Stroomatlas sluis van Wintam

Transcription

1 departement Mobiliteit en Openbare Werken Stroomatlas sluis van Wintam Deelrapport 1 NUMERIEK 2D MODEL _101 WL Rapporten Vlaamse overheid

2 Stroomatlas sluis van Wintam Deelrapport 1 Numeriek 2D model Maximova, T.; Vanlede, J.; Verwilligen, J.; De Maerschalck, B.; Verwaest, T; Mostaert, F. July 2014 WL2014R12_101_1 F-WL-PP10-1 Versie 04 GELDIG VANAF: 12/11/2012

3 This publication must be cited as follows: Maximova, T.; Vanlede, J.; Verwilligen, J.; De Maerschalck, B.; Verwaest, T; Mostaert, F. (2014). Stroomatlas sluis van Wintam: Deelrapport 1 Numeriek 2D model. Versie 4.0. WL Rapporten, 12_101. Waterbouwkundig Laborartorium: Antwerpen, België. Waterbouwkundig Laboratorium Flanders Hydraulics Research Berchemlei 115 B-2140 Antwerp Tel. +32 (0) Fax +32 (0) waterbouwkundiglabo@vlaanderen.be Nothing from this publication may be duplicated and/or published by means of print, photocopy, microfilm or otherwise, without the written consent of the publisher. F-WL-PP10-1 Versie 04 GELDIG VANAF: 12/11/2012

4 Document identification Title: Stroomatlas sluis van Wintam: Deelrapport 1 Numeriek 2D model Customer: Waterwegen en Zeekanaal NV Afdeling Zeekanaal Ref.: WL2014R12_101_1 Keywords (3-5): Numerical model, Wintam, Scheldt estuary Text (p.): 53 Appendices (p.): 70 Confidentiality: Yes Exceptions: Customer Internal Flemish government Released as from: January 2015 No Available online Approval Author Reviser Project Leader Research & Consulting Manager Head of Division Maximova, T. Vanlede, J. Vanlede, J. Verwaest, T. Mostaert, F. Verwilligen, J. De Maerschalck, B. Revisions Nr. Date Definition Author(s) /11/2013 Concept version Maximova, T /12/2013 Substantive revision Vanlede, J.; De Maerschalck, B.; Verwilligen, J /05/2014 Revision customer Van Hoecke, N /07/2014 Final version Vanlede, J. Abstract In this study a detailed model is constructed for the lock of Wintam. For this area it is important to have a well refined grid, which can represent the flow to the lock correctly. Until now this area was simulated only in the NEVLA model, which includes the entire Scheldt estuary. The grid resolution of the overall NEVLA model is too coarse near Wintam for a meaningful application of the model results in the ship simulator of Flanders Hydraulics Research (FHR). Therefore, it was necessary to develop a detailed model for this area. The grid of this model should have a higher resolution than the NEVLA grid and should follow the quay walls better. The detailed model is developed in the TELEMAC software. The use of an unstructured grid in TELEMAC allows a local grid refinement in the study area and results in an accurate representation of the quay walls in the model. The model is calibrated based on the available water level, velocity and discharge measurements. The calibrated model will be used to calculate the flow atlas for the study area. This atlas will present the flood and ebb velocities that are important for vessels. F-WL-PP10-1 Versie 04 GELDIG VANAF: 12/11/2012

5 Contents 1. Introduction Units and reference plane Available measurement data Water levels Water levels available for the simulation period Quality of water level measurements ADCP measurements Discharges The numerical model Overall model Detailed model Software Model domain Land boundary Model grid Bathymetry Boundary conditions Time step Model settings Simulation period Weir at Mechelen Sensitivity analysis Methodology Boundary conditions NEVLA simulations TELEMAC simulations Sensitivity to the velocity diffusivity Sensitivity to the bed roughness Sensitivity to the time step Sensitivity to the grid resolution Model calibration Methodology Cost function Calibration results Quantitative Quality Assessment (calibration dataset) Introduction Water levels Final version WL2014R12_101_1 I

6 Analysis of the high and low waters and time series Harmonic analysis ADCP velocities Discharges Quantitative Quality Assessment (validation dataset) Selection of a validation dataset Validation results Export of the modeled velocity maps Conclusions List of references Tables Figures Appendix 1. Results of the model calibration... A1 Water levels... A1 ADCP velocities... A17 Discharges... A49 Appendix 2. Results of the model validation... A51 Appendix 3. Tidal coefficients... A65 Appendix 4. Statistical parameters... A66 Final version WL2014R12_101_1 II

7 List of tables Table 1. Water level stations used for the model calibration... 3 Table 2. ADCP measurements used for the analysis... 4 Table 3. Available discharge data... 8 Table 4. Channel Mesher parameters Table 5. Applied model settings for the detailed model Table 6. Model runs with different boundary conditions Table 7. Model runs for the sensitivity analysis to the velocity diffusivity Table 8. Model runs for the sensitivity analysis to the bed roughness Table 9. Model runs for the sensitivity analysis to the time step Table 10. Model runs with different grid resolution Table 11. ADCP measurements used for the model calibration Table 12. Discharge data used for the model calibration Table 13. Factors for the calculation of the cost function Table 14. Simulations used for the model calibration Table 15. Cost function of the model runs used for the model calibration Table 16. ADCP measurements used for the model validation Table 17. List of the model settings Table 18. Statistical parameters for the water level time series (simw6_4ref vs. measurements)... A1 Table 19. Statistical parameters for high waters (simw6_4ref vs. measurements)... A1 Table 20. Statistical parameters for low waters (simw6_4ref vs. measurements)... A2 Table 21. Harmonic analysis: Amplitude M2... A2 Table 22. Harmonic analysis: Phase M2... A2 Table 23. Harmonic analysis: Amplitude M4... A3 Table 24. Harmonic analysis: Phase M4... A3 Table 25. Harmonic analysis: Amplitude M6... A3 Table 26. Harmonic analysis: Phase M6... A4 Table 27. Harmonic analysis: Amplitude S2... A4 Table 28. Harmonic analysis: Phase S2... A4 Table 29. Harmonic analysis: Amplitude K1... A5 Table 30. Harmonic analysis: Phase K1... A5 Table 31. Harmonic analysis: Amplitude O1... A5 Table 32. Harmonic analysis: Phase O1... A6 Table 33. Vector differences of model results vs. measurements... A6 Table 34. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Wintam (part 1)... A17 Table 35. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Schelle (28/09/2006)... A19 Final version WL2014R12_101_1 III

