Remote and gateway modes

For member state users it is recommended to use the remote or gateway mode, especially for more demanding tasks, to retrieve and convert data on ECMWF machines and to transfer only the final output files to the local host.

Remote mode

The only difference between both modes is the users working location. In the remote mode you have to login to the ECMWF server and then go to the Run directory as shown above. At ECMWF servers flex_extract is installed in the $HOME directory. However, to be able to start the program you have to load the Python3 environment with the module system first.

# Remote mode
ssh -X <ecuid>@ecaccess.ecmwf.int

# On ECMWF server
[<ecuid>@ecgb11 ~]$ module load python3
[<ecuid>@ecgb11 ~]$ cd flex_extract_vX.X/Run

Gateway mode

For the gateway mode you have to log in on the gateway server and go to the Run directory of flex_extract:

# Gateway mode
ssh <user>@<gatewayserver>
cd <path-to-flex_extract_vX.X>/Run

From here on the working process is the same for both modes.

For your first submission you should use one of the example CONTROL files stored in the Control directory. We recommend to extract CERA-20C data since they usually guarantee quick results and are best for testing reasons.

Therefore open the run.sh file and modify the parameter block marked in the file as shown below:

# -----------------------------------------------------------------
# AVAILABLE COMMANDLINE ARGUMENTS TO SET
#
# THE USER HAS TO SPECIFY THESE PARAMETERS:

QUEUE='ecgate'
START_DATE=None
END_DATE=None
DATE_CHUNK=None
JOB_CHUNK=3
BASETIME=None
STEP=None
LEVELIST=None
AREA=None
INPUTDIR=None
OUTPUTDIR=None
PP_ID=None
JOB_TEMPLATE='job.temp'
CONTROLFILE='CONTROL_CERA'
DEBUG=0
REQUEST=2
PUBLIC=0

This would retrieve a one day (08.09.2000) CERA-20C dataset with 3 hourly temporal resolution and a small 1° domain over Europe. Since the ectrans parameter is set to 1 the resulting output files will be transferred to the local gateway into the path stored in the destination (SEE INSTRUCTIONS FROM INSTALLATION). The parameters listed in the run.sh file would overwrite existing settings in the CONTROL file.

To start the retrieval you only have to start the script by:

./run.sh

Flex_extract will print some information about the job. If there is no error in the submission to the ECMWF server you will see something like this:

---- On-demand mode! ----
The job id is: 10627807
You should get an email per job with subject flex.hostname.pid
FLEX_EXTRACT JOB SCRIPT IS SUBMITED!

Once submitted you can check the progress of the submitted job using ecaccess-job-list. You should get an email after the job is finished with a detailed protocol of what was done.

In case the job fails you will receive an email with the subject ERROR! and the job name. You can then check for information in the email or you can check on ECMWF server in the $SCRATCH directory for debugging information.

cd $SCRATCH
ls -rthl

The last command lists the most recent logs and temporary retrieval directories (usually pythonXXXXX, where XXXXX is the process id). Under pythonXXXXX a copy of the CONTROL file is stored under the name CONTROL, the protocol is stored in the file prot and the temporary files as well as the resulting files are stored in a directory work. The original name of the CONTROL file is stored in this new file under parameter controlfile.

.. code-block:: bash
    :caption: "Example structure of ``flex_extract`` output directory on ECMWF servers."
    pythonXXXXX
    ├── CONTROL
    ├── prot
    ├── work
    │   ├── temporary files
    │   ├── CE000908* (resulting files)

If the job was submitted to the HPC ( queue=cca or queue=ccb ) you may login to the HPC and look into the directory /scratch/ms/ECGID/ECUID/.ecaccess_do_not_remove for job logs. The working directories are deleted after job failure and thus normally cannot be accessed.

To check if the resulting files are still transferred to local gateway server you can use the command ecaccess-ectrans-list or check the destination path for resulting files on your local gateway server.

Local mode

To get to know the working process and to start your first submission you could use one of the example CONTROL files stored in the Control directory as they are. For quick results and for testing reasons it is recommended to extract CERA-20C data.

Open the run_local.sh file and modify the parameter block marked in the file as shown below. The differences are highlighted.

