Cyanobacteria blue-green algae prediction enhancement using hybrid machine learning-based gamma test variable...

Area of Science:

• Environmental Science
• Water Quality Monitoring
• Machine Learning Applications

Background:

• Cyanobacteria blue-green algae (CBGA) blooms pose risks to aquatic ecosystems and human health.
• Accurate modeling of CBGA concentrations is crucial for effective water resource management.
• Traditional water quality variable-based models often show limitations in predicting CBGA dynamics.

Purpose of the Study:

• To evaluate four machine learning (ML) models for CBGA prediction in two US rivers.
• To investigate the effectiveness of signal decomposition techniques in enhancing ML model performance.
• To identify the optimal combination of ML model and decomposition method for CBGA monitoring.

Main Methods:

• Developed and compared four ML models: Artificial Neural Network (ANN), Extreme Learning Machine (ELM), Random Forest Regression (RFR), and Random Vector Functional Link (RVFL).
• Applied signal decomposition methods—Empirical Mode Decomposition (EMD), Variational Mode Decomposition (VMD), and Empirical Wavelet Transform (EWT)—to water quality variables.
• Utilized decomposed components (IMFs and MRA) as input features for the ML models to improve predictions.

Main Results:

• Individually, RFR demonstrated the best predictive accuracy (R ≈ 0.944, NSE ≈ 0.884) among the tested ML models.
• Signal decomposition significantly improved the performance of all ML models.
• The combination of EMD with ANN and RFR yielded the highest accuracy (R ≈ 0.989, NSE ≈ 0.976).

Conclusions:

• Signal decomposition, particularly EMD, is a valuable preprocessing step for enhancing CBGA modeling.
• RFR and ANN models, when combined with decomposition techniques, offer superior performance for predicting CBGA concentrations.
• The proposed framework provides a robust approach for effective cyanobacteria monitoring and management in riverine systems.