11
independently. Phillips (1956) demonstrated this by designing the first successful atmospheric GCM. To allow the oceanic and atmospheric blocks to interact with each other, a
coupling module exchanges information at the air-sea interface. A coupler and the atmospheric and oceanic GCMs
together form the simplest coupled GCM (CGCM). Such a
basic CGCM lacks a number of relevant processes, relating
for example to the land and sea ice components of the climate system or the impact of vegetation. To introduce these
important aspects into the model, CGCMs are “upgraded”
with additional building blocks to form earth system models.
If a basic CGCM is a simple brick house of only one room, a
full-fledged earth system model is a mansion with specialized rooms for different tasks. Important additional building
blocks for an earth system model are modules that simulate
the behavior of sea ice, ice sheets and snow cover on land,
vegetation and other surface processes such as river runoff
into the ocean, atmospheric chemistry, biogeochemistry in
the ocean or even geological processes of varying
complexity.
In order to solve the model equations numerically,
CGCMs need to discretize the real world into finite spatial
and temporal units. The basis for such a discretization is a
three-dimensional grid of grid boxes that each contain a single value of a given variable. The CGCM applies the model
equations to the grid boxes and integrates them forward in
time. Essentially, each grid box is a mini-model that is, however, exchanging information with neighboring grid boxes.
An important characteristic of a model grid is its resolution, i.e. the size of its grid boxes.
5
It defines, among other
things, which processes can be resolved. As an example,
consider the development of cumulus clouds. While cumulus
clouds have a horizontal scale of less than 10 km, state-ofthe-art models use a resolution of about 100 km. On such a
grid CGCMs cannot simulate cumulus clouds directly.
Consequently, the climatic impacts of such clouds have to be
parameterized, i.e. their effect must be captured by the model
in a simpler way that is supported by observations. For convective
6
and mixing processes alone – important aspects of
cumulus clouds -, a number of parameterization schemes
exist that subtly alter the behavior of large-scale processes in
the models.
In addition to horizontal processes, models must be able
to capture vertical motions in the climate system. Cumulus
clouds, for example, extent vertically throughout varying
5 Note that, usually, not all grid boxes of a GCM have the same size,
neither in terms of absolute surface area, nor in terms of longitudinal
and latitudinal extent. A common practice in ocean models, for example, is to refine the latitudinal resolution towards the equator to better
resolve the fine structures of the equatorial oceans. In a similar fashion,
Sein et al. (2016) recently discussed grid layouts for ocean models that
increase their spatial resolution in certain target areas.
6 Convection: upward motion in the atmosphere.
portions of the troposphere, and vertical movement within
clouds is a key factor of precipitation. On a larger scale,
ascending air masses within the ITCZ define an important
aspect of the tropical climate system (cf. Section “The
equatorial atlantic: A climate hot spot”). Models need to be
able to reproduce these vertical movements. They require
vertical layering, giving rise to the three-dimensional structure of a model grid. A common feature of all models is that
their vertical levels are unevenly distributed. Because properties usually change drastically close to the air-sea interface, resolving these strong gradients requires a high
vertical resolution. Conversely, the thickest levels are farthest away from the air- sea interface. In the ocean, the last
model level usually ends at the sea floor; the atmosphere,
however, is not bounded that clearly. Some models only
resolve the troposphere, our “weather” sphere that reaches
up to approximately 15 km, while a number of recent atmosphere models incorporate the stratosphere as well (up to
80 km).
Figure 4 illustrates schematically how the different
“building blocks” of a CGCM work together and how the
real world must be discretized into grid boxes to allow a
numerical solution of the primitive equations.
CGCMs are initialized either from a state of rest – i.e. the
ocean and atmosphere are without motion and only establish
their general circulation patterns during the first stage of the
simulation, the so-called “spin-up” – or from a more specific
state that is generally derived from observations. In both
cases, the model needs time to smooth out initial imbalances
and establish an equilibrium. Additionally, climatically relevant forcing parameters must be prescribed to the model in
the form of boundary conditions. A prominent example of
such a boundary condition is the strength and variability of
the solar forcing, our energy source on earth, or the atmospheric CO 2 concentration.
Climate models are used to address a host of research
questions. They aid scientists in interpreting observations,
infer mechanisms, or provide information on how the climate system might evolve in the future. All of these tasks,
however, require that CGCMs are able to produce a realistic
climate. Due to various limitations, this is not always the
case. A common manifestation of the shortcomings of a climate model is the formation of biases.
