simulation studies on the analysis of radio occultation data
DESCRIPTION
2nd GRAS SAF User Workshop Helsing ø r, Denmark, June 11-13, 2003. Simulation Studies on the Analysis of Radio Occultation Data. Andrea K. Steiner, Ulrich Foelsche, Andreas Gobiet, and Gottfried Kirchengast Institute for Geophysics, Astrophysics, and Meteorology - PowerPoint PPT PresentationTRANSCRIPT
Simulation StudiesSimulation Studies
on theon the
Analysis of Radio Occultation DataAnalysis of Radio Occultation Data
Andrea K. Steiner, Ulrich Foelsche, Andreas Gobiet, and Gottfried Kirchengast
Institute for Geophysics, Astrophysics, and MeteorologyUniversity of Graz (IGAM/UG), Austria
2nd GRAS SAF User WorkshopHelsingør, Denmark, June 11-13, 2003
2003 by IGAM/UG
Simulation Studies on the Analysis of RO DataSimulation Studies on the Analysis of RO Data
OutlineOutline
Properties and Utility of RO Data
End-to-end Simulations of GNSS RO Data
- Atmosphere and ionosphere modeling
- Observation simulations
- Retrieval of atmospheric variables
Simulation Studies
- Empirical error analysis
- Climate monitoring simulation study 2001-2025
- GNSS RO retrieval scheme in the upper stratosphere
- Representativity error study (focus on troposphere)
Summary, Conclusions and Outlook
Simulation Studies on the Analysis of RO DataSimulation Studies on the Analysis of RO Data
Properties andProperties and Utility of RO DataUtility of RO Data
The RO method provides a unique combination of• global coverage (equal observation density above oceans as above land)
• all-weather capability (virtual insensitivity to clouds & aerosols; wavelengths ~20 cm)
• high accuracy and vertical resolution (e.g., T < 1 K at ~1 km resolution)
• long-term stability due to intrinsic self-calibration (e.g., T drifts < 0.1 K/decade)
GNSS Radio occultation observations
• are made in an active limb sounding mode
• exploiting the atmospheric refraction of GNSS signals
• providing measurements of phase path delay for the retrieval of
• key atmospheric/climate parameters such as temperature and humidity.
This is the basis for the utility of RO Data for
• global climate monitoring
• building global climatologies of temperature and humidity
• validation and advancement of climate modeling
• improvement of numerical weather prediction and analysis
Realistic modeling of the neutral atmosphere and ionosphere
ECMWF analysis fields T213L50, T511L60; ECHAM5 T42L39 NeUoG model
Realistic simulations of radio occultation observations Receiver: GNSS Receiver for Atmospheric sounding GRAS LEO satellite: METOP European Meteorological Operational satellite
6 satellite constellation (COSMIC, ACE+ type)
Calculation of excess phase profiles Forward modeling with a sub-millimetric precision 3D ray tracer Observation system simulation including instrumental effects and the raw
processing system
Retrieval of atmospheric profiles in the troposphere and stratosphere dry air retrieval, optimal estimation retrieval (1DVAR) in the troposphere
Simulation tool is the End-to-end GNSS Occultation Performance Simulator
EGOPS (developed by IGAM/UniGraz and partners)
Simulation Studies on the Analysis of RO DataSimulation Studies on the Analysis of RO Data
End-to-end Simulations of GNSS RO DataEnd-to-end Simulations of GNSS RO Data
Empirical Error AnalysisEmpirical Error Analysis Study DesignStudy Design
• Observation day: September 15, 1999
• METOP as LEO satellite with GRAS
receiver
• GPS setting and rising occultation events
• Height range: 1 km to 90 km
• 574 events total
• 300 events globally chosen for study
equally distributed in space and time
• 100 events in each of 3 latitude bands
- low latitudes: -30° to +30°
- mid latitudes: ±30° to ±60°
- high latitudes: ±60° to ±90°
