[ UWME Graphics Selection ] [ UWME Verification & Performance ] [ UWME Publications & Presentations ]

Log of Changes for the UW Mesoscale Ensemble


The GASP member switched from the Australian Bureau of Meteorology's 
APS0 model to their APS1 version. The UWME changeover commenced with 
the 2013031400 UTC forecast cycle.


The NGPS member switched from the US Navy's NOGAPS model to their 
NAVGEM model. The UWME changeover commenced with the 2013031312 UTC 
forecast cycle.


The UWME system was ported from MM5 to WRF and is run in parallel
during a trial period, beginning with the 2010062700 UTC forecast.


The UWME+ system was extended to 72 h forecast lead time and set to
run for both 00 and 12 UTC cycles beginning with the 2007030512 UTC forecast.


The UWME system was extended to 72 h forecast lead time beginning with the
2005082312 UTC forecast cycle.


Ensemble member graphics were updated to match those of the deterministic UW MM5-GFS
run, beginning with the 2005063000 UTC forecast cycle.  Also, several subtle
improvements to the web-page functionality were made.


The UWME system has undergone several major changes.  Beginning with the
2005012512 UTC forecast cycle, all UWME and UWME centroid MM5 forecasts
are using: 
  • Reisner II microphysics
  • CCM2 radiation with a modified effective ice radius
  • NCEP's RUC 20-km soil moisture initialization
  • Dump-bucket soil moisture scheme during the runs
    The Linux cluster running the UWME system has been upgraded and re-configured. Each run now takes about 1.5 hours wall-clock time on 8 Intel Xeon CPUs using MPI. The last UWME member usually finishes by around 0840/2040 UTC. The UWME centroid, mean and spread, and raw probability products are typically available around 0900/2100 UTC.
    The UWME+ system is currently unchanged (although production may be moved slightly earlier due to reduced load). Work is commencing now to update the UWME+ system and details will be available when the system is ready.
    The UWME system has undergone two major changes. Beginning with the
    2004042712 UTC forecast cycle, all UWME, UWME+, and centroid MM5
    forecasts are using nudging on the 36 km domain. Additionally,
    beginning with this same date, the UWME is running full 48 hour MM5
    forecasts on the 12 UTC forecast cycle. No change in naming or web
    availability has been made. 
    The UWME ensemble is up and running again.  Other runs will follow.  Derived
    products should be available in a few days.
    After six weeks, most of the data has been recovered.  However, some data was
    lost.  The ensemble runs will be restored ASAP, but with some changes.  These
    changes include dropping the mirrored runs permanently and renaming the system
    accordingly.  We've adopted UWME, the UW Mesoscale Ensemble.  UWME+ will refer
    to the perturbed-model ensemble (previously called ACMEcore+).  Also there will
    be a new location for the website (www.atmos.washington.edu/~ens/uwme.cgi).
    Catastrophic failure of the disk storage system occurred making the ensemble runs
    Discontinued operational mirrored member runs, updated web page, and changed cron
    jobs to run postprocessing for the PHYS ensemble, not ACME.
    Changed probability calculation method from democratic voting to uniform ranks.
    This provides a better approximation to the probability that is also continuous
    and spatially smoother.  For research applications, complete cases of archived
    probability files were rerun back to 2002103100.
    Note that the raw probability forecasts are still uncalibrated.  We have simply
    used a better method of specifying forecast probability from a set of discrete
    ensemble forecasts.
    Modification to purging scripts to keep past case files from being
      deleted if they are less than 72 hours old.
    Found and fixed a few bad SST perturbation values.  
    pert_sst.F was written to read f8.4 values from the formatted data files 
    (e.g., /home/disk/emm5rt/ensemble/preprocess/SST/SST_pert01_d1.dat).  
    Exponential notation values were in the dat files for values smaller than
    0.0001, so for example 0.000023 was written as 2.3e-0.5.  This was then
    read in as a perturbation value of 2.3.  These values are now reset to 0.0.
    The impact on the ACMEcore+ runs to date is negligable.  There were only
    1 or 2 of these bad values per member, and some were in the bogus land
    area where they make no difference.  Where they were actually over water,
    the surounding perturbations likely washed out their effect.
    Modified ensm_preprocess.pl to fix bug associated with an initial
     inability to locate the server of the ensembles output directory.
     This bug caused the UKMO forecasts to fail from time to time even
     though all of the input grids were available.
    Slight modification to run_ensemble.pl and ensm_preprocess.pl to add
      more diagnostics. 
    Change to crontab entries to allow STDOUT and STDERR to go to same
      file and to cycle the log files every quarter of the year.
    Switched centroid wait time to 10 Z from 11 Z.
    Turning back on 1-hourly output for domain 2 with Rick's mods to
     reset extreme values only every three hours.  2002121800 will be
     first cycle with these mods.
    11 pm
    Commented out changes to ensm_postprocess.iofilter.pl.  They did not
     seem to work (graphics were missing).
    9 pm
    Modified ensm_postprocess.iofilter.pl to use iofilter.pl to call
     graphics creation and GEMPAK file creation scripts.  The
