Supplementary information files for Extraction of connected river networks from multi-temporal remote sensing imagery using a path tracking technique

Supplementary files for Extraction of connected river networks from multi-temporal remote sensing imagery using a path tracking technique. Precise delineation of river networks is important for accurate hydrological and flood modelling. Whilst remote sensing (RS) has showed great potential in monitoring hydrological changes over space and time, the existing RS-based methods extract river networks based on local morphologies and seldom take into account the overall hydrological connectivity of the rivers. The existing methods also commonly neglect the effect of seasonal variation of water surfaces and the existence of temporary water bodies, which deteriorate the precision of positioning river networks. To address these challenges, a new two-stage method is developed to Extract spatiotemporal variation of water surfaces based on Multi-temporal remote sensing Imagery and Delineate connected river networks with improved accuracy (EMID method for short) using a path tracking technique. The EMID method delineates connected river networks using (a) multi-temporal imagery and a Random Forest model to synoptically map the location and extent of water surfaces under different hydrological conditions, and (b) an optimization algorithm to find the best river paths based on water-occurrence frequency. Four drainage basins with various river morphologies are considered to validate EMID. Comparing with alternative methods, the EMID method consistently produces river network results with improved accuracy in terms of stream location, river coverage and network connectivity.

《基于路径跟踪技术从多时相遥感影像提取连通河网》配套补充资料。精准勾勒河网轮廓，对高精度水文与洪水模拟具有重要意义。遥感（Remote Sensing, RS）技术在时空尺度监测水文变化方面展现出巨大潜力，但现有基于遥感的河网提取方法多仅依托局部地貌特征，极少考虑河流整体水文连通性。此外，现有方法通常忽略水面季节性变化与临时水体的影响，这会降低河网定位精度。为应对上述挑战，本研究开发了一种全新两阶段方法：先基于多时相遥感影像提取水面时空变化特征，再通过路径跟踪技术精准勾勒连通河网（简称EMID方法，全称为"基于多时相遥感影像提取水面时空变化与精准勾勒连通河网"）。EMID方法通过两大步骤实现连通河网勾勒：（1）利用多时相遥感影像与随机森林（Random Forest）模型，概绘不同水文条件下水面的位置与范围；（2）依托优化算法，基于水体出现频率筛选最优河道路径。本研究选取四类具有不同河道地貌特征的流域对EMID方法进行验证。与其他同类方法相比，EMID方法在河道位置、河网覆盖度与网络连通性等指标上均能持续获得精度更优的河网提取结果。