This document discusses the development of the new Dutch Continental Shelf Model version 6 (DCSMv6) operational water level forecasting model. DCSMv6 was converted from the original WAQUA model framework to the Delft3D Flexible Mesh (D-Flow FM) engine. While D-Flow FM provided more flexibility, initial runs showed slightly reduced accuracy compared to WAQUA. With further calibration, the accuracy gap could be closed. The new high-resolution DCSMv6 model shows excellent agreement with measurements across the Northwest European Shelf and North Sea but D-Flow FM is currently slower than WAQUA due to differences in parallelization efficiency.

1. An application of Delft3D Flexible Mesh for operational water-level forecasting on the Northwest European Shelf and North Sea (DCSMv6)
Firmijn Zijl
NGHS Symposium, Delft Software Days
November 5, 2014

2. Introduction (Acceptance testing)

3. NGHS acceptance testing
Models considered for acceptance testing
• Official RWS area schematisations for:
•Northwest European Shelf and North Sea
•Rhine-Meuse Entrance
•Meuse River
•Waal River
November 5, 2014
Model
Characteristics
Firth of Clyde
2D, tide surge model
Lake Grevelingen
3D, weakly dynamic system with both temperature and salinity stratification
Neptune (Singapore coastal waters)
3D dynamic estuarine system with salinity and temperature stratification (and intrusion)
Sea of Marmara
3D, both highly and weakly dynamic parts, complex combination of salinity and temperature stratification (CIL)
FSRU Jebel Ali
3D, buoyant plume dispersion

4. Focus of this presentation
•Development of completely redesigned new generation Dutch Continental Shelf Model version 6 (DCSMv6).
•Originally developed using the WAQUA module of SIMONA framework, for numerical modelling of 2D free surface flows
1) Conversion to D-Flow FM* and comparison to WAQUA
Impact on results
Computational time
2) Model improvements made possible by enhanced functionality and flexibility of D-Flow FM
E.g. spatially varying resolution to increase computational speed without loss of accuracy
November 5, 2014
* D-Flow FM: hydrodynamic simulation engine of Delft3D Flexible Mesh

5. Operational forecasting in The Netherlands
November 5, 2014

6. Introduction
The Northwest European Shelf and North Sea Need for accurate, real-time operational water level forecasting:
•On a daily basis, it is important for port operations and to ensure maritime safety on busy shipping routes
•For the coastal regions of the Netherlands, it is crucial, since large areas of the land lie below sea level
•During storm surges, detailed and timely water-level forecasts provided by an operational storm surge forecasting system are necessary to support, e.g. the decision for closure of the movable storm surge barriers in the Eastern Scheldt and the Rotterdam Waterway
November 5, 2014

7. Real-time forecasting in The Netherlands
Developments in real-time flood forecasting in the Netherlands
•Storm Surge Warning Service (SVSD), in close cooperation with the Royal Netherlands Meteorological Institute (KNMI), is responsible for operational forecasts and issuing warnings to coastal authorities in case of high water threats.
•After the November 2006 All Saints storm it was decided that further improvements in model framework were required
• Decision to completely redesign the operational model
• New generation model is part of a comprehensive development to upgrade the operational forecasting system for the North Sea
November 5, 2014

8. DCSMv6 model development
November 5, 2014
tide + surge
surge only

9. Model setup
November 5, 2014

10. DCSMv6 (model grid and bathymetry)
Model setup - computational grid
• Increased spatial coverage
• Uniform cell size of 1.5’ (1/40°) in east-west direction and 1.0’ (1/60°) in north-south direction (~nautical mile)
• Around 106 active grid cells Model setup – bathymetry
• Initially based on NOOS gridded bathymetry data set, supplemented by ETOPO2
• Changes made during calibration
November 5, 2014

11. Model setup (boundary forcing)
Model setup - boundary forcing
• Open boundary with 205 sections
• Distinction made between 2 components of the water level elevation:
(1)Tide, defined in frequency domain (22 constituents)
(2)Surge, as an inverse barometer correction (IBC) based on time and space varying pressure fields
November 5, 2014

12. DCSMv6-ZUNOv4 model setup (meteo forcing)
Model setup - meteo forcing
• Wind speed and air pressure from HIRLAM
model provided (operationally) by KNMI
• Sea surface roughness is calculated using
the Charnock relation (Charnock parameter
0.025)
Model setup - miscellaneous
• Spatially varying manning bed roughness
• Tide Generating Forces (TGF) included
November 5, 2014

