22nd October D4L2 Kerridge 1
Data Comparison Methods
Lecture by B.Kerridge
Rutherford Appleton Laboratory, UK
ESA-MOST DRAGON 2 PROGRAMMEAdvanced Training Course in Atmospheric Remote Sensing
19-24th October 2009, Nanjing University
22nd October D4L2 Kerridge 2
Contents
1. Introduction
2. Radiances
3. Ozone
4. Aerosol
5. Surface networks
6. Summary
22nd October D4L2 Kerridge 3
1. Introduction– Why are data comparisons needed?
1. Comparison of observation with theory is fundamental.2. Validation of satellite data is also essential
→ inform users of data attributes and quality
– Which approaches are used?
– Quantitative comparison of new data set with existing data sets of established attributes and quality (ie resolution, precision & accuracy)
– Spatial & temporal distributions and variances– Biases and standard deviations between new and established data– Sensor-to-sensor, sensor-to-model or -analysis
– What level of stringency is necessary (ie precision and accuracy)?
– Atmospheric variability ← lifetime– Quality of established observing techniques
• What (scientific or other) value are the new observations required to add?• First detection ↔ well-established measurement techniques / sensors
→ Determine level of sophistication
22nd October D4L2 Kerridge 4
2. Radiances / reflectances– The measured quantities (L1 data)
22nd October D4L2 Kerridge 5
Wavelength / nm
R
efle
ctan
ce
GOME coverage
v55 v87v67 v16AATSR channels
Comparison of GOME-1/ATSR-2 & SCIA/AATSR reflectances
SCIA spectra
22nd October D4L2 Kerridge 6
Inter-comparison requires:– Accurate co-location of imager/spectrometer– Spectral averaging of SCIA/GOME-1– Spatial averaging of AATSR/ATSR-2
AATSR / SCIA / GOME spatial coverage
SCIAMACHY
GOME
80 km
40 km
Comparison of GOME-1/ATSR-2 & SCIA/AATSR reflectances
22nd October D4L2 Kerridge 7
Comparison of AATSR, ATSR-2, GOME-1 and SCIAMACHY 670nm Reflectances
Two different orbits on 15th December 2002
22nd October D4L2 Kerridge 8
3. Ozone– A retrieved quantity
22nd October D4L2 Kerridge 9
Direct comparison of MIPAS O3 with HALOE IR solar occultation
• Solar occultation: – High vertical resolution, precision & accuracy– Sparse geographical coverage
• HALOE: – 79 co-locations between 22/07/02-14/12/02– Accuracy: 30-60km 6%; 15-30km 20%– Co-location: <250km on same day
22nd October D4L2 Kerridge 10
High latitude Subtropics
Individual HALOE Profiles
22nd October D4L2 Kerridge 11
0.5 – 50 hPa: +5 – +15% (+/- 20%) >50hPa: +ve biases and increases in RMS
HALOE sampling weighted to N. mid. lats; only 10% in tropics
Ensemble of HALOE Profiles
22nd October D4L2 Kerridge 12
Factors Limiting Direct Comparisons• Precision:
– Direct satellite comparisons underestimate MIPAS precision, due to:
• Imperfect co-location cf atmospheric variability
• Differences in representation of O3 vertical profile
• Precision of the correlative sensor factored in
• Accuracy:– Vertical structure of bias partly reflects different vertical
resolutions of respective sensors
22nd October D4L2 Kerridge 13
Direct Assimilation of MIPAS Data
Ozone: July 10th 2003, 850K level
• Direct assimilation equivalent to
Kalman Smoother, but far lower
computational cost.
• Uses isentropic advection as
modelling constraint.
→ No chemistry or vertical advection
• Field allowed to adjust continually
towards the observations.
→ Analysis unbiased relative to I/P data.
• Applied to MIPAS ozone, methane
and water vapour products (Feb. to
July 2003 v4.61/2)
M.Juckes, RAL
22nd October D4L2 Kerridge 14
MIPAS Ozone Comparisons with Other SensorsMean [ppmv] Standard error [ppmv]
Assimilation allows comparison against other observations which are not coincident.
Biases (left) found to be small in the lower and mid-stratosphere.
Standard error (right) is <10% in most of the stratosphere.
Using a range of independent instruments helps establish confidence.
Hei
ght [
km]
In both graphs, the outer limit of shading is 10% of sample mean profile, inner boundary of shading 1%, transition 5%.
