Yuan Yang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Yuan Yang has a total of 129 co-authors. Here are the top co-authors by number of shared publications:
- Ming Pan
Center for Western Weather and Water Extremes Scripps Institution of Oceanography University of California San Diego La Jolla CA USA
10 shared publications - Yang Hong
The University of Oklahoma, National Weather Center ARRC 4610, 120 David L. Boren Blvd., Norman, United States [US
10 shared publications - Chaopeng Shen
Civil and Environmental Engineering, The Pennsylvania State University, University Park, USA
7 shared publications - Hylke E. Beck
Climate and Livability Initiative, King Abdullah University of Science and Technology, Thuwal, Saudi Arabia
6 shared publications - Guangyuan Kan
State Key Laboratory of Water Cycle and Water Security, Technology Innovation Center for Flood and Drought Prevention and Disaster Reduction of the Ministry of Water Resources, China Institute of Water Resources and Hydropower Research, Beijing 100038, China
4 shared publications - Jiangtao Liu
Civil and Environmental Engineering The Pennsylvania State University University Park PA USA
4 shared publications
Latest Publications
- Open Access
- 24 August 2026
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 53 (16)
SWOT, Empiricism, and River Modeling
Given the insurmountable challenge of measuring all rivers in situ, global river models serve as the foundation of freshwater knowledge past, present, and future. We adopt an inductive empirical framework based on Surface Water and Ocean Topography (SWOT) satellite measurements to assess these models. After controlling for SWOT data quality, we examine 68,347 individual river reaches representing ∼38% of global discharge. We find river models currently struggle in areas of heavy economic development, multi‐channel rivers, arid areas, and many Arctic rivers. After controlling for these expected errors, we find better skill as rivers get wider and that parts of Siberia and China are particularly difficult to model. We also find large variability and spatial heterogeneity to model performance, resisting oversimplification. Our results suggest that leveraging SWOT observations within river models will improve them, but river models must adapt their structure to represent realistic hydraulics to do so.
- 24 August 2026
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 53 (16)
Open AccessSWOT, Empiricism, and River Modeling
Given the insurmountable challenge of measuring all rivers in situ, global river models serve as the foundation of freshwater knowledge past, present, and future. We adopt an inductive empirical framework based on Surface Water and Ocean Topography (SWOT) satellite measurements to assess these models. After controlling for SWOT data quality, we examine 68,347 individual river reaches representing ∼38% of global discharge. We find river models currently struggle in areas of heavy economic development, multi‐channel rivers, arid areas, and many Arctic rivers. After controlling for these expected errors, we find better skill as rivers get wider and that parts of Siberia and China are particularly difficult to model. We also find large variability and spatial heterogeneity to model performance, resisting oversimplification. Our results suggest that leveraging SWOT observations within river models will improve them, but river models must adapt their structure to represent realistic hydraulics to do so.
- Open Access
- 12 August 2026
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 53 (16)
Geospatial Foundation Embeddings as Transferable Catchment Descriptors
Deep learning hydrological models rely on static catchment attributes to characterize basin heterogeneity, but these traditional descriptors suffer from regional inconsistency and limited temporal representativeness. This study evaluates geospatial foundation model embeddings (from AlphaEarth and StefaLand) as transferable catchment descriptors for rainfall‐runoff modeling. Temporal validation across 531 CAMELS basins showed that deep learning models trained on AlphaEarth embeddings alone achieved comparable performance to those trained on traditional attributes (median Nash‐Sutcliffe Efficiency (NSE) 0.779 vs. 0.781). In a global prediction in ungauged basins (PUB) scenario with 3,434 basins, combining embeddings with traditional attributes improved median NSE from 0.683 to 0.720. A dual‐path gated model revealed structured spatiotemporal reliance patterns: models preferentially utilized embeddings during snowmelt transitions and in humid basins, with limited reliance in arid catchments. Independent validation with StefaLand embeddings from a land‐surface foundation model yielded consistent temporal validation results, supporting the broader use of learned embeddings as transferable catchment descriptors.
