gaia-clim ftir oe uncertainties
TRANSCRIPT
![Page 1: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/1.jpg)
Part I. Measurement uncertainty
B. Langerock & M. De MaziereJune 2015
![Page 2: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/2.jpg)
GAIA-CLIM 1
±3%
![Page 3: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/3.jpg)
GAIA-CLIM 1
±3%
![Page 4: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/4.jpg)
GAIA-CLIM 2
±10%
![Page 5: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/5.jpg)
GAIA-CLIM 2
±10%
![Page 6: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/6.jpg)
GAIA-CLIM 3
nan
![Page 7: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/7.jpg)
GAIA-CLIM 4
Definitionscf. GUM [JCG08], P. Green ‘Guide . . . ’, Taylor [Tay97]
. measured quanity: temperature T , pressure P , absorbance I, . . .
. the true value of the measurand: Tt, . . . (subscript t) unknown!
. error on measurand: εT = (T − Tt) unknown!
. error is a stochastic variable with an unknown probability density function
. uncertainty is an estimate for this probability density function
mean of ε, covariance of ε, higher order moments of ε, . . .
![Page 8: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/8.jpg)
GAIA-CLIM 5
Formal definition of 2 uncertainties
. RANDOM uncertainty is an estimate of the covariance of ε
covariance = E[(ε− E[ε])2], or STD for scalars
. SYSTEMATIC uncertainty is an estimate of the mean of ε
mean = E[ε]
![Page 9: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/9.jpg)
GAIA-CLIM 5
Formal definition of 2 uncertainties
. RANDOM uncertainty is an estimate of the covariance of ε
covariance = E[(ε− E[ε])2], or STD for scalars
. SYSTEMATIC uncertainty is an estimate of the mean of ε
mean = E[ε]
Every error has a random and systematic uncertainty!
systematic uncertainty is not completely known (up to a sign)
→ if it is, calibrate the measurement
![Page 10: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/10.jpg)
GAIA-CLIM 6
How to estimate uncertainties?cf [JCG08]
. Type A: the measurand allows repeated measurements... use textbookestimators for the uncertainty on the measurand
the random uncertainty on the mean of a n times repeated measurement reduces with 1√n
compared to the random uncertainty on an individual measurement
![Page 11: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/11.jpg)
GAIA-CLIM 6
How to estimate uncertainties?cf [JCG08]
. Type A: the measurand allows repeated measurements... use textbookestimators for the uncertainty on the measurand
the random uncertainty on the mean of a n times repeated measurement reduces with 1√n
compared to the random uncertainty on an individual measurement
the systematic uncertainty of the mean is the mean of the systematic uncertainties
![Page 12: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/12.jpg)
GAIA-CLIM 6
How to estimate uncertainties?cf [JCG08]
. Type A: the measurand allows repeated measurements... use textbookestimators for the uncertainty on the measurand
the random uncertainty on the mean of a n times repeated measurement reduces with 1√n
compared to the random uncertainty on an individual measurement
the systematic uncertainty of the mean is the mean of the systematic uncertainties
. Type B: other measurand → FTIR measurements
![Page 13: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/13.jpg)
GAIA-CLIM 6
How to estimate uncertainties?cf [JCG08]
. Type A: the measurand allows repeated measurements... use textbookestimators for the uncertainty on the measurand
the random uncertainty on the mean of a n times repeated measurement reduces with 1√n
compared to the random uncertainty on an individual measurement
the systematic uncertainty of the mean is the mean of the systematic uncertainties
. Type B: other measurand → FTIR measurements,. . .
![Page 14: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/14.jpg)
GAIA-CLIM 7
Example: solar zenith angle: numerical computation
. python library: pysolar→ calculates the zenith and azimuth angle of thesun for a position+time
. the online documentation provides an estimate of the error on thecomputed angles:
Compare against US Naval Observatory datapoints...http://aa.usno.navy.mil/data/docs/topocentric.php#notes
![Page 15: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/15.jpg)
GAIA-CLIM 8
SZA error?
