Satellite altimetry reveals intensifying global river water level variability Authors: Chenqi Fang, Di Long, Qi Huang, Jean-François Cretaux, Fabrice Papa, Frédéric Frappart, Huaichuan Liu, Colin J. Gleason, Chunhong Hu Published: 2025 ABSTRACT River water levels (RWLs) are fundamental to hydrology, water resource management, and disaster mitigation, yet the majority of the world’s rivers remain ungauged. Here, using 46,993 virtual stations from Sentinel-3A/B altimetry (2016‒2024), we present a global assessment of RWL variability. We find a median global fluctuation of 3.76 m, with pronounced spatial patterns: significant RWL declines across Central North/South America and Western Siberia, and increases across Africa, Oceania, Eastern/Southern Asia, and Northwestern/Central Europe. Seasonality is intensifying in 68% of basins, as high RWLs become more temporally concentrated. Maximum RWLs are declining by 0.88 cm/yr, while minimum RWLs are rising by 1.43 cm/yr. This convergence is reducing seasonal amplitude globally, with the most pronounced changes in the Americas and Central Africa. These shifts coincide with a recent surge in extreme RWL events, particularly after 2021, signaling growing hydrological instability amid concurrent droughts and floods. Our findings underscore the urgent need for adaptive water management in response to accelerating climate pressures. Introduction River water level (RWL) is a critical variable for understanding the state of surface water and its interactions with the atmosphere, groundwater, ecosystems, and human society1. Accurate monitoring of RWLs is essential for assessing freshwater availability, investigating the mechanisms of floods and droughts, and understanding regional hydrological processes2–4. Although advances in hydrodynamic modeling and in-situ sensor technologies offer complementary tools to traditional gauge networks5–7, much of the world’s river systems remain ungauged, leaving substantial gaps in our understanding of global river dynamics and their response to climate variability. Satellite altimetry has emerged as a transformative technology for hydrological monitoring, offering continuous, large-scale water level observations with high accuracy8–10. Traditionally, satellite altimetry has been focused on lakes and reservoirs due to their expansive surface areas, achieving global-scale measurements with average errors of just 0.1 to 0.2 m11–13. However, monitoring rivers, especially those with widths below 100 m, has posed significant challenges due to the assumption that altimetry performance degrades in small streams14–16. This limitation, compounded by traditional altimetry techniques and algorithms, has resulted in widespread observational gaps in river systems. Recent advancements have challenged these limitations, demonstrating that river width does not necessarily compromise altimetry performance9,17,18. The development of open-loop tracking mode, which bypasses issues caused by complex topography, has further enabled effective monitoring of upstream rivers that were previously overlooked due to tracking loss in the closed-loop mode19. Together, these advancements provide unprecedented opportunities to assess river dynamics globally. The Sentinel-3 (S3) mission, comprising the operational Sentinel-3A (S3A) and Sentinel-3B (S3B) satellites, as well as the planned Sentinel-3C and Sentinel-3D, represents a significant advancement in satellite altimetry. Equipped with Synthetic Aperture Radar (SAR), S3 provides a focused along-track beam within its footprint, offering improved resistance to environmental noise20,21 and enabling more accurate water level retrievals using advanced waveform retracking algorithms (Supplementary Fig. S1). With a dense network of virtual stations (VSs) created where the satellite’s ground tracks intersect river channels, S3 offers more extensive observations than previous missions, such as TOPEX/Poseidon or the Jason series16. Furthermore, S3’s relatively shorter revisit period (~27 days) and radar sensor capabilities enable all-weather, day-and-night data collection, providing a unique balance of spatial and temporal resolution for monitoring riverine systems. Although the latest Surface Water and Ocean Topography (SWOT) mission offers promising wide-swath observations that extend beyond nadir VSs, its limited temporal coverage (2022‒present) and the experimental nature of its novel hydrology products currently constrain in-depth river hydrology analysis compared to well-validated S3 products22,23. Despite its immense potential, the S3 mission remains underutilized in terrestrial hydrology, particularly for small and upstream rivers, leading to their underrepresentation in existing datasets24,25. In this work, we leverage eight years of S3 data (2016‒2024) to construct the most comprehensive RWL dataset across 46,993 VSs globally (25,677 VSs for S3A and 21,316 VSs for S3B). Our study employs an enhanced algorithm optimized for complex riverine environments26, enabling accurate monitoring of rivers ranging from a few meters to several kilometers in width and spanning diverse terrains, including regions over 5,000 m in elevation (Supplementary Fig. S2c, S2d). This study provides a systematic analysis of hydrological variations, including RWL fluctuations25, drying and wetting patterns1, seasonal dynamics27, and extreme events based on S3 data. The urgent need for such analysis is underscored by the increasing frequency of unprecedented droughts and floods28. We aim to bridge observational gaps and advance the understanding of global river dynamics, laying a foundation for improved hydrological research and policymaking in the face of escalating environmental challenges where our knowledge remains limited29–31. Results Global river water levels and their fluctuations Using our waveform retracking algorithm on S3 altimetry data and conducting rigorous quality assessments (see Methods), we generated RWL time series for 46,993 valid VSs across the globe. These VSs include 8520 in North America, 5771 in Europe, 1681 in Oceania, 3440 in Africa, 6368 in South America, and 21,213 in Asia. After filtering out anomalous observations, we retained approximately 1.6 million valid measurements. On average, each S3A VS contributed 37 valid observations, while each S3B VS contributed 30 valid observations, forming a robust dataset for analysis (Supplementary Fig. S2a, S2b). To quantify RWL fluctuations, we calculated the difference between the maximum and minimum observed RWLs at each VS (Fig. 1). Globally, RWLs exhibited a median fluctuation of 3.76 m, lower than the mean by 3.24 m, reflecting the skewing effect of large fluctuations observed at 4110 VSs influenced by reservoirs (average fluctuation: 10.94 m, Supplementary Fig. S3). Although reservoir-affected VSs exhibit distinct hydrological behavior, we include them in our consolidated analysis to present a comprehensive picture of global RWL dynamics. A sensitivity analysis demonstrates that the spatial