-
Notifications
You must be signed in to change notification settings - Fork 3
I. MERRA‐21C
This provide a series of brief tips related to M21C.
Cloning and building the GEOS ADAS software for M21C is no different than cloning and build of any version of GEOSadas; please refer to other items in the GEOSadas Wiki for information on that.
General users trying to start an experiment that will be performed with M21C serving as the baseline - perhaps to implement a necessary modification to M21C (and a corresponding patch); or to study a science variation of what's implemented in M21C - should be interested in starting by simply cloning M21C source code (tag) and reproducing its results for a given period. This GEOSadas Wiki has a section on "How to reproduce GEOSadas results" but for ease of reference this is laid out again in what follows.
For this use the flag -mil when using parallel_build.csh.
To setup an experiment users should simply edit an m21c input file found under: src/Applications/GEOSdas_App/testinputs. For example, simply copy the file m21c_j98.input into a directory of your own; as such this input file starts an experiment associated with the first stream of M21C - on 19971230 21Z. User should be concerned about editing things like:
- Internet plots locations.
- FVHOME - the location of the experiment driving scripts and resources.
- FVICS - the tar ball containing the initial conditions for the experiment (in this case, a file on 19971230 21Z)
- End year-month-day
- Select group - this is the group ID used for SLURM jobs
- The user can also change the experiment name from m21c_j98 to whatever reasonable name (no more that 12 or so characters); in this case, renamed the input file accordingly.
and that's pretty much it. Now don't worry about changing the defaults settings appearing within the square brackets. For example, when changing the group ID the experiment will run under, the lines in the input file above,
Select group: [g2538]
> g0613
need simply change to
Select group: [g2538]
> s0123
where "s0123' is an arbitrary ID used for illustration here.
Once the input file has been edited with entries that make sense to a particular user. Then, simply use the script called runjob found in the bin directory of your build to process the input file. It is advisable, though typically not needed, to set the proper module environment before executing runjob. In summary, two steps to setup the experiment:
- source YOURPATH_TO_GEOSADAS_BIN/g5_modules
- YOURPATH_TO_GEOSADAS_BIN/runjob -l m21c_j98,input
The user will be prompted with about three questions - whose answers should all be y (for yes), and the experiment will be set - be patient, setting up the ensemble requires unfolding a couple of large tar-balls.
To the final question - submit to SLURM - stick to the (default) no. The reason for this is that you might want to edit the two main job scripts for the DAS: m21cA.j and atm_ens.j to fill SLURM specific settings in case you need to, before submitting the job - refer to other parts of the GEOSadas Wiki for more on this; typical edits are addition of specific QOS and specification of TSE (solid disc) location (i.e., redefining FVWORK and ENSWORK in two mains scripts, respectively).
Assuming consistent in the computing environment between the actual M21C experiment and the user's access, pointing to M21C restarts should be enough to ensure reproducibility the first cycle of the DAS (M21C).
In case user wants to reproduce more than simply the first 6-hour cycle, the following additional step is necessary:
- Create a file named given_rndperts.acq and placed inside FVHOME/run/atmens
- This file should have a pointer to the dates of random NMC perturbations used by the ADAS ensemble. For example, in the case of the default setting here - where an experiment is being conducted over the first stream of M21C, this file should have the following content:
/gpfsm/dnb08/projects/p253/archive/e5303_m21c_jan98/atmens/Y%y4/M%m2/e5303_m21c_jan98.rndperts.dates.%y4%m2%d2_%h2z.txt => EXPID.rndperts.dates.%y4%m2%d2_%h2z.txt
where EXPID is the user experiment name. This will tell the DAS scripts to fetch the random dates file used in M21C and use the same perturbation set for the ensemble at any given cycle as the set used in M21C.
By default, the DAS experiments are set to launch forecasts from the 00z analyses (21z restarts). To instruct the DAS to bypass forecasts, i.e., to not do forecasts, simply do:
- cd FVHOME/fcst
- mv g5fcst.j g5fcst.j_
For the actual M21C runs, it is recommended that user set the experiment up under a TSE_staging. In development-like exercises, when the deterministic ADAS and the ensemble run serially (not concurrently), the determinist ADAS can be placed in simply the volatile disk TSE, but the ensemble must be placed in a TSE_staging area. It is recommended that users read the main DAS script and the ensemble main driving script for comments around settings of FVWORK and ENSWORK.
Most exercise on top of M21C will can likely be done replaying to the ensemble of backgrounds available in the original M21C experiment. Users should notice the existence of an input file (to be used as input to runjob) that set up an replay experiments. As with any input file for fvsetup users will need to do minor adaptation to suite their needs (userID, FVHOME, etc - see list of items above). This file is in src/Applications/GEOSadas_App/testsuites/m21c_j98RPY.input.