On October 12–13, a workshop funded by the National Science Foundation was held at Iowa State University in Ames, Iowa with a goal of identifying research needs related to coupled economic and biophysical models within the FEW system. Approximately 80 people attended the workshop with about half representing the social sciences (primarily economics) and the rest from the physical and natural sciences. The focus and attendees were chosen so that findings would be particularly relevant to SBE research needs while taking into account the critical connectivity needed between social sciences and other disciplines.

We have identified several major gaps in existing scientific knowledge that present substantial impediments to understanding the FEW system. We especially recommend research in these areas as a priority for future funding:

1. Economic models of decision-making in coupled systems

Deliberate human activity has been the dominant factor driving environmental and land-use changes for hundreds of years. While economists have made great strides in modeling and understanding these choices, the coupled systems modeling literature, with some important exceptions, has not reflected these contributions. Several paths forward seem fruitful. First, baseline economic models that assume rationality can be used much more widely than they are currently. Moreover, the current generation of IAMs that include rational agents have emphasized partial equilibrium studies appropriate for smaller systems. To allow this approach to be used to study larger systems, the potential for (and consequences of) general equilibrium effects should be studied as well.

Second, it is important to address shortcomings in these models of economic decision-making. Valuable improvements could be gained from developing coupled models that draw insights from behavioral economics. Many decision-makers deviate systematically from actions that would be predicted by strict rationality, but very few IAMs incorporate this behavior, potentially leading to inaccurate predictions about the effects of policies and regulations. Improved models of human adaptation and induced technological change can also be incorporated into coupled models. Particularly for medium to long-run models, decisions about adaptation and technological change will have substantial effects on the conclusions and policy implications, but more compelling methods for incorporating these changes into modeling are sorely needed. In addition, some economic decisions are intrinsically dynamic yet few coupled models explicitly incorporate dynamic models. Economic models that address uncertainty in decision making are also underutilized in coupled models of the FEW system.

2. Coupling models across disciplines

Despite much recent progress, established models for one component of the FEW system often cannot currently produce outcomes that can be used as inputs for models of other components. This misalignment makes integrated modeling difficult and is especially apparent in linking models of natural phenomena with models of economic decision-making. Economic agents typically act to maximize a form of utility or welfare that is not directly linked to physical processes, and they typically require probabilistic forecasts as an input to their decision-making that many models in the natural sciences cannot directly produce.

We believe that an especially promising approach is the development of “bridge” models that convert outputs from one model into inputs for another. Such models can be viewed as application-specific, reduced-form distillations of a richer and more realistic underlying model. Ideally, these bridge models would be developed in collaborative research projects involving economists, statisticians, and disciplinary specialists, and would contribute to improved understanding in the scientific discipline as well.

3. Model validation and comparison

There is little clarity on how models should be evaluated and compared to each other, both within individual disciplines and as components of larger IAMs. This challenge makes larger integrated modeling exercises extremely difficult. Some potential ways to advance are by developing statistical criteria that measure model performance along the dimensions suitable for inclusion in an IAM as well as infrastructure and procedures to facilitate model comparisons. Focusing on the models’ out-of-sample distributional forecasting performance, as well as that of the IAM overall, is especially promising and of particular importance.

Moreover, applications of IAMs tend to estimate the effect of hypothetical future policy actions, but there have been very few studies that have used these models to estimate the effect of past policy actions. These exercises should be encouraged. They offer a well-understood test bed for the IAMs, and also contribute to fundamental scientific knowledge through better understanding of the episode in question. The retrospective nature of this form of analysis also presents the opportunity to combine reduced-form estimation strategies with the IAMs as an additional method of validation.