I Mean error: 0.0736 degreesI Standard deviation of error: 0.124 degrees
![Page 16: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/16.jpg)
GAIA-CLIM 9
Type B error estimates
. find an ensemble of differences
I pysolar angles vs USNOI NCEP temperature profiles vs ECMWF temperature profilesI . . .
. use standard estimators on this ensemble of differences to estimate therandom and systematic uncertainty on the measurand, e.g. SZA, NCEP T,. . .
180 200 220 240 260 280 300 320Temperature [K]
0
20
40
60
80
100
Hei
ght[
km]
NCEPECMWF (stream BF) 10−3
10−2
10−1
100
101
102
103
Pre
ssur
e[h
Pa]
−15 −10 −5 0 5 10
Temperature [K]
0
20
40
60
80
100
Hei
ght[
km]
Mean of difference between ECMWF and NCEP (ZPT) for 2011
10−3
10−2
10−1
100
101
102
103
Pre
ssur
e[h
Pa]
![Page 17: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/17.jpg)
GAIA-CLIM 10
Propagation of uncertaintiesWhy should we propagate uncertainties?
. FTIR measurement is a L1 spectrum
. FTIR target data reports O3 concentration profiles
. the uncertainty on the O3 profile is the propagated uncertainty on themeasured spectrum
If L denotes the (linearized) transformation between the spectrum and thetarget profile:
random covariance on target = L(random covariance on source )L∗
![Page 18: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/18.jpg)
GAIA-CLIM 11
Propagation of systematic uncertainties
. systematic=estimate of the mean error, but sign is missing!
. we can not mathematically propagate the unsigned mean temperatureprofile difference: L|x| 6= |Lx|
180 200 220 240 260 280 300 320Temperature [K]
0
20
40
60
80
100
Hei
ght[
km]
NCEPECMWF (stream BF) 10−3
10−2
10−1
100
101
102
103
Pre
ssur
e[h
Pa]
−15 −10 −5 0 5 10
Temperature [K]
0
20
40
60
80
100
Hei
ght[
km]
Mean of difference between ECMWF and NCEP (ZPT) for 2011
10−3
10−2
10−1
100
101
102
103
Pre
ssur
e[h
Pa]
. systematic ‘covariance’ matrix = E[εT ]iE[εT ]j
![Page 19: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/19.jpg)
GAIA-CLIM 12
. mathematically the systematic uncertainty matrix transforms as acovariance
. the ‘correlation’ coefficients are ±1: just enough information on signchanges to make the propagation work
‘the transformed syst. matrix is the syst. matrix of the transformed error’
. it is NOT a covariance matrix: a systematic uncertainty is an estimate of themean of the error
![Page 20: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/20.jpg)
GAIA-CLIM 12
. mathematically the systematic uncertainty matrix transforms as acovariance
. the ‘correlation’ coefficients are ±1: just enough information on signchanges to make the propagation work
‘the transformed syst. matrix is the syst. matrix of the transformed error’
. it is NOT a covariance matrix: a systematic uncertainty is an estimate of themean of the error
when averaging n measurements (monthly mean),the systematic uncertainty does not decrease with a factor 1√
n
![Page 21: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/21.jpg)
GAIA-CLIM 12
. mathematically the systematic uncertainty matrix transforms as acovariance
. the ‘correlation’ coefficients are ±1: just enough information on signchanges to make the propagation work
‘the transformed syst. matrix is the syst. matrix of the transformed error’
. it is NOT a covariance matrix: a systematic uncertainty is an estimate of themean of the error
when averaging n measurements (monthly mean),the systematic uncertainty does not decrease with a factor 1√
n
. for correct propagation, the random and systematic uncertainties should beprovided as 2D matrices
![Page 22: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/22.jpg)
GAIA-CLIM 13
FTIR reported uncertainties
. we measure L1 date, but we report concentration profiles
. the ‘retrieval method’ is optimal estimation: use the measurement toupdate your best guess on the target profile at the site
