Issuu

13_041_3 FHR reports

Baseline Zuidwestelijke Delta Final calibration of model waqua-schelde_nevla-j07_5

www.flandershydraulicsresearch.be

Baseline Zuidwestelijke Delta Final calibration of model waqua-schelde_nevla-j07_5

Chu, K.; Vanlede, J.; Mostaert, F.

Cover figure © The Government of Flanders, Department of Mobility and Public Works, Flanders Hydraulics Research Legal notice Flanders Hydraulics Research is of the opinion that the information and positions in this report are substantiated by the available data and knowledge at the time of writing. The positions taken in this report are those of Flanders Hydraulics Research and do not reflect necessarily the opinion of the Government of Flanders or any of its institutions. Flanders Hydraulics Research nor any person or company acting on behalf of Flanders Hydraulics Research is responsible for any loss or damage arising from the use of the information in this report. Copyright and citation © The Government of Flanders, Department of Mobility and Public Works, Flanders Hydraulics Research 2020 D/2020/3241/34 This publication should be cited as follows: Chu, K.; Vanlede, J.; Mostaert, F. (2020). Baseline Zuidwestelijke Delta: Final calibration of model waqua-schelde_nevlaj07_5. Version 1.0. FHR Reports, 13_041_3. Flanders Hydraulics Research: Antwerp. IMDC: I/RA/11502/20.014/KCH/ Reproduction of and reference to this publication is authorised provided the source is acknowledged correctly. Document identification Customer: Keywords (3-5): Text (p.): Confidentiality:

Rijkswaterstaat Zeeland Ref.: WL2020R13_041_3 Hydrodynamics; Automatic Calibration; Scheldt 21 Appendices (p.): 4 ‫ ܈‬No ‫ ܈‬Available online

Author(s):

Chu, K.; Vanlede, J.

Control Name Reviser(s):

Signature

Vanlede, J. Getekend door: Joris Vanlede (Signature) Getekend op: 2020-02-24 12:21:43 +01:00 Reden: Ik keur dit document goed

Project leader:

Vanlede, J.

Approval Getekend door: Frank Mostaert (Signature) Getekend op: 2020-02-25 07:46:22 +01:00 Reden: Ik keur dit document goed

Head of Division:

F-WL-PP10-2 Version 7 Valid as from 3/01/2017

Mostaert, F.

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

Abstract The calibration of the WAQUA-Schelde_NEVLA model is done in three steps. Deltares first set up the model based on the NEVLA grid provided by FHR, and did the first step in the automatic calibration by varying the bottom roughness in the Western Scheldt. The subsequent automatic calibration for the Flemish territory was carried out at FHR as the 2nd step, while keeping the roughness field from the first step constant (Chu et al., 2019 & 2020). This report describes the final automatic calibration (3rd step) for the entire model domain starting from the calibrated roughness field from the 2nd step (Chu et al, 2020). It was found that this automatic 3rd step does not further modify the bottom friction or further improve the water level predictions compared to the end result of the 2nd step. This means that in the future, automatic calibration could be safely performed in 2 steps.

Final version

WL2020R13_041_3

III

F-WL-PP10-2 Version 7 Valid as from 3/01/2017

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

Contents Abstract ............................................................................................................................................................ III Contents ............................................................................................................................................................ V List of tables...................................................................................................................................................... VI List of figures ................................................................................................................................................... VII 1

Introduction ............................................................................................................................................... 1

Abbreviations and Conventions ................................................................................................................ 2

2.1

Abbreviations..................................................................................................................................... 2

2.2

Conventions ....................................................................................................................................... 2

Final Calibration with OpenDA .................................................................................................................. 3 3.1

Introduction ....................................................................................................................................... 3

3.2

Cost function ..................................................................................................................................... 3

3.3

Stop Criterion..................................................................................................................................... 5

3.4

Results ............................................................................................................................................... 6

Validation................................................................................................................................................. 11 4.1

Boundary Conditions ....................................................................................................................... 11

4.2

Water Level Timeseries ................................................................................................................... 11

4.3

Harmonic Analysis of Water Levels ................................................................................................. 17

Conclusions .............................................................................................................................................. 20

References ............................................................................................................................................... 21

Appendix A

Introduction of DUD ................................................................................................................ A1

Appendix B

Definition of Statistics.............................................................................................................. A3

Appendix C

Definition of Vector Difference ............................................................................................... A4

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

List of tables Table 1 – Abbreviations used ............................................................................................................................ 2 Table 2 – Standard deviation assigend to the water level stations in the cost function. ................................. 5 Table 3 – Changes of cost function and parameter (manning coefficient) during the final calibration with OpenDA. ............................................................................................................................................................ 7 Table 4 – Comparison of manning coefficient before and after the final calibration with OpenDA. ............... 8 Table 5 – Description of model runs involved in the model validation........................................................... 11 Table 6 – Definition of colour code in terms of bias, RMSE and RMSE0. ........................................................ 11 Table 7 – Comparison of Bias, RMSE and RMSE0 of the complete time series. ............................................. 12

