Abstract

Precise and regularly updated maps of surface water extent are essential for wetland management. Since it is often challenging to obtain water extent information through ground surveys due to accessibility, satellite remote sensing has become a critical and cost-effective tool for acquiring this information in a temporal context. The methods that are commonly used include supervised and unsupervised multispectral classification, density slicing of a single band and spectral indices. However, optical data are substantially affected by cloud cover and solar illumination, thus, radar data comprise an ideal option. This article presents a methodological framework for rapid delineation of waterbodies' boundaries. For that purpose, an object-based approach is implemented. Sentinel-1 data were obtained and the Mean-Shift segmentation algorithm was employed. The proposed methodology produced promising results achieving an overall accuracy of 98%, a producer's accuracy for waterbodies of 90% and a user's accuracy for waterbodies of 95%.

摘要

精确且定期更新的地表水范围地图对于湿地管理至关重要。由于可访问性，通过地面调查获取水域范围信息通常具有挑战性，因此卫星遥感已成为在时间背景下获取此信息的关键且具有成本效益的工具。常用的方法包括有监督和无监督的多光谱分类、单波段的密度切片和光谱指数。然而，光学数据受云层和太阳光照的影响很大，因此，雷达数据是一个理想的选择。本文提出了一个快速划定水体边界的方法框架。为此，实现了基于对象的方法。获得 Sentinel-1 数据并采用 Mean-Shift 分割算法。所提出的方法产生了可喜的结果，实现了 98% 的整体准确度，生产者对水体的准确度为 90%，用户对水体的准确度为 95%。