Analysis of green space in Chongqing and Nanjing, cities of China with ASTER images using object‐oriented image classification and landscape metric analysis

Abstract

Green space is an important urban land use which can enhance the livability of cities. Chinese cities develop rapidly, and increasingly strong emphasis has been put on the provision of better landscape and more green space. We used an object‐oriented approach to classify different land covers in Chongqing and Nanjing, two historical Chinese cities. Suitable segmentation levels were selected by locating break points along the variation of selected object variables. Three segmentation levels were identified for each city. Object variables with good discriminatory power were selected to identify different land covers by making use of their spectral, textural and shape properties. Decision tree classifiers were formulated for classifying images into eight land cover classes. Accuracy of object‐oriented classification was the highest in Chongqing and ranked second in Nanjing. The result was compared to those of maximum likelihood classification, fuzzy classification and linear unmixing classification. Land covers were then generalized as green space for landscape metric analysis. The fragmented nature of green space was discussed. It was revealed that there existed a general lack of green space in old urban centres. With an increasing distance from city centres, more large patches were found.

Acknowledgement

This study is supported by a Research Grant Council CERG Grant RGC4251/03H of Hong Kong.