Adjoint and Data Assimilation Working Group
- 1 Historical Development
- 2 Current Forward Model Code
- 3 Current Adjoint Code
- 4 Documentation
- 5 Distribution and Use
- 6 Giving credit for using the adjoint model
- 7 Get Involved
- 8 Publications
Original work on the adjoint of GEOS-Chem v6 began in 2003, focusing on the adjoint of the offline aerosol simulation. By 2005, the adjoint was expanded to include a tagged CO simulation and a full chemistry simulation; an adjoint of GEOS-Chem v7 was also developed in the following years. Each of these branches of the adjoint code were been constructed in a hybrid fashion using a combination of automatic differentiation software (TAMC, KPP) and manual coding of both discrete and continuous adjoints. They shared many common elements yet had unique features for different applications.
During the summer of 2009, the existing branches were merged and updated to bring the adjoint into alignment with the latest release of GEOS-Chem, v8-02-01. This merged adjoint model is now the standard adjoint code into which all further development efforts will be placed.
Current Forward Model Code
The forward model on which the adjoint is based corresponds to GEOS-Chem v8-02-01 with the following updates:
- KPP solver for gas-phase chemistry (as in GCv8-02-03)
- Implement Bond 2007 BC/OC emissions (as in GCv8-02-02)
- Apply bug fixes from GCv8-02-02 listed here
- Apply bug fixes from GCv8-02-03 listed here
- Apply bug fixes from GCv8-02-04 listed here
Current Adjoint Code
- Meteorological fields
- GEOS-3 needs testing
- model resolution
- 4 x 5
- 2 x 2.5
- Forward model processes
- PBL mixing
- dry deposition
- wet deposition needs GEOS-4 testing
- strat / trop exchange with LINOZ implemented, but will be updated once v8-02-04 is released
- NOy up fluxes
- inorganic aerosol thermodynamics
- sulfate chemistry
- SOA, Dust, sea salt needs doing
- aerosol surface area feedbacks needs updating
- aerosol optical feedbacks needs doing
- all standard emissions included
- Simulation modes
- full chemistry
- tagged CO
- tagged Ox
- Observational Operators
- MOPITT CO column
- SCIAMACHY CO column
- AIRS CO column
- IMPROVE PM2.5 (NO3, SO4, OC, BC) needs updating
- CASTNet (NH4+) needs updating
- GOME / SCIAMACHY NO2 column needs updating
- using KNMI retrieval (Henze)
- using Dalhousie retrieval (Shim)
- TES NH3
- TES O3 needs updating
- Control parameters
- Initial Conditions scaling factors (linear or log)
- Emissions scaling factors (linear or log)
- NH3, primary BC/OC, SO2: anthropogenic, natural, bioburn, biomass, ship
- NOx: soil, lightning, aircraft, anthropogenic, biofuel, bioburn
- all other gas-phase tracers: anthropogenic, biofuel, bioburn
- Adjoint sensitivities
- all implemented control parameters
- w.r.t Reaction Rate Parameters
- w.r.t all emissions
- to AQ attainment metrics needs updating
- to spatiotemporally averaged species concentrations (e.g., arctic O3) needs updating
- Inverse Hessian approximation needs updating
- off-diagonal covariance matrices needs updating
- 3D-Var needs updating
Features may be qualified as:
- needs testing: an implemented feature that we haven't fully used yet
- needs updating: a feature developed with a previous branch that has yet to be updated to GEOS-Chem v8 and the merged adjoint
- needs doing: a feature nobody has tackled the adjoint of yet
Primary code developers
Monika Kopacz, Kumaresh Singh, Changsub Shim, Daven Henze
Adjoint model lead scientist
A user's guide is available. http://adjoint.colorado.edu/%7Edaven/gcadj_std/GC_adj_man.pdf
Meemong Lee has created a detailed flowchart of the inverse model code structure. http://adjoint.colorado.edu/~daven/gcadj_std/flowchart.pdf
Distribution and Use
Code for the adjoint is distributed through a CVS server located at adjoint.colorado.edu. Contact Daven Henze to obtain an account on the server.
Even if your office mate has a copy of the code, the best way to obtain the model is to get a CVS account for yourself and download a version from the repository. So please do not copy code directly from others or pass the code along to third parties. This vastly helps with tracking developments and keeping up with model updates.
