13 November 2012

Post date: Nov 13, 2012 5:31:50 PM

Discussion of how to represent space in agent-based models.

The big distinction is between discrete- and continuous-space representations., i.e., grid-worlds or not. Pros and cons of each approach were discussed. Grid-worlds temptingly economical. Continuous space expensive to compute, but the only option if you want to represent (e.g.) a detailed model of perception. GIS use doesn't commit you either way, but often the continuous space of the native GIS representation is dumped out to a grid or array representation. Also talked about the fact that you could use both techniques on top of each other, e.g., an underlying grid world representing resource density (a la Sugarscape) but a continuous-space representation of exactly where the agents are located on that grid.

See http://simulacra.blogs.casa.ucl.ac.uk/2011/11/agent-based-models-of-geographical-systems/ for a useful summary. Also relevant: http://ncgia.ucsb.edu/projects/abmcss/docs/li_paper.pdf (thanks, Joe).