patterns and key conclusions remain robust when these stations are excluded, indicating that our results reliably capture the dynamics of non-reservoir-affected rivers, primarily in response to recent climate variability on interannual to decadal scales (see Supplementary Text S1). Among continents, South America showed the highest median fluctuation, exceeding other continents by 1.44–2.80 m, largely driven by the Amazon River and its tributaries32. In contrast, Europe exhibited the lowest median fluctuation, 1.18 m below the global median, reflecting weaker seasonal and interannual variations27. However, elevated fluctuations were observed around the Mediterranean and Black Sea regions. Median fluctuations in North America, Oceania, and Africa were comparable (3.22–3.59 m), whereas Asia displayed slightly higher fluctuations, particularly in monsoon-dominated tropical and subtropical regions. While Oceania and Africa exhibited the lowest average fluctuations (<6 m), localized higher variations were observed in southeastern Australia, the Upper Guinea Highlands, and southern Africa. These findings show consistent but more extensive spatial patterns compared with Zhao et al.25, highlighting the enhanced capability of our dataset to capture global RWL variability.Fig. 1Global river water level fluctuations from 2016 to 2024.a Global overview of river water level fluctuations across 46,993 virtual stations. Regional breakdown of fluctuations for each continent: (b) North America, (c) Europe, (d) Oceania, (e) Africa, (f) South America, and (g) Asia. Color bars on the right and bottom of each subfigure indicate the range of median fluctuations within specific latitude and longitude intervals (5° bins). Histograms depict the distribution of fluctuations among virtual stations within each region. Data from both Sentinel-3A and Sentinel-3B are included. Despite Sentinel-3B’s later operational start in 2018, which results in a 0.26 m lower global median fluctuation compared to Sentinel-3A, both satellites show highly consistent spatial patterns (Supplementary Fig. S6). This consistency provides a comprehensive representation of global river water level variability. Global river water level fluctuations from 2016 to 2024. a Global overview of river water level fluctuations across 46,993 virtual stations. Regional breakdown of fluctuations for each continent: (b) North America, (c) Europe, (d) Oceania, (e) Africa, (f) South America, and (g) Asia. Color bars on the right and bottom of each subfigure indicate the range of median fluctuations within specific latitude and longitude intervals (5° bins). Histograms depict the distribution of fluctuations among virtual stations within each region. Data from both Sentinel-3A and Sentinel-3B are included. Despite Sentinel-3B’s later operational start in 2018, which results in a 0.26 m lower global median fluctuation compared to Sentinel-3A, both satellites show highly consistent spatial patterns (Supplementary Fig. S6). This consistency provides a comprehensive representation of global river water level variability. At the basin scale, we constructed monthly time series using RWL anomalies at VSs (see Methods), revealing distinctive hydrological behaviors across different flow regimes (Supplementary Fig. S4a). Basins exhibiting higher fluctuations were primarily located in Southeast Asia (e.g., Salween, Chao Phraya, Mekong, and Red River), South Asia (e.g., Pennar, Narmada, and Krishna), southwestern North America (e.g., Gulf Coast, Río Grande-Bravo, and Mexico Northwest Coast), low-latitude South America (e.g., Amazon, Orinoco, and Sao Francisco), western Africa (e.g., Volta), and southeastern Australia (e.g., Murray-Darling). Spatially, low-latitude coastal rivers exhibited more pronounced fluctuations than those at higher latitudes, largely due to stronger seasonal rainfall patterns driven by monsoons. Mid-latitude rivers in the Southern Hemisphere displayed moderate fluctuations with larger interannual variations, potentially associated with recent La Niña events, which contributed heavy precipitation during 2021‒2023 (Supplementary Fig. S5e, S5f)33. These findings highlight the increasing variability in global RWLs, particularly in climate-sensitive regions under intensifying extreme weather events29,34. Change rates and extremes of global river water levels To assess global RWL responses to recent climate variability, we calculated change rates using a robust least squares method, leveraging the extensive quantity and high quality of valid measurements at each VS (see Methods). Among the 46,993 VSs analyzed, RWL increases (51%) slightly outnumber declines (49%), with a global median change rate of 0.28 cm/yr (Supplementary Fig. S7). Rising RWLs are most prevalent in Asia, Africa, Oceania, and Northwestern/Central Europe, with Asia contributing the largest share due to its significant number of VSs experiencing increases. Notable wetting patterns emerge in Central Siberia, low-latitude regions and the southern margins of Africa, and Australia, corresponding to increased precipitation (Supplementary Fig. S5). However, when considering VSs with significant changes (p < 0.05, 11,688 VSs, 25% of total), RWL declines dominate, with a median rate of -1.63 cm/yr (Fig. 2). These declines are particularly pronounced in South and North America, where over 65% of VSs show decreasing RWLs, with median rates exceeding −12 cm/yr. Severe drying is concentrated in the Amazon Basin, where human-induced climate change primarily drives this pattern35, directly linked to significantly decreased precipitation and increased temperature (Supplementary Fig. S5a). The downstream reaches of the Mississippi River also experienced RWL drying, responding to extremely low precipitation during 2022‒202336. Despite extensive drying conditions, localized RWL wetting occurs in the Caribbean region and southeastern Brazil, influenced by intensifying extreme precipitation events (Supplementary Fig. S5)37–39. Notably, southern Brazil showed patterns consistent with trends in long-term in-situ data during 1980‒2024 (Supplementary Fig. S8), possibly reflecting a climate-warming signal37,40. In contrast, eastern Brazil exhibited discrepancies, suggesting the influence of natural climatic variability. Beyond South America, moderate drying conditions were also observed in the Eastern European Plain, Western Siberia, and the eastern parts of Central Asia.Fig. 2Significant changes in global river water levels (2016–2024).a Global map showing significant river water level change rates (p < 0.05), with red and blue points indicating declines and increases, respectively. These represent 11,688 virtual stations out of the total of 46,993 analyzed. Red rectangles in (a) highlight areas with concentrated declines. Regional views for each continent: b North America, c Europe, d Oceania, e Africa, f South America, and g Asia. Virtual stations from both Sentinel-3A and Sentinel-3B are included, as two satellites show highly consistent spatial