An improved global river vector dataset based on multi-source river data fusion Authors: Yesen Liu, Jianhua Wang, Changjun Liu, Yaohuan Huang, Yuanyuan Liu, Jie Liu, Fuxin Chai, Sheng Chen, Min Li, Wei Qu Published: 2025 ABSTRACT High-precision global river datasets are crucial for hydrology and environmental research. Although several global river datasets have been developed and widely adopted, they often suffer from significant deviations from actual river distributions in many areas. In this study, we propose a multi-source vector river data fusion framework to generate a high spatial accuracy global river dataset with topological information, named GSriver. By integrating high-spatial-resolution but topologically incomplete OpenStreetMap (OSM) waterways with HydroRIVERS and supplementing missing segments with the Global River Topology (GRIT) dataset, GSriver preserves complete river topology while significantly enhancing spatial accuracy. Validation against the high-precision NHDPlus dataset of the United States reveals that GSriver improves spatial accuracy over MERIT, GRIT, and HydroRIVERS by 36.3%, 40.7%, and 56.7%, respectively. More than 40% of nodes in GSriver deviate less than 10 meters from NHDPlus. This approach addresses the spatial accuracy limitations of conventional DEM-derived river datasets by leveraging the advantages of crowdsourced data, offering a scalable and cost-effective solution for constructing large-scale river datasets. GSriver is publicly available at 10.6084/m9.figshare.30119851.v31. Introduction High-precision large-scale river network datasets play a vital role in hydrological modelling, water resources planning, flood hazard assessment, and basin ecosystem preservation2–6. To address this need, a variety of global-scale river datasets have been developed and widely utilized, including HydroRIVERS7, MERIT Hydro8, MERIT Hydro Vector9, Global River Widths from Landsat Database(GRWL)10, Hydrography90m11, Global River Topology(GRIT)12, etc. However, existing global datasets vary significantly in spatial resolution, timeliness, consistency of river delineation, and data integration13,14. The precision limitation across these global river datasets is mainly due to their reliance on Digital Elevation Models (DEMs). For instance, HydroRIVERS, as one of the widely adopted and continuously updated global river datasets, is derived from SRTM DEM data. The inherent limitations of the 90 m resolution of SRTM results in significant spatial accuracy variations, particularly in flat terrains where river channel positioning errors can reach up to ±3.2 km8. The MERIT Hydro datasets, initially released in 201715 and its subsequent vector-based version, MERIT Hydro Vector9, improve global hydrographic data quality in spatial resolution(3 arc-seconds) and vertical accuracy(±2 m) by fusing multi-source DEM data through machine learning. Hydrography90m11 employs a 0.05 km² flow accumulation threshold to establish topological connectivity for 726 million river segments globally, demonstrating a 28% improvement in small stream detection efficiency compared to NHDPlus. In parallel, retrieval approaches from remote sensing imagery are adopted for global river datasets delineation. The Global River Widths from Landsat(GRWL) dataset14 utilizes 30 m resolution Landsat imagery to estimate global river networks, revealing an extended total length of 7.73 million kilometres for rivers exceeding 90 m width. However, this optical remote sensing approach exhibits inherent limitations due to spatial and temporal resolution constraints, resulting in systematic underestimation of narrow and seasonal channels. Comparatively, crowdsourced geographic data offers an alternative approach to generating global river datasets. While this open-source dataset demonstrates relatively high spatial accuracy overall, its data quality exhibits significant regional heterogeneity. Take the OpenStreetMap(OSM) Waterways layer16 as an example, it includes over 27 million river segments globally, which are collected through crowdsourcing efforts. In urban areas, the average positional accuracy of OSM reaches ≤15 meters, comparable to professional surveying standards. However, its positional errors in remote regions can extend to hundreds of meters, with notable distortions in river morphology17. Nonetheless, OSM offers fine spatial detail in many regions, surpassing DEM-derived datasets in visual accuracy. Compared with global river datasets, numerous countries have developed high-resolution river datasets at the national scale, such as the United States’ National Hydrography Dataset (NHD). The NHD’s hierarchical architecture and public-private collaborative governance model have established a benchmark for hydrological data management internationally. The latest iteration, NHDPlus V2, achieves significant advancements by integrating 10-meter resolution DEM with comprehensive watershed data, enabling high-precision modelling of river network topology and dynamic flow simulation. With spatial accuracy conforming to 1:24,000 mapping standards and encompassing over 24 million river segments, this dataset has emerged as a critical validation benchmark for global-scale hydrographic data products12. China has also developed a national river network dataset comprising approximately 520,000 river segments derived from 1:50,000 topographic data (with approximately 30 m resolution). This high-precision dataset has been specifically applied to flood forecasting for mountain torrent disasters, demonstrating particular effectiveness in early warning systems for small and medium-sized rivers in mountainous areas. However, it should be noted that this dataset has not been made publicly available, which limits its potential applications in broader hydrological research and international comparative studies18. Furthermore, the significant disadvantages of these national river datasets are the scale insufficiencies to global hydrological research. Globally, current river datasets are primarily generated through DEM-based, remote sensing-based and crowdsourcing-based approaches. The predominant methodology involves extracting river networks from DEM data based on hydrological flow accumulation principles, which produces river systems with complete topological relationships19. The extraction of river networks from DEM is inherently constrained by limitations related to data resolution, depression artifacts, and topographic complexity in flat terrains. These challenges necessitate iterative pit-filling and flow-direction correction procedures, which often result in channel positional deviations and tributary omission errors. In high-relief mountainous regions, the derived networks frequently exhibit flow-path discontinuities and directional inaccuracies, while low-gradient floodplains suffer from distorted representations due to insufficient elevation variations20. A persistent methodological issue concerns the generation of unnatural vertices(kinks) along extracted channels, producing angular drainage patterns that contradict the smooth sinuosity characteristic of natural fluvial systems. This artifact has been consistently observed across multiple contemporary datasets, revealing a fundamental limitation in current DEM-based extraction algorithms20,21. The other methodological approach involves river network extraction through remote sensing image classification. The limitations of remote sensing-based approach are derived from the spatial resolution constraints and temporal update frequency of satellite imagery, coupled with the seasonal variability of river systems22,23. These factors result in