Abstract:
The urban expansion of Lagos continues unabated and calls for urgent concern. This thesis explored the use of both the conventional and unconventional techniques for modelling land use change. Two conventional methods (ordinary least squares and geographically weighted regression) were based on geographic information systems, while four unconventional methods (logistic regression, artificial neural networks, and two proposed types of support vector machine) were based on cellular automata. These techniques were evaluated using three land use epochs: 1963-1978, 1978-1984, and 1984-2000.
The conventional methods make quite strong statistical assumptions, some of which are shown not to be met by the land use data at hand. Despite this, these methods do exhibit substantial agreement between observed and the predicted maps. The non cellular automata and cellular automata modelling were then implemented with the logistic regression, artificial neural network, support vector machine, and fuzzy support vector machine models, with model parameters set by k-fold cross-validation. The cellular automata predicted maps were more accurate than those of the non cellular automata.
The cellular automata modelling results from the proposed support vector machine and fuzzy support vector machine were compared with those from the geographic information systems based geographically weighted regression, logistic regression, and artificial neural network. The results from the geographic information systems based geographically weighted regression were the best, followed by those from the support vector machine and fuzzy support vector machine, followed by the artificial neural network, and logistic regression. This research demonstrated that the proposed support vector machine and fuzzy support vector machine based cellular automata models are promising tools for land use change modelling.