8 Table 36. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Ballooi (dwars)... A20 Table 37. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Notelaer (langs)... A22 Table 38. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Kruibeke... A23 Table 39. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Boom... A24 Table 40. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Driegoten... A26 Table 41. Statistical parameters for discharges (model vs. measurement)... A49 Table 42. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Wintam (part 2)... A51 Table 43. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Schelle (23/03/2006)... A53 Table 44. Statistical parameters of velocities: calibrated model vs. ADCP measurements at Notelaer (dwars)... A54 Table 45. Typical values of the tidal coefficients for neap, average and spring tides... A65 Table 46. Model qualification based on (Sutherland et al., 2003)... A68 Final version WL2014R12_101_1 IV

9 List of figures Figure 1 - Location of the available ADCP measurements and water level stations in the model domain... 5 Figure 2 - ADCP measurements at Wintam... 5 Figure 3 - ADCP measurements at Schelle... 6 Figure 4 - ADCP measurements at Kruibeke... 6 Figure 5 - ADCP measurements at Boom... 7 Figure 6 - ADCP measurements at Notelaer and Ballooi (10/06 and 11/06/2009)... 7 Figure 7 - ADCP measurements at Driegoten... 8 Figure 8 - Location of the available discharge measurements... 8 Figure 9 - Model grid of the overall 3D NEVLA model (green) and detailed 2D TELEMAC model (blue)... 9 Figure 10 Velocity magnitude and water level at Schelle Figure 11 - Model domain Figure 12 - Grid of the detailed model (the entire model domain) Figure 13 - Detail of the model grid near Wintam Figure 14 - Grid resolution Figure 15 - Grid cell size Figure 16 - An example of the connection of the grids produced by Channel Mesher and by the T3 Mesh Generator Figure 17 - The model grid near the lock and weir in Mechelen Figure 18 - M2 amplitude calculated in the NEVLA runs Figure 19 - M2 amplitude calculated in the TELEMAC runs with different boundary conditions and in the NEVLA runs Figure 20 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw5a simw7_0) Figure 21 - M2 amplitude calculated in the TELEMAC runs with measured water level at the downstream boundary Figure 22 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw6_1 simw6a_1) Figure 23 - M2 amplitude calculated in the runs with different velocity diffusivity Figure 24 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw5_3 simw5) Figure 25 - Velocity for one of the transects at Wintam (max flood) calculated in the runs with different velocity diffusivity. Position of the transect is indicated in figure Figure 26 - M2 amplitude calculated in the runs with different bed roughness Figure 27 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw5_2 simw5) Figure 28 - M2 amplitude calculated in the runs with different time step Figure 29 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw5at simw5a) Figure 30 - Velocity for one of the transects at Wintam (max flood) calculated in the runs with different time steps Figure 31 - Grid resolution used in run simw7_ Figure 32 - Refined grid resolution used in run simw7_0ref Figure 33 - M2 amplitude calculated in runs with different grid resolution (simw7_0 and simw7_0ref) Final version WL2014R12_101_1 V

10 Figure 34 - Velocity for one of the transects at Schelle (max flood) calculated in the runs with different grid resolution. Refined grid resolution is shown on the right Figure 35 - Velocity for one of the transects at Boom (max flood) calculated in the runs with different grid resolution Figure 36 - M2 amplitude calculated in some model runs used for the model calibration Figure 37 - M2 phase calculated in some model runs used for the model calibration Figure 38 - Bias of high waters calculated in some model runs used for the model calibration Figure 39 - Bias of low waters calculated in some model runs used for the model calibration Figure 40 - RMSE of the water level time series calculated in some model runs used for the model calibration Figure 41 - Cost function Figure 42 - Measured water level at Temse Figure 43 - Wintam lock in the NEVLA model Figure 44 - Landboundary of the Scheldt estuary near the lock of Wintam Figure 45 - Model bathymetry (TELEMAC run simw6_4ref) Figure 46 Measured water level at the downstream model boundary Figure 47 - Discharges (from the NEVLA model) at the upstream model boundaries in Scheldt and Nete.. 50 Figure 48 - Measured discharges at the upstream model boundaries in Zenne and Dijle Figure 49 - Location of the downstream boundary of the TELEMAC model Figure 50 - Water level at Kallo and point BC7dnstr calculated in the NEVLA model Figure 51 Measured water level upstream the weir at Mechelen Figure 52 - Bias of high water magnitude (model measurement)... A7 Figure 53 - Bias of low water magnitude (model measurement)... A7 Figure 54 - RMSE of high water magnitude (model vs. measurement)... A8 Figure 55 - RMSE of low water magnitude (model vs. measurement)... A8 Figure 56 - Bias of the water level time series... A9 Figure 57 - RMSE of the water level time series... A9 Figure 58 - M2 amplitude... A10 Figure 59 - M2 phase... A10 Figure 60 - Amplitude ratio M4/M2... A11 Figure 61 - Phase shift 2M2-M4... A11 Figure 62 - Calculated and measured water levels at Antwerp... A12 Figure 63 - Calculated and measured water levels at Schelle... A12 Figure 64 - Calculated and measured water levels at Temse... A13 Figure 65 - Calculated and measured water levels at Tielrode... A13 Figure 66 - Calculated and measured water levels at Sint Amands... A14 Figure 67 - Calculated and measured water levels at Boom... A14 Figure 68 - Calculated and measured water levels at Walem... A15 Figure 69 - Calculated and measured water levels at Mechelen lock... A15 Figure 70 - Calculated and measured water levels at Duffel... A16 Final version WL2014R12_101_1 VI

