Simulation and
Understanding in the
Study of Weather and
Climate
Wendy S. Parker
Durham University
wendy.parker@durham.ac.uk
In the study of weather and climate, the digital computer has allowed scien-
tists to make existing theory more useful, both for prediction and for under-
standing. After characterizing two sorts of understanding commonly sought
by scientists in this arena, I show how the use of the computer to (i) generate
surrogate observational data, (ii) test physical hypotheses and (iii) experi-
ment on models has helped to advance such understanding in signiªcant
ways.
1. Introduction
In 1904, Norwegian physicist Vilhelm Bjerknes published what would
become a landmark paper in the history of meteorology. In that paper, he
proposed that daily weather forecasts could be made by calculating later
states of the atmosphere from an earlier state using the laws of hydrody-
namics and thermodynamics (Bjerknes 1904). He outlined a set of differ-
ential equations to be solved and advocated the development of graphical
and numerical solution methods, since analytic solution was out of the
question.
Using these theory-based equations to produce daily forecasts, however,
turned out to be more difªcult than anticipated. Graphical solution tech-
niques had limited success, and a ªrst attempt to use numerical (ªnite-
difference) methods gave little reason for optimism: it took Lewis Fry
Richardson (1922) more than a month to calculate by hand the six-hour
forecast for a small region, and the results produced were wildly unrealis-
tic. Writing in 1955, atmospheric scientist Jule Charney characterized dy-
namical meteorology in the ªrst half of the twentieth century as “a ªeld in
which belief in a theory was often more a matter of faith than of
Perspectives on Science 2014, vol. 22, no. 3
©2014 by The Massachusetts Institute of Technology
doi:10.1162/POSC_a_00137
336
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
337
experience [. . .] the practicing meteorologist could ignore the results of
theory with good conscience” (1955, p. 798).
The advent of the digital computer in the mid-twentieth century
brought new hope for making theory useful. Perhaps accurate numerical
weather prediction would be possible after all: the computer could per-
form rapidly the calculations required by numerical solution techniques,
and mathematical models incorporating reformulated (“ªltered”) versions
of the theoretical equations might avoid some of the problems that af-
fected Richardson’s forecast.1 But improving weather forecasts was not the
only goal. Scientists also had high hopes for the computer as a device that
would allow them to advance their understanding of the atmosphere and
climate system, in part by allowing them to investigate in new ways the
mechanisms by which salient features and phenomena are produced (see
e.g. Charney 1951, 1955, and quotes below from Lorenz 1967, 1970).
In the half century since its introduction, the computer has proven
valuable indeed for both prediction and understanding in this arena. Of-
ten, progress in weather prediction is emphasized, but in what follows I
discuss how the computer—especially computer simulation—has helped
with the latter goal: understanding. Rather than explore the nature of
scientiªc understanding in general (e.g., de Regt and Dieks 2005, Grimm
2006, Strevens 2008, Ylikoski and Kuorokoski 2010), I show how in
practice computer simulation can promote two sorts of understanding
commonly sought in the study of weather and climate.2
In Section 2, I introduce these two sorts of understanding. Understand-
ing why an event/phenomenon occurs is achieved when scientists obtain an ac-
curate explanation of the occurrence of that event or phenomenon, while
understanding a complex system/phenomenon is more open-ended and involves
both knowledge and know-how. The next three sections discuss particular
ways in which the computer is used to increase, or make progress toward,
these sorts of understanding. Section 3 is concerned with the use of the
computer to produce simulations that serve as surrogate observational
data. Section 4 outlines how the testing of physical hypotheses, especially
hypotheses relevant to explanation, can be performed with the help of the
1. Richardson himself had identiªed some of the problems with his attempt and had
envisioned improved forecasting by numerical methods with the help of thousands of hu-
man “computers,” each responsible for performing by hand a limited set of calculations
(see Richardson 1922). But this human “forecast factory” was never assembled.
2. Though my focus is on meteorology and climate science, much of the analysis is
likely to apply to the use of computer simulation in other ªelds as well. For related discus-
sions focusing on other ªelds, see Bechtel and Abrahamsen 2010 and Ylikoski (this vol-
ume). The present paper also dovetails with the more general discussion of simulation and
understanding given by Lenhard 2009.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
338
Weather and Climate
computer. Section 5 discusses the use of the computer to experiment on
models of the atmosphere and climate system, including models that are
systematically related in a hierarchy. Finally, Section 6 offers some con-
cluding remarks.
2. Understanding Weather and Climate
In the study of weather and climate, as in many other scientiªc ªelds, un-
derstanding is identiªed as a central aim, though there is little explicit
discussion of what the desired understanding consists in. A survey of clas-
sic papers, textbooks and research monographs suggests that there are at
least two notions of understanding in play in this arena.
One sort of understanding is tightly linked to explanation and has as its
target the occurrence of an event or phenomenon, usually deªned in terms
of a small set of salient or essential properties. For instance, scientists
might want to understand why there is an extratropical jet stream—a per-
sistent, narrow region of accelerated air near the top of the troposphere in
the extratropical region, or why severe thunderstorms in the central
United States sometimes split into two separate storms, or why New York
City received a record-setting snowfall last Thursday, rather than a more
typical amount of snow. The desired understanding is achieved by scien-
tists when they arrive at (and perhaps also grasp) an accurate explanation
of the occurrence of the event or phenomenon.3 Usually, the explanations
sought are causal explanations; the stated aim may be to identify the mech-
anism by which a phenomenon of interest is produced (e.g., Charney 1955,
Schneider and Dickinson 1974, Klemp 1987, Markowski 2002), but more
generally the goal is to obtain an accurate causal story—an accurate ac-
count of how a set of causal factors (e.g., forces, processes, conditions) to-
gether produces the event or phenomenon to be explained (Parker 2003,
p. 80; see also Cartwright 1983: Ch.4).4 This sort of understanding will be
referred to as understanding why an event/phenomenon occurs.