significant weather dependency, omission of small tributaries and ephemeral streams, and inadequate representation of narrow channels24. Current mainstream medium-resolution imagery (e.g., 30m-class sensors) fails to detect tributaries narrower than 15 meters or artificial canals. While high-resolution imagery (e.g., 0.5 m) improves feature delineation, it introduces new challenges including limited spatial coverage, exponential growth in data volume, processing inefficiency, and geometric distortions in mountainous terrain with steep slopes. These technical constraints collectively hinder the comprehensive extraction of complete river networks through purely remote sensing-based approaches. Although recent advances have incorporated artificial intelligence algorithms to process complex channel morphologies(e.g., braided rivers, anabranch systems) and demonstrate measurable improvements in remote sensing based approach, these artificial intelligence approaches still require manual correction and fail to provide comprehensive solutions to the fundamental challenges of river network extraction and representation25,26. Moreover, most remote sensing-based datasets also lack network topology, making flow modelling difficult. Above all, existing global river network datasets exhibit complementary strengths and limitations. With the growing availability of diverse geospatial data, the fusion of multi-source river datasets emerges as a promising direction. In this study, we propose a multi-source fusion framework for river data integration, designed to systematically reconcile disparate resolution datasets (from continental-scale to local precision) into a unified hydrographic product. Based on the framework, we develop a global river dataset, GSriver, that fuses OSM Waterways and HydroRIVERS, with GRIT serving as a supplement in areas lacking OSM data. GSriver was rigorously evaluated through comprehensive testing against with three benchmark datasets of HydroRIVERS, OSM Waterways, and GRIT. The objective is to preserve network topology while significantly improving spatial accuracy—leveraging the fine resolution of crowdsourced data and the completeness of DEM-derived datasets. Methods We devised a multi-resolution vector data fusion framework by integrating high-precision coordinate information into existing river networks to produce GSriver. In this study, GSriver is generated by fusing HydroRIVERS with OSM waterways, while utilizing GRIT as a supplementary data source in regions where OSM waterways coverage is incomplete. To these three vector datasets, there are two fundamental technical challenges for data fusion:(1) establishing accurate correspondences between HydroRIVERS and their counterparts in OSM waterways or GRIT, and (2) effectively integrating the high-resolution coordinates from OSM waterways or GRIT into the HydroRIVERS river network topology. To overcome these challenges, our approach implemented several key solutions including HydroRIVERS reaches integration, OSM Waterways identification, multi-data fusion, GRIT supplement, and topology repair. The framework is illustrated in Fig. 1. Given the need to process over 3 million rivers containing more than 500 million coordinate points and to perform multiple spatial topology calculations, our approach involves three key steps. First, we split the data files into individual rivers, treating each river as a separate processing entity. Second, we extract river coordinates into structured text files for processing. Finally, we update the processed results back into the river layer.Fig. 1Framework Architecture. Framework Architecture. Input data We selected HydroRIVERS as the foundational river network dataset due to its widely recognized topological completeness, which has supported numerous scientific studies and applications while undergoing continuous updates. This dataset comprises over 8.47 million reaches, each containing explicit topological attributes in its feature fields. OSM Waterways was chosen as the primary fusion data source. As a crowdsourced dataset, the 2022 version contains approximately 27.22 million rivers. Although this dataset exhibits limitations including incomplete global coverage, regional accuracy inconsistencies, and topological errors, it demonstrates superior spatial precision in many areas - particularly in economically developed and densely populated regions where its positional accuracy often surpasses other global datasets and even approaches surveying-grade precision in certain locations. For regions with missing OSM waterways coverage, GRIT was employed as a secondary fusion source. With its 1.76 million reaches, GRIT effectively compensates for OSM’s coverage gaps, especially in sparsely populated and less developed areas. The spatial accuracy assessment utilized NHDPlus (V2.0) at 1:24,000 scale as the validation benchmark. Covering the United States with 24.43 million reaches, NHDPlus has become a gold standard for river data accuracy evaluation in hydrological research. This selection ensures methodological consistency with existing studies while providing a rigorous precision reference framework. Finally, we conducted a comprehensive accuracy assessment by comparing our newly developed dataset with three established global river products: MERIT, GRIT, and HydroRIVERS. The primary data sources utilized in this study are summarized in Table 1.Table 1Input data.data sourceRelease timepurposeaccessHydroRIVERS(V10)2019①Base River Network②Accuracy Comparisonhttps://www.hydrosheds.org/products /hydroriversOSM waterways2022①Primary Fusion Sourcehttps://www.openstreetmap.org/Global River Topology (GRIT)2024①Supplementary Fusion Source②Accuracy Comparisonhttps://zenodo.org/records/11219313/filesNHDPlus (V2.0)2019①Accuracy Assessmenthttp://nhd.usgs.gov/MERIT Hydro Vector (MERIT)2024①Accuracy Comparison10.6084/m9.fgshare.c.5052635OSM, HydroRIVERS, and GRIT are licensed under the Open Database License (ODbL), Creative Commons Attribution 4.0 International (CC BY 4.0), and Creative Commons Attribution License (CCAL), respectively. These dataset permit use, distribution and reproduction in any medium, provided the original work is properly cited. Table 1Input data.data sourceRelease timepurposeaccessHydroRIVERS(V10)2019①Base River Network②Accuracy Comparisonhttps://www.hydrosheds.org/products /hydroriversOSM waterways2022①Primary Fusion Sourcehttps://www.openstreetmap.org/Global River Topology (GRIT)2024①Supplementary Fusion Source②Accuracy Comparisonhttps://zenodo.org/records/11219313/filesNHDPlus (V2.0)2019①Accuracy Assessmenthttp://nhd.usgs.gov/MERIT Hydro Vector (MERIT)2024①Accuracy Comparison10.6084/m9.fgshare.c.5052635OSM, HydroRIVERS, and GRIT are licensed under the Open Database License (ODbL), Creative Commons Attribution 4.0 International (CC BY 4.0), and Creative Commons Attribution License (CCAL), respectively. These dataset permit use, distribution and reproduction in any medium, provided the original work is properly cited. Input data. ①Base River Network ②Accuracy Comparison ①Supplementary Fusion Source ②Accuracy Comparison OSM, HydroRIVERS, and GRIT are licensed under the Open Database License (ODbL), Creative Commons Attribution 4.0 International (CC BY 4.0), and Creative Commons Attribution License (CCAL), respectively. These dataset permit use, distribution and reproduction in any medium, provided the original work is properly cited. HydroRIVERS Reaches Integration(intHydroRIVERS) HydroRIVERS is structured as discrete river reaches divided at confluences. To align with the conventional concept of a complete river, we first integrated these reaches into continuous