Download - NACP Meeting, New Orleans February 2, 2011
IMPACT OF THE EXPANDING MEASUREMENT NETWORK ON TOP-
DOWN BUDGETING OF CO2 SURFACE FLUXES IN NORTH
AMERICA
Kim Mueller, Sharon Gourdji, Vineet Yadav, Michael Trudeau, Abhishek Chatterjee, Deborah Huntzinger, Arlyn
Andrews, Andrew Schuh, Yoichi Shiga, Kenneth Davis, Britton Stephens, Beverly E. Law, Colm Sweeney, Marc
Fischer, Danilo Dragoni, Doug Worthy, Matt Parker, Mathias Goeckede, Scott Richardson, Natasha Miles, Anna M.
MichalakNACP Meeting, New OrleansFebruary 2, 2011
CO2 observations – fl ask measurements
CO2 observations –continuous measurements
2008log10(ppm/μmolCO2/m2s) log10(ppm/μmolCO2/m2s)
2oo4 2oo5 2oo6 2oo7 2oo8
Inversion
Sources and Sinks,Uncertainti
es
TransportCO2 observations How well can
you match your data
(R)?
How much you trust prior guess (Q)?
Approach to quantify
Q
Approach to quantify
R
Regional Atmospheric Inverse Modeling
Synthesis Bayesian Inverse Modeling (Bayes)
How is the underlying flux field spatially and temporally
correlated (Q)?
Geostatistical Inverse Modeling (GIM)
Coefficients (β)^
R
Q
0 50 100 150 200 250 300 350375
376
377
378
379
380
381
382
383
384Boundary Conditions, June 1 to July 8
"Globalview"CarbonTracker-newCarbonTracker-oldBoundary
Conditions
Subtract
? ?
@ what temporal
scale
?
?
?
Regularization Method
BETTER
COVERAGE
IN SPACE
& TIME
35 towers
10 towers
SPACE TIME
shortaft -3000 obs
1pm - 1050 obs
Inversion
Sources and Sinks
GV-BC
CT-BC
Boundary
Conditions
3hrly
4Ddiurnal
Estimation
Scale
(I) Synthetic data
experiment
(II) Real data (GIM)
experiment
all – 8400 obs
Shortaft ~ afternoon data at short towers, all data at tall towers
Explicit
prior
(III) Real data (Bayes)
experiment
Synthetic Data Experiment Results Using GIMNo explicit prior so experiment test how much information is within atmospheric content in measurements w/out transport error
“Truth”
3hrly Estimation Scale4Ddiurnal Estimation Scale
Adding more measurement in time and space improves both the spatial pattern and grid scale flux estimates.
TIME
SPA
CE
Biggest “bang for the buck” when adding in more data throughout the day with expanded network.
Could draw opposite conclusion if estimating fluxes at a coarser scale.
JUNE (all fluxes post aggregated to monthly scale)
Courtesy of J. RandersonSynthetic Data (GIM)
Real Data (GIM)Experiment Results
No explicit prior - fluxes are based almost solely on atmospheric content in measurements.
WRF-STILT Transport Model (Nehkorn et al., 2010)
Boundary conditions account for the influence of fluxes that occurred outside of the North American domain
Difference of GV-BCs and CT-BCs is approximately 0.5-1ppm with GV-BCs always being lower and therefore are associate with less sinks (more sinks occurred outside of domain of interest)
Boundary Conditions
3hrly Estimation Scale
TIME
SPA
CE
BC
The choice of boundary conditions doesn’t have much impact on monthly grid-scale fluxes except in boreal north
More constraint provided by increasing the number of measurements per day
JUNE,JULY,AUGUST (all fluxes post aggregated to monthly scale)
Real Data (GIM)
Courtesy of S. Ogle, MCI Campaign
InventoryMore constraint provided by the expanded network
Boundary conditional have a large impact on annual totals from MCI
Inversion using CT-BC results in very strong uptake that is not present in inventory estimates
Annual grid-scaleGV-BC (10TN) GV-BC (35TN) CT-BC (35TN)
Real Data (GIM)
Real Data (Bayes) Experiment Results
Used explicit prior (CASA) to see how much atmospheric measurements correct our first guess of grid-scale fluxes
WRF-STILT Transport Model (Nehkorn et al., 2010)
CASA - prior UMich-Bayes (10twrs) UMich-Bayes (35twrs)
Seasonal grid-scale
Start to pull away from explicit prior in South with the use of more towersMore corrections in the SouthWest and stronger sources in agricultural beltNot many deviations from prior in growing season with 10TN compared to 35TN. Corrections across the contiguous US.
