A New Adaptive Remote Sensing Extraction Algorithm for Complex Muddy Coast Waterline

Coastline is an important geographical element of the boundary between ocean and land. Due to the impact of the ocean-land interactions at multiple temporal-spatial scales and the intensified human activities, the waterline of muddy coast is undergoing long-term and continuous dynamic changes. Using traditional remote sensing-based waterline extraction methods, it is difficult to achieve ideal results for muddy coast waterlines, which are faced with problems such as limited algorithm stability, weak algorithm migration, and discontinuous coastlines extraction results. In response to the above challenges, three different types of muddy coasts, Yancheng, Jiuduansha and Xiangshan were selected as the study areas. Based on the Sentinel-2 MSI images, we proposed an adaptive remote sensing extraction algorithm framework for the complex muddy coast waterline, named AEMCW (Adaptive Extraction for Muddy Coast Waterline), including main procedures of high-pass filtering, histogram statistics and adaptive threshold determination, which has the capability to obtain continuous and high-precision muddy coastal waterline. NDWI (Normalized Difference Water Index), MNDWI (Modified Normalized Difference Water Index) and ED (Edge Detection) methods were selected to compare the extraction effect of AEMCW method. The length and spatial accuracy of these four methods were evaluated with the same criteria. The accuracy evaluation presented that the length errors of ED method in all three study areas were minimum, but the waterline results were offset more to the land side, due to spectral similarity, turbid water and tidal flats having similar values of NDWI and MNDWI. Therefore, the length and spatial accuracies of NDWI and MNDWI methods were lower than AEMCW method. The length errors of the AEMCW algorithm in Yancheng, Jiuduansha, and Xiangshan were 14.4%, 18.0%, and 7.7%, respectively. The producer accuracies were 94.3%, 109.6%, and 94.2%, respectively. The user accuracies were 82.4%, 92.9%, and 87.5%, respectively. These results indicated that the proposed AEMCW algorithm can effectively restrain the influence of spectral noise from various land cover types and ensure the continuity of waterline extraction results. The adaptive threshold determination equation reduced the influence of human factors on threshold selection. The further application on ZY-1 02D hyperspectral images in the Yancheng area verified the proposed algorithm is transferable and has good stability.