CO2 Flux Estimation: Past, Present and Future

Ray Nassar ray.nassar@ec.gc.ca

Climate Research Division, Environment Canada

NSERC-CREATE Summer School on Arctic Science, 2013

mailto:ray.nassar@ec.gc.ca

Flux Tower Measurements

• Eddy covariance method can be used to measure localized fluxes for

a forest plot (Baldocchi et al. 1988,

Measuring biosphere-atmosphere

exchange of biologically related

gases with micrometeorological

methods, 69, 5, 1331-1340)

• Multiple assumptions involved in the method, but typically considered

accurate on the scale of forest

plots, although can not reliably be

scaled to city, province/state,

country or continental scales

Forward model: F(x) Measurements: y

Data Assimilation / Inverse Modeling

y = F(x) + e

observations forward model state vector observational error

The state can be the 3D distribution of CO2 in the atmosphere or CO2

emissions/uptake (fluxes). The optimal estimate of the state comes from

combining the model and measurements by weighting each according to

their uncertainties.

(Matrix equation)

Expansion of Surface Networks

http://www.earthnetworks.com/

Measurement, Reporting and Verification (MRV)

Pacala et al. (2010)

The JASON study

‘JASON’ is an elite

advisory group used by

the US government for

security/defense issues

Finkbeiner (2011) JASON past,

present and future: The world’s

most independent defence

science advisers, Nature, 477,

397-399

https://www.fas.org/irp/agency/dod/jason/

2011 Report: Methods for Remote

Determination of CO2 Emissions

An early atmospheric CO2 inversion

Tans, Fung, Takahashi (1990), Observational Constraints on the

Global Atmospheric CO2 Budget, Science, 247, 1431-1439.

Adjusted model to get agreement with the mean latitudinal gradient in

order to identify the latitude of the ‘missing sink’

Science, 282, 442-446, 1998

Fossil fuel emissions were assumed

Response Functions

• Also known as Jacobians, Green’s Functions

• Example: Amazon region, Jan-Mar, mid-troposphere

Atmospheric Tracer Transport

Intercomparison (TransCom) Project

Gurney et al. (2002), Towards robust regional estimates of CO2 sources and

sinks using atmospheric transport models, Nature

• Ensemble based on results from 16 atmospheric transport models and

76 sites for 1992-1996

• Solved for “natural fluxes” using assumed fossil fuel emissions in

regions: 11 ocean, 11 land (1 ice)

• Temperate North American sink ~60% of Fan et al. (1998)

Understanding the limitations

• Transport model errors – no simple measure of the uncertainty in transport, but transport differs in each

model

• Sparse measurement coverage – many regions of the world poorly-sampled by the global networks

(especially the tropics, boreal Asia and Amazon)

• Representativeness errors – size/scale mismatch between point measurements and model grid boxes

CarbonTracker (2007)

CO2 Flux

CO2 Flux Uncertainty

• Comprehensive model of fossil fuel, biomass fires, ocean, biospheric fluxes • Ensemble Kalman Filter approach for data assimilation of CO2 observations • ~10-20 other global inversion/assimilation systems worldwide • CT was the most transparent in terms of methods and results, which were

all publicly available online, so it became the industry standard

Flux estimates from satellite CO2?

• First Observing System Simulation Experiment (OSSE) with satellite CO2: Rayner, P.J. and D.M. O’Brien (2001), The utility of remotely sensed CO2 concentration data in surface source inversions (2001), GRL, 28, 1, 175-178.

• Approach: – Run a model to simulate global 3D CO2 distribution to use as ‘Truth’

– Create ‘synthetic observations’ for a satellite and surface network by sampling the model at hypothetical obs locations/times

– Add different amounts of noise to change satellite CO2 precision

– Test each set of synthetic satellite observations and synthetic surface data in a CO2 flux inversion to assess the ability of the dataset to return the true fluxes

• Conclusion: Monthly mean column CO2 satellite measurements averaged on 8°x10° spatial scales need precision of 2.5 ppm or better to deliver improvement on surface network

OSSE for OCO observations

Conclusion: OCO will offer major improvements in uncertainty reduction over

land, while uncertainty reductions from thermal IR observations (not shown) will

be very modest but still measureable.

