Contextual simulation of landscape based on remotely sensed data

A hierarchical stochastic modeling approach has been developed to provide a general methodology for landscape simulation based on information from multispectral imagery. The model is based on a multiresolution representation whereby a Markov random field is employed to model the region process, a fuzzy approach integrated with a pyramid data structure is used to deal with boundary variation, and natural variability and noise contamination within scene are modeled using class dependent statistical models. The new simulation model is being utilized in conjunction with Landsat TM imagery to provide initial conditions for multitemporal simulations of habitat in western Australia