Progress toward the goal of understanding why an event/phenomenon
occurs is made when scientists obtain what philosopher Peter Railton
3. Without the grasping requirement, this sort of understanding is similar to de Regt’s
(2009) understanding a phenomenon; if the grasping requirement is included, then it is more
similar to Strevens’ analysis (2008; 2013).
4. Occasionally, something like deductive-nomological explanation is referenced, but it
is commonly seen as an inferior sort of explanation. For instance, imagining a perfect nu-
merical simulation of a hurricane, atmospheric scientist Edward Lorenz remarks: “We
might still be justiªed in asking why the hurricane formed. The answer that the physical
laws required a hurricane to form from the given antecedent conditions might not satisfy
us, since we were aware of that fact even before integrating the equations” (1960, pp. 243–
44).
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
339
(1981) calls explanatory information—information that reduces uncertainty
about the form or content of a sought-after explanation. Since causal ex-
planations are typically desired here, explanatory information can include,
among other things: information about causal dependencies; information
about the relative contributions of different causal factors; and informa-
tion about how parts or pieces of a larger mechanism work.
A second sort of understanding is more open-ended and targets com-
plex dynamical phenomena, such as extratropical cyclones and hurricanes,
as well as the atmosphere and climate system as a whole. These targets of
understanding are perceived as rich objects of study, and it is more
difªcult to give an account of what the desired understanding of them
consists in (e.g., what understanding the climate system consists in). It seems
to involve both knowledge and know-how, that is, both knowing things
about the phenomenon or system, such as facts about its structure and dy-
namics, and being able to synthesize and apply that knowledge to answer
correctly additional questions about the phenomenon or system, especially
questions about the effects of interventions on or changes to the system.5
This second sort of understanding, which is discussed in more detail be-
low, will be referred to as understanding a complex phenomenon/system.6
In practice, the knowledge that is partially constitutive of this second
sort of understanding is often summarized in what scientists who study
weather and climate refer to as conceptual models. A conceptual model of X
is a representation of a set of key elements (parts, features, stages) of X as
well as particular relationships (whether spatial, causal, etc.) among those
elements. Conceptual models in meteorology and climate science typically
come in the form of diagrams with associated narrative text. Figure 1, for
instance, shows the diagrammatic portion of a conceptual model of the
hurricane phenomenon, in this case focusing on a period of time in which
the hurricane undergoes eyewall replacement. Conceptual models of parts
of phenomena—such as the cold front in an extratropical cyclone—are
also common; they zoom in to reveal more detail about particular parts of
the larger phenomenon or system. Almost by deªnition, the content of
conceptual models is primarily qualitative, though it is not uncommon for
5. Exercising this ability may involve reasoning about the system in a qualitative or
rough quantitative way or, when questions are beyond the reach of unaided reasoning,
picking out relevant ingredients for simulation studies, etc. This ability to synthesize and
apply existing knowledge thus bears some similarity to abilities emphasized by Ylikoski
(this volume), de Regt (2009) and Lenhard (2009) in their discussions of understanding.
6. The two sorts of understanding are different in kind. Whereas the ªrst is concerned
with “understanding why” the second is concerned with “understanding a system.” Thanks
to Henk de Regt for pushing me to clarify this.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
340
Weather and Climate
Figure 1. Diagrammatic portion of a conceptual model of a hurricane undergo-
ing eyewall replacement (From Houze et al. 2007, Figure 4; see original caption
and accompanying text for further details. Reprinted with permission from
AAAS.).
them to make judicious reference to equations in their narrative text. As
Figure 1 suggests, often they are constructed with the dual aims of de-
scribing and explaining important features of phenomena.
Understanding a particular complex phenomenon/system,
like the
atmosphere or climate system, generally does not have a clear point of
completion; it is achieved only to a greater or lesser extent. It is increased
when a scientiªc community gains signiªcant new knowledge about the
phenomenon or system, or reªnes existing knowledge, or enhances its
ability to synthesize and apply existing knowledge to correctly answer ad-
ditional questions about the phenomenon or system. In connection with
this, note that that uncovering the mechanisms by which salient features
of a complex phenomenon or system are produced (i.e., obtaining this
knowledge) is one important way of increasing understanding of that phe-
nomenon or system. Hence, the ªrst and second sorts of understanding are
not unrelated. Indeed, the ªrst sort of understanding is typically part of
the second. Understanding why storm splitting occurs, for instance, is
partially constitutive of understanding the supercell as a complex dynami-
cal phenomenon. But understanding supercells is not just a matter of ex-
planation; other sorts of knowledge about supercells—such as descriptive
knowledge of their detailed internal composition and structure at differ-
ent stages of development—also is partially constitutive of understanding
of supercells, as is the ability to synthesize and apply such knowledge to
answer additional questions about supercells (e.g., would changing feature
X of the storm environment enhance storm formation and, if so, why?).
According to philosopher Henk de Regt, even when understanding re-
quires only having an adequate explanation, as in the ªrst sort of under-
standing discussed above, it implicitly depends on a kind of pragmatic
understanding (or know-how) as well. In particular, he argues that it
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
341
depends on understanding a theory, which on his view means being skilled
in using the theory to construct suitable models of target systems or phe-
nomena, which in turn are used to develop explanations and/or arrive at
predictions (de Regt 2009, pp. 592–93). We might question whether
scientiªc explanation always requires a theory per se, as de Regt’s analysis
implies, but resolving this matter is not important for present purposes.
What is of interest, rather, is that understanding a theory in de Regt’s
sense seems to be precisely what atmospheric scientists lacked (at least to a
signiªcant degree) before the advent of the computer: they had a set of
theoretical equations that they believed to apply to the atmosphere, but
they were unable to use this theory for many of the explanatory and pre-
dictive purposes that interested them.7 Consequently, the theory was of
limited value for understanding the atmosphere and its phenomena at that
time.