river lines. we selected all reaches within independent watersheds with drainage areas exceeding 50 km², encompassing 52,331 watersheds and 4,408,414 reaches—approximately 52% of all reaches in HydroRIVERS. Then, HydroRIVERS reaches integration was conducted hierarchically at the watershed level. For each watershed, the longest flow path from the outlet to the furthest headwater was identified and merged into a single primary river, which was defined as Primary Rivers (Level 1). Reaches flowing into primary rivers were designated as tributaries of level-2 rivers, and so on recursively. The longest upstream paths from these confluences were traced and merged, iterating until all reaches were processed (Fig. 2). In this study, we produced a reconstructed river network(intHydroRIVERS) comprising 748,629 integrated rivers with improved hydrological continuity and topological consistency, which is compatible with river-level spatial matching and datasets fusion in subsequent steps.Fig. 2Comparison of HydroRIVERS Before and After Integration. Comparison of HydroRIVERS Before and After Integration. OSM Waterways Identification Corresponding to intHydroRIVERS Rivers To be a crowdsourced dataset, the OSM Waterways exhibit significant topological inconsistencies, including discontinuous reaches, improper connections between different watercourse types, dual-line representations of wide rivers, and centerline deviations17,27, which makes extracting corresponding OSM waterway to an identified intHydroRIVERS river extremely challenging. In this study, we employed multiple indices between OSM waterways and intHydroRIVERS to address this challenge, including angular deviation, average distance28,29, and overlap ratio. Initially, OSM waterways reaches within a 2 km buffer zone of each intHydroRIVERS river were selected as potential fusion data. Then, both the intHydroRIVERS rivers and initially selected OSM waterways underwent vertex densification to ensure no two adjacent vertices were spaced less than 100 m apart. These densified vertices were then utilized to calculate multiple indices for OSM Waterways identification.Angle constraint. The initial selected OSM waterways are eliminated with excessive angular deviation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff) from the corresponding river of intHydroRIVERS, which is calculated as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}=\left|{A}_{{osm}}-{A}_{hy{dro}}\right|,$$\end{document}Adiff=Aosm−Ahydro,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{osm}}$$\end{document}Aosm is the direction for each OSM waterways, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{hy{dro}}$$\end{document}Ahydro is the angle index of corresponding river of intHydroRIVERS. In this study, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{osm}}$$\end{document}Aosm is calculated as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{osm}}={\rm{atan}}\left(\frac{{y}_{{end}}-{y}_{{start}}}{{x}_{{end}}-{x}_{{start}}},\times ,180,\div,\pi \right),$$\end{document}Aosm=atanyend−ystartxend−xstart×180÷π,where, atan is the inverse tangent(arctangent) function from trigonometry, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{start}}$$\end{document}xstart and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{{start}}$$\end{document}ystart are the x and y coordinates of the first vertice on the OSM waterways line respectively, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{end}}$$\end{document}xend and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{{end}}$$\end{document}yend are the x and y coordinates of the last vertice on the OSM waterways line respectively.For each OSM waterways reach, we identify the nearest vertice on the intHydroRIVERS river corresponding to the reach’s start and end vertices. Then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{hy{dro}}$$\end{document}Ahydro is calculated based on start and end vertices as follows:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{hy{dro}}={\rm{atan}}\left(\frac{{y}_{b}-{y}_{a}}{{x}_{b}-{x}_{a}},\times ,180,\div,\pi \right),$$\end{document}Ahydro=atanyb−yaxb−xa×180÷π,where, (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{\rm{a}}}$$\end{document}xa, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{y}}}_{{\rm{a}}}$$\end{document}ya) and (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{\rm{b}}}$$\end{document}xb, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{y}}}_{{\rm{b}}})$$\end{document}yb) represent the x, y coordinates of the start and end vertices on the intHydroRIVERS river respectively.In this study, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff is adjusted to constrain the angle within 90 degrees:if, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff≥360, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}-$$\end{document}Adiff− 360,if, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}\ge $$\end{document}Adiff≥ 180, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${360-A}_{{diff}}$$\end{document}360−Adiff,if, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}\ge $$\end{document}Adiff≥ 90, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${180-A}_{{diff}}$$\end{document}180−Adiff.Through comparison with actual data, OSM waterways reaches with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}} > 20$$\end{document}Adiff>20 were eliminated. The step of angle constraint primarily removes fragmental and erroneous tributaries in OSM that flow into main rivers. The schematic result of OSM waterways after angle constraint is shown in Fig. 3.Fig. 3Selected OSM waterways.Average distance filtering. We designed a metric for the average distance to the nearest vertex, based on the “Fréchet distance” calculation method, to measure the similarity between two rivers, which is calculated as follows:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{ave}}=\frac{\mathop{\sum }\limits_{n=1}^{m}{d}_{\min (n)}}{m},$$\end{document}Dave=∑n=1mdmin(n)m,5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\min (n)}=\min \left({d}_{n,1},{d}_{n,2},\ldots \ldots {d}_{n,t}\right),$$\end{document}dmin(n)=mindn,1,dn,2,……dn,t,where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{ave}}$$\end{document}Dave represents the distance from an OSM waterways reach to the corresponding intHydroRIVERS river; m denotes the number of vertices in the OSM reach, n is the vertice index, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\min (n)}$$\end{document}dmin(n) indicates the distance from vertice n of the OSM waterways reach to the corresponding intHydroRIVERS river; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{n,t}$$\end{document}dn,t represents the distance from vertice n of OSM waterways to vertice t of intHydroRIVERS.Simultaneously, we calculated the minimum distance (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{\min }$$\end{document}Dmin) from OSM waterways reaches to corresponding intHydroRIVERS rivers, determined as the smallest value among all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\min (n)}$$\end{document}dmin(n). In this study, OSM waterways reaches with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{ave}} > 1000$$\end{document}Dave>1000 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{\min }$$\end{document}Dmin > 100 were eliminated based on comparison with actual data. Average distance filtering primarily removes reaches in OSM that exhibit similar directional patterns to main rivers but are located at considerable distances, while also filtering out some parallel reaches of main rivers - a phenomenon particularly common in plain areas.Overlap Sifting. In our