Post on 01-Jun-2015
20 juni 2014
3D lagoon modeling using D-Flow FM:
Baía de São Marcos
Lucas Silveira, Jonas Oliveira, Alex Falkenberg, Luana
Taiani, Leticia Nascimento, Joao Dobrochinski
(Arnold van Rooijen)
20 juni 2014
Rio de Janeiro
Florianópolis
Study location:
Baía de São Marcos
Porto Alegre (NL-AUS)
Salvador (NL-SP) Baía de São Marcos
20 juni 2014
Content
• Introduction
• Delft3D modeling
• D-Flow FM curvilinear
• D-Flow FM unstructured
Introduction
• Baía de São Marcos
(Maranhão, NE Brazil)
• Mearim River discharge up
to 2000 m3/s
• Tidal range
• 4-5 m (ocean)
• Up to 7m within bay
• Tidal bore phenomenon
(Pororoca)
• Navigation channel all the
way into the river
20 juni 2014
Terminal Portuario do
Mearim
Introduction
Measurements in 2009
• 2 tidal gauges (Aug-Sep)
• 8 ADCP (Mar 10 – May 28)
Bathymetric surveys:
2007 and 2009
20 juni 2014
Delft3D – model setup
20 juni 2014
Delft3D –model setup
20 juni 2014
• Boundary conditions
• Water level (North)
• Total discharge (South)
• No wind forcing
• Calibration
• C = 65 and 105 m0.5/s
Delft3D – model results
R2 = 0.95
RMSE = 0.40 m
20 juni 2014
R2 = 0.96
RMSE = 0.42 m
Delft3D – model results
20 juni 2014
P2A
P1A
P1B
D-Flow FM curvilinear
• Converted Delft3D model using dflowfmConverter.m (Open Earth
Tools)
• Same grid,
• Same boundary conditions,
• Same model settings,
• Fixed computational timestep,
• 2DH vs. 3D model
20 juni 2014
D-Flow FM curvilinear
• Water levels 2DH model
20 juni 2014
D-Flow FM curvilinear
• Water levels 2DH model
20 juni 2014
D-Flow FM curvilinear
• Velocities 2DH model
20 juni 2014
D-Flow FM curvilinear
20 juni 2014
Delft3D D-Flow FM Delft3D D-Flow FM
water level
Taua2 0.952 0.951 0.402 0.408
Perizes2 0.959 0.956 0.423 0.44
Mean 0.956 0.954 0.413 m 0.424 m
velocity
Canal2A 0.921 0.922 0.322 0.345
P1A 0.966 0.967 0.345 0.322
P1B 0.914 0.95 0.304 0.242
P1C 0.967 0.979 0.25 0.221
P2A 0.921 0.913 0.294 0.306
P2B 0.966 0.955 0.331 0.35
P2C 0.977 0.974 0.292 0.318
P3 0.954 0.964 0.238 0.227
Mean 0.948 0.952 0.306 m/s 0.3 m/s
RMSER2
2DH model
D-Flow FM curvilinear
20 juni 2014
Delft3D D-Flow FM Delft3D D-Flow FM
water level
Taua2 0.953 0.938 0.395 0.46
Perizes2 0.96 0.947 0.423 0.466
Mean 0.956 0.943 0.409 m 0.463 m
velocity
Canal2A 0.876 0.922 0.538 0.328
P1A 0.91 0.965 0.32 0.284
P1B 0.697 0.923 0.607 0.293
P1C 0.809 0.967 0.634 0.249
P2A 0.85 0.919 0.369 0.283
P2B 0.865 0.953 0.531 0.335
P2C 0.845 0.961 0.662 0.324
P3 0.89 0.956 0.411 0.221
Mean 0.843 0.946 0.509 m/s 0.29 m/s
R2 RMSE
3D model
D-Flow FM unstructured
20 juni 2014
• Setting up a new grid using the full flexible
mesh functionality:
• Using DeltaShell user interface
• ‘Curvilinear grids where possible, triangles
etc. where needed’
• Same forcing, settings etc.
D-Flow FM unstructured
Preliminary results
20 juni 2014
Fixed timestep
(30 s)
Auto timestep
D-Flow FM unstructured
Preliminary results
20 juni 2014
D-Flow FM unstructured
Preliminary results
20 juni 2014
D-Flow FM unstructured
20 juni 2014
D-Flow FM unstructured
20 juni 2014
Delft3D D-Flow FM Delft3D D-Flow FM
water level
Taua2 0.952 0.963 0.402 0.44
Perizes2 0.959 0.942 0.423 0.542
Mean 0.956 0.952 0.413 m 0.491 m
velocity
Canal2A 0.921 0.92 0.322 0.399
P1A 0.966 0.913 0.345 0.566
P1B 0.914 0.971 0.304 0.215
P1C 0.967 0.979 0.25 0.201
P2A 0.921 0.796 0.294 0.553
P2B 0.966 x 0.331 x
P2C 0.977 x 0.292 x
P3 0.954 0.955 0.238 0.553
Mean 0.948 0.922 0.306 m/s 0.376 m/s
R2 RMSE
Model efficiency
Processor: Intel(R) Core™i7-3930K CPE @ 3.20GHz 3.20 GHz
Memory (RAM): 16,0 GB
System Type: 64-bit Operating System
Number of Threads: 12
20 juni 2014
Delft3D D-Flow Curvilinear D-Flow Unstructured
2D 258 240 202
2D (auto ts) x 330 321
3D 2248 903 x
3D (auto ts) x 3308 x
• D-Flow FM faster than Delft3D, especially in 3D
• Probably (a.o.) due to OpenMP multicore
Conclusions
20 juni 2014
• Delft3D model of Baía de São Marcos was converted to a D-Flow
FM model
• Curvilinear (Delft3D grid)
• Unstructured grid (from scratch)
• Results Delft3D vs. D-Flow FM curvilinear very similar
• DFM runs a little faster for this setup
• D-Flow FM unstructured model seems to perform similar in
accuracy, however there was a numerical issue halfway the
simulation
Next steps
• Watch the World Cup
• Look into numerical instability in unstructured model (adjust grid)
• Look into 3D model performance
20 juni 2014
Extra slides
Root Mean Square Error (RMSE) (Equation 1):
𝑅𝑀𝑆𝐸 = (𝑦𝑠𝑖𝑚𝑢𝑙𝑎𝑡𝑒𝑑 − 𝑦𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑)
2
𝑛
RMSE-observations Standard deviation Ratio (RSR) (Equation 2) :
𝑅𝑆𝑅 = 𝑅𝑀𝑆𝐸
𝑆𝑇𝐷𝐸𝑉𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑=
(𝑦𝑠𝑖𝑚𝑢𝑙𝑎𝑡𝑒𝑑 − 𝑦𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑)2
(𝑦𝑠𝑖𝑚𝑢𝑙𝑎𝑡𝑒𝑑 −𝑚𝑒𝑎𝑛)2
Determination Coefficient (R²) (Equation 3):
𝑅² = ( (𝑥1 − 𝑥𝑚)(𝑦1 − 𝑦𝑚)
𝑛 − 1
(𝑥1 − 𝑥𝑚)²𝑛 − 1
. (𝑦1 − 𝑦𝑚)²
𝑛 − 1
)2
20 juni 2014
20 juni 2014