298
Air Pollution and Turbulence: Modeling and Applications
inversions (Fan et al., 1999). However, in order to obtain valid estimates from
regularization, one needs to have one or other of the alternatives above applying. A number of workers have identifi ed regularization as an implicit Bayesian
constraint, in that it is making prior assumptions about the small-scale behavior
having low variability.
A high degree of negative autocorrelation occurs between successive zonal fl ux
estimates. This refl ects the ill-conditioning in that the smallest scales are the mostpoorly resolved. It was suggested (Enting, 2002, Section 8.2) that it may be appropriate to communicate results in terms of integrated sources. As well as having plots
that may be easier to interpret, the use of integrated fl uxes facilitates comparison of
results that have been calculated using different discretizations.
A further complication occurs in multi-tracer inversions where the signal-to-noise
characteristics of the different tracers may lead to differences in resolution. For that
example, in the study by Ciais et al. (1995b), the 13 C resolution, and the consequent
land-ocean partitioning, is poorer than the resolution of the total fl ux that derives
from the denser CO 2 data set.
11.3.3 ERRORS IN TRANSPORT MODELS
Section 9.2 of Enting (2002) addresses the issue of transport model error. Model error
is a troublesome issue across a wide range of inverse problems. One initiative for
trace gas inversions has been the TransCom model intercomparison, set up to parallel other earth system intercomparisons such as AMIP and OCMIP. In earth system
science, the term “intercomparison” has come to mean a comparison between models run under equivalent conditions, often with idealized cases designed to reveal
(and hopefully categorize) model differences.
There have been several phases of TransCom, targeted at assessing the effect of
model error on CO 2 inversions:
Phase 1: CO 2 transport This study compared “forward” calculations for the two
main source components: fossil emissions and the seasonal biotic exchange. The
results were published by Law et al. (1996) with further detail in a technical paper
(Rayner and Law, 1995). An important issue was the so-called rectifi er effect (see
Section 11.4.3).
Phase 2: SF 6 transport This was undertaken as a transport experiment using a
compound for which there were good observations and well-known sources (Denning
et al., 1999).
Phase 3: CO 2 inversions This is the “core” TransCom experiment. Gurney et al.
(2002) gave initial summary. This was followed by more detailed studies of annual
means by Gurney et al. (2003), seasonal cycles by Gurney et al. (2004), and a study
of interannual variability by Baker et al. (2006).
Other studies of interannual variability have been reported by Patra et al. (2005)
and Gurney et al. (2008). A consistent result is that interannual variability is determined more precisely than long-term means.
At the time of writing, ongoing TransCom activities are
Non-surface data looking at the role of aircraft data and similar data sets.
© 2010 by Taylor and Francis Group, LLC
Air Pollution and Turbulence: Modeling and Applications
inversions (Fan et al., 1999). However, in order to obtain valid estimates from
regularization, one needs to have one or other of the alternatives above applying. A number of workers have identifi ed regularization as an implicit Bayesian
constraint, in that it is making prior assumptions about the small-scale behavior
having low variability.
A high degree of negative autocorrelation occurs between successive zonal fl ux
estimates. This refl ects the ill-conditioning in that the smallest scales are the mostpoorly resolved. It was suggested (Enting, 2002, Section 8.2) that it may be appropriate to communicate results in terms of integrated sources. As well as having plots
that may be easier to interpret, the use of integrated fl uxes facilitates comparison of
results that have been calculated using different discretizations.
A further complication occurs in multi-tracer inversions where the signal-to-noise
characteristics of the different tracers may lead to differences in resolution. For that
example, in the study by Ciais et al. (1995b), the 13 C resolution, and the consequent
land-ocean partitioning, is poorer than the resolution of the total fl ux that derives
from the denser CO 2 data set.
11.3.3 ERRORS IN TRANSPORT MODELS
Section 9.2 of Enting (2002) addresses the issue of transport model error. Model error
is a troublesome issue across a wide range of inverse problems. One initiative for
trace gas inversions has been the TransCom model intercomparison, set up to parallel other earth system intercomparisons such as AMIP and OCMIP. In earth system
science, the term “intercomparison” has come to mean a comparison between models run under equivalent conditions, often with idealized cases designed to reveal
(and hopefully categorize) model differences.
There have been several phases of TransCom, targeted at assessing the effect of
model error on CO 2 inversions:
Phase 1: CO 2 transport This study compared “forward” calculations for the two
main source components: fossil emissions and the seasonal biotic exchange. The
results were published by Law et al. (1996) with further detail in a technical paper
(Rayner and Law, 1995). An important issue was the so-called rectifi er effect (see
Section 11.4.3).
Phase 2: SF 6 transport This was undertaken as a transport experiment using a
compound for which there were good observations and well-known sources (Denning
et al., 1999).
Phase 3: CO 2 inversions This is the “core” TransCom experiment. Gurney et al.
(2002) gave initial summary. This was followed by more detailed studies of annual
means by Gurney et al. (2003), seasonal cycles by Gurney et al. (2004), and a study
of interannual variability by Baker et al. (2006).
Other studies of interannual variability have been reported by Patra et al. (2005)
and Gurney et al. (2008). A consistent result is that interannual variability is determined more precisely than long-term means.
At the time of writing, ongoing TransCom activities are
Non-surface data looking at the role of aircraft data and similar data sets.
© 2010 by Taylor and Francis Group, LLC