Welcome to the UW Fox Scholars Series

1
GSI The Next Generation, Unified Data
Assimilation System at NCEP
Daryl Kleist
National Centers for Environmental
PredictionEnvironmental Modeling
CenterScience Applications International
2
Thanks to Collaborators

• NCEP
• John Derber, Lidia Cucurull, Dave Parrish, Manuel
  Pondeca, Jim Purser, Russ Treadon, Paul vanDelst,
  Wan-Shu Wu, and others
• GMAO
• Ricardo Todling, Ron Errico, Runhua Yang, Ron
  Gelaro, Yanqiu Zhu, Wei Gu, and others

3
GSI History

• GSI Gridpoint Statistical Interpolation
• The GSI system was initially developed as the
  next generation global analysis system
• Wan-Shu Wu, R. James Purser, David Parrish
• Three-Dimensional Variational Analysis with
  spatially Inhomogeneous Covariances. Mon. Wea.
  Rev., 130, 2905-2916.
• Originally based on SSI analysis system
• Replace spectral definition for background errors
  with grid point representation
• Allows for anisotropic, non-homogenous structures
• Allows for situation dependent variation in
  errors
• Can easily incorporate things as a function of
  topography, water vs. land points, etc.

4
GSI History

• After initial global GSI development, EMC
  management expressed desire for single
  global/regional analysis system
• Simplify exchange of ideas / developments between
  global and regional applications
• Thus, current GSI is an evolutionary combination
  of the global SSI analysis system and the
  regional ETA 3DVAR
• Supports WRF, NCEP, and other infrastructures
• Eventual transition to ESMF

5
GSI Community Collaboration

• Growing number of collaborators / users from
  outside of NCEP (25 registered groups, 60 known
  users)
• NASA GFSC (GMAO), MSFC
• FSL, NESDIS, NCAR
• University of Hawaii, Miami, Oklahoma, Utah,
  Wisconsin
• Periodic updates based on submissions from
  developers (both internal to EMC, and outside
  developers)
• Designed to support global mesoscale forecast
  systems, rapid update cycles, surface analysis,
  and various research problems

6
Overview of GSI

• 3d variational assimilation based on the SSI . .
  .
• Formulated in physical space
• Analysis (Control) Variables
• ?, ?u, Tu, q, Psu, O3 (not ?, ?)
• Multivariate relation differs (non-trivial to use
  things like full nonlinear balance equation on a
  grid)
• Grid point definition of background error
• Spectral definitions replaced with recursive
  filters
• Vertical EOFs replaced with recursive filters
• Background error statistics are a function of
  height and latitude
• Improved efficiency, coding, and documentation

7
Moisture analysis

• Pseudo-relative humidity (Dee and Da Silva, 2002)
• Normalize specific humidity by guess (background)
  saturation specific humidity
• Univariate moisture analysis
• Normalized relative humidity (Holm et al., 2002)
• ?RH / ?(RHb) RHb (?P/Pb ?q /qb - ?T /?b )
• ?(RHb) standard deviation of background error
  as function of RHb
• ?b -1 / ?(RH)/ ?(T)
• multivariate relation between moisture,
  temperature, and pressure

8

• Option 1 univariate
• temperature increment forces large increment in
  RH

• Option 2 multivariate
• temperature increment
• forces increment in q
• much smaller RH increment

9
Multivariate Relation (balance)
Tb G? ?b c? Psb W?
Projection of ? at vertical level 25 onto
vertical profile of balanced temperature (G25)
Percentage of full temperature variance explained
by the balance projection
10
Background Error Covariance (?)
a
b
c

• Standard deviation of variance (interval 0.5 x
  106 m2 s-1)
• Horizontal length scale estimate (interval 100
  km)
• Vertical length scale estimate (interval 1 grid
  unit)

11
Single Observation Analysis
Single zonal wind observation (1.0 ms-1 O-F and
error)
Cross Section at 180o
u increment (black,interval 0.1 ms-1 ) and T
increment (color, interval 0.02 K) from SSI
u increment at (black, interval 0.1 ms-1 ) and T
increment (color, interval 0.02K) from GSI
12
Dynamic Constraint (Jc) in GSI

• 3dVAR Penalty Function
• J Jo Jb Jc
• Penalty term added based on time tendencies
• Attempting to reduce noise and improve upon
  balance in analysis
• Broken into high and low order components
• Based on dynamical initialization ideas of
  Bratseth (1989)