patterns (Supplementary Fig. S9). The format follows Fig. 1. Significant changes in global river water levels (2016–2024). a Global map showing significant river water level change rates (p < 0.05), with red and blue points indicating declines and increases, respectively. These represent 11,688 virtual stations out of the total of 46,993 analyzed. Red rectangles in (a) highlight areas with concentrated declines. Regional views for each continent: b North America, c Europe, d Oceania, e Africa, f South America, and g Asia. Virtual stations from both Sentinel-3A and Sentinel-3B are included, as two satellites show highly consistent spatial patterns (Supplementary Fig. S9). The format follows Fig. 1. At the basin scale, extensive river drying is evident in Central South America, Central North America, the East European Plain, and Western Siberia during 2016–2024 (Supplementary Fig. 4b). RWL declines also affect basins across Madagascar, northern regions of the Southern Africa Plateau, and areas surrounding the Mediterranean and Black Seas, coinciding with intensifying heatwaves since 202141,42 and consistent with reduced precipitation and higher temperatures (Supplementary Fig. S5). These findings highlight the contrast between broad wetting patterns and more intense drying, underscoring heterogeneous hydrological responses and increasing variability in RWLs amid intensifying extreme weather. This pattern is further supported by substantial inconsistencies with long-term in-situ streamflow records, particularly in eastern North America and Europe, suggesting that RWLs are experiencing increasingly frequent anomalies relative to historical baselines (Supplementary Fig. S8). RWL changes have intensified in recent years, particularly during the operational period of S3B (Supplementary Fig. S9). North and South America exhibit accelerated declines (Supplementary Fig. S10a, S10e), with rates in the Amazon and Great Plains exceeding the eight-year average by 4.91–12.53 cm/yr (Supplementary Fig. S11a, S11b). Conversely, Asia and Europe have seen a pronounced increase in RWLs (Supplementary Fig. S10b, S10f). These shifts are closely associated with the growing frequency of extreme RWL events, primarily driven by intensifying climatic variability29. To quantify these events, we identified years with the highest frequency of extreme stages (termed “extreme stage years”) at the basin scale (see Methods). Results reveal that extreme stage years predominantly occur after 2021, accounting for over 47% and 64% of global basins in the annual and wet season analyses, respectively (Fig. 3). Drying basins correspond to recent low stage years (e.g., 2023‒2024 in Central South America and Central North America), while wetting basins align with recent high stage years, particularly post-2021 in Europe, Central Siberia, Africa, and Oceania. Notably, Europe experienced a concentration of high stage years in 2024, associated with extreme spring rainfall43, temporally reversing prior drying patterns (Supplementary Fig. S8, S10b).Fig. 3Extreme stage years across global basins.a Spatial distribution of basins experiencing low (a-1) and high (a-2) stage years based on annual analysis. Extreme stage years represent the year with the highest frequency of extreme low or high water levels. Darker red (a-1) and blue (a-2) colors indicate more recent extreme low and high stage years. The inset histogram shows the percentage of basins with extreme stage years ranging from 2017 to 2024. Panel (a-3) illustrates the temporal lags between low and high stage years identified in (a-1) and (a-2). Red shading indicates basins where the low stage year occurred after the high stage year, while blue shading indicates the opposite. The inset histogram shows the basin count for each lag ranging from −7 to 7 years. b Spatial distribution of basins with extreme low (b-1) and high (b-2) stage years based on wet season analysis, focusing on the year with the highest frequency of extreme water levels during wet seasons. The plot scheme and color definitions follow a. Further details on the definitions of low and high stage years are provided in the Methods section and Supplementary Fig. S13. Extreme stage years across global basins. a Spatial distribution of basins experiencing low (a-1) and high (a-2) stage years based on annual analysis. Extreme stage years represent the year with the highest frequency of extreme low or high water levels. Darker red (a-1) and blue (a-2) colors indicate more recent extreme low and high stage years. The inset histogram shows the percentage of basins with extreme stage years ranging from 2017 to 2024. Panel (a-3) illustrates the temporal lags between low and high stage years identified in (a-1) and (a-2). Red shading indicates basins where the low stage year occurred after the high stage year, while blue shading indicates the opposite. The inset histogram shows the basin count for each lag ranging from −7 to 7 years. b Spatial distribution of basins with extreme low (b-1) and high (b-2) stage years based on wet season analysis, focusing on the year with the highest frequency of extreme water levels during wet seasons. The plot scheme and color definitions follow a. Further details on the definitions of low and high stage years are provided in the Methods section and Supplementary Fig. S13. Global drought conditions peaked in 2023, as indicated by a record frequency of low stage years during the wet season (Fig. 3b-1) and a near-record frequency in the annual analysis (Fig. 3a-1). These droughts significantly reduced peak RWLs, even in continents with overall increasing RWLs (Supplementary Fig. S10). During the Northern Hemisphere summer of 2023 (July‒October), extreme heatwaves41 led to lower RWLs across all six continents compared to average RWLs for the same period in other years, with a median difference of 8.49 cm (Supplementary Fig. S12). The Mississippi and Amazon basins experienced unprecedentedly low RWLs (Supplementary Fig. S11), corroborating reports from the World Meteorological Organization, which identified 2023 as the driest year in the past 33 years44,45. In 2024, the frequency of high stage years reached a global peak, reflecting a growing hydrological instability characterized by concurrent extreme floods and droughts. While global drought conditions showed partial alleviation in 2024, they worsened in the Amazon basin, where RWLs hit new record lows46. These findings underscore that concurrent drying and wetting signals are driving broad RWL anomalies and accelerating variability. Variations in seasonality and seasonal fluctuations Seasonal variations in RWLs are critical for evaluating the impacts of climate change and short-term climate variability on water security and freshwater biodiversity47–49. To quantify these variations, we first calculated apportionment entropy (AE)27 of altimetric RWLs at the basin scale, where higher AE values indicate evenly distributed RWLs throughout the year, and lower values reflect concentrated RWLs