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

List of figures Figure 1 – Diagram working flow for the automatic model calibration. ........................................................... 3 Figure 2 – Map of 34 water level stations involved in the final OpenDA calibration. ....................................... 4 Figure 3 – Evolution of cost function during the final calibration with OpenDA. ............................................. 7 Figure 4 – Comparison of Manning coefficient before and after the final OpenDA calibration. ...................... 9 Figure 5 – Final calibrated bottom roughness map of Manning Coefficient................................................... 10 Figure 6 – Final calibrated bottom roughness map of Manning Coefficient (Zoom in to Upper stream)....... 10 Figure 7 – Bias of complete time series of water levels along the Scheldt. .................................................... 13 Figure 8 – RMSE of complete time series of water levels along the Scheldt. ................................................. 13 Figure 9 – RMSE0 of complete time series of water levels along the Scheldt. ............................................... 14 Figure 10 – Bias of high water levels along the Scheldt. ................................................................................. 14 Figure 11 – RMSE of high water levels along the Scheldt. .............................................................................. 15 Figure 12 – RMSE0 of high water levels along the Scheldt. ............................................................................ 15 Figure 13 – Bias of low water levels along the Scheldt. .................................................................................. 16 Figure 14 – RMSE of low water levels along the Scheldt. ............................................................................... 16 Figure 15 – RMSE0 of low water levels along the Scheldt. ............................................................................. 17 Figure 16 – M2 amplitude along the River Scheldt. ........................................................................................ 17 Figure 17 – M2 phase along the River Scheldt. ............................................................................................... 18 Figure 18 – S2 amplitude along the River Scheldt. .......................................................................................... 18 Figure 19 – S2 phase along the River Scheldt. ................................................................................................. 19 Figure 20 – Vector differences along the River Scheldt. ................................................................................. 19

Final version

WL2020R13_041_3

VII

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

1 Introduction This study aims for a final calibration of the WAQUA-Schelde_NEVLA model for the entire year of 2007. In the summer of 2016, Deltares has finished the calibration and validation of the new WAQUASchelde_Nevla model (waqua-schelde_nevla-j07_5-v4) for the Dutch part of the River Scheldt (Groenenboom et al., 2016). The task of model calibration for the Flemish part is carried out at Waterbouwkundig Laboratorium (Flanders Hydraulics Research or FHR) and reported by Chu et al (2019, 2020). Based on the two calibrations mentioned above, a final calibration of the model for the entire Scheldt is carried out at Deltares. The results are analysed and summarized in this report.

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

2 Abbreviations and Conventions 2.1 Abbreviations Table 1 – Abbreviations used

BeZS

Beneden-Zeeschelde (Lower Sea Scheldt)

BoZS

Boven-Zeeschelde (Upper Sea Scheldt)

FHR

Flanders Hydraulic Research

MET

Middle European Time

NAP

Normaal Amsterdams Peil (Dutch vertical reference level)

NEVLA

Dutch-Flemish hydrodynamic model

OpenDA

Open Data Assimilation

Rijksdriehoekscoördinaten

RMSE

Root Mean Square Error

SIMONA

SImulatie MOdellen NAtte waterstaat

TAW

Tweede Algemene Waterpassing (Belgian vertical reference level)

WES

Westerschelde (Western Scheldt)

2.2 Conventions The following conventions are followed by this report: • • • •

Times are represented in MET. The coordinate reference system, used by the model and for presentation of the model output is RD Parijs, expressed in meters. The vertical reference level used by this project is NAP. NAP is 2.35 m above TAW level. SI units are used.

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

3 Final Calibration with OpenDA 3.1 Introduction The final calibration starts from the calibration run work20 as reported by Chu et al (2020). The friction map from the work20 run is taken as the initial friction map for this study. The final calibration focus on the entire model domain with in total 17 roughness polygons (7 for the Dutch part and 10 for the Flemish part, see Table 4). Figure 1 illustrates the working flow of the 3 steps in automatic model calibration. Figure 1 â€“ Diagram working flow for the automatic model calibration.

3.2 Cost function The cost function used by this study is org.openda.algorithms.SimulationKwadraticCostFunction which is a quadratic cost function over the complete timeseries. It is essentially a total sum of squares, made dimensionless with the measurement uncertainty. đ?‘…đ?‘…đ?‘…đ?‘…đ?‘…đ?‘…đ?‘…đ?‘… đ?‘†đ?‘†đ?‘†đ?‘†đ?‘†đ?‘†đ?‘†đ?‘† đ?‘ đ?‘ đ?‘ đ?‘ đ?‘ đ?‘ đ?‘ đ?‘