13. Calibration approach
November 5, 2014

14. Model calibration
Calibration and validation using tide gauge data at >120 locations
Green dots: radar altimeter cross-over locations Red dots: in-situ tide-gauge locations
November 5, 2014

15. DCSMv6 (against satelite altimeter data)
M2 Amplitude
M2 Phase
November 5, 2014

16. DCSMv6 Model Development
OpenDA-DUD experiment setup and parameters
•Reduction of uncertainty in bottom friction coefficient and bathymetry
•Multiple optimization runs, with increasing length and number of parameters
•Final experiment (DCSMv6)
• 200 control parameters, 12 months, ~100 observations locations
• To achieve desired accuracy many interactions become important (in terms of processes and geographically)
November 5, 2014

17. Results
November 5, 2014

18. Dutch coastal stations
RMSE
(tide)
RMSE (surge)
RMSE
(total)
RMSE
(high water)
RMSE
(low water)
DCSMv5
10.7
7.7
13.1
11.3
11.0
DCSMv6 (WAQUA)
4.1
5.9
7.2
6.6
7.1
DCSMv6 (D-Flow FM)
4.9
6.0
7.8
6.8
7.1
Goodness-of-Fit (in cm) - 13 Dutch coastal stations - entire year of 2007)
November 5, 2014

19. November 5, 2014
DCSMv6 model comparison
Red: Measurement Black: Computation Blue: Residual
Waqua
D-Flow FM

20. Tide representation in the frequency domain
RMS(ΔA)
RMS(ΔG)
RMS(VD)
Q1
0.4
0.4
6.0
8.4
0.5
0.6
O1
0.2
0.4
2.0
2.0
0.5
0.6
P1
0.8
0.7
5.2
5.7
0.9
0.8
K1
0.2
0.2
2.4
1.6
0.4
0.3
N2
0.3
0.4
1.3
2.6
0.5
0.8
M2
1.1
2.3
1.0
1.5
1.9
3.4
S2
1.1
0.6
1.0
1.7
1.2
1.0
K2
0.8
0.5
1.6
3.6
0.8
0.7
M4
0.6
1.0
8.7
12.3
1.5
2.3
M6
0.6
0.7
11.5
11.2
1.3
1.4
A: amplitude in cm, G: phase in °, VD: vector difference in cm
Based on 13 Dutch coastal stations
November 5, 2014

21. Tide representation in the frequency domain
M2 amplitude and phase errors along Dutch coast
WAQUA
D-Flow FM
November 5, 2014

22. November 5, 2014
DCSMv6-ZUNOv4 (Dutch estuaries and Wadden Sea)
Based on 16 stations in Dutch estuaries en Wadden Sea
RMSE
(tide)
RMSE
(surge)
RMSE
(total)
DCSMv6 (WAQUA)
7.1
7.0
9.9
DCSMv6 (D-Flow FM)
9.9
7.1
12.2
Consistency in depth interpolation options is crucial to achieve similar results

23. Conclusions
November 5, 2014

24. Computational time
Comparison of computational time
•Both models use a computational time step of 2 minutes (determined with convergence tests)
•Performance tested on Deltares cluster (quad-core machines)
•No use made of enhanced flexibility of D-Flow FM
•Speed-up is hardware dependent
•More effort required to improve scaling efficiency
November 5, 2014
number of nodes (with 4 cores each)
comp. time [min per day]
number of nodes (with 4 cores each)
speed-up [-]

25. Summary and conclusions
•Completely redesigned, new generation Dutch Continental Shelf Model version 6 (DCSMv6) has been developed
• Year-long simulations show excellent agreement with (shelf-wide) measurements
• Running the model in D-Flow FM reduces the water level representation accuracy
•It is expected that the D-Flow FM accuracy can be improved with limited re- calibration
• For this application D-Flow FM is twice as slow as Simona-Waqua (on one node)
•Enhanced flexibility of D-Flow FM has not been used, but will save computational time
•Scalability of D-Flow FM is rightly getting more attention (hardware dependent) Next step (1): make use of enhanced flexibility of D-Flow FM to reduce number of cells and increase time step by avoiding unnecessary high resolution in the deeper areas off the shelf Next step (2): Acceptance testing with different models, as results are application dependent
November 5, 2014

26. DCSMv6
Questions?
November 5, 2014