13
Dynamic Constraint (incremental) in GSI

• Low order components approximate vertically
  averaged motions
• High order components represent deviations from
  vertical averages
• Using a proxy for time tendency of potential
  temperature
• No low order term
• Pressure correction in high order term

14
Zonally Averaged Tendencies

• Blue guess (six hour model forecast)
• Red analysis with no Jc term
• Green analysis with Jc term on

15
Improved Forecast Skill With Jc

• y -- control
• j with Jc term

16
Too much of meteorological signal removed with
Jc ?
17
Second Attempt at Incremental Dynamic Constraint

• Designed in attempt to operate predominantly on
  the highest frequency components
• SSI has divergency tendency part built in
  (non-operational)

18
Single Ob Results new Jc
19
New Dynamic Constraint

• Seems to have similar improvement in forecast
  skill as was seen with first attempt at adding Jc
  term, without impact larger scales as much
• Caviat small sample so far, work in progress
• Small improvements in precipitation verification ?

20
GSI development Observation errors

• Improved specification of observational errors
• Plan to examine situation dependent
  representativeness errors
• Will increase granularity in the specification of
  observation errors
• For example, all sonde data has same observation
  error independent of sonde type.
• Could (should) vary error as function of sonde
  type
• NCO has found that acars biases are strongly
  equipment dependent
• Adaptive Tuning
• Success in redefining ob errors in regional, may
  extend to global

21
Example from Adaptive Tuning of Observational
Errors
22
GSI development Background errors

• New methods for estimating background error
• Ensemble (Monte Carlo)
• Seems to provide better estimate than NMC,
  especially for regional
• Function fitting or correlation profiles to
  approximate multiple length scale estimates
• Bring computation on-line, adaptive ?
• Anisotropic, situation dependent background
  errors
• 2-dvar capability currently exists in GSI
• Will be used for regional (US) surface analysis
• Extending to full 3d capability, both globally
  and regionally

23
Ridiculously Preliminary Statistics from Ensemble
Estimate
a
b
c

• Variances much smaller
• Horizontal length scale estimates similar
• Vertical length scale estimates smaller
• Balance projections are similar (not shown)

24
Anisotropic vs Isotropic Error Covariances
Error Correlations Plotted Over Utah Topography
Observation influence extends into mountains
indiscriminately
Observation influence restricted to areas of
similar elevation
25
More GSI Development

• New Data
• GPS, AVHRR imagers, SSM/I radiances, AMSR-E
  radiances, SSM/IS, Level 2 Doppler Radar
• Using incremental time tendencies as part of
  control vector in GSI to use observations better
  at correct times
• New CRTM
• Variational quality control
• Model/Guess bias correction
• Integrate surface (including SST) analysis as
  part of atmospheric analysis

26
Status

• Currently GSI producing
• similar quality forecasts in NH and SH and better
  in tropics than SSI for global system
• Superior forecasts for Regional/Mesoscale
  analysis
• Test analyses for surface analysis
• WRF-GSI
• Target June 2005
• GFS-GSI (hybrid)
• Tentative late 2006 / early 2007 (post hurricane
  season, significant work in progress, lots of
  unresolved questions)
• GMAO GEOS5
• Ruc-GSI

27
GSI Parallel vs GFS (6 day)
28
GSI Parallel vs GFS (5 day)
29
GSI Parallel vs GFS (Verification)
30
GSI-Hybrid Day 5 Verification
31
GSI-Hybrid Tropical Wind Vector RMS
32
Precipitation Verification
33
Other GSI Related Developments

• GMAO Re-analysis
• currently performing sweeper run at low
  resolution
• TLM/Adjoint of GSI Development
• Have a version ready, but based on a version of
  GSI from March 2005
• Ready to link up to adjoint of GMAO fv-model

34
Analysis Sensitivity
35
Where to next ?

• First and foremost, we need to take advantage of
  and expand GSI
• Anisotropic, flow-dependent background error
• More intelligent data thinning/selection
• Coupled surface/atmosphere (and ocean) analysis
• More analysis variables, such as precipitation,
  clouds, aerosols
• Strengthen, enhance current collaborations, while
  developing new ones as well
• More involved with university researchers
• 4dVAR ?
• LETKF ?