Using Neural Networks and Landscape Metrics to Map Urban Spatial Growth from Satellite Imagery
ISBN 978-85-88783-11-9
Authors
1Yang, X.; 2Libin, Z.
1FLORIDA STATE UNIVERSITY Email: xyang@fsu.edu
2THOMSON REUTERS LANWORTH Email: lzhou@lanworth.com
Abstract
Over the past several decades, rapid urban growth has been a world-wide phenomenon that can be observed in both developed and developing countries alike. Characterizing the spatial patterns of urban growth can help understand urban morphology and the underlying socio-economic processes. In this paper, we present a method to analyze urban growth patterns based on the combined use of satellite imagery, artificial neural networks, and landscape metrics. The case study site covers the Beijing metropolitan region, one of the world’s fastest growing metropolises during the past four decades. Our method comprises two major components. Firstly, we obtained two dates of satellite imagery acquired by Landsat Enhanced Thematic Mapper Plus (ETM+) and Thematic Mapper (TM) respectively. We produced a land cover map for each of the two dates from the remote sensor images using a pattern classifier called multi-layer-perceptron feed-forward back-propagation neural networks. We further analyzed landscape structural patterns with a selected set of landscape metrics through a moving window approach. Our results have revealed various stages of urban land transformation within the urban core and the surrounding suburbs. This study has demonstrated the utilities of integrating satellite imagery, neural networks and landscape metrics that can provide useful insights into the spatial consequence of urban growth with varying forms of land transformation.