ROMS gang,
From a quick Google Search it seems there are at least 3 sites from which one can download a draft PDF version of a "ROMS" paper, written by Shchepetkin and McWilliams:
[PDF] The Regional Oceanic Modeling System: A Split-Explicit, Free ...
File Format: PDF/Adobe Acrobat - View as HTML
Page 1. The Regional Oceanic Modeling System: A Split-Explicit,
Free-Surface, Topography-Following-Coordinate
Oceanic Model Alexander F. Shchepetkin and James C ...
http:/www.atmos.ucla.edu/~alex/ROMS/roms.pdf - Similar pages
[PDF] The Regional Ocean Modeling System: A Split-Explicit, Free-Surface
File Format: PDF/Adobe Acrobat - View as HTML
Page 1. The Regional Ocean Modeling System: A Split-Explicit,
Free-Surface, Topography-Following-Coordinate
Ocean Model Alexander F. Shchepetkin and James C ...
http://www.myroms.org/Papers/roms.pdf - Similar pages
[HTML] Ocean-Modeling
... 2003. The Regional Ocean Modeling System: A split-explicit, free-surface, topography-following coordinates ocean model, Draft PDF; ...
http://www.ocean-modeling.org/index2.ph ... model=ROMS - 23k -
Cached - Similar pages
The first site (UCLA) has a March 2003 version, while the other two sites (Rutgers & Ocean Modeling) have identical February 2003 versions.
I think it would be useful for the ROMS community to know:
1. What are the differences between the ROMS described in this paper and the ROMS 2.0 available from Rutgers?
2. What is the status of the paper now? Is it submitted, accepted or in press somewhere? If so, is there a more recent version we should be looking at?
Thanks,
Rich
The "ROMS" paper?
Dear All,
The main difference is that Rutgers ROMS (all versions 1.x...2.0/TOMS) use Adams-Bashforth 3 time step for 3D momentum equation (inherited from SCRUM 3.0 code) and LF-TR predictor-corrector (PC) algorithm for tracers. The barotropic mode uses LF-AM3 PC algorithm with some forward-backward feedback, tuned for best possible accuracy (as opposite to stability). All Rutgers ROMS codes store R.H.S. terms for momentum equations in shared arrays to make them available from one time step to the other.
Time stepping algorithms described in the paper are more or less consistent with UCLA codes, which are characterized by alternative form of LF-AM3 PC algorithm for both momentum and tracer equations. Alternative form means that there is no storing of R.H.S. terms: variables are first advanced to n+1/2 step (predictor) and then these values are used to compute R.H.S. again to advance to n+1, hence no interpolation of precomputed R.H.S. is needed. See schematic on Fig. 6 and associated text.
Because of the predictor-corrector scheme for 3D momentum, two versions of coupling with barotropic mode are possible, and in fact historically available: coupling can be done during either predictor or corrector stages of 3D momentum step. These are described and compared in Section 5 of the March 2003 version of the paper (see Fig. 15).
The final version of the code employs corrector-coupled algorithm.
Barotropic Mode: The final version of the code employs Generalized Forward-backward time step for the barotropic mode. A version with alternative form LF-AM3 (with tunable coefficients, but actually it is LF-TR with FB feedback for most 3D applications) is also available for reference, but is considered obsolete. Still, in comparison with Ruthers version, the LF-TR version of UCLA ROMS does not store R.H.S. terms in shared arrays between time steps.
Generally Rutgers ROMS/TOMS family codes share a lot of features with the code described in the paper, but main time stepping engine is not quite there. It actually requires a lot of structural changes in the code to bring it in. It is my understanding that about two or so years ago a decision was made to "freeze" the development of forward code in order to proceed with adjoint. This is explainable (otherwise it would be a logistical nightmare), but to my view it is disappointing because most people in the community use Rutgers code(s) and do not care much about adjoint.