     postprocessing machine is getting swamped and it is possible this
     will fix that.  Moved CMCG and CENT ensemble members back to tahoma.
    4 pm
    Moved CMCG and CENT ensemble members off of tahoma and back to
     ensemble machines in attempt to stabilize MM5 real-time
    Turned back on ensemble cron jobs after data move completed. 9:35 pm PST.
    Major upcoming changes:
    1) Target date for the beginning of PHASE IV is 10/01/2002.  UKMO will be re-instated.
    We are also adding the JMA (Japan) ICs, for a total of 8 core members.  This
    results in the production of 8 mirrored perturbation runs.  Including the
    centroid, we will now have a 17-member ensemble, titled ACME (Analysis-
    Centroid Mirroring Ensemble).
    2) The perturbation factor in the mirroring process will be increased 10% to 1.1.
    This increase accounts for the deficiency in normalized variance that results
    when doubling the number of samples by using a perturbation factor of 1.0.  Thus,
    the normalized variance of the core members will now be matched by the normalized
    variance of the core and mirrored members together.  In idealized experiments with
    1-D sine waves, this slight increase results in mirrored waves that have identical
    spectral amplitudes to the original waves.
    3) Core members and the centroid will also be run at the 1200 UTC cycle.  Thus,
    all products associated with ACMEcore will be generated at the 1200 UTC
    cycle and viewable on the web page.
    4) Also at this time, we will be running 8 perturbed-physics model runs in
    a near-real-time mode.  The ICs for these runs will be from the core
    members.  Access to the regular plots for these runs will only be
    accessible in the department.  However, they will be accessible via the
    web page through the file listing below the graphics selection table.
    This access is the same for the mirrored perturbation runs.
    5) MM5 Ensemble Forecast Website changed to:  http://www.atmos.washington.edu/~emm5rt/ensemble.cgi
    A full description of ACME will be available soon.
    Feed for CMC-GEM data has been sped up, resulting in earlier finishing times
    for that run, the centroid, and the mirrored perturbation runs.  The core
    runs now are usually finished before 1200 UTC, sometimes by 1100 UTC.  The
    centroid and mirrored perturbation runs start firing up an hour after the
    last of the 6 ICs arrive.  This is now much earlier, usually 0700-0730 UTC.
    Feed for UKMO data has been discontinued.  Hoping to re-instate this model
    on or before 10/01/2002, the target starting date for PHASE IV.  Back to 6
    ICs temporarily.
    Seven ensemble perturbation runs are added, called "mirrors".  The ICs for
    these runs are created by reflecting each of the seven core analyses across
    the centroid analysis.  The mirror perturbation runs begin at approximately
    1030 UTC (0230 PST) and finish by about 1600 UTC (0800 PST).
    GASP (Australia) ICs are added to the ensemble.  Total number of ICs now at 7.
    Updated Mean & Stdev plots...these plots should now be available operationally.
    New 10-node (20 processor) Linux PC cluster is online and operational.  First successful
    ensemble forecast runs are made on the new machine.
    Phase III begins.  Mixed-physics members still discontinued until new computer arrives.
    MRF permanently discontinued.  TCWB and UKMO ICs added to the mix.  Switch to version 3.4
    of the MM5 code is made.
    Real-time deterministic run updated to MM5 Version 3.  Added 4 vertical sigma levels.
    Therefore, less processing time is available on the SUN ES6500.  Forced to discontinue
    ETA_TKE and ETA_REISNER ensemble physics runs indefinitely.
    Phase II ends.
    Lowest sigma-level wind speed and direction options are added to the mean & spread plot selection menus.
    Scalar verification statistics are posted for Phase II (Oct. 2000 - Mar. 2001).
    10/1/2000 - 3/31/2001
    Phase II real-time ensemble forecasts are carried out.
    Three physics members are added to the system now totalling eight members.  The
    Eta model serves as IC/LBC for all three physics members.  The physics changes
    are summarized by the table at:
    The Phase II setup of the experiment begins.  The NGM ensemble member is discontinued and
    replaced by the MRF.
    Scalar verification measures posted for Phase I (January-June, 2000) plus
    Winter (JFM) 2000 and Spring (AMJ) 2000 breakdowns.  Statistics include
    Mean Error (bias), Mean Absolute Error, and Root-Mean Squared Error.
    Added ensemble verification page with seasonal and YTD stats.
    Improved the website presentation and included the new option to view the ensemble
    spread plots.  Spread summaries also made available.
    Verification statistics for the ensemble over the 12 km domain made available.
    NCEP made changes to the NGM initialization process which may affect the NGM-MM5
    verification scores.
    Basic description of ensemble system completed.
    Ensemble mean now available starting with the 2000012000 run.
    MM5 Ensemble Forecast Website changed to:  http://www.atmos.washington.edu/~epgrimit/ensemble.cgi
    MM5 Ensemble Forecast Website available to the public in test phase only.
    Humble beginnings of the MM5 Ensemble Forecast Website.

    Questions and comments should be directed to: Eric Grimit

    Last Updated: August 25, 2005

    [UW Atmospheric Sciences] [UW MM5 Real-Time System] [UW Weather Information ]