1. Case study: using GOSAT to estimate US emissions
Daniel J. Jacob
with Alex Turner, Jianxiong Sheng
Turner, A.J., D.J. Jacob, K.J. Wecht, J.D. Maasakkers, E. Lundgren, A.E. Andrews, S.C. Biraud, H. Boesch, K.W. Bowman, N.M. Deutscher, M.K. Dubey, D.W.T. Griffith, F. Hase, A. Kuze, J. Notholt, H. Ohyama, R. Parker, V.H. Payne, R. Sussmann, C. Sweeney, V.A. Velazco, T. Warneke, P.O. Wennberg, and D. Wunch, Estimating global and North American methane emissions with high spatial resolution using GOSAT satellite data, Atmos. Chem. Phys., 15, , 2015. 

2. Building a continental-scale monitoring system for methane emissions in North America
Can we use satellites together with suborbital observations to monitor methane emissions and their trends?
GOSAT
(+soon TROPOMI, GHGSat)
CalNex
INTEX-A
SEAC4RS
1/2ox2/3o grid of GEOS-Chem
Forward model: GEOS-Chem CTM Prior: EDGAR (soon EPA)
A project supported by the NASA Carbon Monitoring System (CMS)

3. The GOSAT observational record
3 cross-track 10-km pixels 250 km apart
orbit
track
250 km
CO2 proxy retrieval [Parker et al, 2011]
Mean single-retrieval precision 13 ppb
Use here data for 6/ /2011

4. GEOS-Chem Chemical Transport Model
Input data
NASA GEOS meteorological fields
other
Model solves 3-D chemical continuity equations
on global or nested Eulerian grid
Modules
emissions
transport
chemistry
aerosols
deposition
sub-surface
Applications
chemical processes, transport, budgets
inverse analyses
radiative forcing
air quality
biogeochemistry …
Model adjoint
Model capabilities:
Tropospheric and stratospheric chemistry, aerosol microphysics, CO2, methane, mercury, various tracers
1980-present GEOS meteorological data, past and future climates (GCMs)
Horizontal resolution: 0.25ox0.3125o (native), 1/2ox2/3o , 2ox2.5o, 4ox5o, other grids
Used by a community of over 100 research groups worldwide

5. GEOS-Chem methane simulation
Global simulation
at 4ox5o resolution
North America simulation at 1/2ox2/3o resolution
1/2ox2/3o GEOS-Chem grid
dynamic
boundary
conditions
72 vertical levels extending to mesosphere
Prior emission inventory from EDGAR + LPJ wetlands
OH concentrations archived from full-chemistry simulation

6. with prior EDGAR4.2+LPJ emissions
First step: conduct GEOS-Chem simulation with prior emissions, compare to GOSAT and suborbital data
GEOS-Chem CTM
with prior EDGAR4.2+LPJ emissions
Satellite (GOSAT)
aircraft+surface data
Are the comparisons consistent?
Three goals:
Indirect validation of satellite data – demonstrate consistency with suborbital data
Correct any biases that would be irreducible in inversions
Identify any anomalous patterns that the inversion would not be able to correct.

7. GEOS-Chem (with prior emissions) compared to suborbital data
HIPPO aircraft data over Pacific
GEOS-Chem
HIPPO
Jan09
Oct-Nov09
Jun-Jul11
Aug-Sep11
Methane, ppbv
Latitude, degrees
NOAA US observations
GEOS-Chem
GEOS-Chem is unbiased for background methane
Concentrations over US are ~30% too low,
to be corrected in inversion
Turner et al. [2015]

8. GEOS-Chem (with prior emissions) compared to GOSAT data
Mean background difference vs. latitude
High-latitude bias could be due to satellite retrieval or GEOS-Chem stratosphere:
in any case, we need to remove it before doing inversion
Turner et al. [2015]

9. Constructing the observational error covariance matrix
𝐲−𝐅 𝐱 𝐀 )
stdev(ԑO)
Estimate observational error variance as residual standard deviation:
difference
mean error to be corrected
In inversion
GOSAT instrument precision is 13 ppb; use it if higher than residual standard deviation
Error correlation length scale ~100 km, can ignore for GOSAT → diagonal matrix

