地球信息科学学报 ›› 2018, Vol. 20 ›› Issue (1): 89-98.doi: 10.12082/dqxxkx.2018.170414

• 遥感科学与应用技术 • 上一篇    下一篇

基于Landsat 8 QA云标识的云影识别方法研究

王蔷1,2(), 黄翀1,*(), 刘高焕1, 刘庆生1, 李贺1, 陈卓然1,2   

  1. 1. 中国科学院地理科学与资源研究所 资源与环境信息国家重点实验室,北京 100101
    2. 中国科学院大学,北京 100049
  • 收稿日期:2017-09-05 修回日期:2017-11-06 出版日期:2018-01-20 发布日期:2018-02-06
  • 通讯作者: 黄翀 E-mail:wangq.15s@igsnrr.ac.cn;huangch@lreis.ac.cn
  • 作者简介:

    作者简介:王 蔷(1992-),女,硕士生,主要从事黄河三角洲土地覆盖研究。E-mail: wangq.15s@igsnrr.ac.cn

  • 基金资助:
    国家自然科学基金项目(41471335、41661144030);资源与环境信息系统国家重点实验室自主创新项目(O88RA303YA)

Cloud Shadow Identification Based on QA Band of Landsat 8

WANG Qiang1,2(), HUANG Chong1,*(), LIU Gaohuan1, LIU Qingsheng1, LI He1, CHEN Zhuoran1,2   

  1. 1. State Key Laboratory of Resources and Environment Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2017-09-05 Revised:2017-11-06 Online:2018-01-20 Published:2018-02-06
  • Contact: HUANG Chong E-mail:wangq.15s@igsnrr.ac.cn;huangch@lreis.ac.cn
  • Supported by:
    Natural Science Foundation of China, No.41471335,41661144030;Innovation Project of State Key Laboratory of Resources and Environment Information System, No.O88RA303YA.

摘要:

Landsat系列卫星数据是对地观测研究中应用最为广泛的遥感数据源之一,但是Landsat数据易受云及云影的影响,因此,在Landsat数据的应用中,云和云影的识别十分关键。美国地质调查局(United States Geological Survey,USGS)在其分发的最新的Landsat 8 数据中新增了一个质量评估(Quality Assessment)波段,能快速提供高精度的云掩膜,然而并不能识别云影。本文在Landsat 8 QA波段云识别基础上,对影像的近红外和短波红外波段进行种子填充变换,提取影像中的潜在云影,采用非监督分类的方法识别影像中的水体,将水体从潜在云影中去除。利用太阳方位角和太阳高度角对云及云影相对位置的影响,对云和云影进行匹配,识别真实的云影。利用全球云和云影验证数据集对本文的云影识别结果进行了精度评价,结果表明:不同生态区域云影识别精度达到87%以上。与Fmask云影检测方法相比,本文方法所需波段数更少,流程简单,简化了云高度估算和视角问题,可以快速、准确地识别云影,对基于Landsat 8数据的定量分析或时序研究有重要价值。

关键词: Landsat 8, 云影, QA波段, L8 Biome, 种子填充变换

Abstract:

The Landsat program began in 1972, providing valuable scientific data for recording surface dynamics. Landsat data is vulnerable to cloud and cloud shadow. Abnormal pixel values caused by cloud and cloud shadow affect scientific calculation. Cloud and cloud shadow detection is the first step to scientific research using remote sensing data. Newly established cirrus band in Landsat 8 OLI data has the capacity to provide cloud mask as quickly as possible, but the cloud shadow hasn’t been marked. A new method for cloud shadow identification in Landsat 8 imagery is proposed in this paper, based on the Landsat collection 1 level-1 quality assessment (QA) band. First, the cloud pixels are identified using cloud mask stored in QA band. Then, flood-fill transformation algorithm is applied to near-infrared (NIR) band and short-wavelength infrared (SWIR) band to identify potential cloud shadow. After this step, cloud shadow can be discriminated from bright features. However, it will be confused with the dark objects such as water bodies. It is necessary to remove water bodies from the potential cloud shadow. Iterative Self-organizing Data Analysis Technique (ISODATA) is further used to distinguish water from potential cloud shadow. Third, the solar elevation angle and the solar azimuth are employed to match the position of cloud and cloud shadow. The solar elevation influences the distance between cloud and cloud shadow, and the solar azimuth affects the relative direction of cloud and cloud shadow. Because the cloud level varies very much, the cloud shadow can be finally identified through matching of cloud and cloud shadow after several iterations of cloud altitude estimation. To assess the accuracy of cloud shadow identification, a new validation dataset “L8 Biome Cloud Validation Masks” is used to test the method. We applied the new method to five biomes (shrubland, barren land, snow/ice, urban area and wetland). The validation results demonstrated that the method performed well in different biomes with the overall accuracy of more than 87%. Especially, the new method achieved an overall accuracy as high as 94.48% in shrubland. In comparison with the Function of mask (Fmask) algorithm, our new algorithm needs fewer Landsat bands but achieves better results, especially in barren land and shrubland with accuracy of 87.99% and 94.48%, respectively (Fmask: 85.38% and 92.02%, respectively). The method proposed here simplifies the process of cloud shadow identification and cloud level estimation, making the QA band of Landsat 8 OLI more valuable. It has the potential to be further developed to produce cloud shadow mask product.

Key words: Landsat 8, cloud shadow, QA band, L8 Biome, flood-fill transformation