An Artificial-Neural-Network-Based Constrained CA Model for Simulating Urban Growth and Its Application

In this research, a constrained cellular automata (CA) model based on Artificial Neural Network (ANN) is developed to simulate and forecast urban growth. As we know, many factors impact urban growth, and relationships among them are complex and non-linear. In geographic CA research, models with different rules and symbology have been developed to simulate those relationships. Yet, in previous done research, it is extremely time-consuming to find proper values of the parameters for CA models through general calibration procedures. As a solution in this research, a neural network can learn from the available urban land-use geospatial data, and deal with redundancy, inaccuracies, and noise. Knowledge and experiences can be easily learnt and stored for further simulation. In this ANN-Urban-CA model, a two-layer Back-Propagation (BP) neural network is integrated into a CA model to seek suitable parameters or weights that match the historical data. The parameters or weights required by CA simulation are automatically determined by the training/learning procedure of the neural network instead of by users, which is a very subjective process. Then each pixel’s probability of urban transformation is determined by the neural network during simulation. Furthermore, a macro-scale socio-economic model is integrated in the CA model to generate the proper demand for urban space in each period in the future. Population is considered as the main factor impacting the demand of urban space. Using population forecasts as exogenous demands, the total number of new urban cells generated by the CA model is constrained. Beijing City is taken as a case study of this ANN-Urban-CA model. Urban growth in the period of 1980-2000 is simulated, and long-term (2001-2015) growth is forecast based on multiple socio-economic scenarios. In conclusion, this ANN-Urban-CA model can simulate and forecast the complex and non-linear spatial-temporal process of urban growth in a reasonably short computational time, with less subjective uncertainty.