A novel approach for forecasting algal bloom: long short-term memory artificial neural network with exponential...

Area of Science:

• Environmental Science
• Data Science
• Marine Biology

Background:

• Sustainable environmental monitoring increasingly utilizes artificial intelligence (AI) for efficiency and cost-effectiveness.
• Algal blooms, indicated by chlorophyll-a concentrations, pose significant threats to marine ecosystems.
• Accurate forecasting of chlorophyll-a is crucial for effective management of algal bloom dynamics.

Purpose of the Study:

• To propose a novel artificial intelligence-based neural network model for forecasting chlorophyll-a concentrations.
• To introduce an arithmetic mean optimization algorithm for training the proposed neural network.
• To evaluate the model's forecasting performance against existing shallow and deep learning methods.

Main Methods:

• Development of a new long short-term memory (LSTM) network integrated with an exponential smoothing method.
• Application of an arithmetic mean optimization algorithm for training the LSTM network.
• Utilizing satellite-derived time series data from 15 Black Sea estuarine monitoring stations for forecasting chlorophyll-a.

Main Results:

• The proposed LSTM model demonstrated superior forecasting performance compared to multiple shallow and deep learning models.
• The model achieved lower root mean square error (RMSE) and mean absolute percentage error (MAPE) across most time series.
• Statistical significance tests (Friedman, Wilcoxon signed-rank) confirmed the model's performance improvements, with a mean rank of 1.06.

Conclusions:

• The developed LSTM neural network provides a reliable and effective framework for modeling chlorophyll-a dynamics.
• The AI-driven approach enhances both the accuracy and robustness of environmental monitoring.
• This method presents a promising tool for early detection and management of harmful algal bloom events in marine ecosystems.