Streamflow Estimates Research Articles

HyG: A hydraulic geometry dataset derived from historical stream gage measurements across the conterminous US

Regional- and continental-scale hydrologic models are increasingly important forecasting tools, yet they rely on highly variable channel parameters (e.g. width, depth, hydraulic resistance) that remain unquantified for millions of stream reaches across the country. Existing hydrologic models utilize over-simplified channel geometries that directly impact the accuracy of streamflow estimates, with cascading effects for water resource and hazard prediction. Here, we present a hydraulic geometry dataset, termed ‘HyG’, derived from discharge field measurements at U.S. Geological Survey (USGS) stream gages across the conterminous United States (CONUS). The HyG dataset includes (1) at-a-station hydraulic geometry parameters, (2) at-a-station hydraulic resistance (Manning’s n) calculated from the Manning equation, (3) daily discharge percentiles, and (4) regionalized downstream hydraulic geometry parameters based on HUC4 (Hydrologic Unit Code 4), derived from a total of 4,051,682 individual measurements from 66,841 total gages. The regionalized HyG dataset can be used directly to improve channel representations in models over CONUS. The original HyG relationships can also be regionalized for finer scales if required.

Streamflow estimation using SWAT model including large storage reservoirs in Godavari River Basin, India

Streamflow estimation using SWAT model including large storage reservoirs in Godavari River Basin, India

Comparison and integration of physical and interpretable AI-driven models for rainfall-runoff simulation

Precise streamflow forecasting in river systems is crucial for water resources management and flood risk assessment. The Tagus Headwaters River Basin (THRB) in Spain is a key hydrological hub, providing regulated flow for agricultural, urban, and energy sectors, and facilitating water transfer to the Segura River Basin, a key semi-arid region. Given the basin's near-total allocation of water resources, accurate streamflow simulations are essential to optimize the socio-economic distribution and ensure sustainable management across both interconnected basins. This study conducts a comprehensive evaluation of the Soil and Water Assessment Tool (SWAT+), support vector regression (SVR), feed forward neural network (FFNN), and long short-term memory (LSTM) models in simulating the rainfall-runoff process at four gauging stations within the THRB. An ensemble machine learning technique is then applied to assess improvements in streamflow estimation. Results revealed that AI-based models significantly surpassed the SWAT+ model in performance. Furthermore, the application of an ensemble technique enhanced the precision of rainfall-runoff modeling by 18 to 26% during the calibration period and 4.1 to 9.2% during the validation period, compared to individual AI-based models. Additionally, the SWAT+ model's precision improved by 44 to 74% and 40 to 55% for the respective periods. The use of Shapley Additive Explanations (SHAP) methodology allowed the results of the ensemble with machine learning to be more interpretable by explaining how each model contributes to the prediction. This research offers significant contributions to hydrological modeling, highlighting the importance of ensemble techniques in elevating predictive accuracy for various river basins.

Using national hydrologic models to obtain regional climate change impacts on streamflow basins with unrepresented processes

Climate change is increasingly impacting water availability. National-scale hydrologic models simulate streamflow resulting from many important processes, but often without processes such as human water use and management activities. This work explores and tests methods to account for such omitted processes using one national-scale hydrologic model. Two bias correction methods, Flow Duration Curve (FDC) and Auto-Regressive Integrated Moving Average (ARIMA), are tested on streamflow simulated by the US Geological Survey National Hydrologic Model (NHM-PRMS), which omits irrigation pumping. A semi-arid agricultural case study is used. FDC and ARIMA perform better for correcting low and high flows, respectively. A hybrid method performs well at both low and high flows; typical Nash-Sutcliffe values increased from <-1.00 to about 0.75. Results suggest methods with which national-scale hydrologic models can be bias-corrected for omitted processes to improve regional streamflow estimates. Utility of these correction methods in simulation of future projections is discussed.

Multi-mission virtual monitoring station for streamflow monitoring and hydrodynamic model calibration

A comprehensive understanding of the streamflow dynamic along the river system is one of the significant components for preserving biodiversity and ensuring sustainable ecological processes. Understanding this requirement, streamflow estimation using remote sensing (RS) has evolved as an alternate approach in the hydrologic literature during the last decade due to its extensive spatiotemporal coverage. However, the existing RS-based approaches have limited applications for deriving the continuous time series in tropical river reaches due to narrow water widths during the low flow and frequent cloud covers during the high flow periods. Therefore, this study proposed frameworks to establish a virtual monitoring station (VMS) where multi-mission satellites are being used to derive continuous streamflow time series. From the multi-mission satellites, the optical RS images are used to develop a Copula-based fusion (CFUS) model by integrating the Frank copula with the 30m × 1-day resolution synthetic Landsat images, derived from the fusion of 250m × 1-day resolution MODIS images and 30m × 16-days resolution Landsat images; whereas the water levels are retrieved from the altimeters to derive discharge using the rating curve. Additionally, the potential utility of the established VMS for hydrodynamic model (MIKE11-NAM-HD) calibration under data-scarce conditions is also demonstrated. Finally, a coupled RS-hydrodynamic (RS-HD) framework is also proposed to establish VMS both at semi-gauged and ungauged sections of the rivers. The efficacy of the aforementioned framework was tested along the lower Brahmani river reach. The results reveal that the advocated framework could perform satisfactorily with the Nash-Sutcliffe efficiency (NSE) and Kling-Gupta efficiency (KGE) ≥0.8. This approach has the potential to be upscaled to other river reaches as one of the next-generation hydrometry for daily streamflow monitoring using high-frequent observation of multi-mission satellites.