during specific periods, leading to pronounced month-to-month variations (see Methods). Basins with distinct wet-dry seasonal cycles, such as those in eastern South America, northern Africa, the Indian Peninsula, the Indochina Peninsula, and Central Siberia, exhibit strong seasonality (Fig. 4a). In contrast, Europe and Western Siberia show weaker seasonality, likely linked to snowmelt contributions during dry seasons, which moderate seasonal stage changes (Supplementary Fig. S10b, S11c)27. From 2016 to 2024, AE values decreased in 68% of global basins, indicating intensified seasonality (Fig. 4b). This shift is especially pronounced in northwestern North America, eastern South America, eastern and southern Asia, and Oceania. Conversely, weakening seasonality is observed in mid- to high-latitude regions, such as the Great Lakes, the East European Plain, and Central Siberia. In tropical regions like the Amazon and Malay Archipelago, reduced seasonality is marked by flattened peak stages induced by extreme droughts (Supplementary Fig. S11a), posing serious threats to diverse rainforest ecosystems.Fig. 4Changes in river water level seasonality and seasonal fluctuations.a Median apportionment entropy values across major hydrological basins, where green areas indicate basins with strong seasonality. b Change rates of apportionment entropy values across basins, with green indicating increased seasonality and brown indicating decreased seasonality. c Change rates of seasonal maximum stages. d Change rates of seasonal minimum stages. e Change rates of seasonal fluctuations, where red and blue shades in (c‒e) denote decreases and increases, respectively. All seasonal metrics are derived from basin-level river stage time series. f Schematic diagram illustrating changes in seasonal river water levels from 2016 to 2024. The inset pie chart categorizes basins into six groups based on changes in seasonal minimum stage (L) and seasonal maximum stage (H): L + H+ (both L and H increase), L-H- (both L and H decrease), L + H- (L increases while H decreases), and L-H+ (L decreases while H increases). Subdivisions of L + H+ and L-H- reflect whether changes in maximum or minimum stages dominate, leading to either increased or decreased seasonal fluctuations. Red and blue colors denote decreases and increases in seasonal fluctuations, respectively. Changes in river water level seasonality and seasonal fluctuations. a Median apportionment entropy values across major hydrological basins, where green areas indicate basins with strong seasonality. b Change rates of apportionment entropy values across basins, with green indicating increased seasonality and brown indicating decreased seasonality. c Change rates of seasonal maximum stages. d Change rates of seasonal minimum stages. e Change rates of seasonal fluctuations, where red and blue shades in (c‒e) denote decreases and increases, respectively. All seasonal metrics are derived from basin-level river stage time series. f Schematic diagram illustrating changes in seasonal river water levels from 2016 to 2024. The inset pie chart categorizes basins into six groups based on changes in seasonal minimum stage (L) and seasonal maximum stage (H): L + H+ (both L and H increase), L-H- (both L and H decrease), L + H- (L increases while H decreases), and L-H+ (L decreases while H increases). Subdivisions of L + H+ and L-H- reflect whether changes in maximum or minimum stages dominate, leading to either increased or decreased seasonal fluctuations. Red and blue colors denote decreases and increases in seasonal fluctuations, respectively. Seasonal fluctuations, defined as the amplitude between seasonal maximum and minimum RWLs within a year, have declined in 68% of global basins (Fig. 4e). This decline is most pronounced in the Americas, Central Africa, Central Europe, Western Siberia, northern China, and Southeast Asia. To identify the drivers of these changes, basins were classified into six categories based on the contributions of seasonal maximum and minimum stages to fluctuations. The dominant pattern (30% of global basins) is a decrease in seasonal maximum stages combined with an increase in seasonal minimum stages, followed by a reduction in both seasonal maximum and minimum stages (22%). Together, these two patterns explain 76% of the observed decreases in seasonal fluctuations. In contrast, increases in seasonal fluctuations are primarily driven by rising seasonal maximum stages and are spatially scattered. Notably, seasonal maximum RWLs are declining at a median rate of 0.88 cm/yr while minimum RWLs show an increase of 1.43 cm/yr, with declines in maximum RWLs being more widespread and significant than those in seasonal minimum stages under the influence of frequent droughts (Fig. 4b, c). The spatial alignment of seasonal RWL changes with overall RWL changes, corresponding to recent precipitation patterns, highlights key river dynamics under intensifying climate variability. Overall, high seasonal RWLs are becoming more concentrated within the year, leading to more pronounced wet-dry contrasts. Simultaneously, the amplitude of seasonal maximum RWLs has diminished, suggesting a shift toward hydrological regimes with reduced seasonal fluctuations (Fig. 4f). Discussion This study presents a comprehensive global-scale assessment of RWLs across 46,993 VSs using Sentinel-3 altimetry data, providing new and fine insights into river dynamics during 2016 − 2024. Our results highlight substantial spatial variability in RWLs, with median fluctuations ranging from 2.58 to 5.38 m across six continents. High fluctuations are concentrated in low- to mid- latitude coastal regions, corresponding to monsoonal patterns and intensifying climate extremes30,33,41. These findings align with previous studies identifying high hydrological variability in the Amazon, Southeast Asia, the Upper Guinea Highlands, and the Nile, based on altimetry and/or in-situ data (Supplementary Fig. S8a)25,50. However, our dataset significantly expands previous work by incorporating an order of magnitude more VSs, extending coverage to medium-to-small rivers (median width: 107 m) and high-altitude regions (>5000 m) (Supplementary Fig. S2c, S2d), thus offering the most detailed global assessment to date. Our analysis reveals distinct wetting-drying patterns, with significant RWL declines observed in Central North America, Central South America, and Western Siberia, while Africa, Oceania, and parts of Asia and Europe have experienced rising RWLs. These patterns correlate strongly with the increasing frequency of extreme hydrological events, particularly in 2022–2024, as confirmed by global water reports44,51. Regionally, the Yangtze and Pearl rivers faced severe floods, while the Amazon and Zambezi rivers experienced record-breaking droughts, highlighting dual drought-flood stressors. Additionally, our results indicate that upstream rivers tend to experience increases in RWL, while downstream sections show declines, patterns that closely align with those