Our idea is about detecting wildfires using animal tracking data. There is already wildfire detection over satellites right now, but it is not accurate enough and it can only scan same area 4-5 times a day. Also its accuracy depends on many factors such as clouds, time of the day and location of the fire relative to satellite’s location.

We know that animals behave different when there is a fire nearby than their daily lives. They all try to survive fire with the abilities they have. And we researched that their behavior is somehow predictable meaning there is a pattern. There are already more than 100,000 animals that are being tracked right now with tracking devices.

Our idea is to use this animal data to make satellite based predictions much more accurate and we can detect the wildfire location even before satellites.

How do we do that: First we feed Machine Learning algorithm with animal tracking and wildfire data that has been collected in past years to train it. The algorithm will be able to find the relation between wildfire and animal behavior. Then we will be able to detect the wildfire occurrences using live animal tracking data and then confirm it with satellite wildfire data.

Challenges we faced: It was really hard to find animal tracking data as most of the scientists made their data private in that 48 hours we found only 3 animal data that we could use. And also we had to do big data analysis as those 3 animal data was 300-400MB each and wildfire data was 700MB for August of 2019. Each time we made a little change in our algorithm and had to test it we had to wait 15mins for the program to load and parse that data.

Resources:

http://www.argos-system.org/

https://www.movebank.org/

https://firms.modaps.eosdis.nasa.gov

https://www.nationalgeographic.com/news/2014/7/140...

https://www.businessinsider.com/wildfires-forest-a...