Are you the P.I. or an administrator for a JSMF grant?

If your are responsible for a managing a current JSMF grant and need information related to submitting reports, click here.

 

Funded Grants

Researcher: Stephen J. Walsh, Ph.D.

Grantee: University of North Carolina at Chapel Hill, Chapel Hill, NC, USA

Researcher: Stephen J. Walsh, Ph.D.

Grant Title: Modeling population-environment interactions in a world heritage site: Comparison of statistical and agent-based modeling approaches to study complex systems

Grant Type: Research Award

Year: 2011

Program Area: Studying Complex Systems

Amount: $410,810

Duration: 4 years

Modeling population-environment interactions in a world heritage site: Comparison of statistical and agent-based modeling approaches to study complex systems

Globalization is not a new phenomenon, but it is posing new challenges to humans and natural ecosystems in the 21st century. From climate change to increasingly mobile human populations to the global economy, the relationship between humans and their environment is being modified in ways that will have long-term impacts on ecological health, biodiversity, and system sustainability as well as the viability of local economies. These challenges are perhaps nowhere more evident than in island ecosystems, as they represent both the greatest vulnerability to globalization and also the greatest scientific opportunity to study the impacts of global changes. The diversity and endemism that characterize island ecosystems, such as the Galapagos Islands of Ecuador, draw attention to science and conservation, but also highlight the threats to the environment caused by tourism and migration as well as global environmental threats from climate change, invasive species, and such global factors as trafficking in shark fins. In the Galapagos, like many other places around the globe, complexity is anchored in the fact that complex systems are often nested within other complex systems, for example in our case global tourism demands, global climate change, migration primarily from the Ecuadorian mainland reflecting attractions to the Galapagos as well as pushes in certain areas of the mainland, invasive species altering the terrestrial landscape, and the fragile ecology and iconic species of the coastal environment. Incorporating multiple complex systems within a theoretical and modeling approach is challenging, and has not received the attention it deserves given that this is the reality in the empirical world.

Within the complexity community, an increasingly popular tool to examine relationships and test scenarios is agent-based models (ABM). An ABM is a type of spatial simulation method that posits equations and other functions describing processes driving social and environmental change and then calibrates the parameters of those relationships to simulate the effects of different forces on outcomes. There are several unique aspects of ABMs that set them apart from more conventional statistical modeling. First, relationships are specified at the micro level. In an ABM, agents, with various attributes, interact with other agents in a dynamic manner via social and spatial networks within a dynamic environment according to a set of algorithms that can be theoretically and/or empirically based. Second, feedback loops and critical thresholds are extensively represented, particularly by allowing potentially nonlinear aggregations of micro behaviors to change larger social units such as communities which, in turn, feedback onto micro behavior. Feedbacks can operate via compositional effects, social networks, spatial relations or through the environment, and can involve learning and adaptation of agents. Third, rather than seeking a statistical or deterministic solution, agent's actions and their effects are solved computationally, allowing the specification of highly nonlinear and complex underlying equations. Stochastic components can be included in the models. Fourth, ABMs allow for analysis of such concepts as synergy, emergence, and tipping points that are more difficult to handle in other approaches.

Nevertheless, ABMs remain controversial. One important critique is that, in practice, when an ABM is parameterized and calibrated, it usually does not account for traditional problems of endogeneity, selectivity, measurement error, and uncertainty. Thus ABM simulations lack credibility across significant portions of the social sciences. Further a tension exists between keeping the models simple so that one can “see” the processes operating and making them sufficiently complex so that results have the possibility of mimicking reality, which in turn runs the risk of turning the model into a “black box.” Relatedly, there are open questions about verifying and validating ABMs, in part because there have not been empirical trials with ABMs and statistical approaches using the same data. Also, comparisons across models are difficult because reporting standards have not emerged and ABMs often use algorithms that are difficult to describe using mathematics. Without exaggerating too much, these controversial aspects of ABM-style modeling have led conventional social scientists to largely ignore results from ABM models, made it difficult to publish results from ABM models in the very best social science journals, and made it difficult to obtain funding to collect the appropriate data to better inform ABM models. Within the ecological communities, the situation is not quite so stark. Rather, there the proliferation of diverse ABMs has made it difficult to compare results, and human behavior tends to be treated in an overly simplified manner.

We will comprehensively attack these problems with a distinguished interdisciplinary team having expertise in statistical modeling, ABMs, and collecting and analyzing complex social and ecological data. We address parameterization, feedback mechanisms, calibration of equations and algorithms, transparency of the model, reporting standards, uncertainty, and validation issues, and by so doing we will bridge disciplinary linguistic, methodological and conceptual divides. If the results convincingly demonstrate the utility of ABMs for studying complex systems, then the impact on understanding broad-scale social and ecological interactions and change will be profound. The research community will be in a better position to go from micro to macro and then back to micro. If our results confirm skepticism about ABMs, then the proposed research will be instructive to the scientific community and to the parts of NIH and NSF that have been seeking to invest in ABMs and in understanding complex systems more generally.

The work proposed here, admittedly at an early stage of development, aims to accomplish two goals. First, using available data supplemented with expert opinion and qualitative data to be collected as part of this project, we will develop an ABM model that has sub-models that incorporate complexity from global tourism, climate change, migration streams, urbanization, invasive terrestrial species, land use/land cover change (LULCC), and the human-environment interactions of a fragile, and increasingly developed, coastal environment. This modeling will not only be a “proof of concept” approach, but will also have results of interest in their own right. Second, we will tackle head-on the skepticism of statistically sophisticated social scientists towards ABM-type models. David Guilkey, a well-known econometrician at the University of North Carolina and a skeptic about the utility of ABMs in causal analysis, will be part of the team designing the ABM. The idea is to highlight all the points at which the ABM we develop is using, data, rules & tools, that would be considered questionable from an econometric perspective, and then, working as a team, design a data collection and analysis approach that could be analyzed both by ABM and econometric experts to see the extent to which the ABM and statistical approaches give similar results, and if not, why not. In short, this is the statistical and modeling equivalent of a “cook off.” Funding for the data collection and analyses for this “cook off” is not part of the present project; rather we would apply to NIH, and possibly NSF for the necessary funds towards the end of this project. The significance of both the proof of concept and being able to bridge the chasm between those using ABM-type models and more conventional statistical models is that serious progress in understanding complex systems will require a better understanding of the strengths and weakness of both approaches, and preferably a functional combination of the two.

During the past three decades, dramatic changes have threatened the social and ecological sub-systems of the Galapagos Islands. Beginning in the 1970’s, the Islands have experienced exponential population growth. Thousands of new residents began to migrate from the mainland attracted by the promise of lucrative opportunities linked to the islands’ rich marine and terrestrial ecosystems and “pushed” by the lack of economic opportunities in many parts of mainland Ecuador. The local population has grown from under 10,000 in 1990 to an estimated 35,000 residents today, and the number of tourists has increased from about 41,000 in 1990 to nearly 185,000 by 2010. In 2007, in response to the direct and indirect effects of the expanding human imprint in the Galapagos, UNESCO placed the islands on the “at risk” list of World Heritage Sites, and, similarly, the Ecuadorian Government declared an “ecological emergency” for the islands.

While it is obvious at a superficial level that jumps in tourism levels and population migration impact the environment and vice versa, less is known about the explicit pathways through which this occurs. The challenge in moving scientific understanding beyond the obvious is graphically illustrated in Figure 1. The three main elements are land (cover and use), resident population, and tourism. Each influences one another. With causality likely going in both directions, untangling effects is extremely difficult, hence the need to consider both ABMs and structural equation modeling.