1 đ??˝đ??˝ = ďż˝ 2

đ?‘&#x;đ?‘&#x;=1

ďż˝ đ?‘ đ?‘ =1

ďż˝

đ?‘›đ?‘›=1

đ?‘ đ?‘ đ?‘ đ?‘ đ?‘ đ?‘ (đ?‘Ąđ?‘Ą) đ?‘œđ?‘œđ?‘œđ?‘œđ?‘œđ?‘œ (đ?‘Ąđ?‘Ą))2 (đ??ťđ??ťđ?‘&#x;đ?‘&#x;,đ?‘ đ?‘ ,đ?‘›đ?‘› âˆ’ đ??ťđ??ťđ?‘&#x;đ?‘&#x;,đ?‘ đ?‘ ,đ?‘›đ?‘› 2 (đ?œŽđ?œŽđ??ťđ??ťđ?‘&#x;đ?‘&#x;,đ?‘ đ?‘ đ?‘œđ?‘œđ?‘œđ?‘œđ?‘œđ?‘œ )

where: H(t) - water level at time t; sim - results obtained from model simulations over the simulation period; obs - observation values; n,Nmax - number of time steps in the time series (52560); s, Smax - number of stations in region r; r, Rmax - number of polygons for which observations are included (17, see Table 4); đ??ˆđ??ˆđ?‘Żđ?‘Żđ?’“đ?’“,đ?’”đ?’” đ?’?đ?’?đ?’?đ?’?đ?’?đ?’? - uncertainties assigned to the observations (2.0 m for Rupel and 0.5 m elsewhere, see Table 2). Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

In this study, time series of water level measurement at 34 stations (as shown in Figure 2) are employed to the OpenDA calibration. The calibration period is the year 2007 (2007/01/01 00:00 â€“ 2007/12/31 23:50) with time step of 10 minutes, giving the total number of time steps equal to 52560 (Nmax). In Chu et al (2019) the entire Rupel basin is represented as one roughness zone with 11 water level measurement stations inside of it. Aggregation into larger roughness zone is a trade-off between a reduction of dimensionality of the problem, and thus fewer iterations in the automatic calibration, and the maximum accuracy that is achievable during calibration. In the cost function, the term đ?œŽđ?œŽđ??ťđ??ť đ?‘œđ?‘œđ?‘œđ?‘œđ?‘œđ?‘œ can be used as the weight đ?‘&#x;đ?‘&#x;,đ?‘ đ?‘

assigned to the water level stations. In Chu et al. (2019) a constant value of 0.5 m was assigned to all stations throughout the estuary. Because the cost function of the automatic calibration takes into account all stations, having one roughness zone with more stations in it, might skew the automatic calibration process.

Therefore Chu et al (2020) explore the effect of changing this value to 2 m for the 11 stations in the Rupel basin. As a result, the contribution of stations in the Rupel basin to the cost function becomes 16 times smaller. Because the Rupel basin is not the zone of interest of the Nevla model (due to limited resolution it is not schematized as accurately as the Scheldt estuary), this is wanted behavior, and so the same approach is adopted in the final 3rd calibration in this study. Figure 2 â€“ Map of 34 water level stations involved in the final OpenDA calibration.

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Table 2 – Standard deviation assigend to the water level stations in the cost function. A higher value corresponds to a lower weight in the cost function No.

Station Name

Vlakte van de Raan

standardDeviation [m] 0.5

No.

Station Name

0.5

Temse

0.5

Tielrode

0.5

Walem

0.5

StAmands

0.5

Dendermonde

0.5

Schoonaarde

0.5

Wetteren

0.5

Melle

0.5

Mechelen Benedensluis

0.5

Hombeek

0.5

Zemst

0.5

Duffel

0.5

Rijmenam

0.5

Lier Molbrug

0.5

Lier Maasfort

2.0

Emblem

Kessel

Westkapelle

Cadzand

Vlissingen

Breskens

Borssele

Terneuzen

Hansweert

Walsoorden

Baalhoek

Bath

Zandvliet

Liefkenshoek

Kallo

Antwerpen

Hemiksem

Boom

standardDeviation [m] 0.5 0.5 2.0 0.5 0.5 0.5 0.5 0.5 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0

3.3 Stop Criterion The inner workings of the DUD algorithm are briefly described in Appendix A. The DUD algorithm has two types of stop criteria: one for the outer iterations, and one for the inner iterations. If the linearization yields a new parameter setting with a lower cost, this is called an outer iteration which can stop on the various criteria:  outerLoop maxIterations = 100 The maximum iteration numbers of the outer loop which is set to 100 in this study. 

absTolerance = 0.001 The maximum absolute difference between the costs resulted from the 2 best parameter estimates |(Jnew – Jprevious)|.

 relTolerance = 0.00001 The maximum relative difference between the costs resulted from the 2 best parameter estimates |(Jnew – Jprevious)|/ Jnew.  relToleranceLinearCost = 0.00001 The maximum relative difference between the linearized costs resulted from the 2 best parameter estimates |(Jnew-Jlinear)/(Jprevious- Jlinear)|; However if the linearization does not produce a better estimate, DUD will perform a line search until a better estimate is found. The procedure is stopped when one of the stopping criteria of inner iteration is fulfilled.  lineSearch maxIterations = 30

