Segmentation-based and rule-based spectral mixture analysis for estimating urban imperviousness

For detailed estimation of urban imperviousness, numerous image processing methods have been developed, and applied to different urban areas with some success. Most of these methods, however, are global techniques. That is, they have been applied to the entire study area without considering spatial and contextual variations. To address this problem, this paper explores whether two spatio-contextual analysis techniques, namely segmentation-based and rule-based analysis, can improve urban imperviousness estimation. These two spatio-contextual techniques were incorporated to a classic urban imperviousness estimation technique, fully-constrained linear spectral mixture analysis (FCLSMA) method. In particular, image segmentation was applied to divide the image to homogenous segments, and spatially varying endmembers were chosen for each segment. Then an FCLSMA was applied for each segment to estimate the pixel-wise fractional coverage of high-albedo material, low-albedo material, vegetation, and soil. Finally, a rule-based analysis was carried out to estimate the percent impervious surface area (%ISA). The developed technique was applied to a Landsat TM image acquired in Milwaukee River Watershed, an urbanized watershed in Wisconsin, United States. Results indicate that the performance of the developed segmentation-based and rule-based LSMA (S-R-LSMA) outperforms traditional SMA techniques, with a mean average error (MAE) of 5.44% and R2 of 0.88. Further, a comparative analysis shows that, when compared to segmentation, rule-based analysis plays a more essential role in improving the estimation accuracy.