Investigating saltwater intrusion is vital for optimal use of estuarine water resources. Presently, diverse data-driven models, mainly neural network models, have been employed to predict tidal estuarine salinity. However, the high nonlinearity, randomness, and instability of salinity sequences pose challenges for accurate estuarine salinity forecasting. In this paper, a multi-factor salinity prediction model using an enhanced Long Short-Term Memory (LSTM) network was proposed, based on measured data from Cangqian and Qibao stations in the Qiantang Estuary during 2011-2012. To improve prediction accuracy, input variables of the model were determined through Grey Relational Analysis (GRA) combined with estuarine dynamic analysis, and hyperparameters were optimized using a multi-strategy Improved Sparrow Search Algorithm (ISSA). Additionally, the model was applied to forecast salinity under different runoff conditions, analyzing the sensitivity of salinity to upstream discharge. Experimental result shows that compared to other models (BP, GRU, LSTM, SSA-LSTM), the new proposed ISSA-LSTM model has smaller errors and higher prediction accuracy, with NSE improving over 5% and other metrics (MAP, MAPE, RMSE) improving over 10%. Thus, the model provides a practical solution for the rapid and precise prediction of estuarine salinity.