ECMWF reanalysis using GPS RO data

1
ECMWF reanalysis using GPS RO data

• Sean Healy
• Shinya Kobayashi, Saki Uppala, Mark Ringer and
  Mike Rennie

2
Outline

• Background why are GPS RO measurements useful
  given that we already have millions (ATOVS, AIRS,
  IASI) of satellite radiance measurements
  available per day.
• Assimilation options for NWP and reanalysis
  Bending angle or refractivity.
• Reanalysis applications Some results from
  reanalysis experiments and the use of CHAMP GPSRO
  measurements in ERA Interim.
• Impact on bias correction of other sensors.
• Summary.

3
Background

• Why are GPSRO measurements useful for NWP and
  reanalysis, given that we already have millions
  of satellite radiances available?
• Globally distributed
• Good vertical resolution
• All-weather capability
• Can be assimilated without bias correction
• Quite simple to forward model (in comparison with
  radiances)

4
Information content studies GPSRO/IASI
Collard and Healy (2003)
5
Vertical resolution
6
Assimilation Options

• NWP centres tend to assimilate quantities that
  are as close to the raw measurement as possible.
• We do not assimilate temperature and humidity
  retrievals derived from GPSRO measurements.
• Experience with satellite radiances suggests that
  the error characteristics of retrievals are
  difficult to model.

7
Assimilation Options (2)

• Operational NWP centres assimilate either bending
  angle or refractivity. At ECMWF we assimilate
  bending angle profiles with a 1D operator. It
  evaluates
• The bending angles can be evaluated efficiently
  in terms of standard integrals (Gaussian Error
  function).
• Extrapolating NWP model above the model top.
  ECMWF model goes up to 80 km. The bending above
  the model top for a ray with a tangent height at
  40 km is 0.05 microradians.

Refractive index
Impact parameter
(xnr)
8
Assimilating refractivity

• Assimilating refractivity requires an additional
  step in the processing of the observation.
  Refractivity is derived with an Abel transform of
  the bending angle profile
• One of the problems of this approach is that the
  Abel transform requires bending angles to
  infinity.
• Processing centres have their own methods for
  extrapolating/smoothing the noisy bending angle
  profile, but most adopt what is called
  statistical optimization.

9
Statistical Optimization

• The bending angles used in the Abel transform are
  the weighted average of the observed values and
  bending angle values simulated with a climatology
  or NWP model (eg, MSIS, CIRA or ECMWF!).

Simulated BAs
Observed BAs
Error cov. matrix for simulated BAs
Error cov. matrix for observed BAs
statistically optimized bending angle.
10
REFRACTIVITY
Produced by Michael Rennie (Met Office)
Comparing CHAMP measurements processed at UCAR
and GFZ.
BENDING ANGLE
11
Statistical optimization (cont)

• The statistical optimization step introduces
  number of new degrees of freedom. How do we
  specify the error covariance matrices, chose the
  climatology/NWP information etc? This may be a
  significant factor in the refractivity
  differences above 25 km from different processing
  centres.
• Ultimately, we do not want to assimilate
  information from MSIS or CIRA into the ECMWF
  analysis/reanalysis, so bending angle
  assimilation is the preferred option.

12
Operational implementation of GPSRO at ECMWF
ECMWF started assimilating GPSRO data
operationally on December 12, 2006. Initial
implementation neutral in the troposphere, but
good improvement in the stratospheric temperature
scores. Clear improvement in the bias in
operational fit to radiosonde temperature
measurements.
13
Reanalysis applications ERA Interim

• After operation assimilation of GPSRO
  measurements we started considering use in
  ERA-Interim.
• ERA-Interim will cover the period 1989 to present
  day. The reanalysis group have reprocessed up to
  late 2002 so far. (A public website will be
  launched in March and there will be article in
  the next ECMWF newsletter.)
• Improvements since ERA-40 include
• Use of 4D-Var
• Variational Bias correction (VarBC) of satellite
  radiances
• Improved moisture analysis.

14
Reanalysis applications(Shinya Kobayashi)

• The first reanalysis experiments ran from July
  2002 to February 2003 using CHAMP measurements.
• Shinya investigated how GPSRO modifies the
  analyses produced with and without a recent
  correction he had introduced to the RTTOV Zeeman
  splitting (AMSU-A Channel 14 weighting function
  too high).
• The error in the RTTOV Zeeman splitting appears
  to have been the main source the wave-like bias
  in the mean temperature analyses over the poles
  in the operational and ERA analyses.

