一种基于NDISI的复合权重波段双差值不透水面提取指数

针对现有遥感指数提取不透水面取结果中混有沙地、裸土等噪声的问题,本文在传统NDISI的形式基础上,提出一种新型的复合权重双差值不透水面指数（Composite-Weighted Double-Difference Impervious Surface Index, CWDDISI）。通过波段的2次差值扩大不透水面和裸地的光谱表现差距,并以植被指数和夜光灯数据作为约束权重,以此提高热红外波段中的不透水面信息比重的同时降低噪声地物的干扰。本文利用Landsat8 OLI-TIR、Landsat7 ETM⁺以及Sentinel-2A光谱数据,结合珞珈一号、DMSP-OL以及VIIRS/DNB夜光数据,选取广州市、西安市、咸阳市以及深圳市、北京市为实验区展开对比实验。研究结果：① CWDDISI具有很好的多区域适用性。在2018年的数据集上,相较于NDISI,CWDDISI在以山地为主的广州市试验区和以平原为主的西安市、咸阳市实验区中,其不透水面提取总精度分别提高了6.02%和7.56%,Kappa系数提高了0.078和0.104; ② CWDDISI具有很好的多时相数据适用性。实验选取2002年和2016年的Landsat7 ETM⁺多时相数据,以深圳市和北京市为实验区展开对比;相较于NDISI,CWDDISI的不透水面总精度分别提高了1.74%和2.13%,Kappa系数分别增加了0.028和0.076。通过实验对比结果可证明,CWDDISI能够克服传统不透水面指数难以区分不透水面信息和裸土区域的问题,为后续不透水面指数的研究提供参考价值。

A NDISI-based Double-Differenced Remote Sensing Index with Composite-Weights for Impervious Surface Information Estimation

The existing impervious surface spectral indices tend to suffer from the disturbance of sand and bare land, which leads to the unsatisfied results in impervious area extraction. To handle this problem, based on Normalized Difference Impervious Surface Index (NDISI), this paper proposes a Composite-Weighted Double-Difference Impervious Surface Index (CWDDISI). CWDDISI increases the gap of spectral feature between the Impervious Surface (IS) and bare land by measuring two difference calculations with the specific three bands. Then, by integrating the Normalized Difference Vegetation Index (NDVI) and night-time light luminosity, CWDDISI achieves the constrain principle, which works to enhance impervious surface information while depressing bare land characteristics in the thermal band. In this paper, remote sensing data of Landsat8 OLI-TIR, Landsat7 ETM+, Sentinel-2A, Luojia1-01, DMSP-OL, and VIIRS/DNB were used. Guangzhou, Shenzhen, Xian, Xianyang, and Beijing were selected as study areas. Comparative experiments showed that CWDDISI performed well across different study areas and landforms. For the dataset of 2018, compared with NDISI, CWDDISI improved the overall accuracy and Kappa value by 6.02% and 0.078, respectively, at Guangzhou, where mountains dominate. In the meantime, CWDDISI improved the overall accuracy and Kappa value by 7.56% and 0.104, respectively, at Xian and Xianyang, where flatlands dominate. For the Landsat ETM+ data in 2002 and 2016, compared with NDISI, the overall accuracy and the Kappa value of CWDDISI had improved by 1.74% and 0.028, respectively, in Shenzhen and improved by 2.13% and 0.076, respectively, in Beijing. After analyzing all these comparative results, this study found that CWDDISI can successfully overcome the confusion between the impervious surface and bare land, which is difficult for most other impervious surface indices. This study provides valuable reference for future researches in impervious surface estimation.