approach, OSM waterways with low overlap ratio with corresponding intHydroRIVERS rivers are further discarded. To obtain the overlap ratio, we calculated the distance between each vertice on intHydroRIVERS rivers and every vertice on each corresponding OSM waterway reach. The nearest vertice of OSM waterway and corresponding OSM waterway reach are identified. Subsequently, the count of nearest vertices (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{near}}$$\end{document}Cnear) is tallied for each OSM waterways reach, and the ratio of nearest vertices to total vertices per OSM waterways reach is calculated as follows:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_{{near}}=\frac{{C}_{{near}}}{{C}_{{total}}},$$\end{document}Pnear=CnearCtotal,where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{near}}$$\end{document}Cnear represents the number of nearest vertices for a given OSM waterways reach, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{total}}$$\end{document}Ctotal is the total number of vertices in that OSM waterways reach. In this study, OSM waterways reaches with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_{{near}}$$\end{document}Pnear < 0.8 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{near}}$$\end{document}Cnear < 5 were eliminated. Overlap Sifting primarily removes shorter OSM reaches and those with only partial overlap with intHydroRIVERS rivers. Angle constraint. The initial selected OSM waterways are eliminated with excessive angular deviation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff) from the corresponding river of intHydroRIVERS, which is calculated as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}=\left|{A}_{{osm}}-{A}_{hy{dro}}\right|,$$\end{document}Adiff=Aosm−Ahydro,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{osm}}$$\end{document}Aosm is the direction for each OSM waterways, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{hy{dro}}$$\end{document}Ahydro is the angle index of corresponding river of intHydroRIVERS. In this study, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{osm}}$$\end{document}Aosm is calculated as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{osm}}={\rm{atan}}\left(\frac{{y}_{{end}}-{y}_{{start}}}{{x}_{{end}}-{x}_{{start}}},\times ,180,\div,\pi \right),$$\end{document}Aosm=atanyend−ystartxend−xstart×180÷π,where, atan is the inverse tangent(arctangent) function from trigonometry, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{start}}$$\end{document}xstart and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{{start}}$$\end{document}ystart are the x and y coordinates of the first vertice on the OSM waterways line respectively, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{end}}$$\end{document}xend and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{{end}}$$\end{document}yend are the x and y coordinates of the last vertice on the OSM waterways line respectively.For each OSM waterways reach, we identify the nearest vertice on the intHydroRIVERS river corresponding to the reach’s start and end vertices. Then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{hy{dro}}$$\end{document}Ahydro is calculated based on start and end vertices as follows:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{hy{dro}}={\rm{atan}}\left(\frac{{y}_{b}-{y}_{a}}{{x}_{b}-{x}_{a}},\times ,180,\div,\pi \right),$$\end{document}Ahydro=atanyb−yaxb−xa×180÷π,where, (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{\rm{a}}}$$\end{document}xa, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{y}}}_{{\rm{a}}}$$\end{document}ya) and (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{\rm{b}}}$$\end{document}xb, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{y}}}_{{\rm{b}}})$$\end{document}yb) represent the x, y coordinates of the start and end vertices on the intHydroRIVERS river respectively.In this study, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff is adjusted to constrain the angle within 90 degrees:if, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff≥360, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}-$$\end{document}Adiff− 360,if, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}\ge $$\end{document}Adiff≥ 180, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${360-A}_{{diff}}$$\end{document}360−Adiff,if, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}\ge $$\end{document}Adiff≥ 90, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${180-A}_{{diff}}$$\end{document}180−Adiff.Through comparison with actual data, OSM waterways reaches with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}} > 20$$\end{document}Adiff>20 were eliminated. The step of angle constraint primarily removes fragmental and erroneous tributaries in OSM that flow into main rivers. The schematic result of OSM waterways after angle constraint is shown in Fig. 3.Fig. 3Selected OSM waterways. Angle constraint. The initial selected OSM waterways are eliminated with excessive angular deviation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff) from the corresponding river of intHydroRIVERS, which is calculated as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}=\left|{A}_{{osm}}-{A}_{hy{dro}}\right|,$$\end{document}Adiff=Aosm−Ahydro,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{osm}}$$\end{document}Aosm is the direction for each OSM waterways, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{hy{dro}}$$\end{document}Ahydro is the angle index of corresponding river of intHydroRIVERS. In this study, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{osm}}$$\end{document}Aosm is calculated as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{osm}}={\rm{atan}}\left(\frac{{y}_{{end}}-{y}_{{start}}}{{x}_{{end}}-{x}_{{start}}},\times ,180,\div,\pi \right),$$\end{document}Aosm=atanyend−ystartxend−xstart×180÷π,where, atan is the inverse tangent(arctangent) function from trigonometry, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{start}}$$\end{document}xstart and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{{start}}$$\end{document}ystart are the x and y coordinates of the first vertice on the OSM waterways line respectively, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{end}}$$\end{document}xend and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{{end}}$$\end{document}yend are the x and y coordinates of the last vertice on the OSM waterways line respectively. For each OSM waterways reach, we identify the nearest vertice on the intHydroRIVERS river corresponding to the reach’s start and end vertices. Then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{hy{dro}}$$\end{document}Ahydro is calculated based on start and end vertices as follows:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{hy{dro}}={\rm{atan}}\left(\frac{{y}_{b}-{y}_{a}}{{x}_{b}-{x}_{a}},\times ,180,\div,\pi \right),$$\end{document}Ahydro=atanyb−yaxb−xa×180÷π,where, (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{\rm{a}}}$$\end{document}xa, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{y}}}_{{\rm{a}}}$$\end{document}ya) and (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{\rm{b}}}$$\end{document}xb, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{y}}}_{{\rm{b}}})$$\end{document}yb) represent the x, y coordinates of the start and end vertices on the intHydroRIVERS river respectively. In this study, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff is adjusted to constrain the angle within 90 degrees: if, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff≥360, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}-$$\end{document}Adiff− 360, if, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}\ge $$\end{document}Adiff≥ 180, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${360-A}_{{diff}}$$\end{document}360−Adiff, if, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}\ge $$\end{document}Adiff≥ 90, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}}$$\end{document}Adiff=\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${180-A}_{{diff}}$$\end{document}180−Adiff. Through comparison with actual data, OSM waterways reaches with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{diff}} > 20$$\end{document}Adiff>20 were eliminated. The step of angle constraint primarily removes fragmental and erroneous tributaries in OSM that flow into main rivers. The schematic result of OSM waterways after angle constraint is shown in Fig. 3.Fig. 3Selected OSM waterways. Selected OSM waterways. Average distance filtering. We designed a metric for the average distance to the nearest vertex, based on the “Fréchet distance” calculation method, to measure the similarity between two rivers, which is calculated as follows:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{ave}}=\frac{\mathop{\sum }\limits_{n=1}^{m}{d}_{\min (n)}}{m},$$\end{document}Dave=∑n=1mdmin(n)m,5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\min (n)}=\min \left({d}_{n,1},{d}_{n,2},\ldots \ldots {d}_{n,t}\right),$$\end{document}dmin(n)=mindn,1,dn,2,……dn,t,where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{ave}}$$\end{document}Dave represents the distance from an OSM waterways reach to the corresponding intHydroRIVERS river; m denotes the number of vertices in the OSM reach, n is the vertice index, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\min (n)}$$\end{document}dmin(n) indicates the distance from vertice n of the OSM waterways reach to the corresponding intHydroRIVERS river; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{n,t}$$\end{document}dn,t represents the distance from vertice n of OSM waterways to vertice t of intHydroRIVERS.Simultaneously, we calculated the minimum distance (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{\min }$$\end{document}Dmin) from OSM waterways reaches to corresponding intHydroRIVERS rivers, determined as the smallest value among all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\min (n)}$$\end{document}dmin(n). In this study, OSM waterways reaches with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{ave}} > 1000$$\end{document}Dave>1000 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{\min }$$\end{document}Dmin > 100 were eliminated based on comparison with actual data. Average distance filtering primarily removes reaches in OSM that exhibit similar directional patterns to main rivers but are located at considerable distances, while also filtering out some parallel reaches of main rivers - a phenomenon particularly common in plain areas. Average distance filtering. We designed a metric for the average distance to the nearest vertex, based on the “Fréchet distance” calculation method, to measure the similarity between two rivers, which is calculated as follows:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{ave}}=\frac{\mathop{\sum }\limits_{n=1}^{m}{d}_{\min (n)}}{m},$$\end{document}Dave=∑n=1mdmin(n)m,5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\min (n)}=\min \left({d}_{n,1},{d}_{n,2},\ldots \ldots {d}_{n,t}\right),$$\end{document}dmin(n)=mindn,1,dn,2,……dn,t,where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{ave}}$$\end{document}Dave represents the distance from an OSM waterways reach to the corresponding intHydroRIVERS river; m denotes the number of vertices in the OSM reach, n is the vertice index, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\min (n)}$$\end{document}dmin(n) indicates the distance from vertice n of the OSM waterways reach to the corresponding intHydroRIVERS river; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{n,t}$$\end{document}dn,t represents the distance from vertice n of OSM waterways to vertice t of intHydroRIVERS. Simultaneously, we calculated the minimum distance (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{\min }$$\end{document}Dmin) from OSM waterways reaches to corresponding intHydroRIVERS rivers, determined as the smallest value among all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\min (n)}$$\end{document}dmin(n). In this study, OSM waterways reaches with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{ave}} > 1000$$\end{document}Dave>1000 or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{\min }$$\end{document}Dmin > 100 were eliminated based on comparison with actual data. Average distance filtering primarily removes reaches in OSM that exhibit similar directional patterns to main rivers but are located at considerable distances, while also filtering out some parallel reaches of main rivers - a phenomenon particularly common in plain areas. Overlap Sifting. In our approach, OSM waterways with low overlap ratio with corresponding intHydroRIVERS rivers are further discarded. To obtain the overlap ratio, we calculated the distance between each vertice on intHydroRIVERS rivers and every vertice on each corresponding OSM waterway reach. The nearest vertice of OSM waterway and corresponding OSM waterway reach are identified. Subsequently, the count of nearest vertices (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{near}}$$\end{document}Cnear) is tallied for each OSM waterways reach, and the ratio of nearest vertices to total vertices per OSM waterways reach is calculated as follows: Overlap Sifting. In our approach, OSM waterways with low overlap ratio with corresponding intHydroRIVERS rivers are further discarded. To obtain the overlap ratio, we calculated the distance between each vertice on intHydroRIVERS rivers and every vertice on each corresponding OSM waterway reach. The nearest vertice of OSM waterway and corresponding OSM waterway reach are identified. Subsequently, the count of nearest vertices (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{{near}}$$\end{document}Cnear) is tallied for each OSM waterways reach, and the ratio of nearest vertices to total vertices per OSM waterways reach is calculated as follows: Moreover, OSM waterway reaches containing river islands were further eliminated. These reaches are characterized in the map as relatively short sections whose start and end points connect to the same OSM waterways. Data fusion of intHydroRIVERS and OSM waterways Once the corresponding OSM waterways were identified, we implemented a data fusion to improve the spatial precision of HydroRIVERS rivers while preserving topological integrity. This process involved several steps:Establishing vertice-pairs. The vertice-pairs between intHydroRIVERS and OSM were matched based on the distance from each vertice of intHydroRIVERS rivers to the nearest identified OSM waterways reach vertices. The two vertices with the minimum distance are determined as the first vertice-pair and excluded from the subsequent matching process. The nearest-neighbour vertice-pairs matching continues in a loop until the distance exceeds 1000 meters.Coordinate Replacement. Based on the vertice-pairs, the coordinates of intHydroRIVERS vertices are replaced with those of OSM waterways vertices. It should be noted that direct replacement may cause artificial crisscrossing of river lines due to the presence of meandering rivers. To resolve this, we assigned sequential indices to the vertices of both intHydroRivers river and OSM waterways reach from start to end. During coordinate replacement, the sequential relationship between OSM and intHydroRIVERS vertices must maintain consistent increases