Testing
Home Introduction Methodology Testing Output Plots Validation Future Work Papers Comments

 

Real-time Testing

    The kilo-member ensemble system was fully automated for real-time testing during the 2002 and 2003 seasons for the Atlantic and Eastern Pacific basins. Following is the current ensemble configuration:


Resolution of GFS ensemble fields     2 x 2 deg

Resolution of MUDBAR                     3 nested grids, each with 32x32 grid points, resulting in grid spacing of 193 km, 97 km, and 48 km

Deep layer means                           shallow (850-500 hPa), medium (850-350 hPa), deep (850-200 hPa), and very deep (1000-100 hPa)

Equivalent phase speed (for gamma) 50 m/s, 100 m/s, 300 m/s

Size/strength (Vmax)                       15 m/s, 30 m/s, 50 m/s

Motion vector perturbations              1 m/s (fast and right, slow and right, slow and left, fast and left, and no perturbation)

Ensemble size                                1980 members


The operational ensemble run on a 450 MHz Pentium Linux PC.  Each forecast case takes approximately 4600 seconds (76 min) to run. When I reran the 2003 cases on my 2 GHz Pentium PC at home, also a Linux box, each case took approximately 2000 seconds (33 min). Post processing is accomplished via a Korn Shell scripting system that utilizes a mix of Fortran programs and NCAR Command Language (NCL) scripts. 

Currently Atlantic storms run first, then the Eastern Pacific run.  The output plots are generated and are automatically posted to this web site. Here is the timing associated with the current configuration (all times are MDT):

8:00 PM                   This is the time for which the forecast is valid -- NHC operational storm data are put into ATCF format

Midnight - 3:00 AM    GFS ensembles run on IBM/SP at NCEP (operational storm information is used in a storm relocation algorithm for the control member of the GFS ensemble, but not the positively and negatively perturbed members)

4:00 AM                   GFS ensemble fields are uploaded to local machine

4:15 AM                   Deep layer mean fields are produced on local machine

4:30 AM                   The operational storm information estimates from night before are uploaded to local machine

4:35 AM                   Kilo-ensemble starts running, taking about 2 hours per case, including post processing

6:30 AM                   If from first storm case becomes available on web at approximately this time

8:30 AM                   Output from second case gets posted to web at approximately this time

10:45 PM                 Output from third case available

1:00 PM                   Output from fourth case available

 

Go on to output plots . . .