- 12 August 2026
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 53 (16)
Open AccessGeospatial Foundation Embeddings as Transferable Catchment Descriptors
Deep learning hydrological models rely on static catchment attributes to characterize basin heterogeneity, but these traditional descriptors suffer from regional inconsistency and limited temporal representativeness. This study evaluates geospatial foundation model embeddings (from AlphaEarth and StefaLand) as transferable catchment descriptors for rainfall‐runoff modeling. Temporal validation across 531 CAMELS basins showed that deep learning models trained on AlphaEarth embeddings alone achieved comparable performance to those trained on traditional attributes (median Nash‐Sutcliffe Efficiency (NSE) 0.779 vs. 0.781). In a global prediction in ungauged basins (PUB) scenario with 3,434 basins, combining embeddings with traditional attributes improved median NSE from 0.683 to 0.720. A dual‐path gated model revealed structured spatiotemporal reliance patterns: models preferentially utilized embeddings during snowmelt transitions and in humid basins, with limited reliance in arid catchments. Independent validation with StefaLand embeddings from a land‐surface foundation model yielded consistent temporal validation results, supporting the broader use of learned embeddings as transferable catchment descriptors.
- Open Access
- 9 July 2026
- Published by Elsevier in Journal of Hydrology: Regional Studies
- Vol. 66
Parsimonious climatic and physiographic controls explain the spatial variability of the Budyko parameter
- 9 July 2026
- Published by Elsevier in Journal of Hydrology: Regional Studies
- Vol. 66
Open AccessParsimonious climatic and physiographic controls explain the spatial variability of the Budyko parameter
- Open Access
- 6 February 2026
- Published by Wiley
High-throughput AI Video Surveys Enable Reproducible Multiscale Sediment Size Mapping, with Implications for Hydrobiogeochemical Parameterization
Sediment size regulates flow resistance, sediment transport, stream–groundwater exchange and microbial respiration, yet its distributions remain poorly mapped at scales relevant to hydrobiogeochemical (HBGC) processes. Existing methods are labor-intensive, site-limited and difficult to reproduce, leaving the role of sediment heterogeneity in HBGC dynamics poorly constrained. Here we present a computer-vision artificial intelligence (AI) framework that uses drone and smartphone videos to map centimeter-resolution grain size percentiles with automated uncertainty quantification. Three models, i.e., Scale AI, Anthro AI and Grain AI, are developed to estimate pixel size and uncertainty, remove anthropogenic objects, and measure individual grains, respectively. Working together, they achieved 3–4% error and 4.9% reproducibility error in median grain size mapping and sampling efficiencies up to 173 square-meters per second, nearly five orders of magnitude faster than traditional surveys. Incorporating the AI-derived heterogeneous grain-size fields into established HBGC relationships shows that accounting for centimeter-scale heterogeneity changes HBGC metrics by 15-52% in terms of mean absolute relative errors and up to 60-220% relative errors in extreme conditions, when compared to estimates using reach-averaged grain size, with the largest effects at fine scales and in non-perennial streams. Our results reveal a spatiotemporal scale- and flow intermittency-dependent role of sediment heterogeneity in HBGC function and provide a high-throughput approach to map it across river networks in a reproducible and quality-controlled manner.
- 6 February 2026
- Published by Wiley
Open AccessHigh-throughput AI Video Surveys Enable Reproducible Multiscale Sediment Size Mapping, with Implications for Hydrobiogeochemical Parameterization
Sediment size regulates flow resistance, sediment transport, stream–groundwater exchange and microbial respiration, yet its distributions remain poorly mapped at scales relevant to hydrobiogeochemical (HBGC) processes. Existing methods are labor-intensive, site-limited and difficult to reproduce, leaving the role of sediment heterogeneity in HBGC dynamics poorly constrained. Here we present a computer-vision artificial intelligence (AI) framework that uses drone and smartphone videos to map centimeter-resolution grain size percentiles with automated uncertainty quantification. Three models, i.e., Scale AI, Anthro AI and Grain AI, are developed to estimate pixel size and uncertainty, remove anthropogenic objects, and measure individual grains, respectively. Working together, they achieved 3–4% error and 4.9% reproducibility error in median grain size mapping and sampling efficiencies up to 173 square-meters per second, nearly five orders of magnitude faster than traditional surveys. Incorporating the AI-derived heterogeneous grain-size fields into established HBGC relationships shows that accounting for centimeter-scale heterogeneity changes HBGC metrics by 15-52% in terms of mean absolute relative errors and up to 60-220% relative errors in extreme conditions, when compared to estimates using reach-averaged grain size, with the largest effects at fine scales and in non-perennial streams. Our results reveal a spatiotemporal scale- and flow intermittency-dependent role of sediment heterogeneity in HBGC function and provide a high-throughput approach to map it across river networks in a reproducible and quality-controlled manner.
