Press Release: Data Quest First Results
USING AI AND SATELLITE DATA TO TACKLE BUSHFIRES – FIRST RESULTS
The Bushfires Data Quest 2020 is live-streaming a showcase on how bushfire, machine learning and Earth-observation researchers can help solve the bushfire crisis (Wed 16 Sept 2.30pm http://dataquest.live). As the 2020/21 fire season approaches in Australia and New Zealand, and fires ravage the US West Coast, we are all looking for new ways to understand and respond to the threat. Four interdisciplinary teams are revealing their results from an intensive research sprint, applying modern data-science and artificial intelligence (AI) techniques to mitigate the effect of bushfires.
1. Can we use AI and satellite data to improve the resolution of Australia’s fuel moisture? Yes we can. The team proved that, in theory, we can develop maps nearly 50 times more detailed and update them dynamically every five days.
A closely related question is: can we predict the fire risk or ‘burnability’ of the landscape? The team built an AI system that creates a fire risk map based solely on satellite data. This fine-scaled (20m) map could be used to identify the most dangerous areas and triage fuel-removal efforts. By incorporating additional information on terrain and weather, the same system can predict the spread of the fire from an ignition point. Usually this type of modelling requires a supercomputer, but the team showed that AI models can make similar predictions on a laptop in seconds. Such tools in the hands of first responders would enable faster, more tactical and effective fire-fighting.
2. Can we use AI and geostationary satellites (which sit over Australia and deliver an ‘always on’ view) to provide early warning on fire ignition? This is also possible. The team developed an initial test that demonstrated near-real-time fire-warnings could come from space in the future. This demonstration has so far only been tested on historical data, but the workflow is very promising.
3. Once we have an ignition, can we help fire-fighters and communities understand the evolution of the bushfire? This challenge is a bold vision and the physics and co-factors of bushfire progress are complex. However, the team showed that an AI system could match the performance of models on supercomputers - which are stretched thin during fire season. This opens up the possibility of more localised prediction capabilities, meaning fast and accurate prediction tools in the hands of first responders, and local decisions makers. Such tools offer the potential of more tactical and effective response.
4. Can we understand the level of fire threat more effectively by tracking smoke and cloud patterns? The last result looked at the behaviour of smoke haze and fire clouds - pyrocumulonimbus - which are also visible from space and detectable with AI. Changes to the colour and behaviour of smoke can indicate the fire becoming ‘extreme’: the fire front accelerating outwards and sending up pyrocumulonimbus clouds that create their own storm weather. Detection of this phase-shift from space and the ability to warn fire-fighters and communities on the ground could also be a lifesaver.
The goal of the Bushfire Data Quest was to demonstrate ‘solvability’. During the short research sprint, the teams aimed to show key challenges are solvable by applying artificial intelligence (AI) and space data to bushfire science. The results of the Data Quest demonstrate paths to deployable tools for use by fire managers.
Dr Cormac Purcell, Program Director at Trillium Technologies, which created Data Quest explained what’s next for these results.
“These AI methods have potential to deliver insight faster and with great accuracy when properly deployed, as they can continually learn from new data. The next step is to develop prototypes and run controlled validation tests with stakeholder involvement from across the science and bushfire community”.
“I am so proud of what the teams have been able to achieve. I am excited and deeply affected by the potential for applied research of this kind to make significant and useful contributions to the lives of individuals and communities across Australia and New Zealand. Data Quest may be based in scientific inquiry, but its success is built on a deeply human contribution as everyone involved applied themselves to helping tackle the threat to all our lives. I would like to thank all our partners, researchers, faculty and collaborators for the immense combined effort. We showed we truly are better together.”
Adrian Turner CEO of Minderoo Foundation’s Fire and Flood Resilience initiative and the Challenge Partner of the Bushfire Data Quest 2020 described how the Data Quest’s results fit into Minderoo Foundation’s moon-shot vision to reduce harm caused by fire and floods across Australia.
“The Data Quest results have shown how much can be achieved over a short period when bringing brilliant minds to collaborate on difficult challenges. We applaud and welcome the progress made, which fits with Minderoo Foundation’s Fire Shield mission and its goal of ensuring bushfires don’t become disasters”, said Mr Turner
“Fire Shield aims to reduce the scale and impact of bushfires by finding new methods to detect, monitor and extinguish dangerous blazes within an hour, anywhere in Australia, by 2025. The work of the Data Quest 2020 teams has shown real potential to contribute to this goal”.
In his remarks to open the show case, NSW Deputy Premier John Barilo, welcomed the progress.
"In a country with a high bushfire risk, satellite images and data are increasingly playing a vital role in protecting our communities and environment.”
“Not only do satellites help detect bushfires, they allow us to predict their movement and assess the damage they cause.”
“The NSW Government is proud to support the Bushfire Data Quest and welcome the unique solutions developed to manage and prevent bushfire events”
Artificial Intelligence is particularly useful for detection and prediction. Once trained, the systems are able to maintain vigilance over enormous data volumes and make accurate predictions on future events, based on observations past examples. AI systems can also incorporate new data as it becomes available, becoming more robust over time. While training takes time, making a prediction usually takes only seconds. This is why the fusion of geospatial data from space and artificial intelligence holds so much promise.
During the 2020 Data Quest the potential of space data and AI for bushfire defence was explored across the following challenge areas (1) Fuel assessment - in other words, the dryness and characteristics of the bush itself, which informs both the likelihood of a destructive fire and its progress. (2) Early detection - crucial for rapid response and triage. (3) Fire behaviour - how a fire evolves is critical for co-ordinated response planning.
Data Quest 2020 is the inaugural event in Australia and is based on a successful formula used for interdisciplinary research in the United States and Europe.