Over recent decades,increasing anthropogenic activities in the Strait of Georgia(SOG)have heightened the demand for enhanced environmental protection measures.This study presents a novel approach to improve the predic...Over recent decades,increasing anthropogenic activities in the Strait of Georgia(SOG)have heightened the demand for enhanced environmental protection measures.This study presents a novel approach to improve the prediction accuracy of water temperature and salinity dynamics in the strait through advanced machine learning techniques,offering valuable theoretical support for environmental planning,ecosystem management,and sustainable fisheries.We developed an innovative forecasting model by integrating empirical mode decomposition(EMD)with long short-term memory(LSTM)neural networks.The EMD-LSTM model demonstrated exceptional performance,achieving a strong Pearson correlation coefficient(>0.8)with observational data across three monitoring stations.Comparative analysis revealed the model’s superior predictive accuracy and adaptability over conventional backpropagation neural network(BPNN)and standalone LSTM approaches,with its advantages becoming increasingly evident in extended forecasting periods.The integration of time-domain multi-scale analysis with neural network architecture not only improved forecasting precision but also enhanced model interpretability by elucidating the spatial-temporal variations in water temperature and salinity patterns across different monitoring sites.This advanced forecasting framework shows significant potential for supporting high-precision marine environmental predictions in the SOG region,contributing to more effective marine resource management and conservation strategies.展开更多
基金The National Natural Science Foundation of China under contract Nos 52239005,52322109 and U22A2012the R&D Program of Guangdong Provincial Department of Science and Technology under contract No.2024B1212040004.
文摘Over recent decades,increasing anthropogenic activities in the Strait of Georgia(SOG)have heightened the demand for enhanced environmental protection measures.This study presents a novel approach to improve the prediction accuracy of water temperature and salinity dynamics in the strait through advanced machine learning techniques,offering valuable theoretical support for environmental planning,ecosystem management,and sustainable fisheries.We developed an innovative forecasting model by integrating empirical mode decomposition(EMD)with long short-term memory(LSTM)neural networks.The EMD-LSTM model demonstrated exceptional performance,achieving a strong Pearson correlation coefficient(>0.8)with observational data across three monitoring stations.Comparative analysis revealed the model’s superior predictive accuracy and adaptability over conventional backpropagation neural network(BPNN)and standalone LSTM approaches,with its advantages becoming increasingly evident in extended forecasting periods.The integration of time-domain multi-scale analysis with neural network architecture not only improved forecasting precision but also enhanced model interpretability by elucidating the spatial-temporal variations in water temperature and salinity patterns across different monitoring sites.This advanced forecasting framework shows significant potential for supporting high-precision marine environmental predictions in the SOG region,contributing to more effective marine resource management and conservation strategies.