- 4 December 2025
- Published by American Geophysical Union (AGU) in Global Biogeochemical Cycles
- Vol. 39 (12)
Underestimated Carbon Losses as Dissolved Organic Carbon in Forested Headwater Streams
Large amounts of dissolved organic carbon (DOC) are transported laterally from uplands and wetlands to headwater streams; however, the global magnitude of this flux and its role in the carbon (C) budget remain unclear. By compiling 20,403 DOC concentration observations and applying machine learning models, we estimated the annual DOC flux from forested headwater streams north of 30°S to be 116.2 Tg C yr −1 (90% CI: 86.5–145.2). Higher yields were observed in tropical regions (3.8 g C m −2 yr −1 ), where vegetation productivity is high, and in boreal regions (3.1 g C m −2 yr −1 ), where peatland cover and soil organic C stocks are substantial. When comparing our results with global terrestrial C flux estimates upscaled from eddy covariance data, we found that the fraction of net ecosystem production (NEP) lost annually as DOC ranged from negligible (<0.1%) to 20.7%, with discharge explaining 34% of the variation. Our study suggests that neglecting the lateral export of DOC could lead to an overestimation of NEP in forested headwater catchments, a bias that is further amplified by increased discharge.
- 4 December 2025
- Published by American Geophysical Union (AGU) in Global Biogeochemical Cycles
- Vol. 39 (12)
Underestimated Carbon Losses as Dissolved Organic Carbon in Forested Headwater Streams
Large amounts of dissolved organic carbon (DOC) are transported laterally from uplands and wetlands to headwater streams; however, the global magnitude of this flux and its role in the carbon (C) budget remain unclear. By compiling 20,403 DOC concentration observations and applying machine learning models, we estimated the annual DOC flux from forested headwater streams north of 30°S to be 116.2 Tg C yr −1 (90% CI: 86.5–145.2). Higher yields were observed in tropical regions (3.8 g C m −2 yr −1 ), where vegetation productivity is high, and in boreal regions (3.1 g C m −2 yr −1 ), where peatland cover and soil organic C stocks are substantial. When comparing our results with global terrestrial C flux estimates upscaled from eddy covariance data, we found that the fraction of net ecosystem production (NEP) lost annually as DOC ranged from negligible (<0.1%) to 20.7%, with discharge explaining 34% of the variation. Our study suggests that neglecting the lateral export of DOC could lead to an overestimation of NEP in forested headwater catchments, a bias that is further amplified by increased discharge.
- Top Cited PapersOpen Access
- 16 June 2025
- Published by American Geophysical Union (AGU) in Water Resources Research
- Vol. 61 (6)
Global Daily Discharge Estimation Based on Grid Long Short‐Term Memory (LSTM) Model and River Routing
To expand the spatial coverage of the conventional Basin Long Short‐Term Memory (LSTM) model for river discharge estimation beyond pre‐selected individual locations, we developed a discharge modeling scheme, Grid LSTM‐RAPID, to estimate discharge for every river reach worldwide. Grid LSTM‐RAPID applies LSTM runoff estimation to the grids (0.25°), small rectangular hydrological response units (HRUs) rather than basins (irregularly shaped HRUs of any size), and then routes the grid runoff over all reaches on a global river network using the RAPID routing model. It largely maintains the strong performance of Basin LSTM over gauged basins and achieves a median Kling‐Gupta Efficiency (KGE) of 0.653 for small basins out‐of‐sample both temporally and spatially (0.688 for out‐of‐sample temporally), and a median KGE of 0.592 for other basins with larger areas and less data quality. Compared to Basin LSTM, Grid LSTM‐RAPID loses about 0.03 in median KGE for basins out‐of‐sample in both time and space in exchange for global all‐reach coverage without heavy cost. Despite this tradeoff, it significantly outperforms a well‐calibrated process‐based benchmark model. Using the new scheme, we create an improved global reach‐level daily discharge data set from 1980 to near present named GRADES‐hydroDL, which is openly shared at https://www.reachhydro.org/home/records/grades‐hydrodl.
