Improving Urban Impervious Surface Mapping by Linear Spectral Mixture Analysis and Using Spectral Indices

As a key indicator of urban built-up areas, impervious surfaces have been frequently analyzed in the studies of urbanization and environmental impacts. For various science and policy applications, it is necessary to accurately estimate and map urban impervious surface areas. Although linear spectral mixture analysis (LSMA) can provide spatial distribution and quantitative fraction information on urban impervious surfaces, using conventional LSMA alone to accurately extract impervious surfaces remains a great challenge. Misclassifications often occur when urban impervious surfaces are estimated by addition of low-albedo and high-albedo fraction images. The low-albedo fraction may relate to features other than impervious surface, such as water bodies and shaded areas, whereas the high-albedo fraction may be spectrally confused with dry soil and other land features. These issues can be acute in mapping of fractional impervious surfaces when the intra-variability of urban impervious surfaces is also considered. To solve these issues, an improved LSMA method was explored in this study using aided spectral indices to estimate urban impervious surface based on a study of Guangzhou, southern China. Three spectral indices were employed, namely, normalized differencing built-up index (NDBI), normalized differencing bare-soil index (NDBaI), and albedo. The improved LSMA method included three analytical steps: (1) High-albedo and low-albedo fraction images were extracted from satellite imagery using conventional LSMA; (2) NDBI, NDBaI, and albedo indices were then employed to remove misclassification pixels in the high-albedo and low-albedo fraction images by establishing thresholds for each index; and (3) accuracy of the impervious surface maps was assessed by using root mean square error (RMSE), mean absolute error (MAE), and systematic error (SE). The results indicate that the overall RMSE (0.10) was achieved for the impervious surface map estimated using the improved LSMA compared to 0.15 when conventional LSMA was employed. The improved LSMA method provides a way to enhance urban impervious surface estimation when LSMA is applied, a common approach for median spatial-resolution image mapping.