Port and channel sedimentation: quantifying uncertainties using an emulator

Dechichi, L. (2016) Port and channel sedimentation: quantifying uncertainties using an emulator.

Abstract:All around the world, expensive computationally models have been used in feasibility studies in order to predict the amount of port siltation in a certain area. Depending on how many millions of cubic meters of fine sediments brought into the navigation channel (by the tides, waves and rivers) need to be dredged, the construction of such a port can become impractical due to high maintenance costs. But if the uncertainties associated with this natural process are taken into account, for a better prediction, then those models cannot be used due to the many runs required to compute these uncertainties. The goal of this study is to quantify these uncertainties in port siltation using an emulator. An emulator is a simpler model which aims to generate an output as close as possible from the original model but with a lower computational time. In order to reach the goal of this study, a model and two emulators with a lower horizontal resolution will be used. The model and the two emulators are termed as sixteen grid cells model, eight grid cells model and four grid cells model due to the number of grid cells along their respective navigation channel width. Moreover, four steps will be applied as follows. Firstly, the uncertain input parameters are selected with help of expert judgement and literature review. Secondly, we choose the one-at-a-time method (OAT) to determine among the uncertain parameters which contributed most to the uncertain in the model output, named as qualitative sensitivity analysis. Thirdly, we apply the Latin Hypercube Sampling method (LHS) to sample the previously selected input parameters which are used in the model runs for uncertainty quantification. Finally, we investigate mapping functions that, along with the emulator, are capable to translate the amount of uncertainty in port siltation computed by the model. Regarding the sensitivity analysis, the uncertain input parameters to which the siltation is the most sensitive are: bed roughness, settling velocity and significant wave height, according to the model and the two emulators. However, the four grid cell model presented very low siltation values when compared to the sixteen and eight grid cells models. Because of this, it is not used in uncertainty quantification. With respect to the uncertainty quantification, the eight grid cells model present a smaller amount of uncertainty in the model output when compared to the sixteen grid cells model. This is an unrealistic result proving that the eight grid cells emulator cannot be used for uncertainty quantification. As for the identification of mapping functions, four mapping functions are identified and verified through box plots. From this study, three important conclusions can be higlighted. First, the eight grid cells model is a good replacement for the sixteen grid cells model with respect to siltation computation in the navigation channel. Second, the four grid cells model can be used for a ‘parameter screening’ in qualitative sensitivity analysis. Third, the eight grid cell model can be used for uncertainty quantification if the specified mapping functions are applied. Summarizing, it is possible to quantify the uncertainty in a navigation channel using an emulator as long as a mapping function is applied. With attention to the fact that the channel width should be around 10 grid cells, otherwise the hydrodynamics and morphodynamics are not well represented by the emulator.
Item Type:Essay (Master)
Faculty:ET: Engineering Technology
Subject:56 civil engineering
Programme:Civil Engineering and Management MSc (60026)
Link to this item:http://purl.utwente.nl/essays/71008
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page