.. code-block:: bash
   :emphasize-lines: 16,20,23
   # -----------------------------------------
   # AVAILABLE COMMANDLINE ARGUMENTS TO SET
   #
   # THE USER HAS TO SPECIFY THESE PARAMETERs:
   #
   QUEUE=''
   START_DATE=None
   END_DATE=None
   DATE_CHUNK=None
   JOB_CHUNK=None
   BASETIME=None
   STEP=None
   LEVELIST=None
   AREA=None
   INPUTDIR='./Workspace/CERA'
   OUTPUTDIR=None
   PP_ID=None
   JOB_TEMPLATE=''
   CONTROLFILE='CONTROL_CERA'
   DEBUG=0
   REQUEST=0
   PUBLIC=0

.. code-block:: bash
   :emphasize-lines: 16,20,23
   # -----------------------------------------
   # AVAILABLE COMMANDLINE ARGUMENTS TO SET
   #
   # THE USER HAS TO SPECIFY THESE PARAMETERs:
   #
   QUEUE=''
   START_DATE=None
   END_DATE=None
   DATE_CHUNK=None
   JOB_CHUNK=None
   BASETIME=None
   STEP=None
   LEVELIST=None
   AREA=None
   INPUTDIR='./Workspace/CERApublic'
   OUTPUTDIR=None
   PP_ID=None
   JOB_TEMPLATE=''
   CONTROLFILE='CONTROL_CERA.public'
   DEBUG=0
   REQUEST=0
   PUBLIC=1

This would retrieve a one day (08.09.2000) CERA-20C dataset with 3 hourly temporal resolution and a small 1° domain over Europe. The destination location for this retrieval will be within the Workspace directory within Run. This can be changed to whatever path you like. The parameters listed in run_local.sh would overwrite existing settings in the CONTROL file.

To start the retrieval you then start the script by:

./run_local.sh

While job submission on the local host is convenient and easy to monitor (on standard output), there are a few caveats with this option:

1. There is a maximum size of 20GB for single retrieval via ECMWF Web API. Normally this is not a problem but for global fields with T1279 resolution and hourly time steps the limit may already apply.
2. If the retrieved MARS files are large but the resulting files are relative small (small local domain) then the retrieval to the local host may be inefficient since all data must be transferred via the Internet. This scenario applies most notably if etadot has to be calculated via the continuity equation as this requires global fields even if the domain is local. In this case job submission via ecgate might be a better choice. It really depends on the use patterns and also on the internet connection speed.

Public datasets

The main difference in the definition of a CONRTOL file for a public dataset is the setting of the parameter DATASET. This specification enables the selection of a public dataset in MARS. Otherwise the request would not find the dataset. For the two public datasets CERA-20C and ERA-Interim an example file with the ending .public is provided and can be used straightaway.

CONTROL_CERA.public
CONTROL_EI.public

For CERA-20C it seems that there are no differences in the dataset against the full dataset, while the public ERA-Interim has only analysis fields every 6 hour without filling forecasts in between for model levels. Therefore it is only possible to retrieve 6-hourly data for public ERA-Interim.

CERA

For this dataset it is important to keep in mind that the dataset is available for the period 09/1901 until 12/2010 and the temporal resolution is limited to 3-hourly fields. It is also a pure ensemble data assimilation dataset and is stored under the enda stream. It has 10 ensemble members. The example CONTROL files will only select the first member (number=0). You may change this to another number or a list of numbers (e.g. NUMBER 0/to/10). Another important difference to all other datasets is the forecast starting time which is 18 UTC. Which means that the forecast in CERA-20C for flux fields is 12 hours long. Since the forecast extends over a single day we need to extract one day in advance and one day subsequently. This is automatically done in flex_extract.

ERA 5

This is the newest re-analysis dataset and has a temporal resolution of 1-hourly analysis fields. Up to date it is available until April 2019 with regular release of new months. The original horizontal resolution is 0.28125° which needs some caution in the definition of the domain, since the length of the domain in longitude or latitude direction must be an exact multiple of the resolution. It might be easier for users to use 0.25 for the resolution which MARS will automatically interpolate. The forecast starting time is 06/18 UTC which is important for the flux data. This should be set in the CONTROL file via the ACCTIME 06/18 parameter in correspondence with ACCMAXSTEP 12 and ACCTYPE FC.

