Algal bloom forecasting leveraging signal processing: A novel perspective from ensemble learning.

Area of Science:

• Environmental Science
• Data Science
• Applied Mathematics

Background:

• Accurate forecasting of algal blooms is crucial for effective management and mitigation strategies.
• Standalone models struggle with the complex time-frequency dynamics of algal blooms.
• Existing methods often lack the robustness needed for real-world environmental monitoring.

Purpose of the Study:

• To develop an advanced ensemble framework for improved algal bloom forecasting.
• To integrate signal processing techniques with machine learning for enhanced predictive accuracy.
• To create a robust model for forecasting algal dynamics across various temporal and spatial scales.

Main Methods:

• Utilized the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to decompose non-stationary algal dynamics.
• Employed an ensemble of four distinct machine learning models to learn from the decomposed components.
• Developed the CEEMDAN-Hybrid-Ensemble (CHES) model, combining signal processing and machine learning.

Main Results:

• The CEEMDAN-Hybrid-Ensemble (CHES) model significantly improved forecasting performance, increasing validation R² by an average of 75% compared to standalone machine learning models.
• Achieved high forecasting accuracy across multiple time resolutions (hourly, daily, biweekly) in diverse water bodies (rivers and lakes).
• Demonstrated stable multi-step forecasting capabilities with high validation R² and low root-mean-square-error (RMSE).

Conclusions:

• The integration of CEEMDAN with an ensemble of machine learning models offers a powerful approach for accurate algal bloom forecasting.
• The developed CHES model provides a robust and versatile tool for environmental monitoring and early warning systems.
• This study highlights the significant benefits of ensemble methods in capturing complex environmental dynamics.