validation of tes methane with hippo observations
DESCRIPTION
Application to Adjoint Inverse Modeling of Methane Sources. Validation of TES Methane with HIPPO Observations. Atmospheric and Environmental Research, Inc. 9 November 2010. Kevin J. Wecht, DJ Jacob, SC Wofsy , EA Kort , A Eldering , JR Worden, SS Kulawik , GB Osterman , VH Payne. - PowerPoint PPT PresentationTRANSCRIPT
Validation of TES Methane with HIPPO Observations
Kevin J. Wecht, DJ Jacob, SC Wofsy, EA Kort, A Eldering, JR Worden, SS Kulawik, GB Osterman, VH Payne
9 November 2010Atmospheric and
Environmental Research, Inc
Application to Adjoint Inverse Modeling of Methane Sources
Atmospheric CH4: RF and trends
0.6 0.3 [W m-2]
Emission-based Radiative Forcing
Radiative forcing from CH4 emissions ≈ 50% radiative forcing from CO2 emissionsCause of slowdown in CH4 growth rate remains unclear, but is likely driven by emissions.
CH
4 [pp
b]
?
Bottom-up CH4 Emission Inventories
Annual CH4 Emissions (IPCC) [Tg a-1]
Bottom-up inventories do not capture slowdown in CH4 growth rate or confidently apportion emissions between source types.1978 2005
EDGAR CH4
Anthropogenic Emissions [Tg a-1]350
300
?
CH4
[ppb]
Intelligent climate change legislation requires more tightly constrained sources.
200
100
200
100
50
150
Inverse Methods Help to Constrain Emissions
[ppb]
GEOS-Chem and NOAA GMD surface methane
Forward
Model
Inverse Model
CH 4
Forward Model:GEOS-Chem 3-D chemical transport model (CTM)
Satellite observations represent potentially huge source of information on CH4 emissions
GEOS-ChemNOAA GMD
Adjoint Inverse Modeling of Methane Sources
Adjoint inverse analysis
OPTIMIZATION OF SOURCES
GEOS-Chem CTM
ValidationHIPPO
Methane(Wofsy, Kort)
Apriori Sources
GEOS-Chem Adjoint
EDGAR v4, Kaplan, GFED 2, Yevich and Logan [2003]
HIPPO Methane provides:• Large number of profiles• Wide latitudinal coverage• Remote from sources
(reduces colocation error)
TES Methane (Worden, Kulawik, Payne)
Other Methane
Remote Sensing of CH4
Wavelength, μm20 1015 89 7
CH4
Radi
ance
, 10-3
W m
2 sr-1
Jacob 1999
Terrestrial Radiation Spectrum
Calibrated Radiance
Inte
nsity
Wavelength
CH4
Retrieved Atmospheric CH4
Pres
sure
Mixing ratio [ppb]
Inverse Model
1 TES “global survey”16 orbits, 26 hours, 2300 observations
15-16 global surveys per month
8 km
5 km
One obs
I validate the TES CH4 product and evaluate its utility for use in inverse modeling.
TES V004 CH4
• Methane retrieval 7.658 – 7.740 μm• Averaging kernels peak 200-400 hPa• Degrees of Freedom for Signal 0.5-2.0• ~50% of TES observations pass CH4 quality filter• Unit of comparison for this work is
Representative Tropospheric Volume Mixing Ratio (RTVMR)
Rows of Typical Averaging Kernel
RTVMR (Payne et. al. 2009)• Map retrieval to 4-level pressure grid defined by AK.• Concentration at 2nd pressure level is the RTVMR.
0 0.05 0.10 1600 1700 1800 1900 [ppb]
Pres
sure
[hPa
]
10
1000
100
aprioriTES
Corresponding CH4 Retrieval RTVMR Grid
Reduce all profiles to 1 piece of information representative of tropospheric mixing ratio.
