Day two at the NSF/ESRC Agenda Setting Workshop on Agent Based Modelling of Complex Spatial Systems in Santa Barbara.

Currently presenting is David O’Sullivan, ex of our lab at CASA.

Agent-based complex systems are dynamic networks of many interacting agents; examples include ecosystems, financial markets, and cities. The search for general principles underlying the internal organization of such systems often uses bottom-up simulation models such as cellular automata and agent-based models –Grim et al 2005.

Complex systems are irreducible – a compact model of a complex system is possible. In general, specific to complex systems is the interest in the spatial locations of the outcomes, such as patterns, processes and the relationship between them.

Verification of any model of an open system is impossible and validation is not the same thing. In a similar fashion, calibration and confirmation are also not the same thing (as verification, or as one another), the argument therefore is for multiple models.

We therefore know that the end product is difficult to analysis and therefore do the models merely outcome what we expect? Keith Bevan has a proposal for a coherent philosophy for modelling the environment, while Steve Bankes advocates working with multiple models.

Agent Based Models models play multiple roles throughout the scientific process, as such it is hard to pinpoint the nature of the exploration. It is also of note that real world choices are made depending on the outcomes of models.

The talk ends with a series of questions – What type of science is Agent Based Modelling? Are tools making the problem work or can they be part of the solution? Does GIS have a role to play?

I would add that it needs to be asked why is the tool set is so complicated? The role of GIS and agents was demonstrated yesterday by ESRI and the level of complexity is concerning, as is the quality of the visualisation.

Interestingly, people do not seem to be aware of the type of model systems running in current game engines and therefore wth full 3D visualisation….

******

Next up is Dawn Parker of George Mason University looking at what researchers seek from Agent Based Models. In general is it a better understanding of the process behind real world systems. Yet models are complex, should they be simplified?

Take a look at Dawn’s SLUDGE model (Simulated Land Use Dependent on Edge Effect Externalities). SLUDGE is a simle combined cellular automaton and agent-based model designed to study the joint influcent of distance-dependent spatial externalities and transportation costs on patterns of land use.

******

Coffee Break 🙂 – For a full rundown on GIS and Agents see the CASA GIS and Agent Based Modelling Blog.

Sitting in the Upham Hotel, Santa Barbara at a two day workshop on Agent-Based Modelling.

An interesting day of presentations ahead including a look into representations of complex modeling and a demonstration from David Maguire of ESRI.

Depending on the relevance we will update this post with details and thoughts..

First up is Mike Worboys, University of Maine looking into the technology which is driving effective agent based modelling.

Technology is weaving itself into everyday life until it becomes part of the general human experience, with each new innovation comes new challenges for modelling systems. Temporal progressions infer changes but we often miss occurrent entities and their relationships when developing models….

Distributed models are inherently complex, even if described by simple rules. Rule systems are not enough we need more ‘mathematical’ principles upon which such systems can be expressed. Credo: Process algebras and their extensions that handle mobility provide a possible next step from cellular rules.

Following on is Marina Alberti from the University of Washington on Modelling Complex Urban Landscape Dynamics: A Pattern-Orientaed Approach. Take a look at the research site for papers and details.

Agent based models are being used at Washington to model emerging phenomena resulting from the interactions of individual entities. Their agents exhibit complex behaviour and learning algorithms based on a hierarchical model.

Dynamic Probabilistic Relations Models (DPRM) are a particular type of nth order Markov model allowing for representation of different types of states and complex relations -each relation is described by its own set of variables. Washington’s Bio-complexity Conceptual Model is structured with demographic and a micro simulation of human behaviour, represented by UrbanSim. The interaction between land user and land cover is driven by a number of spatial interactions.

In general the model provides a probability of the transition of rural/semi-rural to urban areas based on past observations and the generated output.

The discussion is subsequently opened up to the floor moving onto both Activity Theory and emerging phenomena in terms of Complex Systems.