access_T_AMIP_ACCESS_1.3 - ACCESS-NRI/accessdev-Trac-archive GitHub Wiki
ACCESS 1.3 N96L38(CABLE) Transpose-AMIP II procedure
SUMMARY: Basic steps to run accessdev ACCESS 1.3 T-AMIP
This section shows the minimum number of steps needed to do a single ACCESS 1.3 T-AMIP run. These steps are explained in more detail in the following sections.
- setup initial files on raijin:
- login to raijin eg (note we are using p66/glr548 as the example project/user login)
ssh -X [email protected]
- create a directory where you will keep the initial condition files for your run
mkdir /short/p66/glr548/tamip_ic
- decide which T-AMIP date you want to use (here I select 0600 UTC on 2009-01-16) and in this dir create soft links to the corresponding initial dump file and the appropriate SST and SICE ancillary files
cd /short/p66/glr548/tamip_ic
ln -s /projects/access/AMEL/TransposeAMIPII/access1.3/sstice/iceN96_20090101nd59 tamipice
ln -s /projects/access/AMEL/TransposeAMIPII/access1.3/sstice/sstN96_20090101nd59 tamipsst
ln -s /projects/access/AMEL/TransposeAMIPII/access1.3/na1.3ic/na1.3tamip2009-01-16-h06 tamipic
Note: wed29oct14: T-AMIP data is in /g/data/p66/glr548/tamip_sam/access1.3 and is also now in /projects/access/AMEL/TransposeAMIPII/access1.3
- get a copy of the ACCESS 1.3 T-AMIP RUN job on accessdev:
- login to accessdev and open umuix
ssh -X [email protected]
umuix
-
in umuix cp glr548 ACCESS 1.3 T-AMIP RUN job vafrb to your own job eg vahdb
-
edit your new job (vahdb) to point to the directory of the initial condition files you created via links on rajin above
umuix -> vahdb -> Input -> Time_convention: change
TAMIP_DATA from /short/p66/glr548/tamip_ic
to your new dir (in my case I leave it as /short/p66/glr548/tamip_ic)
-
submit your new job:
umuix -> vahdb -> Check/Save/Process/Submit
Check == you will get 3 errors, but just ignore these - in fact you need not do this step
Save/Process == this saves and then processes the job and so creates the umui_jobs/vahdb directory
/home/548/glr548/umui_jobs/vahdb
and this holds the job parameter control files
CNTLALL FCM_BLD_COMMAND INITHIS SIZES UMSUBMIT
CNTLATM FCM_EXTR_SCR PPCNTL STASHC USR_FILE_OVRDS
CNTLGEN FCM_MAIN_SCR PRESM_A SUBMIT USR_MACH_OVRDS
CONTCNTL FCM_UMUI_BASE_CFG RECONA UAFILES_A
EXT_SCRIPT_LOG FCM_UMUI_MODEL_CFG SCRIPT UAFLDS_A
Submit == submits the job
-
check on run and its outputs on raijin:
- login to raijin
ssh -X [email protected]
- Check on the running job
qstat -u $USER
-
using 64 processors the single 5day forecast should take ~4min
Standard Error and Output will be in your home_dir/um_output eg in my case
cd /home/548/glr548/um_output
vahdb000.vahdb.d14301.t153149.leave
Data output files will be in
DATAM # DATAW /short/$PROJECT/$USER/$RUNID eg /short/dp9/glr548/vahdb in my case
cd /short/dp9/glr548/vahdb
vahdb.astart = start dump
vahdba.p* = forecast output files which show 3hrly outputs for the 5day forecast
-
you can view the pp output, plot them or convert them to netcdf using xconv eg
cd /short/dp9/glr548/vahdb
xconv vahdba.pjk91g5
We can create U10m and Q1.5m postscript plots after 3hrs into the run, and they are shown in the section "Example of ACCESS 1.3 T-AMIP outputs" below.
Note: As of wed29oct14 the T-AMIP data which is in /g/data/p66/glr548/tamip_sam/access1.3 is also now also in /projects/access/AMEL/TransposeAMIPII/access1.3 so all projects should be able to access it
In the sections below we explain the procedure in more detail but the above steps are all you need to start running a T-AMIP job.
Preliminary steps
The following steps, covering those listed in the T-AMIP experimental design, were taken to setup the ACCESS T-AMIP procedure. These do not need to be re-done by new users, but are here to explain how we setup the run.
- (1) Model state variables are to be initialised from ECMWF YOTC analyses available from http://data-portal.ecmwf.int/data/d/yotc. Guidance on carrying out the interpolation can be found in the ECMWF IFS documentation.
To do this we follow the procedure of Roff et al. (2012) and download four 3D fields (U, V, T and Q) and four surface fields (Surface pressure, Skin Temperature, Geopotential and Land-sea mask with the latter two being invariant) and combine them into one grib file. This is then supplied as the start “dump”, or initial conditions, for the UM reconfiguration run job (saacb on the Bureau’s ngamai computer) to produce a N96L38 um-formatted initial condition file. Note: this is not the final dump file we need for ACCESS 1.3 T-AMIP, as discussed below.
