Mesh Orientation and Cell Size
Sensitivity in 2D SWE Solvers
Duncan Kitts (Duncan.kitts@bmtglobal.com)
Greg Collecutt
Shuang Gao
Phillip Ryan
Bill Syme
RiverFlow 2020
Eliminating Grid Sensitivity in Fixed Grid Solvers
Introduction
 Understanding flood risk increasingly
important
 2-Dimensional hydraulic modelling software
commonly used
 High resolution DTM data becoming
available but computational constraints limit
it’s use
 2D Computational Cell > DTM Cell
2D Hydraulic Modelling
Conventional Approach to assigning cell elevations
• Current Cell Volume and Face Schema
• Cell volume based on single cell centre elevation (or
average)
• Cell face based on elevation of single cell side
elevation point
• Need a grid cell size to reflect underlying
topography
Storage of a flat
square cell
ZU
ZV
ZC
Rectangular
channel to next cell
Rectangular
channel to next cell
TUFLOW HPC
Conventional Approach to assigning cell elevations
Cell volume elevation point
Cell face elevation point
Cell Volume / Face Schema
TUFLOW HPC
Conventional Approach to assigning cell elevations
Cell Volume / Face Schema
Cell volume elevation point
Cell face elevation point
TUFLOW HPC
Cell Resolution Effects
Topography 1m Grid Cell 20m Grid Cell
Topography
20m Cell Size with single
cell elevation value
20m Cell Size with sub-grid
sampling
TUFLOW HPC
With SGS
Cell volume / elevation curve
Cell face width / elevation curve
Sub-Grid Cell Volume / Face
Schema
SGS Benchmarking
Rectangular Channel Test
Numerical experiment set-up
• A rectangular channel with length of 1000m and width of 100m
• Flow rate = 100 m3/s
• Depth = 1 m
• Slope = 0.0009
• Manning’s n = 0.03
Flow
L = 1000 m
W = 100m
Uniform flow of U = 1m/s, d = 1m
Theoretical water level and energy
slope using Manning’s equation
d = 1m
Rotated Channel Test
TUFLOW HPC
15° 30° 45°0°
H [m] H [m] H [m] H [m]
The more mis-aligned the mesh is with the flow
direction, the poorer the simulation output
compared to those expected theoretically.
Rotated Channel Test
TUFLOW HPC with SGS
15° 30° 45°0°
H [m] H [m] H [m] H [m]
With SGS, simulated output compares well to
theoretically expected output with sensitivity to
mesh orientation eliminated.
Mesh Size Sensitivity
TUFLOW HPC
50m
H [m]
With a large mesh size, not a great comparison
with theoretically expected outputs.
Mesh Size Sensitivity
TUFLOW HPC
50m25m
H [m]
With a large mesh size, not a great comparison
with theoretically expected outputs.
Mesh Size Sensitivity
TUFLOW HPC
50m25m10m
H [m]
As the mesh size is refined, start to converge on
theoretical outputs.
Mesh Size Sensitivity
TUFLOW HPC
50m25m10m5m
H [m]
As the mesh size is refined, start to converge on
theoretical outputs.
5m
Mesh Size Sensitivity
TUFLOW HPC SGS
H [m]
With SGS, good convergence with theoretical
outputs….
5m
Mesh Size Sensitivity
TUFLOW HPC SGS
10m
H [m]
With SGS, good convergence with theoretical
outputs….
5m
Mesh Size Sensitivity
TUFLOW HPC SGS
10m25m
H [m]
With SGS, good convergence with theoretical
outputs….for range of mesh cell sizes.
5m
Mesh Size Sensitivity
TUFLOW HPC SGS
10m25m50m
H [m]
With SGS, good convergence with theoretical
outputs….for range of mesh cell sizes.
Eliminates sensitivity to mesh size.