- 16 June 2025
- Published by American Geophysical Union (AGU) in Water Resources Research
- Vol. 61 (6)
Top Cited PapersOpen AccessGlobal Daily Discharge Estimation Based on Grid Long Short‐Term Memory (LSTM) Model and River Routing
To expand the spatial coverage of the conventional Basin Long Short‐Term Memory (LSTM) model for river discharge estimation beyond pre‐selected individual locations, we developed a discharge modeling scheme, Grid LSTM‐RAPID, to estimate discharge for every river reach worldwide. Grid LSTM‐RAPID applies LSTM runoff estimation to the grids (0.25°), small rectangular hydrological response units (HRUs) rather than basins (irregularly shaped HRUs of any size), and then routes the grid runoff over all reaches on a global river network using the RAPID routing model. It largely maintains the strong performance of Basin LSTM over gauged basins and achieves a median Kling‐Gupta Efficiency (KGE) of 0.653 for small basins out‐of‐sample both temporally and spatially (0.688 for out‐of‐sample temporally), and a median KGE of 0.592 for other basins with larger areas and less data quality. Compared to Basin LSTM, Grid LSTM‐RAPID loses about 0.03 in median KGE for basins out‐of‐sample in both time and space in exchange for global all‐reach coverage without heavy cost. Despite this tradeoff, it significantly outperforms a well‐calibrated process‐based benchmark model. Using the new scheme, we create an improved global reach‐level daily discharge data set from 1980 to near present named GRADES‐hydroDL, which is openly shared at https://www.reachhydro.org/home/records/grades‐hydrodl.
- Top Cited PapersOpen Access
- 10 April 2025
- Published by American Geophysical Union (AGU) in Water Resources Research
- Vol. 61 (4)
High‐Resolution National‐Scale Water Modeling Is Enhanced by Multiscale Differentiable Physics‐Informed Machine Learning
The National Water Model (NWM) is a key tool for flood forecasting, planning, and water management. Key challenges facing the NWM include calibration and parameter regionalization when confronted with big data. We present two novel versions of high‐resolution (∼37 km2) differentiable models (a type of hybrid model): one with implicit, unit‐hydrograph‐style routing and another with explicit Muskingum‐Cunge routing in the river network. The former predicts streamflow at basin outlets whereas the latter presents a discretized product that seamlessly covers rivers in the conterminous United States (CONUS). Both versions use neural networks to provide a multiscale parameterization and process‐based equations to provide a structural backbone, which were trained simultaneously (“end‐to‐end”) on 2,807 basins across the CONUS and evaluated on 4,997 basins. Both versions show great potential to elevate future NWM performance for extensively calibrated as well as ungauged sites: the median daily Nash‐Sutcliffe efficiency of all 4,997 basins is improved to around 0.68 from 0.48 of NWM3.0. As they resolve spatial heterogeneity, both versions greatly improved simulations in the western CONUS and also in the Prairie Pothole Region, a long‐standing modeling challenge. The Muskingum‐Cunge version further improved performance for basins >10,000 km2. Overall, our results show how neural‐network‐based parameterizations can improve NWM performance for providing operational flood predictions while maintaining interpretability and multivariate outputs. The modeling system supports the Basic Model Interface (BMI), which allows seamless integration with the next‐generation NWM. We also provide a CONUS‐scale hydrologic data set for further evaluation and use.
- 10 April 2025
- Published by American Geophysical Union (AGU) in Water Resources Research
- Vol. 61 (4)
Top Cited PapersOpen AccessHigh‐Resolution National‐Scale Water Modeling Is Enhanced by Multiscale Differentiable Physics‐Informed Machine Learning
The National Water Model (NWM) is a key tool for flood forecasting, planning, and water management. Key challenges facing the NWM include calibration and parameter regionalization when confronted with big data. We present two novel versions of high‐resolution (∼37 km2) differentiable models (a type of hybrid model): one with implicit, unit‐hydrograph‐style routing and another with explicit Muskingum‐Cunge routing in the river network. The former predicts streamflow at basin outlets whereas the latter presents a discretized product that seamlessly covers rivers in the conterminous United States (CONUS). Both versions use neural networks to provide a multiscale parameterization and process‐based equations to provide a structural backbone, which were trained simultaneously (“end‐to‐end”) on 2,807 basins across the CONUS and evaluated on 4,997 basins. Both versions show great potential to elevate future NWM performance for extensively calibrated as well as ungauged sites: the median daily Nash‐Sutcliffe efficiency of all 4,997 basins is improved to around 0.68 from 0.48 of NWM3.0. As they resolve spatial heterogeneity, both versions greatly improved simulations in the western CONUS and also in the Prairie Pothole Region, a long‐standing modeling challenge. The Muskingum‐Cunge version further improved performance for basins >10,000 km2. Overall, our results show how neural‐network‐based parameterizations can improve NWM performance for providing operational flood predictions while maintaining interpretability and multivariate outputs. The modeling system supports the Basic Model Interface (BMI), which allows seamless integration with the next‐generation NWM. We also provide a CONUS‐scale hydrologic data set for further evaluation and use.
