Coastal and inland water pixels extraction algorithm (WiPE) from spectral shape analysis and HSV transformation applied to Landsat 8 OLI and Sentinel-2 MSI

REMOTE SENSING OF ENVIRONMENT

卷:223

Article

Dat Dinh Ngoc^1,2,3  Loisel, Hubert^2,3  Jamet, Cedric^2,3  Vantrepotte, Vincent^2,3  Duforet-Gaurier, Lucile³  Chung Doan Minh¹  Mangin, Antoine⁴ 

[1] VAST, Space Technol Inst, 18 Hoang Quoc Viet, Hanoi, Vietnam

[2] Univ Sci & Technol Hanoi, VAST, LOTUS, 18 Hoang Quoc Vie, Hanoi, Vietnam

[3] Univ Lille, Univ Littoral Cote dOpale, Lab Oceanol & Geosci, CNRS,UMR 8187,LOG, 32 Ave Foch, Wimereux, France

[4] ACRI ST, F-06904 Sophia Antipolis, France

关键词: Medium spatial resolution sensors;    Coastal and inland waters;    Water pixel extraction;    Cloud mask;   

DOI  :  10.1016/j.rse.2019.01.024

来源: Elsevier

【 摘 要 】

Identification of water pixels over natural water bodies is a prerequisite step prior to applying algorithms dedicated to the estimation of bio-optical properties of surface waters from remote sensing observations. For visible remote sensing sensors, clouds affect the quantity and quality of the observations, directly by hiding part of the scene and indirectly by their shadows. A certain level of confusion could occur for detection of clouds over turbid (i.e. bright) waters and for detection of their shadows over any kind of surface water. Some algorithms exist but their performance is not satisfactory, especially over turbid waters where cloud-free pixels are sometimes classified as cloud or land, leading to a loss of data. This is particularly important for medium spatial resolution observations such as those performed by the Operational Land Imager (OLI) sensor on Landsat-8 or the Multispectral Instrument (MSI) on Sentinel-2 (a and b). In the frame of this study, we developed a two-step algorithm for the extraction of water pixels (referred to as WiPE) for these medium spatial resolution sensors. In contrast to other approaches based on the top of atmosphere (TOA) reflectance, this algorithm uses the Rayleigh-corrected TOA reflectance (rho(rc)(lambda)) as input parameter allowing the spectral signature of each object to be better characterized. The first step, based on the rho(rc)(lambda) spectral shape analysis of each object, allows water pixels to be discriminated from cloud, vegetation, barren land, and constructions pixels. The second step, in which the rho(rc)(lambda) spectra are transferred into the Hue-Saturation-Value space, greatly improves the detection of cloud shadow over waters. This second step, based on the processing of the whole image, does not require any knowledge on the location and altitude of clouds. Thin clouds are identified during the two steps of the algorithm. This algorithm has been successfully tested over a broad range of environments. WiPE, specifically designed for the extraction of water pixels, generally shows better performance over turbid waters than the standard algorithm developed for Landsat imagery (Fmask). This is explained by the fact that Fmask does not specifically focus on the detection of water pixels, but on the masking of cloud, cloud shadow, and snow over land and water.

【 授权许可】

Free