Development of a regional model for catchment-scale suspended sediment yield estimation in ungauged rivers of the Lower Mekong Basin

Abstract

There have been many attempts to accurately predict water and sediment yields from river basins, an issue which is particularly critical for sediments as limited monitoring stations exists worldwide. This research examines the performance of catchment similarity methods, based on basin physical characteristics, to estimate monthly suspended sediment yield time series (SSY). The analysis was based on datasets from 16 gauged catchments in the Lower Mekong Basin, Southeast Asia. Different methods were compared to determine the similarity between the catchments: (1) the rank-accumulated similarity technique (CSI1); (2) the sum of absolute differences between catchment descriptors, normalized by their ranges (CSI2); (3) the Euclidean distance computed by multidimensional scaling in which each set of catchment descriptors is normalized by its standard deviation (CSI3); and (4) the Euclidean distance computed by multidimensional scaling in which catchment descriptors are normalized to a specific range (CSI4). Catchment descriptors include information on hydrology, topography, morphology, land use/cover and soil type. The evaluation was conducted using the leave-one-out re-sampling procedure. The concept of this procedure is that each of the gauged catchments is in turn considered as pseudo ungauged catchment, while the remaining ones are regarded as candidate donor catchments. The model predictive accuracy was measured by three statistical indicators: Nash–Sutcliffe efficiency (NSE), percent bias (PBIAS), and the ratio of the root mean square error to the standard deviation of observed data (RSR). A satisfactory result (NSE>0.50, PBIAS±55% and RSR≤0.70) was obtained in all 16 modeled catchments when the regionalization was performed using the CSI4 measure. Dependent on the statistical distribution pattern of catchment descriptor datasets being used, CSI3 was slightly inferior with 15 satisfactorily modeled catchments. CSI2, which was less sensitive to outlier candidate donor catchments, performed well in 14 catchments, inferior to both CSI4 and CSI3. With only 10 satisfactorily modeled catchments, CSI1 provided the worst regionalization solution. Relied upon the high quality results offered by the CSI4-based regionalization approach, a regional SSY model, i.e. an integration of 16 locally calibrated models, was established. With a strong predictive power, it could be applied simply to estimate catchment-scale SSY in the ungauged rivers of the Mekong basin.