The computer helped scientists to improve this situation, but how?
The computer did not tell scientists which models to construct—it did
not tell them how to simplify and idealize the equations of hydrodynam-
ics, thermodynamics, radiative transfer, etc. to arrive at mathematical
models that were easier to work with computationally but still realistic
enough to be useful (see e.g., the quasi-geostrophic model developed in
Charney 1947). But the computer did make it possible to estimate solu-
tions to analytically-intractable equations using numerical methods,
something that took ages, and thus was practically infeasible, when calcu-
lations were performed by hand. In doing so, it helped scientists to see
what followed (or failed to follow) from the physical assumptions reºected
in different mathematical models of the atmosphere or climate system.
Moreover, because the equations of these models are meant to describe in
an approximate way how conditions in the atmosphere or climate system
change over time, their repeated solution for small time steps with the
help of the computer could produce simulations, i.e., representations of the
temporal evolution of the atmosphere or climate system.
In the next three sections, I discuss in more detail how using the com-
puter to reveal the implications of modelling assumptions and to produce
simulations has helped to advance understanding in the study of weather
and climate. Given the two sorts of understanding just identiªed, any of
the following would increase, or count as progress toward, understanding:
(a) obtaining explanatory information, such as information about causal
dependencies or causal relationships; (b) obtaining descriptive knowledge
7. See, however, de Regt and Dieks (2005, pp. 153–54) on “PV-thinking” (potential-
vorticity thinking) for an example of how simpliªed theory can be employed in the service
of understanding a limited range of atmospheric phenomena.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
342
Weather and Climate
of the structure or evolution of a phenomenon or system of interest—the
sort of knowledge that is commonly summarized in conceptual models;
(c) enhancing a scientiªc community’s ability to synthesize and apply ex-
isting knowledge to answer additional questions about the phenomenon
or system. I will suggest that the computer can help with all of these, but
I will most often highlight connections with (a).
Before moving on, however, some additional targets of understanding
should be mentioned: the occurrence of events and phenomena in a simula-
tion; simulated complex dynamical phenomena; and model atmospheres and
climate systems. The mathematical models used to produce simulations in
the study of weather and climate include variables that are given physical
interpretations—they stand for temperature, pressure, wind speed, etc.
Many atmospheric scientists view simulation results as if they were obser-
vations of a hypothetical atmosphere or climate system and identify phe-
nomena and events that occur “in the simulation.” Often, they seek to un-
derstand the occurrence of these simulated phenomena and events, just as
they seek to understand the occurrence of real phenomena and events in
Earth’s atmosphere or climate system; the aim is to explain how the phe-
nomenon or event in the simulation—the “jet stream” or “record snow-
fall” in the simulation—would be produced by the causal factors repre-
sented in the simulation. Likewise, atmospheric scientists sometimes seek
to better understand a model atmosphere or climate system, or a simu-
lated hurricane, just as they aim to better understand the real atmosphere
and climate system and real hurricanes. In general, however, this model-
directed understanding is desired as a means to better understanding
weather and climate in the real world.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
3. Computer Simulation Results as Surrogate Observational Data
One way to learn about phenomena is to study them observationally, i.e.,
to collect data regarding their properties using instruments or even the
naked eye. Sometimes, these data provide information that leads to sig-
niªcant progress in developing accurate explanations and conceptual mod-
els. For example, when there are several competing accounts of the mecha-
nism by which a phenomenon is produced, trustworthy observational data
might strongly support one of the proposed mechanisms while indicating
that the others are untenable. In such a situation, the observational data
provide valuable explanatory information.8
Yet obtaining desired observational data in the study of weather and
8. What counts as explanatory information for a given scientist, of course, will depend
on her background knowledge and cognitive capacities; she must be able to see that the in-
formation obtained is relevant to an explanatory goal.
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
343
climate can be quite difªcult. Routine observations tend to be made at rel-
atively widely-spaced locations and only a few times a day, while many
phenomena and events of interest occur on smaller spatiotemporal scales.
In addition, some of these phenomena and events—like supercell thunder-
storms and hurricanes—involve intense and dangerous conditions. Radar
can provide some information about conditions within a supercell from
afar, but it is difªcult to make in-situ observations of temperature, humid-
ity and other conditions inside such a storm; one must be willing to brave
high winds, lightning, heavy rain, hail and perhaps a tornado, and mea-
surements still usually will cover only a relatively limited spatiotemporal
domain.
Computer simulations, by contrast, provide values for every variable in
the model, at every spatial grid point, for every time step in the simula-
tion. Moreover, simulation results can be examined from the comfort of
the computer lab, without risking life and limb, and, in many cases, can
be obtained for substantially less cost than observational data collected in
specialized observing campaigns. So it is not surprising that atmospheric
scientists sometimes analyze simulation output in place of real-world data,
learning which features of a phenomenon of interest arise in what order in
the simulation, how key atmospheric ªelds (e.g., temperature, pressure,
vorticity) are structured at particular time steps, how they change from
one time step to the next, etc. (see Figure 2 for an illustration).9 Here, ad-
vanced visualization tools, which allow scientists to plot and animate re-
sults in ways that make desired information more salient, are especially
valuable (see also Winsberg 1999). Just as traditional observational data
can aid the development of conceptual models and can favor one proposed
explanation over another, so too can simulation results analyzed as surro-
gate observational data.