reported in a previous study52 and supported by our analysis (Supplementary Fig. S14). This consistency, together with extensive validation of our global RWL dataset against over 10,000 gauges, underscores the robustness of our methodology (see Supplementary Text S2, S3). Furthermore, we find that 68% of global basins exhibit intensified seasonality and reduced seasonal fluctuations. These changes are primarily driven by a concentration of high RWLs within wet seasons and suppressed seasonal peak stages. This shift suggests that extreme droughts and rising temperatures51 disproportionately impact seasonal maximum stages. We further observe a tendency for reservoirs to retain more water in recent years, potentially reflecting a regional adaptation to frequent climate extremes. This is evidenced by lower RWL change rates when excluding reservoir-affected VSs, particularly across Europe, South America, and Asia (Supplementary Fig. S15, S16). Moreover, recent RWL signals in some regions deviate largely from long-term trends since 1980 (e.g., eastern North America, eastern Brazil, and Europe), suggesting that short-term climate variability, superimposed on ongoing climate change and human activities, is compounding hydrological pressures (Supplementary Fig. S8). Such growing hydrological variability poses mounting risks to aquatic ecosystems, fish migration, agricultural irrigation, and urban water security53,54, underscoring the urgency of adaptive water management strategies. While previous studies have examined large-scale hydrological variations using in-situ observations, their findings may not fully capture global patterns, particularly outside North America and Europe27,55. As a result, important knowledge gaps remain, including broad wetting in Asia, significant drying in Western Siberia, and heterogeneous seasonal variability, largely due to the scarcity of gauges. Our results fill these gaps, providing a substantially more comprehensive assessment of river drying in 2023 than is currently documented in global and local water reports (Supplementary Fig. S12). This highlights the role of satellite altimetry as not only a powerful complement to ground data but also, in many data-scarce regions, as an indispensable tool for advancing hydrological understanding. Looking ahead, advancements in satellite altimetry and remote sensing offer opportunities to establish denser and more continuous global river monitoring networks in near real-time. Although this study establishes an extensive VS network, current low-resolution mode and SAR altimetry observations remain limited to discrete nadir points, constraining spatial continuity. The SWOT mission provides a transformative opportunity to overcome these limitations by delivering wide-swath measurements of water surfaces. Integrating SWOT and Sentinel-3 data with hydrological models and AI-driven forecasting systems56–59 could enable a new generation of global river monitoring platforms. These advances would not only enhance water resource management but also strengthen disaster preparedness and climate adaptation strategies. Future research should further examine how anthropogenic drivers, including dam operations and land-use change, interact with climate-driven hydrological shifts, thereby advancing a more comprehensive understanding of river dynamics under global change. Methods The data processing workflow in this study comprises three main components: (1) generation of global virtual stations (VSs), (2) retrieval of river water levels (RWLs) from Sentinel-3 altimetry data, and (3) postprocessing analysis of RWL fluctuations, change rates, extreme low and high stages, and seasonal variations. Each of these components is detailed below. Data The data used in this study consist of Sentinel-3 altimetry, the Surface Water and Ocean Topography Mission River Database (SWORD), and several ancillary datasets. Sentinel-3 altimetry data and SWORD form the backbone of the global RWL dataset, while the ancillary datasets provide crucial support for postprocessing analyses. Sentinel-3 altimetry Sentinel-3 data were acquired using the Synthetic Aperture Radar Altimeter (SRAL), which has a temporal resolution of 27 days and an equatorial cross-track resolution of 104 km. SRAL captures reflected echoes within beams with an along-track size of 300 m, effectively reducing environmental noise interference (Supplementary Fig. S1). The European Space Agency (ESA) provides Sentinel-3 data in three processing levels: Level-0 (raw telemetered data), Level-1 (Level-0 data corrected for instrumental effects), and Level-2 (Level-1 data further corrected for geophysical effects). For RWL retrieval, we utilized Level-2 Non-Time Critical Hydrology Thematic Products, which offer precise orbital data and comprehensive corrections. These products are available through the Copernicus Data Space Ecosystem (last accessed: November 23, 2024). Data from September 2016 onwards for S3A and December 2018 onwards for S3B were processed using our improved algorithm, which incorporates radar waveforms, geolocations, acquisition times, and geophysical and propagation corrections to obtain accurate RWLs (see below). Surface Water and Ocean Topography Mission River Database (SWORD)60 The Surface Water and Ocean Topography (SWOT) Mission River database (SWORD) serves as a global reference for river networks, supporting the SWOT mission. It encompasses 213 thousand reaches and 10.7 million nodes across global rivers, integrating hydrological observations for each SWOT revisit cycle. SWORD was developed by combining several global hydrography datasets, including the Global River Widths from Landsat (GRWL)61, MERIT Hydro62, and HydroBASINS63. SWORD version 15 was utilized in this study, with its nodes within the S3 orbital range used to generate VSs along global rivers (detailed below). Each SWORD node includes prior attributes, such as river width, elevation, and their associated uncertainties, which form the basis for filtering RWL anomalies during subsequent retrieval processes. Additionally, the unique identifier (ID) in SWORD enables precise alignment of our VSs with the corresponding nodes in SWORD. Ancillary data (1) HydroBASINS63. This dataset provides global-scale, consistently sized, and hierarchically nested sub-basin boundaries for hydrological analyses. We used the major hydrological basins to aggregate the results of VSs, reflecting the hydrological clustering of RWL variations across basins. (2) In-situ RWLs and discharge. Daily gauge measurements from the United States Geological Survey (USGS), the Global Runoff Data Center (GRDC), and the Information Center of the Ministry of Water Resources of the People’s Republic of China (2016‒2024) were jointly used to evaluate RWLs obtained from S3 altimetry data and hydrological patterns derived from this study. Discharge records since 1980 were employed to extract long-term hydrological signals under climate change. (3) HydroRIVERS63. This dataset, which provides global