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

The maximum iteration number of the inner loops which is set to 30 in this study. ďƒ˜ maxRelStepSize = 5 Maximum size of relative step. This indicates how much the parameters may change from one iteration to the other. In practice, the searching step is preferably not set too large so that the searching is still in the surroundings of the initial guess. When the norm of Îą (see Appendix A) is larger than maxRelStepSize, Îą will be reduced in such a way that: Îą_new = |Îą| / maxRelStepSize. ďƒ˜ backtracking shorteningFactor (Îˇ) = 0.5 Factor for shortening step size. ďƒ˜ startIterationNegativeLook = 3 Maximum number of iterations before searching in opposite direction. This determines when parameter đ?œˇđ?œˇ = Âą1 changes sign.

3.4 Results

Table 3 shows the changes of cost function and calibration parameter (in this case Manning coefficient) during the final OpenDA calibration after 29 iterations. The first column shows the iteration numbers of the automatic calibration, the second column shows the cost function as calculated in Â§3.2. Columns 3 to 19 shows the changes to the manning coefficient in the 17 roughness polygons. For N degrees of freedom (the polygons in which a roughness value needs to be assigned), DUD requires a first set of (N+1) runs to establish the sensitivity of the model. Therefore the first 18 runs are the preparation for the optimization algorithm. The actual optimization procedure (e.g. linearization and line search) starts from iteration number 19. The cost function drops gradually until iteration number 26, at which the minimum cost function is found of 31234.08. Afterwards the cost function starts oscillating, therefore the automatic calibration is terminated by the user. The automatic calibration did not stop automatically because the default stop criterion was too strict for this particular case. This calculation shows that none of the stop criterion is met after 26 iterations. đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’? đ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Žđ?’Ž = 4 < 100

đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’“đ?’“đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚đ?’‚ = |đ?‘›đ?‘›đ?‘›đ?‘›đ?‘›đ?‘› đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘? âˆ’ đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘? đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?| = |31234.08 âˆ’ 31234.82| = 0.74 > 0.001

đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“ =

|đ?‘›đ?‘›đ?‘›đ?‘›đ?‘›đ?‘› đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘? âˆ’ đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘? đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?| |đ?‘›đ?‘›đ?‘›đ?‘›đ?‘›đ?‘› đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?|

đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’†đ?’†đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’“đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’?đ?’? =

|31234.08 âˆ’ 31234.82| 31234.08

|đ?‘›đ?‘›đ?‘›đ?‘›đ?‘›đ?‘› đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘? âˆ’ đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™đ?‘™ đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?| |đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘? đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?tâˆ’đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘–đ?‘– đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?đ?‘?|

= 2.39E âˆ’ 5 > 1.0E âˆ’ 5

|31234.08 âˆ’ 31026.28| |31234.82 âˆ’ 31026.28|

= 1.0 > 1.0E âˆ’ 5

The evolution of the cost function during the final OpenDA calibration are demonstrated in Figure 3. The cost function is slightly reduced during the entire calibration study. Table 4 and Figure 4 compare the Manning coefficients before and after OpenDA calibration. The values of the Manning coefficients are almost unchanged. The final calibrated roughness map of Manning Coefficient is demonstrated in Figure 5 and Figure 6.

