skip to primary navigationskip to content

Alex Hagen-Zanker: Abstract and background reading

        Titile image Alex Hagen Zanker



We propose a Cellular Automata (CA) urban growth modelling framework and apply it to the English towns Oxford and Swindon. The model calibration is fully automated and a novel application of Markov Chain Monte Carlo Approximate Bayesian Computation (MCMC-ABC). The CA land use change model is formulated with a minimum of required parameters and calibrated from urban genesis onwards. Results suggest that our proposed calibration method effectively captures fundamental characteristics of urban spatial structure and can identify different modes of urban growth. The urban growth modelling framework predicts land use changes well agreeing with empirical observations, overcoming long-standing problems associated with calibration and application of complex and stochastic models over short time periods and with imperfect data.


Cellular Automata (CA) models are widely used to study complex land use change processes and human-nature interactions. The calibration of the CA model remains a difficult task due to the complex relationship between model parameters and emerging spatiotemporal dynamics. One complication is that model calibration is often based on a short period of available data: there may be insufficient information to characterise the dynamic processes and furthermore, the signal-to-noise ratio of real changes versus apparent changes may be problematic. We propose and implement a CA urban growth modelling framework consisting of a model calibration and prediction stage. In particular, the model calibration stage overcomes the problem of short calibration periods by extending the period backwards to the point of original urban genesis. The prediction stage then applies the estimated parameters to simulate urban growth given contemporary land cover data. The model calibration is fully automated and combines innovations in its key components: the estimation method, the land use change model, and the goodness-of-fit measure. The estimation method is a novel application of Markov Chain Monte Carlo Approximate Bayesian Computation (MCMC-ABC), which provides a probabilistic framework to not only estimate parameters for a stochastic model, but also quantify parameter uncertainty. The land use model conceptualizes urban growth as the competition between drivers of urban agglomeration and the preservation of scarce non-urban land resources at multiple spatial scales, with a minimum of required parameters. The goodness-of-fit measure is inspired by the literature on fractal urban form and evaluates correspondence in spatial patterns by comparing the spatial distribution of built density using kernel density estimates of urban land. We demonstrate the feasibility of the proposed framework to automatically calibrate CA parameters and quantify model uncertainty. The proposed model calibration and urban growth modelling framework push boundaries in CCA calibration and modelling to calibrate from urban genesis onwards, automatic CA calibration, CCA model structure uncertainty quantification, designing a minimum CCA, and proposing a GOF measure of spatial structure. When applied over an independent validation period, it is shown that the calibrated parameters for two English towns Oxford and Swindon effectively captured the different modes of urban growth in both towns. For Oxford it correctly reproduces the scattered and constrained growth pattern and for Swindon it produces more compact concentric growth patterns in agreement with real urban land use changes. The identification of growth modes has both a theoretical significance - we show that existing land use patterns can be an important indicator of future trajectories, and a practical significance - we can provide spatial planners with insight in alternative future trajectories yielding more insight in the decision space and the cumulative effect of parcel-by-parcel planning decisions.