Ocean modeling at NCEP

1
Ocean modeling at NCEP
Hendrik L. Tolman NOAA / NWS / NCEP / EMC Marine
Modeling and Analysis Branch Hendrik.Tolman_at_NOAA.
gov
2
RTOFS-Atlantic
1

• HYCOM application
• Primitive equation with free surface.
• Sub-grid scale parameterizations. Vertical and
  horizontal eddy viscosity and mixing. Diapycnal
  mixing.
• Tides river outflow (USGS, RIVDIS).
• Atmospheric fluxes (GFS).
• North Atlantic grid with 4-7km US coastal
  resolution (ETOPO2, NGDC)
• Open boundaries T,S from climatology
• SSH and barotropic velocity from tidal model
  (TPX06) and climatology.

3
RTOFS-Atlantic
2
4
RTOFS-Atlantic
3

• 2007 model updates
• Revised vertical grid to 26 layers
• Higher resolution in the shallow waters.
• Better resolution on the shelf break.
• Better representation of Denmark Iceland
  overflows.
• Resolving 4 vertical dynamical modes in major
  sub-basins
• Improved barotropic / baroclinic inputs at open
  boundaries
• Updated Climatology (NCEP version 6)
• Mean dynamic topography (Rio 5)
• Historical transports

5
RTOFS-Atlantic
4

• 2007 model updates conted
• Revise river inflow data (location and strength)
  from USGS
• Remove noise in net heat flux
• Update of model algorithms
• Surface initialized Montgomery Potential
• Employ two invariant external mode boundary
  conditions.
• Stabilization of sigma_star.
• Enforced salinity minimum by refreshing the water
  column

6
RTOFS-Atlantic
5

• 2007 model updates conted
• Major upgrades to data assimilation
• SST spatially varying bias removal algorithm
• SSH absolute sea surface height
• 2D 1D approach, conserving momentum.
• TS assimilation of vertical profiles of
  temperature and salinity
• 2D density, temperature and layer thickness
  anomaly, preserves volume, momentum updates
  mass and heat.
• New model output
• GRIB files on native grids.
• GRIB files sets for selected sub-regions.

7
RTOFS-Atlantic
6

• SSH without (upper-left panel) and with SSH
  assimilation (lower-left panel) from
  RTOFS-Atlantic
• Data JASON, GFO and RIO5 mean dynamic
  topography.
• U. of Colorado analyses. Uses also ENVISAT data.

8
RTOFS-Atlantic
7
Mean Gulf Stream path from old model (left) tends
to overshoot the annual mean path derived from
altimetry data near 72? W as compared to the
Gulf Stream location in the new implementation
(right)
9
RTOFS-Atlantic
8

• Product distribution
• FTP site (ftpprd)
• Nowcast and 5-day forecasts available for 3 days.
• NOMADS server
• Nowcasts and 5-day forecasts available for 30
  days. Tools to extract sub-regions available.
• Targeted to become operational in 2008.
• NODC Archives
• Long term archival. Under construction.

10
Coupled HWRF-HYCOM
1
IC/BC
HWRF O(9 km)
GFS O(25 km)
interaction
RTOFS (HYCOMBasin) O(417 km)
Hurricane (HYCOMRegional) O(814 km)
IC/BC
11
Coupled HWRF-HYCOM
2
12
Coupled HWRF-HYCOM
3

• Examples from Rita (2005).
• Model forced with GFS only.
• Fully coupled system is in testing mode.

13
Where to go

• Original plan called for regional West Coast and
  Hawaii models.
• We need a global model first
• Much time lost with boundary conditions of
  regional model.
• More essential element of national backbone
• High resolution global model can be backbone for
  all other NCEP requirement (such as HWRF).
• NCEP focus on US coast, eventually higher
  resolution around the US coast.
• Needs to fit in the larger picture.
• Will be HYCOM based for several years,

14
NCEP considerations
1

• For short-term ocean forecasts to have value for
  the NWS, the model needs to be eddy resolving
  (1/12 spatial resolution or better, Gulf Stream
  and eddy prediction). In this context, SSH
  assimilation is crucial.
• For ocean modeling in the context of HWRF
  hurricane modeling, the same holds true for the
  limited ocean domain used in such models. Nesting
  in similar large scale ocean models would appear
  beneficial.
• For coupled atmosphere-ocean modeling in the
  context of weather forecasting, lower resolution
  ocean models may be acceptable initially.
• Lower resolution ocean models have been used in
  climate modeling, the necessity of high
  resolution models for climate now is a topic of
  research.

15
NCEP considerations
2

• NCEPs operational Atlantic HYCOM system (the
  Real Time Ocean Forecast System, RTOFS)
• 1200x1684x26 grid points (1/12 spatial
  resolution).
• 8 nodes with 128 processor cores (IBM power 5
  mist/dew).
• 20min data assimilation, 23min for 24h nowcast
  and 107min for 5 day forecast.
• Running daily on the 00z model cycle only.
• This fits comfortably on our system, but we
  lack the appropriate computational power to
  provide new initial conditions for a model or
  model update in a reasonable time. This hampers
  development and makes model recovery tricky.

16
NCEP considerations
3

• A candidate for global model would be the Navy
  1/12 global model.
• 4500x3928x32 global grid with Arctic bipolar
  patch Base resolution is 1/12.
• 24 nodes with 379 HYCOM processes.
• 0.9 wall hours for NCODA analysis step, 0.9 wall
  hours per HYCOM forecast day (IBM power 5
  Kraken)
• NRL runs this system on a weekly basis.

Courtesy of Alan Wallcraft (NRL)
17
NCEP considerations
4

• NCEP could tentatively run an analysis cycle, a
  hindcast day and a single forecast day of this
  Navy model running in a time slot similar to
  RTOFS-Atlantic using the other three model cycles
  on mist/dew (8 nodes, 3h per cycle).
• NCEP cannot support initialization of this model
  or serious research using this model on the
  presently available computer resources at NCEP.
• Eddy resolving global ocean modeling at NCEPs
  present computer system is only feasible in a
  close partnership with other entities such as
  NRL, FSU, etc .....