11 Figure 71 - Bias of velocity magnitude and direction at Wintam part 1 (model vs. ADCP measurements). A28 Figure 72 - RMSE of velocity magnitude and direction at Wintam part 1 (model vs. ADCP measurements)... A29 Figure 73 - Time series of the measured and modeled velocity magnitude and direction at Wintam (part 1)... A30 Figure 74- Bias of velocity magnitude and direction at Schelle (model vs. ADCP measurements (28/09/2006))... A31 Figure 75- RMSE of velocity magnitude and direction at Schelle (model vs. ADCP measurements (28/09/2006))... A32 Figure 76 - Time series of the measured (28/09/2006) and modeled velocity magnitude and direction at Schelle... A33 Figure 77 - Bias of velocity magnitude and direction at Ballooi (dwars) (model vs. ADCP measurements)a34 Figure 78 - RMSE of velocity magnitude and direction at Ballooi (dwars) (model vs. ADCP measurements)... A35 Figure 79 - Time series of the measured and modeled velocity magnitude and direction at Ballooi (dwars)... A36 Figure 80 - Bias of velocity magnitude and direction at Notelaer (langs) (model vs. ADCP measurements)... A37 Figure 81 - RMSE of velocity magnitude and direction at Notelaer (langs) (model vs. ADCP measurements)... A38 Figure 82 - Time series of the measured and modeled velocity magnitude and direction at Notelaer (langs)... A39 Figure 83 - Bias of velocity magnitude and direction at Kruibeke (model vs. ADCP measurements)... A40 Figure 84 - RMSE of velocity magnitude and direction at Kruibeke (model vs. ADCP measurements)... A41 Figure 85 - Time series of the measured and modeled velocity magnitude and direction at Kruibeke... A42 Figure 86 - Bias of velocity magnitude and direction at Boom (model vs. ADCP measurements)... A43 Figure 87 - RMSE of velocity magnitude and direction at Boom (model vs. ADCP measurements)... A44 Figure 88 - Time series of the measured and modeled velocity magnitude and direction at Boom... A45 Figure 89 - Bias of velocity magnitude and direction at Driegoten (model vs. ADCP measurements)... A46 Figure 90 - RMSE of velocity magnitude and direction at Driegoten (model vs. ADCP measurements)... A47 Figure 91 - Time series of the measured and modeled velocity magnitude and direction at Driegoten... A48 Figure 92 - Calculated and measured discharges at Kruibeke... A49 Figure 93 - Calculated and measured discharges at Driegoten... A50 Figure 94 - Calculated and measured discharges at Boom... A50 Figure 95 - Bias of velocity magnitude and direction at Wintam (part 2) (model vs. ADCP measurements)... A56 Figure 96 - RMSE of velocity magnitude and direction at Wintam (part 2) (model vs. ADCP measurements)... A57 Figure 97 - Time series of the measured and modeled velocity magnitude and direction at Wintam (part 2)... A58 Figure 98 - Bias of velocity magnitude and direction at Schelle (model vs. ADCP measurements (23/03/2006))... A59 Figure 99 - RMSE of velocity magnitude and direction at Schelle (model vs. ADCP measurements (23/03/2006))... A60 Final version WL2014R12_101_1 VII

12 Figure Time series of the measured (23/03/2006) and modeled velocity magnitude and direction at Schelle... A61 Figure Bias of velocity magnitude and direction at Notelaer (dwars) (model vs. ADCP measurements)... A62 Figure RMSE of velocity magnitude and direction at Notelaer (dwars) (model vs. ADCP measurements)... A63 Figure Time series of the measured and modeled velocity magnitude and direction at Notelaer (dwars)... A64 Figure Definition of straight and oblique setup (after Adema, 2006).... A66 Final version WL2014R12_101_1 VIII

13 1. Introduction In this study a detailed model is constructed for the lock of Wintam. For this area it is important to have a well refined grid which can represent the flow to the locks correctly. Until now this area was simulated only in the NEVLA model, which includes the entire Scheldt estuary. The grid resolution of the overall NEVLA model is too coarse near Wintam for a meaningful application of the model results in the ship simulator of Flanders Hydraulics Research (FHR). Therefore, it was necessary to develop a detailed model for this area. The grid of this model should have a higher resolution than the NEVLA grid and should follow the quay walls better. The TELEMAC model software is used for the 2D calculations in this project. The model domain is discretized into an unstructured grid that is locally refined in the study area. Parallel computing is used to decrease the computational time. The calibrated model will be used to calculate the flow atlas near the lock of Wintam. This atlas will present the flood and ebb velocities that are important for vessels sailing in the study area. Final version WL2014R12_101_1 1