ERA-Interim

This re-analysis dataset will exceed its end of production at 31st August 2019! It is then available from 1st January 1979 to 31st August 2019. The etadot is not available in this dataset. Therefore flex_extract must select the GAUSS parameter to retrieve the divergence field in addition. The vertical velocity is the calculated with the continuity equation in the Fortran program calc_etadot. Since the analysis fields are only available for every 6th hour, the dataset can be made 3 hourly by adding forecast fields in between. No ensemble members are available.

Operational data

This is the real time atmospheric model in high resolution with a 10-day forecast. This means it underwent regular adaptations and improvements over the years. Hence, retrieving data from this dataset needs extra attention in selecting correct settings of parameter. See :ref:`ref-tab-dataset-cmp` for the most important parameters. Nowadays, it is available 1 hourly by filling the gaps of the 6 hourly analysis fields with 1 hourly forecast fields. Since 4th June 2008 the eta coordinate is directly available so that ETA should be set to 1 to save computation time. The horizontal resolution can be up to 0.1° and in combination with 137 vertical levels can lead to troubles in retrieving this high resolution dataset in terms of job duration and quota exceedence. It is recommended to submit such high resolution cases for single day retrievals (see JOB_CHUNK parameter in run.sh script) to avoid job failures due to exceeding limits.

CONTROL files for normal daily retrievals with a mix of analysis and forecast fields are listed below:

CONTROL_OD.OPER.4V.eta.global
CONTROL_OD.OPER.FC.eta.global
CONTROL_OD.OPER.FC.eta.highres
CONTROL_OD.OPER.FC.gauss.highres

These files defines the minimum number of parameters necessary to retrieve a daily subset. The setup of field types is optimal and should only be changed if the user understands what he does. The grid, domain and temporal resolution can be changed according to availability.

Pure forecast

It is possible to retrieve pure forecasts exceeding a day. The forecast period available depends on the date and forecast field type. Please use MARS catalogue to check the availability. Below are some examples for 36 hour forecast of Forecast (FC), Control forecast (CF) and Calibration/Validation forecast (CV). The CV field type was only available 3-hourly from 2006 up to 2016. It is recommended to use the CF type since this is available from 1992 (3-hourly) on up to today in 1-hourly temporal resolution. CV and CF field types belong to the Ensemble prediction system (ENFO) which contain 50 ensemble members. Please be aware that in this case it is necessary to set the specific type for flux fields explicitly, otherwise it could select a default value which might be different from what you expect!

CONTROL_OD.ENFO.CF.36hours
CONTROL_OD.ENFO.CV.36hours
CONTROL_OD.OPER.FC.36hours

Half-day retrievals

If a forecast for just half a day is wanted it can be done by substituting the analysis fields also by forecast fields as shown in files with twiceaday in it. They produce a full day retrieval with pure 12 hour forecasts twice a day. It is also possible to use the operational version which would get the time information from ECMWF's environmental variables and therefore get the newest forecast per day. This version uses a BASETIME parameter which tells MARS to extract the exact 12 hours upfront to the selected date. If the CONTROL file with basetime in the filename is used this can be done for any other date too.

CONTROL_OD.OPER.FC.eta.basetime
CONTROL_OD.OPER.FC.operational
CONTROL_OD.OPER.FC.twiceaday.1hourly
CONTROL_OD.OPER.FC.twiceaday.3hourly

Ensemble members

The retrieval of ensemble members were already mentioned in the pure forecast section and for CERA-20C data. In this flex_extract version there is an additional possibility to retrieve the Ensemble Long window Data Assimilation (ELDA) stream from the real-time dataset. This model version has (up to May 2019) 25 ensemble members and a control run (number 0). Starting from June 2019 it has 50 ensemble members. Therefore we created the possibility to double up the 25 ensemble members (before June 2019) to 50 members by taking the original 25 members from MARS and subtracting 2 times the difference between the member value and the control value. This is done by selecting the parameter DOUBLEELDA and set it to 1.

CONTROL_OD.ELDA.FC.eta.ens.double
CONTROL_OD.ENFO.PF.ens

Specific features

rrint

Decides if the precipitation flux data uses the old (0) or new (1) disaggregation scheme. See :doc:`Documentation/disagg` for explanaition.

