by_adr - Fotolia
In January 2017, a series of wildfires roared across Chile. The fires killed at least 11 people and destroyed thousands of homes. They burned well over a million acres of land and cost the government of Chile hundreds of millions of dollars.
While the 2017 fires were some of the worst the South American country had seen in modern times, wildfires are fairly common in Chile. To help prevent them, the country has turned to tech vendor Entel Ocean, which uses AI to help prevent and detect wildfires.
Fire detection in Chile
Entel Ocean is the digital unit of Entel, the largest telecommunications provider in Chile. The company offers a diverse portfolio of products, including cloud, analytics and IoT platforms.
Working with the government of Chile, Entel Ocean installed IoT sensors on trees in some of Chile's forests to help detect and predict forest fires, like the ones in 2017 that ravaged the country. The "noses," as the company calls them, collect a range of data on their surroundings, including humidity levels, temperature and the concentration of certain materials, said Leonor Ferrebuz, service line manager at Entel Ocean.
The sensors relay that information back to an IoT platform in Entel Ocean's offices. Using predictive analytics and machine learning, the company can use the data the "noses" collect to identify how possible forest fires are likely, as well as identify if a forest fire, even a small one, is active.
The platform enables the detection of fires 12 minutes quicker than traditional metrics, which includes people stationed in watchtowers in the forest visually keeping an eye out for smoke.
Automated machine learning
"It's pretty amazing to us that you can accelerate the data and AI components with DataRobot," Ferrebuz said. "It's impressive the way it accelerates and helps it improve and reduce the training cycle using this technology."
The Entel Ocean team working on the forest fire detection system has experience with numerous AI platforms, Ferrebuz explained. They chose DataRobot after evaluating each platform, she said.
While the platform has worked well for Entel Ocean, the fire detection product got off to a rough start. The product needs data from the "noses" to work efficiently. After setting the "noses" up, it took a while for them to collect the necessary amount of data to work well.
Leonor FerrebuzService line manager, Entel Ocean
"The best way to train the model is to be in the forest," Ferrebuz said.
Originally, the product had a fairly large number of false positives, she said, with factors like rainfall or barbecues potentially raising flags. Those have fallen significantly over time and are expected to continue to fall.
Still, with the system predicting fires 12 minutes before other methods, and it limiting the danger fires pose to human visually trying to detect them, Ferrebuz expects other governments to potentially look into using it to protect their forests.
"Today, forest fires are detected primarily by vision," she said. "We give insight before the eye or the camera can see."
Dig Deeper on AI technologies
Vodafone heads consortium to develop use cases for world’s first 5G marine-focused testbed
Benioffs and Salesforce put $300m into combating climate change on eve of COP26
AI carbon footprint: Helping and hurting the environment
Dryad Networks gains funding for large-scale wildfire detection IoT network