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Table 3 â€“ Changes of cost function and parameter (manning coefficient) during the final calibration with OpenDA. Iteration Cost A-310 A-311 A-433 A-436 A-437 A-440 A-441 A-450 A-455 A-457 A-459 A-461 A-462 A-463 A-464 A-465 A-467 1 31284.66 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 31704.84 0.001 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 31298.83 0 0.001 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4 31263.16 0 0 0.001 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5 31381.65 0 0 0 0.001 0 0 0 0 0 0 0 0 0 0 0 0 0 6 31307.6 0 0 0 0 0.001 0 0 0 0 0 0 0 0 0 0 0 0 7 31363.96 0 0 0 0 0 0.001 0 0 0 0 0 0 0 0 0 0 0 8 31290.98 0 0 0 0 0 0 0.001 0 0 0 0 0 0 0 0 0 0 9 31339.36 0 0 0 0 0 0 0 0.001 0 0 0 0 0 0 0 0 0 10 31413.23 0 0 0 0 0 0 0 0 0.001 0 0 0 0 0 0 0 0 11 31364.91 0 0 0 0 0 0 0 0 0 0.001 0 0 0 0 0 0 0 12 31442.23 0 0 0 0 0 0 0 0 0 0 0.001 0 0 0 0 0 0 13 31372.14 0 0 0 0 0 0 0 0 0 0 0 0.001 0 0 0 0 0 14 31711.36 0 0 0 0 0 0 0 0 0 0 0 0 0.001 0 0 0 0 15 31420.08 0 0 0 0 0 0 0 0 0 0 0 0 0 0.001 0 0 0 16 31362.45 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.001 0 0 17 31270.83 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.001 0 18 31313.77 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.001 19 31240.14 1.87E-06 1.77E-04 1.49E-04 1.61E-04 9.24E-05 6.39E-05 8.34E-05 -1.86E-05 -9.04E-05 -7.57E-06 -5.14E-05 7.36E-05 -6.02E-05 1.99E-05 -4.61E-06 1.56E-04 1.47E-04 20 31244.95 -4.95E-06 1.87E-04 1.56E-04 1.76E-04 9.10E-05 6.36E-05 7.97E-05 -3.39E-05 -1.08E-04 -1.73E-05 -6.58E-05 7.27E-05 -8.90E-06 7.31E-06 -1.76E-05 1.63E-04 1.50E-04 21 31251.8 -1.54E-06 1.82E-04 1.52E-04 1.69E-04 9.17E-05 6.37E-05 8.15E-05 -2.63E-05 -9.92E-05 -1.24E-05 -5.86E-05 7.31E-05 -3.46E-05 1.36E-05 -1.11E-05 1.59E-04 1.49E-04 22 31250.29 1.65E-07 1.80E-04 1.51E-04 1.65E-04 9.21E-05 6.38E-05 8.24E-05 -2.25E-05 -9.48E-05 -1.00E-05 -5.50E-05 7.34E-05 -4.74E-05 1.68E-05 -7.86E-06 1.58E-04 1.48E-04 23 31235.44 1.02E-06 1.78E-04 1.50E-04 1.63E-04 9.23E-05 6.39E-05 8.29E-05 -2.06E-05 -9.26E-05 -8.78E-06 -5.32E-05 7.35E-05 -5.38E-05 1.83E-05 -6.24E-06 1.57E-04 1.48E-04 24 31234.82 3.79E-07 2.00E-04 1.64E-04 1.83E-04 8.93E-05 6.34E-05 8.00E-05 -4.61E-05 -1.23E-04 -2.26E-05 -7.43E-05 7.60E-05 -1.51E-05 4.40E-06 -2.65E-05 1.73E-04 1.58E-04 25 31238.72 3.87E-07 2.10E-04 1.71E-04 1.90E-04 8.70E-05 5.94E-05 7.88E-05 -6.20E-05 -1.48E-04 -3.78E-05 -2.19E-05 6.70E-05 -1.55E-05 -5.00E-08 -3.99E-05 1.80E-04 1.64E-04 26 31234.08 3.83E-07 2.05E-04 1.68E-04 1.87E-04 8.81E-05 6.14E-05 7.94E-05 -5.41E-05 -1.35E-04 -3.02E-05 -4.81E-05 7.15E-05 -1.53E-05 2.17E-06 -3.32E-05 1.76E-04 1.61E-04 27 31240.26 4.00E-07 2.19E-04 1.77E-04 1.98E-04 8.71E-05 5.83E-05 7.73E-05 -7.48E-05 -1.62E-04 -4.22E-05 -2.55E-05 7.06E-05 -1.60E-05 4.94E-06 -4.99E-05 1.87E-04 1.70E-04 28 31243.13 3.91E-07 2.12E-04 1.72E-04 1.92E-04 8.76E-05 5.99E-05 7.83E-05 -6.44E-05 -1.49E-04 -3.62E-05 -3.68E-05 7.11E-05 -1.56E-05 3.56E-06 -4.15E-05 1.82E-04 1.65E-04 29 31236.08 3.87E-07 2.08E-04 1.70E-04 1.90E-04 8.79E-05 6.06E-05 7.89E-05 -5.93E-05 -1.42E-04 -3.32E-05 -4.25E-05 7.13E-05 -1.55E-05 2.87E-06 -3.74E-05 1.79E-04 1.63E-04

Figure 3 â€“ Evolution of cost function during the final calibration with OpenDA.

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Table 4 â€“ Comparison of manning coefficient before and after the final calibration with OpenDA.

Roughness Polygons

Water Level Stations

A-310

Vlakte van de Raan; Westkapelle; Cadzand; Vlissingen;Breskens

A-311 A-433 A-436 A-437 A-440 A-441 A-450 A-455 A-457 A-459 A-461

Terneuzen;Borssele Hansweert Walsoorden;Baalhoek Bath Zandvliet Liefkenshoek;Kallo Antwerp Hemiksem Temse

A-462

StAmands; Dendermonde; Schoonaarde

A-463 A-464 A-465

Wetteren Melle Tielrode

A-467

Boom; Wallem; Hombeek; Zemst; Mechelen_Benedensluis; Rijmenam; Duffel; Lier_Molbrug; Lier_Maasfort; Emblem; Kessel

Manning Coefficient Manning Coefficient (Initial Guess) (After final calibration) Run: Work20 Run: Work26 Offshore

Western Scheldt

Lower Sea Scheldt

0.0217

0.0213 0.0198 0.0304 0.0249 0.0261 0.0147 0.0293 0.0217 0.0235 0.0225 0.0240

0.0215 0.0199 0.0306 0.0250 0.0262 0.0148 0.0293 0.0216 0.0235 0.0224 0.0240

0.0137

0.0161 0.0196 0.0263

0.0161 0.0196 0.0265

0.0221

0.0222

Upper Sea Scheldt

Rupel Basin

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Figure 4 â€“ Comparison of Manning coefficient before and after the final OpenDA calibration.