Houweling et al. (2004), Inverse modeling of CO2 sources and sinks using

satellite data: a synthetic inter-comparison of measurement techniques and their

performance as a function of space and time, Atmos. Chem. Phys. 4, 523–538.

Flask Inversion OCO Inversion

sx/sxpri = uncertainty reduction

R. Nassar, D.B.A. Jones, S.S. Kulawik, J.R. Worden, K.W. Bowman, R.J. Andres, P.

Suntharalingam, J.M. Chen, C.A.M. Brenninkmeijer, T.J. Schuck, T.J. Conway, D.E. Worthy (2011),

Inverse modeling of CO2 sources and sinks using satellite observations of CO2 from TES and surface

flask measurements, ACP, 11, 6029-6047.

2006Annual

mean

CO2 Fluxes by Combining TES and Flask data

TES and flask data together give the

best agreement with independent ship

(NOAA) and aircraft (CARIBIC) flask

data as a result of the complementary

vertical and horizontal information

Civil Aircraft for the Regular Investigation of the atmosphere

Based on an Instrument Container (CARIBIC)

TES and surface CO2 measurements are complementary

ship station aircraft

Nassar et al. (2011), ACPD, 11, 4263-4311

GOSAT Assimilation Results

• Maksyutov et al. (2013), Regional CO2 flux estimates for 2009-2010 based on GOSAT and ground-based CO2 observations, Atmos. Chem. Phys., in press.

• Basu et al. (2012), Global CO2 fluxes estimated from GOSAT retrievals of total column CO2, Atmos. Chem. Phys. Discuss. 13, 4535-4600.

• Deng et al. Inferring Regional Sources and Sinks of Atmospheric CO2 from GOSAT XCO2 data, submitted to Atmos. Chem. Phys. Discuss.

Future Directions

• Increase measurement coverage

• Improve model transport

• Improve model representation of all prior fluxes

• Improve data assimilation methods

• Adapt systems to accept multiple streams of data

Observing System Simulation Experiment (OSSE)

• Objective: Compare the potential information contributed for constraining CO2 surface fluxes

at Boreal and Arctic latitudes from HEO vs. LEO

• Approach:

– Run a model (GEOS-Chem) simulation to

obtain a CO2 distribution to use as the ‘Truth’

– Create ‘synthetic observations’ for a LEO mission (GOSAT) and a

HEO mission (PHEOS-WCA) by sampling the model at hypothetical

observation locations/times then adding noise

– Assimilate each set of synthetic observations to get optimized

estimates of CO2 fluxes and assess posterior flux precision and bias

relative to the ‘Truth’

Nassar, Sioris, Jones, McConnell, Satellite observations of CO2 from a Highly

Elliptical Orbit (HEO) for studies of the northern high latitude carbon cycle,

submitted to JGR

Generating Synthetic Observations

Each region/apogee: 48 scans for 100 sec each, consisting of 56x56 array of 10x10

km2 pixels. Checkerboard pattern of data-thinning to meet downlink requirement, and

observations every other repeat cycle to accommodate other observing priorities

Three-APogee (TAP) orbit

Recreated GOSAT orbit using SPENVIS (ESA), 3 cross

track obs: orbit track, ±263 km, glint subsolar±20°

2 satellites, 8h apart in co-planar 16h orbit

Apogee ~43,500 km, Perigee ~8100 km

3 Apogees/day (8:00 and 16:00 local time)

observing ±4 h from apogee giving up to

16 h of data per 48 h per region

Solar Zenith Angle, Albedo, Cloud

Spectral Albedo Values from MODTRAN

*Advanced Very High Resolution Radiometer

AVHRR* 1°x1° surface types

Signal-to-noise ratio depends on albedo of surface for each band.

Nadir XCO2 retrievals over ocean, sea-ice and old snow are assumed unlikely.

Used GOSAT v2.0 averaging kernels and covariances for these surface types, scaled

according to SNRs (reduced precision by factor of 2 over seasonal snow).

