Pattern-oriented calibration and validation of urban growth models: Case studies of Dublin, Milan and Warsaw

Abstract

Urban growth models are established to simulate complex dynamic processes of urban development, such as urban sprawl. According to the pattern-oriented modelling (POM) paradigm, recently gaining weight in ecology as a strategy for modelling complex systems, patterns at multiple scales should be considered to reflect the underlying processes of a complex system. Yet, calibration and validation of urban growth models is typically performed with a goal function of locational (cell-by-cell) agreement only, thus not in line with POM. We therefore examined POM as an approach to calibrate and validate (constrained) cellular automata for the European cities Warsaw, Milan, and Dublin. For Milan and Warsaw, the model structures identified with POM outperformed reference solutions calibrated on a single pattern with improvements up to 25% and 30%, respectively. For Dublin, no good model structure was found, but POM did help to recognize this problem, while locational agreement only failed to do so. Furthermore, the model structures identified with POM were more diverse, i.e. including more driving factors. In these diverse structures, the importance of the neighborhood effect relative to the infrastructure and land use effects reflected the polycentricity of the city as well as its type of sprawl: from monocentric edge expansion in Dublin to in-between ribbon sprawl in Warsaw to polycentric infill development in Milan. We conclude that POM improves the robustness of urban growth model calibration and validation, and obtains more dependable information about the processes driving urban sprawl that may serve the design of instruments to limit it.