Climate Change CycleGAN

In 2019, researchers from Montreal Institute for Learning Algorithms (MILA), Microsoft, and ConscientAI Labs (a Sri Lanka based tech start-up) developed a tool that enables people to visualize potential future impacts of flooding due to climate change on their own home and community. The initial prototype uses an application of Machine Learning, called CycleGAN, which stands for Cycle Generative Adversarial Network and refers to a method by which a computer can modify an image’s appearance in an efficient and accurate way. In the past, a computer could learn to translate an image only by being fed numerous paired images to train the model. For example, it would have required many photos of locations both before and during a flooding event, which are rare if not impossible to find. The CycleGAN allows the computer to translate images based on a few unpaired images (e.g. a photo of a house and an image of a flood at a different location). This technique allows for any photograph to automatically be translated into a predicted flood scene, showing how a person’s house might look in 2050 based on existing climate models. This tool is still at its prototype stage and it has only successfully been employed for visualizing single-family homes surrounded by lawns. While the tool aims to raise awareness about the reality of climate change and the importance of acting to prevent its impacts, it runs the risk of discouraging users about the fate of their neighbourhood and opting to leave, if they have the choice. In the future, the team hopes to expand the prototype to incorporate the potential impacts of other climate events such as droughts and floods. The researchers also plan to enable users to experiment with different lifestyle decision-making scenarios in order to visualize how the outcome may change as a result of their actions.

Schmidt, V., Luccioni, A., Mukkavilli, S. K., Balasooriya, N., Sankaran, K., Chayes, J., & Bengio, Y. (2019). “Visualizing the consequences of climate change using cycle-consistent adversarial networks“.

Brownlee, J. (2019, August 16). “A Gentle Introduction to CycleGAN for Image Translation“.


Citizen Science, Climate Change, Data, Ecological Modelling, Ecological Monitoring, Industry/Natural Commodities, Regulation