- Open Access
- 26 September 2024
- Published by Wiley
High-resolution national-scale water modeling is enhanced by multiscale differentiable physics-informed machine learning
The National Water Model (NWM) is a key tool for flood forecasting and planning and water management. Key challenges facing NWM include calibration and parameter regionalization when confronted with big data. We present two novel versions of high-resolution (~37 km2) differentiable models (a type of physics-informed machine learning): one with implicit, unit-hydrograph-style routing and another with explicit Muskingum-Cunge routing in the river network. The former predicts streamflow at basin outlets whereas the latter presents a discretized product that seamlessly covers rivers in the conterminous United States (CONUS). Both versions used neural networks to provide multiscale parameterization and process-based equations to provide structural backbone, trained them together (“end-to-end”) on 2,807 basins across CONUS, and evaluated them on 4,997 basins. Both versions show the great potential to elevate future NWMs for extensively calibrated as well as ungauged sites: the median daily Nash-Sutcliffe efficiency (NSE) of all 4,997 basins is improved to around 0.68 from 0.49 of NWM3.0. As they resolve heterogeneity, both greatly improved simulations in the western CONUS and also in the Prairie Pothole Region, a long-standing modeling challenge. The Muskingum-Cunge version further improved performance for basins >10000 km2. Overall, our results show how neural-network-based parameterizations can improve NWM performance for providing operational flood predictions while maintaining interpretability and multivariate outputs. We provide a CONUS-scale hydrologic dataset for further evaluation and use. The modeling system supports the Basic Model Interface (BMI), which allows seamless integration with the next-generation NWM.
- 26 September 2024
- Published by Wiley
Open AccessHigh-resolution national-scale water modeling is enhanced by multiscale differentiable physics-informed machine learning
The National Water Model (NWM) is a key tool for flood forecasting and planning and water management. Key challenges facing NWM include calibration and parameter regionalization when confronted with big data. We present two novel versions of high-resolution (~37 km2) differentiable models (a type of physics-informed machine learning): one with implicit, unit-hydrograph-style routing and another with explicit Muskingum-Cunge routing in the river network. The former predicts streamflow at basin outlets whereas the latter presents a discretized product that seamlessly covers rivers in the conterminous United States (CONUS). Both versions used neural networks to provide multiscale parameterization and process-based equations to provide structural backbone, trained them together (“end-to-end”) on 2,807 basins across CONUS, and evaluated them on 4,997 basins. Both versions show the great potential to elevate future NWMs for extensively calibrated as well as ungauged sites: the median daily Nash-Sutcliffe efficiency (NSE) of all 4,997 basins is improved to around 0.68 from 0.49 of NWM3.0. As they resolve heterogeneity, both greatly improved simulations in the western CONUS and also in the Prairie Pothole Region, a long-standing modeling challenge. The Muskingum-Cunge version further improved performance for basins >10000 km2. Overall, our results show how neural-network-based parameterizations can improve NWM performance for providing operational flood predictions while maintaining interpretability and multivariate outputs. We provide a CONUS-scale hydrologic dataset for further evaluation and use. The modeling system supports the Basic Model Interface (BMI), which allows seamless integration with the next-generation NWM.
- Open Access
- 6 July 2024
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 51 (13)
Contribution of Mesoscale Convective Systems to Floods in the East Asian Summer Monsoon Region
The quantitative relationship between Mesoscale Convective Systems (MCSs) and floods over East Asia has not been established. In this study, MCSs are clustered into four types with Self‐Organizing Map approach. Floods in June‐August of 2000–2021 are linked with different types of MCS by automated algorithms we constructed. We find that among the major floods (potential flood peak periods), 91% (87%) are related to MCS, 65% (78%) are dominated by MCS, and 38% (20%) are dominated by multi‐types of MCS. Types 1 and 2 MCS have higher flood‐inducing efficiencies than common MCS (Type‐4). Type‐1 MCS, characterized by the least number (2% of the total number), the largest precipitation volume, longest lifetime, slowest moving, strongest precipitation, can most efficiently produce floods. Type‐2 MCS, characterized by the second largest precipitation volume, more numerous than Type‐1 particularly over land, can induce floods not only relatively efficiently but also more frequently than Type‐1.