river vectors, was used to align gauging stations with VSs, facilitating the validation of our global RWL dataset, and to assign river orders (ORD_FLOW) to VSs, enabling investigation of RWL change rates by river size. (4) GeoDAR64. This global dataset includes 24,783 dams and 21,515 reservoirs. It was used to identify VSs influenced by reservoirs. (5) Global Precipitation Climatology Project (GPCP)65. This dataset provides monthly rainfall at 2.5° × 2.5° resolution (1979‒present), integrating observations from gauges, satellites, and soundings. We calculated average precipitation from 2010–2023 for each grid cell to establish a baseline for identifying wet months. This baseline was then used to set thresholds for waveform retracking and to filter valid measurements during outlier removal. Additionally, wet months were critical for the extreme year analysis. Precipitation during our study period was used to interpret RWL changes. (6) GISS Surface Temperature Analysis (GISTEMP)66. This dataset provides global surface temperature anomalies at 2° × 2° resolution (1980–present). Monthly data during 2016–2024 were used to interpret RWL changes in relation to temperature variability. (7) Connectivity Status Index54. This dataset quantifies the degree of human intervention across global rivers within the HydroRIVERS framework and was used to validate the dominance of climate drivers on our observed RWL patterns. Generation of global virtual stations VSs are the primary measurement targets for satellite altimeters, serving as the foundation for constructing time series of RWLs. In this study, we established VSs by creating a 2-km-wide buffer around the standard tracks of S3 (accessible at https://www.aviso.altimetry.fr/en/missions/current-missions.html). Nodes from SWORD that fell within this buffer were identified as potential VS candidates. With an approximate 200 m spacing between adjacent nodes, this process initially yielded multiple neighboring candidates within the same river reach (typically about 10 km in length). To minimize overlap between VSs within the same reach and to avoid redundancy in river stage data, we imposed a minimum interval of nine nodes (~2 km) between adjacent VSs. For instance, if candidate IDs in a reach ranged from 1 to 35 (downstream to upstream), we selected candidates with IDs 10, 20, and 30 as VSs. For reaches with fewer than 20 candidates, we selected the candidate corresponding to the median ID as the VS. This procedure is illustrated in Supplementary Fig. S17. The selected VSs inherently retained the basic attributes of SWORD nodes, such as river widths and elevations, streamlining their integration into the analysis. This strategy eliminated the need to manually merge these attributes from other hydrography datasets and facilitated linkage with SWOT observations via unique IDs. Compared to previous VS creation methods based on static water masks11,25, our approach is more efficient and adaptable. In total, 81,071 VSs were generated. Subsequently, we created 1-km-radius buffers around these VSs to identify satellite footprints containing water surface information. Establishment of global river water level dataset We employed the Improved Multiple Subwaveform Analysis (IMSA) algorithm26 to retrieve RWLs from S3 altimetry data and to construct time series at global VSs. The IMSA algorithm has been demonstrated to perform effectively in complex riverine environments. Its core functionalities involve waveform retracking and outlier removal, with minor adjustments made to enhance its applicability to global rivers, which are detailed below. This algorithm was implemented for each VS independently. Water level retrieval from Sentinel-3 radar waveforms To retrieve RWLs from Sentinel-3 radar waveforms, we first filtered the satellite footprints located within the buffers of global VSs and extracted waveforms corresponding to these footprints. Suitability for over-water waveforms was determined using Pulse Peakiness (PP) and maximum power. PP was calculated using Eq. (1):1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{PP}}}}=\frac{31.5{P}_{\max }}{{\sum }_{i=5}^{n-5}{P}_{i}}$$\end{document}PP=31.5Pmax∑i=5n−5Piwhere n represents the total number of gates in the range window, and Pi denotes the waveform power of the ith gate. The first four and last four gates, typically indicative of environmental noise, were excluded from the total power calculation8. Once suitable over-water waveforms were identified, waveform retracking was applied. This process sequentially extracted subwaveforms and used the leading-edge slope and the peak power to remove noise-induced subwaveforms. Reflection gates were identified based on the number of detected subwaveforms. For waveforms with a single subwaveform, reflection gates were adaptively positioned using seasonally transformed threshold approaches. Note that the wet season for each grid cell was defined as the five consecutive months with the highest precipitation, based on GPCP data. For waveforms with multiple subwaveforms, the algorithm initially identified the subwaveform group most likely to contain the target river’s signal, before finely locating gates through further exploration within this group. This strategy effectively mitigated interference from environmental noise, thereby contributing significantly to accurate river surface locations across diverse riverine environments. The retracked range (Rretracked) was calculated using Eq. (2), and the water surface elevation (WSE) was determined after applying geophysical and propagation corrections, including pole tides, solid Earth tides, ionospheric delays, and both dry and wet tropospheric delays, as shown in Eq. (3)67.2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{{{\rm{retracked}}}}}={R}_{{{{\rm{unretracked}}}}}+({G}_{{{{\rm{r}}}}}-{G}_{{{{\rm{nominal}}}}})\times {R}_{0}$$\end{document}Rretracked=Runretracked+(Gr−Gnominal)×R03\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{WSE}}}}={H}_{{{{\rm{sat}}}}}-{R}_{{{{\rm{retracked}}}}}-\left({{{\rm{dry}}}}+{{{\rm{wet}}}}+{{{\rm{iono}}}}+{{{\rm{set}}}}+{{{\rm{pt}}}}\right)-{H}_{{{{\rm{geoid}}}}}$$\end{document}WSE=Hsat−Rretracked−dry+wet+iono+set+pt−Hgeoidwhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{{{\rm{retracked}}}}}$$\end{document}Rretracked and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{{{\rm{unretracked}}}}}$$\end{document}Runretracked are the retracked and unretracked ranges, respectively; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{0}$$\end{document}R0 is the width of one gate; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${G}_{{{{\rm{r}}}}}$$\end{document}Gr is the retracked gate, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${G}_{{{{\rm{nominal}}}}}$$\end{document}Gnominal is the nominal gate; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H}_{{{{\rm{sat}}}}}$$\end{document}Hsat is the satellite height relative to the ellipsoid; WSE represents the river water surface elevation relative to the