The case of supercell thunderstorms illustrates this nicely. Relatively
high-resolution simulations of supercells were developed in the early
1980s, drawing on some of the same theoretical foundations used in
weather forecasting (i.e. ºuid dynamics and thermodynamics) but focus-
ing on a smaller spatiotemporal scale. While traditional observations of
the complicated inner workings of supercells were difªcult to make, these
increasingly-sophisticated successors provided
simulations and their
“complete kinematic and thermodynamic data both in and around a [sim-
ulated] storm” (Klemp 1987, p. 372). Upon examining and analyzing
these data with the help of advanced visualization techniques, atmospheric
scientists developed new hypotheses about the mechanisms responsible for
9. This is what I mean by using simulation results as “surrogate” observational data;
they are a surrogate or stand-in for “real” observational data.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
344
Weather and Climate
Figure 2. Vertical cross section of conditions 75 minutes into a simulation of
convection when the convection is isolated (left) and along a frontal boundary
(right). The plot shows rainwater density (shading), wind speed and direction in
the plane (arrows) and normal to the plane (contours), locations of prominent
minima and maxima ((cid:2)/(cid:3)), and the location of the 298°K potential temperature
isotherm (dark grey line). (From Jewitt and Wilhelmson 2006: Figure 12. Re-
printed with kind permission from the American Meteorological Society.)
salient features of supercells, such as their tendency to propagate to the
right of the mean environmental wind (Rotunno and Klemp 1985). In ad-
dition, some existing hypotheses about supercell dynamics were called
into question because they appeared inconsistent with what was observed
to happen in simulations (see e.g., Klemp 1987, p. 395; Houze 1994,
p. 294, on the role of the rear-ºank downdraft in the transition of the
supercell to its tornadic phase), while others were supported by the simu-
lation results.
In general, conceptual models and explanations developed by analyzing
simulation results as surrogate observational data—such as the explana-
tion of storm propagation mentioned above—are treated as how-possibly or
how-plausibly models and explanations, pending further empirical investi-
gation.10 For instance, a review article on supercells, written shortly after
the ªrst wave of high-resolution simulation studies, draws heavily on
these studies but cautions that:
[. . .] although these [simulation] models have demonstrated good
qualitative agreement with observed storms, some of the
10. See Machamer et al. (2000) and Craver (2006) for further discussion of how-possibly,
how-plausibly and how-actually explanations of the mechanistic variety.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
345
mechanisms derived from the detailed analyses of simulated storms
must still be tested against future data that will be obtained from
the increasingly sophisticated storm-observing systems (Klemp
1987, p. 372).
Fifteen years later, a more specialized review article expresses the same
sentiment, but only after attributing to simulation studies “signiªcant ad-
vances in our understanding of supercells” (Markowski 2002, p. 870). By
this time, some of the mechanisms derived from detailed analyses of simu-
lated storms had been accepted and, while others remained more hypo-
thetical, they provided starting points for further investigation—starting
points that might well have been lacking if atmospheric scientists had not
been able to “look inside” simulated supercells, examining their detailed
inner workings. It was not possible to do this for real supercells, given the
limited availability of observational data, and simply inspecting the theo-
retical equations used in storm simulations provides little insight into the
complex dynamical evolution of supercells. Indeed, as the same review
notes, “it is probable that some conclusions drawn from simulation results
never could have been made from observations or theory alone” (Mar-
kowski 2002, p. 870).
Thus, as the case of supercells illustrates, the use of simulation results
as surrogate observational data can promote both sorts of understanding
identiªed in Section 2. It does this in part by facilitating the development
of descriptive and explanatory hypotheses. These hypotheses provide a
starting point for further empirical investigation and may eventually be
accepted as correct, as when how-possibly or how-plausibly explanations
become accepted as how-actually explanations in light of subsequent em-
pirical investigation. (The design of such empirical investigation is itself
often strongly inºuenced by what the simulation results indicate, e.g.
about where in the storm one should look to ªnd evidence of a particular
structure or process.) Without simulation results that stand in for observa-
tional data, even how-plausibly explanations for some phenomena might
remain out of reach for much longer.
4. Testing Hypotheses
A second important way in which the computer helps to advance under-
standing in the study of weather and climate is by facilitating tests of
hypotheses relevant to explanation—tests that in many cases would not
otherwise be feasible. As Jule Charney put it not long after high-speed
digital computers were introduced:
The radical alteration that is now taking place [in dynamical mete-
orology] is due not merely to the ability of the machine to solve
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
346
Weather and Climate
known equations with known initial and boundary conditions but
even more to its ability to serve as an inductive device. [. . .] The
machine, by reducing the mathematical difªculties involved in car-
rying a physical argument to its logical conclusion, makes possible
the making and testing of physical hypotheses in a ªeld where con-
trolled experiment is still visionary and [physical] model experi-
ment difªcult, and so permits a wider range of inductive methods
(Charney 1955, pp. 798–99).
Here, the physical hypotheses of interest concerned the mechanisms re-
sponsible for salient, large-scale features of the atmosphere, such as the
high- and low-pressure systems that regularly populate the middle lati-
tudes. But the value of the computer for testing hypotheses about climate
change was also emphasized early on. Even as computer models of the cli-
mate system were in their early days, atmospheric scientist Edward Lorenz
suggested that “perhaps there should be a center for climatic change hy-
pothesis testing” (Lorenz 1970, p. 328), where the tests would be carried
out with the help of computers.11
But how can the computer facilitate hypothesis testing? And what sorts
of hypotheses can be tested? According to the Charney passage, the com-
puter facilitates testing by “reducing the mathematical difªculties in-
volved in carrying a physical argument to its logical conclusion” (1955,
p. 798). Put differently, the computer allows scientists to see what follows
(or fails to follow) from the physical assumptions reºected in different
mathematical models of the atmosphere or climate system—models for
which analytical solutions are out of reach. This allows scientists to test
hypotheses about the sufªciency of different sets of causal factors (processes,
conditions, forces) for producing a phenomenon or event or feature of the
atmosphere/climate system, and it can also facilitate tests of hypotheses re-
garding necessary causal factors.
