AI in Bushfire Management: Briefly.

Homepage Topic References

Bushfires are some of the most devastating natural disasters, not just to homes and people, but to the environment as a whole [Podur et al, 2003].
In recent times, with the development of technology and observational devices such as satellites, the focus of research efforts can move more to the strategies used to prevent fires from spreading and causing undue damage to life, infrastructure, and the land on which we live.

Current methods of tracking objects used in cameras, such as facial recognition and traffic cameras are insufficient for these purposes [Filkov & Prohnaov, 2019].
Thermograms have been successfully used to predict, track, and label individual firebrands produced by wildland fires [Thomas JC et al, 2017].
These data points may also be usable in urban areas, with structural fires having a similar structure in their cinders and spread [Suzuki et al, 2012], but as urban fires do not tend to last as long as bushfires, will not be as useful for those ends.

Tracking of smoke particles is a further way to track and model fire spread [Satir et al,2016].
Accuracy would be paramount for this system, and false negatives are too costly to allow happen in any capacity [Tien Bui et al, 2017].
Laser sheet imaging detection has been used for some time, and using machine learning to find subtle correlations between smoke particles may even be able to differentiate between two fires of equal size but differing intensity and spread [Shu et al, 2006].
Prioritization of more virulent fires is paramount to such a system being effective.

While modelling the methods of spread of secondary fires may be integral to fire management, modelling the spread of the fire itself is even more important [Green et al, 1995].
By scanning the landscapes involved, and assigning labels to a given area based on whether or not it has been affected by fires, a map of sorts is built, allowing assessment of the fire’s spread at a glance [Zheng, et al, 2017].
Currently, algorithms used to model these areas lack accuracy. With time and research, these could be refined to produce more reliable results to take in provided data and appropriate variables to produce maps of the fire’s spread that could be used to predict and prevent the spread of further flame [Niessen & Blumen, 1988].
Machine learning algorithms such as these could be produced with millions of fire’s worth of data, without any lives being lost, running simulation after simulation with slightly tweaked variables until one algorithm proves successful, consistent, and efficient, comparing the output to that of actual recorded bushfire spreads [Hernandez Encinas et al, 2007].
This mass of data would form the core of any fire spread prediction, and even if the accuracy of the system was impeccable, multiple instances would be run with identical conditions for ensured precision.

If these two aspects of disasters could be integrated, then it would become possible for people affected by fire to simply open up an application on one’s phone, select their location, and send the details to an active database.
This database could then be able to be visualised in a virtual map overlayed with the latest satellite imagery, and used to convey AI-generated procedures, routes, and general information to responders [Dimopoulou & Giannikos, 2004].
Co-ordination between various agencies acting towards the same goal is a vital one, and must be available to all relevant parties, to prevent contradicting information which only serves to confuse those seeking help (Cohen et al, 2005, p. 16).