CMAQ Adjoint Meeting Notes

Jan. 11, 2011

(notes from Shannon Capps)

1- Two subgroups were formed after the last talk and each held a call.

a) Cloud and aq chem (Jaemeen):

- With desire for functioning model by February, it was decided that Shunliu would attempt first; Shunliu has worked on it some; adjoint is developed but not tested

- Jaemeen is testing the forward code now; she's been able to run everything through KPP (including equilibrium stuff.

- Deposition velocities: Do they (for a couple of species) depend on concentrations? Will including bi-directional flux require further development? Considered a minor issue for now, maybe later.

- Peter: planning for the future to checkpoint the deposition velocity rather than CGRID to save the velocity itself for the backward solution

b) Observational operator (Havala):

- OMI NO2 column within CMAQ almost working; need to run in parallel (observational operator) and the approach that is taken right now is OpenMP parallelization within MPI

- Useful currently for comparing CMAQ to OMI

- Will consult Randall or Lok about their method for removing dependence from TM4 priors by using native CTM (CMAQ's) profile: does avg kernel get used after the suggested adjustment? and for data assimilation would this adjustment really matter?

- Havala will forward document describing code to entire adjoint development group

- Current code reads raw HDF file; would it make more sense to process the data first and then read that into the operator, which would be more amenable to parallel processing?

- Resolving interactions of hybrid MPI and OpenMP not too far

- Tianfeng: preprocessing would prevent having to reprocess; information for preprocessing would be simple to separate and use

- Goal (Tianfeng/Amir): Have future product that uses preprocessing code that can separate ingested data from observation

- Currently, data assimilation or emissions inversion is not yet implemented; are there limitations in the code? What would change when we take this obs op into the data assimilation framework? Need to plan ahead.

- OMI NO2 columns are just one example of observations and this code is one direction to try. Operators would be necessary for other types of operations

2- other issues:

- Porting gradients from v.4.5.1 to the current (Tianfneg [emissions] and Peter [BCs]) — starting to be taken on in the near future

- Running the code: Tianfeng - downloaded but has not built. Peter - downloaded but not run yet. Peter has noticed some options missing (ACM2 inline, discrete eddy)

- Still some issues in builder, particularly with dependencies arising from KPP. Should look to see how the build process can resolve (or automate) these dependencies (Matt?)

- Notable areas that need more work in the current version: Gas phase chemistry (Amir) and Aerosol checkpointing (Jaroslav)

- Issue of priority and whether it is to have gas-phase running or completing aerosols.

Next Conference Call: February 8, 2011 at 2 pm EST

Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License