GEOS-5 Cloud Fraction

0.5x0.67

Calculated Solar Zenith Angle (SZA) for each obs and only retained when SZA < 85°

0 1

Co

mp

ari

so

n o

f C

ov

era

ge 1°x

1°

Comparison of Coverage 1°x1°: Winter/Summer

SNR Comparison to Estimate Precision

Instrument Altitude

(km)

Aperture

(cm)

Scan

Time

(s)

1.61

mm

res

(cm-1)

Assumed

albedo at

1.61 mm

SZA SNR

at

1.61

mm

Source

TANSO-FTS 665.96 6.8 4 0.20 0.3 30° > 300 Yokota et al. (2009) SOLA

60° > 228 Calculated

PHEOS-FTS

Optimal

(85 kg)

41,200

(max

43,500)

15.0 100 0.25 0.4 60° > 150 Phase-A closure report

(2012)

0.3 > 134 Calculated

PHEOS-FTS

All-band

(45 kg)

41,200

(max

43,500)

10.0 100 0.25 0.4 60° > 100 Phase-A closure report

(2012)

0.3 > 91 Calculated

7598.0)30cos()60cos(

8660.040.030.0

SZA change

Albedo change

Biospheric CO2 Flux Uncertainties

Lower annual flux uncertainties from HEO

Biospheric CO2 Flux Biases

Lower annual flux biases from HEO

Could we detect CO2 emissions from permafrost thaw?

• Simulated slow emission of CO2

from permafrost thaw: 0.2 PgC Jul-

Sep over 6 million km2 (Schuur et

al. suggest 0.8-1.1 PgC/yr by 2100)

• Generated synthetic obs from this simulation and carried out an

inversion to quantify these emissions, assuming no prior knowledge

• Although the total (biospheric + permafrost perturbation) fluxes

could be constrained within ~2% and assigned to the proper spatial

region, distinguishing permafrost emissions from background

biospheric fluxes is much like the challenge of separating biospheric

fluxes from fossil fuel emissions, making complementary

measurements (i.e. CH4, chlorophyll fluorescence, NDVI, Leaf Area

Index (LAI), soil moisture or freeze/thaw, …) extremely valuable

GEO Carbon Strategy (2010)

Recommends an Integrated Global

Carbon Observing System

including satellite and in situ / flask

GHG measurements

• Adapt GEM operational Ensemble Kalman Filter (EnKF) from EC’s weather and air quality forecast systems to CO2 (CH4) flux estimation

– Optimize both CO2 concentrations and CO2 fluxes

– Ensemble approach: perturb initial concentrations, fluxes,

meteorology, model error, etc. giving multiple ensemble

members to estimate the sizes of error components

Environment Canada Carbon Assimilation System (EC-CAS)

Led by Saroja Polavarapu (EC)

also Ray Nassar (EC), Dylan Jones (UofT)

• Forward model development – Apply best available inventories for fossil fuels (national + shipping/aviation),

biomass (biofuel) burning, and ocean flux

– Implement Canadian Terrestrial Ecosystem Model (CTEM)

• Observations – Accept multiple streams of data: atmospheric / land, remote sensing / in situ

– CO2 in situ data and satellite observations (NIR/TIR sounders)

– Implement quality control and bias correction schemes, and rigorously account

for representativeness errors for each instrument and species

GEM

Numerous other CO2 assimilation systems

• Mixture of 4DVar and EnKF methods

• CarbonTracker, CarbonTracker-EU, NASA-CMS, MPI-Jena, NIES, ….

• Intercomparison exercises (TransCom, CarboScope)

Summary and Conclusions

• CO2 flux estimation using inversion methods has slowly evolved since ~1990

• Moving beyond estimation of CO2 biospheric fluxes at sub-continental scales (~22-64) to reliable estimation of all CO2 fluxes at policy-relevant scales should eventually

be achievable by development of capabilities like those recommended in the NAS

and JASON reports:

• Important areas for improvement include: – Greatly increased CO2 data coverage with minimal biases, likely requiring a

constellation of satellite along with complementary ground-based

measurements

– Improved modelling of prior fluxes

– Implementation of sophisticated assimilation methods from weather forecasting

– Utilization of additional datasets (veg remote sensing, chlorophyll, CO, 14C, etc.)

– Improved understanding of model transport uncertainties

– Rigorous accounting/propagation of all uncertainties to the final estimates