Data-aware forecast of harmful algal blooms with model error

Area of Science:

• Environmental Science
• Water Resource Management
• Computational Biology

Background:

• Harmful algal blooms (HABs) pose significant risks to human health, aquatic ecosystems, and water resource utilization.
• Effective HAB management necessitates comprehensive historical analysis, accurate short-term forecasting, and proactive on-site lake management strategies.

Purpose of the Study:

• To develop and evaluate a novel modeling framework for short-term forecasting of harmful algal blooms (HABs).
• To compare the efficacy of Kalman Filter (KF) and Gaussian Processes (GP) approaches in estimating model error for HAB prediction.

Main Methods:

• A mechanistic cyanobacteria-growth model was integrated with error estimation techniques, specifically Kalman Filter (KF) and Gaussian Processes (GP).
• The framework utilized daily observations of bacteria concentration, temperature, phosphorus, nitrogen, and irradiance for model error estimation.
• The developed models were applied to forecast HABs in Cheney Reservoir, Kansas, using available high-frequency historical data.

Main Results:

• Both KF and GP variants provided accurate three- and seven-day forecasts for HABs, with accuracy diminishing over longer forecast durations.
• The Gaussian Processes (GP) variant demonstrated superior prediction accuracy compared to the Kalman Filter (KF) approach.
• The proposed modeling framework significantly outperformed model-free statistical approaches and machine learning techniques, reducing relative error by at least 50% and increasing the coefficient of determination by 70%.

Conclusions:

• The developed modeling framework offers a robust and accurate solution for short-term HAB forecasting.
• The study highlights the potential of mechanistic models combined with advanced error estimation techniques for improved water quality management.
• Both KF and GP methods provide valuable tools for HAB prediction, with GP offering higher accuracy at increased computational cost.