Design and Validation of the GMAO OSSE
Ronald M. ErricoNikki Prive’
King-Sheng Tai
Progress report presented to the GMAO 23 Feb. 2012
Acknowledgements
Runhua Yang Meta Sienkiewicz Jing Guo Ricardo Todling Will McCartyArlindoda Silva Joanna Joiner Joe Stassi Ops Group et al.
Outline:
1.Methodology2.Assimilation metrics3.Forecast metrics4.Application: Characterization of analysis error5.Application: Consideration of future instruments 6.Summary
Time
Analysis AnalysisAnalysis
Analysis AnalysisAnalysis
Real Evolving Atmosphere, with imperfect observations. Truth unknown
Climate simulation, with simulated imperfect “observations.” Truth known.
Observing System Simulation Experiment
Data Assimilation of Real Data
Applications of OSSEs
1. Estimate effects of proposed instruments (and their competing designs)on analysis skill by exploiting simulated environment.
2. Evaluate present and proposed techniques for data assimilation by exploiting known truth.
Validation of OSSEs
Compare a variety of statistics that can be computedin both real assimilation and OSSE contexts.
These statistics vary in their importance and in their abilityto be tuned in the OSSE context.
Methodology
ECMWF Nature Run
1. Free-running “forecast” 2. Operational model from 20063. T511L91 reduced linear Gaussian grid (approx 35km)4. 10 May 2005 through 31 May 20065. 3 hourly output6. SST and sea ice cover is real analysis for that period
GEOS-5.7.1p2 (GSI 3DVAR)Resolution 0.5x0.625 degree grid, 72 levels (eta coordinate) Evaluation for 1-31 July 2005Spin-up starts15 June 2005 from a previous 2-day accelerated spin upConventionalobservations include: prepbufrwithout GOES radiances or precipitation observationsRadiance observation include: HIRS-2 (N14), HIRS-3 (N16,17), AMSU-A (N15,16,AQUA; without channel 1-3,15), AMSU-B (N15,16,17),AIRS (AQUA), MSU (N14)Radiance bias correction initialized with OSSE spun up file
Assimilation System
Simulation of Observations
1. All observations created using bilinear interpolation horizontally, log-linear interpolation vertically, linear interpolation in time2. Radiance observations created using CRTM version 1.23. Clouds and precipitation are treated as elevated black bodies4. No use of NR snow coverage5. Locations for all “conventional” observations given bycorresponding real ones, except no drift for RAOBS 6. SATWNDS not associated with trackable features in NR
Probability of significant cloud
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Cloud fraction f
Pro
bab
ility
P(f
)
Dependency of results on errorsConsider Kalman filter applied to linear model
Daley and Menard 1993 MWR
Simulation of Explicit Observation Errors
1.Some representativeness error implicitly present2.Gaussian noise added to all observations3.AIRS errors correlated between channels4. Observational errors for SATWND and non-AIRS radiances horizontally correlated (using isotropic, Gaussian shapes)5. Conventional soundings and SATWND observational errors vertically correlated (Gaussian shaped in log-p coordinate)6. Tuning parameters are error standard deviations, fractions ofvariances for correlated components, vertical and horizontal length scales
AMSU-A NOAA-15 RAOB
RAOB GOES-IR Winds
Standard deviations ofObservation errors
In GSI error tables (solid circles or lines)
vs.
