 150 years combined. That’s a lot of sky (both blue and cloudy!) with which to create a unique climate archive, with a very personal twist.
Using this incredible and growing archive we have launched a project that aims to unlock the climate record in these images and to make it personal by allowing you to create your own climate history from the images stored on your phone, or in a dusty shoebox. So our photo algorithms use state-of-the-art deep learning techniques to extract a climate signal from images whether they are online, in your phone or even from the art in a gallery. Our outdoor experiences and memories as recorded in the photographs we take, or in the art we create, are influenced by the weather. We have developed tools to unlock this. Now your own images are location and time specific, but put them with millions of others and they become powerful, allowing us to reconstruct specific summers, winters or even temporal trends in different places or in time countries.
We are just starting out and have our first prototype running, which automatically classifies every single pixel in an image into cloud, sky or other and samples the blue tones. We are currently using measures of cloudiness and colour intensity within sky segments to get a record of the weather in an image, but we’re also experimenting with cloud textures and other automatically descriptors that the algorithms came up with on their own.
Why does it matter? Making historic and contemporary climate change real and tangible to the public is a priority to altering perceptions and behaviours about climate and our changing planet. It also provides a way of appreciated landscape art in a different way, the manifestation of climate in art has been well documented[2] and this project simply continues this trend bringing it up to date. It also showcases the potential of machine learning and big data to create new perspectives on old problems.
As we say we are starting out and will shortly launch a citizen-science project, via a dedicated website that will allow you to scan and upload your images and for the climate data to be extracted. We need your help to test safely within BU our algorithms and to collect some initial data. In short we need help to break our system before we launch it on the world! The system also learns and the more data we can get at this initial stage the more it can learn – better to send an 16 year old kid out into the world than one with the learning of a 4 year old! So put another way we would like your help in educating our algorithms.
You can access our site on campus only at: http://dec251.bmth.ac.uk/
Go online and upload some of your pictures and give it a try. We are particularly interested in pictures from Bournemouth/Poole over the last twenty years.
Marcin Budka, Bastian Fraune, Sally Reynolds, Matthew Bennett
[1] According to Mary Meeker’s annual Internet Trends report
[2] Bonacina, L.C.W., 1939. Landscape meteorology and its reflection in art and literature. Quarterly Journal of the Royal Meteorological Society 65, 485-498.
Neuberger, H., 1970. Climate in art. Weather 25, 46-56.
Thornes, J.E., 2000. A brief history of weather in European landscape art. Weather 55, 363-375.
Robinson, P.J., 2005. Ice and snow in paintings of Little Ice Age winters. Weather 60, 37-41.
Thinking differently about the past to try to see a brighter future: The role of counterfactual thinking in depression.
Conversation article: London Marathon – how visually impaired people run
Horizon Europe News – December 2023