How Australia is using AI, sensors, and more to tackle bushfires

The New South Wales and federal governments are both evaluating and helping develop new technologies to better detect, model, and communicate around devastating bushfires.

australia bushfires 2019 rfs
NSW Rural Fire Service

At the end of August 2020, the New South Wales government announced plans to use advanced technologies to accelerate the detection and response to prevent bushfires, following last summer’s fires that resulted in 33 lives lost—26 of those in NSW—the loss of more than 3,000 homes, 17 million hectares burned and the loss of a billion mammals, birds and reptiles across the country.

The NSW government is working with local and international organisations to investigate how it can use data from multiple satellites and local sensor networks to create algorithms that will help detect fires earlier, predict fire behaviour, and help emergency services respond more effectively to protect homes, people and nature.

Currently it counts on Spark, a modeling tool developed by CSIRO’s Data61 unit for simulating fire spread. It also counts on a combination of data from satellite imagery, machine learning through the use of structured data, and predictions, combined with data provided from first responders on the ground—which is still pen- and paper-based in most cases. There are some sensors in use, but there are also people who will climb up towers daily and watch for signs of smoke.

To advance the state of bushfire detection and analysis, the NSW government engaged the Minderoo Foundation as an enabler between the government and organisations such as IT companies and universities.

Minderoo Foundation Fire Fund CEO Adrian Turner said the organisation wants to drive a step change in how Australia deals with fires and floods and “lift Australia to be the global leader in fire and flood resilience by 2025”.

The foundation has organised a “working group” of 40 organisations that includes ASX-listed companies, insurers, banks, telecommunications providers. It has also partnered with the National Bushfire Recovery Agency, the Threatened Species Commissioner and the Australasian Fire and Emergency Service Authority (AFAC).

Limitations of current tech used for fire prevention

Turner explained to Computerworld Australia that one of the main limitations of the technologies currently in use is around the efficiency of the algorithms, which depends on the computational resources needed to do the modelling. This is done with the use of supercomputers but Turner expects that one day this will be possible from laptops.

Another limitation is the speed with how the information is updated and delivered to first responders and also those who can make decisions remotely and optimise where resources are placed and reduce harm. And lastly is accuracy from the models.

Turner explained:

The problem with models is you need good structured data to train the model. And then these fires were so intense that they burned with characteristics that we haven’t seen before. Because of the intense heat, because there’s so much fuel and the fuel load was so dry, it created its own mini weather system. 
So the existing tools that were in use for fire spread and prediction modelling didn’t take into account those characteristics. And some of the assumptions around for example, how fire spreads across the open paddocks. Some of those assumptions when correcting the models, they played out differently in reality.

And this is why Minderoo is working on getting a combination of structured data to train the algorithms with and ground truth data to be fed back so that the algorithms are constantly improving.

How can technology help with fire detection and prevention

The NSW government also released the final report of the NSW Bushfire Inquiry to help understand what happened during the 2019-20 bushfire season and how it was different from previous seasons. The report has 76 recommendations, both those for immediate action and those that need to begin now as they will take longer to complete.

“The total tally of fire generated thunderstorms in south-eastern Australia since the early 1980s increased from 60 at the end of 2018-19 to almost 90 at the end of the 2019-20 bush fire season—an increase of almost 50 percent in one bushfire season. Fire-generated thunderstorms are extremely dangerous phenomena that produce extreme winds, lightning, tornadoes and black hail,” stated the report.

The report recommends the government establishes a spatial technology acceleration program to maximise the information available from the various remote sensing technologies currently in use and to plan for inclusion of new remote-sensing systems that can sense precisely and rapidly through heavy smoke, cloud, fog and dust.

Turner said there is a lot of advancement to be made in sensor technology, and this technology is dependent on network connectivity, battery life and other considerations. But there is already emerging technology for sensing lightning strikes and determining whether that lightning strike it’s likely to cause fire ignition.

There is a need for improved telecommunications so the community can access information in a timely manner, as well as to improve fire-fighting capability. This includes several steps such as improving power backup arrangements, expanding fire information apps, and improving firefighter access to radio public safety networks.

The report also recommends the creation of a bushfire technology fund, an increase in the number of trained fire behaviour, and training of more meteorologists in fire behaviour so there are more expert resources available. It also recommends an accelerated roll-out of mobile data terminals into all fire-fighting vehicles to improve delivery of briefings and incident information and intelligence to field commanders.

“Over the longer term, some major changes are needed. We need to push available technologies harder, especially fire science, remote sensing, data science and artificial intelligence to equip us better to understand what happens during a bushfire and respond more quickly,” stated the report.

When it comes to drones and how they can be used, Minderoo’s Turner said drones will initially be likely used for surveillance. But there is a cost element as drones that can see through smoke and fly with winds of up to 100km/h have to be very rugged;  at the moment these mostly are military-grade drones mostly, and it is actually cheaper to fly an aeroplane.

Bushfire Data Quest investigates the use of AI to help fight fires

In early August, a group that included the federal and the NSW governments ran the Bushfire Data Quest — a week-long research sprint—to study if AI, data from satellites, and local sensor networks can detect bushfires early. Bushfire Data Quest saw machine learning specialists worked with bushfire researchers and leading data scientists.

“We think there is opportunity for further development in using things like computer vision and thermal imagery for detection of bushfires,” Turner said. “And it’s a stream of work that we’re very focused on. The detection response is all geared around new methods for sensing, new methods for making sure that information is shared to the right people at the right time to deliver situational awareness. And the Data Quest project is just the first step towards accelerating development of new techniques for situational awareness.”

Turner said that is also important to reinforce the human element of this work, whether it is the community or first respondents, and the impact and importance of the people. “Technology can make a big difference, but it is also about the people.”

Copyright © 2020 IDG Communications, Inc.

7 inconvenient truths about the hybrid work trend
Shop Tech Products at Amazon