and 940°N), where the models gave much higher
inventories than the ⌬C* calculations.
There is to date no basin-wide estimate of
Pacific inventories based on the recent WOCE
data, although this is presently being prepared (R.
Feely, personal communication). A rough estimate
of the Pacific contribution based on an earlier version of a preformed-CO 2 estimation technique and
various historical data sets gave a total of about
31.3 PgC for 1980 (Chen, 1993). When scaled similarly to the Atlantic estimates given above this suggests an inventory of about 37 PgC in 1989. (Note
that this estimate is presented only to allow a
rough comparison with the other basin inventories
and is subject to revision based on the WOCE data,
using improved analysis techniques.)
Summary
The differences and uncertainties associated with
the various Excess CO 2 separation techniques (e.g.
those highlighted by the discussion of the Indian
Ocean I1 section distributions) emphasize the
model and assumption sensitivity of Excess CO 2
separation. The factors underlying Excess CO 2 estimation differences, and, particularly, which calculation approach gives a better indication of the ‘true’
values, remain unclear (Coatanoan et al., 2001).
The identification and resolution of such differences
is important, however, if the goal of using such distributions to guide model improvement is to be
attained. As Orr et al. (2000) has pointed out: ‘we
validate model predictions to understand where
they may be wrong and how to improve them; we
should do the same for data-based estimates such as
those for anthropogenic CO 2 … ’. There may be significant utility in attempting to evaluate the various
Excess CO 2 calculation techniques using modelgenerated data as input. The current generation of
three-dimensional ocean carbon GCMs could be
used to generate known target fields of Excess and
natural carbon. Effects of spatial and temporal variability, gas exchange and complex mixing could be
simulated in the creation of these fields. Finally, sets
of ‘model data’ similar to the data collected during
the CO 2 Survey could be generated and the various
Excess CO 2 techniques applied in order to reconstruct the known or target fields. In this way, the
relative ability of various approaches could be evaluated. An analysis of this type is currently underway for the Indian Ocean (C. Sabine, personal
communication).
6.3.4.2 Time series comparison methods
Background
The preformed methods described above seek to
estimate the total oceanic inventory of Excess CO 2 :
i.e. all the extra carbon added since about 1750.
Several fundamental problems with most such
approaches exist that can introduce significant
uncertainty into the estimation of Excess CO 2 .
Wallace (1995) showed that evaluation of the
Excess CO 2 increase through the direct comparison
of TCO 2 data collected over decadal or longer
intervals is now possible. Specifically changes in
TCO 2 that are not associated with analytical error
or natural variability in other physical or biogeochemical factors (e.g. temperature, salinity, AOU,
alkalinity) can be attributed to the build-up of
Excess TCO 2 . In any such comparison, variability
of these parameters is encountered and has to be
corrected for before the Excess CO 2 change can be
identified. Two separate such approaches have to
date been applied to WOCE data sets.
A multivariate approach
Wallace (1995) introduced an empirical, multivariate technique to remove the influence of natural
variability on TCO 2 concentration differences. In
order to compare two data sets collected at different times, the following steps were proposed:
¥ a multivariate predictor equation is established
based on regional TCO 2 data collected at one
point in time (T1), through regression with
‘oceanographically related’ predictor variables
such as potential temperature, AOU, salinity,
total alkalinity, nutrient concentrations, etc. On
basin scales these regressions explain most of
the observed subsurface TCO 2 variance other
than that arising from analytical error.
¥ This predictor equation is applied to independent variables as they were measured at a different point in time (T2, e.g. 10 years earlier or
later) and a TCO 2 distribution is predicted.
¥ The predicted TCO 2 distribution, calculated
from the T1 predictor equation and the T2
independent variables, is compared with the T2
observed TCO 2 concentrations and the residuals
calculated.
¥ The multivariate predictor equation is assumed
to account for natural TCO 2 variability (e.g.
that associated with changes in temperature,
dissolved oxygen, salinity, etc.). The residuals
6.3 Storage and Transport of Excess CO 2 in the Oceans
511
Wallace
inventories than the ⌬C* calculations.