Rectangular Channel Test
Deep Sided Channels Unaligned to Mesh
Mesh not aligned
with deep banks
• Distorts streamlines
• Artificial energy losses;
steepens gradient
Traditional Solutions
• 1D channel with cross-section
(time-consuming; full 2D solution compromised)
• Flexible mesh (quadrilaterals aligned with banks)
• Much finer gridded mesh (much longer run times)
• or…
Rectangular Channel Test
Deep Sided Channels Unaligned to Mesh
In conclusion:
SGS
• Conforms with Manning’s equation at all orientations
• Resolves limitation of using gridded meshes along deep sided
channels
• Removes requirement for flexible mesh, higher resolution or 1D
representation.
• Significantly reduces mesh sensitivity
• More accurate flood models and outputs


SGS Benchmarking
U-Bend Flume Test – Experiment Set-up
• Flume experiment conducted by De Vriend (1978)
Flow
H along outer bank
centre line
inner bank
Flow
• Flow rate = 0.189 m3/s
• Downstream H = 0.18 m
• Manning’s n estimated to be 0.0115 ~ 0.0125
R = 4.25m
W = 1.7m
U-Bend Flume Test
TUFLOW HPC 2019 without SGS
• TUFLOW HPC Defaults
H [m]
Use of fixed grid causes edge effects
representing curved wet-dry boundary.
U-Bend Flume Test
TUFLOW FV (Flexible Mesh)
• TUFLOW FV
H [m]
Flexible mesh can be aligned with
wet-dry boundary…although can be
time consuming to do so.
U-Bend Flume Test
TUFLOW HPC without SGS vs TUFLOW FV
TUFLOW HPC TUFLOW FV
H [m]
U-Bend Flume Test
TUFLOW HPC with SGS
• TUFLOW HPC with SGS
H [m]
Partially wet cells/faces smooth out the flow field
SGS approaches eliminates the edge
effects at wet-dry boundary.
U-Bend Flume Test
TUFLOW HPC with SGS vs TUFLOW FV
TUFLOW HPC with SGS TUFLOW FV
H [m]
As good as a well-designed
flexible mesh model!!
TUFLOW HPC
34cm
U-Bend Flume Test – Mesh Size Sensitivity
TUFLOW HPC-Without SGS
17cm10cm5cm
H [m]
• Slightly higher than experimental data due to
fixed grid “saw-tooth” effects.
TUFLOW HPC
34cm
U-Bend Flume Test – Mesh Size Sensitivity
TUFLOW HPC-Without SGS
17cm10cm
H [m]
• Slightly higher than experimental data due to
fixed grid “saw-tooth” effects.
TUFLOW HPC
34cm
U-Bend Flume Test – Mesh Size Sensitivity
TUFLOW HPC-Without SGS
17cm
H [m]
• Slightly higher than experimental data due to
fixed grid “saw-tooth” effects.
• Edge effects worsen with increasing mesh
size.
TUFLOW HPC
34cm
U-Bend Flume Test – Mesh Size Sensitivity
TUFLOW HPC-Without SGS
H [m]
• Slightly higher than experimental data due to
fixed grid “saw-tooth” effects.
• Edge effects worsen with increasing mesh
size.
TUFLOW HPC with SGS
U-Bend Flume Test – Mesh Size Sensitivity
TUFLOW HPC-With SGS
H [m]34cm17cm10cm
Good agreement with experimental data…
5cm
H [m]
TUFLOW HPC with SGS
U-Bend Flume Test – Mesh Size Sensitivity
TUFLOW HPC-With SGS
H [m]34cm17cm10cm
Good agreement with experimental data…
H [m]
TUFLOW HPC with SGS
U-Bend Flume Test – Mesh Size Sensitivity
TUFLOW HPC-With SGS
H [m]34cm17cm
H [m]
Good agreement with experimental data across
range of mesh sizes.
TUFLOW HPC with SGS
U-Bend Flume Test – Mesh Size Sensitivity
TUFLOW HPC-With SGS
H [m]34cm
Good agreement with experimental data across
range of mesh sizes.
H [m]
U-Bend Flume Test
Conclusions
H [m]
In conclusion:
Sub-Grid Sampling...
• Resolves limitation of using gridded meshes along curved channels
• Removes requirement for flexible mesh or higher resolution.