To see why, suppose an atmospheric scientist hypothesizes that H:
Causal factors {c1 . . . cn} are jointly sufªcient for producing P, a partic-
ular phenomenon or event. Even if the scientist cannot say exactly how
{c1 . . . cn} would produce P, she might test H by building a mathematical
11. The climate system is usually deªned to include the atmosphere, ocean, land sur-
face and cryosphere. Today’s state-of-the-art global climate models incorporate not only at-
mospheric models similar to those used in weather forecasting, but also representations
of these other component systems. Historically, climate modeling grew out of atmo-
spheric modeling, with the ªrst global atmospheric modeling results obtained by Norman
Phillips in a project aimed to construct a “dynamic climatology” (see Charney 1955; Phil-
lips 1956).
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
347
model that accurately represents the mutual interactions among {c1 . . . cn}
and then checking whether that model entails the occurrence of P. Com-
puters help with the latter step, i.e., checking whether the model entails
the occurrence of P. Of course, typically the computer delivers only ap-
proximate solutions to the modelling equations of interest, so it is impor-
tant to consider whether the occurrence/non-occurrence of P in a simula-
tion is a product of errors introduced by the methods used to estimate
solutions or by programming mistakes. But when there is reason to think
that such errors did not interfere in this way, and that {c1 . . . cn} and other
important system processes have been adequately represented via the
modelling equations, then the simulations produced can provide evidence
regarding H. In particular, if the simulations produce something closely
resembling P, then this is evidence for H; if they do not produce anything
like P, then this is evidence against H.12
In 1970, when Lorenz was writing, hypotheses about climate change
mainly concerned the causes of past ice ages and very fundamental ques-
tions about the climate system, such as whether there might be more than
one semi-stable climate for Earth even when important factors like incom-
ing solar energy and the chemical composition of the atmosphere are held
constant. Lorenz describes how the computer could help test the latter sort
of hypothesis:
Meanwhile, it is of interest to ask what would happen if we took
the mathematical models which are currently being used to simu-
late climate, without any modiªcation to accommodate existing cli-
matic change hypotheses, and performed experiments lasting centu-
ries or more. Would climatic changes be revealed? If we include as
one hypothesis of climatic change the proposition that no processes
other than those commonly considered in short-range weather fore-
casting are needed to bring about changes in climate, we would be
testing this hypothesis. (Lorenz 1970, p. 328)
In other words, we would be testing the hypothesis that H: The physical
processes represented in 1970-era weather forecasting models are suf-
ªcient to produce changes in climate. (Models then used to simulate cli-
mate were very similar to the models used in short-range weather forecast-
ing at the time.)
For a more recent example, consider a hypothesis about the causes of
12. Of course, there is still an empirical dimension to these tests—there is an empirical
phenomenon or event or feature to be accounted for. The computer helps with the step in
testing that involves deriving a prediction or conclusion from the set of equations.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
348
Weather and Climate
late twentieth century global warming, H1: Estimated changes in natural
forcing factors are sufªcient to produce most of the global warming ob-
served to occur during the second half of the twentieth century; changes in
greenhouse gas emissions and other human-related factors need not have
contributed much. To test H1, scientists might run today’s state-of-the-art
climate models, allowing natural forcing factors (i.e., changes in solar out-
put and volcanic aerosols) to vary in accordance with twentieth century es-
timates but holding ªxed all anthropogenic forcing factors. Of interest
would be whether global warming similar in magnitude to that observed
during the latter half of the twentieth century occurred in the simulations.
In fact, such simulations have been produced, and they do not show
warming similar to that observed to occur during the second half of the
twentieth century (see Figure 3b). Insofar as natural forcing factors and
important climate system processes are adequately represented in today’s
climate models, this ªnding is evidence against H1.13
By contrast, simulations produced with state-of-the-art climate models
that include representations of both anthropogenic and natural forcing
factors do show changes in global mean temperature that (roughly) match
those estimated from twentieth century observations (see Figure 3a). Inso-
far as the identiªed forcing factors and important climate system processes
are adequately represented in today’s climate models, this ªnding supports
hypothesis H2: These anthropogenic and natural
forcing factors are
sufªcient to produce (roughly) the changes in global mean temperature
observed to occur over the course of the twentieth century. Moreover, to
the extent that there is also good reason to think that all major natural
forcing factors have been identiªed, the two sets of simulations together
also support H3: Anthropogenic forcing factors are necessary to account for
observed twentieth century global warming.
As the examples illustrate, the conclusion that simulation results pro-
vide evidence of the sufªciency or necessity of a set of causal factors rests
on some signiªcant assumptions, for instance, that the causal factors of in-
terest, as well as important system processes, are adequately represented in
the models used. In the case of hypotheses about necessary factors, there is
also the assumption that all of the plausible candidate factors have been
identiªed. The difªculty in justifying these assumptions will vary from
case to case. Some of the easier cases are those in which hypotheses concern
the sufªciency of a relatively small set of causal factors for producing P,
13. Here, I mean adequately represented for purposes of discerning the major causes of
late twentieth century global warming; a model might be adequate for this purpose but
not, say, for giving precise quantitative predictions of long-term regional and local changes
in climate.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
349
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Figure 3. Global mean surface temperature anomalies over the 20th century
from observations (black) and from simulations (thin light grey) when the simu-
lations include (a) both natural and anthropogenic forcing factors and (b) natural
forcing factors only. Anomalies comprising a given time series are calculated rela-
tive to the global mean surface temperature for that simulation (or from observa-
tions) during the period 1900–1950. The heavier grey line shows the average
anomaly in the simulations. (Adapted from Climate Change 2007: The Physical
Science Basis. Working Group I Contribution to the Fourth Assessment Report
of the Intergovernmental Panel on Climate Change, Figure 9.5 (a). Reprinted
with kind permission from Cambridge University).
