Parameters and Tuning
In terms of models, especially the global climate ones, it’s all about the parameters: the “fuzzy boundary between data, theory, and algorithms, a place deep within the practical, everyday work of general circulation modeling where modelers combine measured quantities with code to calculate the effects of processes too small, too complex, or too poorly understood to be modeled directly” (Edwards 2010:337). Parameters are “quantities that may be more or less arbitrarily assigned values for purposes of the problem at hand” (2010:338) – they are a best guess, largely based on the limits of physical properties (in the case of global climate models, GCM). Most physical properties in the atmosphere require parameters in constructing models; examples include radiative transfer (gases and solids that absorb and radiate heat); cloud cover (at different elevations). Any given GCM contains hundreds of parameterizations. The model itself comes from the interaction of parameters with initial conditions and other constraints (equations of motion).
What makes some models work better than others is tuning: “adjusting the values of coefficients and even, sometimes, reconstructing equations in order to produce a better overall model result” (Edwards 2010:342). Tuning isn’t done randomly, but is usually based on the modelers evaluation of how to mathematically represent physical properties. The end result, in terms of modeling, is that parameterization and tuning are more ‘art’ than ‘science,’ and that its grounding in physical theory and observational data can be very different from model to model.
Edwards uses Simon Shackley’s own model (or typology) of global climate modelers to show the gulf between modelers, climate seers vs. model constructors. In short, seers want to understand specific processes while constructors want to reproduce the climate in the model. Seers tune their models following intuition, while constructors want to find the complexities and more processes. In general, modelers see tuning as a necessary evil, as long as tuned variable does not violate observed behavior – the ends justifies the means.
The issue with models, then is not technically validation or verification then, but evaluation. Comparing models to actual climate observations is technically not “validation.” Models are inherently inductive reasoning, so do not follow normal science’s method of falsification (from Karl Popper) “The fact that a model agrees-even perfectly-with observations does not guarantee that all of the principles it embodies are true” (Edwards 2010:347-348). This is the difference between deductive and inductive reasoning (or the difference between fictional detectives Columbo and Sherlock Holmes). However, just because deductive reasoning follows falsifiability does not mean that this is the only way (or is even a way) to truth.
“Distinguishing evaluation and confirmation from validation or verification helps to clarify the proper role of models in forecasting climatic change: not as absolute truth claims or predictions, but as heuristically valuable simulations or projections” (2010:352).
Comparing models to each other has resulted in a number of ways that models have become not only a way to analyze climate change (or life in general) but an essential part of our infrastructure. These ways include: an increasing standardization of models; the modular building of models; and the shift from systems to infrastructure through a networking of isolated systems. The main point is that models are not just theories. Because of the model-data symbiosis, observations are already embedded in models through parameterization.
Climate in the Policy Realm
According to Edwards, three elements were necessary to make global climate change a policy issue:
“Policymakers care much less about the past than about the future. The overriding questions for them take the form “What will happen if … ?” For most policymakers or policy institutions even to notice any given issue among the thousands that confront them daily generally requires at least three things: a crisis, a constituency that cares about the crisis, and a theory of why the crisis is occurring and how to resolve it” (2010:357).
GCM’s have become policy tools today because of a number of factors outside climatology. As we discussed earlier, in the 1960s, global warming wasn’t a policy issue; similarly, modeling did not have full scientific (nor public) legitimacy (largely because of limits in computing power). A number of key books made climate into a policy issue, including: Rachel Carson’s Silent Spring (1962), The Limits to Growth (1972), and Our Common Future (1987). In brief, Carson made the public aware of environmental degradation; the Club of Rome (business people, scientists, politicians) used models to show policy people the environmental cost of exponential growth of population/resource consumption; and the Brundtland Report (UN-initiated group) made sustainability the way to go.
Complex systems as expressed in models were the basis of The Limits to Growth:
“systems as complex as cities are “counterintuitive,” in the sense that policies developed to fix problems often end up making them worse. In a classic example, citizens and downtown businesses complain about a lack of available parking. The city responds by building parking structures, but the result is gridlock on the streets when the structures empty simultaneously at the end of the workday. In effect, policymakers tend to see and treat symptoms rather than causes. This occurs because “a complex system is not a simple feedback loop where one system state dominates the behavior, [but] a multiplicity of interacting feedback loops . . . controlled by nonlinear relationships” (Edwards 2010:367).
Part of the reasoning in The Limits to Growth was that there was a “worse before better” pattern in correcting complex systems like the earth’s environment. Essentially, this highly influential book showed that our finite planet could not sustain exponential growth (in people, consumption, and pollution), and that models (whether natural scientific or social scientific) were the way to demonstrate the need for social and political change. Given complex systems, whether in economics or climatology, models were ways to collect inconsistent, incomplete data from a variety of sources and transforming them into a picture of what may be. So it is not an accident that models became a factor in shaping everyday life at the same time that global climate change became an issue. There were, of course a number of historical crises that set off environmental conservation as a policy issue: SST, aerosol/ozone hole, nuclear winter; for the US, famine/drought, Peruvian fishery/Soviet wheat collapse. These crises resulted in environmental issues becoming national security issues, geopolitical stability.