Acta Geographica Sinica ›› 2010, Vol. 65 ›› Issue (6): 665-675.doi: 10.11821/xb201006004

Previous Articles     Next Articles

Kernel Principal Components Analysis Based Cellular Model for Restructuring and Predicting Urban Evolution

FENG Yong-jiu1, LIU Miao-long2, TONG Xiao-hua2, LIU Yan3, HAN Zhen1   

  1. 1. College of Marine Sciences, Shanghai Ocean University, Shanghai 201306, China;
    2. Department of Surveying and Geo-informatics, Tongji University, Shanghai 200092, China;
    3. National Institute of Education, Nanyang Technological University, Singapore 637616
  • Received:2009-08-07 Revised:2010-03-15 Online:2010-06-25 Published:2010-06-25
  • Supported by:

    National Natural Science Foundation of China, No.40771174; Key Project on Science & technology of Ministry of Education of China, No.209047; Key Program of Science and Technology Commission of Shanghai Municipality, No.08230510700; Special Research Funds for Selection and Cultivation of Outstanding Young Teachers of Shanghai Universities, No.ssc09018


Simulating and restructuring complex non-linear process of urban evolution with cellular automata plays a significant role in urban land use planning and decision-making. By using conventional methods, it is difficult to retrieve reasonable CA transition rules to capture the dynamic process of urban expansion and evolution. Based on kernel principal components analysis approaches (KPCA), non-linear dimension reduction can be executed on spatial variables with multi-collinearity by kernel method projection in the high-dimensional feature space, therefore, a novel CA model based on KPCA with explicit CA parameters is built which can well reflect the nonlinear nature of urbanization. In a geographical modelling framework called as SimUrban developed in a GIS environment, a fast growing area, Jiading District of Shanghai Municipality, is successfully simulated from 1989 to 2006, and the spatial pattern of the urban areas of 2010 is predicted. The simulation results demonstrate that the urban expansion occurred on the fringe areas of urban center and main roads, which reflects the impacts of the first two components extracted from KPCA approaches and highly accords with the actual development. To evaluate the performances of the KPCA-CA model, confusion matrix and area control indexes are used to assess the accuracies of the simulation results. The overall accuracy 80.67% and Kappa coefficient 61.02% illustrate that the simulation results produced by the KPCA-CA model are well matched with the actual urban evolution of Jiading District. Compared with a cellular model based on linear PCA approach, the simulated results generated by the cellular model based on KPCA have higher accuracies.

Key words: urban evolution simulation, cellular automata, kernel principal components analysis, accuracy analysis, Jiading District of Shanghai