HIAPER Pole-to-Pole Observation Program (HIPPO)
HIPPO I - interpolated methanePacific transect (flights 3-7 of 10)Dominant variability with latitudeLittle variability in vertical
• Five missions:- Jan, Nov 2009- April 2010, June & Aug 2011
• > 700 vertical profiles by finish- 80% surface – 330 hPa- 20% surface – < 200 hPa
• Methane Instrument Properties- Frequency: 1 Hz- Accuracy/Precision: 1.0/0.6 ppb
Pres
sure
[hPa
]
1000
800
600
400
200
Latitude0 20 6040-60 -20-40 80
1600 1775 1950[ppb]
HIPPO I HIPPO II
Using HIPPO to Define Coincidence Criteria
Mean Bias
HIPP
O I
Residual Standard Dev.
[ppb
]
# Observations
Coincidence requirements of +/- 750km and +/- 24h are sufficient.
Standard error of the mean
24 hour12 hour
24 hour12 hour
~ 250 km
HIPPO profiles
TES observation 5 x 8 km
Validation characterizes mean bias and residual error.Residual error contains contributions from: 1) error in the retrieval 2) colocation error r
-6 h+20 h
250 500 750 250 500 750 250375
750500625Distance [km]Distance [km]
N
HIPPO v. TES by Latitude
RTVM
R [
ppb]
RTVM
R [
ppb]HIPPO 1
Jan 9-30,2009
HIPPO 2Oct 22 – Nov 20,
2009
Positive bias and significant noise, but latitudinal gradient roughly captured.
TES
HIPPO w/ TES op.HIPPO 250-450 hPa
# TES = 204# HIPPO = 76
# TES = 183# HIPPO = 72
HIPPO v. TES Difference by Latitude
HIPPO 1Jan 9-30,
2009
HIPPO 2Oct 22 – Nov 20,
2009
RTVM
R [
ppb]
RTVM
R [
ppb] TES – HIPPO over land
TES – HIPPO over ocean
Mean Bias
HIPPO I: bias = 73.7 ppb, residual std dev = 43.1 ppb ≈ 3 x self-reported errorHIPPO II: bias = 59.9 ppb, residual std dev = 48.4 ppb marginally sig. land-ocean diff
Distribution of TES Residual Errors
HIPPO I1/9 - 1/30,
2009
HIPPO II10/22 – 11/20,
2009
Normally distributed errors are important for derivation of inverse cost function.
ScatterplotsError distributions
TES
RTVM
R [p
pb] 2000
1900
1800
1700
2000
1900
1800
1700TES
RTVM
R [p
pb]
HIPPO RTVMR [ppb]1700 1800 1900
N = 76 r = 0.65
N = 72 r = 0.59
bias = 73.3σ = 43.1
bias = 59.9σ = 48.4
The Ability of TES to Capture Latitudinal Gradients
TES captures HIPPO I & II latitudinal gradients on a scale of ~20°
HIPPO I 15° Latitude Bins
HIPPO I and TES GradientsTESHIPPO
Latitude0 20 6040-60 -20-40
Gra
dien
t [pp
b / °
lat]
0
2
-2
4
RTV
MR
[ppb
] 1900
1800
1700
Bin Size [° latitude]0 20 6040 80
[ppb
/ °la
t]
0
1
2
3
4
5
mean( |HIPPO| )mean( |TES – HIPPO difference| )
Adjoint Inverse Modeling of Methane Sources
Adjoint inverse analysis
OPTIMIZATION OF SOURCES
GEOS-Chem CTM
ValidationHIPPO
Methane(Wofsy, Kort)
Apriori Sources
GEOS-Chem Adjoint
EDGAR v4, Kaplan, GFED 2, Yevich and Logan [2003]
HIPPO Methane provides:• Large number of profiles• Wide latitudinal coverage• Remote from sources
(reduces colocation error)
TES Methane (Worden, Kulawik, Payne)
Other Methane
Adjoint Inverse Modeling
Analytical Inversion Adjoint InversionJ(x)
xx*xa
Jacob 2007
Define cost function
Adjoint approach allows us to constrain emissions at the resolution of the model.
