Chlorophyll soft-sensor based on machine learning models for algal bloom predictions.

Area of Science:

• Environmental Science
• Aquatic Ecology
• Data Science

Background:

• Harmful algal blooms (HABs) pose significant risks to public health and aquatic ecosystems.
• Manual water monitoring methods have limitations for effective water safety planning.
• High-frequency monitoring is crucial for timely detection and management of HABs.

Purpose of the Study:

• To develop a data-driven chlorophyll-a (Chl-a) soft-sensor using automatic high-frequency monitoring (AFHM) and machine learning (ML).
• To create an inexpensive and efficient system for inferring Chl-a fluorescence from low-cost variables.
• To support manual sampling efforts in water bodies susceptible to HABs.

Main Methods:

• Collected extensive water quality data (temperature, pH, EC, battery) over three years at 15-minute intervals.
• Developed ML-based soft-sensors using compact, energy-efficient algorithms for Chl-a inference.
• Applied input and output aggregations to enhance ML model performance and developed an alert system for Chl-a levels.

Main Results:

• The developed Chl-a soft-sensors accurately inferred fluorescence using low-cost input variables.
• The system demonstrated effectiveness in a freshwater reservoir setting.
• An alert component successfully triggered at a 10 µg/L Chl-a threshold.

Conclusions:

• Chl-a soft-sensors offer a rapid and economical solution for HAB monitoring.
• This technology can significantly enhance water safety planning and early warning systems.
• AFHM systems combined with ML provide a scalable approach for managing aquatic ecosystems.