14 2. Units and reference plane Time is expressed in CET (Central European Time). Depth, height and water levels are expressed in meter TAW (Tweede Algemene Waterpassing). Bathymetry is positive below the reference plane in SIMONA (NEVLA model) and it is positive above the reference plane in TELEMAC. Water levels are positive above the reference plane. The horizontal coordinate system is RD Parijs. Final version WL2014R12_101_1 2

15 3. Available measurement data 3.1. Water levels Water levels available for the simulation period The water levels in 11 different stations are analyzed. Table 1 shows the list of the stations for which measured water levels are available for the simulation period. Figure 1 shows the location of the measurement stations. 10 minutes time series of the water level measurements (m NAP, CET) were retrieved from the Hydro Meteo Centrum Zeeland database (HMCZ, for Antwerp. These data were converted to the TAW reference plane. Measured water levels for other Belgian stations were available from the Hydrologic Information Centre (HIC) (YAKU database, The measurements from the HIC database are stored as one-minute time series in UTC (Coordinated Universal Time), and were converted to 10 minute time series in CET (Central European Time). Table 1. Water level stations used for the model calibration Station Data source 1 Antwerpen HMCZ 2 Schelle 3 Temse 4 Tielrode 5 Sint Amands 6 Boom HIC 7 Walem 8 Mechelen lock 9 Mechelen upstream weir 10 Rijmenam 11 Duffel Quality of water level measurements There is a problem with the measurement of low waters at Temse. The measurement instrument at this station is located in a muddy environment, measuring the level of the mud instead of the water level around low water (figure 40). Therefore, only high waters are analyzed at this station. Water levels at Mechelen upstream the weir and Rijmenam are not used for the model calibration because realistic control of the weir is not possible in the model (see chapter ) ADCP measurements Available ADCP (Acoustic Doppler Current Profiler) measurements located in the model domain are described in table 2 and figure 1. The tidal coefficient, k, in the table is calculated as the ratio of the tidal amplitude during the analyzed period to the amplitude of the average tide for the period from 1991 to More information about the tidal coefficients k is given in Appendix 3. Final version WL2014R12_101_1 3

16 Table 2. ADCP measurements used for the analysis Location Date and time (CET) Tide Project Used for Figure Wintam 13/02/ :25 20:39 k=1.21 (spring) 12_101 Stroomatlas sluis van Wintam one part of the measurement is used for the calibration, another part is used for the validation figure 2 Schelle 23/03/ :55 21:45 k=0.83 (neap) HCBS IMDC 2006 March validation figure 3 Schelle 28/09/ :34 19:32 k=0.97 (average) HCBS IMDC 2006 September calibration figure 3 Kruibeke 26/05/ :27 18:19 Boom 22/06/ :58 19:44 k=1.14 (spring) k=1.09 (spring) Moneos Raaien 2009 calibration figure 4 Moneos Raaien 2009 calibration figure 5 Ballooi ( dwars') Notelaer ( dwars ) Notelaer ( langs ) 10/06/ :42 19:22 11/06/ :15 19:53 10/06/ :00 19:42 k=1.02 (average) k=0.99 (average) k=1.02 (average) 713_21 Vervolgstudie inventarisatie en historische analyse van slikken en schorren langs Zeeschelde (Aqua Vision, 2010a) calibration validation calibration figure 6 Driegoten 23/06/ :24 20:14 k=1.11 (spring) Moneos Raaien 2009 (Aqua Vision, 2010b) calibration figure 7 Final version WL2014R12_101_1 4

17 Figure 1 - Location of the available ADCP measurements and water level stations in the model domain Figure 2 - ADCP measurements at Wintam Final version WL2014R12_101_1 5

18 Figure 3 - ADCP measurements at Schelle Figure 4 - ADCP measurements at Kruibeke Final version WL2014R12_101_1 6

19 Figure 5 - ADCP measurements at Boom Figure 6 - ADCP measurements at Notelaer and Ballooi (10/06 and 11/06/2009) Final version WL2014R12_101_1 7

20 Figure 7 - ADCP measurements at Driegoten 3.3. Discharges Table 3 gives an overview of the discharge measurements that can be used for the model calibration. The location of the cross sections is shown in figure 8. Table 3. Available discharge data Name of cross section Date Kruibeke 26/05/2009 Boom 22/06/2009 Driegoten 23/06/2009 Figure 8 - Location of the available discharge measurements Final version WL2014R12_101_1 8

21 4. The numerical model 4.1. Overall model The NEVLA model (Maximova et al., 2009; Vanlede et al (concept version)) was developed at Flanders Hydraulics Research for the Western Scheldt, the Sea Scheldt and tidally influenced tributaries. The model runs in the SIMONA software and it includes a broad coastal area and all Flemish tidal rivers, such as Scheldt, Durme, Rupel, Nete (Beneden, Grote and Kleine), Dijle and Zenne. These rivers are represented up until their tidal border (Vanlede et al., 2009). The calibration of the 2D NEVLA model was executed in Maximova et al. (2009), the calibration of the 3D NEVLA model is still in progress and is described in Vanlede et al. (2013, concept version). The grid resolution of the NEVLA model is too coarse (55 x 37 m to 35 x 30 m) (figure 41) near the lock of Wintam for a meaningful implementation of the model results into the ship simulator at FHR. Within the framework of this project a TELEMAC 2D model (with a significantly finer grid resolution than NEVLA) is developed. This model is calibrated based on the comparison of the calculated and measured water levels, velocities and discharges. The grids of the overall and detailed models are presented in figure 9. These models are linked by nesting at the upstream boundaries. The upstream boundary conditions for the detailed model are defined based on the output of the NEVLA 3D run simg104. The measured water level is specified at the downstream boundary of the detailed model. This combination of the boundary conditions produced the best results (see chapter 5.2). The NEVLA run is performed for a period from 21/05/2009 to 26/06/2009. More information about the overall model is given in Vanlede et al. (2013, concept version). Figure 9 - Model grid of the overall 3D NEVLA model (green) and detailed 2D TELEMAC model (blue) 4.2. Detailed model Software The model for this project is developed in the TELEMAC software, which is based on the finite element method. The model domain is discretized into an unstructured grid of triangular elements and it can be locally refined in the study area. Therefore, the complex geometry of the study area can be taken into account. Parallel computing is used to decrease the computational time. Final version WL2014R12_101_1 9