• Significantly reduces mesh sensitivity
• More accurate flood models and outputs
Real World Applications
South Johnstone River Catchment
South Johnstone River Case Study
• Whole catchment model
• Direct rainfall
• Historic major event calibration
20km
Real World Applications
South Johnstone River Catchment
Flow Calibration
SGS provides more comparable
results across range of mesh sizes
compared to conventional approach
Real World Applications
Plynlimon Catchment
Plynlimon Catchment
• Research catchment in mid Wales managed
by Centre of Hydrology, Bangor
• Heavily instrumented with long rainfall and
flow records
• Soils/land cover well mapped
• High resolution LiDAR data for majority of
catchment
Plynlimon TUFLOW Model
• Direct rainfall
• Moderate event
Real World Applications
Plynlimon Catchment Application
Plynlimon Catchment Model
• Direct rainfall
• Moderate event
• Compare sensitivity of mesh size to
observed flow data (red line)
With conventional approach:
• Excellent cell size convergence
• Spread with mesh resolution
TUFLOW conventional approach
shows excellent cell size convergence
but show mesh dependency
Real World Applications
Plynlimon Catchment Application
Plynlimon Catchment Model
• Direct rainfall
• Moderate event
• Compare sensitivity of grid size to observed
flow data (red line)
With conventional approach:
• Excellent cell size convergence
• Spread with mesh resolution
With SGS approach:
• Limited variation with mesh resolution
SGS provides near identical results
across range of mesh sizes
Conclusions
• Sub-Grid Sampling has potential to eliminate sensitivity to both:-
• Mesh Orientation
• Mesh Size
• Benefits seen at range of scales and applications:-
• Whole of catchment modelling
• Urban environments
• Easier Calibration and uncertainty analysis
• Long term continuous simulation can be run rapidly:-
• Natural Flood Management
• Sustainable Urban Drainage Systems
• Improved Forecasting

Mesh_Orientation_and_Cell_Size_Senstivity_in_2D_SWE_Solvers

  • 1.
    Mesh Orientation andCell Size Sensitivity in 2D SWE Solvers Duncan Kitts (Duncan.kitts@bmtglobal.com) Greg Collecutt Shuang Gao Phillip Ryan Bill Syme RiverFlow 2020
  • 2.
    Eliminating Grid Sensitivityin Fixed Grid Solvers Introduction  Understanding flood risk increasingly important  2-Dimensional hydraulic modelling software commonly used  High resolution DTM data becoming available but computational constraints limit it’s use  2D Computational Cell > DTM Cell
  • 3.
    2D Hydraulic Modelling ConventionalApproach to assigning cell elevations • Current Cell Volume and Face Schema • Cell volume based on single cell centre elevation (or average) • Cell face based on elevation of single cell side elevation point • Need a grid cell size to reflect underlying topography Storage of a flat square cell ZU ZV ZC Rectangular channel to next cell Rectangular channel to next cell
  • 4.
    TUFLOW HPC Conventional Approachto assigning cell elevations Cell volume elevation point Cell face elevation point Cell Volume / Face Schema
  • 5.
    TUFLOW HPC Conventional Approachto assigning cell elevations Cell Volume / Face Schema Cell volume elevation point Cell face elevation point
  • 6.
    TUFLOW HPC Cell ResolutionEffects Topography 1m Grid Cell 20m Grid Cell
  • 7.
    Topography 20m Cell Sizewith single cell elevation value 20m Cell Size with sub-grid sampling
  • 8.
    TUFLOW HPC With SGS Cellvolume / elevation curve Cell face width / elevation curve Sub-Grid Cell Volume / Face Schema
  • 9.
    SGS Benchmarking Rectangular ChannelTest Numerical experiment set-up • A rectangular channel with length of 1000m and width of 100m • Flow rate = 100 m3/s • Depth = 1 m • Slope = 0.0009 • Manning’s n = 0.03 Flow L = 1000 m W = 100m Uniform flow of U = 1m/s, d = 1m Theoretical water level and energy slope using Manning’s equation d = 1m
  • 10.
