February 2008
The spectrum of environmental policy challenges--from climate change to nuclear waste storage to coastal shoreline erosion--depend on sophisticated forecasting and modeling techniques. How sound and reliable are our environmental models? What are the inherent limits of environmental science when attempting to forecast the future under different policy regimes? Are there ways to improve environmental forecasting for policymaking purposes?
On February 26, 2008, Daniel Botkin, a research professor in the Department of Ecology, Evolution and Marine Biology at the University of California, Santa Barbara, and Orrin Pilkey, the James B. Duke Professor Emeritus of Geology at Duke University, discussed past performance in environmental modeling. J. Scott Armstrong, a forecasting expert and professor of marketing at the Wharton School, and Jim Manzi, CEO of Applied Predictive Technologies, discussed the strengths and weaknesses of climate models. Stephen F. Hayward, AEI's F. K. Weyerhaeuser Fellow, and Kenneth P. Green, a resident scholar at AEI, moderated.
Panel 1--Past Performance: Success and Failures in Environmental Modeling
Orrin Pilkey
Duke University
When reviewing environmental modeling, it is important to make the distinction between qualitative and quantitative models. Qualitative models are very useful, but quantitative models of earth surface processes cannot produce accurate answers. Society is becoming more quantitative. Even if all the parameters behind the physical processes on the surface of the earth were understood, the future outcome of those processes cannot be accurately predicted. For example, the intensity of storms in one decade does not predict the intensity of those that will occur in the next.
The typical way of validating a model is to apply the model to past decades, but the successful reproduction of an event in one validation does not mean it is correct because of its ordering complexity. It is possible to falsify, but not verify. Society is convinced that the future can be accurately modeled and uncertainty can be done away with, so oftentimes negative aspects of a model are excluded. Models cannot be applied beyond their prediction time.
There are five types of models. First, there is a black box model. The use of this model is foolish because it produces inaccurate results catered to the clients' needs. Second, there is the model that uses the Brunn Rule, but it has no connection with nature. Third, there is a model that the U.S. Army Corps of Engineers uses, but it also produces inaccurate answers. Fourth, there are quantitative models that are inaccurate, but are useful as qualitative indicators. And fifth, there are qualitative models, which are the best types. However, there is little dialogue between modelers and nonmodelers, and this produces misunderstanding and inaccuracies.
Daniel Botkin
University of California, Santa Barbara
Environmental forecasting is used to predict the future and learn the implications of these predictions. The use of Ockham's Razor is the best way to begin creating an environmental forecast model. Ockham's Razor starts with the simplest explanation that is consistent with observations and then adds complexities. Unfortunately, the interest in using models is often in the politicization of the model and the attempt to advocate a position and persuade others of it. Modeling often fails for these reasons. However, an important use of models is to understand the implications of one's own assumptions. To create useful and valid models, a modeler should use Ockham's Razor and be prepared for opposition from the public. A useful insight that challenges common knowledge may be received with little interest and resistance from people who do not want to change their way of thinking.
Panel II--Strengths and Weaknesses of Climate Models
J. Scott Armstrong
Wharton School
There are many weaknesses in current climate models. Current models do not provide forecasts; they provide the modeler's own speculations or scenarios. Given the complexity and uncertainty involved, it is unlikely that these models will prove to be a useful guide. Scientific forecasts are derived from evidence-based models. To create a scientific forecast, several principles apply: it is important to avoid complex models and unaided expert judgment and to be conservative when uncertainty is high. In order to have an accurate forecast for global warming, there are several contingencies. First of all, there must be accurate forecasts in long-term temperature change, the effects of that change, and the effects of feasible policy changes. Climate modelers claim that their models are scenarios, not forecasts, but continue to refer to them as forecasts or predictions. Climate "experts" use models to present their own opinions and make adjustments to suit their assumptions. When making forecasts for complex and uncertain future events, experts have no advantage over nonexperts. Processing facts is most important.
After auditing the current climate change models, most of the forecasting principles were contravened, and there is not a single scientific forecast to support global warming. Forecasts by climate experts are of little value. Climate may change in the future, but because of the uncertainties that exist, the most sensible forecast right now is no change. To produce accurate results, modelers need to use scientific approaches to climate forecasts, avoid alternative sources of bias, consider alternative explanations, examine empirical evidence, use valid empirically based methods, provide full disclosure, present findings clearly, and obtain peer review.
Jim Manzi
Applied Predictive Technologies
It is important to segment the challenges faced in predicting global warming into four steps. First, a modeler must make a prediction of growth, whether it be in terms of population or the economy. Second, a modeler must translate the growth scenarios into impacts on the climate, such as temperature change. Third, a modeler must translate those climate changes into economic or other costs. Finally, a modeler must translate that to utility by examining the trade-off of giving up money today in the expectation of more dollars in the future without those actions. When creating prediction models, the focus is placed on step two. When assuming a constant weather pattern--and the concentration of CO2 is doubled in the atmosphere--global temperature will rise one degree centigrade. Climate modelers go through a process of building and then validating models by comparing two older models. It is empirically impossible to isolate the data and insulate the parameter settings from the knowledge of the performance of the real system. There is no empirical demonstration of global climate models.
There has not been a comparison among a new model, a past model, and the actual change in temperature. This is because the duration of the measurement period is long enough that it makes this difficult. Over the history of climate modeling, climate sensitivity estimates are three degrees centigrade. The scientific consensus is that climate sensitivity is between 1.5 and 4.5 degrees centigrade. This forecast, however, has never been validated.
If it is assumed that the global climate models are correct, a four degree increase equals the loss of 3 percent GDP over time. With this assumption, the drastic action proposed is not required. If the amount of climate sensitivity demonstrated in the laboratory is correct, massive mitigation and aggressive reduction of emissions is not needed. It is important to be cautious, however, when using probability distribution. At a practical level, it is difficult to quantify disaster risk, and we lack the ability to predict those impacts. Therefore, research must focus on the nonquantifiable but nontrivial possibility of disaster by assuming that scenario and working backwards. It is imperative to think practically and improve prediction capabilities. Climate modelers are smart and dedicated but unsupervised, so validation studies are essential.
AEI intern Lauren Jones prepared this summary.
