1) Boundary conditions representing air flowing into North America from other regions contribute significantly to uncertainties in atmospheric CO2 mixing ratios, especially at seasonal timescales.
2) Fossil fuel emissions uncertainties are another major source of uncertainty in CO2 mixing ratios when analyzed at annual timescales.
3) Flux tower measurements of ecosystem carbon uptake and atmospheric CO2 concentration data provide consistent results on biases in biogeochemical model simulations, but concentrations cannot fully disentangle diurnal biases identified by flux towers.
300003-World Science Day For Peace And Development.pptx
Lauvaux, Thomas: Uncertainty-based analysis on constraining continental carbon exchanges from atmospheric greenhouse gas mixing ratios
1. Constraining continental carbon exchanges from atmospheric CO2
mixing ratios: an uncertainty-based analysis.
Thomas Lauvaux
LSCE - IPSL, France
ICOS Science Conference, Online, 15 Sept 2020
2. Land Carbon uptake over North America: NACP report
2007 2012 2015 2018
Estimates of the North American Carbon Sink in this Century. These estimates, in teragrams of carbon (Tg C) per
year, are derived from inventory analysis, atmospheric inversion models (AIMs), and terrestrial biosphere models
(TBMs). From SOCCR 2 report, 2018.
https://carbon2018.globalchange.gov/downloads/SOCCR2_Ch2_North_American_Carbon_Budget.pdf
3. Net Ecosystem Exchange of CO2 over Western Europe
Monteil et al., in review, ACPD
Annual carbon uptake similar to North
America (~0.5PgC)
Biogeochemical models tend to be lower
than inverse estimates (-0.2 to -0.4PgC)
Seasonality agrees well but amplitude of
the seasonal cycle varies widely
4. Atmospheric inversions of CO2 fluxes over the Corn Belt
Schuh et al., 2013, GCB
Strong uptake in a smaller region, the Corn Belt (~120
TgC)
Agricultural inventory and inverse estimates agree well
but with large uncertainties (~35%)
No seasonality here with annual inventory
5. Posterior errors
Atmospheric inversion versus direct error propagation
Prior CO2
fluxes
Fossil fuel
emissions
CO2 boundary
inflow
Adjoint/Direct model
Posterior CO2
fluxes
Prior errors No error Boundary
errors
Transport errors
Inversion framework for domain-limited problems
6. Posterior errors
Prior CO2
fluxes
Fossil fuel
emissions
CO2 boundary
inflow
Adjoint/Direct model
Posterior CO2
fluxes
Prior errors No error Boundary
errors
Transport errors
Inversion framework for domain-limited problems Direct error propagation framework
for domain-limited problems
Prior errors Fossil fuel
errors
Boundary
errors
Transport errors
Prior errors
in CO2
mixing ratios
Fossil fuel
errors in CO2
mixing ratios
Boundary
errors in CO2
mixing ratios
Atmospheric inversion versus direct error propagation
7. Atmospheric inversions of CO2 fluxes over North America
Main science question:
Can we invert the biogenic CO2 fluxes alone? and neglect everything else...
Ensemble of transport, boundary conditions, and surface fluxes (biogenic and fossil fuel):
1. Ensemble of General Ciruclation models (e.g. GEOS-Chem, TM5) coupled to WRF-Chem over North
America
Collaborators: A. Schuh (CSU), D. Baker (CSU), A. Jacobson (NOAA), J. Liu (JPL)
2. Ensemble of transport realizations varying model physics and adding perturbations (SKEBS, Shutts, 2005,
QJRMS)
3. Ensemble of biogenic surface CO2 fluxes from MsTMIP (Huntzinger et al., 2013, GMD) and CASA
(Zhou et al., 2020, JGR: Bio.)