and something resembling P does appear in the simulations. For instance,
early investigations found that phenomena resembling extratropical cy-
clones, the large low-pressure systems that bring poor weather in the mid-
dle latitudes, would develop in simulations made with simple atmo-
factors (see
spheric models that represented a reduced set of causal
350
Weather and Climate
Charney 1955, pp. 800–1), supporting the hypothesis that those factors
are sufªcient for cyclogenesis.14 On the other hand, with a system as com-
plex as the climate system, about which there is substantial but still rather
limited knowledge, it is not easy to argue persuasively that all candidate
causes of a climate phenomenon have been identiªed, which is required
when testing whether a particular causal factor is necessary. The rejection
of H1 above is resisted by some individuals precisely on the grounds that
there may be other important natural forcing factors—such as cosmic
rays—that are not represented, or not represented adequately, in today’s
models.15
So how does using computers to test hypotheses about necessary and
sufªcient causal factors advance understanding? Once again, the clearest
links have to do with explanation: testing these hypotheses sometimes
provides explanatory information. Learning that a set of causal factors is
sufªcient for producing phenomenon P indicates that a causal or mecha-
nistic explanation of P in terms of those causal factors should be possible;
one can then being pursuing such an explanation.16 Learning that the set
of factors is not sufªcient for producing P can prevent one from wasting
time trying to ªnd such an explanation. Moreover, learning that particular
causal factors are necessary for producing P can strongly constrain the
space of possible explanations of P (e.g., the range of mechanism descrip-
tions) that should be taken seriously.
5. Experimenting on Models / Exploring Hierarchies
The atmosphere and climate system are made up of numerous nonlinear
and interactive processes, making it difªcult to infer causal relationships
by simply observing these systems in action. By experimenting on a com-
puter simulation model of the atmosphere or climate system in various
ways—“turning off” particular physical processes, varying the values of
parameters, etc.—and comparing the simulations produced with and
without these interventions, scientists investigate the contributions of the
in producing (simulated)
manipulated processes
and parameters
14. Charney concluded that these simulation studies had determined the actual cause of
cyclogenesis (see Charney 1955, p. 801), but this conclusion would seem unwarranted,
since existing rival hypotheses had not been tested.
15. It is beyond the scope of this paper to examine past and ongoing scientiªc debates
about additional natural forcing factors. Such debates do, however, merit further attention
from philosophers of science and scholars in STS. For more on the assumptions made in
contemporary detection and attribution studies, see Solomon et al. 2007, Parker 2010, and
Petersen 2012.
16. There is no guarantee that such an explanation will be correct, of course, but at
least one’s efforts are directed in a potentially fruitful way.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
351
phenomena of interest. For instance, a scientist might include more realis-
tic topography in a climate model in order to learn whether this makes
much difference to the amount of (simulated) annual precipitation that falls
over the United States or to other outcomes of interest. What is learned in
this way about dependencies in the model can provide a starting point for
identifying the mechanisms by which real-world phenomena are produced
and for reasoning about the effects of particular interventions. In this way,
it can promote both sorts of understanding identiªed in Section 2.
In fact, a closely related approach to advancing understanding of
the atmosphere and climate system was explicitly recommended when the
digital computer ªrst came on the scene. The idea was to start with
simpliªed models that represent in an idealized way a reduced set of causal
factors thought to be particularly important in shaping system behavior,
and then gradually increase the models’ complexity by including repre-
sentations of additional causal factors and/or by representing previously-
included factors more realistically (see e.g., Charney 1949, Phillips 1956,
Lorenz 1960; see also Dahan Dalmedico 2001). By observing how the
simulated atmosphere or climate system changed with the addition of
each new causal factor, such as frictional drag or a primitive hydrologic cy-
cle, scientists could develop a storehouse of information about depend-
encies in simpler models, which would serve as a resource for constructing
explanations of the behavior of more complex models and of the real atmo-
sphere and climate system.17
In other words, understanding was to be advanced by constructing and
experimenting on a hierarchy of models of increasingly complexity, not by run-
ning the most comprehensive and “realistic” computer simulation model
possible. The latter was considered unlikely to help much with the goal of
identifying the causal contributions of different factors:
The total behavior of the [atmospheric] circulation is so complex
that the relative importance of various physical features, such as the
Earth’s topography and the presence of water, is no more evident
from an examination of numerical solutions than from direct obser-
vations of the real atmosphere. (Lorenz 1967, p. 134)
On Lorenz’s view, “it is only when we use systematically imperfect equa-
tions or initial conditions that we can begin to gain further understanding
17. This strategy calls to mind uses of “false” models in biology identiªed by Wimsatt
(1987, pp. 30–31): “An oversimpliªed model may act as a starting point in a series of
models of increasing complexity and realism” and “An oversimpliªed model may provide a
simpler model for answering questions about the properties of more complex models that
also appear in the simpler case, and answers derived here can sometimes be extended to
cover the more complex models.”
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
352
Weather and Climate
of the phenomena which we observe” (1960, p. 244). Doing so can show
us how things would be different—in the simulated system and perhaps
in the real system—if particular factors were absent or changed.
So have atmospheric scientists followed the hierarchies-of-models
approach to advancing understanding? That is, have collections of
systematically-related models been carefully developed and intensively
studied, to provide a foundation for explaining the observed behavior of
the atmosphere and climate system and for reasoning about how weather
and climate would be different if conditions were changed in various
ways? Only to a limited extent, it would appear. While signiªcant prog-
ress via a hierarchy-of-models approach was made early on, in recent de-
cades a tremendous increase in computing power and a growing interest
in predicting future climate change has led atmospheric scientists to focus
their efforts on the development of more and more comprehensive and de-
tailed models, without attending carefully to how they relate to other
models and without investing comparable effort in understanding simpler
ones.