A bias is a systematic difference between the modeled
and the observed climate. This difference can occur in any
statistical property of any model variable. While standard
biases are routinely monitored during the development and
application of climate models, non-obvious biases may be
present in simulations that look fine otherwise. Consider, for
example, SST in a given location. While routine bias controls may have found a realistic mean SST, closer inspection
could reveal that SST anomalies tend to be too high. Because
positive and negative anomalies cancel each other out on
Can Climate Models Simulate the Observed Strong Summer Surface Cooling in the Equatorial Atlantic?
independently. Phillips (1956) demonstrated this by designing the first successful atmospheric GCM. To allow the oceanic and atmospheric blocks to interact with each other, a
coupling module exchanges information at the air-sea interface. A coupler and the atmospheric and oceanic GCMs
together form the simplest coupled GCM (CGCM). Such a
basic CGCM lacks a number of relevant processes, relating
for example to the land and sea ice components of the climate system or the impact of vegetation. To introduce these
important aspects into the model, CGCMs are “upgraded”
with additional building blocks to form earth system models.
If a basic CGCM is a simple brick house of only one room, a
full-fledged earth system model is a mansion with specialized rooms for different tasks. Important additional building
blocks for an earth system model are modules that simulate
the behavior of sea ice, ice sheets and snow cover on land,
vegetation and other surface processes such as river runoff
into the ocean, atmospheric chemistry, biogeochemistry in
the ocean or even geological processes of varying
complexity.
In order to solve the model equations numerically,
CGCMs need to discretize the real world into finite spatial
and temporal units. The basis for such a discretization is a
three-dimensional grid of grid boxes that each contain a single value of a given variable. The CGCM applies the model
equations to the grid boxes and integrates them forward in
time. Essentially, each grid box is a mini-model that is, however, exchanging information with neighboring grid boxes.
An important characteristic of a model grid is its resolution, i.e. the size of its grid boxes.
5
It defines, among other
things, which processes can be resolved. As an example,
consider the development of cumulus clouds. While cumulus
clouds have a horizontal scale of less than 10 km, state-ofthe-art models use a resolution of about 100 km. On such a
grid CGCMs cannot simulate cumulus clouds directly.
Consequently, the climatic impacts of such clouds have to be
parameterized, i.e. their effect must be captured by the model
in a simpler way that is supported by observations. For convective
6
and mixing processes alone – important aspects of
cumulus clouds -, a number of parameterization schemes
exist that subtly alter the behavior of large-scale processes in
the models.
In addition to horizontal processes, models must be able
to capture vertical motions in the climate system. Cumulus
clouds, for example, extent vertically throughout varying
5 Note that, usually, not all grid boxes of a GCM have the same size,
neither in terms of absolute surface area, nor in terms of longitudinal
and latitudinal extent. A common practice in ocean models, for example, is to refine the latitudinal resolution towards the equator to better
resolve the fine structures of the equatorial oceans. In a similar fashion,
Sein et al. (2016) recently discussed grid layouts for ocean models that
increase their spatial resolution in certain target areas.
6 Convection: upward motion in the atmosphere.
portions of the troposphere, and vertical movement within
clouds is a key factor of precipitation. On a larger scale,
ascending air masses within the ITCZ define an important
aspect of the tropical climate system (cf. Section “The
equatorial atlantic: A climate hot spot”). Models need to be
able to reproduce these vertical movements. They require
vertical layering, giving rise to the three-dimensional structure of a model grid. A common feature of all models is that
their vertical levels are unevenly distributed. Because properties usually change drastically close to the air-sea interface, resolving these strong gradients requires a high
vertical resolution. Conversely, the thickest levels are farthest away from the air- sea interface. In the ocean, the last
model level usually ends at the sea floor; the atmosphere,
however, is not bounded that clearly. Some models only
resolve the troposphere, our “weather” sphere that reaches
up to approximately 15 km, while a number of recent atmosphere models incorporate the stratosphere as well (up to
80 km).
Figure 4 illustrates schematically how the different
“building blocks” of a CGCM work together and how the
real world must be discretized into grid boxes to allow a
numerical solution of the primitive equations.
CGCMs are initialized either from a state of rest – i.e. the
ocean and atmosphere are without motion and only establish
their general circulation patterns during the first stage of the
simulation, the so-called “spin-up” – or from a more specific
state that is generally derived from observations. In both
cases, the model needs time to smooth out initial imbalances
and establish an equilibrium. Additionally, climatically relevant forcing parameters must be prescribed to the model in
the form of boundary conditions. A prominent example of
such a boundary condition is the strength and variability of
the solar forcing, our energy source on earth, or the atmospheric CO 2 concentration.
Climate models are used to address a host of research
questions. They aid scientists in interpreting observations,
infer mechanisms, or provide information on how the climate system might evolve in the future. All of these tasks,
however, require that CGCMs are able to produce a realistic
climate. Due to various limitations, this is not always the
case. A common manifestation of the shortcomings of a climate model is the formation of biases.
A bias is a systematic difference between the modeled
and the observed climate. This difference can occur in any
statistical property of any model variable. While standard
biases are routinely monitored during the development and
application of climate models, non-obvious biases may be
present in simulations that look fine otherwise. Consider, for
example, SST in a given location. While routine bias controls may have found a realistic mean SST, closer inspection
could reveal that SST anomalies tend to be too high. Because
positive and negative anomalies cancel each other out on
Can Climate Models Simulate the Observed Strong Summer Surface Cooling in the Equatorial Atlantic?