22 The Blue Kenue software (Canadian Hydraulics Centre, 2011) is used for the grid and bathymetry generation. The Fudaa software is used to generate the boundary condition files. The model results are visualized in MATLAB Model domain The tidal excursion length scale was analyzed in order to define the minimum area that should be included within the model domain. The velocity magnitude at Schelle measured by ADCP on 28/09/2006 was used for the analysis. The following formulas for the tidal excursion length scale were used: T2 ebb L ebb = Vdt, T1 ebb T1 ebb T2 flood L flood = Vdt + Vdt, T1 flood T2 ebb (Length scale for flood is calculated as a sum of 2 integrals because flood is divided in 2 parts). Figure 10 Velocity magnitude and water level at Schelle which gives the following characteitis length scales for flood and ebb: L flood = 18.7 km and L ebb = 20.4 km Figure 11 shows the location of the model boundaries according to this calculation (green lines). Blue lines in the figure show the boundaries of the chosen model domain. Dijle and Zenne are included completely in the model. The measured discharges for these rivers will be specified as upstream boundary conditions. Final version WL2014R12_101_1 10

23 Land boundary Figure 11 - Model domain The land boundary of the Scheldt estuary (file GrensSchorDrooggebiedSchelde.ldb) is accurate enough near the lock of Wintam (figure 42). It can be used to define the outline of the model during grid generation Model grid The downstream boundary of the detailed model is located between Kallo lock and Oosterweel. The model has five upstream boundaries, which are located: - between the bend of Kramp and Dendermonde in the Upper Sea Scheldt; - between Duffel and Lier in the Nete; - Durme, Zenne and Dijle rivers are included in the model until their tidal boundary (figure 12). The grid resolution is presented in figure 14. It is 10 m near the lock of Wintam. The total number of nodes in the grid is The total number of grid cells is The distribution of the grid cell size vs. number of cells is presented in figure 15. Final version WL2014R12_101_1 11

24 Figure 12 - Grid of the detailed model (the entire model domain) Figure 13 - Detail of the model grid near Wintam Final version WL2014R12_101_1 12

25 Figure 14 - Grid resolution Figure 15 - Grid cell size Final version WL2014R12_101_1 13

26 Hard points were used in Blue Kenue during the grid generation to define the location of the measurement stations. Hard points establish the positions of fixed nodes within the mesh (Canadian Hydraulics Centre, 2011). The T3 Channel Mesher was used to produce triangular meshes of unbranched channels (Canadian Hydraulics Centre, 2011). In this study it was used to generate the meshes for Zenne, Dijle and Nete rivers. These meshes were afterwards incorporated into the mesh produced by the T3 Mesh Generator (figure 16). The use of the Channel Mesher helps to obtain a sufficient grid resolution (for an accurate representation of the flow) with a smaller number of grid cells. The parameters used in the Channel Mesher are described in table 4. The Cross Channel Node Count represents the number of nodes that span the channel, including the edge nodes. The Along Channel Interval represents the space between nodes along the edges of the channel (Canadian Hydraulics Centre, 2011). Table 4. Channel Mesher parameters River Cross Channel Node Count Parameter Along Channel Interval (m) Nete 6 30 Dijle 6 20 Zenne 6 20 Only unbranched parts of Zenne, Dijle and Nete were produced by the Channel Mesher. Other parts of the model grid were generated by the T3 Mesh Generator (figure 17). Figure 16 - An example of the connection of the grids produced by Channel Mesher and by the T3 Mesh Generator Final version WL2014R12_101_1 14

27 Bathymetry Figure 17 - The model grid near the lock and weir in Mechelen The bathymetry of the Sea Scheldt, Rupel and its tributaries is defined based on the samples provided by Flemish Hydrography and W&Z (Waterwegen en Zeekanaal). TAW is used as a vertical reference (figure 43). Samples from 2012 are available for the Sea Scheldt and Durme; samples from 2013 are used to define the bathymetry of the Rupel river and Wintam lock. Samples from 2010 are used for Dijle and Beneden Nete. The LIDAR data from 2010 are used to define the bathymetry of the intertidal areas. The bathymetry is missing for the upstream part of Durme, Zenne, Dijle and Nete. It is defined based on the bathymetry of the 3D NEVLA model and based on the old samples (used to define the bathymetry of the 2D NEVLA model). The bathymetry is deepened in several cells at the upstream model boundaries to ensure that the discharge boundaries never become dry and to prevent model runtime errors Boundary conditions The choice of boundary conditions was optimized during the calibration procedure (chapter 6.3). The 10 minute time series of the measured water level at Kallo is defined at the downstream boundary (figure 44). This measurement station is located close to the downstream boundary of the TELEMAC model (see chapter 5.2.2). The gaps in the measured data are filled with the calculated water level (from the NEVLA model). The 10 minute time series of the discharges calculated in NEVLA are defined at the upstream boundaries of the TELEMAC model in the Upper Sea Scheldt and Nete (figure 45). The discharges from NEVLA are multiplied by -1 because the ebb flow is negative in SIMONA at these locations while it is positive in TELEMAC (the inflow at the upstream boundary is positive). Measured discharges are defined at Zenne and Dijle (figure 46) because these rivers are completely included in the model domain (until their tidal border). The discharge at Durme is zero and therefore no boundary condition is defined there Time step The time step used for the model simulations is 3 s. It was chosen based on the sensitivity analysis taking into account the Courant number (chapter 5.5). Final version WL2014R12_101_1 15