. the optimal estimation uses
I tools from statisticsI an iterative scheme that compares a model spectrum with measured
spectrum
![Page 23: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/23.jpg)
GAIA-CLIM 14
xFTIR = xa +A(xt − xa)
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
ppmv
5
10
15
20
25
30
35
40
Heig
ht
[km
]
apriori profile
retrieved profile
![Page 24: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/24.jpg)
GAIA-CLIM 14
xFTIR = xa +A(xt − xa)
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
ppmv
5
10
15
20
25
30
35
40
Heig
ht
[km
]
apriori profile
retrieved profile
0.10 0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.30
5
10
15
20
25
30
35
40
Heig
ht
[km
]
CO.MIXING.RATIO.VOLUME AVK plot between [2.5,40.0]km (DOF=2.50)ppv/ppv units, LA.REUNION.MAIDO 2014/02/11 04:07:28
![Page 25: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/25.jpg)
GAIA-CLIM 15
Uncertainty components
. [Rod00]:
error on target = (A− In)error apriori +Gnoise +GKperror model params
. smoothing uncertainty: (A− In)error apriori
. measurement uncertainty: Gnoise
. model parameters uncertainties: GKperror model params for eachparameter p (temperature, SZA, ILS, . . . )
![Page 26: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/26.jpg)
GAIA-CLIM 15
Uncertainty components
. [Rod00]:
error on target = (A− In)error apriori +Gnoise +GKperror model params
. smoothing uncertainty: (A− In)error apriori
. measurement uncertainty: Gnoise
. model parameters uncertainties: GKperror model params for eachparameter p (temperature, SZA, ILS, . . . )
. other uncertainties . . .
![Page 27: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/27.jpg)
GAIA-CLIM 16
. Leading uncertainty sources: for O3 M. Schneider ([SRH+08]), O. Garcıa([GSR+12]), C. Vigouroux ([VDMD+08, VBC+15]), . . .
![Page 28: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/28.jpg)
GAIA-CLIM 17
GEOMS reported uncertainties
. NDACC provides a random and systematic uncertainty
. we do not report the smoothing uncertainty!
. we expect the data user to apply the AVK to the comparable data:elimination of the apriori in the comparison
The purpose of a comparison=true vs ref
![Page 29: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/29.jpg)
GAIA-CLIM 17
GEOMS reported uncertainties
. NDACC provides a random and systematic uncertainty
. we do not report the smoothing uncertainty!
. we expect the data user to apply the AVK to the comparable data:elimination of the apriori in the comparison
The purpose of a comparison=true vs ref=xt − xref
xFTIR = xt +
![Page 30: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/30.jpg)
GAIA-CLIM 17
GEOMS reported uncertainties
. NDACC provides a random and systematic uncertainty
. we do not report the smoothing uncertainty!
. we expect the data user to apply the AVK to the comparable data:elimination of the apriori in the comparison
The purpose of a comparison=true vs ref=xt − xref
xFTIR = xt + (A− In)error apriori +Gnoise +GKperror model
xFTIR = xa +A(xt − xa) +Gnoise +GKperror model
![Page 31: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/31.jpg)
GAIA-CLIM 17
GEOMS reported uncertainties
. NDACC provides a random and systematic uncertainty
. we do not report the smoothing uncertainty!
. we expect the data user to apply the AVK to the comparable data:elimination of the apriori in the comparison
The purpose of a comparison=true vs ref=xt − xref
xFTIR = xt + (A− In)error apriori +Gnoise +GKperror model
xFTIR = xa +A(xt − xa) +Gnoise +GKperror model
xsref = xa +A(xref − xa)← smooth the reference profile
![Page 32: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/32.jpg)
GAIA-CLIM 17
GEOMS reported uncertainties
. NDACC provides a random and systematic uncertainty
. we do not report the smoothing uncertainty!