This state of affairs has not gone unnoticed. Recently, atmospheric sci-
entist Isaac Held (2005) expressed concern over the growing gap between
what can be simulated with today’s climate models and what is under-
stood about the climate system and its dynamics. He calls for renewed ef-
forts to develop “hierarchies of lasting value”—sets of systematically-
related “elegant” models, the careful study of which can provide a founda-
tion for understanding the real atmosphere and climate system.18 Con-
structing hierarchies of lasting value will not be easy, he suggests. Unlike
molecular biologists, who in their quest to understand human biology
at the molecular level are provided by nature with a ready-made,
evolutionarily-connected hierarchy of model organisms, ranging from bac-
teria to fruit ºy to mouse to man, atmospheric scientists must construct
their hierarchies from scratch; nevertheless, they should try to identify
“the E.coli of climate models” as well as models of intermediate complex-
ity that they take “just as seriously as do the biologists who map out every
single connection in the nervous system of the snail” (Held 2005,
p.p. 1610, 1614). Despite the challenges associated with a hierarchies-of-
models approach, on Held’s view, “there are no alternatives if we want to
understand the climate system and our comprehensive climate models”
(Held 2005, p. 1610).19
18. By “elegant” models, he means ones that include only what is necessary to “to cap-
ture the essence of a particular source of complexity” in the atmosphere or climate system
(Held 2005, p. 1613).
19. While this is a strong claim, it is difªcult to see how else desired understanding of
these complex systems would be achieved, given humans’ cognitive limitations. Of course,
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
353
As noted above, experimenting on a computer model of the atmosphere
or climate system can provide explanatory information and can aid and in-
form reasoning about the effects of particular interventions on the real sys-
tem. Exploration of a hierarchy of models of the atmosphere or climate
system involves extensive experimentation on models, with similar
beneªts. Moreover, in the long run, such exploration/experimentation can
create a familiarity with the behavior of model atmospheres and climate
systems, and with the ways in which different causal factors shape their
behavior, that can enhance scientists’ ability to synthesize and apply exist-
ing knowledge to answer additional questions about their real counter-
parts (i.e., can help to develop this know-how).20, 21 Exploration of a hierar-
chy of systematically-related models is, and is explicitly recognized to be,
a long-term strategy for advancing understanding of the atmosphere and
climate system—a strategy that involves leveraging knowledge about and
experience with simpler models to make progress in understanding more
complex models and systems.
6. Concluding Remarks
In the half century since its introduction, the computer has helped to
transform the study of weather and climate. Its impact has been felt not
just in weather prediction, where dramatic increases in forecast skill have
been achieved, but also in basic research that aims to advance understand-
ing of weather and climate phenomena and of the atmosphere and climate
system as a whole.
Three ways in which the computer has helped to advance understand-
ing in this arena were identiªed above. First, the computer has been used
to produce simulations that supply surrogate observational data, aiding
the development of conceptual models and explanations. Second, it has
been used to facilitate tests of explanatory hypotheses, especially hypothe-
ses about the causal factors that are necessary or sufªcient for producing a
phenomenon of interest. Third, the computer has been used to experiment
particular questions about the atmosphere and climate system might be answered in other
ways—e.g., by experimenting on a single model. But understanding the atmosphere or cli-
mate system (in the sense of understanding a complex phenomenon/system as discussed in Section
2) aims at more comprehensive knowledge and know-how.
20. Here, I refer to the know-how involved in carrying out qualitative reasoning about
the system, or in deciding which processes to represent, and in what manner, in a mathe-
matical model of the system, etc., given the goal of correctly answering some additional
question about the system (see Section 2).
21. As Lenhard (2009, p. 173) suggests, with experimentation on models, one may de-
velop “the ability to recognize [. . .] qualitatively characteristic consequences of modelling
assumptions even though the modelling dynamic remains partly opaque.”
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
354
Weather and Climate
on models, including models related systematically in a hierarchy; this not
only can reveal information that aids the search for explanations but also
can create a familiarity with the behaviour of model atmospheres and cli-
mate systems that enhances one’s ability to synthesize and apply existing
knowledge to answer additional questions about these systems.
The discussion also revealed, however, some limitations or caveats on
these uses of the computer to advance understanding. For instance, con-
ceptual models and explanations developed by analyzing simulation re-
sults as surrogate observational data are treated as how-possibly or how-
plausibly models and explanations, pending further empirical investiga-
tion; they are not, nor should they be, immediately accepted as how-
actually models and explanations. Likewise, testing hypotheses about the
sufªciency or necessity of sets of causal factors requires some signiªcant
assumptions—for instance, that the causal factors of interest, as well as
important system processes, are adequately represented in the models
used—and these assumptions are sometimes difªcult to justify.
Despite these limitations and caveats, the computer is now ªrmly es-
tablished as an important tool for advancing understanding in the study of
weather and climate. This paper has provided a preliminary overview of
some of the ways in which the computer is used to advance understanding
in this arena. These practices involving simulation, both in the study of
weather and climate and in other ªelds, merit additional attention from
philosophers of science and scholars in STS; further analysis and detailed
case studies no doubt will shed more light on how use of the digital
computer is enriching and transforming the quest for understanding in
science.
References
Bechtel, William, and Adele Abrahamsen. 2010. “Dynamic Mechanistic
Explanation: Computational Modeling of Circadian Rhythms as an Ex-
emplar for Cognitive Science.” Studies in History and Philosophy of Science
41: 321–333. doi:10.1016/j.shpsa.2010.07.003.
Bjerknes, Vilhelm. 1904. “Das Problem der Wettervorhersage, betrachtet
vom Standpunkte der Mechanik und der Physik.” Meteorologische Zeit-
schrift 21: 1–7.
Cartwright, Nancy. 1983. How the Laws of Physics Lie. Oxford: Oxford
University Press.
Charney, Jule G. 1947. “Dynamics of Long Waves in a Baroclinic West-
erly Current.” Journal of Meteorology 4: 135–162.
Charney, Jule G. 1949. “On a Physical Basis for Numerical Prediction of
Large-Scale Motions in the Atmosphere.” Journal of Meteorology 6: 371–
385.
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
Perspectives on Science
355
Charney, Jule G. 1951. “Dynamical Forecasting by Numerical Process.”