- (2) Atmospheric composition, solar forcing and land use should be as the final year of the CMIP5 AMIP experiment (2008).
The atmospheric mixing ratios of CO2, CH4 (Methane), N2O, CFC11, CFC12, CFC113, HCFC22, HFC125 and HFC134A for the T-AMIP period have been set to the same values as used in the ACCESS 1.3 AR5 RCP4.5 experiment. Similarly, the solar forcing and the volcanic forcing are also set to the values from the AR5 RCP4.5 run.
- (3) SSTs from ECMWF YOTC should be used in the hindcasts.
The 0.75o resolution SST and Seaice YOTC fields were downloaded for the relevant dates as grib files and were then processed via: converted to netcdf using cdo operators; split into separate SST and Seaice files using nco operators; the SST fields were then extrapolated over land using convsh scripts; converted to ancillary format using Xancil; then interpolated in space to N96L38 resolution using the UM Central Ancillary Program.
- (4) Land surface models should be initialised using one of the following methods:
- Initialise from fields produced by a land surface assimilation system (e.g. ECMWF or GLDAS)
- Initialise using a suitable climatology: e.g. from GSWP2 or derived from the model’s AMIP simulation.
- Initialised with a nudging method as described by Boyle et al. (2005).
We initialize the LSM soil moisture and temperature fields from the ACCESS 1.3 AMIP climatology. With CABLE the normal soil moisture, temperature and frozen fraction fields are just diagnostics set to the grid box average of the tiled CABLE fields. Thus for initializing the model we need the tiled fields but will also get the diagnostic fields for consistency. The 3 diagnostic fields, and corresponding UM STASH codes, are: SOIL MOISTURE CONTENT IN A LAYER, DEEP SOIL TEMP AFTER TIMESTEP, and FROZEN SOIL MOISTURE FRAC AFTER TS (STASH codes 9, 20, and 215). The 25 prognostic fields are: SOIL LAYER 1-6 TEMPERATURE ON TILES, SOIL MOISTURE LAYER 1-6 (ON TILES), FROZEN SOIL MOIST FRAC LYR 1-6 (TILES), SNOW TEMPERATURE LAYER 1-3 (ON TILES), SNOW AGE (ON TILES), and SNOW DEPTH LAYER 1-3 (ON TILES) (STASH codes 301:306, 307:312, 313:318, 323:325, 330, 332:334).
All these fields are collected into a netcdf monthly mean LSM file which is then used to initialize the T-AMIP runs. The remaining land surface soil and vegetation fields are either initialized to zero, calculated within the model or set to time independent values from ancillary files in the reconfiguration step. See README file for details.
- (5) Aerosols concentrations should either be initialised using a climatology calculated from the model’s AMIP simulation, or initialised using the nudging method of Boyle et al. (2005).
As we do not have aerosol climatologies we have taken the aerosol concentrations from the ancillary files that were used for the CMIP5 experiment. As the T-AMIP period is in 2008-9 and CMIP5 runs cease in 2005, we have extended the aerosol concentrations to this period by using those ancillaries from the ACCESS 1.3 AR5 RCP4.5 experiment where needed.
- (6) Non-state variable prognostics which spin-up quickly (such as cloud fraction for models with a prognostic scheme) can either be initialised from zero, or initialised using the nudging method of Boyle et al. (2005).
We initialize to zero all the cloud fields: QCF AFTER TIMESTEP, CONV CLOUD BASE LEVEL NO. AFTER TS, CONV CLOUD TOP LEVEL NO. AFTER TS, CONV CLOUD LIQUID WATER PATH, QCL AFTER TIMESTEP, AREA CLOUD FRACTION IN EACH LAYER., BULK CLOUD FRACTION IN EACH LAYER, LIQUID CLOUD FRACTION IN EACH LAYER, and FROZEN CLOUD FRACTION IN EACH LAYER (STASH codes: 12, 14, 15, 16, 254, 265, 266, 267, and 268)
- (7) The AGCMs submitted should be the same (both in terms of physics and resolution) as those used for the CMIP5 AMIP experiment in order to compare model biases across timescales.
The ACCESS model used for this is the ACCESS 1.3 atmospheric (AMIP) configuration as used in the CAWCR submission to CMIP5. See the Library of standard ACCESS experiments and test cases, Climate experiment configurations, ACCESS 1.0 and 1.3 AMIP configurations https://trac.nci.org.au/trac/access/wiki/ACCESS_AMIP_testcases for more information.
Allow for CABLE in ACCESS 1.3
Unfortunately ACCESS 1.3 cannot use the initial condition dump file created in (1) yet. This is because the reconfigured file is configured to run with the four soil levels and nine surface types of the MOSES land-surface scheme which is used in ACCESS 1.0 and not the six soil levels and seventeen tiles used in the CABLE land-surface scheme run in ACCESS 1.3.