10%
22nd October D4L2 Kerridge 15
Assimilation of MLS ozone profiles: comparison of analyses with ozone sondes
0 5 10 15 20 25 30
Ozone in mPa
1000800600500400300
200
1008060504030
20
1086543
2
Pre
ssure
hPa
Month = 200708 ( 10 sondes)Neymayer (Lat = -70.7, Lon = -8.3)
Ozone profiles from sondes and analyses
Sonde CTRL MLS
0 5 10 15 20 25 30
Ozone in mPa
1000800600500400300
200
1008060504030
20
1086543
2
Pre
ssure
hPa
Month = 200709 ( 8 sondes)HOHENPEISSENBERG (Lat = 47.8, Lon = 11.0)
Ozone profiles from sondes and analyses
Sonde CTRL MLS
0 5 10 15 20 25 30
Ozone in mPa
1000800600500400300
200
1008060504030
20
1086543
2
Pre
ssure
hPa
Month = 200709 ( 3 sondes)NAHA (Lat = 26.2, Lon = 127.7)
Ozone profiles from sondes and analyses
Sonde CTRL MLS
• 1st July – 30th Sept 2007• Ozone data actively assimilated into ECMWF operational system:
• CTRL: SCIAMACHY total column (KNMI)• MLS: SCIAMACHY total column + MLS O3 profile
→ Agreement with ozonesonde profiles <200hPa improved by assimilating MLS
R.Dragani, ECMWF
22nd October D4L2 Kerridge 16
)( DIAL aafinal xxAxx
TES a priori profile
DIAL or sonde profile
TES averaging kernel
Comparison of TES Retrieved O3 Profileswith Ozonesondes & Airborne DIAL
22nd October D4L2 Kerridge 17
TES O3 – Ozonesondes
Worden et al.
22nd October D4L2 Kerridge 18
TES O3 – Ozonesondes (contd.)
TES V1 TES V2
22nd October D4L2 Kerridge 19
Airborne DIAL Measurements • DIAL profiles ozone simultaneously
above & below DC-8 aircraft
• DIAL profiles <0.15o (lat/long) from TES profile averaged & interpolated to TES p grid.
• Missing data in DIAL profile → TES a priori
•DIAL: accuracy <10% (2 ppbv)•Vertical resolution of 300 m.
Richards et al
22nd October D4L2 Kerridge 20
TES O3 – DIAL
TES V2 TES V3
– Care needed in comparisons near tropopause, where O3 vertical gradient strong.– Using AKs, TES bias in upper troposphere seen to be reduced in V3
22nd October D4L2 Kerridge 21
Comparison of OMI integrated ozone column to ECMWF analyses
• 70-day period from 12 Aug 2005
• Assimilation of operational obs:– ozonesondes – aircraft – satellite IR rads– SBUV/2 partial columns– SCIAMACHY columns – MLS ozone profiles
– OMI data:– No sunglint – SZA< 84 deg– Coincidence criterion: within ± 1 hour of analysis time
Analysis by S.Migliorini, U.Reading
22nd October D4L2 Kerridge 22
Total column ozone comparisons
• Total column should be compared to
ˆ ( )Ta a a x x
ana anaˆ ( )T
a a a x x
ECMWF analyses firstinterpolated to OMI pixel locations
ˆ ( )a a x x x A x x ε
T g x T Ta g A Tx g εIntegrated column:
Retrieved profile:
22nd October D4L2 Kerridge 23
OMI ozone layer averaging kernels
a < 1
15 Oct 2005 12 UTC
nadir pixel
22nd October D4L2 Kerridge 24
OMI total column ozone
22nd October D4L2 Kerridge 25
Simulation with ECMWF 3-D analysis
22nd October D4L2 Kerridge 26
Integrated Column Differences (%)
For 70 day’s data: [OMI-ECMWF] bias -3.773 % and RMS 5.2 %
22nd October D4L2 Kerridge 27
4. Aerosol
22nd October D4L2 Kerridge 28
Height-Integrated Aerosol
• Aerosol optical thickness (AOT), (), at = 550 and 870 nm
Cloud screening & handling of surface BRDF critical for aerosol retrieval from satellite sensors
22nd October D4L2 Kerridge 29
Comparisons of AATSR & SEVIRI with MODIS and MISR
MODIS & MISR on EOS-Terra Day-time equator crossing time
similar to Envisat (~10am) Terra ascending node in daytime
whereas Envisat descending, → Observing times at high latitude
quite different to AATSR.
SEVIRI on MSG samples hourly, though viewing geometry a fixed function of geographical location
22nd October D4L2 Kerridge 30
Impact of coincident sampling on AATSR – MODIS Comparison in Sept’04
Cor=.64 Cor=.48 Cor=.56SD=.16
Cor=.71SD=.08
22nd October D4L2 Kerridge 31
C=0.86 C=0.53 C=0.80 C=0.45
SEVIRI
MODISMISR
SEVIRI aerosol optical thickness comparisons with MODIS & MISRJuly 2005
22nd October D4L2 Kerridge 32
5. Ground-based networks
22nd October D4L2 Kerridge 33
Aeronet
AErosol RObotic NETwork. http://aeronet.gsfc.nasa.gov
Standardized instruments, calibration & processing.
