The Averaging Kernel of CO2 Column Measurements by the Orbiting Carbon Observatory (OCO), Its Use in Inverse Modeling, and Comparisons to AIRS, SCIAMACHY, and Ground-based FTIR Brian Connor1, Zhiming Kuang2, Geoff Toon3, David Crisp3, Stephen Wood1,

1
The Averaging Kernel of CO2 Column Measurements
by the Orbiting Carbon Observatory (OCO), Its Use
in Inverse Modeling, and Comparisons to AIRS,
SCIAMACHY, and Ground-based FTIRBrian Connor1,
Zhiming Kuang2, Geoff Toon3, David Crisp3,
Stephen Wood1, Chris Barnet4, and Michael
Buchwitz51 National Institute of Water and
Atmospheric Research, Lauder, New Zealand2
University of Washington3 Jet Propulsion
Laboratory, Pasadena, CA4 5 University of
Bremen, Bremen, Germany
The Orbiting Carbon Observatory (OCO) is planned
as the first satellite instrument dedicated
exclusively to measurement of CO2 for studies of
the carbon cycle. It will measure the column
weighted mixing ratio of CO2 , XCO2, at high
spatial resolution (3 km2) with a global repeat
cycle of 16 days. If such measurements can be
made at very high precision, they can be used by
models of global emission and transport to infer
CO2 sources and sinks on a global scale. For this
reason, OCO will target a precision of 1 ppm
(0.3) for monthly regional averages, throughout
the globe. Such high precision places
unprecedented demands on both instrument and
analysis. Our purpose here is to assess the
uncertainties introduced by non-uniform
sensitivity of the measurement as a function of
height (i.e. imperfect averaging kernels), and to
show how these uncertainties can be mitigated in
use of the data. The sensitivity of a
space-based CO2 measurement varies with height
due to the physics of spectroscopy and radiative
transfer interacting with atmospheric properties,
and to instrument characteristics such as the
spectral bandpass, resolution, and noise. Thus it
is necessary to take the profile shape and
variability with height into account in
retrieving the column and in comparing it to
models and to other measurements for
validation. The sensitivity as a function of
height may be described by the averaging kernel
ac   ac ?XCO2/?x(z) F(K,S?,Sa)   where x(z)
is the true CO2 profile, S? is the covariance
matrix of the measurements, Sa is the covariance
of the a priori CO2 profile, and   K ?y/?x for
spectral measurements y. K is the Jacobian or
weighting function matrix. The variance of
smoothing error results in a column measurement
is   ss2 (ac -1) T Se (ac -1)   where Se is
the covariance of the real atmosphere. Note that
this is a source of error which is inevitable
given a variable atmosphere in the presence of an
imperfect averaging kernel. Our best current
estimate of a global covariance for CO2 was
derived from a multi-year series of NOAA aircraft
data acquired at Carr, CO (P. Tans, private
communication), which has then been scaled to
provide a standard deviation in XCO2 of 12 ppm
(3 Dufour Breon, 2003).
The OCO averaging kernel depends on the solar
zenith angle, surface albedo, and aerosol optical
depth, as illustrated below. Variations in the
kernel are much larger at high zenith angles,
where the signal-to-noise ratio is poorer.
Zenith angle 35
85 Albedo .06 .06 Aerosol t
.05 .2 .05 .2 Error (ppm) Noise .52 .52 2.
52 2.15 Smoothing .33 .30 1.00 1.33 Total .63 .
59 2.70 2.52 Albedo .24
.24 Aerosol t .05 .2 .05 .2 Error
(ppm) Noise .26 .26 .93 1.04 Smoothing .15 .15
.44 .41 Total .30 .30 1.04 1.11
Projected CO2 Error (ppm) Estimated
uncertainties in XCO2 are shown to the right for
representative zenith angles, albedo, and optical
depth. Noise and smoothing error will have
significantly different effects. While noise
errors are truly random, smoothing error, due to
the imperfect averaging kernels shown above, is
likely to be geographically coherent, and thus
result in systematic biases. Uncompensated, these
could have serious effects on source/sink
determination.