Lidia Cucurull UCARNOAAJCSDANCEP

1
Use of COSMIC observations in the NCEP/JCSDA data
assimilation system
Lidia Cucurull (UCAR-NOAA/JCSDA/NCEP) Jo
hn Derber, Russ Treadon
(NOAA/NCEP) James G. Yoe
(NOAA/NESDIS) Bill Kuo (UCAR)
2
Motivation

• To develop the total infrastructure (codes,
  scripts, etc.) necessary to monitor and
  assimilate radio-occultation (RO) observations at
  NOAA.
• Work schedule enables complete preparation of
  NCEP data assimilation system in time for COSMIC
  launch (estimated Dec 2005 or Jan 2006).

3
Getting COSMIC data to Weather Centers
This system is currently under development by
UCAR, NESDIS, UKMO
4
What has been done?

• Analysis of the differences between the different
  GPS processing centers to assure high quality RO
  retrievals at CDAAC.
• Comparison of retrievals of refractivity between
  the different processing centers and the NASA
  GMAO (formerly, DAO) analysis system (PSAS). Work
  done in collaboration with Paul Poli and Joanna
  Joiner. (Results presented at the GPS RO Data
  Users Workshop, MD, October 2003).
• Implementation of the local refractivity operator
  in the Gridpoint Statistical Interpolation (GSI)
  analysis system.
• Ability to ingest refractivity profiles in the
  system.
• Compute the innovation vector with CHAMP data
  (Forward Model).
• Tangent Linear and Adjoint codes (implemented and
  tested).
• Impact studies using a single observation, a
  single profile of refractivity and all the
  available profiles at a given analysis time.

5
GSI Analysis System

• Developed by NOAA/EMC and others JCSDA partners
  including NASA/GMAO.
• System is still under development.
• Planned to replace the NCEPs current operational
  Spectral Statistical Interpolation (SSI) analysis
  system and regional data assimilation system
  prior to availability of the COSMIC data.
• Characteristics
• The background error covariance matrix is defined
  in a grid space. This allows the definition of
  spatially varying covariance structures.
• T254 (nx512,ny256 for the linear gaussian grid)
  with 64 levels in the vertical (from surface to
  about 0.27 hPa). Experiments are conducted in a
  lower resolution grid T62 (nx192,ny94 for the
  quadratic gaussian grid) with 28 levels in the
  vertical.
• MPP code running on IBM-SP.

6
GSI Analysis System (II)

• The background or first guess is a 3,6,9 hour
  forecast.
• Analysis variables u-wind, v-wind, temperature,
  ln(surface pressure), specific humidity, ozone,
  surface skin temperature, and coefficients for
  bias correction for satellite radiance and
  precipitation data and for the background state.
• The analysis assimilates most conventional and
  satellite data.
• Analysis procedure
• 2 external iterations in the first one, the
  solution is linearized around the forecast. In
  the second one, the solution is linearized around
  the previous iteration.
• For each external iteration, there is a series of
  internal iterations to create the analysis. The
  forward model is applied to the solution (or
  forecast) interpolated in time and space to the
  observation locations.
• The current solution is iteratively updated by
  use of a nonlinear conjugate gradient algorithm
  to find the solution which minimizes the cost
  function.
• Linearization of the forward model.
• Adjoint of the forward model.

7
Forward Operator

• Model variables of pressure, temperature and
  specific humidity are interpolated to
  refractivity observation locations.
• Model refractivity is computed from the
  interpolated values.
• Tangent Linear Model consistency with Forward
  Model
• FM(x) Forward model acting on x
• FM(xDx)perturbed Forward model acting on
  xDx
• TL(x,Dx) Tangent Linear model acting on Dx
  (at x)

8
Tangent Linear Test

• Obs Iter Pos ratio Neg ratio
• 1 1 0.909090909 -0.909090909
• 1 2 0.952380952 -0.952380952
• 1 3 0.975609756 -0.975609756
• 1 4 0.987654321 -0.987654321
• 1 5 0.993788820 -0.993788820
• 1 6 0.996884735 -0.996884735
• 1 7 0.998439938 -0.998439938
• 1 8 0.999219360 -0.999219360
• 1 9 0.999609528 -0.999609528
• 1 10 0.999804726 -0.999804726
• 1 11 0.999902353 -0.999902353
• 1 12 0.999951174 -0.999951174
• 1 13 0.999975587 -0.999975587
• 1 14 0.999987793 -0.999987793
• 1 15 0.999993897 -0.999993897

9
Adjoint Test

• Abs(TL-AD)/TL lt MP
• Specific humidity
• 0.2633763648317E04 0.2633763648317E04
  -0.1726606528179E-15
• Temperature
• -0.5465696094271E00 -0.5465696094271E00
  -0.0000000000000E00

10
Single observation Test

• One random observation from a CHAMP profile.
• Coordinates lat70.216,lon125.38,height6.40km
  (p438mb)
• The observation error has been modified to
  account for representativeness error. The ratio
  (O-B)/? 1
• The horizontal and vertical structure of the
  increments are determined by the scales and
  structure of the background error covariance
  matrix.
• The impact is not only found in temperature and
  humidity. The balance constraint specified in the
  background error statistics drives increments on
  the wind field.

11
(No Transcript)
12
(No Transcript)
13
Single profile

• CHAMP profile located at lon259.288, lat-53.933

N Fractional difference
14
N fractional differences
15
Analysis Background
16
Profiles in 6hr assimilation window

• Analysis Time 2002080812
• 51 profiles from CHAMP
• Observations of refractivity rejected if (O-B)gt 5
  error(O)
• Experiments
• Assimilation of profiles of refractivity only.
• Assimilation of profiles of refractivity and all
  the conventional observations.
• Assimilation of conventional observations only.

17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
Outlook

• (1) QC and monitoring
• Develop (real time) Monitoring System (O-B) for
  calibration/testing of the radio occultation
  observations.
• QC checks (like account for SR).
• (2) Error
• Examine representativeness error.
• Adjustment of the background error covariance
  matrix according to the results of the
  assimilation of profiles of refractivity.
• (3) Experiments
• Compute statistics of the assimilation
  experiments with CHAMP.
• Assess the impact of the RO observations of
  refractivity when combined with all other
  available observations.
• Superob (average in the simplest case) vertical
  levels appropriate to model vertical resolution.
• Test other Forward Models
• Local bending angle
• Non-local linear observational operator
  (bending?, phase?)
• OSSE Studies.