Time stepping of UCLA and Rutgers codes were identical to untill early 1999, when UCLA swithced to the predictor-corrector codes for 3D momentum (about 60% gain in stability limit at the expense of 20% increase in computational cost --- rhs3D and prsgrd repeated twice). The Generalised FB step for barotropic mode is exactly twice as efficient relatively to the original LF-AM3. The Rutgers version will incorporate these changes in the near future.
The paper is being reviewed for Ocean Modelling. It passed first review round (with positive outcome, also turned out to be a lot of typos), now is been finalized (hopefully the very last time) and prepared to be sent back to the journal within a week.
The final form is expected to be somewhat shorter. For example, description of predictor-coupled algorithm has been excluded along with the left part of Fig. 15. This is done mainly to satisfy the editor's demand to keep the size of the paper in check.
Since ROMS doesn't have a dedicated user manual yet, we kept several versions of the manuscript in the web site because it has very useful information to the user. It is likely that this type of information will be included in the ongoing web documentation of the model. The difference between February and March versions of the manuscript is very small.
Alexander Shchepetkin
The main difference is that Rutgers ROMS (all versions 1.x...2.0/TOMS) use Adams-Bashforth 3 time step for 3D momentum equation (inherited from SCRUM 3.0 code) and LF-TR predictor-corrector (PC) algorithm for tracers. The barotropic mode uses LF-AM3 PC algorithm with some forward-backward feedback, tuned for best possible accuracy (as opposite to stability). All Rutgers ROMS codes store R.H.S. terms for momentum equations in shared arrays to make them available from one time step to the other.
Time stepping algorithms described in the paper are more or less consistent with UCLA codes, which are characterized by alternative form of LF-AM3 PC algorithm for both momentum and tracer equations. Alternative form means that there is no storing of R.H.S. terms: variables are first advanced to n+1/2 step (predictor) and then these values are used to compute R.H.S. again to advance to n+1, hence no interpolation of precomputed R.H.S. is needed. See schematic on Fig. 6 and associated text.
Because of the predictor-corrector scheme for 3D momentum, two versions of coupling with barotropic mode are possible, and in fact historically available: coupling can be done during either predictor or corrector stages of 3D momentum step. These are described and compared in Section 5 of the March 2003 version of the paper (see Fig. 15).
The final version of the code employs corrector-coupled algorithm.
Barotropic Mode: The final version of the code employs Generalized Forward-backward time step for the barotropic mode. A version with alternative form LF-AM3 (with tunable coefficients, but actually it is LF-TR with FB feedback for most 3D applications) is also available for reference, but is considered obsolete. Still, in comparison with Ruthers version, the LF-TR version of UCLA ROMS does not store R.H.S. terms in shared arrays between time steps.
Generally Rutgers ROMS/TOMS family codes share a lot of features with the code described in the paper, but main time stepping engine is not quite there. It actually requires a lot of structural changes in the code to bring it in. It is my understanding that about two or so years ago a decision was made to "freeze" the development of forward code in order to proceed with adjoint. This is explainable (otherwise it would be a logistical nightmare), but to my view it is disappointing because most people in the community use Rutgers code(s) and do not care much about adjoint.
Time stepping of UCLA and Rutgers codes were identical to untill early 1999, when UCLA swithced to the predictor-corrector codes for 3D momentum (about 60% gain in stability limit at the expense of 20% increase in computational cost --- rhs3D and prsgrd repeated twice). The Generalised FB step for barotropic mode is exactly twice as efficient relatively to the original LF-AM3. The Rutgers version will incorporate these changes in the near future.
The paper is being reviewed for Ocean Modelling. It passed first review round (with positive outcome, also turned out to be a lot of typos), now is been finalized (hopefully the very last time) and prepared to be sent back to the journal within a week.
The final form is expected to be somewhat shorter. For example, description of predictor-coupled algorithm has been excluded along with the left part of Fig. 15. This is done mainly to satisfy the editor's demand to keep the size of the paper in check.
Since ROMS doesn't have a dedicated user manual yet, we kept several versions of the manuscript in the web site because it has very useful information to the user. It is likely that this type of information will be included in the ongoing web documentation of the model. The difference between February and March versions of the manuscript is very small.
Alexander Shchepetkin