aspect that is common to TOGA and WOCE is the
overriding importance of having access to highquality winds. TOGA addressed this issue by supporting an array of moorings in the tropical Pacific
(TAO; McPhaden, 1995) that would provide accurate estimates of the surface wind forcing in the
equatorial Pacific. As noted previously, the WOCE
strategy sought global estimates. In order to use
the wind data effectively, better quantification of
the errors must be obtained. In the more sophisticated applications, this information on errors
allows a more consistent blending of the assimilated surface information (including the wind forcing) with subsurface and other data.
An outstanding question concerns the relative
importance of various data sets for climate prediction (Fig. 7.4.2). At this time, the greatest value for
ENSO prediction is attached to the tropical mooring array, but further research is required to quantify better the relative impact of each data set. For
example, the results of Roemmich and Gilson
(2000) demonstrate the synergy of XBT measurements and altimetry. In the tropical region, can
altimetric measurements reduce some of the subsurface sampling requirements for climate prediction? Will the next generation surface wind
scatterometer data, and improved models, also
lessen the need for in-situ data? Conversely, we
might argue that the combination of Argo, mooring data and XBTs provides for much of the
requirements and that remote sensing should be
regarded as partly redundant. These are all difficult issues and will require substantial further
research.
7.4.4.4 Beyond seasonal-to-interannual
prediction
I will not dwell to any great extent on climate prediction and variability problems beyond seasonalto-interannual prediction (see WCRP, 1998, for a
discussion), either on shorter time scales or at
longer time scales. However, it is useful to consider the approach that is being adopted for
intraseasonal-to-seasonal problems. While the
focus is on process studies and, to a large extent,
surface and non-oceanographic observations, the
legacy from TOGA and WOCE is being promoted
as a needed background for these studies. In other
words, by preserving the effective components
of these observing systems (such as the SOOP in
the Indian Ocean), CLIVAR is able to develop
research strategies without the need to ‘reinvent’
the basic observing system. Using Fig. 7.4.1 to picture this schematically, we can envisage a range of
process studies being placed ‘on top’ of the combined basic and CLIVAR observing systems; the
process studies can be planned and implemented
safe in the knowledge that basic requirements are
being met. The WOCE data and their continuation
provide a baseline for future studies of variability
and prediction, at both shorter and longer time
scales (Section 7.4.5).
7.4.5 The mean state and long-term
change
Interest in climate change issues intensified
through the 1980s and the prospect of climate
change influenced the design and establishment of
WOCE. The development and publication of the
First Assessment Report of the Intergovernmental
Panel on Climate Change (IPCC, 1990) coincided
with the initial implementation phase of WOCE. It
was inevitable that WOCE was influenced by the
First Assessment and that subsequent Assessments
were shaped to some extent by WOCE results.
The principal focus of this chapter is prediction
of climate and the related observations needed for
initialization, and it is useful to extend this examination to longer time scales and to predictions on
decadal-to-centennial time scales. In theory, the
principal role of data assimilation is to constrain
(correct) the model evolution so that it more accurately mimics variations of the real world. In practice, however, the unconstrained mean state of
seasonal prediction models is usually far from reality, so the most immediate impact of the data is
often correcting the mean state of the ocean
(Anderson et al., 2001).
We might anticipate that knowledge of the
mean state would also be important for prediction
and estimation of long-period variations. Indeed,
recent research has highlighted the importance of
the initialization procedure for climate change
integrations (Bryan, 1998; Stouffer and Dixon,
1998). Bryan (1998) found the NCAR climate
model adjusted on two time scales after initialization: a ‘fast’ mode, where the upper ocean, atmosphere and sea ice sought equilibrium, and a slow
‘drift’ mode that appeared to involve the slow
adjustment of the deep ocean. Gordon et al. (2000)
successfully initialized the Hadley Centre coupled
7.4 Ocean and Climate Prediction – the WOCE Legacy
595
Smith
overriding importance of having access to highquality winds. TOGA addressed this issue by supporting an array of moorings in the tropical Pacific
(TAO; McPhaden, 1995) that would provide accurate estimates of the surface wind forcing in the
equatorial Pacific. As noted previously, the WOCE
strategy sought global estimates. In order to use
the wind data effectively, better quantification of
the errors must be obtained. In the more sophisticated applications, this information on errors
allows a more consistent blending of the assimilated surface information (including the wind forcing) with subsurface and other data.