4. Ensemble of fossil fuel CO2 emissions from the FFDAS (Rayner et al., 2010, JGR)
8. Boundary conditions: uncertainties in the CO2 inflow
Boundary inflow
Surface fluxes
Regional inversion of carbon sources and sinks
- Atmospheric transport and surface CO2 fluxes at higher resolution
- Boundary conditions become part of the inverse problem
9. Boundary conditions: uncertainties in the CO2 inflow
Propagation of uncertainties from global models
Coupling of four different global CO2 models optimized using atmospheric CO2 concentrations
- CarbonTracker 2018 NOAA inversion (A. Jacobson, NOAA Boulder)
- PCTM 4D-var inversion (D. Baker, CSU)
- GEOS-Chem EnKF inversion (A. Schuh, CSU)
- Carbon Monitoring System (J. Liu, NASA JPL)
How large are the uncertainties at seasonal / daily timescales?
10. Decomposition of CO2 boundary conditions over North America
- Seasonal scale variability shows
lagged growing season
- Similar amplitude
- No annual bias
- Boundary inflow varies across
the domain
Where does it matter the most?
- Differences among GCM’s
Are these differences significant
compared to surface fluxes?
- Two temporal scales (at least):
seasonal and synoptic
Can we characterize them
accurately?
Wisconsin Tall tower
Wisconsin Tall tower
11. Monthly air flow and CO2 mixing ratios (here January 2010) over
North America
Flow of CO2 from boundary conditions over North America
Monthly air flow and CO2 mixing ratios (here July 2010) over
North America- Large scale flow drives the spatial patterns in CO2
mixing ratios from the boundaries
- Main flow over North America from North and West,
associated with the polar jet stream
Lauvaux et al., in prep.
12. Synoptic systems and CO2 boundary conditions over North America
Winter
Summer
Jet Stream
Summer N.Am.
monsoon
13. Synoptic systems and CO2 boundary conditions over North America
Time series of modeled CO2 mixing ratios (in black) and
potential temperature (in gold) by WRF-Chem at 27km coupled
to CarbonTracker 2016 for July 2010.
14. Transport model and GCM uncertainty on CO2 boundary conditions
Model-model differences across four GCM’s and 10 different WRF-Chem
SKEBS configurations for 2010
- Large spatial variability across North America
- Synoptic scale errors due to atmospheric model
(location of syntopic systems) combined with errors
from the boundary inflow
- Transport bias remains low (<1ppm) but random
errors are large (up to 4ppm) esp. in summer. Bias
comes from the boundary inflow itself.
16. Posterior uncertainty associated with total fossil fuel CO2 emissions
(percentage standard deviation) at 0.1 degree resolution (year 2002)
Fossil fuel uncertainty at sub-national scale
Fossil Fuel Data Assimilation System
(FFDAS)
- Propagation of national-scale emissions
uncertainties (downscaling), monthly
- Further downscaled with imposed diurnal
variability
- No variability in the spatial distribution at
sub-national scale
Asefi-Najafabady et al., 2014, JGR: Atmos.
18. Biogenic flux uncertainties: ensemble of biogeochemical models
Monthly Mean of NEE (July 2010)
MsTMIP project: inter-comparison of biogeochemical
Terrestrial Models with similar driver data and predictive LAI
Global simulations at 0.5 degree resolution providing monthly
Net Ecosystem Exchange, downscaled to 3-hour fluxes
Main NEE spatial features over North America remain highly
uncertain
For example: absence of Corn Belt in half of the members
Net Ecosystem Exchange from 15 different MsTMIP
biogeochemical terrestrial models
Feng et al., 2019, JGR: Atmos.
Huntzinger et al., 2013, GMD
19. Error spatial distribution in atmospheric mixing ratios
biogenic flux
errors
Boundary
errors
Transport
errors
- Flux and transport errors are highly correlated (caused
by the CO2 flux distribution)
- Boundary errors located in the northern part of the
domain, and follow the jet stream over North America
- Flux errors larger than transport errors, while boundary
errors remain smaller (RMSD of 3ppmv)
Feng et al., 2019, JGR: Atmos.
21. Temporal aggregation of errors: daily to annual estimates
- Transport errors decrease following a random
distribution
- Boundary errors remain significant until the annual scale
- Flux errors are largest up to the annual scale
- Fossil fuel errors are second largest, and similar to
biospheric errors annually
Feng et al., 2019, GRL
23. From the surface to the atmosphere:
Are concentration and flux data telling the same story?