28 Model settings The most important model settings are described in table 5. The complete list of the model settings is given in table 17. Table 5. Applied model settings for the detailed model Parameter Time step Initial condition Number of layers in the vertical Version TELEMAC Salt transport Wind Computational time Roughness formula Bed roughness value Velocity diffusivity Treatment of the linear system Value 3 s Constant elevation = 7 m TAW 1 (2D model) TELEMAC 6.0 (Linux) Off Off about 14 hours (12 CPU s) (simulation period of 1 month) Manning m -1/3 s uniform 10-4 m²/s 2: wave equation Free surface gradient compatibility 0.9 Continuity correction Turbulence model Type of advection Solver true 1: Constant viscosity Method of characteristics 7: GMRES The parameter Treatment of the linear system specifies the treatment of the Saint-Venant equations (1 original equations; 2 wave equation). In most cases, option 2 is recommended and offers the optimum in terms of CPU time (EDF-R&D, 2010). The free surface gradient compatibility was set to 0.9 (it is a recommended value) to avoid free surface wiggles (for example in areas with strong bathymetric gradients). The main consequence is a slightly altered compatibility between depth and velocity in the continuity equation (EDF-R&D, 2010). The continuity correction was used. Residual mass errors (of the order of a few percent) may appear when using boundary conditions with imposed depth. Indeed the continuity equation is not solved for these points and is replaced by the equation depth = imposed value. Therefore, the resultant discharge is not properly computed and leads to error. The use of the continuity correction helps in correcting the velocity at these points so that the overall continuity is verified (EDF-R&D, 2010). The default turbulence model and type of advection were used. Final version WL2014R12_101_1 16

29 GMRES solver (Generalised Minimum RESidual method) is used for the calculations. The analysis in Smolders et al., (2013) showed that this solver produces good results. The model runs are stable Simulation period The TELEMAC model is nested into a larger NEVLA model (which runs in SIMONA). The NEVLA model is calibrated for Therefore, one month in 2009 was chosen for the calibration and validation of the TELEMAC model. The simulation period is chosen based on the analysis of the available ADCP measurements. Several ADCP measurements are available in May and June 2009 (Kruibeke, Boom, Ballooi, Notelaer and Driegoten) (table 2). Therefore, the simulation period used in this study is chosen from 24/05/ :00 to 25/06/ :00. The spin-up period is 1 day (from 24/05/ :00 to 25/05/ :00). For the model validation an independent dataset will be used (see chapter 8). Comparable tides for the measurements used for the validation will be found inside the same period as the one used for the calibration Weir at Mechelen In reality there is a weir near Mechelen (figure 1) which is controlled automatically or by hand. In the case of low discharges the weir is controlled automatically. If the discharges are high (the difference between water level upstream and downstream is less than 80 cm), the weir is manually controlled (Adema, 2006). Analysis of the measured water levels at Mechelen upstream the weir (figure 49) shows that the weir was closed in the beginning of the simulation period and open in the second half of the period. The weir is implemented in the model bathymetry: the level of the crest is 4.33 m TAW (this value is taken from the NEVLA model (NAP converted to TAW; opposite sign)). Since realistic control of the weir is not possible in the model, it was decided not to include stations Mechelen upstream the weir and Rijmenam in the analysis. The weir has only a local effect and it does not affect water levels and flow velocities in the study area. Final version WL2014R12_101_1 17

30 5. Sensitivity analysis 5.1. Methodology The sensitivity analysis of the TELEMAC model is performed in order to understand the impact of different values of model parameters on the model results. The results of the reference run are compared with a simulation where only one of the parameters is changed. The results of the sensitivity analysis give necessary information for the model calibration Boundary conditions NEVLA simulations The boundary conditions for the TELEMAC simulations were defined based on the output of the NEVLA model. Two different NEVLA runs (described in Vanlede et al., 2013 (concept version)) were used for the analysis: - simg104 - simg105 (= simg104 with corrected harmonic components; adapted grid (always dry cells are deleted); more recent bathymetry of the Rupel tributaries; weir at Mechelen). The M2 amplitude calculated in these NEVLA runs is compared with the measurements in figure 18. The amplitude is underestimated in simg104 at all stations. The difference is about 10 cm at Kallo lock (where the downstream boundary of the detailed TELEMAC model is located). The results of simg105 are closer to the measurements. Therefore, it is expected that the nesting of the TELEMAC model into simg105 will give better results than the nesting in simg104. Figure 18 - M2 amplitude calculated in the NEVLA runs Final version WL2014R12_101_1 18

