to marine observations or to atmospheric analyses
of surface conditions appear to have uncertainties
of 30 W m
92
. The causes of such uncertainty are
unclear but recent work suggests that the amount
of heat loss over strong western boundary currents
may be significantly underestimated in the flux climatologies (Josey et al., 1999). Despite substantial
improvements in the observational databases and
in the flux parameterizations over the past two
decades, the conclusion remains the same as that
delivered by the Cage Group (Bretherton et al.,
1982): heat fluxes computed by bulk formula are
uncertain by 25–30 W m
92 . While such uncertainty may be tolerable for seasonal or local
climate analyses, an uncertainty of 30 W m
92 in
air–sea heat exchange means that an integration of
fluxes over an area 4000 km by 4000 km will
result in an uncertainty in ocean heat transport
divergence of 0.5 PW; and larger areas will have
proportionately larger uncertainties.
6.1.3 Residual method estimates of
ocean heat transport
Until recently, the residual method of subtracting
atmospheric energy transport from the radiation requirement at the top of the atmosphere
was applied exclusively to zonal averages (e.g.
Carissimo et al., 1985). While the errors in incoming and outgoing radiation are about 5 W m
92
, and
because the globally averaged incoming and outgoing radiation balance within 10 W m
92
, a major
uncertainty in the residual method is considered to
be in the atmospheric energy transport estimates.
In several respects, the atmospheric energy transport derived from analysis of the observational
rawinsonde network by Oort and collaborators
(Carissimo et al., 1985) seems too small. First,
subtracting such atmospheric transport from the
radiation requirement yielded ocean heat transports that seem unreasonably large (Talley, 1984;
Bryden et al., 1991). Second, comparisons with
Geophysical Fluid Dynamics Laboratory (GFDL)
model synthesis of the enhanced First GARP
(Global Atmospheric Research Programme)
Global Experiment (FGGE) observational data set
and with a year-long European Centre for
Medium Range Weather Forecasts (ECMWF)
model synthesis for 1986 showed that the rawinsonde analysis yielded maximum poleward atmospheric transports about 1 PW smaller than the
model syntheses (Masuda, 1988; Michaud and
Derome, 1991). Thus, ocean heat transports
derived from the residual method have been uncertain by at least 1 PW due to the uncertainty in
atmospheric energy transport.
Trenberth and Solomon (1994) made a major
advance with the residual method by applying it
pointwise, subtracting atmospheric flux divergence
from the net radiation at the top of the atmosphere,
to derive the spatial distribution of surface heat
flux. While Trenberth and Solomon used Earth
Radiation Budget Experiment (ERBE) radiation
data and ECMWF analyses from a single year
(1988), Keith (1995) applied similar techniques
using nearly four years of ERBE data (1985–88),
three years of ECMWF analyses (1989–92) and one
year of National Meteorological Center (NMC)
analyses (1991–92). The patterns of derived surface
heat flux over the open ocean are visually similar to
maps of air–sea heat exchange derived from bulk
formulae (Fig. 6.1.5). But there are also clear problems in the patterns of residual surface heat flux.
Most notably, the annual average surface heat flux
over land is not zero; instead, it exhibits fluctuations of up to <100 W m
92
, with a suggested error
of 30 W m
92 (Trenberth and Solomon, 1994). Such
an error in the residual method would be of the
same magnitude as that in bulk formula estimates
of air–sea exchange. Both Trenberth and Solomon
and Keith argue that the large errors in surface heat
flux over land may be due to problems in representing orographic effects in the atmospheric models
that lead to dipoles of strong positive and negative
fluxes near topography, and that the surface heat
flux over the relatively smooth oceans may be more
accurately represented. While possibly true over the
open ocean, there are indications that land effects
invade the ocean along some boundaries, particularly off South America.
