In this study, a new long short-term memory with an exponential smoothing method is proposed, and an arithmetic mean optimization algorithm-based training strategy is introduced for the proposed neural networks.
To evaluate the effectiveness of this framework, comprehensive experiments are conducted by comparing the proposed neural network with several shallow and deep learning models.
The results show that the proposed neural network model achieves superior forecasting performance in terms of root mean square error and mean absolute percentage error across the majority of the analysed time series.
The proposed neural network model attains the best overall ranking with a mean rank of 1.06 and significantly outperforms all competing models under the Holm–Bonferroni correction.
The findings indicate that the proposed neural network improves both forecasting accuracy and robustness, providing a reliable and effective framework for modelling chlorophyll-a dynamics.
Artificial intelligence-based methods have gained increasing attention for sustainable environmental monitoring due to their efficiency and low operational cost. In this study, a new long short-term memory with an exponential smoothing method is proposed, and an arithmetic mean optimization algorithm-based training strategy is introduced for the proposed neural networks. The resulting proposed neural network method is used to forecast chlorophyll-a concentrations, a key indicator of algal bloom dynamics, using satellite-derived time series data from 15 estuarine monitoring stations in the Black Sea region. To evaluate the effectiveness of this framework, comprehensive experiments are conducted by comparing the proposed neural network with several shallow and deep learning models. The results show that the proposed neural network model achieves superior forecasting performance in terms of root mean square error and mean absolute percentage error across the majority of the analysed time series. Furthermore, statistical significance analyses based on the Friedman and Wilcoxon signed-rank tests confirm that the observed performance improvements are statistically significant. The proposed neural network model attains the best overall ranking with a mean rank of 1.06 and significantly outperforms all competing models under the Holm–Bonferroni correction. The findings indicate that the proposed neural network improves both forecasting accuracy and robustness, providing a reliable and effective framework for modelling chlorophyll-a dynamics. This approach offers a promising tool for the early detection and management of algal bloom events in marine ecosystems.