There is to date no basin-wide estimate of
Pacific inventories based on the recent WOCE
data, although this is presently being prepared (R.
Feely, personal communication). A rough estimate
of the Pacific contribution based on an earlier version of a preformed-CO 2 estimation technique and
various historical data sets gave a total of about
31.3 PgC for 1980 (Chen, 1993). When scaled similarly to the Atlantic estimates given above this suggests an inventory of about 37 PgC in 1989. (Note
that this estimate is presented only to allow a
rough comparison with the other basin inventories
and is subject to revision based on the WOCE data,
using improved analysis techniques.)
Summary
The differences and uncertainties associated with
the various Excess CO 2 separation techniques (e.g.
those highlighted by the discussion of the Indian
Ocean I1 section distributions) emphasize the
model and assumption sensitivity of Excess CO 2
separation. The factors underlying Excess CO 2 estimation differences, and, particularly, which calculation approach gives a better indication of the ‘true’
values, remain unclear (Coatanoan et al., 2001).
The identification and resolution of such differences
is important, however, if the goal of using such distributions to guide model improvement is to be
attained. As Orr et al. (2000) has pointed out: ‘we
validate model predictions to understand where
they may be wrong and how to improve them; we
should do the same for data-based estimates such as
those for anthropogenic CO 2 … ’. There may be significant utility in attempting to evaluate the various
Excess CO 2 calculation techniques using modelgenerated data as input. The current generation of
three-dimensional ocean carbon GCMs could be
used to generate known target fields of Excess and
natural carbon. Effects of spatial and temporal variability, gas exchange and complex mixing could be
simulated in the creation of these fields. Finally, sets
of ‘model data’ similar to the data collected during
the CO 2 Survey could be generated and the various
Excess CO 2 techniques applied in order to reconstruct the known or target fields. In this way, the
relative ability of various approaches could be evaluated. An analysis of this type is currently underway for the Indian Ocean (C. Sabine, personal
communication).
6.3.4.2 Time series comparison methods
Background
The preformed methods described above seek to
estimate the total oceanic inventory of Excess CO 2 :
i.e. all the extra carbon added since about 1750.
Several fundamental problems with most such
approaches exist that can introduce significant
uncertainty into the estimation of Excess CO 2 .
Wallace (1995) showed that evaluation of the
Excess CO 2 increase through the direct comparison
of TCO 2 data collected over decadal or longer
intervals is now possible. Specifically changes in
TCO 2 that are not associated with analytical error
or natural variability in other physical or biogeochemical factors (e.g. temperature, salinity, AOU,
alkalinity) can be attributed to the build-up of
Excess TCO 2 . In any such comparison, variability
of these parameters is encountered and has to be
corrected for before the Excess CO 2 change can be
identified. Two separate such approaches have to
date been applied to WOCE data sets.
A multivariate approach
Wallace (1995) introduced an empirical, multivariate technique to remove the influence of natural
variability on TCO 2 concentration differences. In
order to compare two data sets collected at different times, the following steps were proposed:
¥ a multivariate predictor equation is established
based on regional TCO 2 data collected at one
point in time (T1), through regression with
‘oceanographically related’ predictor variables
such as potential temperature, AOU, salinity,
total alkalinity, nutrient concentrations, etc. On
basin scales these regressions explain most of
the observed subsurface TCO 2 variance other
than that arising from analytical error.
¥ This predictor equation is applied to independent variables as they were measured at a different point in time (T2, e.g. 10 years earlier or
later) and a TCO 2 distribution is predicted.
¥ The predicted TCO 2 distribution, calculated
from the T1 predictor equation and the T2
independent variables, is compared with the T2
observed TCO 2 concentrations and the residuals
calculated.
¥ The multivariate predictor equation is assumed
to account for natural TCO 2 variability (e.g.
that associated with changes in temperature,
dissolved oxygen, salinity, etc.). The residuals
6.3 Storage and Transport of Excess CO 2 in the Oceans
511
Wallace