Algal classification and Chlorophyll-a concentration determination using convolutional neural networks and...

Area of Science:

• Environmental Science
• Water Quality Monitoring
• Applied Artificial Intelligence

Background:

• Harmful algal blooms (HABs) increasingly compromise drinking water quality.
• Chlorophyll-a (Chl-a) is a key indicator of algal biomass, but current measurements face limitations in accuracy and environmental adaptability.
• Accurate quantification of algae and Chl-a is crucial for effective water resource management.

Purpose of the Study:

• To develop an advanced method for accurate algal classification and chlorophyll-a (Chl-a) concentration determination.
• To overcome the challenges of existing Chl-a measurement techniques in natural aquatic environments.
• To enhance the reliability of water quality monitoring systems for harmful algal blooms.

Main Methods:

• Integration of convolutional neural networks (CNNs) with three-dimensional fluorescence data matrices.
• Development of an algal classification model tested on thirteen algal species.
• Implementation of Chl-a concentration models for mixed algal solutions in various water backgrounds (ultrapure, reservoir, and natural lake water).

Main Results:

• Algal classification model achieved over 99.5% accuracy, focusing on pigment regions.
• Chl-a models showed varying Mean Absolute Percentage Errors (MAPEs) across different water types, with significant improvement after calibration in complex environments.
• Model performance was influenced by the intensity and location of algal pigment fluorescence peaks.

Conclusions:

• The developed CNN-based approach offers a novel and accurate solution for algal identification and Chl-a quantification in diverse aquatic settings.
• This method holds significant potential for practical applications in safeguarding drinking water quality from algal blooms.
• Further research and calibration are essential for optimizing performance in highly variable natural waters.