Introduction

Situation awareness is key to mitigating and responding to disasters to help save life and protect the natural environment. An abundance of data is available; from satellites to tweets, from UAVs to in-situ sensors, from expert knowledge to mathematical hazard models, and making sense of all this in a timely, accurate manner is critical to effective disaster management. By developing novel machine learning approaches to data fusion this project aims to combine all this data together, remove unreliable data and identify informative sources.

Explaining the science

Disaster management

The work is developing technology that exploits recent advances in crowdsourcing, Bayesian classifier combination, deep learning, heterogeneous data fusion, and scalable inference, to provide products to be used for disaster policy, planning, and response.

Also developing foundational, hybrid Bayesian deep learning and classifier combination approaches to exploit human-agent collectives, in order to interpret satellite imagery for applications as far ranging as illegal logging and micro-insurance.

Project aims

Using novel machine learning approaches to data fusion to generate maps and reports of where help is needed, what help is needed, and where help is available. This information is passed to first responders, NGOs (e.g. insurance providers), National Disaster Management Agencies, farmers, and others affected by the disaster to help formulate policy and plan response effectively, efficiently, and accurately.

The project involves working closely with both industry and end-users, folding in UK expertise and in-country partners to co-design and deliver effective situation awareness technology in the form of dashboards, apps, and other related outputs.

Applications

The project's work is currently being developed for use in Malaysia, Ethiopia, and Kenya through UK Space Agency International Partnership Programme projects with Airbus and the Satellite Applications Catapult. The work in Malaysia is in collaboration with flood hazard modellers there to improve the accuracy of their models.

Recent updates

May 2019

University of Oxford publishes update on the project's impact. The technology produced in the project is being opened up to other applications:

  • Working with the charity Save the Elephants, using satellite images to help resolve human-elephant conflicts in Kenya
  • Working with the Brazilian government to identify the location of dams in the country, many of which are built by private companies without the government’s knowledge
  • Working with earth scientists to identify seismic faults which are extremely hard to spot on the ground, but which are responsible for a large number of medium-sized earthquakes that cause substantial damage

2018

Previous work, which mapped settlements using satellite imagery interpreted through crowdsourcing following the Nepal Earthquake in 2015, was used by disaster responders to target the delivery of aid and was featured at the World Economic Forum in Davos.

Organisers

Dr Steven Reece

Senior Research Associate, Machine Learning Research Group, University of Oxford

Contact info

[email protected]

Funders

Collaborators