Use of the adjoint model code follows standard practice for GEOS-Chem. It is expected that any developments that come of individual applications based on this community model will eventually be given back to the community by incorporation of new developments into the standard adjoint code. New development should be submitted to Daven Henze for inclusion in the standard adjoint model code.
Giving credit for using the adjoint model
We aim to make distribution of adjoint model code as immediate as possible. A consequence is that many features may not yet be publicly documented. Therefore, giving code developers due credit is of utmost importance. At this point, any use of the GEOS-Chem model adjoint should include as co-authors the primary developers: Daven Henze, Monika Kopacz, Kumaresh Singh and Changsub Shim. In trade, we provide user support. As the model matures and user support become less of an issue, these requirements will be relaxed.
Authors of new additions to the standard code will be given due credit on the first round of publications to come of their development. Hence, additional co-authors may also be required if the application uses features that are newly developed by additional developers.
Lastly, it is requested that users of the GEOS-Chem adjoint consider offering co-author ship to several additional scientists who have contributed to the adjoint code support and progress, such as Kevin Bowman, Adrian Sandu and Daniel Jacob.
Overall, if you have any questions about authorship, even for a conference presentation, please contact Daven Henze.
Support and development of the GEOS-Chem adjoint has clearly become a broad undertaking. In attempt to focus such developments, we've created a GC-adjoint mailing list. If you wish to become involved, you should add yourself to this list by sending an email to email@example.com with the word subscribe in the subject or body, or you may click here for browser sign-up: https://lists.seas.harvard.edu/mailman/listinfo/geos-chem-adjoint).
Kevin Bowman organizes bimonthly conference calls concerning current adjoint model activities.
- Kopacz, M., D.J. Jacob, J.A. Fisher, J. A. Logan, L. Zhang, I. A. Megretskaia, R. M. Yantosca, K. Singh, D. K. Henze, J. P. Burrows, M. Buchwitz, I. Khlystova, W. W. McMillan, J. C. Gille, D. P. Edwards, A. Eldering, V. Thouret, and P. Nedelec (2010): Global estimates of CO sources with high resolution by adjoint inversion of multiple satellite datasets (MOPITT, AIRS, SCIAMACHY, TES), Atmos. Chem. Phys., 10, 855-876. http://www.atmos-chem-phys.net/10/855/2010/acp-10-855-2010.pdf
- Kopacz, M., D. J. Jacob, D. K. Henze, C. L. Heald, D. G. Streets, and Q. Zhang (2009), A comparison of analytical and adjoint Bayesian inversion methods for constraining Asian sources of CO using satellite (MOPITT) measurements of CO columns, J. Geophys. Res., doi:0.1029/2007JD009264. http://acmg.seas.harvard.edu/publications/KopaczJGR2009_2007JD009264.pdf
- Henze, D. K., J. H. Seinfeld and D. T. Shindell (2009), Inverse modeling and mapping U.S. air quality influences of inorganic PM2.5 precursor emissions with the adjoint of GEOS-Chem, Atmos. Chem. Phys., 9, 5877-5903.
- Zhang, L., D. J. Jacob, M. Kopacz, D. K. Henze, K. Singh, and D. A. Jaffe (2009), Intercontinental source attribution of ozone pollution at western U.S. sites using an adjoint method, Geophys. Res. Lett., 36, L11810, doi:10.1029/2009GL037950
- Henze, D. K., A. Hakami and J. H. Seinfeld (2007), Development of the adjoint of GEOS-Chem, Atmos. Chem. Phys., 7, 2413-2433.
- Adjoint inversion of CO sources using combined MOPITT, SCIAMACHY and AIRS CO columns, presented by Monika Kopacz at the COSPAR Scientific Assembly, Montreal, July 18, 2008. http://acmg.seas.harvard.edu/presentations/powerpoints/mak2008/COSPAR_MKopacz_July2008.ppt
- Singh, K., P. Eller, A. Sandu, D. K. Henze, K. Bowman, M. Kopacz, and M. Lee (2009), Towards the construction of a standard geos-chem adjoint model, ACM High Performance Computing Conference.