AI-Powered System Predicts Harmful Algal Blooms in Real Time

Harmful algal blooms (HABs) pose a significant threat to public health and aquatic ecosystems. Researchers at Hohai University in Changzhou have developed an innovative solution to predict these blooms in real time using a combination of automated monitoring systems (AMS) and machine learning algorithms.

HABs are often triggered by excess nutrients like nitrogen and phosphorus, and they thrive in warm water conditions, typically above 20°C. Current monitoring methods, such as satellite and manual monitoring, have limitations, including low temporal resolution and high operating costs. The BloomSense system addresses these issues by deploying sensor buoys that collect data at 15-minute intervals.

The study, published in the special issue of 'New World: Advancing Water Applications Through Machine Learning and Artificial Intelligence,' was conducted in the eutrophic freshwater reservoir As Conchas in southwestern Spain. The system uses a hybrid machine learning model that includes Random Forest, SMOTE, ResNet-18, and LSTM algorithms. These components work together to predict the probability of Chl-a concentrations exceeding critical levels.

Key parameters for predicting HABs include water temperature, pH, electrical conductivity (EC), and chlorophyll-a (Chl-a) concentration. The system also monitors the battery charge level of the buoys to ensure reliable operation under limited energy resources.

The model was tested on two data sets: one from stable inland waters and another from dynamic coastal areas. It reduced the mean absolute error (MAE) by 26.2 percent compared to classical models and increased the F1-score by 70.2 percent in classifying the occurrence of a bloom. The system performed well even in unstable conditions, making it a robust solution for real-world applications.

The study concluded that continuous monitoring with buoy sensors effectively replaces the limitations of satellite detection. Using low-cost input parameters (temperature, pH, EC, Chl-a) makes the solution scalable and suitable for resource-constrained environments. The integrated warning system allows for rapid response to potential harmful algal blooms, enhancing water resource management.

This new hybrid machine learning model combines deep feature extraction with sequential analysis, optimizing input variable selection and data balancing. It has proven to be energy-efficient, low-cost, and accurate, capable of withstanding environmental variability and performing well under different hydrological conditions. Future research will focus on the impact of battery charge level on data quality and overall system performance.