Nguyen S. Q., Nguyen L. C., Ngo-Duc T., Ouillon Sylvain. (2024). Applying a machine learning-based method for the prediction of suspended sediment concentration in the Red river basin. Modeling Earth Systems and Environment, [Early access], p. [18 p.]. ISSN 2363-6203.
Titre du document
Applying a machine learning-based method for the prediction of suspended sediment concentration in the Red river basin
Nguyen S. Q., Nguyen L. C., Ngo-Duc T., Ouillon Sylvain
Source
Modeling Earth Systems and Environment, 2024,
[Early access], p. [18 p.] ISSN 2363-6203
Knowledge of sediment transport is important to understand the transportation and recycling of elements and matter in the Earth system. Usually, sediment transport in rivers is characterized by suspended sediment concentration (SSC) and river discharge (Q). However, SSC measurements are often inadequate in many river systems, such as the Red River basin in Vietnam. In this study, we performed a Tributary-based Downstream gauge Estimation (TDE) machine learning (ML) approach to estimate SSC at Son Tay hydrological station based on Q and SSC monthly data from three upstream stations, one per tributary of the Red River over a 14-years period (2000-2013). A comparative analysis of four ML algorithms, including Multiple Linear Regression (MLR), Elastic Net (EN), Random Forest (RF), and Support Vector Machines (SVM) was conducted. Results showed that when using both Q and SSC of the three upstream stations, the SVM algorithm with linear kernel exhibited the highest accuracy (r2 = 0.87 and RMSE = 64.7 g m-3). The performance of the TDE-ML was seasonally dependent, with higher accuracy in the high-flow period. This approach also revealed that SSC measured at Yen Bai station (Thao River) had the highest contribution to the prediction of SSC at Son Tay station meanwhile Vu Quang station (Lo River) contributed the least to downstream SSC. Furthermore, new dams have been impounded during the 14-years period. Although the global performance of the RF method was slightly less than SVM with linear kernel, it was the only one able to fairly estimate SSC in the most recent 6-years period affected by new forcing.
Plan de classement
Sciences fondamentales / Techniques d'analyse et de recherche [020]
;
Hydrologie [062]
;
Géologie et formations superficielles [064]
