Acta Geodaetica et Cartographica Sinica ›› 2023, Vol. 52 ›› Issue (10): 1772-1783.doi: 10.11947/j.AGCS.2023.20220244

• Cartography and Geoinformation • Previous Articles     Next Articles

Considering the spatial multi-scale neighborhood effect and time dependence into cellular automata model for urban growth simulation

WANG Haijun1,2, CHANG Ruihan1, LI Qiyuan1, ZHOU Xiaoyan1, WANG Quan1, ZENG Haoran1, LIU Yining2,3, YUE Zhaoxi2,3   

  1. 1. School of Resource and Environmental Science, Wuhan University, Wuhan 430079, China;
    2. Key Laboratory of Spatial-Temporal Big Data Analysis and Application of Natural Resources in Megacities, MNR, Shanghai 200063, China;
    3. Shanghai Surveying and Mapping Institute, Shanghai 200063, China
  • Received:2022-04-11 Revised:2022-08-23 Published:2023-10-31
  • Supported by:
    Funded by Key Laboratory of Spatial-Temporal Big Data Analysis and Application of Natural Resources in Megacities, MNR (No. KFKT-2022-10);The National Natural Science Foundation of China (No. 42171411)

Abstract: Under the strategic background of promoting new-type urbanization and implementing territorial space planning in the new era, urban growth research has gradually become a hot issue. The current urban growth simulation based on cellular automata (CA) lacks the analysis of multi-scale neighborhood effect of urban space, and the expression of time-dependent influence of long-term urban evolution process in transformation rules is not perfect, which simplifies the spatio-temporal dependence of urban growth. Future planning implementation scenarios cannot be simulated and deduced to serve national spatial planning. To solve the above problems, this paper constructs a CA model of urban growth (hereinafter referred to as Deep-CA), which takes spatial multi-scale neighborhood effect (3DCNN) and time dependence into account (ConvLSTM). Firstly, 3DCNN combining ordinary convolution and empty convolution is used to extract the multi-scale neighborhood effect of urban space. Then, ConvLSTM neural network is used to assimilate historical information, and the time dependence of long time series is considered. Thus, the suitability probability of urban growth is obtained. The land use data and its driving factors in Beijing from 1995 to 2015 are used to verify the scientific nature and applicability of the proposed CA model. The data from 1995 to 2010 are used for model training to simulate the urban scope in 2015. At the same time, the accuracy of simulation results is compared with the three traditional methods: artificial neural network-CA, logistic regression-CA and maximum entropy-CA. Compared with the traditional CA model, the simulated FoM index of Deep-CA in Beijing increases by about 4% in 2015, the simulation effect of Deep-CA on urban global and local morphology is good, and the patch fragmentation is low. The experimental results show that Deep-CA can accurately obtain long-term spatio-temporal dependence, thus further improving the simulation authenticity of urban growth CA model.

Key words: urban growth, spatial multi-scale neighborhood effect, time dependence, deep learning, cellular automata

CLC Number: