UN World Challenge Winning Team Shares Their Story
Wildfire AWARE started off as a pure passion project. University of York Computer Science exams had just ended in June 2017 giving students a few weeks off before their internships began. One member of the team, Vishal Soomaney was starting to get bored and was looking to work on an interesting project. He found another team member, Peter Jupp and they began brainstorming ideas. This is when a devastating wildfire in Portugal occurred, taking many lives and causing massive destruction.
They began thinking about why appropriate prevention measures weren’t taken beforehand, “With all the data we have access to, why can’t we predict such wildfires and take preventative measures beforehand?”. That is when they decided to work on a project attempting to predict Wildfires using Artificial Intelligence and Big Data. Shortly after, Flinn Dolman joined the team as well.
Most of the work was done in the evenings as all team members were working full-time (9am — 5pm) in an internship so the project became quite stressful as time went on, especially when bugs were found and had to be fixed late at night. The team specifically recall a bug that caused the application to go down shortly before the NASA Europa Challenge judging: they spent all night working to fix until finally realising it was caused by a satellite going offline. Despite all the late nights and constant stress, the efforts paid off when Wildfire Aware tested the model on historical wildfire datasets and found that it had a 74.8% wildfire prediction accuracy. Wildfire Aware went on to win the RCUK Smart Cities Prize 2017 and were finalists in the NASA Europa Challenge 2017 powered by Ultrahack.
In 2018, the team was still working full-time at their internships and were constantly raising their Software Engineering skills. Vishal, Peter and Flinn spent time looking at the weaknesses of the application and decided to rework everything with a new system architecture. The frontend and backend were fully rewritten to increase speed and efficiency as well as to add more features. The machine learning model was redeveloped using Gradient Boosted Decision Trees rather than a Support Vector Machine based approach. A lot more data was used as well, this included usage of 1.88 million historical wildfires along with a massive amount of weather data. Although theoretically the updated approach should have increased efficiency and prediction accuracy, there was no way to tell for sure until they actually tested it. This led to quite a bit of stress and anxiety since all of the work could have been for nothing. Fortunately, the effort ended up paid off when the updated application produced a 78% prediction accuracy. Wildfire Aware went on to apply for the UN World Challenge 2018 and the Microsoft Azure challenge and were delighted to have made it into the finals. From trying reindeer for the first time at the fantastic Aino restaurant in Helsinki’s city centre, to relaxing at the saunas in the Nuuksio National Park, the team had a great time during the event.
The event also provided a great chance for networking, the team had wonderful opportunities to speak to representatives from IBM, Microsoft, UNtil, the ESA and CGI amongst others. Wildfire Aware are continuing to speak to many of these groups even now and hope to arrange long-term collaboration with them. “We also became great friends with a lot of the other amazing teams at the finals, many of whom gave us great advice on how we can grow and improve our application,” says Vishal.
Just like most teams, Wildfire Aware didn’t want to get our hopes up but held on to the dream of potentially winning the competition. Their dream ended up coming true and they won both the UN World Challenge main prize and the Microsoft Azure challenge. The events that followed the awards ceremony were just as surprising, whilst victory had yet to fully sink in, the guys were being interviewed by Microsoft and multiple different sponsors wanted to take pictures with them. “With this experience, if software engineering doesn’t work out, we might have a potential future in modelling!” they say.
Wildfire Aware is currently looking to work with governments and fire agencies around the world and hope to be saving as many people as possible from these natural disasters in the near future. If you’d like to talk to them, leave a comment down below, and we will connect you to the team.