or decrease. Vertices violating this directional consistency are excluded to prevent topological artifacts.Geometry Refinement. When replacing adjacent vertices in intHydroRIVERS, if the corresponding OSM vertices belong to the same OSM reach but are non-adjacent, the intermediate OSM vertices between them are inserted into intHydroRIVERS to refine the geometry. This operation ensures geometric alignment between the intHydroRIVERS and OSM waterways.Isolated Vertices Filtering. The isolated vertices in intHydroRIVERS are further checked. In this study, the short sequence of replaced or unreplaced vertices ( <5 vertices) were excluded to avoid introducing noise, which is validated that retaining such small clusters is unnecessary. Establishing vertice-pairs. The vertice-pairs between intHydroRIVERS and OSM were matched based on the distance from each vertice of intHydroRIVERS rivers to the nearest identified OSM waterways reach vertices. The two vertices with the minimum distance are determined as the first vertice-pair and excluded from the subsequent matching process. The nearest-neighbour vertice-pairs matching continues in a loop until the distance exceeds 1000 meters. Establishing vertice-pairs. The vertice-pairs between intHydroRIVERS and OSM were matched based on the distance from each vertice of intHydroRIVERS rivers to the nearest identified OSM waterways reach vertices. The two vertices with the minimum distance are determined as the first vertice-pair and excluded from the subsequent matching process. The nearest-neighbour vertice-pairs matching continues in a loop until the distance exceeds 1000 meters. Coordinate Replacement. Based on the vertice-pairs, the coordinates of intHydroRIVERS vertices are replaced with those of OSM waterways vertices. It should be noted that direct replacement may cause artificial crisscrossing of river lines due to the presence of meandering rivers. To resolve this, we assigned sequential indices to the vertices of both intHydroRivers river and OSM waterways reach from start to end. During coordinate replacement, the sequential relationship between OSM and intHydroRIVERS vertices must maintain consistent increases or decrease. Vertices violating this directional consistency are excluded to prevent topological artifacts. Coordinate Replacement. Based on the vertice-pairs, the coordinates of intHydroRIVERS vertices are replaced with those of OSM waterways vertices. It should be noted that direct replacement may cause artificial crisscrossing of river lines due to the presence of meandering rivers. To resolve this, we assigned sequential indices to the vertices of both intHydroRivers river and OSM waterways reach from start to end. During coordinate replacement, the sequential relationship between OSM and intHydroRIVERS vertices must maintain consistent increases or decrease. Vertices violating this directional consistency are excluded to prevent topological artifacts. Geometry Refinement. When replacing adjacent vertices in intHydroRIVERS, if the corresponding OSM vertices belong to the same OSM reach but are non-adjacent, the intermediate OSM vertices between them are inserted into intHydroRIVERS to refine the geometry. This operation ensures geometric alignment between the intHydroRIVERS and OSM waterways. Geometry Refinement. When replacing adjacent vertices in intHydroRIVERS, if the corresponding OSM vertices belong to the same OSM reach but are non-adjacent, the intermediate OSM vertices between them are inserted into intHydroRIVERS to refine the geometry. This operation ensures geometric alignment between the intHydroRIVERS and OSM waterways. Isolated Vertices Filtering. The isolated vertices in intHydroRIVERS are further checked. In this study, the short sequence of replaced or unreplaced vertices ( <5 vertices) were excluded to avoid introducing noise, which is validated that retaining such small clusters is unnecessary. Isolated Vertices Filtering. The isolated vertices in intHydroRIVERS are further checked. In this study, the short sequence of replaced or unreplaced vertices ( <5 vertices) were excluded to avoid introducing noise, which is validated that retaining such small clusters is unnecessary. Moreover, each intHydroRIVERS river vertice is annotated to indicate whether it originates from intHydroRIVERS or OSM waterways, which is recorded in the final dataset’s attribute fields. Then, the updated river retained geometry connectivity while adopting improved spatial detail of multi-sources datasets. GRIT datasets supplement To overcome regional heterogeneity of data quality in OSM, GRIT was applied as a supplemental dataset for fusion. We first segmented the rivers and extracted the unmodified portions based on the recorded vertex types (OSM waterways or HydroRIVERS). For all river segments where no modifications were made to the vertices, the same workflow as for OSM waterways was applied, replacing intHydroRIVERS vertices with GRIT vertices. However, gaps or overlaps may exist between rivers fused from GRIT and OSM waterways. To maintain connectivity between adjacent intHydroRIVERS reaches, we identified the closest points between neighbouring intHydroRIVERS sections, and then removed all intermediate vertices between these reference points in both reaches and connected the nearest nodes of adjacent river segments to maintain connectivity between the two segments. River topology repair Following the data fusion of intHydroRIVERS rivers with OSM waterways and GRIT, we conducted a topological repair process to ensure that the fused dataset maintained a consistent, connected, and navigable river network. In this study, we identified the nearest vertices from the upper-level river(tributary) to the lower-level river(mainstream), then discarded all vertices after these nearest vertices in the upper-level river, and finally connected the last vertex of the tributary to the nearest vertices of the mainstream, thereby ensuring spatial connectivity between all rivers. This may potentially remove significant portions of tributary channels. We advise readers to be mindful of this issue when using the data. Rivers that were straightened due to topology repair are recorded in the TopoCheck field of the dataset, with the attribute value “End Repair”. Through the aforementioned processing steps, a new river dataset named GSriver was generated. A comparison between GSriver and multi-sources datasets of HydroRIVERS, OSM and GRIT is shown in Fig. 4.Fig. 4Comparison between GSriver and multi-sources datasets of HydroRIVERS, OSM and GRIT. Comparison between GSriver and multi-sources datasets of HydroRIVERS, OSM and GRIT. Data Record The dataset is available at Figshare1 and consists of six zip files, which are river vector data of the world’s continents. Due to the file size and shapefile format limitations, Russia is saved separately as an individual file. The final GSriver dataset was compiled with complete attribution table. We incorporate additional attributions of river length, fusion ratio, river name and river hierarchy into each original GSriver river (Table 2). The attributions of fusion ratio including the proportion of vertices from OSM waterways and GRIT for each river. To river name, which is always lacked in most DEM-derived river datasets, the information from OSM is applied. In this study, the most frequently occurring name of multiple OSM is selected for each river. Additionally, we include a field storing all associated OSM reach names for each river. For Level 1 rivers, names have been manually verified against actual names, and we plan to gradually refine names for other rivers using more accurate datasets in