31 TELEMAC simulations The model runs used for the sensitivity analysis to the boundary conditions are described in table 6. The same bed roughness and velocity diffusivity are used in all simulations: m -1/3 s and 10-4 m²/s (default value). Table 6. Model runs with different boundary conditions Boundary conditions Model run Downstream Upstream (Scheldt and Nete) Upstream (Zenne and Dijle) simw5a Water level from simg104 Discharge from simg104 simw7_0 Water level from simg105 Discharge from simg105 simw6_1 Measured water level at Kallo Discharge from simg104 Measured discharges simw6a_1 Measured water level at Kallo Discharge from simg105 Downstream and upstream boundary conditions from NEVLA Figure 19 shows the M2 amplitude calculated in the TELEMAC runs with different boundary conditions derived from the different NEVLA models. The amplitude is too low in both TELEMAC simulations. In run simw5a (with the boundary conditions from simg104) the M2 amplitude is 17 cm lower than the measurement at Antwerp and it is 43 cm lower at Sint Amands. The amplitude is represented better in NEVLA run simg105 than in simg104 at most stations. Therefore, it is expected that the nesting of the TELEMAC model into simg105 will give better results. However, the energy dissipation in simw7_0 (with the boundary conditions from simg105) is also very high. The amplitude at Kallo and Antwerp is accurate in simg105 (NEVLA) (figure 18). However, in simw7_0 the difference at Antwerp is already 10 cm. It is related to the energy loss between the downstream boundary (where the water level from simg105 is defined) and Antwerp. The difference in the M2 amplitude increases upstream up to 50 cm at Sint Amands. Figure 19 - M2 amplitude calculated in the TELEMAC runs with different boundary conditions and in the NEVLA runs Final version WL2014R12_101_1 19

32 The map of differences in velocity magnitude between the runs is presented in figure 20. The velocity magnitude is very similar in both simulations. Figure 20 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw5a simw7_0) Measured water level downstream The downstream boundary of the TELEMAC model is located near the Kallo lock. The analysis of the NEVLA results shows that the calculated water levels at Kallo are very similar to the water levels in the point BC7dnstr (located at the downstream boundary of the detailed TELEMAC model) (figure 49, figure 50). Therefore, the measured water levels at Kallo can be used as a downstream boundary condition in the TELEMAC model. The calculated discharges from the overall model are specified upstream in the Scheldt and Nete (simw6_1 and simw6a_1, table 6). The tidal amplitude at Antwerp is represented slightly better in runs simw6_1 and simw6a_1 (the difference is 8 cm) than in the runs with the water level from NEVLA at the downstream boundary (simw5a and simw7_0). However, it is still underestimated at all stations (figure 21). The amplitude is represented slightly better in run simw6_1 than in run simw6a_1. The map of differences in velocity magnitude between the runs is presented in figure 22. The velocity magnitude is very similar in both simulations. Final version WL2014R12_101_1 20

33 Figure 21 - M2 amplitude calculated in the TELEMAC runs with measured water level at the downstream boundary Figure 22 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw6_1 simw6a_1) 5.3. Sensitivity to the velocity diffusivity The model runs used for the sensitivity analysis to the velocity diffusivity are described in table 7. The overall viscosity coefficient (molecular + turbulent viscosity) is provided in the TELEMAC model as the velocity diffusivity, which has a default value of 10-4 m²/s (the minimum value is 10-6 m²/s, corresponding to the molecular viscosity of water) (EDF-R&D, 2010). A change of the viscosity affects the velocity profile Final version WL2014R12_101_1 21

34 along the cross section. When viscosity increases, the velocity profile becomes less convex and the horizontal velocity gradients decrease. It is expected that an increase of the viscosity results in a decrease of the tidal amplitude and velocity in the river channel (higher energy dissipation). Table 7. Model runs for the sensitivity analysis to the velocity diffusivity Model run Bed roughness (m -1/3 s) Velocity diffusivity (m²/s) Type of advection Turbulence model simw (default) simw5_ simw5_5 1 Method of characteristics Constant viscosity simw5_3 2 Figure 23 shows the M2 amplitude calculated in the runs with different viscosity. An increase of the velocity diffusivity results in a decrease of the tidal amplitude due to higher energy dissipation in the model domain. The results of the runs with a default velocity diffusivity of 10-4 m²/s and a minimum value of 10-6 m²/s are very similar to each other (simw5 and simw5_4). This means that during the calibration process it is not possible to increase the tidal amplitude in the model by lowering the velocity diffusivity below its default value. Changing of the velocity diffusivity does affect the velocity profile over a cross section. The flow velocities in the river channel decrease, velocities in the intertidal areas increase (figure 24, figure 25). A velocity diffusivity of 10-4 m²/s will be used for the model calibration. Figure 23 - M2 amplitude calculated in the runs with different velocity diffusivity Final version WL2014R12_101_1 22

35 Figure 24 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw5_3 simw5) Figure 25 - Velocity for one of the transects at Wintam (max flood) calculated in the runs with different velocity diffusivity. Position of the transect is indicated in figure 2 Final version WL2014R12_101_1 23

36 5.4. Sensitivity to the bed roughness The model runs used for the sensitivity analysis to the bed roughness are described in table 8. Considering energy dissipation, changes in bed roughness have similar effect on the model results as changes in viscosity (velocity diffusivity). A decrease of the bed roughness results in a lower energy dissipation, and therefore an increase of the tidal amplitude (figure 26) and velocity in the main channel (figure 27). Table 8. Model runs for the sensitivity analysis to the bed roughness Model run Bed roughness (m -1/3 s) Velocity diffusivity (m²/s) simw simw5_ (default) simw5_ Figure 26 - M2 amplitude calculated in the runs with different bed roughness Final version WL2014R12_101_1 24

