TM5 4DVAR meeting 4-9-2014
Dec 29, 2015
TM5 4DVAR meeting4-9-2014
Agenda
• TM5 4DVAR Main branch
• Priorities LOT3
• TM5 mp
• New feature wish list
• Meeting format etc.
Main branch
• Where? [SF ..?]
• Who maintains it?
• What should it contain?
• How to organize updates?
4D-var Tree
default sourishmultitracer
sanderCH4 project
“cy1”
JRC version
SourceForge versions
• current versions considered in “merge” project
defaultmulti tracer
sourishmulti tracer
sanderCH4 project
“cy1”
JRC version
SourceForge versions
defaultmulti tracerCH4 project
defaultmulti tracerCH4 projectJRC features
Merging procedure
• Compare codes, identify differences• Difference found ?
– Add to discussion page on SourceForge Wiki (t.b.d.)– If difference does not alter result:
• Make one version the new standard
– If difference does alter result:• Initially keep both alive, add temporary flag to rcfile• After discussion decide if:
– both stay supported, with new flag in rcfile– one version becomes new standard, other is removed
Benchmark tests
• Two benchmark test:
PyShell test JRC testProvided by: Sander H. Peter B.Tracer(s) CH4 CH4Optim Emissions
Bias parameters ?EmissionsInitial concentrations
Observations PointGOSAT “PROXY”
Point
Time period 2 months 14 months
• While testing, the merge versions should reproduce both benchmarks exactly.
F.U.Q. - Frequently Unanswered Questions• Why is the PyShell code inside TM5 ?
base/branches/var4d-mt/src/ bin/ rc/ optimizer/ py/ f2py/scripts/rc/
TM5/base/branches/whatever/src/ bin/ rc/PyShell/base/branches/var4d-mt/optimizer/ py/ f2py/ scripts/ rc/
Alternative:
+ better separation of model versions+ split in 4D-var settings and model related settings
F.U.Q. - Frequently Unanswered Questions (2)• Is TIPP still used ?
– Put it on SourceForge in a ‘tools’ directory ?(also put ‘vpp’ in there?)
• Are emission time profiles still used? (for example for fire emissions at daily resolution)Now categories could have different time resolutions already.
Wish list (long term)• bias parameter optimization
– satellite data:-> JRC approach will be done within framework project-> SRON approach part of “sander” branch?
– point observations (N2O) : t.b.d.
• Rodenbeck scheme (for coupling regional models)• Monte Carlo approach for uncertainty estimates• log-normal p.d.f.’s for uncertainty in parameters• …
TM5mp & 4DVAR
• Motivation: Modern computers & 4DVAR
• Allow higher resolution simulations, more years, more tracers, etc.
• Draw back: No zoom
• But PyShell facilitates switching between model versions
Next …
• Do we agree that TM5mp 4DVAR has high priority?
• If so, how to organize this.
New feature wish list 1
• GM: Simplify RunTM5.py• GM: Get rid of the shared pool of rc-keys.• GM: ‘simple’ rather than ‘fancy’ python
coding.• GM: Simplify the code tree (e.g. get rid of
branches in the main branch).• GM: Document Python classes and methods.• GM: Get rid of the logging module.
• SB: Read one day of meteo at once• SB: Fix phlb_t in the adjoint• SB: Implement Chevalliers 4DVAR time
splitting• SB: Let VPP calculate the AK• SB: Fix problem with adjoint and zoom• SB: 3 hourly meteo in 4DVAR
New feature wish list 2
• Look for volunteers.
• Pay Arjo?
• We could use these meetings to evaluate progress and prioritize tasks.
How to deal with the ‘new feature wish lists’
• Frequency
• Size
• Telecon tool
• Archiving / Minutes
• Else?
About these meetings