Accurate estimation of salinity is critical for water resource management in deltas and coasts. Traditional methods such as numerical models often rely on physical transport processes, leading to significant uncertainty in modeling parameter estimation. Data-driven models like long short-term memory (LSTM) offer an effective alternative. This study implements the LSTM model to estimate salinity at multiple locations in the Vietnamese Mekong Delta. Hourly tidal data from Vung Tau and discharge data from Chau Doc and Tan Chau were applied as inputs, with salinity data from six locations as outputs. The model is trained and tested using data collected from 01/01/2014 to 30/06/2017, before being applied. The model’s accuracy was evaluated using several statisitcal indicators, including Nash–Sutcliffe efficiency (NSE), Pearson’s correlation coefficient (r), mean error (ME), mean absolute error (MAE), and using root-mean-square error (RMSE). The findings indicated that the LSTM model accurately reproduced salinity, with dimensionless errors between 0.84 and 0.99, and dimensional errors from -0.31 to 0.38 PSU. These results demonstrate the reliability and generalizability of LSTM models for salinity estimation in the Vietnamese Mekong Delta. Moreover, the integration of wavelet and wavelet coherence analyses provided novel insights into the temporal structure and multiscale interactions of salinity with key hydrodynamic drivers, such as river discharge and tidal forcing. This study contributes to the growing body of literature advocating for hybrid and machine learning-based modeling approaches in hydro-environmental science, offering scalable, interpretable, and efficient tools for forecasting and decision support in data-scarce coastal regions worldwide.