37 Figure 27 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw5_2 simw5) 5.5. Sensitivity to the time step Model runs used for the sensitivity analysis to a time step are described in table 9. The relationship between the time step and the grid size has an effect on the model accuracy and stability. The time steps for the analysis were chosen based on the Courant number. The Courant number expresses the ratio of the distance traveled by an advected particle in one time step to the size of the grid cell. This number should be smaller than 1 to ensure the numerical stability. The Courant number is calculated as C = V t x 1 where V is velocity, t is the time step, x is the grid size. It is calculated for the minimum grid size (10 m) and for the velocity of 2.5 m/s. Table 9. Model runs for the sensitivity analysis to the time step Model run Time step (s) Courant number (max value) Computational time (hours : min) simw5a :05 simw5at :35 Figure 28 shows the M2 amplitude calculated in the runs with different time steps. The map of differences in velocity magnitude between the runs is presented in figure 29. The model results and measurements are presented for one of the transects in figure 30. The velocity magnitude is very similar in both simulations. It was chosen to use a time step of 3 s for this study. Final version WL2014R12_101_1 25

38 Figure 28 - M2 amplitude calculated in the runs with different time step Figure 29 - Difference in velocity magnitude (m/s) for max flood at Wintam (simw5at simw5a) Final version WL2014R12_101_1 26

39 Figure 30 - Velocity for one of the transects at Wintam (max flood) calculated in the runs with different time steps 5.6. Sensitivity to the grid resolution Table 10 gives an overview of the model runs used for the sensitivity analysis to the grid resolution. In run simw7_0ref the grid is refined downstream Schelle, between Wintam and Baasrode and between Wintam and Walem (figure 30). Table 10. Model runs with different grid resolution Model run Grid resolution (m) Boundary conditions Time step (s) Courant number Computational time simw7_0 figure 28 From simg105 about 5 hours simw7_0ref figure 29 From simg105 about 14 hours Final version WL2014R12_101_1 27

40 Figure 31 - Grid resolution used in run simw7_0 Figure 32 - Refined grid resolution used in run simw7_0ref Final version WL2014R12_101_1 28

41 An increase of the grid resolution results in an increase of the M2 amplitude in the model (figure 33). It increases by 3 cm at Antwerp to 18 cm at Sint Amands in the Sea Scheldt and by 10 to 18 cm in the Rupel and its tributaries. In the run with an increased grid resolution the modeled velocities slightly increase in the river channel and decrease in the intertidal areas (figure 34, figure 35), which would correspond to a decrease in velocity diffusivity. The strong effect of the adaptation of the grid resolution on the calculated water levels is probably related to the artificial diffusion of the TELEMAC model. The artificial diffusion of an upwind advection scheme is about U*dx/2 where U is the velocity and dx is the mesh size (Jean-Michel Hervouet, TELEMAC forum). Therefore, a decrease of the mesh size results in a significant decrease of the artificial diffusion and the model becomes less dissipative. The calibration process was started with a lower grid resolution (figure 31) to decrease the computational and post-processing time. The final grid resolution was chosen during the calibration (chapter 0). Figure 33 - M2 amplitude calculated in runs with different grid resolution (simw7_0 and simw7_0ref) Final version WL2014R12_101_1 29

42 Figure 34 - Velocity for one of the transects at Schelle (max flood) calculated in the runs with different grid resolution. Refined grid resolution is shown on the right. Figure 35 - Velocity for one of the transects at Boom (max flood) calculated in the runs with different grid resolution Final version WL2014R12_101_1 30

43 6. Model calibration 6.1. Methodology The main objective of the model calibration in this project is to improve the model accuracy for the velocities near the lock of Wintam. For the model calibration different simulations are performed with different parameters of the 2D hydrodynamic model. Bed roughness is used as a calibration parameter. A constant velocity diffusivity of 10-4 m²/s (default) is used in all the calibration runs (because lowering of this parameter did not affect the model results (chapter 5.3)). Results of the model simulations are compared with the measured water levels, velocities and discharges. The measured water levels are compared with the model results for the period from 25/05/2009 to 25/06/2009. Harmonic analysis of the tide is performed and statistical parameters (bias, RMSE, RMSE 0 ) are calculated for high and low waters and for the time series of water levels (more information is given in Appendix 4). The ADCP measurements and discharges are compared with the model results for the tides similar to the tides observed during the measurements (table 11, table 12). More information about the tidal coefficients k is given in Appendix 3. For Wintam and Schelle comparable tides are found inside the simulation period based on the smallest RMSE 0 between the tides observed during the measurement and during the modeling period. This is done in order to obtain a better representation of velocities in the model. Velocities at Kruibeke, Boom, Ballooi, Notelaer and Driegoten were measured during the period represented in the model. Therefore, they are compared with the model results for the same tides as the tides observed during the ADCP measurements. The model accuracy is evaluated based on the history and vector plots of velocities, statistical parameters and plots of the discharge histories. Table 11. ADCP measurements used for the model calibration Measurement Tref of measured tide k Antwerp Tref of comparable tide k Antwerp RMSE 0 Wintam (part 1)* 13/02/ : /05/ : Schelle 28/09/ : /06/ : Kruibeke 26/05/ : /05/ : Boom 22/06/ : /06/ : Ballooi ( dwars transverse profile) Notelaer ( langs longitudinal profile) 10/06/ : /06/ : /06/ : /06/ : Driegoten 23/06/ : /06/ : *The ADCP measurements at Wintam are divided into two parts. One part is used for the model calibration and the other one is used for the model validation. Measurement Table 12. Discharge data used for the model calibration Tref of measured tide k Antwerp Tref of comparable tide k Antwerp RMSE 0 Kruibeke 26/05/ : /05/ : Boom 22/06/ : /06/ : Driegoten 23/06/ : /06/ : Final version WL2014R12_101_1 31