As with Dec-Feb, start to pull away from explicit prior in South with expanded network
Real Data (Bayes)
Annual Budgets (synthetic, GIM and Bayes)
WRF-STILT Transport Model (Nehkorn et al., 2010)
GIM
Annual biospheric budgets for NA
Synthetic data experiments indicates that with the expanded measurement network, we should be able to recover annual budget using GIM. No boundary conditions needed but did simulate real measurement gaps.
Spread of the budgets due to boundary conditions is wide (>1GtC/year). This spread may be exacerbated by the setup of GIM to recover 3hrly fluxes for the year.
The Bayesian results have less of a spread of the estimates due to choice of boundary conditions but still wider than the differences between estimates from the smaller and expanded network.The impact of the boundary conditions was also apparent in the 2004 results.
The ability of the expanded measurement network to budget continental sources and sinks is hampered by the influence of boundary conditions. The spread is likely the same if not wider when using more data is space.
Orchidee courtesy of D.N. Huntzinger and Interim-Synthesis TeamCarbonTracker courtesy of A. Jacobsen and NOAA
Bayes
35TN(08)10TN(08)
9TN(04)
2004 results courtesy of S. Gourdji
Conclusions
1. Estimate fluxes to account for underlying variability in transport or flux field (e.g. 3hrly)
2. Use more observations from more times of the day1. Need a method to verify simulated atmospheric
transport at these additional times3. Better means of validating our boundary conditions (A.E.
Andrews has new version available)
Can the expanded observational network help us to identify sources and sinks at regional scales?
Results look promising but more work to be done …
4. Improve atmospheric transport models5. Better ways to assess uncertainty
1. Assess at what spatial and temporal scales we can trust estimates
To help maximize the extent to which the inversion can extract information content of measurements need to:
ACKNOWLEDGEMENTS Other Contributors:
NOAA-ESRL: Adam Hirsch, Andy JacobsenAER: Thomas Nehrkorn, John Henderson, Janusz EluszkiewiczNACP-Interim Sythensis Team Members
NASA NAS: technical support staff (Johnny Chang and others)Funding: NASA (NNX06AE84G Constraining North American Fluxes
of Carbon Dioxide and Inferring their Spatiotemporal Covariances through Assimilation of Remote Sensing and Atmospheric Data in Geostatistical Framework)
QUESTIONS? Check out: The Top-Down Constraint on North American CO2 Fluxes: and Inter-comparison of Region Inversion Results for 2004, Gourdji et al., 2004Friday at 8:50am
Check out: Come to the data-assimilation side meeting (Yadav & Michalak) Form 5:15-6:15 in the Lafitte Room
Annual difference between fluxesusing GVBCs and CTBCs
Boundary conditions account for the influence of fluxes that occurred outside of the North American domain
Difference of GV-BCs and CT-BCs is approximately 0.5-1ppm with GV-BCs always being lower and therefore are associate with less sinks (more sinks occurred outside of domain of interest)
Even though we saw big differences in MCI region with choice of boundary conditions, differences are the greatest in the West Coast and under-constrained regions of the continent.
Post aggregated fluxesJune More spatial locations reduces
the spread of the monthly budgetand improves ability to recover the“truth”
More spatial data reduces spread but still a lot of variability in estimatesassociated with different setup choices
MidContinental Intensive9 measurement
locations
MCI9 measurement
locations
Average grid-scale diagnostics
Post aggregated scales
Boundary conditions only shift seasonal cycle up or down. Not shown.More difference in growing season with choice of observations to use throughout the day with more measurement locations.Temporal aggregation error has an influence at aggregated areas size of MCI but less so at continental scale.
Post aggregated scales
1. What is the information content of the expanding measurement network in terms of budgeting sources and sinks?
2. How does the inversion setup influence our ability to extract the information from the measurements?
SPACE & TIMEO
ther
Esti
mate
s
Courtesy of S. Gourdji
Courtesy of J. Randerson