Influence of multiple spatiotemporal resolutions on the performance of urban growth simulation models

Tingting Xu; Heng Su; Biao He; Aohua Tian; Jianing Guo

doi:10.1016/j.isci.2023.108540

Influence of multiple spatiotemporal resolutions on the performance of urban growth simulation models

iScience. 2023 Nov 30;27(1):108540. doi: 10.1016/j.isci.2023.108540. eCollection 2024 Jan 19.

Authors

Tingting Xu^{1

2}, Heng Su^{1

2}, Biao He^{2

3}, Aohua Tian¹, Jianing Guo¹

Affiliations

¹ Chongqing University of Posts and Telecommunications, Chongqing, China.
² Guangdong-Hong Kong-Macau Joint Laboratory for Smart Cities, Shenzhen, China.
³ Shenzhen University, Shenzhen, China.

Abstract

The study developed a framework to investigate the impact of multiple spatial and temporal resolutions on urban growth simulation. The research utilized the convolutional long short-term memory (ConvLSTM) model and three regular models and data from 2009 to 2017 to simulate the urban area of Liangjiang New District in 2018 and determine the optimal spatiotemporal resolution for urban expansion models. The results indicated that the ConvLSTM model has the best simulation result and the ideal temporal resolution for Liangjiang district is to include the previous two years of data, with an optimal spatial resolution of 90 m and a spatiotemporal simulation zone within a two-year time step and 100 × 100 spatial information filter. At this combination, the kappa value of the ConvLSTM is 0.87 which is about 5% higher than others. Our findings revealed that the characteristics of input data can have a significant impact on simulation results and should be carefully considered during the simulation process.

Keywords: Artificial intelligence; Geography; Machine learning; Urban planning.