Simulated observables are phase delays and amplitudes
– Phase delays for the GPS carrier signals in L band: L1 (~1.6 GHz), L2 (~1.2 GHz)
– Atmospheric phase delay (after correction for ionosphere): LC (illustrated above)
– LC phase rms error of ~2 mm at 10 Hz sampling rate conservatively reflects
METOP/GRAS-type performance
~ 1 mm Mesopause
~ 20 cm Stratopause
~ 20 m Tropopause
~ 1 – 2 km Surface
Empirical Error AnalysisEmpirical Error Analysis Simulated ObservablesSimulated Observables
• Interpolation of retrieved (xretr) and ‘true’ co-located (xtrue) atmospheric profiles
to a L60 vertical grid with the uppermost level at ~65 km/0.1 mbar
(inspection at levels 900 mbar < p < 0.75 mbar; 1 km < z < 50 km)
• Difference profiles:
• Bias:
• Bias-free profiles:
• Error Covariance Matrix:
• Standard Deviation:
• Correlation Matrix:
jjii
ijij
ii
Tbiasfreek
biasfreek
kbiasfreek
k
trueretr
SS
SR
S
xxN
xx
kxN
xxx
with
1
1
Events of .No,1,1
R
s
S
b
b
Empirical Error AnalysisEmpirical Error Analysis Error Analysis MethodError Analysis Method
Relative StdDev:8 < h < 35 km: 0.3% – 1%
3 < h < 8 km: < 8%
h > 35 km: < 5%
Relative Bias: 5 < h < 38 km: < 0.1%
5 > h > 38 km: < 0.5%
Covariance Matrix Model: Sij = s2 exp(-|zi-zj|/L)
RReell.. SSttddDDeevv ss CCoorrrr.. LLeennggtthh LLzz << zz__ttrrooppoo a1 + a2*x-3 a1 = 0.140
a2 = 483.20.5 km
zz >> zz__ttrrooppoo a3*exp(a4*x) a3 = 0.043a4 = 0.093
0.5 km
Empirical Error Analysis Bending Angle Error - MSIS StatOptBending Angle Error - MSIS StatOpt
Relative StdDev:
5 < h < 40 km: 0.1% – 0.75%
5 > h > 40 km: < 2%
Relative Bias:
2.5 < h < 40 km: < 0.1%
h > 40 km: < 0.3%
Covariance Matrix Model: Sij = s2 exp(-|zi-zj|/L)
RReell.. SSttddDDeevv ss CCoorrrreellaattiioonn LLeennggtthh LLzz << zz__ttrrooppoo a1 + a2*x-1 a1= -0.221
a2 = 4.4612 km
zz >> zz__ttrrooppoo a3*exp(a4*x) a3 = 0.019 a4 = 0.087
linearly decreasing from 2 kmat z_tropo to 1 km at z = 60 km
Empirical Error Analysis Refractivity ErrorRefractivity Error
Summer seasons (JJA) during 2001 to 2025
ECHAM5-MA with resolution T42L39 (64x128 grid points, 2.8°resolution)
6 LEO satellites, 5x5yrs
Dry air temperature profiles retrieval in the troposphere and stratosphere to establish a set of realistic simulated temperature measurements.
An statistical analysis of temporal trends in the “measured” states from the simulated temperature measurements (and the “true” states from the modeling, for reference).
An assessment of how well a GNSS occultation observing system is able to detect climatic trends in the atmosphere over the coming two decades.
Testbed for setup of tools and performance analysis: JJA 1997
Objective is to test the capability of a small GNSS occultation observing
system for detecting temperature trends within the coming two decades
Climate Monitoring Simulation StudyClimate Monitoring Simulation Study
Study DesignStudy Design
Atmosphere model: ECHAM5-MA (MPIM Hamburg) Model resolution: T42L39 (up to 0.01hPa/~80km)Model mode: Atmosphere-only (monthly mean SSTs)Model runs: 1 run with transient GHGs+Aerosols+O3
1 control run (natural forcing only)
Change monitoring: In JJA seasonal averageT fields as they evolve from 2001 to 2025Domain: 17 latitude bins of 10 deg width
34 height levels from 2 km to 50 km vertical resolution 1 – 2 km core region 8 km to 40 km
Date: July 15, 1997; UT: 1200 [hhmm]; SliceFixDim=Lon: 0.0 [deg] Mean T field in selected domain: “True” JJA 1997 average temperature
Climate Monitoring Simulation StudyClimate Monitoring Simulation Study
Atmosphere ModelingAtmosphere Modeling
Ionosphere model: NeUoG model (IGAM/UG)Model type: Empirical 3D, time-dependent, sol.activity-dependent modelMode: Driven by day-to-day sol.act. variability (incl. 11-yrs solar cycle, etc.)