    Rotated Channel Test TUFLOWHPC 15° 30° 45°0° H [m] H [m] H [m] H [m] The more mis-aligned the mesh is with the flow direction, the poorer the simulation output compared to those expected theoretically.
  • 11.
    Rotated Channel Test TUFLOWHPC with SGS 15° 30° 45°0° H [m] H [m] H [m] H [m] With SGS, simulated output compares well to theoretically expected output with sensitivity to mesh orientation eliminated.
  • 12.
    Mesh Size Sensitivity TUFLOWHPC 50m H [m] With a large mesh size, not a great comparison with theoretically expected outputs.
  • 13.
    Mesh Size Sensitivity TUFLOWHPC 50m25m H [m] With a large mesh size, not a great comparison with theoretically expected outputs.
  • 14.
    Mesh Size Sensitivity TUFLOWHPC 50m25m10m H [m] As the mesh size is refined, start to converge on theoretical outputs.
  • 15.
    Mesh Size Sensitivity TUFLOWHPC 50m25m10m5m H [m] As the mesh size is refined, start to converge on theoretical outputs.
  • 16.
    5m Mesh Size Sensitivity TUFLOWHPC SGS H [m] With SGS, good convergence with theoretical outputs….
  • 17.
    5m Mesh Size Sensitivity TUFLOWHPC SGS 10m H [m] With SGS, good convergence with theoretical outputs….
  • 18.
    5m Mesh Size Sensitivity TUFLOWHPC SGS 10m25m H [m] With SGS, good convergence with theoretical outputs….for range of mesh cell sizes.
  • 19.
    5m Mesh Size Sensitivity TUFLOWHPC SGS 10m25m50m H [m] With SGS, good convergence with theoretical outputs….for range of mesh cell sizes. Eliminates sensitivity to mesh size.
  • 20.
    Rectangular Channel Test DeepSided Channels Unaligned to Mesh Mesh not aligned with deep banks • Distorts streamlines • Artificial energy losses; steepens gradient Traditional Solutions • 1D channel with cross-section (time-consuming; full 2D solution compromised) • Flexible mesh (quadrilaterals aligned with banks) • Much finer gridded mesh (much longer run times) • or…
  • 21.
    Rectangular Channel Test DeepSided Channels Unaligned to Mesh In conclusion: SGS • Conforms with Manning’s equation at all orientations • Resolves limitation of using gridded meshes along deep sided channels • Removes requirement for flexible mesh, higher resolution or 1D representation. • Significantly reduces mesh sensitivity • More accurate flood models and outputs  
  • 22.
    SGS Benchmarking U-Bend FlumeTest – Experiment Set-up • Flume experiment conducted by De Vriend (1978) Flow H along outer bank centre line inner bank Flow • Flow rate = 0.189 m3/s • Downstream H = 0.18 m • Manning’s n estimated to be 0.0115 ~ 0.0125 R = 4.25m W = 1.7m
  • 23.
    U-Bend Flume Test TUFLOWHPC 2019 without SGS • TUFLOW HPC Defaults H [m] Use of fixed grid causes edge effects representing curved wet-dry boundary.
  • 24.
    U-Bend Flume Test TUFLOWFV (Flexible Mesh) • TUFLOW FV H [m] Flexible mesh can be aligned with wet-dry boundary…although can be time consuming to do so.
  • 25.
    U-Bend Flume Test TUFLOWHPC without SGS vs TUFLOW FV TUFLOW HPC TUFLOW FV H [m]
  • 26.
    U-Bend Flume Test TUFLOWHPC with SGS • TUFLOW HPC with SGS H [m] Partially wet cells/faces smooth out the flow field SGS approaches eliminates the edge effects at wet-dry boundary.
  • 27.
    U-Bend Flume Test TUFLOWHPC with SGS vs TUFLOW FV TUFLOW HPC with SGS TUFLOW FV H [m] As good as a well-designed flexible mesh model!!
  • 28.
