Modeling input errors to improve uncertainty estimates for one-dimensional sediment transport models

Abstract

Bayesian methods have recently been applied to one-dimensional sediment transport models to assess the uncertainty in model predictions due to uncertainty in the parameter values. However, these approaches neglect any uncertainties in the model inputs, which might play a substantial role. The objective of this research is to include uncertainties in sediment transport model inputs and evaluate their contributions to the overall uncertainty in the model predictions. To accomplish this goal, simple error models are developed for the input data and integrated into an existing Bayesian method. Five types of input data are considered: input discharges, rating curves, vertical and horizontal distances in cross-sections, and benchmark elevations that define the longitudinal profile of the reach. The input errors are modeled using Gaussian distributions, and the means and standard deviations are treated as uncertain parameters that are estimated within the Bayesian framework. The Bayesian approach is coupled to the Sedimentation and River Hydraulics-One Dimension (SRH-1D) model and used to simulate a 23-km reach of the Tachia River in Taiwan. When input uncertainties are included, the prediction ranges change substantially and cover more of the available observations, which suggests the uncertainty is better represented when input errors are considered. The results also indicate that the errors in the benchmark elevations have the largest impact on the uncertainty of the predictions among those considered.