Crowdsourcing and Machine Learning for Disaster Relief and Resilience

This project builds on a strong history of successful, impactful STFC-supported research, applying this research within the world-leading Zooniverse citizen science platform to humanitarian and disaster management issues in countries that require Official Development Assistance. The Planetary Response Network is a partnership led by the Zooniverse, the Machine Learning Group at the University of Oxford, and the response and resilience charity Rescue Global. Since 2015 the PRN has deployed crowdsourcing projects to classify multiple kinds of damage following major natural disasters in Nepal, Ecuador, and multiple Caribbean nations including Dominica and Antigua & Barbuda. This project seeks to improve on the successes of those projects by incorporating feedback from ground-responders partnered with Rescue Global and from a recent multi-agency report which clearly articulated the unique needs of crowdsourced projects in humanitarian response applications. Thanks to STFC support, the Zooniverse has well-established platform infrastructure that can fully address these needs; the modest additional support requested in this project will bring high value for money by adding targeted high-impact features to the Zooniverse platform. These features include a pipeline to rapidly process pre- and post-event satellite images into classifiable "subjects" for the crowd, application of STFC-supported machine learning research to pre-classification of images, incorporation of STFC-supported advanced algorithms for real-time human-machine classification, and intuitive visualisation of consensus results so that decision makers and responders on the ground can easily interpret damage maps and maximise situational awareness, leading to better allocation of resources and aid, faster restoration of infrastructure, and a significant positive impact on societies preparing for and recovering from natural disasters.

This project benefits both academic researchers and society as a whole by building on the products of past STFC research and applying these products to the humanitarian sphere. The deployments enabled by this project are intended to provide rapid, accurate, and ongoing high-value information about evolving conditions before, during, and after a major crisis, especially in countries that require Official Development Assistance. Local, regional and national responders and decision makers benefit from the significantly improved risk awareness and situational awareness provided by the Planetary Response Network damage maps. The additional information can then be used to more efficiently assess risk, allocate aid and resources, and both preserve and more rapidly repair infrastructure. These actions can save lives, and have additional benefits such as helping the local economy recover more quickly after a disaster.

Classifications produced by the Zooniverse have proven to be of high value to the wider academic community in a variety of disciplines; this is anticipated to continue with this project. In particular, machine learning researchers may find both the raw classifications and the processed consensus damage maps extremely useful in pursuing further advances in human-machine computation problems.

Additionally, the project's results will be directly applicable to researchers in other fields using geo-tagged data, especially those will benefit from rapid, real-time classification of geo-tagged data, such as conservationists. As the software work in this project will be added to the open-source Zooniverse platform, researchers may use it to build crowdsourcing projects that capitalise directly on the tools created in this project. The open availability of these tools within the public Zooniverse platform maximises the long-term impact of the project.