    TUFLOW HPC 34cm U-Bend FlumeTest – Mesh Size Sensitivity TUFLOW HPC-Without SGS 17cm10cm5cm H [m] • Slightly higher than experimental data due to fixed grid “saw-tooth” effects.
  • 29.
    TUFLOW HPC 34cm U-Bend FlumeTest – Mesh Size Sensitivity TUFLOW HPC-Without SGS 17cm10cm H [m] • Slightly higher than experimental data due to fixed grid “saw-tooth” effects.
  • 30.
    TUFLOW HPC 34cm U-Bend FlumeTest – Mesh Size Sensitivity TUFLOW HPC-Without SGS 17cm H [m] • Slightly higher than experimental data due to fixed grid “saw-tooth” effects. • Edge effects worsen with increasing mesh size.
  • 31.
    TUFLOW HPC 34cm U-Bend FlumeTest – Mesh Size Sensitivity TUFLOW HPC-Without SGS H [m] • Slightly higher than experimental data due to fixed grid “saw-tooth” effects. • Edge effects worsen with increasing mesh size.
  • 32.
    TUFLOW HPC withSGS U-Bend Flume Test – Mesh Size Sensitivity TUFLOW HPC-With SGS H [m]34cm17cm10cm Good agreement with experimental data… 5cm H [m]
  • 33.
    TUFLOW HPC withSGS U-Bend Flume Test – Mesh Size Sensitivity TUFLOW HPC-With SGS H [m]34cm17cm10cm Good agreement with experimental data… H [m]
  • 34.
    TUFLOW HPC withSGS U-Bend Flume Test – Mesh Size Sensitivity TUFLOW HPC-With SGS H [m]34cm17cm H [m] Good agreement with experimental data across range of mesh sizes.
  • 35.
    TUFLOW HPC withSGS U-Bend Flume Test – Mesh Size Sensitivity TUFLOW HPC-With SGS H [m]34cm Good agreement with experimental data across range of mesh sizes. H [m]
  • 36.
    U-Bend Flume Test Conclusions H[m] In conclusion: Sub-Grid Sampling... • Resolves limitation of using gridded meshes along curved channels • Removes requirement for flexible mesh or higher resolution. • Significantly reduces mesh sensitivity • More accurate flood models and outputs
  • 37.
    Real World Applications SouthJohnstone River Catchment South Johnstone River Case Study • Whole catchment model • Direct rainfall • Historic major event calibration 20km
  • 38.
    Real World Applications SouthJohnstone River Catchment Flow Calibration SGS provides more comparable results across range of mesh sizes compared to conventional approach
  • 39.
    Real World Applications PlynlimonCatchment Plynlimon Catchment • Research catchment in mid Wales managed by Centre of Hydrology, Bangor • Heavily instrumented with long rainfall and flow records • Soils/land cover well mapped • High resolution LiDAR data for majority of catchment Plynlimon TUFLOW Model • Direct rainfall • Moderate event
  • 40.
    Real World Applications PlynlimonCatchment Application Plynlimon Catchment Model • Direct rainfall • Moderate event • Compare sensitivity of mesh size to observed flow data (red line) With conventional approach: • Excellent cell size convergence • Spread with mesh resolution TUFLOW conventional approach shows excellent cell size convergence but show mesh dependency
  • 41.
    Real World Applications PlynlimonCatchment Application Plynlimon Catchment Model • Direct rainfall • Moderate event • Compare sensitivity of grid size to observed flow data (red line) With conventional approach: • Excellent cell size convergence • Spread with mesh resolution With SGS approach: • Limited variation with mesh resolution SGS provides near identical results across range of mesh sizes
  • 42.
    Conclusions • Sub-Grid Samplinghas potential to eliminate sensitivity to both:- • Mesh Orientation • Mesh Size • Benefits seen at range of scales and applications:- • Whole of catchment modelling • Urban environments • Easier Calibration and uncertainty analysis • Long term continuous simulation can be run rapidly:- • Natural Flood Management • Sustainable Urban Drainage Systems • Improved Forecasting