In order to create ACCESS 1.3 initial conditions python scripts are used to copy the atmospheric fields from the ACCESS 1.0 reconfigured file created in (1) onto a basic ACCESS 1.3 AMIP dump file which has suitable levels and tiles. Similar scripts also copy the zeroed cloud fields discussed in (6) and LSM fields for the month of the dump file date taken from the climatology created in (4). A full description of how all the dump fields are initialized is in the README file.
ACCESS 1.3 T-AMIP run procedure
On NCI accesscollab/accessdev the T-AMIP ACCESS 1.3 run job is uaoob/vafrb, respectively, which are copies of sabmf, the AR5 ACCESS 1.3 AMIP job. We then changed this experiment by: applying the Atmospheric composition, solar forcing and land use changes in (2); point to the SST and SICE field ancillary files created in (3) and the aerosol ancillaries mentioned in (5); and point to the ACCESS 1.3 initial condition dump file discussed in the paragraph above.
The executable used is the NeCTAR Climate and Weather Science Laboratory standard exe that comes with the original umui job (sabmf). However if you wish to change the source code then you can copy the build job for this executable the standard NeCTAR job saadb and then compile it to create an executable. I have done this in my accesscollab/accessdev jobs uaoog/vafra, respectively.
This run job can then be run from the UMUI or we can use a raijin script (run_tamip_fcast00 for uaoob and described in the README file, or run_tamip_fcast_acdev for vafrb) to submit several jobs in series. The job runs on 64 processors and takes ~4minutes to complete a 5 day hindcast for one date, so ~4 hours for a full T-AMIP run over all 64 dates.
Example of ACCESS 1.3 T-AMIP outputs
Plots of the U10m and Q1.5m produced after 3 hours by ACCESS 1.3 T-AMIP for starting date 0600 UTC on 2009-01-16 (as described above) are:
Bi et al. (2013), referred to as B13 below, described the ACCESS 1.0 and 1.3 Coupled Models (CM) and presented results from their CMIP5 simulations, including some from the ‘present climate’ which is the 30-year average over 1976-2005 from their 20th century historical simulations. Using the above procedure we have calculated corresponding figures for the ACCESS 1.3 T-AMIP runs (over all 64 dates), as well as for a full ACCESS 1.3 AMIP run, for comparison.
The figure below shows the (a) AMIP and (b) T-AMIP surface air temperature (SAT) bias, relative to ERA-Interim reanalysis 1979-2008 data. If compared to the ACCESS 1.3 CM results in B13 Fig. 3(b) then the T-AMIP and AMIP land mass biases are very similar.
The SAT biases indicate that the ACCESS T-AMIP experimental procedure is able to simulate the climate biases seen in the ACCESS AMIP and CM runs. This suggests that ACCESS T-AMIP could be a valuable tool in examining the development of these systematic biases in the ACCESS model.
ACCESS 1.3 T-AMIP scripts, datasets, sample outputs and documentation
All the ACCESS 1.3 T-AMIP scripts, datasets and example outputs discussed above - and covered in more detail in the README file - can be found on raijin at:
raijin at /g/data/p66/glr548/tamip_sam/access1.3 and /projects/access/AMEL/TransposeAMIPII/access1.3
This directory holds sub-directories:
- era-recon: holds the saacb reconfiguration dump files created from the YOTC grib files in directory grib
- obs: yotcu10v10T2*nc hold the corresponding YOTC files U10m, V10m and T2m fields used as “truth” for the 5day simulation surface forecasts discussed in Figs 1, 2 and 3 in Roff (2014)
- ../docs: holds a pdf copy of the paper "ACCESS Transpose-AMIP: Experimental Procedure and Preliminary Results", Greg Roff (accepted CAWCR Res.Let., Oct 2014) which describes in more detail the Transpose-AMIP II experimental design and its application to the ACCESS 1.3 model as well as a companion README file with even more detail and referred to above - these are in files tamip_access1.3_CAWCRresLet.pdf and tamip_access1.3_README.pdf, respectively
- example_output: example output files for ACCESSS 1.3 T-AMIP run for date 2009071500 as well as the postprocessed file 2009071500.nc produced by postproc00_tamip
- ../grib: holds the combined YOTC T-AMIP grib files used in the reconfiguration to create the era-recon dump files
- lsm: holds the scripts and files needed to overwrite the dump file with CABLE LSM fields using the AMIP CABLE climatology ts_clim_a1.3N96L38amip_cableFLDS.nc
- na1.3c: holds the ACCESS 1.3 T-AMIP ic dump files and the raijin python scripts used to create these from the saacb reconfigured dump files in ../era-recon
- sstsice: holds the YOTC sst+ice fields and the scripts used to create the T-AMIP sst+ice ancillary files: sstN96_20081001nd61 sstN96_20090101nd59 sstN96_20090401nd61 sstN96_20090701nd62 iceN96_20081001nd61 iceN96_20090101nd59 iceN96_20090401nd61 iceN96_20090701nd62
- tamip_run: which holds the ACCESS 1.3 T-AMIP automatic run script (run_tamip_fcast00), the corresponding control files directory (new) and the postprocessing script (postproc00_tamip).
Back to ACCESS Model Experiment Library
[19] Review and updates by greg Roff
For further support on this suite, please contact [email protected]