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Figure 5 â€“ Final calibrated bottom roughness map of Manning Coefficient.

Figure 6 â€“ Final calibrated bottom roughness map of Manning Coefficient (Zoom in to Upper stream).

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

4 Validation In this Chapter, the model performance is evaluated with the in-house postprocessing tool VIMM written in MATLAB. VIMM is a toolbox developed in-house at Flanders Hydraulics to assist the modeller during calibration and validation of hydraulic models. We compare the model performance between 2 different model runs (Table 5). Table 5 – Description of model runs involved in the model validation.

Run004

Run work20 as reported by Chu et al (2020 as shown in Table 4 (before final calibration).

Run005

Run work26 as shown inTable 4 (after final calibration).

4.1 Boundary Conditions Boundary conditions aren’t changed during the 3-step calibration. They are nested from DCSMv6-ZUNOv4.

4.2 Water Level Timeseries Table 7 compare the statistics of Bias, RMSE and RMSE0 (see definitions in Appendix B) of the complete time series. The statistical values are color-coded by the definition shown in Table 6. Figure 7 to Figure 15 present the statistical comparison between Run004 and Run005. In general the differences on water level predictions between Run004 and Run005 are negligible throughout the estuary, except at Hombeek and Zemst (e.g. the RMSE is reduced by 1.5 cm and 2.7 cm respectively). Run004 and Run005 are identical in model setups except the bottom friction map, apart from which Run004 ran at FHR with 32 processors while Run005 ran at Deltares with 20 processors. Normally the model results are not (and should not be) dependent on the number of processors. However some tests at FHR have shown that the Zenne river is sensitive to the number of processors used by the model.

Table 6 – Definition of colour code in terms of bias, RMSE and RMSE0.

Legend

Final version

|Bias| [cm] 0-5 5-10 10-15 15-20 >20

RMSE [cm] 0-5 5-10 10-15 15-20 >20

WL2020R13_041_3

RMSE0 [cm] 0-5 5-10 10-15 15-20 >20

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Table 7 – Comparison of Bias, RMSE and RMSE0 of the complete time series. Complete TimeSeries Stations

Run004 BIAS

RMSE

Run005 RMSE_0

BIAS

RMSE

Run005 – Run004 RMSE_0

BIAS [cm]

RMSE [cm] RMSE_0 [cm]