the future. Furthermore, the attribution of river hierarchy, which is retrieved based on an improved Gravelius classification method30, is also included. In GSriver, global rivers are classified to 8 levels. Current river classification methods are based on natural attributes, yet these attributes are unsuitable for map representation. For instance, many rivers that flow directly into the sea are classified as first-order rivers despite their short lengths. This results in major rivers not being prioritized for display on maps. Therefore, we have introduced a new river classification approach. Firstly, we identified 11 primary rivers with main stems longer than 5,000 km as Level 1, which are Missouri River, Nile River, Congo River, Ob River, Yenisei River, Lena River, Heilongjiang River, Yangtze River, Huanghe River, Lancang River. Tributaries directly draining into these Level 1 rivers are designated as Level 2, those draining into Level 2 rivers as Level 3, and so forth, extending the hierarchy up to Level 9. Then we calculated average length of each level within these 11 watersheds, which was applied as the criterion of determining the river hierarchy of main stems in other basins. Subsequent tributaries are classified as similar as Gravelius classification method. The new river classification information is stored in the “Rclassdisplay” field and shown in Fig. 5.Table 2Attributes Description of GSriver.NameTypeDescriptionOBJECTintegerID Field in MDB Vector Layer FilesshapeBinaryStores geometric information of river linesRiverIDLONGUpstream HydroRIVERS reach ID (from HYDRO_ID field), serving as the unique identifier for the riverNextdownLONGCode of the receiving river (0 for level-1 rivers indicating sea/lake termination or inland rivers)MAINRIVIDLONGWatershed ID (from MAINRIV_ID field in HydroRIVERS)HydrocountLONGNumber of original HydroRIVERS reaches merged into this riverRclassINTEGERRiver classification levelRclassdisplayINTEGERPrimarily used for map display. First, rivers within the 11 major global river basins are classified by order of magnitude. Then, based on the average length of rivers at each order, the classification levels for main rivers in other basins are determined. Subsequently, tributaries within these basins are classified in sequence.RlenthDOUBLERiver length(km)HydrolenthDOUBLEOriginal HydroRIVERS reach length before fusion(km)RcatchDOUBLEDrainage area of this river (excluding tributary areas) (km²)TCatchDOUBLETotal drainage area of all tributaries(km²)OSMratioDOUBLEPercentage of vertices from OSM waterways (2 decimal places)GRITpercentDOUBLEPercentage of vertices from GRIT (2 decimal places)HydropercentDOUBLEPercentage of vertices from HydroRIVERS (2 decimal places)RnameSTRINGRiver name (Some names have been modified based on other datasets)NameListSTRINGAll associated OSM waterway namesOrignCNTSTRINGCountry of river originTopoCheckSTRINGTopological restoration has been performed at the river’s terminus, and attention should be paid to the potential issue of straightening meandering rivers.Fig. 5Classification of GSriver. Table 2Attributes Description of GSriver.NameTypeDescriptionOBJECTintegerID Field in MDB Vector Layer FilesshapeBinaryStores geometric information of river linesRiverIDLONGUpstream HydroRIVERS reach ID (from HYDRO_ID field), serving as the unique identifier for the riverNextdownLONGCode of the receiving river (0 for level-1 rivers indicating sea/lake termination or inland rivers)MAINRIVIDLONGWatershed ID (from MAINRIV_ID field in HydroRIVERS)HydrocountLONGNumber of original HydroRIVERS reaches merged into this riverRclassINTEGERRiver classification levelRclassdisplayINTEGERPrimarily used for map display. First, rivers within the 11 major global river basins are classified by order of magnitude. Then, based on the average length of rivers at each order, the classification levels for main rivers in other basins are determined. Subsequently, tributaries within these basins are classified in sequence.RlenthDOUBLERiver length(km)HydrolenthDOUBLEOriginal HydroRIVERS reach length before fusion(km)RcatchDOUBLEDrainage area of this river (excluding tributary areas) (km²)TCatchDOUBLETotal drainage area of all tributaries(km²)OSMratioDOUBLEPercentage of vertices from OSM waterways (2 decimal places)GRITpercentDOUBLEPercentage of vertices from GRIT (2 decimal places)HydropercentDOUBLEPercentage of vertices from HydroRIVERS (2 decimal places)RnameSTRINGRiver name (Some names have been modified based on other datasets)NameListSTRINGAll associated OSM waterway namesOrignCNTSTRINGCountry of river originTopoCheckSTRINGTopological restoration has been performed at the river’s terminus, and attention should be paid to the potential issue of straightening meandering rivers. Attributes Description of GSriver. Classification of GSriver. Data Overview The data fusion framework generated a global river vector dataset of GSriver composed of 748,629 rivers, which is stored in mdb format with the WGS84 (World Geodetic System 1984) coordinate system. In GSriver, 4,408,414 reaches within 52,331 watersheds with drainage areas exceeding 50 km² from HydroRIVERS were utilized, accounting for 52% of all HydroRIVERS reaches and 62% of total reach length. To OSM waterways and GRIT, 1,053,962 reaches and 1,131,328 reaches were extracted for data fusion, respectively. Statistics on river vertices show that GSriver is comprised of a total of 382,912,493 vertices, including 257,739,597 vertices (57.3%) from OSM waterways,125,172,896 vertices (27.8%) from GRIT and 67,108,236 vertices (14.9%) from original HydroRIVERS. Accordingly, 550,991 GSriver rivers are data fused geometries, representing 73.6% of all rivers in GSriver (Figs. 6 and 7). As shown in Fig. 6 and Fig. 7, regions with higher proportions of OSM waterways primarily correspond to areas with greater population density and economic development, while GRIT present a significant complement in remote regions where OSM waterways data was unavailable, sparse, or failed to meet quality thresholds.Fig. 6Proportion of OSM waterways in GSriver.Fig. 7Proportion of GRIT in GSriver. Proportion of OSM waterways in GSriver. Proportion of GRIT in GSriver. Technical Validation In general, GSriver present a high data quality. We checked the rivers positions of GSriver by overlapping them with satellite imagery, which demonstrated strong spatial agreement. Figure 8 shows an example of a region randomly selected in Mississippi River Basin, where visual inspection revealed GSriver matched well with the remote sensing image of Google Earth with spatial resolution of 1 m.Fig. 8Comparison between GSriver and high-resolution remote sensing images (Google Earth≈1 m). Comparison between GSriver and high-resolution remote sensing images (Google Earth≈1 m). To evaluate the spatial accuracy and geometric reliability of GSriver, we conducted the validation using high-precision national datasets of NHDPlus and comparisons with existing global products of MERIT, GRIT and HydroRIVERS. The independent and observed NHDPlus vector dataset of the United States, with 1:24,000-scale spatial resolution, is considered as one of the highest spatial accuracy national scale river datasets publicly available and is often used as a benchmark for testing the accuracy of river datasets. Over 760000 river segments in the Mississippi River Basin from NHDPlus were selected, which were integrated into 13048 complete rivers. In order to ensure the consistency of the spatial accuracy comparison standards, 10917 rivers from NHDPlus were applied for validation by excluding those without corresponding rivers in MERIT and GRIT. By interpolating the