Scholarlink

<p>The estuary is an important ecological system that faces significant challenges from salinity intrusion, which has had significant effects on human activities. Accurate prediction of water salinity plays an important role in the contribution of irrigation water to agricultural development, especially in the context of climate change and the rise of sea level. The objective of this study is the application of machine learning, namely long- and short-term memory (LSTM), gated recurrent units (GRUs), and hybrid machine learning (LSTM-GRU and LSTM-GRU- Sailfish optimiser algorithm (SFO)) to predict water salinity in the Hau River estuary, in Mekong detail in Vietnam. More specifically, (i) this study develops a hybrid model by combining LSTM, GRU, and LSTM-GRU networks, optimised by the Sailfish Optimizer (SFO) algorithm. (ii) this study compares the proposed models to determine the best model to predict soil salinity in the study area; (iii) this study analyses the trend of salinity change to support decision makers or local authority for the optimisation of water resource management. In this study, the water level and the discharge from the river were used to predict the water and 10 days were selected to predict one, five, seven and ten days ahead. seven, The results showed that the LSTM-GRU-SFO model outperformed other models in terms of accuracy. Specifically, this model achieved the best performance with the value of R² = 0.894, RMSE = 3.18, MAE = 1.37 for the prediction and R² = 0.880, RMSE = 3.11, MAE = 1.34 for the prediction of five days, R<sup>2</sup> = 0.854, RMSE = 3.46, MAE = 1.47 for the prediction of seven days and R² = 0.815, RMSE = 3.87, MAE = 1.42 for the prediction of ten days. This accuracy presents that the LSTM-GRU-SFO model effectively captures not only temporal salinity but also high reliability over all prediction horizons. The results of this study justify the effectiveness of machine learning to predict water salinity in estuaries. The knowledge obtained in this study can provide important support for water irrigation intelligence management strategies not only in Vietnam but also in other regions of the world.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Salinity Intrusion Prediction in the Estuary Using Machine Learning: Vietnam’s Mekong Delta Tested for Global Study

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Huu Duy Nguyen

摘要

The estuary is an important ecological system that faces significant challenges from salinity intrusion, which has had significant effects on human activities. Accurate prediction of water salinity plays an important role in the contribution of irrigation water to agricultural development, especially in the context of climate change and the rise of sea level. The objective of this study is the application of machine learning, namely long- and short-term memory (LSTM), gated recurrent units (GRUs), and hybrid machine learning (LSTM-GRU and LSTM-GRU- Sailfish optimiser algorithm (SFO)) to predict water salinity in the Hau River estuary, in Mekong detail in Vietnam. More specifically, (i) this study develops a hybrid model by combining LSTM, GRU, and LSTM-GRU networks, optimised by the Sailfish Optimizer (SFO) algorithm. (ii) this study compares the proposed models to determine the best model to predict soil salinity in the study area; (iii) this study analyses the trend of salinity change to support decision makers or local authority for the optimisation of water resource management. In this study, the water level and the discharge from the river were used to predict the water and 10 days were selected to predict one, five, seven and ten days ahead. seven, The results showed that the LSTM-GRU-SFO model outperformed other models in terms of accuracy. Specifically, this model achieved the best performance with the value of R² = 0.894, RMSE = 3.18, MAE = 1.37 for the prediction and R² = 0.880, RMSE = 3.11, MAE = 1.34 for the prediction of five days, R2 = 0.854, RMSE = 3.46, MAE = 1.47 for the prediction of seven days and R² = 0.815, RMSE = 3.87, MAE = 1.42 for the prediction of ten days. This accuracy presents that the LSTM-GRU-SFO model effectively captures not only temporal salinity but also high reliability over all prediction horizons. The results of this study justify the effectiveness of machine learning to predict water salinity in estuaries. The knowledge obtained in this study can provide important support for water irrigation intelligence management strategies not only in Vietnam but also in other regions of the world.