Westhinder

-0.3

2.6

-0.3

2.6

0.0

Vlakte van de

-1.4

4.6

4.4

-1.4

4.6

4.4

0.0

Westkapelle

-0.4

4.1

-0.4

4.1

0.0

Cadzand

2.0

5.8

5.5

2.0

5.8

5.4

0.0

Vlissingen

-1.4

4.8

4.6

-1.4

4.8

4.6

0.0

Breskens

-0.6

5.0

4.9

-0.6

5.0

4.9

0.0

Borssele

-0.5

5.2

-0.5

5.2

0.0

Terneuzen

-1.6

5.7

5.5

-1.6

5.8

5.5

0.0

0.1

Hansweert

1.1

5.5

5.4

1.1

5.6

5.5

0.0

Walsoorden

0.7

6.0

5.9

0.7

6.1

6.0

0.0

0.1

Baalhoek

1.6

6.2

6.0

1.5

6.3

6.1

0.0

0.1

Bath

4.0

8.3

7.2

4.0

8.2

7.1

0.0

-0.1

Zandvliet

2.5

9.0

8.6

2.4

8.7

8.4

0.0

-0.2

Liefkenshoek

4.7

8.8

7.4

4.6

8.7

7.4

0.0

-0.1

Kallo

6.2

10.1

8.0

6.1

10.0

8.0

-0.1

0.0

Antwerpen

6.3

10.4

8.3

6.2

10.3

8.3

-0.1

Hemiksem

-1.5

8.2

8.0

-1.6

8.1

8.0

-0.1

0.0

-0.1

Boom

-0.3

8.1

-0.4

8.0

-0.1

Temse

0.7

8.0

0.6

8.0

-0.1

0.0

Tielrode

-0.5

10.0

9.9

-0.6

10.0

-0.1

0.0

Walem

1.0

8.2

8.1

0.9

8.0

-0.1

StAmands

-5.9

16.9

15.8

-6.1

16.8

15.6

-0.1

-0.2

Dendermonde

8.7

21.6

19.8

8.5

21.6

19.8

-0.2

0.0

0.1

Schoonaarde

-7.0

15.1

13.4

-7.1

15.2

13.4

-0.2

0.1

0.0

Wetteren

-1.6

16.1

-1.8

16.2

16.1

-0.2

0.0

-0.2

0.0

Melle

4.6

22.2

21.7

4.4

22.2

21.7

MechelenSluis

12.3

23.0

19.4

12.3

23.2

19.7

0.0

0.2

0.3

Hombeek

10.1

20.9

18.2

8.0

19.3

17.6

-2.1

-1.5

-0.7

Zemst

13.3

30.0

26.9

9.8

27.3

25.5

-3.5

-2.7

-1.4

Duffel

-3.5

10.1

9.5

-3.5

10.3

9.6

0.0

0.2

Rijmenam

-2.3

23.3

23.2

-2.3

23.4

23.2

0.0

0.1

Lier Molbrug

-17.0

23.3

15.9

-17.1

23.3

15.9

-0.1

0.0

-0.1

Lier Maasfort

-17.7

25.5

18.3

-17.8

25.5

18.2

-0.1

0.0

-0.1

Emblem

-16.7

25.2

18.9

-16.6

25.0

18.7

0.0

-0.2

-8.7

19.6

17.5

-8.8

19.6

17.5

0.0

Kessel

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Figure 7 â€“ Bias of complete time series of water levels along the Scheldt.

Figure 8 â€“ RMSE of complete time series of water levels along the Scheldt.

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Figure 9 â€“ RMSE0 of complete time series of water levels along the Scheldt.

Figure 10 â€“ Bias of high water levels along the Scheldt.

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Figure 11 â€“ RMSE of high water levels along the Scheldt.

Figure 12 â€“ RMSE0 of high water levels along the Scheldt.

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Figure 13 â€“ Bias of low water levels along the Scheldt.

Figure 14 â€“ RMSE of low water levels along the Scheldt.

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Figure 15 â€“ RMSE0 of low water levels along the Scheldt.

4.3 Harmonic Analysis of Water Levels Figure 16 to Figure 19 compare the model predicted tidal components of M2 and S2 along the River Scheldt. Figure 20 shows the vector difference (see definition in Appendix C). In general Run004 and Run005 produce identical predictions on water level in the frequency domain throughout the estuary. Slight differences on M2 amplitude are observed at at Hombeek and Zemst. Figure 16 â€“ M2 amplitude along the River Scheldt.

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Figure 17 â€“ M2 phase along the River Scheldt.

Figure 18 â€“ S2 amplitude along the River Scheldt.

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5 Figure 19 â€“ S2 phase along the River Scheldt.

Figure 20 â€“ Vector differences along the River Scheldt.

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

5 Conclusions The project goal is the autimatic calibration of the WAQUA-Schelde_NEVLA model for the entire year of 2007. Automatic model generation is performed through Baseline (based on an existing grid). The hydraulic model is run in Simona-Waqua. Automatic calibration is performed through OpenDA. This report describes the final automatic calibration (3rd step) for the entire model domain starting from the calibrated roughness field from the 2nd step (described in Chu et al., 2019 & 2020). The main findings are:  The 3rd automatic calibration with OpenDA leads only to a limited further reduction of cost function.  The final calibrated bottom roughness map also doesn’t differ much from the previous calibration by Chu et al (2020).  In the future, automatic calibration could be safely performed in 2 steps.

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

6 References Chu, K.; Vanlede, J.; Decrop, B.; Mostaert, F. (2019). Baseline Zuidwestelijke Delta: Set-up and Calibration of model waqua-schelde_nevla-j07_5. Version 2.0. FHR Reports, 13_041_1. Flanders Hydraulics Research: Antwerp. IMDC: I/RA/11502/18.003/KCH/ Chu, K.; Vanlede, J.; Decrop, B.; Mostaert, F. (2020). Baseline Zuidwestelijke Delta: Automatic re-calibration of waqua-schelde_nevla-j07_5 with lower weights in the Rupel basin. Version 2.0. FHR Reports, 13_041_2. Flanders Hydraulics Research: Antwerp. Groenenboom, J.; van der Kaaij, Th.; Plieger, R. (2016). Schelde-estuarium, WAQUA model 5e generatie Werkzaamheden 2016. Deltares Report. Delft, the Netherlands. OpenDA (2016). OpenDA User Documentation. The OpenDA Association.