vertices of these rivers from GSriver, MERIT, GRIT and HydroRIVERS to a 10 meter interval. As shown in Fig. 9, GSriver achieves the lowest median error and narrowest interquartile range (IQR), indicating consistently high spatial precision. The mean distance for GSriver(68.25 m) is also significantly lower than MERIT (107.14 m), GRIT (115.10 m), and HydroRIVERS (157.79 m), reflecting reduced bias. In contrast, HydroRIVERS exhibits the highest variability and average error, confirming its limitations in flat terrain where DEM-derived flowlines often misalign. These results demonstrate that GSriver not only improves average accuracy but also reduces uncertainty in river positioning.Fig. 9Distance statistics between different river datasets and NHDPlus. Distance statistics between different river datasets and NHDPlus. We further analyzed the distance distribution falling within different distance bands from the high-precision NHDPlus dataset (Fig. 10). GSriver outperforms all three datasets in spatial alignment, with 80.1% rivers located within a distance of less than 100 m from NHDPlus, compared to 72.1%(MERIT), 74.5%(GRIT), and 11.3%(HydroRIVERS). The proportion of rivers with a distance of less than 100 m even reaches 97.7%. In contrast, HydroRIVERS exhibits significantly larger positional deviations, with over 70% rivers falling in the 100–200 m range. GSriver also shows the smallest proportion with offsets exceeding 200 m, demonstrating a substantial improvement in geometric fidelity. Moreover, we analyzed the relationship between GSriver’s spatial accuracy and the OSM fusion ratio, revealing that GSriver’s accuracy improves as the fusion percentage increases. Areas with high OSM fusion ratios are primarily located in flat, densely populated regions.Fig. 10Distribution of distances of four river datasets to NHDPlus. Distribution of distances of four river datasets to NHDPlus. Additionally, we counted the percentage of river vertices within incremental 10-meter distance bands from the NHDPlus (Fig. 11). GSriver shows a clear lead in spatial precision, with 40.3% of its vertices located within 10 meters of NHDPlus, which is substantially higher than GRIT (31.3%), MERIT (16.0%), and HydroRIVERS (24.6%). Furthermore, 69.3% of GSriver vertices fall in the range of distance less than 50 m. Conversely, other three datasets of GRIT, MERIT and HydroRIVERS showed flatter distribution, with significantly fewer vertices concentrated near NHDPlus and higher proportions beyond 50 m. These results underscore GSriver enhanced accuracy at fine spatial scales, benefiting applications such as floodplain modeling and localized hydrological analysis.Fig. 11Vertices Statistics of distances of four datasets to NHDPlus. Vertices Statistics of distances of four datasets to NHDPlus. To further validate the dataset, we conducted a nationwide assessment using river data at a scale of 1:50,000. Compared with HydroRIVERS, GSriver shows a reduction in mean deviation of 92.8 meters, indicating a substantial improvement in positional accuracy. This enhancement arises from the fusion of multiple data sources: high-spatial-resolution but topologically incomplete OpenStreetMap (OSM) waterways were integrated with HydroRIVERS, while missing river segments were supplemented using the Global River Topology (GRIT) dataset. The resulting GSriver dataset retains the topological structure of HydroRIVERS, including flow direction, confluences, and HydroIDs, while incorporating the refined geometries of OSM and GRIT. Low-relief plains are always the places with dense river-network and OSM data, where DEM-derived HydroRIVERS flowlines showed low spatial accuracy. By substituting these DEM-derived planforms with high-resolution OSM/GRIT geometries, GSriver preserves river topology while improving the spatial accuracy. Usage Notes The GSriver dataset represents a significant advancement in global hydrography by addressing the limitations of existing river datasets through multi-source vector data fusion. Unlike previous datasets that rely primarily on digital elevation models (DEMs) or satellite imagery, GSriver effectively integrates the spatial detail of OpenStreetMap (OSM) in well-mapped regions while leveraging the global continuity of GRIT and the topological consistency of HydroRIVERS. Validation against the NHDPlus dataset demonstrated that GSriver substantially reduces vertice-level positioning errors while preserving river network connectivity. The reorganization of river features into hydrologically complete river lines further improves both interpretability and standardization. Besides, the reclassification of rivers as complete hydrological units better aligns with conventional usage and traditional definitions of rivers. Furthermore, the modular data fusion framework enables continuous integration of higher-resolution spatial data, demonstrating strong scalability and innovation. Despite these improvements, several limitations should be acknowledged. The accuracy of OSM data varies widely depending on contributor activity, which can introduce inconsistencies in sparsely mapped areas. For instance, the United States contains over 6.6 million of the 27 million total OSM river reaches, substantially exceeding other countries’ coverage. In GSriver, approximately 23.6% of river reaches remain unmodified and retain their original lower spatial accuracy. Additionally, while GRIT helps supplement sparsely mapped areas, it still inherits the limitations of DEM-based approaches in flatlands and deltaic systems. Validation also revealed a few residual topological inconsistencies in plain regions such as the Huai River Basin in China. To address these limitations, each river in GSriver includes metadata fields for source and fusion ratio to aid transparency and user interpretation. To maintain consistency across the global dataset, this version of GSriver intentionally excludes high-resolution river data (e.g., NHDPlus) that are only available at the national or watershed scale. However, as such datasets become increasingly accessible, future development of GSriver may include the integration of region-specific high-resolution river vectors to further enhance detail and accuracy. Additionally, community-based validation and editing interfaces could help improve regional quality and maintain temporal relevance. Future development of GSriver may include the integration of high-resolution remote sensing imagery and the refinement of matching algorithms using machine learning. Additionally, community-based validation and editing interfaces could help improve regional quality and maintain temporal relevance. Compared to existing global river datasets, this dataset offers distinct advantages in the following scenarios: (1) In hydrological application such as flood forecasting, GSriver’s high spatial precision improves the accuracy of parameters such as river length, thereby enhancing the reliability of flood prediction in small-to-medium river basins; (2)For high-precision multi-source data analysis, such as overlaying GSriver with water bodies derived from 30-meter resolution remote sensing imagery, enabling refined water body delineation and multi-scale hydrological assessments. (3) For large-scale cartographic mapping, GSriver’s hierarchical approach and high spatial accuracy ensure effective integration with remote sensing imagery, topographic features, and urban settlement data, supporting high-quality visualization and thematic mapping across regional to global scales. Rights and permissions Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third-party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.