An outstanding question concerns the relative
importance of various data sets for climate prediction (Fig. 7.4.2). At this time, the greatest value for
ENSO prediction is attached to the tropical mooring array, but further research is required to quantify better the relative impact of each data set. For
example, the results of Roemmich and Gilson
(2000) demonstrate the synergy of XBT measurements and altimetry. In the tropical region, can
altimetric measurements reduce some of the subsurface sampling requirements for climate prediction? Will the next generation surface wind
scatterometer data, and improved models, also
lessen the need for in-situ data? Conversely, we
might argue that the combination of Argo, mooring data and XBTs provides for much of the
requirements and that remote sensing should be
regarded as partly redundant. These are all difficult issues and will require substantial further
research.
7.4.4.4 Beyond seasonal-to-interannual
prediction
I will not dwell to any great extent on climate prediction and variability problems beyond seasonalto-interannual prediction (see WCRP, 1998, for a
discussion), either on shorter time scales or at
longer time scales. However, it is useful to consider the approach that is being adopted for
intraseasonal-to-seasonal problems. While the
focus is on process studies and, to a large extent,
surface and non-oceanographic observations, the
legacy from TOGA and WOCE is being promoted
as a needed background for these studies. In other
words, by preserving the effective components
of these observing systems (such as the SOOP in
the Indian Ocean), CLIVAR is able to develop
research strategies without the need to ‘reinvent’
the basic observing system. Using Fig. 7.4.1 to picture this schematically, we can envisage a range of
process studies being placed ‘on top’ of the combined basic and CLIVAR observing systems; the
process studies can be planned and implemented
safe in the knowledge that basic requirements are
being met. The WOCE data and their continuation
provide a baseline for future studies of variability
and prediction, at both shorter and longer time
scales (Section 7.4.5).
7.4.5 The mean state and long-term
change
Interest in climate change issues intensified
through the 1980s and the prospect of climate
change influenced the design and establishment of
WOCE. The development and publication of the
First Assessment Report of the Intergovernmental
Panel on Climate Change (IPCC, 1990) coincided
with the initial implementation phase of WOCE. It
was inevitable that WOCE was influenced by the
First Assessment and that subsequent Assessments
were shaped to some extent by WOCE results.
The principal focus of this chapter is prediction
of climate and the related observations needed for
initialization, and it is useful to extend this examination to longer time scales and to predictions on
decadal-to-centennial time scales. In theory, the
principal role of data assimilation is to constrain
(correct) the model evolution so that it more accurately mimics variations of the real world. In practice, however, the unconstrained mean state of
seasonal prediction models is usually far from reality, so the most immediate impact of the data is
often correcting the mean state of the ocean
(Anderson et al., 2001).
We might anticipate that knowledge of the
mean state would also be important for prediction
and estimation of long-period variations. Indeed,
recent research has highlighted the importance of
the initialization procedure for climate change
integrations (Bryan, 1998; Stouffer and Dixon,
1998). Bryan (1998) found the NCAR climate
model adjusted on two time scales after initialization: a ‘fast’ mode, where the upper ocean, atmosphere and sea ice sought equilibrium, and a slow
‘drift’ mode that appeared to involve the slow
adjustment of the deep ocean. Gordon et al. (2000)
successfully initialized the Hadley Centre coupled
7.4 Ocean and Climate Prediction – the WOCE Legacy
595
Smith