update on hydrodynamic model comparisons
DESCRIPTION
Update on hydrodynamic model comparisons. Marjy Friedrichs and Carl Friedrichs Aaron Bever (post-doc) Leslie Bland (summer undergraduate student). Methods: Target diagrams ( Jolliff et al., 2009). Total RMSD 2 = Bias 2 + unbiased RMSD 2. mean. s easonal variability. Bias. - PowerPoint PPT PresentationTRANSCRIPT
Update on hydrodynamic model comparisons
Marjy Friedrichs and Carl FriedrichsAaron Bever (post-doc)Leslie Bland (summer undergraduate student)
Methods: Target diagrams (Jolliff et al., 2009)
1
1
-1
-1 Unbiased RMSD
Bias
Total RMSD2 = Bias2 + unbiased RMSD2
mean seasonalvariability
x > 0overestimates
variability
y > 0: overestimates
mean
Methods: Target diagrams (Jolliff et al., 2009)
1
1
-1
-1 Unbiased RMSD
Bias
Total RMSD2 = Bias2 + unbiased RMSD2
mean variability
x > 0overestimates
variability
y > 0: overestimates
mean
Methods: Target diagrams (Jolliff et al., 2009)
1
1
-1
-1 Unbiased RMSD/stdev(obs)
Bias/stdev(obs)
Normalization by standard deviation of observations
outer circle: Model-data misfit = variability in data
x > 0overestimates
variability
y > 0: overestimates
mean
Methods: Target diagrams (Jolliff et al., 2009)
1
1
-1
-1 Unbiased RMSD/stdev(obs)
Bias/stdev(obs)
Normalization by standard deviation of observations
outer circle: Model-data misfit = variability in data
x > 0overestimates
variability
y > 0: overestimates
mean
Methods: Target diagrams (Jolliff et al., 2009)
1
1
-1
-1 Unbiased RMSD/stdev(obs)
Bias/stdev(obs)
Normalization by standard deviation of observations
outer circle: Model-data misfit = variability in data
x > 0overestimates
variability
y > 0: overestimates
mean
model does worse than the mean of the data
Model simulationsOriginal simulations (summer
‘10)◦CH3D (P. Wang)◦EFDC (J. Shen)◦ChesROMS (W. Long)◦CBOFS2 (L. Lanerolle)
New ‘consistent forcing’ simulations (this week!)◦EFDC (J. Shen)◦CBOFS2 (L. Lanerolle)◦UMCES ROMS (Y. Li)
Preliminary model comparisons (Summer ‘10)Initially examined salinity at the
halocline (max dS/dz) as a function of bathymetric error, latitude, salinity, oxygen, bottom depth
Conclusion: For all four models, model skill (total RMSD) is primarily a function of mean salinity and/or latitude
New model comparisons (Fall ‘10)Best match over ±12 hour time
windowAdditional variables:
dS/dz at max dS/dz z of max dS/dz S at max dS/dz
New ‘consistent forcing’ simulations
New model comparisons (Fall ‘10)Best match over ±12 hour time
windowAdditional variables:
dS/dz at max dS/dz z of max dS/dz S at max dS/dz
New ‘consistent forcing’ simulations
= inst match = best
match(over 24h)
SurfaceSalinity
New model comparisons (Fall ‘10)Best match over ±12 hour time
windowAdditional variables:
dS/dz at max dS/dz z of max dS/dz S at max dS/dz
New ‘consistent forcing’ simulations
Salinity (psu)
Stratification=
max dS/dz
depth of max dS/dz
Salinity (psu)
Salinity at max dS/dz
Salinity (psu)
New model comparisons (Fall ‘10)Best match over ±12 hour time
windowAdditional variables:
dS/dz at max dS/dz z of max dS/dz S at max dS/dz
New ‘consistent forcing’ simulations
Surface Salinity
EFDC CBOFS2
New forcing: Slight improvement in CBOFS2 resultsSlight degradation in EFDC results
Red = First runsBlack = New or consistent forcing runs.The EFDC results here are without showing the far outliers.
= old results
= new results
max dS/dz
old forcing
new forcing
Salinity (psu)
max dS/dz
newforcing
Salinity (psu)
Salinity (psu)
depth ofmax dS/dz
old forcing
new forcing
Salinity (psu)Salinity (psu)
depth of max dS/dz
newforcing
Salinity (psu)
Salinity (psu)
Salinity atmax dS/dz
old forcing
new forcing
Salinity (psu)
Salinity (psu)
Salinity atmax dS/dz
newforcing
Salinity (psu)
Salinity (psu)
Next steps for hydrodynamic comparisons? Why does CH3D produce superior stratification? ◦Vertical grid structure? C&D canal?
Bathymetry?◦other?
Next model runs◦Atmospheric forcing? ◦Boundary conditions on shelf?
Additional metrics