Go to Paper
Return to GeoComputation 99 Index
The Application of Cellular Automata Modeling for Enhanced Land Cover Classification in the Ecuadorian Amazon
MESSINA, Joseph P. (email@example.com), WALSH, Stephen J. (firstname.lastname@example.org), VALDIVIA, Gabriela, and TAFF, Gregory, University of North Carolina-Chapel Hill, Department of Geography, CB#3220, Chapel Hill, N.C. 27599-3220
Key Words: cellular automata, Ecuador, landscape, remote sensing
The Ecuadorian Amazon region is experiencing rapid development as the result of both resource exploitation and official settlement programs. Frontier settlement in the study site lends itself to a clearer understanding of landuse and landcover change. The landscape of the region is heterogenous and differs structurally in the distribution of species, energy, and materials among the patches, corridors, and matrix present. This particular region is well suited for evaluation as the variables defining settlement expansion and resource development are well controlled and tightly defined.
The discrimination of landcover and landuse has historically been accomplished using single or possibly multidate remotely sensed imagery, systematic ground truthing, or a combination of both. The complexity of the landscape in the Ecuadorian Amazon is such that ground truthing in the historic sense is not easily accomplished, if at all possible, in any systematic manner. With this research, we propose to demonstrate the utility of cellular automata modeling by enhancing the landcover characterization of the region.
Landsat Thematic Mapper data spanning the 10-year period from 1986-96 is used as base data for the development of an initial landcover classification scheme. Traditional classification procedures and validation routines are applied on the collected remotely sensed data. These traditional methods are used as a comparative baseline to evaluate the statistical validity of the cellular automaton enhancement method.
The cellular automata model rules used for this research are more complex than those of a typical cellular automaton, and involve the use of multiple data sources, including topography, road networks, and existing settlement distributions, and their modification over time. The existing settlement patterns are defined by an official government frontier settlement program and exhibit consistency not found in other Amazonian regions. Using traditional urban geography parameters, the model is redefined for the region and allowed to run providing a prior probability component to the classification scheme. Furthermore, the model's control parameters are allowed to self-modify in order to adapt itself to the circumstances it generates, in particular, over the productive lifespan of given settlement and agricultural components. The combination of traditional classification techniques with enhanced cellular automata modeling will significantly improve the local landcover classification accuracy and may prove extendible to widely varying regional conditions.