Time-Series Analysis of Medium-Resolution, Multisensor Satellite Data for Identifying Landscape Change

Abstract

The overall goal of this study is to use medium-resolution satellite imagery to determine recent changes in the landscape of the coastal zone near Sanya in the Province of Hainan, China. A search for suitable satellite imagery revealed that the only way to identify the changes was to use data from three different sensors acquired over a 12-year time period: a 1987 Landsat 5 Thematic Mapper (TM) image, a 1999 Landsat 7 Enhanced Thematic Mapper Plus (ETM� ) image, and two SPOT 2 High Resolution Visible (HRV) images acquired in 1991 and 1997. Given that the Landsat and SPOT images have different spatial resolutions and that the spectral bands cover somewhat different spectral ranges, the challenge was how to combine the images in digital format to be able to detect subtle changes in the landscape. Measures of brightness, greenness, and the normalized difference vegetation index (NDVI) were explored using standardized principal components analysis (PCA). Approximately 38 percent of the scene was occupied by water, so tests were performed with the water included and also with the water masked out to remove these low-variance pixels. Factor loadings and input-band contributions were used to interpret component images. Results show that PCA of the visible bands, representing brightness, is the superior approach for identifying new urban features in the landscape. For identification of changes to vegetation, the near-infrared (NIR) bands outperformed NDVI. Selected standardized PCA images with visible and NIR bands are recommended for identifying general changes to an urban landscape using a time-series of imagery acquired by different satellite sensors. Benefits of using a mask are believed to be dependent upon study-site characteristics.