geoid (EGM2008); dry, wet, iono, set, and pt are corrections for dry and wet tropospheric delays, ionospheric delays, solid Earth tides, and pole tides, respectively; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H}_{{{{\rm{geoid}}}}}$$\end{document}Hgeoid is the geoid height above the ellipsoid. Water level outlier removal The algorithm employs a comprehensive outlier removal process to ensure the reliability of RWL data while preserving both long-term and short-term variability. Initially, a large sliding window spanning 10 cycles (~9 months) is used to compute the average water surface elevation (WSE1) from the 15th and 85th percentiles of stage measurements. Measurements near the center of this window that fall below \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{WSE}}}}}_{1}-0.2\times ({{{{\rm{WSE}}}}}_{{{{\rm{wet}}}}}-{{{{\rm{WSE}}}}}_{{{{\rm{dry}}}}})$$\end{document}WSE1−0.2×(WSEwet−WSEdry) during the wet season or exceed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{WSE}}}}}_{1}+0.2\times ({{{{\rm{WSE}}}}}_{{{{\rm{wet}}}}}-{{{{\rm{WSE}}}}}_{{{{\rm{dry}}}}})$$\end{document}WSE1+0.2×(WSEwet−WSEdry) during the dry season are flagged as outliers and removed. Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{WSE}}}}}_{{{{\rm{wet}}}}}$$\end{document}WSEwet and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{WSE}}}}}_{{{{\rm{dry}}}}}$$\end{document}WSEdry represent the 85th and 15th percentile RWLs during the wet and dry seasons, respectively. To refine the analysis and capture short-term RWL variability, a smaller sliding window spanning two cycles (~2 months) is applied. With this window, the median stage (WSE2) and median absolute deviation (MAD) are calculated to provide a finer resolution of RWLs. Measurements are retained as valid data points for each cycle if they satisfy multiple criteria: being within \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.1\times ({{{{\rm{WSE}}}}}_{{{{\rm{wet}}}}}-{{{{\rm{WSE}}}}}_{{{{\rm{dry}}}}})$$\end{document}1.1×(WSEwet−WSEdry) of the prior surface elevation, within \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.75\times ({{{{\rm{WSE}}}}}_{{{{\rm{wet}}}}}-{{{{\rm{WSE}}}}}_{{{{\rm{dry}}}}})$$\end{document}0.75×(WSEwet−WSEdry) of WSE1, and within five times the MAD of WSE2. Additionally, data points with a standard deviation below 2 m are considered valid. This dual-window approach ensures that short-term anomalies are effectively filtered out while preserving long-term changes. To address hooking effects68‒where RWLs near the riverbank appear lower than those at the river center‒a parabolic fit is applied to valid along-track measurements (Eq. 4). This adjustment is referenced against the median stage (Eq. 5)16. Consecutive corrected median RWLs from these adjustments are compiled to establish RWL time series at each VS, ensuring a robust representation of RWL dynamics across both short and long time scales.4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat{{{{\rm{WSE}}}}}=a{s}_{0}^{2}+b{s}_{0}+c$$\end{document}WSE^=as02+bs0+c5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{WSE}}}}}_{{{{\rm{corr}}}}}=\overline{\overline{{{{\rm{WSE}}}}}}+({{{\rm{WSE}}}}-\widehat{{{{\rm{WSE}}}}})$$\end{document}WSEcorr=WSE¯¯+(WSE−WSE^)where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat{{{{\rm{WSE}}}}}$$\end{document}WSE^ is the fitted stage; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{WSE}}}}}_{{{{\rm{corr}}}}}$$\end{document}WSEcorr is the corrected stage; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{\overline{{{{\rm{WSE}}}}}}$$\end{document}WSE¯¯ is the median stage; s0 is the serial number of valid measurements along the track; and a, b, and c are parameters fitted using the least square method. Dataset analysis With the global RWL dataset established, we conducted a comprehensive analysis focusing on RWL fluctuations, change rates, extreme stage years, and seasonal dynamics, as detailed below. Analyses of RWL fluctuations and change rates were performed at both VS and basin scales. In contrast, extreme stage years and seasonal dynamics were examined only at the basin scale due to the large uncertainties arising from temporally uneven altimetric measurements at the VS scale. To ensure the robustness of our findings, VSs with fewer than 10 measurements or monitoring periods shorter than four years were excluded, resulting in a refined dataset of 46,993 VSs. At the basin scale, RWLs at individual VSs were normalized to their average level to facilitate integration of RWL variations across different locations69. A monthly sliding window approach was used to capture the median of all valid stage anomalies, providing representative monthly RWLs for each basin. This method enabled the construction of basin-level RWL time series, with examples shown in Supplementary Fig. S11. Basin-level analyses were conducted exclusively on these time series, ensuring reliable insights into basin-scale dynamics while minimizing the influence of individual VS anomalies or outliers. Note that VSs exhibiting fluctuations greater than 25 m (i.e., ~4% of global VSs, typically reflecting strong human regulation or erroneous observations) were considered invalid and excluded from the basin-level RWL construction. Basins were included in the analysis only if they contained at least 35 valid VSs, and monthly stages were recorded only when the sliding window incorporated at least three measurements. Ultimately, 137 basins worldwide met these criteria and were retained for analysis. RWL fluctuation RWL fluctuation at each VS was quantified as the difference between the maximum and minimum RWLs, reflecting the total stage amplitude over the monitoring period. Seasonal fluctuations, representing intra-annual variability, were calculated as the amplitude between the seasonal maximum and minimum stages (i.e., the maximum and minimum RWLs recorded within a year). At the basin scale, overall and seasonal fluctuations were derived from the basin-level RWL time series, enabling a broader perspective on variability across entire river basins. RWL change rate RWL change rate captures multi-year variability in river stages over the monitoring period, indicating both the direction and magnitude of changes. A robust linear regression model70 was used to estimate the annual rate of change (slope) for all valid VSs. This approach minimizes the influence of outliers by iteratively adjusting regression parameters based on a weighted loss function. The statistical significance of the change was evaluated using the corresponding p-value (two-tailed test). For the regression, Tukey’s loss function71 was applied to manage residuals:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{a}(u)=\left\{\begin{array}{cc}{u}^{2}/2-{u}^{4}/\left(4{a}^{2}\right),& {{\mbox{if}}}{|u|}\le a\\ {a}^{2}/4,\hfill & \quad{\mbox{otherwise}}\end{array}\right.