Author(s): Ngo Tran Sang; Doan Van Binh; Pham Thi Mai Thy; Sameh Ahmed Kantoush; Tran Thi An; Nguyen Dinh Vuong

Linked Author(s): SAMEH KANTOUSH

Keywords: Salinity intrusion; Remote sensing; Machine learning; Vietnamese Mekong Delta; Sentinel-2

Abstract: Salinity intrusion is among the most serious environmental problems, shaping the livelihoods and ecosystems of worldwide deltas. In coastal and estuarine areas such as the Vietnamese Mekong Delta (VMD)—a rice bowl of Vietnam—salinity intrusion has emerged as a serious constraint on agricultural productivity and sustainable development. The phenomenon exhibits highly irregular spatiotemporal patterns, mainly driven by upstream hydropower operations, climate change, extreme weather events, and sea -level rise. This study evaluated salinity intrusion in the VMD by integrating machine learning onto Sentinel-2 imagery. We analyzed 1,092 Sentinel-2 images for 2015–2023 and salinity data from 40 hydrological stations. Combining them, 2,230 salinity data were used to develop three advanced machine learning algorithms, namely Decision Tree (DT), Support Vector Regression (SVR), and Random Forest Regressor (RFR), with 70% dataset for training (1,561) and 30% for testing (669). We found that the RFR model outperformed the other two models, achieving R² = 0.84, NSE = 0.84, RMSE = 3.65 g/L, pBIAS = 3.51%, and MAE = 2.88 on the test dataset. The results highlight the great potential of combining remote sensing (RS) and machine learning (ML) techniques for salinity monitoring and prediction, supporting adaptive water resources management and climate resilience in the VMD.

DOI: https://doi.org/10.64697/iahr.proc.ise2026.abs.184

Year: 2026