HR: 0800h
AN: H31D-1341    [Abstracts]
TI: Sediment load estimation using statistical distributions with streamflow dependent parameters
AU: Mailhot, A
EM: alain_mailhot@ete.inrs.ca
AF: Institut National de la Recherche Scientifique, Centre Eau, Terre et Environnement, 490 rue de la Couronne, Québec, QC G1K9A9 Canada
AU: * Rousseau, A N
EM: alain_rousseau@ete.inrs.ca
AF: Institut National de la Recherche Scientifique, Centre Eau, Terre et Environnement, 490 rue de la Couronne, Québec, QC G1K9A9 Canada
AU: Talbot, G
EM: guillaume_talbot@ete.inrs.ca
AF: Institut National de la Recherche Scientifique, Centre Eau, Terre et Environnement, 490 rue de la Couronne, Québec, QC G1K9A9 Canada
AU: Quilbé, R
EM: renaud_quilbe@ete.inrs.ca
AF: Institut National de la Recherche Scientifique, Centre Eau, Terre et Environnement, 490 rue de la Couronne, Québec, QC G1K9A9 Canada
AB: The classical approaches to estimate sediment and chemical loads are all deterministic: averaging methods, ratio estimators, regression methods (rating curves) and planning level load estimation methods. However, none of these methods is satisfactory since they are often inaccurate and do not take into account nor quantify uncertainty. To fill this gap, statistical methods have to be investigated. This presentation proposes a new statistical method in which sediment concentration is assimilated to a random variable and is described by distribution functions. Three types of distributions are considered: Log-Normal, Gamma and Weibull distributions. Correlation between sediment concentrations and streamflows is integrated to the model by assuming that distribution parameters (mean and coefficient of variation) are related to streamflow using several different functional forms: exponential, quadratic and power law forms for the mean, constant and linear for the coefficient of variation. Parameter estimation is realized through maximization of the likelihood function. This approach is applied on a data set (1989 to 2004) from the Beaurivage River (Quebec, Canada) with weekly to monthly sampling for sediment concentration. A comparison of different models (selection of a distribution function with functional forms relating the mean and the coefficient of variation to streamflow) shows that the Log-Normal distribution with power law mean and coefficient of variation independent of streamflow provides the best result. When comparing annual load results with those obtained using deterministic methods, we observe that ratio estimators values are rarely within the [0.1, 0.9] quantile interval. For the 1997-2004 period, ratio estimator values are almost systematically smaller than the 0.1 quantile. This could presumably be due to the small number of sediment concentration samples for these years. This study suggests that, if deterministic methods such as the ratio estimator are useful as a first approximation to estimate a mean annual load over a long period, they are clearly not adapted to estimate loads at a shorter time step, especially for infrequent concentration sampling and when correlation between streamflow and sediment concentration is poor. The properties of the statistical distributions represent interesting alternatives to single, intermittent observations with unknown uncertainties (i.e. not quantified). This approach is particularly well suited for the calibration of watershed models and will be applied for other contaminants such as nutrients and fecal coliforms.
DE: 1815 Erosion
DE: 1847 Modeling
DE: 1862 Sediment transport (4558)
DE: 1871 Surface water quality
SC: Hydrology [H]
MN: Fall Meeting 2005