Applying bootstrap resampling to quantify uncertainty in fluvial suspended sediment loads estimated using rating curves

Knowledge of suspended sediment loads in rivers is essential for catchment management purposes and study of landscape evolution. Obtaining such information is difficult because of the often sparse physical sampling of sediment concentrations and high temporal variability in discharge and sediment concentration, as well as strong hysteretic effects and temporal and spatial variations in catchment condition. Here bootstrap and Monte Carlo resampling techniques are used to calculate suspended sediment loads using sediment rating curves. The method proposed avoids the need to apply a bias correction factor to load estimates calculated using rating curves defined by least squares regression of log-transformed data. The algorithm also quantifies uncertainty in suspended sediment load estimates arising from uncertainty in the shape of the rating curve and the residual scatter in the data. Applied to 11 gauging stations in the catchment of Lake Burragorang in Australia, the method produced suspended sediment yields consistent with other Australian observations. The majority of the sediment delivered to the lake comes via the lower Wollondilly River (251(-49)(+264) kt a(-1)), with lesser contributions from the Kowmung (114(-30)(+62) kt a(-1)) and Cox's (63(-14)(+81) kt a(-1)) rivers. Despite uncertainty in the shape of the rating curve at high flows, days with the largest flow clearly transport most of the suspended sediment delivered to the reservoir over the long term, with 40-85% of the total load being transported in < 1% of the time. This bootstrap-based method could be applied to the calculation of other constituent fluxes in rivers where discrete sampling of constituent concentration serves as the principal data.