University of Twente Student Theses


Comparing model performance between the hydrodynamic models SOBEK and TYGRON

Renswoude, R.C. (2020) Comparing model performance between the hydrodynamic models SOBEK and TYGRON.

[img] PDF
Abstract:Since May 2018, TYGRON presents a 2D hydrodynamic model in their geo-design platform. TYGRON proves to be valuable in modelling overland flow in urban and rural areas. However, the model performance of TYGRON in an applied river case is not fully tested. To study the model performance of TYGRON in a river study, a comparison is made with the reference case of the Overijsselse Vecht in the 1D (main channel)/2D (floodplains) SOBEK model of Regional Water Authority (R.W.A.) Vechtstromen between the German border-weir Hardenberg. The goal of this thesis is to analyse the extent to which TYGRON can be used for a river study and which practical/hydrodynamic problems are encountered. This is done by comparing the following 5 aspects of model performance: the accurate simulation of 1) flood water levels, 2) inundation and 3) flow velocities, 4) realistic model sensitivity to the calibrated parameters and 5) how to implement a measure. 1) To study the performance of TYGRON to accurately simulate water levels, TYGRON is calibrated by changing the hydraulic roughness of the main channel and the floodplains for the 1/4Q (average winter scenario) and T10 (flood frequency of 1/10 years) discharge scenario, respectively. Reaching the design water level in the 1/4Q scenario was not possible since the difference between the simulated water level and the design water level was 1.67 m at the main channel roughness of 0.025 m1/3/s (lowest roughness value for a river described by the table of Chow, (1959)). Calibration of the T10 scenario in TYGRON results in a smaller difference between the simulated water level (10.66 m) and the design water level (10.36 m) compared to the SOBEK model (9.92 m). The malfunctioning of weirs in a wide river section is one of the reasons why the 1/4Q scenario was not possible to calibrate on hydraulic roughness. Weirs connect with one grid cell centre point causing the flow to be simulated past the weirs instead of over the weirs when the river is wider than the connected grid cell. Another reason can be allocated to the larger simulated water levels in TYGRON, namely the high influence of numerical viscosity in a square grid cell. A square grid may increase the influence of numerical viscosity in a meandering river profile (i.e. the course of Overijsselse Vecht) and hence result in large simulated water levels. Furthermore, it is not possible to obtain the actual simulated water levels as a result output in TYGRON. In TYGRON water levels are defined by the sum of the simulated water depth and bed level. To retrieve water levels from the grid overlay the measuring tool must be used. However, the simulated water depths are simulated based on reconstructed bathymetry. This results in irregular water levels in the length profile since the original bathymetry data is exported with the measuring tool. 2) Five inundation images in the floodplains of De Haandrik and Hardenberg of the 2018 flood event are used to validate the performance to simulate inundation. Although the discharge event of 2018 is overestimated in TYGRON, another clear difference can be seen from the SOBEK simulation. In SOBEK it is difficult to relate the inundation from the images to the simulated inundation on a specific location because of the large 25x25 m grid results in a lower bathymetry accuracy and hence a rough inundation prediction. The 2018 event in TYGRON is overestimated, locally TYGRON simulates the inundation according to the flood images. 3) The performance of the flow velocities is qualitatively analysed in the river bend and floodplains at Hardenberg. In literature, it is described that high flow velocities occur in the outer bend of the main channel and gradually decrease towards the inner bend (e.g. Luchi et al., 2011; Sukhodolov, 2012). Due to the missing 2D flow components in the main channel, SOBEK is not fit to correctly predict the expected flow velocity pattern in the main channel of the Overijsselse Vecht. Furthermore, the low resolution of the used grid in SOBEK (25x25 m) results in an over-discretization of the bathymetry and hence the flow velocities in the floodplains are generalised. The TYGRON model computes unexpected high flow velocities at both sides of the main channel. The steep slope near the banks of the main channel causes a wrong estimation of the flow velocity between two adjacent cells resulting in an overshoot. The overshoot is inherent to the used algorithm in the 2D scheme which is currently under development at TYGRON (TYGRON, 2019). 4) A sensitivity analysis is executed on the calibrated parameters (i.e. weir dimensions, hydraulic roughness and grid cell size). Analysing the flow over the weir indicates that the weirs in TYGRON are not correctly implemented. The sensitivity analysis on the weir’s dimensions shows that at De Haandrik the discharge over the weir is highly influenced by changing the dimensions in the 1/4Q scenario. Three floodplain roughness scenarios were analysed in the T1 and T10 discharge scenario in TYGRON and SOBEK. Before this analysis can be executed the Chézy coefficients from SOBEK are converted to Manning values to implement them in TYGRON (TYGRON can only consider Manning roughness values). This analysis showed that the SOBEK model is not sensitive by changing the Chézy coefficient with 20%, the water levels are only slightly increased in the T10 scenario. However, for TYGRON, in contrast to SOBEK, changes in the hydraulic roughness of the floodplains had a major influence on the simulated water levels. Three different grid cell sizes (1x1, 2x2 and 5x5 m) were analysed in TYGRON for the 1/4Q and T10 discharge scenarios. The results show that in the 1/4Q scenario the water level slope is more similar to the water level slope simulated by SOBEK when using a 1x1 m than a 2x2 m grid cell size. In the T10 discharge scenario, the 1x1 m grid shows comparable simulated water levels to the 2x2 m grid. However, the computation time in TYGRON is significantly increased from 1 hour to 4-6 hours when using a 1x1 m compared to a 2x2 m grid. Simulation with a 5x5 m grid shows a distorted result and as some of the inlets (functioning as upstream boundary condition) were turned off by overlapping connection points. 5) To analyse how easily a measure can be implemented in TYGRON and predict the hydraulic effects, a side-channel is implemented in the case of the Overijsselse Vecht. TYGRON can change the used elevation model by lowering/raising the absolute or relative height values and therefore, only separated elements in the height can be changed. This makes it difficult to adjust a side-channel since corrections to the elevation model cannot be undone. On the other hand, there is an option in TYGRON to exchange geodata such as height elements (GeoTIFF) and object elements (GeoJSON). This makes it possible to design a certain measurement in another software program (e.g. GIS or AutoCAD) and implement the design in TYGRON to analyse the hydraulic effects. Based on this study, it can be concluded that, at the moment, TYGRON is not suitable for a river study like the river Overijsselse Vecht, although extreme discharge conditions can be predicted with more accuracy compared to average and low discharge scenarios where the influence of river weirs is significant. The following possible reasons can be mentioned why TYGRON is not yet suitable for a river study: 1) the absence of water levels as result output, 2) incorrectly simulation of weir dependent river sections, 3) the non-optimal functioning of boundary conditions and 4) the influence of numerical viscosity by the square grid shape. In the update of 9 May 2020 of the current TYGRON model, structures can be implemented over an area instead of one grid cell centre point, which may improve the simulation of flow in river scenarios where the influence of weirs is significant. In case TYGRON wants to expand the application of their water module in river studies, it is recommended to include water levels as result output. Furthermore, grid cell sizes lower than 1x1 m probably improve the flow distribution over the grid cells in the downstream direction and hence decrease the influence of numerical viscosity and friction in a square grid cell. However, additional problems in hydrodynamic modelling may occur when simulating in such small grid cell sizes (e.g. overshoot in the simulated flow velocities). It is recommended to analyse if flow distribution indeed improves in a square high-resolution grid and which hydraulic problems may occur at simulating in such high-resolution.
Item Type:Essay (Master)
Faculty:ET: Engineering Technology
Programme:Civil Engineering and Management MSc (60026)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page