15
Use of GPSRO in ERA experiments(Shinya
Kobayashi)
Normalised (observation background) bending
angle departures in Polar regions. (normalised
with ob error)
GPSRO passive
GPSRO active
Red old RTTOV coefficients for AMSU-A Black new
RTTOV coefficients for AMSU-A
16
1228 New RTTOV, no GPSRO 1230 New RTTOV
GPSRO 1231 Old RTTOV , no GPSRO 1236 Old
RTTOV GPSRO
17
Value of GPSRO a satellite radiance bias
correction perspective

• The importance of GPSRO measurements was noted in
  the report by working group 3 at the
  ECMWF/EUMETSAT NWP-SAF workshop on bias
  correction in data assimilation (2005)
  (Proceedings available from ECMWF).
• The GPSRO measurements have clearly improved
  biases in the operational ECMWF stratospheric
  temperature analyses. Why - good vertical
  resolution and assimilated without bias
  correction.
• All satellite radiance measurements (AIRS, IASI)
  are bias corrected to the model. In fact, a good
  bias correction scheme is a key component for
  obtaining positive impact from these radiance
  data. ECMWF has an adaptive, variational bias
  correction (VarBC) model.
• VarBC can only work if other data (e.g.,
  radiosondes, GPSRO) are assimilated without bias
  correction. They are anchor points and they
  stabilise the VarBC scheme. VarBC assumes the NWP
  model is unbiased.
• GPSRO should improve the assimilation of
  satellite radiances by reducing model biases.

18
ECMWF/EUMETSAT NWP SAF Bias correction workshop
Working Group 3 Operational implementation of
bias correction

• Reference network
• Bias correction schemes need to be grounded by a
  reference. This is particularly important for
  adaptive schemes that may have the potential to
  drift. Therefore a network of observations
  approximating the true state of the atmosphere
  is required. The data must be compared to the NWP
  models, either passively to monitor any drift or
  actively to anchor the data assimilation system.
  
• Recommendation to the NWP centres to identify a
  part of the global observing system (e.g. high
  quality Radio-sondes, GPS Radio-Occultation) as
  reference network which is actively assimilated
  but NOT bias corrected against an NWP system.

19
VarBC Dee, QJRMS (2007), 131, pp 3323-3343

• Bias corrected radiances are assimilated.
• VarBC assumes an unbiased model.

In the 4D-Var, we minimize an Augmented cost
function, where the bias coefficients are
estimated.
20
Experiment in a simplified NWP system

• Period June 15th to August 31st, 2007.
• CONTROL Assimilates all conventional
  measurements AMSU-A and MHS measurements from
  the METOP-A satellite.
• COSMIC As control, but with all COSMIC
  measurements assimilated.
• How do the COSMIC measurements modify the
  evolution of the bias correction of AMSU-A
  radiances?

21
How the bias correction evolves for AMSU-A,
channel 9 (SH)
CONTROL (NO COSMIC MEASUREMENTS)
COSMIC MEASUREMENTS ASSIMILATED
22
Fit to radiosonde and COSMIC measurements (SH)
Radiosonde temperature
CONTROL red COSMIC black
Bending angle COSMIC-4 (normalised
O-B Departures)
23
Summary

• Tried to highlight the value of GPSRO
  measurements for NWP.
• Assimilate bending angle or refractivity. I would
  argue that bending angle is better because it
  circumvents the statistical optimization step.
• Reanalysis experiments Recent improvement/correct
  ion to the RTTOV model has improved the fit to
  CHAMP bending angle measurements. CHAMP is now
  being used in ERA-Interim.
• Assimilation experiments in a reduced NWP system
  have shown that the COSMIC bending angles
  anchor the bias correction for the AMSU-A upper
  tropospheric and lower stratospheric channels.

24
Temperature bias caused by aircraft (NH)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
q0
29
Bias in Dry temperature
Specific humidity (Kg/Kg)
30
20 km
12 km
8 km
5 km
31
Fig 3, Foelsche et al (2008), JGR