- 6 July 2024
- Published by American Geophysical Union (AGU) in Geophysical Research Letters
- Vol. 51 (13)
Open AccessContribution of Mesoscale Convective Systems to Floods in the East Asian Summer Monsoon Region
The quantitative relationship between Mesoscale Convective Systems (MCSs) and floods over East Asia has not been established. In this study, MCSs are clustered into four types with Self‐Organizing Map approach. Floods in June‐August of 2000–2021 are linked with different types of MCS by automated algorithms we constructed. We find that among the major floods (potential flood peak periods), 91% (87%) are related to MCS, 65% (78%) are dominated by MCS, and 38% (20%) are dominated by multi‐types of MCS. Types 1 and 2 MCS have higher flood‐inducing efficiencies than common MCS (Type‐4). Type‐1 MCS, characterized by the least number (2% of the total number), the largest precipitation volume, longest lifetime, slowest moving, strongest precipitation, can most efficiently produce floods. Type‐2 MCS, characterized by the second largest precipitation volume, more numerous than Type‐1 particularly over land, can induce floods not only relatively efficiently but also more frequently than Type‐1.
- Open Access
- 14 October 2023
- Published by Wiley
Global Daily Discharge Estimation Based on Grid-Scale Long Short-Term Memory (LSTM) Model and River Routing
Accurate global river discharge estimation is crucial for advancing our scientific understanding of the global water cycle and supporting various downstream applications. In recent years, data-driven machine learning models, particularly the Long Short-Term Memory (LSTM) model, have shown significant promise in estimating discharge. Despite this, the applicability of LSTM models for global river discharge estimation remains largely unexplored. In this study, we diverge from the conventional basin-lumped LSTM modeling in limited basins. For the first time, we apply an LSTM on a global 0.25° grid, coupling it with a river routing model to estimate river discharge for every river reach worldwide. We rigorously evaluate the performance over 5332 evaluation gauges globally for the period 2000-2020, separate from the training basins and period. The grid-scale LSTM model effectively captures the rainfall-runoff behavior, reproducing global river discharge with high accuracy and achieving a median Kling-Gupta Efficiency (KGE) of 0.563. It outperforms an extensively bias-corrected and calibrated benchmark simulation based on the Variable Infiltration Capacity (VIC) model, which achieved a median KGE of 0.466. Using the global grid-scale LSTM model, we develop an improved global reach-level daily discharge dataset spanning 1980 to 2020, named GRADES-hydroDL. This dataset is anticipated to be useful for a myriad of applications, including providing prior information for the Surface Water and Ocean Topography (SWOT) satellite mission. The dataset is openly available via Globus.
- 14 October 2023
- Published by Wiley
Open AccessGlobal Daily Discharge Estimation Based on Grid-Scale Long Short-Term Memory (LSTM) Model and River Routing
Accurate global river discharge estimation is crucial for advancing our scientific understanding of the global water cycle and supporting various downstream applications. In recent years, data-driven machine learning models, particularly the Long Short-Term Memory (LSTM) model, have shown significant promise in estimating discharge. Despite this, the applicability of LSTM models for global river discharge estimation remains largely unexplored. In this study, we diverge from the conventional basin-lumped LSTM modeling in limited basins. For the first time, we apply an LSTM on a global 0.25° grid, coupling it with a river routing model to estimate river discharge for every river reach worldwide. We rigorously evaluate the performance over 5332 evaluation gauges globally for the period 2000-2020, separate from the training basins and period. The grid-scale LSTM model effectively captures the rainfall-runoff behavior, reproducing global river discharge with high accuracy and achieving a median Kling-Gupta Efficiency (KGE) of 0.563. It outperforms an extensively bias-corrected and calibrated benchmark simulation based on the Variable Infiltration Capacity (VIC) model, which achieved a median KGE of 0.466. Using the global grid-scale LSTM model, we develop an improved global reach-level daily discharge dataset spanning 1980 to 2020, named GRADES-hydroDL. This dataset is anticipated to be useful for a myriad of applications, including providing prior information for the Surface Water and Ocean Topography (SWOT) satellite mission. The dataset is openly available via Globus.