1. Carbon Model-Data Fusion
Britton Stephens
National Center for Atmospheric Research
Atmospheric Technology Division
CDAS Team: Dave Schimel, Britt Stephens, David Baker, Steve Aulenbach,
Jennifer Oxelson, Dave Brown, Roger Dargaville
Main research in instruments for O2 and CO2. . .
Got into modeling by thinking about how modelers were using CO2, what type of data would be most useful, and then . . .

2. Are we presently model or data limited? Model-data fusion
Data are sparse but models can’t handle high variability
Model-data fusion
Synthesis inversion
Data assimilation
Parameter estimation
“Introduction of observations into a modeling framework, to provide:
Estimates of model parameters
Uncertainties on parameters and model output
Ability to reject a model” (Michael Raupach)

3. Challenges to carbon model-data fusion
Limited concentration data, far from sources
Vertical and horizontal model coarseness
“Representativeness” or model-data mismatch
Boundary-layer stable-layer height errors
Spatial flux heterogeneities
 must weight measurements appropriately S

4. Regional and smaller scales are critical for linking to underlying processes
TEMPERATURE (C)
(IPCC, 2001)
(NRCS/USDA, 1997)
(NRCS/USDA, 1997)
CHLOROPHYLL
(SeaWIFS, 2002)

5. Unresolved variance presently contains most of the information on regional- and smaller-scale fluxes

6. . . . but to use on a global scale requires a new approach.
Even biased high-frequency measurements do better than long-term means. . .
(Rachel Law, submitted
to Tellus, 2001)
. . . but to use on a global scale requires a new approach.

7. Data-assimilation Ingests data at the time of observations
Can handle very large data streams
Used extensively in weather prediction and satellite analysis
Can assimilate multiple data types
In situ concentrations
Satellite concentrations
Satellite environmental data (e.g. standing water)
Direct flux measurements
Inventory data
Methods are relatively complex
Error statistics are not produced as easily

8. Differences between CH4 and CO2
Assimilation of CH4 may be easier because:
Fluxes are much more unidimensional
Diurnal rectification of sources not an issue
Ocean fluxes are much less significant
Satellite measurements may be more feasible
However. . .
Spatial structure of sources are highly local
In situ measurements are more challenging

9. Carbon Data-Model Assimilation (C-DAS)
Bring together observationalists and modelers to form an integrated approach to improving our understanding of the global carbon cycle.
Initial effort: Network design exercises based on a selected assimilation modeling strategy.
Ongoing: Further development of the assimilation tool and support for testing and

10. Carbon Data-Model Assimilation (C-DAS)
Overview of CDAS
Users
http-Based
Interface
DODS
Aggregation
Server
Simulated
Observing
System
GrADS-
DODS
Server
Simulated
CO2
Observations
Reference Global
Atmospheric CO2
4D VAR
Assimilation
System

11. Carbon Data-Model Assimilation (C-DAS)
Overview of CDAS: Production of Reference Atmospheric CO2
Users
http-Based
Interface
Annual
Land Model
Fluxes
(0.5o)
DODS
Aggregation
Server
Diurnal & Seasonal Cycle Model
Simulated
Observing
System
GrADS-
DODS
Server
Ocean
Model
Fluxes
(2o )
Atmospheric Transport
Model
Reference
Global
Atmospheric
CO2
Simulated
CO2
Observations
Reference Global
Atmospheric CO2
2.5o, resolution
25 vertical levels, 1 hour Dt, & 365 days = 2.6TB
Industrial
Fluxes
(1o )
4D VAR
Assimilation
System

12. Carbon Data-Model Assimilation (C-DAS)

13. Carbon Data-Model Assimilation (C-DAS)
CDAS Application: Data Volumes
Users
2.6 TB
http-Based
Interface
DODS
Aggregation
Server
Simulated
Observing
System
GrADS-
DODS
Server
Simulated
CO2
Observations
Reference Global
Atmospheric CO2
200 MB
4D VAR
Assimilation
System
Global Estimate,
11 North American
Bioregions

14. Carbon Data-Model Assimilation (C-DAS)
Overview of CDAS: retrieval of fluxes using data assimilation
Users
http-Based
Interface
4D VAR
Assimilation
System
Atmospheric
Transport
Model
Estimated
Annual Fluxes
(Bioregional)
DODS
Aggregation
Server
Retrieved
CO2
Observations
Simulated
Observing
System
GrADS-
DODS
Server
Adjoint of
Atmospheric
Transport
Model
Compare
1st Guess fluxes
Simulated
CO2
Observations
Reference Global
Atmospheric CO2
Input Global
Atmospheric
CO2 fluxes
Optimizer
4D VAR
Assimilation
System

16. Flux corrections using existing CO2 network
Hour = 6
Hour = 7
Hour = 5
Hour = 4
Hour = 2
Hour = 3
Hour = 1

17. Flux corrections constrained by regional patterns
Month = 9
Month = 10
Month = 12
Month = 8
Month = 11
Month = 7
Month = 3
Month = 2
Month = 4
Month = 5
Month = 6
Month = 1

18. Potential applications for CH4
What is the optimal network expansion?
Continuous vs. flask measurements
Value of satellite concentrations for various sensors
Proximity of measurements to sources
Accuracy and resolution vs. density of measurements
What other types of data can we assimilate?
Satellite water distributions
Direct flux measurements
Inventory data
Can we assimilate CO2 and CH4 together?
 Primary requirement is people