An AI Algorithm Built to Count and Grade the "Blue Tears" Glowing Along China's Coasts

On warm summer nights along China's Fujian coast and elsewhere in East Asia, the ocean sometimes glows blue. The source is noctiluca scintillans - a single-celled marine organism that produces bioluminescent light when disturbed by wave action. The phenomenon has been called "blue tears," and it draws tourists in large numbers. It also poses ecological risks. Dense blooms of these organisms can deplete oxygen in the water, create conditions hazardous for marine life, and leave beaches unsafe.

Managing the tension between the tourism appeal of blue tears and their ecological impact requires something the field has lacked until recently: a reliable way to measure them. How intense is a particular bloom? How large is the glowing area? How quickly is it moving? Without quantitative answers to those questions, issuing forecasts or managing visitor access is largely guesswork.

A team led by Professor LI Jianping at the Shenzhen Institutes of Advanced Technology of the Chinese Academy of Sciences has addressed that measurement gap. Working with collaborators from the Ministry of Natural Resources, they developed a real-time video monitoring algorithm called BT-YOLO, published in Ecological Informatics. The algorithm analyzes footage from coastal surveillance cameras and segments the glowing areas at the pixel level, producing quantitative measurements of bloom size and intensity.

From Detection to Quantification

Existing methods for tracking blue tears typically detect the presence or absence of a bloom - a binary answer. BT-YOLO goes further. By performing pixel-level segmentation, it distinguishes the glowing bloom area from the dark surrounding water with enough precision to calculate the area and intensity of individual patches of bioluminescence within a single video frame.

This distinction - from detection to quantification - is what makes forecasting possible. A bloom-detection system can tell you that blue tears are present tonight. A bloom-quantification system can tell you whether tonight's bloom is twice as large as last night's, whether it is concentrated in a specific bay, and how rapidly it is expanding. Those are the inputs that a forecasting model needs.

"We have built precise 'scales' and 'rulers' to measure 'blue tears'," said Professor LI. "Once the coastal surveillance camera network is deployed, this algorithm will allow us to perform rapid quantification and move closer to an operational forecasting system."

Potential Applications Beyond Bioluminescence

The BT-YOLO framework is designed to be adaptable. The underlying computer vision approach - detecting and segmenting anomalous features in ocean surface imagery from coastal cameras - applies to other marine phenomena that have historically been difficult to monitor continuously. Red tide algal blooms, which can produce toxins dangerous to marine life and humans who consume shellfish, produce characteristic color changes that video analysis could track. Marine debris patches, plastic accumulations, and unusual turbidity events are other potential targets for the same type of automated visual monitoring.

The team describes their current work as laying a foundation. The BT-YOLO algorithm has been validated on existing footage, but a full operational forecasting system will require deployment across a network of coastal cameras, collection of sufficient observational data to train predictive models, and integration with oceanographic data on currents, temperature, and nutrients that drive bloom formation.

The authors acknowledge that further validation using live camera network data is needed before the system can support reliable forecasting. Coastal lighting conditions, camera angles, weather interference, and the variation in bloom color across different water conditions will all need to be addressed as the system moves from research to operational use.