→Geographically distributed observations of aerosol spectral optical depths & retrieved products
22nd October D4L2 Kerridge 34
Aeronet • Each ground station has CIMEL sun-photometer
’s: 340, 380, 440, 500, 675, 870, 940 and 1020 nm– Widths: 2nm at 340nm, 4 at 380nm, others 10nm.– Narrow FOV ~1o
• Direct sun measurement every 15 mins
• AOT from direct sun extinction → Beer’s law – Corrected for Rayleigh scat. + trace-gas abs.– Rel. acc. cf other photometers <0.004– Abs. acc. <0.01-0.02
• [Angstrom coefficient fitted to AOT at 440, 500, 675, 870nm]
22nd October D4L2 Kerridge 35
• Satellite grid-box containing AERONET station identified.
• For each day, AOT (and Angstrom coeff.) extracted within ±20km from this satellite pixel.
• Aeronet measurements for each station extracted within ±30 minutes of satellite measurements.
• For both Aeronet and satellite sensor, no. of valid retrievals, means and standard deviation stored
• Minimum of 4 matches required.
• If SD of either satellite or Aeronet exceeds 0.15, comparison discounted.
Satellite – Aeronet Comparison
22nd October D4L2 Kerridge 36
SEVIRI – Aeronet individual time series comparison
22nd October D4L2 Kerridge 37
SEVIRI
AeronetMODIS
MISR
SEVIRI – Aeronet time series comparison for North Atlantic Ocean
22nd October D4L2 Kerridge 38
SEVIRI – Aeronet: statistical comparison
C=.2SD=.07
C=.84SD=.04
C=.38SD=.12
C=.81SD=.07
22nd October D4L2 Kerridge 39
SEVIRI – Aeronet: Europe & Africa
C=.64SD=.06
C=.36SD=.15
– SEVIRI performance reasonable over sea and coasts– High-reflectance land sites (Africa) are problematic
22nd October D4L2 Kerridge 40
– Representativeness – point location vs satellite field of view
– Only five aerosol types in retrievals– aerosol composition varies continuously
– Cloud flagging may be too stringent– upper limit on allowed AOT set too low
– High land surface reflectance not modelled well
Some Factors Limiting Aeronet Comparisons
22nd October D4L2 Kerridge 41
Network for the Detection of Atmospheric Composition Change (NDACC)
>70 high-quality stations observing stratosphere and upper troposphere
→ impact of stratospheric changes on troposphere and on global climate LIDAR profiles:Raman lidar - water vapor Differential Absorption Lidar (DIAL) - O3 Backscatter lidars - aerosol Raman and Rayleigh lidars - temperature
Microwave radiometers: ozone, water vapor, and ClO profiles
UV/VISIBLE SPECTROMETERS: column ozone, NO2 (OClO and BrO) FTIR SPECTROMETERS: column ozone, HCl, NO, NO2, ClONO2, and HNO3
DOBSON/BREWER: column ozone
SONDES: ozone and aerosol profiles
UV SPECTRORADIOMETERS: UV radiation at the ground
22nd October D4L2 Kerridge 42
22nd October D4L2 Kerridge 43
• Network of ground-based FTS• Near-IR solar absorption spectrometry• 4,000–14,000 cm−1 at 0.02 cm−1 resolution • Analysis with standard algorithm (GFIT)
– Non-linear least squares → scale profile
→ CO2, CH4, O2 & other columns
• O2 to convert column densities to pressure- • weighted column average mixing ratios
• Complementary surface in-situ at each site
• To usefully constrain global carbon budget – directly, and through validation of satellite column measurements – precision of 0.1% required!
• SCIAMACHY & GOSAT validation
→ Stringent accuracy reqs. eg:- solar tracking- correction for source fluctuations- FTS instrument line shape
-permanently monitored with HCl cell in solar beam.
- spectroscopic parameters- retrieval algorithms
Total Column Carbon Observing Network - TCCON
22nd October D4L2 Kerridge 44
– Variability reflects atmospheric lifetimes: CH4 ~10yrs, N2O >100yrs
– Correction needed for stratospheric column variability
Sherlock et al
22nd October D4L2 Kerridge 45
6. Summary– Comparison of observation with theory is fundamental– Validation of satellite data also essential– Important to compare like-with-like as far as possible
→ Apply observation operators to model fields and in assimilation→ Account for sensor vertical smoothing & a priori also in
comparisons with profiles observed at higher resolution
– Techniques of increasing sophistication are being used→ to compare with models and other observations→ to quantify value added in assimilation
– Production of long-term satellite data-sets on the essential climate variables will depend on surface networks for ground truth.