Improving Accuracy of Impervious Surface Extraction Based on a Threshold Hierarchical Method (THM)

Abstract

Impervious surface area (ISA) is an important representation of urban area. It is very popular to extract ISA by using linear spectral mixture analysis (LSMA). However, there are still some defects in this method: underestimated in areas with a large amount of ISA. Hence, we designed a threshold hierarchical method (THM) to test this underestimation and understand which scale is the best to mixture. The capacity of the THM and the optimal threshold in the impervious surface extraction are the focus in this work. In THM model, the medium-resolution image (Landsat 8 OLI) and the high-resolution image (Gaofen-2, GF-2) were used, the LSMA and the object-oriented method (OOM) were applied for the area with a larger amount of impervious surfaces, which was extracted from the Landsat 8 OLI image after finishing the LSMA procedure by a threshold of the ISA abundance data, the GF-2 image was employed to extract the ISA by OOM. The results show that the THM had the capacity to achieve higher ISA extraction accuracy and ameliorate the ISA underestimate problem.