Annual Urban Expansion Extraction and Spatio-Temporal Analysis Using Landsat Time Series Data: A Case Study of Tianjin, China

In this study, we propose a new method of extracting urban land expansion using annual Landsat time series data and then analyze spatio-temporal dynamics of urban expansion obtained using the proposed method. The proposed method includes two major stages, i.e., one-class classification and subsequent spatio-temporal refinement. A one-class classification method, one-class support vector machine, was first adopted to classify Landsat image of each year, producing initial urban land classification results. These initial classification results constitute a time sequence of urban land. A spatio-temporal refinement method was then applied to improve the initial urban classification results (and urban land sequence) and to obtain a more accurate urban expansion result. The proposed method was evaluated using Landsat images over Tianjin area, China, from 1990 to 2014. Results showed that the proposed spatio-temporal refinement method significantly improved initial classification results for all 25 years, with increases in overall accuracy from 4.3% to 11.7%. Compared to an existing method, the proposed method achieved both higher classification accuracies for individual years and higher accuracy of urban expansion. The spatio-temporal analysis of the results revealed that Tianjin area experienced an accelerating double-centric urban expansion process during 1990-2014. The main urban expansion type was edge-expansion and urban expansion was spatially concentrated along railways and the coastline.