?

cwc

Decides if the total cloud water content will be retrieved (set to 1) in addition. This is the sum of cloud liquid and cloud ice water content.

addpar

With this parameter an additional list of 2-dimensional, non-flux parameters can be retrieved. Use format param1/param2/.../paramx to list the parameters. Please be consistent in using either the parameter IDs or the short names.

doubleelda

Use this to double the ensemble member number by adding further disturbance to each member.

debug

If set to 1 all temporary files were kept at the end. Otherwise everything except the final output files will be deleted.

request

This produces an extra csv file mars_requests.csv where the content of each mars request of the job is stored. Useful for debugging and documentation.

mailfail

At default the mail is send to the mail connected with the user account. Add additional email addresses if you want. But as soon as you enter a new mail, the default will be overwritten. If you would like to keep the mail from your user account, please add ${USER} to the list ( comma seperated ) or mail addresses.

Hints for definition of some parameter combinations

Field types and times

This combination is very important. It defines the temporal resolution and which field type is extracted per time step. The time declaration for analysis (AN) fields uses the times of the specific analysis and (forecast time) steps have to be 0. The forecast field types (e.g. FC, CF, CV, PF) need to declare a combination of (forescast start) times and the (forecast) steps. Both of them together defines the actual time step. It is important to know the forecast starting times for the dataset to be retrieved, since they are different. In general it is enough to give information for the exact time steps, but it is also possible to have more time step combinations of TYPE, TIME and STEP because the temporal (hourly) resolution with the DTIME parameter will select the correct combinations.

?

.. code-block:: bash
   :caption: Example of a setting for the field types and temporal resolution.
    DTIME 3
    TYPE AN FC FC FC AN FC FC FC
    TIME 00 00 00 00 12 12 12 12
    STEP 00 03 06 09 00 03 06 09

Vertical velocity

The vertical velocity for FLEXPART is not directly available from MARS. Therefore it has to be calculated. There are a couple of different options. The following parameters are responsible for the selection. See :doc:`Documentation/vertco` for a detailed explanation. The ETADIFF, OMEGA and OMEGADIFF versions are only recommended for debugging and testing reasons. Usually it is a decision between GAUSS and ETA, where for GAUSS spectral fields of the horizontal wind fields and the divergence are to be retrieved and used with the continuity equation to calculate the vertical velocity. For ETA the latitude/longitude fields of horizontal wind fields and eta-coordinate are to be retrieved. It is recommended to use ETA where possible due to a reduced computation time.

??

.. code-block:: bash
    :caption: Example setting for the vertical coordinate retrieval.
    GAUSS 0
    ETA 1
    ETADIFF 0
    DPDETA 1
    OMEGA 0
    OMEGADIFF 0

Grid resolution and domain

?

The grid and domain selection depends on each other. The grid can be defined in the format of normal degrees (e.g. 1.) or as in older versions by 1/1000. degrees (e.g. 1000 for 1°). After selecting the grid, the domain has to be defined in a way that the length of the domain in longitude or latitude direction must be an exact multiple of the grid. The horizontal resolution for spectral fields will be set by the parameter RESOL. For information about how to select an appropriate value you can read the explanation of the MARS keyword here and in this table.

?

.. code-block:: bash
    :caption: Example setting for a northern hemisphere domain with a grid of ``0.25°``.
    GRID 0.25
    RESOL 799
    SMOOTH 0
    UPPER 90.
    LOWER 0.
    LEFT -179.75
    RIGHT 180.

Flux data

The flux fields are accumulated forecast fields all the time. Since some re-analysis dataset nowadays have complete set of analysis fields in their temporal resolution it was important to define a new parameter set to define the flux fields since the information could not be taken from TYPE, TIME and STEP any longer. Select a forecast field type ACCTYPE, the forecast starting time ACCTIME and the maximum forecast step ACCMAXSTEP. The DTIME parameter defines the temporal resolution for the whole period.

?

.. code-block:: bash
   :caption: Example setting for the definition of flux fields.
    DTIME 3
    ACCTYPE FC
    ACCTIME 00/12
    ACCMAXSTEP 36

.. toctree::
    :hidden:
    :maxdepth: 2