Bushfires Data Quest 2020 Showcase
The Bushfires Data Quest 2020 brought together the best scientific minds in bushfire research, machine-learning and Earth-observation. The Data Quest showcase of results are from four Data Quest interdisciplinary teams, illustrating what is solvable with the help of machine learning.
The showcase also provided opportunities to hear the latest thinking about how research of this kind can help communities across Australia and New Zealand get ready for Bushfires season.
FrontierSI Data Scientist, Dr Caitlin Adams, participated in the Data Quest, where her team investigated how to predict moisture in vegetation fuel (a key driver in fire behaviour) on a 20-metre scale, using radar and optical data from the Sentinel satellites.
Bushfires Data Quest 2020 Showcase
The digital showcase presented results showing how machine learning and bushfires specialists measure fuel loads accurately, detect fires earlier and better predict fire behaviour.
Below is a summary of FrontierSI insights from the showcase:
The devastating bushfire season has prompted Australia to think deeply on how we could better leverage technology to aid fire prevention, response, and recovery. This is evidenced by the Bushfires Royal Commission, as well as the New South Wales Bushfire Enquiry. Both have identified that there is a clear opportunity to improve how we manage bushfires by applying modern analytic approaches, like machine learning, to the high spatial- and temporal-resolution remote sensing data available today.
The Data Quest 2020 program was designed with all of this in mind. It asked whether we could apply machine learning to satellite imagery to improve our management of fires before, during, and after they occur. What makes it unique is that it asked Australian and New Zealand researchers to explore this question in just a single week. The program put together four interdisciplinary teams, with each researcher having a background in machine learning, Earth observation, or a mixture of both.
FrontierSI Data Scientist, Dr Caitlin Adams, participated in the Data Quest, where her team investigated how to predict moisture in vegetation fuel (a key driver in fire behaviour) on a 20-metre scale, using radar and optical data from the Sentinel satellites.
The initial results from Caitlin’s team were promising; their machine learning approach was trained to predict fuel moisture from Earth observation data. This is beneficial as machine learning models could provide an alternative approach to physics-based models, which can be too time-intensive to run predictions in real-time. However, the team highlighted that there are currently not enough ground measurements of fuel moisture to harness the power of machine learning approaches, which are most successful when applied to large volumes of data. One way to combat this could be to run a citizen science project, sourcing measurements of vegetation fuel moisture from across Australia.
The Data Quest program brought multidisciplinary teams together to tackle challenging problems and allow them to validate new ideas before investing significant time and effort in applying them. The presence of researchers with different backgrounds can lead to completely new approaches; for example, another team used image stacking techniques from Astronomy to enhance the detection of fires in satellite imagery. This program has a good first step in understanding how Earth Observation and machine learning can be better leveraged in the bushfire space, but there is more to do. Connecting the right organisations to understand their needs and opportunities will be critical if we are to see the advances made during the Data Quest put into practice.