Pp. 470–482 in Compendium of Meteorology. Edited by Thomas F.
Malone. Boston: American Meteorological Society.
Charney, Jule G. 1955. “Numerical Methods in Dynamical Meteorology.”
Proceedings of the National Academy of Science 41(11): 798–802.
Craver, Carl F. 2006. “When Mechanistic Models Explain.” Synthese 153:
355–376.
Dahan Dalmedico, Amy. 2001. “History and Epistemology of Models:
Meteorology (1946–1963) as a Case Study.” Archive for the History of the
Exact Sciences 55: 395–422.
de Regt, Henk W., and Dennis Dieks. 2005. “A Contextual Approach to
Scientiªc Understanding.” Synthese 144: 137–170.
de Regt, Henk W. 2009. “The Epistemic Value of Understanding.” Philos-
ophy of Science 76: 585–597.
Grimm, Stephen R. 2006. “Is Understanding A Species of Knowledge?”
British Journal for the Philosophy of Science 57: 515–535.
Held, Isaac. 2005. “The Gap between Simulation and Understanding in
Climate Modeling.” Bulletin of the American Meteorological Society 86:
1609–1614.
Houze, Robert. 1994. Cloud Dynamics. San Diego: Academic Press.
Houze, Robert et al. 2007. “Hurricane Intensity and Eyewall Replace-
ment.” Science 315: 1235–1239.
Jewitt, Brian F., and Robert B. Wilhemson. 2006. “The Role of Forcing in
Cell Morphology and Evolution within Midlatitude Squall Lines.”
Monthly Weather Review 134: 3714–3734.
Klemp, Joseph B. 1987. “Dynamics of Tornadic Thunderstorms.” Annual
Reviews of Fluid Mechanics 19: 369–402.
Lenhard, Johannes. 2009. “The Great Deluge: Simulation Modeling and
Scientiªc Understanding.” Pp. 169–186 in Scientiªc Understanding.
Edited by Henk W. de Regt, Sabina Leonelli and Kai Eigner. Pitts-
burgh: University of Pittsburgh Press.
Lorenz, Edward N. 1960. “Maximum Simpliªcation of the Dynamic
Equations.” Tellus 12: 243–254.
Lorenz, Edward N. 1967. The Nature and Theory of the General Circulation of
the Atmosphere. Geneva: World Meteorological Organization.
Lorenz, Edward N. 1970. “Climatic Change as a Mathematical Problem.”
Journal of Applied Meteorology 9: 325–329.
Machamer, Peter K., Lindley Darden and Carl F. Craver. 2000. “Thinking
about Mechanisms.” Philosophy of Science 57: 1–25.
Markowski, Paul M. 2002. “Hook Echoes and Rear-Flank Downdrafts:
A Review.” Monthly Weather Review 130: 852–876.
Parker, Wendy S. 2003. Computer Modeling in Climate Science: Experiment,
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
356
Weather and Climate
Explanation, Pluralism. PhD Dissertation. Pittsburgh: University of
Pittsburgh.
Parker, Wendy S. 2010. Comparative process tracing and climate change
ªngerprints. Philosophy of Science 77: 1083–1095.
Petersen, Arthur C. 2000. Philosophy of Climate Science. Bulletin of the
American Meteorological Association 81: 265–271.
Petersen, Arthur C. 2012. Simulating Nature, 2nd edition. Boca Raton:
CRC Press.
Phillips, Norman A. 1956. “The General Circulation of the Atmosphere:
a Numerical Experiment.” Quarterly Journal of the Royal Meteorological
Society 82: 123–164.
Railton, Peter. 1981. “Probability, Explanation, and Information.”
Synthese 48: 233–256.
Richardson, Lewis F. (1922) 1965. Weather Prediction by Numerical Process.
New York: Dover.
Rotunno, Richard and Joseph B. Klemp. 1985. “On the Rotation and
Propagation of Simulated Supercell Thunderstorms.” Journal of the At-
mospheric Sciences 42: 271–92.
Schneider, Stephen H. and Robert E. Dickinson. 1974. “Climate
Modeling.” Reviews of Geophysics and Space Physics 12: 447–493.
Solomon, Susan et al. (eds.) 2007. Climate Change 2007: The Physical Science
Basis. Contribution of Working Group I to the Fourth Assessment Re-
port of the Intergovernmental Panel on Climate Change. New York:
Cambridge University Press.
Strevens, Michael. 2008. Depth: An Account of Scientiªc Explanation. Cam-
bridge, MA: Harvard University Press.
Strevens, Michael. 2013. No Understanding Without Explanation. Forth-
coming in Studies in History and Philosophy of Science A. http://dx.doi.org/
10.1016/j.shpsa.2012.12.005.
Wimsatt, William C. 1987. “False models as means to truer theories.”
Pp. 23–55 in Neutral Models in Biology. Edited by in Matthew Nitecki
and Antoni Hoffman. Oxford: Oxford University Press.
Winsberg, Eric. 1999. “Sanctioning Models: The Epistemology of Simula-
tion.” Science in Context 12: 275–292.
Ylikoski, Petri, and Jaakko Kuorikoski. 2010. “Dissecting Explanatory
Power.” Philosophical Studies 148: 201–219.
Ylikoski, Petri. 2013. Agent-Based Simulation and Sociological Under-
standing. Perspectives on Science (this volume).
l
D
o
w
n
o
a
d
e
d
f
r
o
m
h
t
t
p
:
/
/
d
i
r
e
c
t
.
m
i
t
.
/
e
d
u
p
o
s
c
/
a
r
t
i
c
e
–
p
d
l
f
/
/
/
/
2
2
3
3
3
6
1
7
8
9
9
1
8
p
o
s
c
_
a
_
0
0
1
3
7
p
d
.
/
f
b
y
g
u
e
s
t
t
o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3
