Algal blooms forecasting with hybrid deep learning models from satellite data in the Zhoushan fishery
Wenxiang Ding, Changlin Li
Abstract
Algal blooms are increasingly frequent in coastal areas, posing a significant threat to coastal ecosystems. The Zhoushan fishery, one of the most affected regions along the Chinese coast, faces severe challenges from algal blooms. In this study, Convolutional Neural Network (CNN), Long Short-term Memory (LSTM) and hybrid CNN-LSTM deep learning models were constructed to forecast chlorophyll (Chl) concentrations and algal blooms from satellite data. The hybrid CNN-LSTM model outperformed the individual models, achieving the highest determination coefficient and the lowest root mean square error for Chl concentration forecasts. It also excelled in predicting algal blooms, with the highest probability of detection and Heidke skill score, effectively capturing the trends in algal bloom development. In areas with high Chl concentration, the Chl parameter significantly influences model forecasts, while meridional wind and current are the main influence factors in the regions with medium and low Chl concentration. The powerful algal bloom forecast provided by the hybrid CNN-LSTM model offers valuable support for the efficient management and sustainable development of the Zhoushan fishery.