AGCI Session II: Characterizing and Communicating Scientific Uncertainty

Session Chairs: Dr. Richard H. Moss and Dr. Stephen H. Schneider

July 31 to August 8, 1996

---

Models in the Policy Arena

Paul Edwards

Stanford University

Stanford, California

---

Climate change would not exist as a political issue without models, Edwards contends, launching a discussion of computer models, uncertainty and the politics of climate change. Uncertainty in these models has important political uses, both positive and negative, by different parties. Models don't and probably won't ever control policy choices because there are other policy constraints that are too powerful, Edwards says.

There are a variety of uncertainties and epistemological problems with climate models including the limitations of computational methods, scaling issues, parameterizations and tuning, flux adjustments, uncertainty cascades (where uncertainties from one model feed into other models), limitations of data, blurring of model results and data (some "data" are in fact products of other models), and validation and verification problems. It is unlikely that these uncertainties will be eliminated.

Climate models are purely mathematical constructs that don't work in ways that are analogous to models in other fields. It is routinely reported in climate research that "experiments with the models" conclude ... But this is a new scientific paradigm one that excepts a mathematical model as a representation of global climate. Another problem concerns so-called "global" data. Very few data sets are actually global; most are regional or very scattered in their global coverage. Models have forced researchers to try to provide data on a global uniform grid and this has resulted in spotty data being filtered through models to produce "global" sets using techniques including smoothing, interpolating and gridding. But global data are required for validation and calibration of models. Thus, data used to validate one class of models are themselves the product of other models. Further, all data go through filters set up by humans. One such filter resulted in the famous "missed ozone hole," when the computer was programmed to reject data outside a certain range of values.

Climate change would not exist as a political issue without models, Edwards contends, launching a discussion of computer models, uncertainty and the politics of climate change.

While models don't control policy, they do play an important role in politics and policy making. One role of models for climate science has been to build an increasingly large community around the climate change issue in which many groups and elements have come to play a role. Successful political processes succeed by enrolling allies. Scientists can't simply write up their results and publish them in journals. They also need to draw connections with other scientific disciplines, the fields of energy and economics, and with government agencies and politicians. Computer models, and particularly their graphics, have been a powerful tool for this kind of enrollment. The primary goal of models in politics is to play a heuristic role as opposed to a strong predictive role.

The Club of Rome Example

The Club of Rome's well-known modeling project that was published as Limits to Growth, grew from Jay Forrester's ideas about complex systems: that they are counterintuitive, nonlinear, and impossible for unaided minds to grasp. He also believed that policies often worsen problems because "complex systems resist most policy changes." Models of such systems are insensitive to changes in most parameters. For all of these reasons, a model is needed to reveal leverage points which are likely not to be where you think they are.

Forrester saw models as the policy solution, and felt that models could serve policy purposes even without good data. He said that "the barrier to progress in social systems is not lack of data. We have vastly more information than we use in an orderly and organized way. The barrier is deficiency in the existing theories of structure." He believed models should be comprehensive. And he saw growth as a developmental phase, not a constant. He believed that continued exponential growth was impossible. Pointing to the importance of metaphor in the policy arena, Edwards recalled the exponential growth curve that became the icon for the Limits to Growth work.

The primary goal of models in politics is to play a heuristic role as opposed to a strong predictive role.

In 1970 and 1971, the first modeling studies of truly global environmental problems were created: the Study of Critical Environmental Problems (SCEP) and the Study of Man's Impact on Climate (SMIC). These models raised for the first time anthropogenic global climate change as a major policy issue. Both studies recommended new methods for gathering global information in standard ways, integration of existing monitoring programs, and a global network of monitors. Following these efforts, climate simulation and global scale observation drove each other. Before computers, there was too much data. After computer models, there was not enough.

In 1970, Forrester attended the first general meeting of the Club of Rome, having been invited via member Carroll Wilson, organizer of SCEP and SMIC. The Club of Rome was considering a computer model of world "problematique," which included issues with global dimensions such as population, resources and environment. Further developing Forrester's approach, they created a model called World 1. The world was divided into five major subsystems: natural resources, population, pollution, capital, and agriculture. A rapid work-up, with system structure and dynamics of greatest importance, was performed. In complexity, it was equivalent to a global average energy balance model. It exhibited the characteristic typical of all Forrester's models: overshoot and collapse. Edwards says that in fact, it is very difficult to produce a policy that does not exhibit this pattern.

The System Dynamics Group at Massachusetts Institute of Technology developed the model further, producing World 2 and World 3, the latter with over 120 interdependent variables. These models were calibrated to historical trends. Edwards says that the models were not particularly good but that they did provoke efforts to gather data and develop new models that were better. They concluded that exponential growth rates were unsustainable and that catastrophic collapse would come around 2050. In 1972, their results were published in Limits to Growth which sold 7 million copies in 30 languages. It was heavily criticized, especially by economists, but earned international respect for the Club of Rome in other circles. It had few direct policy impacts, despite the Club of Rome's promotional campaign to governments. It did, however, have a large impact on world public opinion, perhaps as much as Paul Ehrlich's Population Bomb, Edwards believes.

Limits to Growth was purely heuristic; rather than trying to make specific predictions, it tried to make three general points: the world is a system, exponential growth can't continue, and a comprehensive approach is necessary. It also helped establish a models-for-policy tradition of which climate scientists are the inheritors. This tradition includes a hybrid science/policy community and an interdisciplinary /transdisciplinary approach.

Limits to Growth tried to make three general points: the world is a system, exponential growth can't continue, and a comprehensive approach is necessary.

All of these early models confirmed what the creators already believed that the world faced huge problems that had to be addressed or we would face catastrophe. A huge modeling community was thus created and blossomed in the 1970s, including the International Institute for Applied Systems Analysis. After the 1972-73 energy crisis, a number of energy models and global economic models were developed. The 1980s saw a renaissance of models-for-policy with models designed to simulate the enhanced greenhouse effect. It also saw the first integrated assessment model, IMAGE, which is a direct descendant of the world dynamics models.

A great advantage of the current regional uncertainty of climate models is that self-interest can't creep in.

Most policies that make it through the policy process have certain characteristics, Edwards says: they have a narrow focus, a high probability of success, short-term payoffs, they are tangible, easily perceived, have widely derived benefits, perceived affordability, and feedbacks. The climate change issue has none of these characteristics. The case for policy becomes even weaker when it is taken to the regional and local levels where we cannot even tell if policies will have any observable impact on global climate. If models progress to identifying regional winners and losers, that could make policy responses even less likely. A great advantage of the current regional uncertainty of climate models is that self-interest can't creep in; NIMBY is a robust phenomenon, Edwards says.

Edwards concludes that models do have a role to play in policymaking. They can be used for retrospective policy evaluation, helping to determine if a policy worked by comparing what actually happened to model results of what would have happened in the absence of the policy. Uncertainty can play a role in raising money to do further research involving models. He stresses that models-for-policy should be used heuristically, not predictively, and that modelers should take this into account when addressing the policy community.