Pywr-DRB: An open-source Python model for water availability and drought risk assessment in the Delaware River Basin

The Delaware River Basin (DRB) in the Mid-Atlantic region of the United States is an institutionally complex water resources system that provides drinking water for 13.5 million people, plus water for energy, industry, recreation, and ecosystems. This paper introduces Pywr-DRB, an open-source Python model exploring the impacts of reservoir operations, transbasin diversions, and minimum flow targets on water availability and drought risk in the DRB. Pywr-DRB draws on streamflow estimates from emerging data resources, bridging advances in large-scale hydrologic modeling with an improved representation of the basin's evolving water infrastructure and management institutions. Our detailed model diagnostic assessment demonstrates that Pywr-DRB provides substantial improvements over sole use of hydrologic models in capturing the DRB's dynamics. We also explore how water management alters model-derived risk estimates for low flows and water demand shortfalls. Our approach to diagnostic benchmarking and water systems modeling is broadly applicable to other major basins.

Hydrological modelling using SWAT for the assessment of streamflow dynamics in the Ganga River basin.

Growing concerns over water availability arise from the problems of population growth, rapid industrialization, and human interferences, necessitating accurate streamflow estimation at the river basin scale. It is extremely challenging to access stream flow data of a transboundary river at a spatio-temporal scale due to data unavailability caused by water conflicts for assessing the water availability.Primarily, this estimation is done using rainfall-runoff models. The present study addresses this challenge by applying the soil and water assessment tool (SWAT) for hydrological modelling, utilizing high-resolution geospatial inputs. Hydrological modelling using remote sensing and GIS (Geographic Information System) through this model is initiated to assess the water availability in the Ganga River basin at different locations. The outputs are calibrated and validated using the observed station data from Global Runoff Data Centre (GRDC). To check the performance of the model, Nash-Sutcliffe efficiency (NSE), percent bias (PBIAS), coefficient of determination (R2), and RSR efficacy measures are initiated in ten stations using the observed and simulated stream flow data. The R2 values of eight stations range from 0.82 to 0.93, reflecting the efficacy of the model in rainfall-runoff modelling. Moreover, the results obtained from this hydrological modelling can serve as valuable resources for water resource planners and geographers for future reference.

A national-scale hybrid model for enhanced streamflow estimation – consolidating a physically based hydrological model with long short-term memory (LSTM) networks

Abstract. Accurate streamflow estimation is essential for effective water resource management and adapting to extreme events in the face of changing climate conditions. Hydrological models have been the conventional approach for streamflow interpolation and extrapolation in time and space for the past few decades. However, their large-scale applications have encountered challenges, including issues related to efficiency, complex parameterization, and constrained performance. Deep learning methods, such as long short-term memory (LSTM) networks, have emerged as a promising and efficient approach for large-scale streamflow estimation. In this study, we have conducted a series of experiments to identify optimal hybrid modeling schemes to consolidate physically based models with LSTM aimed at enhancing streamflow estimation in Denmark. The results show that the hybrid modeling schemes outperformed the Danish National Water Resources Model (DKM) in both gauged and ungauged basins. While the standalone LSTM rainfall–runoff model outperformed DKM in many basins, it faced challenges when predicting the streamflow in groundwater-dependent catchments. A serial hybrid modeling scheme (LSTM-q), which used DKM outputs and climate forcings as dynamic inputs for LSTM training, demonstrated higher performance. LSTM-q improved the mean Nash–Sutcliffe efficiency (NSE) by 0.22 in gauged basins and 0.12 in ungauged basins compared to DKM. Similar accuracy improvements were achieved with alternative hybrid schemes, i.e., by predicting the residuals between DKM-simulated streamflow and observations using LSTM. Moreover, the developed hybrid models enhanced the accuracy of extreme events, which encourages the integration of hybrid models within an operational forecasting framework. This study highlights the advantages of synergizing existing physically based hydrological models (PBMs) with LSTM models, and the proposed hybrid schemes hold the potential to achieve high-quality large-scale streamflow estimations.