24. Model Input Product Resolution Frequency Source
NDVI MOD13C2 5600 m monthly NASA MODIS
Precipitation MERRA-2 0.625°× 0.5° monthly NASA MERRA
Temperature MERRA-2 0.625°× 0.5° monthly NASA MERRA
PAR MERRA-2 0.625°× 0.5° monthly NASA MERRA
Longwave radiation MERRA-2 0.625°× 0.5° monthly NASA MERRA
Land cover type MCD12Q1 500 m N/A NASA MODIS
Soil type
(clay, sand and silt, %)
MsTMIP soil
map
0.25°× 0.25° N/A MsTMIP
CASA flux ensembles over North America
(Without disturbance legacies; Monthly; 5.6 km; 2001 – 2016)
April 2016 April 2016 April 2016
Biogenic flux uncertainties: from the surface to the atmosphere
Zhou et al., 2020, JGR:Bio.
25. Parameter Description Default Min Max
Emax Maximum light use efficiency 0.55 0.60 0.50
Topt Optimal temperature of photosynthesis Toptdef Toptdef – 2 Toptdef + 2
Q10 Parameter driving the exponential dependency of the heterotrophic respiration on temperature 1.40 1.20 1.60
Ksurfmet turnover rate of surface metabolic carbon pool 14.80 13.32 16.28
Ksurfstr turnover rate of surface structual carbon pool 3.90 3.51 4.29
Ksurfmic turnover rate of surface microbial carbon pool 6.00 5.40 6.60
Ksoilmet turnover rate of soil metabolic carbon pool 18.50 16.65 20.35
Ksoilstr turnover rate of soil structural carbon pool 4.80 4.32 5.28
Ksoilmic turnover rate of soil microbal cabron pool 7.30 6.57 8.03
Kcwd turnover rate of CWD carbon pool 0.10 0.09 0.11
Kslow turnover rate of slow carbon pool 0.20 0.18 0.22
Karmored turnover rate of armored carbon pool 0.0045 0.00 0.00
Model parameters:
Parameter Sensitivity Analysis
Biogenic flux uncertainties: from the surface to the atmosphere
Three parameters explain 95%
of the observed variability:
Emax, Topt, and Q10
Parametric uncertainty
represents 90% of the model
sensitivity in CASA (much
larger than driver uncertainties)
Spin-up (pool equilibrium) not
tested here
Zhou et al., 2020, JGR:Bio.
26. Calibration of CASA ensemble using Ameriflux data
Zhou et al., 2020, JGR:Bio.
Map of Ameriflux sites used for the calibration
27. MsMTIP versus CASA ensemble: monthly deviation and correlation
MsTMIP models tend to uner-estimate the
seasonal variability
CASA ensemble is more correlated to flux
data thanks to diagnostic LAI
Spatial distribution is similar across the
CASA ensemble
MsTMIP models
(prognostic LAI)
CASA ensemble
(MODIS-derived LAI)
Zhou et al., 2020, JGR:Bio.
29. CT2017
CASA mean
Low Emax (E1)
Medium Emax (E2)
High Emax (E3)
Flux biases as seen by Ameriflux data (summertime only)
Feng et al., in prep.
Courtesy of Dr. Sha Feng, PennState University
30. CT2017
CASA mean
Low Emax (E1)
Medium Emax (E2)
High Emax (E3)
Flux biases compared to concentration biases (summertime only)
CASA ensemble minus Ameriflux
observation
(within the concentration tower
footprints)
WRF concentrations (coupled to
CASA ensemble) minus NOAA
towers
Feng et al., in prep.
Positive bias (uptake is too low)
consistent with both observation
types
Ranking of Emax simulations is
consistent (medium Emax is best)
Daily NEE from CarbonTracker is
optimal but flux data suggested
compensating biases (day versus
night)
Courtesy of Dr. Sha Feng, PennState University
31. Boundary conditions represent a signifciant fraction of the observed atmospheric mismatch at
seasonal timescales
Fossil fuel emissions contribute to a large fraction of the uncertainty in CO2 mixing ratios at the
annual timescale
Flux tower data and concentration provide consistent biases and Emax ranking but
atmospheric observations are unable to disentangle the diurnal biases (new tracers?)
Conclusions and perspectives