Landsat 8 OLI image based terrestrial water extraction from heterogeneous backgrounds using a reflectance homogenization approach

• Published in: Remote sensing of environment 2015-12, Vol.171, p.14-32
• Main Authors: Yang, Yuhao, Liu, Yongxue, Zhou, Minxi, Zhang, Siyu, Zhan, Wenfeng, Sun, Chao, Duan, Yuewei
• Format: Article
• Language: English
• Subjects: Fuzzy clustering technique; Landsat OLI; Reflectance homogenization; Seasonal analysis; Surface water extraction; Water index
• ISSN: 0034-4257; 1879-0704
• DOI: 10.1016/j.rse.2015.10.005

Surface waters are fundamental resources for terrestrial life, yet they are not free of both natural and anthropogenic influences at global-scale. An accurate and robust method to extract water bodies is critical to effectively manage these irreplaceable resources. Conventional methods are frequently limited in terms of the uncertainty related to the coarse resolution and regional reflectance heterogeneity of satellite images. The fuzzy clustering method (FCM) considering local spatial information has a proven capability to compensate for these limitations. Nevertheless, this technique is highly sensitive to immense false signals in original satellite images. Therefore, a systematic surface water extraction method by taking advantage of the complementarity between a water index (WI) and a modified FCM (WIMFCM) was designed in this study to improve the water extraction accuracy, the rationale of which is a background reflectance bias correction. Applications were performed to sixteen test sites varying from coasts to inland waters to comprehensively evaluate the reliability of the WIMFCM using the Landsat-8 Operational Land Imager (OLI) images. Results showed the WIMFCM improved the accuracy of water extraction in comparison to alternative methods in terms of kappa coefficients (KCs) and total classification errors (TEs). Overall, the mean KC of the WIMFCM was 0.94 and its mean TE was 11.39%, compared with original WI (KC=0.89, TE=19.84%) and a support vector machine (SVM) method (KC=0.89, TE=22.39%). In addition, a seasonal analysis revealed the WIMFCM could maintain consistent reliability throughout the year, demonstrating its potential for accurate dynamic water monitoring associated with seasonal water probability mask. Additional tests using Moderate-Resolution Imaging Spectroradiometer (MODIS) data revealed the WIMFCM could be extended to large-scale regions, as well as be used in near-real time surface water body extraction. The findings of this study offer a new method to improve target detection accuracy under reflectance heterogeneous environments. •We designed a method to improve water detection accuracy in heterogeneous contexts.•It was validated from different regions globally and various seasons.•It was not limited to specific water bodies and sensors with similar spectral bands.•Advantages of water index technique and a fuzzy clustering method were complemented.•Accuracies were thoroughly analyzed in accordance with various water body types.