Solar activity prescription: Representative day-to-day F107 values (weekly history averages)Future F107 data (2001-2025): from past data of solar cycles 21, 22, and 23 (1979-1999)
Month: July; UT: 1200 [hhmm]; SAc/F107: 120; SliceFixDim=Lon: 0.0 [deg] Solar activity 1996-2025: day-to-day F107 values and monthly mean values
Climate Monitoring Simulation StudyClimate Monitoring Simulation Study
Ionosphere ModelingIonosphere Modeling
Sampling into 17 equal area latitude Bins – 85°S to 85°N (10°lat x 15°lon at equator) – No. of occultation events > 50 per Bin for each JJA season (max. 60/Bin)
No. of occultation events per Bin and month – light gray: June events only – light&medium gray: June+July events – light&medium&dark gray: June+July+August
Climate Monitoring Simulation StudyClimate Monitoring Simulation Study
Observation Simulations - Spatial SamplingObservation Simulations - Spatial Sampling
Typical example of T profile errors (~50 events)
Retrieval of 50-60 Tdry air profiles per latitude Bin• Temperature errors < 0.5 K within upper troposphere and lower stratosphere for individual T profiles• Errors in TAv for ~50 events < 0.2 K (8 km < z < 30 km)
Climate Monitoring Simulation StudyClimate Monitoring Simulation Study
Temperature Profiles - Temperature TrendsTemperature Profiles - Temperature Trends
Temperature trends estimation
• using TJJA Av
• Time period 2001 to 2025
• Latitude x height slices (17 x 34 matrix)
Detection tests on temperature trends
• in the model run with transient forcings
• in the control run for comparison
• relative to estimated natural variability
Bias error in temperature climatology Total observational error
2
12
2
N
TTT
stddevijbias
ijobsij
true
jretrj
ii
biasij TT
NInterpT
1
Climate Monitoring Simulation StudyClimate Monitoring Simulation Study
Performance analysis: Observational errorPerformance analysis: Observational error
Sampling error for the selected events• Difference between the “sampled” JJA
average T field (from the “true” T profilesat the event locations) and the “true” one
• ~55 selected events per Bin (total ~1000)
Sampling error if all events used• Difference “sampled”-minus-“true” JJA
average T field using all occultationevents available in the Bins
• ~750 events per Bin (~13 000 in total)
Climate Monitoring Simulation StudyClimate Monitoring Simulation Study
Performance Analysis: Sampling ErrorPerformance Analysis: Sampling Error
2
122
sam
ijobsij
totalij TTT
Total climatological error (observational plus sampling error)
Climate Monitoring Simulation StudyClimate Monitoring Simulation Study
Performance Analysis: Total Climatological ErrorPerformance Analysis: Total Climatological Error
Total climatological error for all eventsTotal climatological error for selected events
• GNSS occultation based JJA T errors are expected to be < 0.5 K in most of the core region (8–40 km) northward of 50°S.
• 2001–2025 JJA T trends are expected to be > 0.5 K per 25 yrs in most of the core region northward of 50°S. (ECHAM4 T42L19 GSDIO experiment)
Significant trends (95% level) expected to be detectable within 20 yrs in most of the core region Aspects to be more clearly seen in the long-term: ionospheric residual errors, sampling errors, performance southward of 50°S (high-latitude winter region)
Exemplary simulated temperature trends 2001–2025
Climate Monitoring Simulation StudyClimate Monitoring Simulation Study
Perspectives for the Full Experiment 2001-2025Perspectives for the Full Experiment 2001-2025
Total climatological error of test-bed season
GNSS RO retrieval scheme in the upper stratosphereGNSS RO retrieval scheme in the upper stratosphere
Empirical Background Bias CorrectionEmpirical Background Bias Correction
• Background data: bending angle derived from MSISE-90 model
• Error covariance matrices:
Background B: 20% error, exponential drop off with correlation length L = 6 km
Observation O: rms deviation of o from b between 70-80 km, L = 1 km
• Basic scheme: Search the best fit bending angle profile in the climatology
• Advanced scheme: Linearly fitting of the background to the observation in addition to the basic scheme (background B: 15% error)
• Result: In general the effect of fitting is small - background bending angles are modified by < 1%, negligible effect on temperature profiles. In extreme cases background bending angles are modified up to ~15%, seen in temperature profiles (1 K level) down to 20 km.
