Artificial intelligence-based water quality assessment from remote sensing images – a case study of the Thi Vai river

Abstract

The Thị Vải River is an important estuarine system in southeastern Vietnam that faces increasing water-quality pressure from industrialization and human activities, particularly in relation to total suspended solids (TSS). Conventional monitoring is often limited by sparse sampling and high operational costs, making remote sensing and machine learning promising alternatives for water-quality assessment. This study evaluated the performance of Random Forest (RF) and Gradient Tree Boosting (GTB) models for estimating TSS using Landsat-8 and Sentinel-2 surface reflectance imagery integrated with in-situ observations collected during 2017–2021. Satellite data were processed in Google Earth Engine, where water bodies were extracted using the Normalized Difference Water Index (NDWI), and model performance was assessed using R2, RMSE, and MAE. Temporal transferability was further examined by applying trained models to imagery from different years. The results showed that RF consistently outperformed GTB across dry and rainy seasons, achieving R2 values of 0.60–0.90 compared with 0.10–0.75 for GTB. RF also demonstrated moderate temporal transferability, indicating potential for TSS prediction in data-limited environments. Spatial prediction maps revealed elevated TSS concentrations near industrial and navigation-influenced areas. Overall, the findings demonstrate the feasibility of integrating remote sensing and machine learning for cost-effective and spatially continuous TSS monitoring in estuarine river systems.