天然河流以一定的边界形态承载着水流及物质的输移，对河流的描述和研究往往通过研究河流信息实现，如提取河流断面形态等几何信息，研究流速、流量等水流特征信息等。但发育在山区的众多河流因地势险要无法仅通过现场量测获取足够的基础信息，是典型的缺资料地区。随着遥感技术的飞速发展，通过多源遥感数据提取山区等缺资料地区河流的几何信息，并以此为基础获取河流水系的水流特征量，逐渐成为丰富河流基础信息和开展河流地貌演化及水文过程模拟等的主要方式。但现有提取河流几何信息和水流特征量的研究多通过经验估算或基于单一数据源，丰度、精度较低，对江河源区、山区河流的效果极差，难以满足日益增长的河流管理与研究的需求。 本文基于多源遥感数据，建立了三维河流几何信息提取方法，并通过各级河流几何信息提升了水文模型的模拟效果，建立了无资料流域河流碳排放量的估算方法。取得的主要成果为：（1）以多源光学卫星影像为数据源，基于融合自动化法建立了山区河流表面信息自动提取方法。方法既可用于面积小于104 km2小流域，又适用于小流域组成的流区或大型流域，推广性强。不同区域河流提取精度均超90%，提取能力及精度优于现有方法。同时，以河流表面积为例，高分辨率（2m）较中分辨率（30m）提取结果增加了5.4倍，证实了提升分辨率对充分提取河流信息的重要性。（2）基于高分辨率河流表面信息，建立了非监督的河宽自动提取方法，提取精度优于半个像素。基于河宽及卫星水位数据，建立了具备一定推广性的山区河流断面概化模型，提取了各级河流的断面形态，与河流表面结合获得了各级河流三维几何信息。（3）开发了国产高分辨率卫星影像的批量预处理算法，基于500余景国产GF-1影像及千景Sentinel-2影像，建立了黄河上游上段25万km2融合分辨率的平滩河流表面信息及河宽数据库，所得河流表面积较现有研究提升3.1倍。（4）基于各级河流三维几何信息改进了水文模型，提升模拟精度超25%。结合模拟所得流域各级河流的水流特征量估算了典型间歇性河流年CO2排放量，纠正了现有研究因高估河流表面积及CO2排放时间导致的严重高估问题。本研究可为联合使用多源遥感数据开展缺资料地区大范围、高精度的河流几何信息提取提供有效方法，为基于河流水系几何特征的河流地貌演化、水文过程模拟及河流碳排放等研究提供基础数据。
Natural rivers carry water and materials within a certain boundary geometry. Research on rivers often involves extracting geometric information of river surfaces and boundaries or hydraulic characteristics such as flow velocity and discharge. Many rivers of China developed in mountainous areas which cannot be measured in the field to obtain enough basic data, for the mountainous areas of complex topography and difficult to access. So mountainous rivers are typical poorly gauged basins, and it is difficult to conduct relevant research. With the rapid development of remote sensing technology, extracting geometric information of mountainous rivers and obtaining hydraulic characteristics based on this information has gradually become the primary way to enrich basic river information and conduct studies on river geomorphology and hydrological processes. However, existing methods for extracting river geometric information and hydraulic characteristics often rely on empirical estimation or single data sources, resulting in low richness and accuracy, and poor performance for river management and research in river source areas and mountainous regions.In this study, a three-dimensional river geometric information extraction method was established based on multi-source remote sensing data, and the simulation effect of hydrological models was improved through the enhancement of geometric information abundance at all river orders. An estimation method for the carbon emissions of data-poor river basins was also developed. The main achievements include: (1) Using multi-source optical satellite images as data sources and a fusion automation method to establish an automatic extraction method for mountainous river surfaces. This method is suitable for small watersheds with an area less than 104 km2, as well as for larger watersheds composed of small watersheds, with strong applicability. The extraction accuracy of rivers in different regions is over 90%, and the extraction capability and accuracy exceed existing methods. At the same time, taking areas of river surfaces as an example, the high-resolution (2m) extraction result is 5.4 times higher than the medium resolution (30m), demonstrating the importance of increasing resolution to fully extract river geometric information. (2) Based on high-resolution river surfaces, an unsupervised automatic extraction method for river width was established with an accuracy exceeding half a pixel. Based on river width and satellite water level data, a generalization model for cross-section morphology of mountainous rivers was established, and the cross-section morphology of all river orders was extracted, which was then combined with river surfaces to obtain three-dimensional geometric information of rivers at all orders. (3) A batch preprocessing algorithm for domestic high-resolution satellite images was developed, and a database of surface information and river width for the Upper Yellow River with an area of 250,000 km2 was established based on more than 500 china-made GF-1 images and thousands of Sentinel-2 images, resulting in a 3.1-fold increase in area of river surfaces compared to existing research. (4) Improved hydrological models based on the three-dimensional geometric information of all order rivers increased simulation accuracy by more than 25%. Based on the hydraulic characteristics of all order rivers obtained through simulation, the annual CO2 emissions of typical intermittent rivers were estimated, correcting the serious overestimation caused by overestimating area of river surfaces and CO2 emission time in existing research. This study provides an effective method for extracting river geometry on a large scale with high accuracy in poorly gauged basins using multi-source remote sensing data, and provides basic data for studies on river geomorphology, hydrological processes simulation, and river carbon emissions based on river system geometric features.
