关键词: 时序信息, 不透水面, 先验知识, 级联提取, 精度差异, Landsat OLI, 波谱特征, 遥感指数

Abstract:

Impervious surface acts as an important technical indicator of urban characterization and regional urbanization dynamics. Its location, patch size, and spatial distribution are widely used by the communities of surface hydrothermal cycle and energy balance. However, traditional methods usually adopt a single time-phase image to retrieval the impervious surface information without considering the rich information implied in multiple time-phase images. Therefore, this paper presented a method to extract the impervious surface by integrating features in time-series images within a year. We generated the Normalized Difference Vegetation Index (NDVI), Modified Normalized Difference Water Index (MNDWI), and Normalized Difference Built-up Index (NDBI) from time-series Landsat-8 OLI surface reflectance images to analyze the intra-annual features of multi-temporal spectrum curves between typical objects and summarize principles in describing changes of impervious surface from multiple time phases. We also combined the prior knowledge of effective impervious surface information to improve the extraction accuracy. Finally, we used field survey data and the 30-m impervious map generated from the 2-m GF-1 image by screen digitalization to validate the accuracy of results using three combinations of images as input (i.e., single time-phase image, four images of seasons, and multi-temporal images). The results show that the impervious surface extracted from the single-phase image had the lowest accuracy, while the extraction accuracy using multi-temporal images was the highest with an overall accuracy of 93.66% and a Kappa coefficient of 0.81. The extraction accuracy using the four images of seasons was in the middle. Furthermore, the presented method showed potential advantages of effectively identifying impervious surfaces in rural areas. Our method provides a new idea for the impervious surface extraction through integrating temporal spectral features of impervious surface.

Key words: time-series features, impervious surface, prior knowledge, cascade retrieval, accuracy comparison, Landsat OLI, spectrum characteristic, remote sensing index