Trenberth and Solomon (1994) suggested that
residual method ocean heat transport estimates
could be made by masking the land and integrating
the surface heat flux just over ocean areas. Their
integration of surface heat flux over the entire
ocean yielded a net heating of the oceans
by 0.7 PW, suggesting a globally averaged bias
toward heating the ocean of only 2 W m
92
, which
is substantially smaller than the global bias in bulk
formula estimates of air–sea exchange and substantially smaller than the radiation uncertainties of 5
and 10 W m
92 referred to above. Both Trenberth
SECTION 6 LARGE-SCALE OCEAN TRANSPORTS
458
of surface conditions appear to have uncertainties
of 30 W m
92
. The causes of such uncertainty are
unclear but recent work suggests that the amount
of heat loss over strong western boundary currents
may be significantly underestimated in the flux climatologies (Josey et al., 1999). Despite substantial
improvements in the observational databases and
in the flux parameterizations over the past two
decades, the conclusion remains the same as that
delivered by the Cage Group (Bretherton et al.,
1982): heat fluxes computed by bulk formula are
uncertain by 25–30 W m
92 . While such uncertainty may be tolerable for seasonal or local
climate analyses, an uncertainty of 30 W m
92 in
air–sea heat exchange means that an integration of
fluxes over an area 4000 km by 4000 km will
result in an uncertainty in ocean heat transport
divergence of 0.5 PW; and larger areas will have
proportionately larger uncertainties.
6.1.3 Residual method estimates of
ocean heat transport
Until recently, the residual method of subtracting
atmospheric energy transport from the radiation requirement at the top of the atmosphere
was applied exclusively to zonal averages (e.g.
Carissimo et al., 1985). While the errors in incoming and outgoing radiation are about 5 W m
92
, and
because the globally averaged incoming and outgoing radiation balance within 10 W m
92
, a major
uncertainty in the residual method is considered to
be in the atmospheric energy transport estimates.
In several respects, the atmospheric energy transport derived from analysis of the observational
rawinsonde network by Oort and collaborators
(Carissimo et al., 1985) seems too small. First,
subtracting such atmospheric transport from the
radiation requirement yielded ocean heat transports that seem unreasonably large (Talley, 1984;
Bryden et al., 1991). Second, comparisons with
Geophysical Fluid Dynamics Laboratory (GFDL)
model synthesis of the enhanced First GARP
(Global Atmospheric Research Programme)
Global Experiment (FGGE) observational data set
and with a year-long European Centre for
Medium Range Weather Forecasts (ECMWF)
model synthesis for 1986 showed that the rawinsonde analysis yielded maximum poleward atmospheric transports about 1 PW smaller than the
model syntheses (Masuda, 1988; Michaud and
Derome, 1991). Thus, ocean heat transports
derived from the residual method have been uncertain by at least 1 PW due to the uncertainty in
atmospheric energy transport.
Trenberth and Solomon (1994) made a major
advance with the residual method by applying it
pointwise, subtracting atmospheric flux divergence
from the net radiation at the top of the atmosphere,
to derive the spatial distribution of surface heat
flux. While Trenberth and Solomon used Earth
Radiation Budget Experiment (ERBE) radiation
data and ECMWF analyses from a single year
(1988), Keith (1995) applied similar techniques
using nearly four years of ERBE data (1985–88),
three years of ECMWF analyses (1989–92) and one
year of National Meteorological Center (NMC)
analyses (1991–92). The patterns of derived surface
heat flux over the open ocean are visually similar to
maps of air–sea heat exchange derived from bulk
formulae (Fig. 6.1.5). But there are also clear problems in the patterns of residual surface heat flux.
Most notably, the annual average surface heat flux
over land is not zero; instead, it exhibits fluctuations of up to <100 W m
92
, with a suggested error
of 30 W m
92 (Trenberth and Solomon, 1994). Such
an error in the residual method would be of the
same magnitude as that in bulk formula estimates
of air–sea exchange. Both Trenberth and Solomon
and Keith argue that the large errors in surface heat
flux over land may be due to problems in representing orographic effects in the atmospheric models
that lead to dipoles of strong positive and negative
fluxes near topography, and that the surface heat
flux over the relatively smooth oceans may be more
accurately represented. While possibly true over the
open ocean, there are indications that land effects
invade the ocean along some boundaries, particularly off South America.
Trenberth and Solomon (1994) suggested that
residual method ocean heat transport estimates
could be made by masking the land and integrating
the surface heat flux just over ocean areas. Their
integration of surface heat flux over the entire
ocean yielded a net heating of the oceans
by 0.7 PW, suggesting a globally averaged bias
toward heating the ocean of only 2 W m
92
, which
is substantially smaller than the global bias in bulk
formula estimates of air–sea exchange and substantially smaller than the radiation uncertainties of 5
and 10 W m
92 referred to above. Both Trenberth
SECTION 6 LARGE-SCALE OCEAN TRANSPORTS
458