10. Inversion strategy GOSAT observations, 2009-2011 Dynamic boundary
conditions
Analytical inversion
with 369 Gaussians
Adjoint-based inversion
at 4ox5o resolution
correction factors to EDGAR v4.2 + LPJ prior
Inversion assumes 50% error standard deviation on prior emissions
Turner et al. [2015]

11. State vector chosen to balance smoothing & aggregation error
Native-resolution 1/2ox2/3o
emission state vector x (n = 7096)
Reduced-resolution
state vector x (here n = 8)
Aggregation matrix 
x =x
observation
aggregation
smoothing
total
Choose n = 369 for negligible aggregation error; allows analytical inversion with full error characterization
Posterior error
depends on choice
of state vector
dimension
Mean error s.d., ppb
, ,000
Number of state vector elements
Posterior error covariance matrix:
Aggregation Smoothing Observation
Turner and Jacob [2015]

12. Using radial basis functions (RBFs) with Gaussian mixing model as state vector
Example: dominant Gaussian elements describing emissions in Southern California
State vector of 369 Gaussian 14-D pdfs optimally selected from similarity criteria in native-resolution state vector
Each 1/2ox2/3o grid square is unique linear combination of these pdfs
This enables native resolution (~50x50 km2) for major sources and much coarser resolution where not needed
Turner and Jacob [2015]

13. Inversion results and information content
Turner et al. [2015]

14. Evaluation of posterior emissions with independent data sets in contiguous US
GEOS-Chem simulation
with posterior vs. prior emissions
Comparison of California results
to previous inversions of CalNex data
(Los Angeles)
Turner et al. [2015]

15. Correction factors to bottom-up EDGAR inventory
CONUS anthropogenic emission of Tg a-1 vs. EPA value of 27 Tg a-1
Is the underestimate due to livestock or oil/gas emissions or both?
Turner et al. [2015]

16. Optimized top-down inventory
CONUS anthropogenic emission of Tg a-1 vs. EPA value of 27 Tg a-1
Is the underestimate due to livestock or oil/gas emissions or both?
Turner et al. [2015]

17. Use prior emission pattern for source type attribution
EDGAR + LPJ wetlands
Problem is that EDGAR patterns have large errors
total: 63 Tg a-1
wetlands: 20
livestock: 14
oil/gas: 11
waste: 10
coal: 4
Turner et al. [2015]

18. Methane emissions in US: comparison to previous studies, attribution to source types
Ranges from
prior error
assumptions
2004
SCIAMACHY
2007
surface,
aircraft
GOSAT
EPA national inventory underestimates anthropogenic emissions by 30%
Livestock is a contributor: oil/gas production probably also
What is needed to improve source attribution in future?
Better observing system (TROPOMI)
Better bottom-up inventory (gridded EPA inventory, wetlands)
Turner et al. [2015]

19. Current evaluation of Turner et al
Current evaluation of Turner et al. (2015) results with SEAC4RS aircraft data over Southeast US (Aug-Sep 2013)
Data below 2 km from Don Blake, UC Irvine

20. Using SEAC4RS data to correct Turner et al. (2015) emissions
aircraft data
Optimal emission estimate
from Turner et al. [2015]
Correction factors
Inversion
used as prior estimate
GEOS-Chem CTM
0.25ox0.3125o resolution
Sheng et al., in prep.

21. SEAC4RS inversion provides better fit to SEAC4RS data…
Scaling factor
Prior
Posterior
Sheng et al., in prep

22. …and also to NOAA surface data in Sep-Oct 2013…
SCT
WKT
SGP
Prior
Posterior
Sheng et al., in prep

23. …and to GOSAT data (but noisy because of data sparsity)
September-October 2013
Prior
Posterior
Could emissions have gone down in 2013?
Need to examine consistency between GOSAT data for 2013 and
Sheng et al., in prep

24. Potential of TROPOMI and GEO-CAPE to improve observing system
OSSE applied to SEAC4RS viewing domain and 2-month period
Diagonal elements of averaging kernel matrix A:
DOFS=68.7
DOFS=9.7
SEAC4RS
GEO-CAPE
TROPOMI
DOFs = tr(A) : number of pieces of information to constrain emissions
DOFS=34.1
Sheng et al., in prep