I will start by rejecting them, and bring in the question of the structure of the models as my argument. They smell geopolitical rather than scientific"
http://www.geosci-model-dev.net/8/1221/2015/gmd-8-1221-2015.pdf"5 Conclusions
These software architecture diagrams show, in a broad sense,
how climate models work: how the climate system is divided
into components and how these components communicate
with each other. They also illustrate the similarities and differences
between the eight models we have analyzed. Some
models, particularly in North America, exhibit a high level of
encapsulation for each component, with all communication
managed by the coupler. Other models, particularly in Europe,
implement a binary atmosphere–ocean architecture that
simplifies the coupling process. Institutions focus their efforts
on different climatic processes, which eventually cause
different components and subcomponents to dominate each
model’s source code. However, not all models are completely
independent of each other: modeling groups commonly exchange
pieces of code, from individual routines up to entire
components. Finally, climate models vary widely in complexity,
with the total line count varying by a factor of 20
between the largest GCM and the smallest EMIC we analyze
(Fig. 9). Even when restricting this comparison to the
six GCMs, there is still a factor of 7 variation in total line
count.
Our analysis also offers new insights into the question
of model diversity, which is important when creating multimodel
ensembles. Masson and Knutti (2011) and Knutti et al.
(2013) showed that models from the same lab tend to have
similar climatology, even over multiple model generations.
We believe this can be explained, at least in part, in terms of
their architectural structure and the distribution of complexity
within the model. As Knutti et al. (2013) suggest, “We
propose that one reason some models are so similar is because
they share common code. Another explanation for the
similarity of successive models in one institution may be that
different centers care about different aspects of the climate
and use different data sets and metrics to judge model ‘quality’
during development.” Our analysis offers preliminary evidence
to support both of these hypotheses. We hypothesize
further that the relative size of each component within an
Earth system model indicates the relative size of the pool of
expertise available to that lab in each Earth system domain
(once adjustments are made for components imported from
other labs). The availability of different areas of expertise at
each lab may provide a sufficient explanation for the clustering
effects reported by Masson and Knutti (2011) and Knutti
et al. (2013). Furthermore, the two analyses are complementary:
while our analysis looks at model code without considering
its outputs, Masson and Knutti (2011) and Knutti et al.
(2013) analyze model outputs without looking at the code.
Our diagrams may prove to be useful for public communication
and outreach by their host institutions. The inner
workings of climate models are rarely discussed in the media,
even by science reporters; as such, these pieces of software
are fundamentally mysterious to most members of the public.
Additionally, the diagrams could be used for communication
between scientists, both within and across institutions. It can
be extremely useful for climate scientists, whether they are
users or developers of coupled models, to understand how
other modeling groups have addressed the same scientific
problems. A better understanding of the Earth system models
used by other institutions may open doors for international
collaboration in the years to come
