Long-term daily prediction of saltwater intrusion based on the long memory-double autoregressive model

作者：Kairong Lin, Shuai Wei, Tongfang Li, Tian Lan, Xinjun Tu · 发表于：Journal of Hydroinformatics · 年份：2025 · DOI：10.2166/hydro.2025.047 · 被引用次数：2 · 研究领域：Hydrological Forecasting Using AI、Underwater Acoustics Research、Water Quality Monitoring Technologies

ABSTRACT Existing research on saltwater intrusion faces key challenges, including limited accuracy and difficulty in long-term prediction. This study introduces a long memory-double autoregressive (LMDAR) model based on fractional differencing, designed for daily prediction of hours exceeding the chloride threshold. Four salinity monitoring stations in the Pearl River Estuary were selected as case study stations. The model's performance was evaluated and compared with the double autoregressive (DAR) model, the long short-term memory (LSTM) model, and the gated recurrent unit (GRU) model using Nash–Sutcliffe efficiency (NSE), mean absolute error, and percent bias across different lead times. The results indicate that in short-term predictions (1 and 3 days), the prediction accuracy of the LMDAR, LSTM, and GRU models is comparable, with NSE values exceeding 0.7 at all four example stations and reaching up to 0.93. In long-term predictions (7, 15, and 30 days), the LMDAR model consistently achieved NSE values above 0.51 across the four stations, with the highest reaching 0.82. The LMDAR thus proves to be an effective tool for predicting hours exceeding the chloride threshold, demonstrating strong short-term performance and reliable long-term prediction capability.