For adding obs. errors (open circles or dashed lines)
Assimilation Metrics
Locations of QC-accepted observations for AIRS channel 295 at 18 UTC 12 July
Real
Simulated
Standard deviations of QC-accepted O-F values (Real vs. OSSE)
AIR AQUAAMSU-ANOAA-15
RAOBU Wind
RAOB q
Horizontal correlations of O-FAMSU-A NOAA-15 Chan 6 GLOB RAOB T 700 hPa NHET
GOES-IR SATWND 300 hPa NHET GOES-IR SATWND 850 hPa NHET
Square roots of zonal means of temporal variances of analysis increments
T OSSET Real
U OSSEU Real
Time meanAnalysis increments
T 850 hPa
U 500 hPa
OSSE Real
Forecast Metrics
Real ControlOSSE Control
Mean anomaly correlations
OSSE vs Real Data: Forecasts
Real ControlOSSE Control
January Forecasts
U-Wind RMS error: July
Real ControlOSSE Control
Solid lines: 24 hour RMS error vs analysisDashed lines: 120 hr forecast RMS error vs analysis
T RMS error: July
Solid lines: 24 hour RMS error vs analysisDashed lines: 120 hr forecast RMS error vs analysis Real Control
OSSE Control
Real OSSE
Real OSSE
Real OSSE
Impact of obs errors on forecast
July Adjoint: dry error energy norm
July Adjoint: dry error energy norm
Analysis and Forecast ErrorVs. Nature Run
RMS OSSE Error vs NR, Temperature
RMS Error vs NR, U wind
400 hPa analysis error, U (m/s)
July mean error vs NR
A-B at 400 hPa
Application: Characterization of analysis error(as an example of the kinds of calculations that can be performed)
Square roots of zonal means of temporal variances of U wind analysis error
Fractional reduction of zonal means of temporal variances of analysis errors compared with background errors
uT
Horizontal spectra of analysis and analysis error (Spectra of time-mean fields subtracted)
Rotational Wind 200 hPa Divergent Wind 200 hPa
KE(J/kg)
KE(J/kg)
wave number n wave number n
Horizontal correlation length scales for v wind(Explicit bkg error vs. NMC method estimate)
South Pacific Tropical Pacific
Application: Consideration of future instruments
ESA Aeolus• Active measurements of
wind from space– Measuring the Doppler shift
imposed on lidar backscatter
• Launch delayed till late 2013
• 408 km, dawn-dusk, sun-synchronous orbit
• Wind measurements 90° off-track, 35° off-nadir– Nearly u @ equator
• No spaceborne heritage for DA development
• Utilize OSSE framework to generate realistic Aeolus proxy data – Lidar Performance Analysis
Simulator (LIPAS), via KNMI (G.-J. Marseille & A. Stoffelen)
• Need proper sampling (spatial & vertical), yield, and errorcharacteristics
Assimilation Results
Zonal Wind RMS Difference (ms-1)
Large reductions of RMS seen over tropics•Plots show reduction in analyzed u-wind RMS (compared to the NR as truth) by adding Aeolus observations to GMAO OSSE– Negative values indicate
improvement•Existing Observations primarily of mass field (passive satellite radiances)•Winds cannot be inferred through balance assumptions
Assimilation Results• Impact at 200 hPa consistent
through troposphere• Less impact towards surface
– Less observations– Increased contamination
• Lesser impacts in N./S. hemispheres – Winds through mass-wind
balance• Less impact in mass fields,
but still generally positive (not shown)
Tropics (-30º to 30º)NH Extratropics (30º to 90º)SH Extratropics (-90º to -30º)
Conclusions, Caveats, and Future Efforts• These results are overstated!– Error in simulated DWL observations inherently understated
• No added representativeness error– Error in Yield
• Lack of clouds in NR not accounted for in DWL obs simulation
• Shows expected result that largest impact is expected in tropics
• Future work – Include incorporation of L2B processing into DA System (GSI)
• Software release delayed due to change in instrument operations• A necessary level of processing for use of data in near-realtime• Will be included as part of outer loop processing
– Redo experiment w/ simulator updated for change in instrument ops• Simulator update in progress @ KNMI
Summary
Summary
1. Fairly easy to match O-F covariances2. Harder to match analysis increment statistics 3. Hardest to match forecast error metrics 4. Present OSSE framework validates reasonably well, but …5. Some correctable deficiencies6. Some puzzles
Continuing Development
1. New SATWND observation location determination based on NR2. Fix surface wind observation simulator3. Improve specification of land surface parameters for CRTM4. Additional radiance observations5. GPS observations6. More complete tuning of conventional observation errors7. Estimation of model error in OSSE framework8. Use of other NR models9. Demonstration of OSSE applications10. Further examination of NMC method for estimating B
Extras
‘Perfect’ Initial Conditions
• A grid of soundings with no error is ingested into the GSI– One sounding every other latitude and longitude from NR– Soundings extend to the top of the atmosphere– Ingested as type RAOB (120/220)– Intention is to drive the OSSE toward a ‘perfect’ initial
condition
• Cycled from 15 Dec to 30 January• Forecasts daily at 00Z
U-Wind RMS error vs NR
OSSE PerfectOSSE Control
Solid lines: analysis RMS errorDashed lines: 120 hr forecast RMS error
T RMS error vs NR
Solid lines: analysis RMS errorDashed lines: 120 hr forecast RMS error
OSSE PerfectOSSE Control
OSSE PerfectOSSE Control
Top Related