. we expect the data user to apply the AVK to the comparable data:elimination of the apriori in the comparison
The purpose of a comparison=true vs ref=xt − xref
xFTIR = xt + (A− In)error apriori +Gnoise +GKperror model
xFTIR = xa +A(xt − xa) +Gnoise +GKperror model
xsref = xa +A(xref − xa)← smooth the reference profile
xFTIR − xsref = A(xt − xref) +Gnoise +GKperror model
![Page 33: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/33.jpg)
GAIA-CLIM 18
0.10 0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.30
5
10
15
20
25
30
35
40
Heig
ht
[km
]FTIR.CO LA.REUNION.MAIDO AVK (rel to apriori) DOF=2.76
0.0 0.2 0.4 0.6 0.8 1.0
CO.MIXING.RATIO.VOLUME_ABSORPTION.SOLAR [ppv]1e 7
5
10
15
20
25
30
35
40
Heig
ht
[km
]
FTIR.CO LA.REUNION.MAIDO: 2015 Feb 14 04:15:24 against g4e2AN
NORS profile
NORS apriori profile
MACC original profile
MACC smoothed profile
MACC regridded (no smoothing)
![Page 34: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/34.jpg)
GAIA-CLIM 19
References
[GSR+12] O. E. Garcıa, M. Schneider, A. Redondas, Y. Gonzalez, F. Hase,T. Blumenstock, and E. Sepulveda. Investigating the long-termevolution of subtropical ozone profiles applying ground-based ftirspectrometry. Atmospheric Measurement Techniques,5(11):2917–2931, 2012.
[JCG08] JCGM/WG2. Evaluation of measurement data — Guide to theexpression of uncertainty in measurement. Working Group 2 ofthe Joint Committee for Guides in Metrology, 2008.
[Rod00] Clive D. Rodgers. Inverse Methods for Atmospheric Sounding :Theory and Practice (Series on Atmospheric Oceanic andPlanetary Physics). World Scientific Publishing Company, June2000.
![Page 35: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/35.jpg)
GAIA-CLIM 20
[SRH+08] M. Schneider, A. Redondas, F. Hase, C. Guirado,T. Blumenstock, and E. Cuevas. Comparison of ground-basedbrewer and ftir total column o3 monitoring techniques.Atmospheric Chemistry and Physics, 8(18):5535–5550, 2008.
[Tay97] J.R. Taylor. An Introduction to Error Analysis: The Study ofUncertainties in Physical Measurements. Physics - Chemistry -Engineering. University Science Books, 1997.
[VBC+15] C. Vigouroux, T. Blumenstock, M. Coffey, Q. Errera, O. Garcıa,N. B. Jones, J. W. Hannigan, F. Hase, B. Liley, E. Mahieu,J. Mellqvist, J. Notholt, M. Palm, G. Persson, M. Schneider,C. Servais, D. Smale, L. Tholix, and M. De Maziere. Trends ofozone total columns and vertical distribution from ftirobservations at eight ndacc stations around the globe.Atmospheric Chemistry and Physics, 15(6):2915–2933, 2015.
![Page 36: GAIA-Clim FTIR OE uncertainties](https://reader030.vdocuments.net/reader030/viewer/2022020203/588f0f991a28ab764f8b7dda/html5/thumbnails/36.jpg)
GAIA-CLIM 21
[VDMD+08] C. Vigouroux, M. De Maziere, P. Demoulin, C. Servais, F. Hase,T. Blumenstock, I. Kramer, M. Schneider, J. Mellqvist,A. Strandberg, V. Velazco, J. Notholt, R. Sussmann, W. Stremme,A. Rockmann, T. Gardiner, M. Coleman, and P. Woods.Evaluation of tropospheric and stratospheric ozone trends overwestern europe from ground-based ftir network observations.Atmospheric Chemistry and Physics, 8(23):6865–6886, 2008.