Section 1.2: The Components of Climate Research
9
1.2.2 Numerical Experimentation
Numerical experimentation with large "nature-like" models ofthe ocean, atmosphere and other climate-relevant systems (such as sea.-ice) is the youngest
actor on the stage. After the pioneering work in the 1960's (Smagorinsky,
1963; Smagorinsky et al., 1965) with atmospheric models and ocean models
(Bryan and Cox, 1969), coupled models were developed (Manabe and Bryan,
1969). Later, numerical experiments were able to show that a significant part
of the observed atmospheric variabily can be reproduced by an atmospheric
GCM forced with 15 years of observed sea-surface temperature (Lau, 1985).
And more recently, numerical models were instrumental in revealing the potential danger of anthropogenic modifications of the ttopospheric greenhouse
gas concentrations (Manabe et al., 1994; Cubasch et al., 1992).
Numerical experimentation is also successful in attracting attention and
confidence in the scientific community - one reason is that such models appear
as tools which can answer , at least in principle, all quest ions in a "physically
consistent" manner. There is still a lot to do. The hydrodynamic part of the
models is based on first principles, such as the conservation of mass, energy
and angular moment um, but the discretization in space and time introduces
errors that cannot be assessed easily. The irreversible thermodynamic processes are described by parameterizations which should ac count for the net
effect of processes such as turbulence in the boundary layer, convection or
clouds. Discretization and the indeterminacy of our understanding of the
physical, biological and hydrological processes affect parameterizations even
more severely than the hydrodynamics. The package of these parameterizations is usually called the "physics" of the models - but this "physics" is a
mix of equations which have been fitted to satisfy first principles, intuition,
results of field campaigns or to improve the overall performance of the model
on a large scale (see, for instance, Delsol et al., 1971, Miyakoda and Sirutis,
1977 and more recently, Sirutis and Miyakoda, 1990, Tiedtke, 1986; Roeckner
et al., 1992).
1.2.3 Statistical Analysis
Statistical analysis is required for the interpretation of observed and simulated data sets. The need to use statistical techniques originates from the
large phase space of climate (see, e.g., Chapter 13). The advantage of statistical approaches is that they allow one to deal directly with information
about the real system. While certain crucial processes might be misrepresented in a model, observed data are reflecting the influence of all relevant
processes. Unfortunately, observed data reflect also all irrelevant processes,
which create the noise from which the real signal is to be discriminated.
The danger is that the discrimination between signal and noise may not
be successful. Then random features of the noise are mistaken as a true
signal - and the literat ure is full of such cases. A problem specific to the
9
1.2.2 Numerical Experimentation
Numerical experimentation with large "nature-like" models ofthe ocean, atmosphere and other climate-relevant systems (such as sea.-ice) is the youngest
actor on the stage. After the pioneering work in the 1960's (Smagorinsky,
1963; Smagorinsky et al., 1965) with atmospheric models and ocean models
(Bryan and Cox, 1969), coupled models were developed (Manabe and Bryan,
1969). Later, numerical experiments were able to show that a significant part
of the observed atmospheric variabily can be reproduced by an atmospheric
GCM forced with 15 years of observed sea-surface temperature (Lau, 1985).
And more recently, numerical models were instrumental in revealing the potential danger of anthropogenic modifications of the ttopospheric greenhouse
gas concentrations (Manabe et al., 1994; Cubasch et al., 1992).
Numerical experimentation is also successful in attracting attention and
confidence in the scientific community - one reason is that such models appear
as tools which can answer , at least in principle, all quest ions in a "physically
consistent" manner. There is still a lot to do. The hydrodynamic part of the
models is based on first principles, such as the conservation of mass, energy
and angular moment um, but the discretization in space and time introduces
errors that cannot be assessed easily. The irreversible thermodynamic processes are described by parameterizations which should ac count for the net
effect of processes such as turbulence in the boundary layer, convection or
clouds. Discretization and the indeterminacy of our understanding of the
physical, biological and hydrological processes affect parameterizations even
more severely than the hydrodynamics. The package of these parameterizations is usually called the "physics" of the models - but this "physics" is a
mix of equations which have been fitted to satisfy first principles, intuition,
results of field campaigns or to improve the overall performance of the model
on a large scale (see, for instance, Delsol et al., 1971, Miyakoda and Sirutis,
1977 and more recently, Sirutis and Miyakoda, 1990, Tiedtke, 1986; Roeckner
et al., 1992).
1.2.3 Statistical Analysis
Statistical analysis is required for the interpretation of observed and simulated data sets. The need to use statistical techniques originates from the
large phase space of climate (see, e.g., Chapter 13). The advantage of statistical approaches is that they allow one to deal directly with information
about the real system. While certain crucial processes might be misrepresented in a model, observed data are reflecting the influence of all relevant
processes. Unfortunately, observed data reflect also all irrelevant processes,
which create the noise from which the real signal is to be discriminated.
The danger is that the discrimination between signal and noise may not
be successful. Then random features of the noise are mistaken as a true
signal - and the literat ure is full of such cases. A problem specific to the