XCO2 Retrieval Algorithms

1
XCO2 Retrieval Algorithms OCO Science Team
L.Brown, B. Connor, D. Crisp, S. Doney, I. Fung,
D. Jacob, C. Miller, D. OBrian, S.Pawson, J.
Randerson, P. Rayner, R. Salawitch, S.Sander, G.
Stephens, P. Tans, G. Toon, P. Wennberg, S.
Wofsy, and Y.Yung
Standard Algorithm Performance Achievable
Precision of XCO2

Realistic, end-to-end, Observational System
Simulation Experiments have been performed using
a prototype retrieval algorithm to determine the
achievable precision of XCO2 by OCO. The
prototype retrieval algorithm performed well. It
was stable and the results agree well with those
predicted by linear analysis. This is expected
given the weakly nonlinear nature of this
retrieval problem. Based on these experiments, we
conclude that the achievable precision from a
single clear sky nadir sounding (?lt0.3) is
0.3-2.5 ppm. The actual value depends on the
surface albedo, aerosol loading, and solar zenith
angle (SZA). Sun glint observations over the
ocean will yield higher SNR and smaller errors
than the nadir observations.
As an example, we show on the left results from
the prototype retrieval algorithm for 750
individual nadir soundings simulated using
baseline OCO flight instrument specifications.
Atmospheric and surface properties were varied
randomly over the full range of expected values,
with sub-visual cirrus and thin to moderate
aerosols of different types, including sulfate,
dust, and soluble organic aerosols, for SZA of
35 and 75. INSET Distribution of XCO2 errors
(ppm) for each case.
Alternative Retrieval Method (by Regression)
Physics behind the Achievable Precision
Advantage and Disadvantage
While an inverse problem can be generally written
as XCO2F-1(I1,I2,.In), where F-1 represents
the inverse operation, and Ii is the measured
radiance at wavelength i. It can sometimes be
approximated by g(XCO2)?aif(Ii) where g and f
and known simple functions, and the coefficients
ai can be determined by fitting the data where
XCO2 and the spectra are simultaneously available
from either validation campaigns or radiatve
transfer models. This method has been used
successfully to estimate surface pressure from
O2-A band. A simple model is explored
here XCO2?aiIi where Ii is the normalized
radiance. More sophisicated forms for g and f are
being explored. Potential use of neural network
in learning the functional forms from data is
being considered. Ref1 OBrien

• After the coefficients are obtained, the
  operational retrieval is simply a vector
  multiplication and is extremely efficient. No
  radiative transfer model calculation is required.
• Not as physically based as the standard one and
  yield slightly lower precision.

• Simultaneous retrieval of 0.76, 1.58 and 2.06 ?m
  spectra effectively separates CO2 column changes
  from changes in the other state variables.
  Broadly speaking
• CO2 1.58 ?m band is well suited for retrieving
  CO2 column abundances
• CO2 2.0 ?m band CO2, cloud, aerosol, water vapor,
  temperature
• O2 A-band spectrometer constrains clouds,
  aerosols, and surface pressure
• In particular, the simultaneous use of the strong
  CO2 2.0 ?m and O2 A-bands constrains the
  wavelength dependent optical properties of
  aerosols

Standard retrieval Regression retrieval
Land, ?0.1
Ocean, ?0.1
Ocean, ?0.2
Performance