UC Santa Cruz researchers build AI to prevent drownings

In partnership with NOAA and funded by UC Santa Cruz’s Center for Coastal Climate Resilience, Alex Pang and his team are working on algorithms — sets of programmed instructions — that can monitor shoreline change, identify rip currents, and alert lifeguards of potential hazards. They hope to improve beach safety and ultimately save lives. 

 

SANTA CRUZ — As the winter swell approaches, UC Santa Cruz researchers are developing potentially life-saving artificial intelligence technology.

In partnership with the National Oceanic and Atmospheric Administration and funded by UC Santa Cruz’s Center for Coastal Climate Resilience, Alex Pang and his team are working on algorithms — sets of programmed instructions — that can monitor shoreline change, identify rip currents, and alert lifeguards of potential hazards. They hope to improve beach safety and ultimately save lives.

The initial seed of inspiration took root while Pang was windsurfing with his friends. “They would point out a rip, and I would look in the water and say, ‘That’s just water,’” says Pang. Since rip currents are difficult for the untrained eye to detect, Pang thought, computer technology might be useful. He officially started the rip current detection project in 2015, one year before UC Santa Cruz lost two students to drowning.

“That put more pressure on realizing this capability,” says Pang.

According to NOAA, a rip current is a “powerful, narrow channel of fast-moving water.” It pulls unsuspecting swimmers into deep water, where they risk fatigue and drowning after trying to fight the current. Santa Cruz Marine Safety reports 10 drowning deaths in the past decade, including two in 2023. The National Weather Service ranks rip currents as the third most dangerous of all weather hazards, just behind heat and flooding.

Rip currents can be difficult to detect from shore and sometimes appear unexpectedly. Pang’s team explored many different methods of rip detection and ultimately decided to use a machine-learning-based system similar to the obstacle detection systems used in self-driving cars. Machine learning is a type of artificial intelligence that describes the ability of a machine to make decisions based on information it has been given. Scientists showed their rip current detector a collection of images, some with rip currents and some without, to train the system to recognize the common attributes of a rip current. After training, the detector can find rip currents in live video streams.

Pang’s team partnered with NOAA to develop and refine its detection methods. With the help of the Santa Cruz Harbor Office, O’Neill Sea Odyssey and the U.S. Coast Guard, researchers installed a streaming webcam at Walton Lighthouse with views of Seabright and Twin Lakes beaches. Pang’s team is now using the machine learning model to process and detect rip currents on these beaches via the live video feed.

Pang’s rip current detection model will create a rip current observation data set that can validate and improve NOAA’s existing forecast model developed by Greg Dusek, physical oceanographer at NOAA. The forecast model takes information on the wave height, wave direction, tide and the presence of sand bars, and calculates the risk of a rip current developing. “It predicts the likelihood of a hazardous rip current from zero to 100%, similar to other weather forecasts,” says Dusek.

Once Pang’s detector model is reliable and there aren’t too many false flags, the research team plans to develop an alert system that lifeguards can customize based on their needs. “Ideally, the system will send alerts to lifeguards only if there are people detected in the rip. If necessary, it can distinguish between people and surfers,” says Pang.

The ultimate goal is to have a national network of cameras operating in real-time with rip current detection. The detection accuracy at each location will improve over time as data are collected, and different environmental and weather conditions are factored in.

Another potential development could be a phone app that would do the same job. “You could bring your smartphone to a beach without internet, and take videos of the water. The app would detect a potential rip current and then contribute to the data pool,” says Pang.

This technology does not eliminate the need for boots on the ground. The Santa Cruz Fire Department is responsible for patrolling the beach by the Boardwalk. “This would still require a dispatch of responding rescuers to help someone that is caught in a life-threatening situation,” says Brian Thomas, marine safety captain for the Santa Cruz Fire Department. “This system may help get resources started if a bystander did not see the person in distress and call 911. For that specific scenario, it could have a benefit.”

Other public safety officials think it could help distribute resources more effectively. “I think it’s a tremendous idea. I see this as being an added tool that would help us in terms of monitoring, outreach, and education of the public,” says Santa Cruz Fire Chief Robert Oatey.

Local surf instructor Che’ Jordan, who often finds himself performing water rescues on larger swell days, advises that the technology could be useful, but people should still learn to identify rip currents for themselves. “Technology always has its glitches, and when it comes to Mother Nature, those glitches can cost lives,” said Jordan.

Despite AI’s potentially lifesaving contributions to public safety, the researchers want to be clear that even if the model doesn’t identify a rip current, that doesn’t mean there’s no danger. According to the United States Lifesaving Association, no matter your location or the weather conditions, the best thing you can do is swim near a lifeguard.

“This is just one tool to give you an idea about what hazards might be present. If you see something yourself, and it looks dangerous, it very well may be,” says Dusek.