Artificial intelligence is transforming the way we learn, how we treat disease, and the very nature of work. Now, it's tackling a problem that has long plagued Hawaiʻi residents: making our weather apps accurate for once.
Peter Sadowski is an associate computer science professor at the University of Hawaiʻi at Mānoa. He's collaborating with atmospheric scientists to incorporate AI statistical techniques into weather forecasting.
"It's a good time to be working in this area," Sadowski said. "There are real applications."
Traditional two-week weather forecasts rely on big, expensive computational simulations of the atmosphere conducted by the U.S. government.
"Hawaiʻi is particularly vulnerable to errors in this forecast because we have these tall mountains that create all these different weather patterns in our state," Sadowski said.
Artificial intelligence can refine those simulations to make hyper-local forecasts for Hawaiʻi. The implications of this work go beyond whether you should pack an umbrella when you leave the house. It could help state officials predict where the next wildfire will be.
AI can use data on things like rainfall, soil moisture and wind speeds to map wildfire risk across Hawai'i.
"The recipe for wildfires is dry areas with lots of brush that can catch fire easily, but also high winds," Sadowski said. "So one of the problems we've been looking at here is to use machine learning to help predict wind."
Lawmakers are looking to jump-start that research. Companion bills in the state Senate and House would establish a two-year program at UH to develop an official AI wildfire forecast system.
The university supports the measure and stated in testimony that it has "the full range of skill sets and research, as well as the climate data, necessary to develop this important tool for Hawaiʻi’s decision makers."
Several states are already using AI as part of their wildfire prevention strategy.
The California Department of Forestry and Fire Protection (CAL FIRE) has partnered with UC San Diego to launch an AI-powered fire detection program. It employs a network of cameras placed on remote mountain tops that take panoramic photos every two minutes.
AI then scans each photo looking for any changes that could be a sign of fire. If such a change is detected, the system sends an alert to CAL FIRE officials.
"We're trying to basically improve the time from when a fire is first detected to when we can quickly get resources to the scene," said Andy Emerick, an assistant chief with CAL FIRE.
CAL FIRE's goal is to keep 95% of California's wildfires to 10 acres or less, which means early detection is key.
Generally, emergency responders rely on reports through 911 to know when a fire has started. Emerick said that their new AI system often catches a fire before the fire dispatchers receive any 911 calls.
"About 20% of the time, that artificial intelligence is beating the call dispatch system, which is pretty remarkable," Emerick said.
The San Francisco-based startup Pano AI has seen similar success with AI wildfire surveillance. They've installed fire detection cameras in nine states, as well as several locations in Australia.
CEO Sonia Kastner previously worked on the Nest Cam, a home security camera that uses AI to tell the difference between residents and intruders. She said their Pano program is doing the same thing, but for fires.
"Sometimes there's a perception that wildfire technology is cutting edge, but actually, we're taking technologies that were developed for other industries, and we're just bringing it into the wildfire and emergency management space," Kastner said.
Both the CAL FIRE and Pano AI's systems rely mainly on 5G connectivity to upload photos from remote locations. That could be an obstacle to implementation in Hawaiʻi where such connectivity is still comparatively limited, said tech expert and Bytemarks Café host Burt Lum.
“Let's say you’re driving along the Hana Highway, you can barely keep regular cell service,” said Lum.
But Kastner said there is plenty to be excited about when it comes to AI-powered wildfire strategies, from modeling how fires will behave to identifying risks like dry fuels and faulty power lines before fires start.
"We're lucky to be part of an ecosystem of exciting innovators in wildfire mitigation," Kastner said.
CAL FIRE is also exploring other ways to use artificial intelligence throughout its department.
Emerick said the approach to fighting fires needs to change because the fires themselves are fundamentally changing. Between 2018 and 2021, Emerick said California has seen a significant increase in the size of fires.
"Fires that used to burn maybe 1,000 acres a day are now burning anywhere between 10,000-20,000, even 100,000 acres in a day sometimes," said Emerick.
Emerick said AI tools are becoming essential for emergency managers "to make the most sound decisions possible."