$$\end{document}ρa(u)=u2/2−u4/4a2,if∣u∣≤aa2/4,otherwisewhere u denotes the residual, and a is a tuning constant, typically set to 4.685. This regression model was also applied to basin-level RWL time series to derive change rates at the basin scale, providing a comprehensive assessment of RWL dynamics globally. Low/High stage year detection To evaluate the impact of extreme climatic conditions on river systems, we identified years with the highest frequency of extreme RWLs. Two types of analyses were conducted: (1) Annual analysis: Extreme low and high stages were determined based on the 25th and 75th percentile RWLs, calculated from basin-level time series. The proportions of low and high stages were computed for each year from 2017 to 2024. The years with the highest proportions of low or high stages were designated as extreme low and high stage years, respectively; and (2) Wet season analysis: During wet seasons, the 75th percentile RWL was used as a threshold to define extreme high stages. The year with the highest proportion of high stages during the wet season was identified as the wet season’s extreme high stage year, while the year with the lowest proportion was identified as the extreme low stage year, indicating significant drought conditions. These thresholds were validated through a sensitivity test to ensure robust representation of global extremes (see Supplementary Text S4). The year 2016 was excluded due to its short observation period (less than four months). Years with at least seven monthly stage observations were included in the annual analysis, while the wet season analysis required at least three stage observations during the wet season. Extreme stage years were determined for basins with at least six valid years of data, resulting in 134 and 133 basins included in the annual and wet season analyses, respectively. This detection process is illustrated in Supplementary Fig. S13. RWL seasonality and variation To assess RWL seasonality, apportionment entropy (AE) was employed to analyze the distribution of RWLs across months27. Lower AE values indicate stronger seasonality, where high RWLs are concentrated in specific months. However, due to missing monthly RWL data in some years, adjustments were made to ensure comparability. AE was normalized by dividing it by its theoretical maximum value, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\log }_{2}(n)$$\end{document}log2(n), scaling it between 0 and 1:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X={\sum }_{i=1}^{n}{x}_{i}$$\end{document}X=∑i=1nxi8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{AE}}}}=-\left(\mathop{\sum }_{i=1}^{n}\frac{{x}_{i}}{X}{\log }_{2}\left(\frac{{x}_{i}}{X}\right)\right)\div{\log }_{2}(n)\times 100\%$$\end{document}AE=−∑i=1nxiXlog2xiX÷log2(n)×100%where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{i}$$\end{document}xi represents the adjusted stage for the ith measurement, X is the sum of adjusted stages, and n is the number of measurements within a year. Stages were referenced to their respective minimum RWLs to better capture variations in strong and weak seasonality. AE values, as well as other seasonal metrics (seasonal minimum and maximum stages, and seasonal fluctuations), were calculated only for years with at least eight valid monthly RWLs. The median AE across valid years was used to represent basin-level seasonality. Change rates for these four seasonal metrics were derived using the robust linear regression method described earlier. This calculation was implemented for all 137 basins. Limitations and opportunities While our dataset achieves satisfactory global accuracy of approximately 0.6 m (see Supplementary Text S2), which adequately supports in-depth hydrological analysis, several limitations underscore areas for future improvement. Data quality at some VSs is affected by environmental and topographic conditions. For instance, in urban areas, reflective surfaces such as bright roofs and other infrastructure can contaminate waveforms, reducing measurement accuracy. Similarly, rivers with complex underlying features, such as sandbars, braided channels, or highly dynamic riverbeds, pose challenges for accurate retrieval18. In mountainous regions, the accuracy of altimetric measurements declines due to satellite orbital drift, which introduces elevation discrepancies between altimetric footprints26. This issue is particularly evident in steep-sloped rivers or those that run parallel to the satellite’s ground track72. Frozen rivers present additional challenges, as snow and ice interfere with waveform retracking algorithms, further complicating the retrieval process73. Fortunately, technological advancements such as advanced SAR modes and Open-Loop Tracking Command techniques19,74,75 (accessible at https://www.altimetry-hydro.eu/oltc) offer promising solutions by enhancing the accuracy of water surface elevations for most rivers. The development of more sophisticated algorithms tailored to these specific scenarios could further improve data robustness and quality, addressing many of the current limitations and reducing the sensitivity of hydrological signals to measurement noise. The relatively low temporal resolution of S3 data, with its 27-day revisit period, remains another significant constraint. This limitation restricts our ability to fully capture river dynamics76, thereby leading to underestimation of RWL fluctuations and overestimation of RWL change rates (see Supplementary Text S3). To mitigate its interference with our findings, we integrated VS stages within basins and constructed monthly continuous basin-level RWL time series. Patterns observed at this balanced scale can better represent common hydrological behaviors for most rivers, particularly their seasonal responses. To capture river dynamics in a more spatiotemporally fine-grained and precise manner, observation densification by fusing multi-source data becomes essential, which helps reduce uncertainties caused by data gaps. Emerging satellite missions, particularly the Surface Water and Ocean Topography (SWOT) mission, are poised to significantly enhance the spatial and temporal resolution of river observations. The integration of satellite observations with hydrological and data-driven models offers a powerful pathway to refine our understanding of river dynamics77, particularly the underlying mechanisms and runoff processes under recent climate extremes. The interplay between natural hydrological processes, climate variability, and anthropogenic influences is inherently complex54,78. While we have examined the impact of reservoirs, one of the major anthropogenic factors, the full extent of other human activities on fluvial dynamics, and the combined effects of river-related factors remain unknown and require further exploration. This knowledge is critical for developing effective strategies for global water resources management to address the growing frequency and severity of extreme weather events. Supplementary information Supplementary Information Transparent Peer Review file Supplementary Information Transparent Peer Review file Source data Source Data Source Data