When toxic algae blooms sweep across lakes and rivers, the people living nearby often have no fast way to know if their water is safe to drink. Now, a team of researchers from Florida and South Korea has built a computer system that can help cheap, handheld sensors detect these poisons accurately — without needing a lab full of expensive equipment.
The researchers, from the University of Central Florida and Hanbat National University in South Korea, created a machine learning program that teaches simple sensors to work reliably even when testing water from different rivers, lakes, or estuaries. Their system uses a type of artificial intelligence called XGBoost to make sense of sensor readings that would otherwise be muddled by differences in water quality.
"This framework eliminates the need for repeated sample-specific calibration, reducing time, labor and sensor consumption," said Professor Jungsu Park of Hanbat National University, who led the study along with Professor Woo Hyoung Lee of the University of Central Florida.
The team tested their system at 27 different water sites across Florida, collecting 201 measurements from rivers, estuaries, and transitional waterways. They trained the computer model to account for factors like water acidity, cloudiness, and electrical conductivity — all the things that can throw off a sensor's reading. The result: the system correctly detected toxin levels 89 percent of the time.
That matters because the toxins in question — called microcystins — are no small concern. Produced by blue-green algae during harmful blooms that are becoming more common as the climate changes, these poisons can damage the liver and have been linked to higher risks of liver and colon cancer. The World Health Organization sets the safe limit at just 1 microgram per liter in drinking water.
Traditional lab testing for these toxins is accurate but slow and expensive. Portable sensors offer a faster, cheaper option — but they have long struggled to give reliable results when the water being tested has different chemistry from the water they were calibrated for. This new framework bridges that gap by using the sensor's measurements along with simple water quality data to predict toxin levels accurately no matter where the sample comes from.
As algal blooms grow more frequent and severe with rising global temperatures, having a practical way to monitor water safety could make a real difference for communities near affected waterways. The researchers say their system could be adapted for other water quality monitoring needs, potentially helping protect both drinking water supplies and recreational waters around the world.
The findings were published in the journal Water Research.