Computationally impractical for my goals
Twin Tests (OSSE) for Inversion of Methane Sources1-D example
x = CH4 emissions, F(x) = CH4 [ppb] F(xtrue) true stateF(xpert) apriori, perturbedF(xpost) post. best guess
F(x)
distance
F(x)
distance
F(xtrue) true statePseudo-observations
Why perform twin tests?• Test the inverse calculation• Understand the effects of TES error and apriori• Find optimum length of assimilation period
Twin test methodology:• Generate pseudo-observations from “true” state• Perform inversion with purposefully incorrect apriori• Goal: recover true atmospheric state by adjusting emissions
Through twin tests, we can evaluate the utility of TES CH4 for retrieving CH4 emissions.
Twin Tests (OSSE) for Inversion of Methane Sourceswith TES CH4
Why perform twin tests?• Test the inverse calculation• Understand the effects of TES error and apriori• Find optimum length of assimilation period
The twin test methodology• “Observe” GEOS-Chem with TES• Add error to pseudo-observations of magnitude
determined by HIPPO validation• Begin adjoint inversion with incorrect apriori• Goal: to recover original emission field.
GEOS-Chem RTVMR July 2008
w/ error
A priori emissions
= -50% = +50%
# of Observations July 2008 total = 22038 ppb
2.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Twin Tests Results
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%0
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%02.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%02.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%02.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%02.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%02.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%02.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%02.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%02.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%02.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Tests Results
Iteration #0 5 10 15 20 25
Cos
t Fun
ctio
n
-50% +50%0
Convergence!
2.0 4.0 6.00
July 2008True CH4 Emissions [102 Mg]
Assimilate July 2008 observations 22,038 total
Observational Error = 40ppb
Twin Test Results
Assimilation Length # observationsTop 1 month 22038Bottom 2 months 40060
TES alone is not sufficient for use in a global adjoint inversion
0 +50%-50%
Optimize July 2008 emissions, assimilating data for variable lengths of time.
Additional work:• Test longer assimilation periods• Understand how the solution changes as a function of assimilation length
New TES CH4
Update this slide for new retrieval • Extend spectral domain of retrieval• Joint retrieval for CH4, N2O, H2O, HDO• “N2O correction” to mitigate systematic errors
• New retrieval has more DOFS and sensitivity lower in troposphere(key for inverse modeling)
New Retrieval has sensitivity lower in troposphere, some vertical information.
New TES CH4
During HIPPO I & II (Jan/Nov 2009)
DOFS have strong seasonal and latitudinal dependence
RTVMR-like levels
New TES CH4
First look comparison to HIPPO I
• Reproduce HIPPO TES difference by latitude and error characterization slides here• Maybe reproduce whether TES can capture latitudinal gradients.• Make a slide on whether TES captures vertical gradients as measured
by HIPPO (let dCH4/dz ~ diff b/wn upper and lower pieces of info)
Error <= that of old TES CH4 retrieval. Real benefits: higher DOFS, lower sensitivity
TES – HIPPO difference by latitude
Upper Troposphere
Mid-lower Troposphere
TES – HIPPO over landTES – HIPPO over ocean
ppbppb
ppbppb
• Further validation of New TES CH4 for use in inverse modeling of CH4 sources.• Evaluate utility of new TES product in constraining CH4 emissions.• Assimilate new TES product to optimize emissions.
• Joint SCIAMACHY/TES assimilation.– SCIAMACHY provides complimentary sensitivity and coverage to TES
• Evaluate inversion results with NOAA GMD surface observations
• Incorporate additional satellite CH4 products (AIRS, GOSAT) and focus on North America.
Future Work: Optimal Estimation of CH4 Sources
Evaluate utility of new TES CH4 for inverse modeling.Assimilate new TES and additional satellite CH4 to optimize emissions.
Old TES CH4 - Most recent public release• TES captures latitudinal gradient in HIPPO data at ~20° resolution• TES is biased high and residual instrument error is 3 x self-reported range• Colocation error in validation is negligible• Enabling Inverse Modeling:
– Quantification of bias, characterization of error– Robust latitudinal gradient with greater coverage than surface stations– TES alone is not sufficient for use in global adjoint inversion
New TES CH4
• Sensitivity lower in troposphere (important for inverse modeling)• Error <= that of old TES CH4
Future Work• Assimilate new TES CH4
• Evaluate inversion results with NOAA GMD surface observations• Incorporate additional satellite CH4 products and focus on N.A.
Thank You!