12. Running forecasts

With the method described by [Marchesiello, 2008], CROCO can be used to downscale global forecasts in order to provide higher resolution forecasting. To do so, It is advised to start from building a regional configuration following the dedicated tutorials and testing it with climatological or interannual forcing after editing crocotools_param.m, croco.in, cppdefs.h, param.h. Then, specifically for the forecast system, you will need to set the forecast parameters in crocotools_param.m, define cpp key ROBUST_DIAG in cppdefs.h and recompile the model using jobcomp, then edit croco_forecast.in. Finally, run the script run_croco_forecast.bash that will download global forecast (ocean and atmosphere), create specific forcing files using the Forecast_tools Matlab package, then run croco for the current period.

12.1. Strategy of Forecast_tools

CROCO_TOOLS allows running both inter-annual and real-time simulations. In the latter case, we rely on operational global ocean circulation models for the initial and lateral boundary conditions and operational global atmospheric models (introduced through bulk formulations) for surface forcing. In order to limit the volume of data transferred over the Internet, we use OPeNDAP (Matlab) or Motuclient (Python) and extract only the necessary subgrid. We use the U.S. NCEP/NOAA data base for atmospheric forcing and the European CMEMS data for oceanic forcing.

In a regional domain with low intrinsic variability where the circulation is directly forced by surface fluxes (away from fronts and eddies), initialization is of lower importance. However, if oceanic intrinsic variability with low predictability is dominant (e.g., mesoscale eddies with monthly time-scale), the model may well provide a statistically reliable image of the eddy field, but out of phase at any given time. An initialization method is therefore needed to bring ocean fields into phase with real time. Here, we do not perform data assimilation, but rely on the global CMEMS forecast products, which assimilates satellite altimetry and in-situ data. Newtonian nudging is used to adjust CMEMS data to CROCO dynamics in a downscaling procedure.

The strategy for a hindcast/forecast cycle is as follows. The simulation starts at t0-hdays (hdays=1, t0 being the present time) and ends at t0+fdays (fdays=3). The model is run using interpolated data from CMEMS for lateral boundary conditions and NCEP for surface forcing. CMEMS data or a CROCO restart file (from a previous forecast) is used for initialization at t0-1, while nudging assimilates CMEMS data in the interior domain (with a nudging time-scale of about 10 days). A Shell script (run_croco_forecast.bash) manages the whole procedure: download and pre-processing, forecast simulations, post-processing (plot_forecast_croco.m), data storage and preparation of the next forecast cycle.

12.2. Set forecast parameters

  1. Edit crocotools_param.m section 8

    % 8 - Parameters for the forecast system
    %     --> select OGCM name above (mercator ...)
    %     --> don't forget to define in cppdefs.h:
    %                    - ROBUST_DIAG
    %                    - CLIMATOLOGY
    %                    - BULK_FLUX
    %                    - TIDES if you choose so, but without TIDERAMP
    FRCST_dir = [FORC_DATA_DIR,'Forecast/'];  % path to local OGCM data directory
    % Number of hindcast/forecast days
    if strcmp(OGCM,'ECCO')
    elseif strcmp(OGCM,'mercator')
    % Local time= UTC + timezone
    timezone = +2;
    % Add tides
    add_tides_fcst = 1;       % 1: add tides
    %  MERCATOR case:
    %  =============
    %  To download CMEMS data: set login/password (http://marine.copernicus.eu)
    %  and path to croco's motuclient python package;
    %  or set pathMotu='' (empty) to use your own motuclient
    %  Various sets of data are proposed in the Copernicus web site
    if strcmp(OGCM,'mercator')
      user     = 'XXX';
      password = 'XXX';
      pathMotu =[CROCOTOOLS_dir,'Forecast_tools/'];
      mercator_type=1;   % 1 -->  1/12 deg Mercator forecast
                         % 2 -->  high-resolution regional grids
      if mercator_type==1
        product_id_fcst={'cmems_mod_glo_phy_anfc_0.083deg_P1D-m', ...
                         'cmems_mod_glo_phy-cur_anfc_0.083deg_P1D-m', ...
                         'cmems_mod_glo_phy-thetao_anfc_0.083deg_P1D-m', ...
      elseif mercator_type==2
       product_id_fcst={'cmems_mod_med_phy-ssh_anfc_4.2km_P1D-m', ...
                        'cmems_mod_med_phy-cur_anfc_4.2km_P1D-m', ...
                        'cmems_mod_med_phy-tem_anfc_4.2km_P1D-m', ...

    Make sure that OGCM=mercator (section 7), choose the hindcast/forecast period hdays/fdays (start with default),set your time zone and then your login/password (http://marine.copernicus.eu) to access CMEMS data through motuclient python package in Forecast_tools directory.


    CMEMS url and files may sometimes change name, resulting in system failure until we make the necessary adjustments.

  2. Edit croco_forecast.in, and set specific parameters

    Simulation time: typically 4 days (hdays+fdays)

    time_stepping: NTIMES   dt[sec]  NDTFAST  NINFO
                    480      720      60       10

    Output frequency: 1 day (set NRST,NWRT,NAVG according to time step dt)

    restart:          NRST, NRPFRST / filename
                       120    0
    history: LDEFHIS, NWRT, NRPFHIS / filename
                 T     120    0
    averages: NTSAVG, NAVG, NRPFAVG / filename
                 1     120    0

    Nudging to forecast data (on the order of 10 days)

    nudg_cof:    TauT_in, TauT_out, TauM_in, TauM_out  [days for all]
                    1.       10.      1.      360.

12.3. Compiling

  1. Edit cppdefs.h for defining forcing and nudging procedures:

    # define BULK_FLUX
                  /* Lateral Forcing */
    # define CLIMATOLOGY
    # ifdef CLIMATOLOGY
    #  define ZCLIMATOLOGY
    #  define M2CLIMATOLOGY
    #  define M3CLIMATOLOGY
    #  define TCLIMATOLOGY
    #  define ZNUDGING
    #  define M2NUDGING
    #  define M3NUDGING
    #  define TNUDGING
    #   define ROBUST_DIAG
    # endif

    ROBUST_DIAG sets non-zero nudging coefficients tauT_out for nudging to CMEMS forecast (T,S) data.

  2. Re-compile the model:

    ./jobcomp > jobcomp_forecast.log

12.4. Running the script

  1. Copy run_croco_forecast.bash from SCRIPTS/Plurimonths_scripts to local directory.

  2. Edit run_croco_forecasts.bash.

    Check in particular the locations of matlab executable and loaddap library (used to extract NCEP atmospheric forecast data)

    export TOOLSDIR=$HOME/croco/croco_tools/Forecast_tools
    export RUNDIR=${PWD}
    export MATLAB=/usr/local/bin/matlab
    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/loaddap-3.5.2/lib

    Check the forecast cycle options (default choice is generally fine but you may want to set RESTART=1 after the first forecast cycle)

    # Get forcing Files from DODS SERVER and process them for CROCO
    # PRE_PROCESS=1 ==> do the work (0 otherwise)
    export PRE_PROCESS=1
    # Perform Iterations for convergence of CROCO and OGCM
    # ITERATION=1 ==> do several hindcasts using nudging
    export ITERATION=0
    # Restart from previous forecast
    export RESTART=0
    # Clean results of previous forecasts
    export CLEANING=0
    # ONERUN process: ONERUN=1: make one run for hindcast/forecast
    #                 ONERUN=0: make two runs for hindcast and forecast
    export ONERUN=1
  3. Run the script (from Terminal or from Crontab)