An Object-based Land-use Cellular Automata Model to Overcome Cell Size and Neighborhood Sensitivity

Cellular automata (CA) are individual-based spatial models increasingly used to simulate the dynamics of natural and human systems and forecast their evolution. Despite their simplicity, they can exhibit extraordinary rich behavior and are remarkably effective at generating realistic simulations of land-use patterns and other spatial structures. However, recent studies have demonstrated that the standard raster-based CA models are sensitive to spatial scale, more specifically to the cell size and neighborhood configuration used for the simulation. To overcome spatial scale dependency, a novel object-based CA model has been developed where space is represented using a vector structure in which the polygons correspond to meaningful geographical entities composing the landscape under study. The proposed object-based CA model allows the geometric transformation of each polygon, expressed as a change of state in part or in totality of its surface, based on the influence of its respective neighbors. In addition, the concept of dynamic neighborhood has been implemented where the neighborhood relationships among objects are defined semantically, that is two objects are neighbors if they are separated by 0, 1 or more objects whose states favor the state transition between them. This flexible neighborhood definition removes any restriction of distance to identify neighborhood relationships among objects, therefore overcoming the neighborhood configuration sensitivity present in the traditional raster CA models. The model was tested to simulate the land-use/land-cover changes in a sub-area of the Elbow river watershed, located in Southwest Alberta, Canada. The results reveal that the object-based CA model generates an adequate evolution of the geographic objects and a spatial configuration of the landscape patches that is more realistic than the one produced by a conventional rasterbased CA model. The model also produces land-use patterns that are very similar to the reference maps.