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

Appendix A

Introduction of DUD

The DUD (Doesn’t Use Derivative) algorithm (Ralston and Jennrich, 1978) is a derivative-free algorithm for nonlinear least squares. It can be seen as a Gauss-Newton method, in the sense that it transforms the nonlinear least square problem into the well-known linear square problem. DUD evaluates and optimizes uncertain model parameters by minimizing a cost function. The parameter values corresponding to the minimum value of the cost function are considered as the optimal parameter values for the given analysis. For N calibration parameters, DUD requires (N+1) set of parameters estimates. The affine function for the linearization is formed through all these (N+1) guesses. Note that the affine function gives exact value at each of the (N+1) points. The resulting least square problem is then solved along the affine function to get a new estimate, whose cost is smaller than those of all other previous estimates. If it does not produce a better estimate, the Dud will perform different steps, like searching in opposite direction and/or decreasing searching-step, until a better estimate is found. Afterwards, the estimate with the largest cost is replaced with the new one and the procedure is repeated for the new set of (N+1) estimates. The procedure is stopped when one of the stopping criteria is fulfilled. The general DUD implementation is given by the following steps: •

Start running first guess and modify each parameter;

•

Linearize the model around these values and solve the linear problem;

•

If this is an improvement, update linearization with new point;

•

Or, do a line-search (only until there is improvement).

Suppose there is one uncertain model parameter P ϵ R and a set of n observations y ϵ Rn. The cost function J is expressed as a sum of squares as J(P) = [y-f(P)]T [y-f(P)] f is the model function. Note that calculating J() involves doing one model run, and post-processing the results to calculate the cost function. The DUD optimization starts with an unperturbed run (with initial guess P1) and one sensitivity run (P2). The corresponding function values f(P1) and f(P2) are stored in memory. The parameters are sorted according to their cost function J(P) from high to low. Let’s assume that in this case P1 results in a higher cost function compared with P2. A linear approximation L(α) is built of the function f around P2: L(α) = f(P2) + α ΔF

where ΔF = f(P1) - f(P2)

In the cost function, the model evaluation f(P) is substituted by the linearized function L(α). The linearized cost function J(α) becomes: J(α) = [y- f(P2) – α ΔF]T [y- f(P2) – α ΔF] We then solve for α that minimizes the linearized cost function J(α). After some algebra, this gives: α = (ΔFTΔF)-1 ΔFT[y – f(P2)] The new parameter estimation of P3 is then based on • • •

P2, the difference between the two best estimates (ΔP = P1-P2), and the parameter α.

Final version

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

P3 = P2 + Îą Î”P If J(P3) is less than J(P2), then P1 and J(P1) are tossed out from memory, P2 is replaced by P3, and the two estimates are again sorted according to their corresponding cost function (P2 and P3). This is called the outer iteration. If J(P3) is not less than J(P2), then a line search is done in between P2 and P3: P3 = P2 + Îí (P3-P2) The step size Îí is calculated as: Îí = ďż˝

đ?&#x;?đ?&#x;?, đ?’Šđ?’Š = đ?&#x;Žđ?&#x;Ž đ?’Šđ?’Š đ?œˇđ?œˇ Ă— (đ?œźđ?œź) , đ?’Šđ?’Š > đ?&#x;?đ?&#x;?

The line search stops until P3 is reached for which J(P3) is less than J(P2), if that exists. This is called the inner iteration. The parameter đ?œˇđ?œˇ = Âą1 determines whether the line search is continued in a positive (+) or negative (-) direction. In this study, negative search starts when i>3 (startIterationNegativeLook=3). Îˇ is the shorteningFactor = 0.5 in this study. The procedure is stopped as soon as one of the stopping criteria is fulfilled (see Â§3.3). The best fit will be given by the value of P with the lowest cost J. For a more detailed descriptions of the DUD algorithm, the reader is referred to OpenDA (2016).

WL2020R13_041_3

Final version

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

Appendix B

Definition of Statistics

Water levels The Bias of water level represents the average deviation of the differences between model predicted water level and measurement. The RMSE of water level is a measure of the spread of the predicted values level around the measurement. It corresponds to a sample standard deviation. The RMSE0 is the bias corrected root mean square error which describes the forecast errors not associated with the bias. The mathematical expressions are listed below. y and x represent modelled and measured values respectively and n is the number of samples.

Bias=

y−x n

RMSE =

∑ (y − x ) 1

Final version

RMSE0 =

∑ ((y − x ) − ( y − x )) 1

WL2020R13_041_3

Baseline Zuidwestelijke Delta - Final calibration of model waqua-schelde_nevla-j07_5

Appendix C

Definition of Vector Difference

The vector difference analysis combines the results from different tidal components regarding both amplitude and phase. In short vector difference is a unified variable with one value describing the model accuracy from harmonic point of view. The mathematical expression of vector difference is shown as below.

= es

∑

N i =1

[Ac,i cos( φc,i ) − Am,i cos( φm,i )] 2 + [Ac,i sin( φc,i ) − Am,i sin( φm,i )] 2

where es is the vector difference calculated at a certain station. c and m represent the model computed and measured value. A and φ represent the tidal amplitude and phase. i represents the number of tidal components.

WL2020R13_041_3

Final version

DEPARTMENT MOBILITY & PUBLIC WORKS Flanders hydraulics Research Berchemlei 115, 2140 Antwerp T +32 (0)3 224 60 35 F +32 (0)3 224 60 36 waterbouwkundiglabo@vlaanderen.be www.flandershydraulicsresearch.be