Classifying eutrophication spatio-temporal dynamics in river systems using deep learning technique

Area of Science:

• Environmental Science
• Water Quality Management
• Machine Learning Applications

Background:

• Eutrophication, driven by algal blooms, severely degrades water quality in South Korea.
• The Trophic State Index (TSIko) is used for management, but mechanistic models face calibration and nonlinearity challenges.
• Deep learning models, particularly CNNs, offer a promising alternative for extracting water quality variables without prior knowledge.

Purpose of the Study:

• To develop and optimize a CNN model for classifying the Trophic State Index (TSIko) in South Korean rivers.
• To assess the performance of the CNN model against conventional machine learning approaches.
• To generate a eutrophication map for major South Korean rivers using the optimized CNN model.

Main Methods:

• A CNN model was constructed and optimized using nine years (2014-2022) of water quality data from the Han, Guem, Yeongsan, and Nakdong Rivers.
• Model performance was validated using the F1 score, a measure of classification accuracy.
• The validated CNN model was used to simulate spatial and temporal variations of the eutrophication index.

Main Results:

• The CNN model achieved high F1 scores: 0.922 (oligotrophic), 0.950 (mesotrophic), 0.964 (eutrophic), and 0.896 (hypertrophic).
• The CNN model demonstrated superior performance compared to conventional machine learning models.
• A eutrophication map revealed spatio-temporal dynamics, particularly in the Yeongsan and Nakdong Rivers.

Conclusions:

• CNN models are effective tools for analyzing eutrophication conditions across various spatial and temporal scales.
• The developed CNN model provides a robust and accurate method for classifying TSIko in major South Korean rivers.
• This approach enhances water quality management strategies by accurately mapping eutrophication dynamics.