Published works

Land-use change models are used to understand the wide-ranging impacts that land-use changes have on a region. Effective modelling of land-use changes must capture multiple, mutually influential drivers. A common framework for modelling land-use changes uses Cellular Automata (CA), which have seen a growth in application driven by the availability of generic modelling platforms, shifting the focus of research about Land-Use Cellular Automata (LUCA) models from development to application, with a particular focus on calibration. Calibration of LUCA models is complex, as land-use change is a path-dependent process with uncertain outcomes captured by a number of model parameters. Of note are LUCA models that use a transition potential, which are traditionally calibrated using a manual approach, a process that is time-consuming and lacks objectivity. Hence, there has been a focus on the development of automatic calibration methods for these types of models. To automate calibration, metrics are used to capture two separate properties of performance: locational agreement, the match of pixels between simulated outputs and the corresponding observed data, and landscape pattern structure, the inferred realism of land-use change processes captured by the difference between the observed and simulated landscape patterns. The primary objective of this research is to develop improved automatic calibration methods for transition potential based LUCA models. There are two common approaches, optimisation-based and process-specific. The major contributions of this body of work are the development of improved versions of each type of approach, and the development of a hybrid method combining the advantages of the two approaches. First, a generic multi-objective optimisation framework for automatic calibration of transition potential LUCA models was developed in Paper 1 (Chapter 2) that allows for the exploration of trade-offs between the model performance objectives. Second, a process-specific semi-automatic calibration method that integrates objective analysis with discursive input to facilitate efficient calibration of neighbourhood rules (the main calibration parameter for this type of model) within a limited computational budget was developed in Paper 2 (Chapter 3). Finally, a generic framework for hybrid automatic calibration, which integrates domain knowledge into a multi-objective optimisation approach, was developed in Paper 3 (Chapter 4). The utility of each method was demonstrated via case study applications, showing promising potential for future applications of LUCA models to support long term planning and policy development.