What did we develop?

Live Web App:https://firebusters-fireless.web.app/

Until recently, fires were mostly identified by human reporting through public hotlines, detecting smoke from a watchtower by fire lookouts or by ground and aerial surveillance by means of an aircraft. These methods are however limited by the human error, accessibility of the area and geographical location. Even with the abundance of remote sensing data from satellites, the application of automated electronic detection of wildfires is limited.  

We developed an application that leverages the Machine Learning and Deep Learning techniques and the abundance of open satellite resources to build a model that can effectively detect fires and predict their future spread. The system then sends out alerts to the areas nearby that could be affected by the reported fire. 

Additionally, the application also includes a communication system to leverage the human resource on the ground. It allows us to enroll volunteers to aid firefighting efforts. The volunteers can aid any citizens affected by the wildfires or those in regions susceptible to the wildfire spread as determined by out fire spread prediction model.

How does it work? 

We use the MODIS data from the NASA Aqua and Terra Satellites. The MODIS data gives us the NDVI (Vegetation Index), Thermal Anomalies and Land Surface Temperatures. RTMA (Real-Time Mesoscale Analysis) data gives us weather data such as wind speed, direction, pressure, humidity, etc.

1 .Prediction and detection of wildfires:

To predict and detect wildfires, we train our model with the NDVI, surface temperature and RTMA data as predictors and the Thermal Anomalies data as the target labels. Once the model is trained, it is evaluated to see how accurately it can predict the anomalies using the known Thermal anomalies for validation.

2.Modelling the spread of a forest fire once it has started

The next step would be to accurately predict the spread of the fire so that firefighters can take measures to prevent the spread to those regions and contain the fire. With accurate fire spread prediction, the firefighters can focus their resources and be more effective at putting the fire out. We have researched several papers which outline many novel techniques to model the forest fire spread factoring in all the environment and weather conditions.

3.Enabling Communication

The web interface provides a feature where it allows people living in affected areas to help by enrolling as volunteers on our website. A person in an affected area or a fire susceptible area can get in touch with a volunteer nearby and gain information about mitigation and protection measures to take, evacuation strategies, if any.

How this useful? When a region is struck by disaster, the authorities and emergency responses are strained and may not be able to cater to all the affected people. The volunteers nearby can help take the load of the emergency services and authorities so that they can focus on getting the fire out. They will have to prioritize severe cases. Our network of volunteers can pitch in here and help reduce the stress on our emergency services. Any non-essential assistance or services can be carried out by the volunteers.

Why is it important? 

In California, the U.S. Forest Service spends about $200 million per year to suppress 98% of wildfires and up to $1 billion to suppress the other 2% of fires that escape the initial attack and become large.

Australia faced the worst bush fires ever seen. Intense bush fires cause the death, by incineration or smoke suffocation, of large numbers of native animals and insects that are unable to avoid the flame. Half a billion animals perished. Not only that, this disaster cost Australia close to $100 billion. So, good fire detection and spread modelling not only helps save the environment and fauna, but it also helps save the infrastructure destroyed and the resources invested in it. It saves the government its financial resources which could be better invested in research and other projects that benefit the economy and the environment.

What do you hope to achieve?

Detecting wildfires early on or better yet predict the possibility of ignition of a wildfire at a location can assist the authorities to facilitate a quick and efficient response. The conventional methods such as human scouting of locations for fires or smoke, or aerial surveillance of forests can be inefficient and expensive. It can be very challenging to cover large forest regions and would require devoting a large number of resources and manpower to constantly monitor fire-prone regions. 

A more cost-effective measure would be to deploy our remote sensing application which constantly monitors the latest satellite feeds to detect any fires and/or identify and alert susceptible regions.