• Method: Inverse covariance weighting statistical optimization of observed bending angle o with background bending angle b
)()( 1bobopt ααOBBαα
GNSS RO retrieval scheme in the upper stratosphereGNSS RO retrieval scheme in the upper stratosphere
Test-bed Results with Advanced RetrievalTest-bed Results with Advanced Retrieval
Enhanced background bias correction:Inverse covariance weighting optimization with search & fitError reduction in the southern high latitudesand above 30 km.
Basic scheme:Inverse covariance weighting optimization with searchBackground MSISE-90
Mean dry temperature bias of GNSS CLIMATCH test-bed season
Representativity Error StudyRepresentativity Error Study
Study DesignStudy Design
Azimuth Sectors
– Sector 1: 0° < |Azimuth| < 10° – Sector 2: 10° < |Azimuth| < 20° – Sector 3: 20° < |Azimuth| < 30° – Sector 4: 30° < |Azimuth| < 40° – Sector 5: 40° < |Azimuth| < 50°
581 occ. events in total (1 day MetOp/GRAS), ~100 in each sector, during 24 hour periodECMWF analysis field T511L60 (512x1024)
Reference Profiles - vertical vs tangent point trajectories
Representativity Error StudyRepresentativity Error Study
Tangent Point TrajectoriesTangent Point Trajectories
Occultation events are never vertical Average elevation angle in the height interval 2-3 km: Sector 1: 6.6°, Sector 3: 4.9°, Sector 5: 3.2°
Representativity Error StudyRepresentativity Error Study
Temperature Errors as ExampleTemperature Errors as Example
Vertical ReferenceProfile
Retrieved3D TangentPoint Trajectory
“True” 3D Tangent Point Trajectory
Retrievedminus “True”3D TangentPoint Trajectory
All Events All Events
All Events All Events
Simulation Studies on the Analysis of RO DataSimulation Studies on the Analysis of RO Data
Summary,Summary, Conclusions and Outlook (1)Conclusions and Outlook (1)
An empirical error analysis of realistically simulated RO data provides errorcharacteristics for key atmospheric variables. Simple analytical functions for covariance matrices were deduced for bending angle and refractivity, which can be used as total observational error covariance matrices for data assimilation systems.
A representativity error study shows that the comparison of RO profiles with vertical reference profiles introduces large representativity errors, especially in the lower troposphere. The average zenith angle of the tangent point trajectory near the Earth’s surface is about 85°. Errors decrease significantly if the retrieved profiles are compared to reference profiles along a tangent point trajectory deduced purely from observed data.
An advanced GNSS RO retrieval scheme in the upper stratosphere was developed including background profile search and empirical background bias correction. It was successfully tested with simulation data and is currently under evaluation with CHAMP data.
Simulation Studies on the Analysis of RO DataSimulation Studies on the Analysis of RO Data
Summary,Summary, Conclusions and Outlook (2)Conclusions and Outlook (2)
A climate monitoring simulation study for the years 2001-2025 is ongoing. The preliminary results for the test-bed season suggest that the expected temperature trends over the coming two decades could be detected in most parts of the upper troposphere and stratosphere.
Based on our simulation studies we aim to built first real RO based global climatologies from the CHAMP and SAC-C missions.
Current multi-year single RO sensors such as on CHAMP, SAC-C, GRACE, and METOP are important initial components for starting continuous RO based climate monitoring. As a next step, constellations like COSMIC and ACE+ need to be implemented with high priority.