AI to assist Aussie firefighters in bushfire prediction and response
In this thought piece, NSSN Engineer Dr Ayu Saraswati reflects on some of the applications of Artificial Intelligence in bushfire prevention, mitigation and response.
Bushfires are a seasonal occurrence in Australia. However, in summer of 2019/2020, the nation experienced an unprecedented fire season.
The New South Wales (NSW) bushfire inquiry report into the 2019/2020 season (released in July 2020), highlighted unusual fire events such as fires burning down forested regions, unpredictable fire behaviour, and a 50% increase in fire-generated thunderstorms (compared to the year before).
The report listed recommendations to improve NSW capability in bushfire planning, preparation, and response. It recognised the need to improve the available technology, especially in remote sensing, data science, and artificial intelligence (AI).
AI technology can be leveraged for timely automated fire detection so that evacuation and suppression efforts can be made safely and quickly. Timely detection is at the essence in bushfire response, as CSIRO reported, fire area could burn up to 100 m2 of vegetation for every 10-minute delay.
Grass root efforts in Kangaroo Island, South Australia revealed enhancing Australia’s remote sensing capability could significantly improve our ability to detect and contain bushfires. Although large data sets of satellite images from multiple sources are often freely accessible, Kangaroo Island fires proved the unprocessed data failed to provide any immediate insights for the emergency responders. It is in circumstances as such that AI can play a critical role in containing bushfires.
More recently, AI technology has been used to increase the resolution of the Himawari-8 satellite images so that smaller fires can be detected before they spread. The model also significantly reduces the number of false-positives, allowing emergency responders to focus their resources on high-risk areas.
Australian researchers have used a method called image stacking, previously used in astronomy to increase Himawari-8 satellite’s image quality and resolution. This method involves training a Convolutional Neural Network (CNN) on the terrain and landscape to detect smaller fires.
AI technology has also been used to process and make sense out of the data produced by satellite images and on the ground sensors. Fireball, a Queensland-based company, has shown its AI technology can detect a fire within 3-5 minutes of ignition by processing satellite images to detect fire ignition and sensors on towers to confirm the detection. Their published work details a smoke detection camera based on neural network architecture InceptionV3 that fuses with satellite-based data to achieve precise fire detection.
NSSN member university researchers are also hard at work to tackle this Grand Challenge. Researchers at the Australian National University (ANU) are developing the first Australian satellite for bushfire detection. In this case, AI technology will assist with both detecting fires and measuring and predicting the contributing factors — such as fuel moisture — that could lead to fires.
There are numerous possibilities for AI technology to enhance bushfire planning, preparation, and response. AI technology can be implemented to predict where and when ignition occurs prior to a fire event or can provide locals with information about fire risk based on the level of fuel moisture around their properties.
Although AI technology has proven to be a game-changer for bushfire detection, mitigation and response, complex challenges have slowed down the technology to be wildly adapted and commercialised.
One of such challenges is in the machine learning component, which is the lack of vetted ground truth data that is required to train the machine learning models. The other challenge is to communicate, share and translate the result of these AI technologies to the firefighters and affected communities on the ground. NSSN member university researchers are well-positioned to find solutions for such challenges.
Bushfires are one of the four Grand Challenges announced by NSSN as part of the Network’s Grand Challenges Program launched in 2020.
The NSSN Grand Challenges aim to tackle Australia’s most intractable problems by bringing together the smart sensing expertise of eight research powerhouses across NSW and the ACT with industry and government.
Advances in smart sensing technologies can detect the earliest signs of a fire, improve onsite communication for firefighters, and monitor the effects of fires on our community. By combining the world-class research across our network, the NSSN aims to formulate a comprehensive response to this grand challenge.
To learn more about NSSN capabilities in bushfire prevention, mitigation and response, please click here.
To partner with NSSN, please click here.
This thought piece was written by NSSN Engineer Dr Ayu Saraswati.
Media: Shahrzad Abbasi - 0466 548 145