JCSDA GPS RO ASSIMILATION Error Characteristics

1
JCSDA GPS RO ASSIMILATIONError Characteristics

• Martin S Lohmann

2
This presentation

• Observation errors (important to specify
  covariances)
• Assimilation strategies below superrefraction
  layers
• QC
• BUFR files at NESDIS/NCEP

3
RO Errors

• Above approx. 20-25 km
• Errors are dominated by background/ionospheric
  noise and errors in the 1. Guess
• Errors are not related to the neutral atmosphere
• Dynamic error estimation can be used
• From approx. 5 km to 20-25 km
• Errors are dominated by along track horizontal
  variations
• The large scale variations result in
  representativeness errors
• Small scale variations (turbulence) are
  measurement errors
• Dynamic error estimation?
• 0 to approx. 5 km
• Errors are dominated by along track horizontal
  variations and superrefraction
• (tracking errors - CHAMP)
• Dynamic error estimation?
• Superrefraction
• For non-local operators it is straightforward
  determine the corresponding measurement errors.
  In this case the height ranges indicated above
  are likely to change

4
Dynamic error statistics (20-40 km) vs. previous
studies
5
Refractivity errors for a single occultation
6
Bending angle errors for a single occultation
7
Strategy for error estimation below 20 km

• If possible use dynamic error estimation to
  increase impact of RO data
• Look for possible correlation between observation
  errors and some signal property, e.g. FSI phase
  or amplitude fluctuations. Also a useful approach
  for fine-tuning of QC
• For the First assimilation attempt we will use
  fixed error estimates based on (Kou et al. 2004)

Empirical relation
8
Strategies for handling superrefraction

• Refractivity
• Refractivity is biased below superrefraction
  layers
• Detect height of a likely superrefraction layer
  and discard observations or consider
  observations as a lower bound below that height
• Likely superrefraction layers can be detected
  from
• Measured N gradient
• Model N gradient
• Model-observation bias

• Local bending angle
• Measured bending angles are unbiased below
  superrefraction layers
• BUT the problem is ill-conditioned as different
  model states will correspond to the same bending
  angle profile
• Model resolves superrefraction ambiguity

9
QC Strategies, fine-tuning of QC criteria

• First assimilations will be based on current
  CDAAC QC
• Profiles with large deviations between background
  and observations will be rejected
• Profiles which pass the CDAAC QC, but fails the
  NCEP sanity check may point out weaknesses in
  CDAAC QC
• Profiles which pass the NCEP sanity check but are
  rejected by CDAAC QC, will indicate that the
  CDAAC QC could be to strict
• QC flags are very important

10
NCEP RO BUFR-izing

• Currently preparing NCEP BUFR files from CDAAC
  BUFR files
• (a) Prepare NCEP BUFR tables from CDAAC
  BUFR tables
• (b) Generate NCEP RO BUFR data files using
  NCEP BUFR tables
• (c) Test the generated NCEP BUFR FILES for
  consistency with
• the corresponding CDAAC BUFR FILES
• Intermediate plan to test assimilation of the
  NCEP BUFR files in the NCEP operational DAS
• Issues/Questions
• Are the CDAAC BUFR tables/files in final form ?
• May be dependent on UKMO
• Are the CDAAC BUFR files already tested for
  consistency with their parent NetCDF files ?

11
Summary and outlook

• Above 20 km, error covariances for both optimized
  bending angles and refractivity, including
  off-diagonal terms, are currently estimated as
  part of CDAAC
• Error Statistics based on these error estimates
  are found to be in good agreement with other
  studies
• The possibility of using dynamic error estimates
  below 20 km will be investigated in the future
• It is expected that bias caused by
  superrefraction can be avoided by assimilating
  bending angles instead of refractivity
• CDAAC QC will be fine-tuned based on feed-back
  from the first assimilations of CHAMP
  occultations
• NESDIS is currently preparing NCEP BUFR files
  from CDAAC BUFR files