-
Notifications
You must be signed in to change notification settings - Fork 3
F. Replay Strategies within ADAS
The idea of replaying the ADAS to a given analysis is applicable to multiple circumstances. The primary situation is one of trying to obtain extra output from a cycling ADAS that has perhaps been missed when the experiment first ran. Another type of replay arises when changes being implemented in the system are not expected to affect the ensemble too drastically so that in a hybrid methodology would simply replay to ensemble to that of an existing experiment; also useful to get a first glimpse of what changes might bring to the system without having to incur the cost of running the ensemble. Other cases may represent a desire to study the consequences of changes in the aerosol analysis without bothering with direct impact to the meteorological analyses; or vice versa, when the user simply wants to replay to existing aerosol analyses without bothering about the effects of changes to aerosols due to changes in the meteorology, and finally further motivations that may arise from the creativity of every user.
In recent years, the ADAS replay strategy that has been mostly exercised and found to be important in getting preliminary results from first-order testing is that of ensemble-replay. Clearly, this strategy applies to systems that run some form of hybrid ensemble-variational methodology (either 3D or 4D).
Replaying a new experiment to the existing ensemble of another is controlled by a file named atmens_replay.acq. This is a single-line file providing the templated name of the location of a existing set of ensemble background fields. Say, the user has an experiment named MYEXP and wants to replay the ensemble to that of GEOS FP f5271_fp, the acquire file atmens_replay.acq should look like
/home/dao_ops/f527_fp/run/.../archive/atmens/Y%y4/M%m2/f5271_fp.atmens_ebkg.%y4%m2%d2_%h2z.tar => MYEXP.atmens_ebkg.%y4%m2%d2_%h2z.tar
and it should be placed under FVHOME/run. Notice the redirect notation (=>) in line above responsible for the automatic rename of the external file to a name consistent with the user experiment name.
It is opportune to point out that this ensemble-replay strategy is employed in the calculation of FSOI. Since GEOS ADAS FSOI strategy involves re-running the analysis solver, the ensemble of background fields need to be made available for the analysis sensitivity run. By default, in this case, the replay is a self replay, that is, one that retrieves the ensemble from from the cycling ADAS. In the case, of setting the forward ADAS to replay to an external ensemble, it is necessary to edit the file FVHOME/asens/atmens_asens.acq and get it to be consistent with setting in FVHOME/run/atmens_replay.acq (e.g., as above). Similarly, if Standalone Analyses will be exercised within the scope of the user's experiment with the ensemble being replayed to an external analyses, the file FVHOME/anasa/atmens_replay.acq needs to equal that under FVHOME/run.
One more point on replaying the ensemble of backgrounds that might be useful to users to know is that the resolution on the ensemble is not an issue. Differences in experiment resolution and that of the replayed ensemble are automatically addressed on the fly.
Replaying ADAS to given analysis can be accomplished by renaming an acquire file found under FVHOME/run/, that is,
mv noreplay.acq atmens_replay.acq
Typically, this file would need to be edited and adapted to the needs of the user; its defaults are to replay to MERRA-2 analyses which would require some extra work by the user.