| Cellular Automata (CA) are one of several types of models used to simulate urban growth, as proposed by Tobler. Despite their evolution and wide-spread usage in academia, there are still problems with CA that need resolution. Many of these problems deal with the behavior of CA, how it (behavior) can be quantified, and why some models exhibit chaotic behavior under certain transition rules.; Using the SLEUTH Urban Growth Model, as a test case for developing a method of creating alternate futures from model parameters, the behavior of the model across the parameter, metric, and behavioral space was examined. Using self-organizing maps, the metric space of the model was evaluated across the entire parameter space using multiple datasets, determining the optimal metrics for evaluating goodness of fit. Next, using the new measures of goodness of fit, the behavior of the model was examined across the entire parameter space using a modified version of the cellular disorder measure. It was expected that the behavioral space would have some areas of chaotic behavior while maintaining a high goodness of fit; this was not the case. With the gained knowledge of the parameter, metric, and behavioral spaces, clustering techniques were implemented to dissect the parameter space into distinct regions for forecasting. During this investigation, areas within the parameter space were found where the model overfit the data, and produced no growth during forecasting despite rapid growth during calibration. After eliminating these areas, the parameter space was reclustered into five areas that equally fit the data, and forecasting using these areas created five different, but equally probably futures.; This dissertation demonstrated that through exhaustive exploration of model parameter, metric, and behavioral spaces, that multiple equally probably model forecasts can be found for this and other CA. It also illustrates how overfit is a common problem in modeling, and that there are major research areas that need attention before CA can be used in real-world planning and decision-making. Only once there is complete understanding of the mechanical nature of a model can it be used properly and reach its promise of "honesty in modeling."... |