The influence of human activities on streamflow reductions during the megadrought in central Chile

Abstract. Since 2010, central Chile has experienced a protracted megadrought with annual precipitation deficits ranging from 25 % to 70 %. An intensification of drought propagation has been attributed to the effect of cumulative precipitation deficits linked to catchment memory. Yet, the influence of water extractions on drought intensification is still unclear. Our study assesses climate and water use effects on streamflow reductions during a high-human-influence period (1988–2020) in four major agricultural basins. We performed this attribution by contrasting observed streamflow (driven by climate and water use) with near-natural streamflow simulations (driven mainly by climate) representing what would have occurred without water extractions. Near-natural streamflow estimations were obtained from rainfall–runoff models trained over a reference period with low human intervention (1960–1988). Annual and seasonal streamflow reductions were examined before and after the megadrought onset, and hydrological drought events were characterized for the complete evaluation period in terms of their frequency, duration, and intensity. Our results show that before the megadrought onset (1988–2009) the mean annual deficits in observed streamflow ranged between 2 % and 20 % across the study basins and that 81 % to 100 % of those deficits were explained by water extractions. During the megadrought (2010–2020), the mean annual deficits in observed streamflow were 47 % to 76 % among the basins. During this time, the relative contribution of precipitation deficits on streamflow reduction increased while the contribution of water extractions decreased, accounting for 27 % to 51 % of the streamflow reduction. Regarding drought events during the complete evaluation period, we show that human activities have amplified drought propagation, with almost double the intensity of hydrological droughts in some basins compared to those expected by precipitation deficits only. We conclude that while the primary cause of streamflow reductions during the megadrought has been the lack of precipitation, water uses have not diminished during this time, causing an exacerbation of the hydrological drought conditions and aggravating their impacts on water accessibility in rural communities and natural ecosystems.

Evaluation of the support vector regression (SVR) and the random forest (RF) models accuracy for streamflow prediction under a data-scarce basin in Morocco

Streamflow prediction is a key variable for water resources management. It becomes more important in semi-arid regions such as the Tensift river basin in Morocco, where water resources are facing a severe drought and the demand is continuously increasing. The present analysis focuses on evaluating Machine Learning techniques, namely support vector regression (SVR) and Random Forest (RF) against the multiple linear regression (MLR) for daily streamflow forecasting in the mountainous sub-basin of Rheraya between 2003 and 2016. The results show that SVR performed best, followed by RF and MLR. In measurable terms and regarding mean performance, SVR exhibited the higher Nash–Sutcliffe efficiency score (NSE = 0.59) and a lower root mean squared error (RMSE = 1.18 m3s-1\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ ext {m}^3\\,\ ext {s}^{-1}$$\\end{document}) compared to RF (NSE = 0.53, RMSE = 1.18 m3s-1\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ ext {m}^3\\,\ ext {s}^{-1}$$\\end{document}) and MLR (NSE = 0.54, RMSE = 1.01 m3s-1\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\ ext {m}^3\\,\ ext {s}^{-1}$$\\end{document}). Furthermore,the available time series was too short to properly capture the full range of streamflow variability, which reduced the prediction performance outside of the calibration conditions. These findings suggest that ML algorithms, particularly SVR, can provide accurate streamflow estimation useful for water resources management when trained on a representative period. The results highlight the capacity of Machine Learning algorithms, specifically SVR, to augment streamflow prediction for enhanced water resource management in arid regions.

On the Challenges of Simulating Streamflow in Glacierized Catchments of the Himalayas Using Satellite and Reanalysis Forcing Data

Abstract Hydrologic assessment of climate change impacts on complex terrains and data-sparse regions like High Mountain Asia is a major challenge. Combining hydrological models with satellite and reanalysis data for evaluating changes in hydrological variables is often the only available approach. However, uncertainties associated with the forcing dataset, coupled with model parameter uncertainties, can have significant impacts on hydrologic simulations. This work aims to understand and quantify how the uncertainty in precipitation and its interaction with the model uncertainty affect streamflow estimation in glacierized catchments. Simulations for four precipitation datasets [Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG), Climate Hazards Group Infrared Precipitation with Station (CHIRPS), ERA5-Land, and Asian Precipitation–Highly Resolved Observational Data Integration Toward Evaluation (APHRODITE)] and two glacio-hydrological models [Glacio-Hydrological Degree-Day Model (GDM) and Hydrological Model for Distributed Systems (HYMOD_DS)] are evaluated for the Marsyangdi and Budhigandaki River basins in Nepal. Temperature sensitivity of streamflow simulations is also investigated. Relative to APHRODITE, which compared well with ground stations, ERA5-Land overestimates the catchment average precipitation for both basins by more than 70%; IMERG and CHIRPS overestimate by ∼20%. Precipitation uncertainty propagation to streamflow exhibits strong dependencies to model structure and streamflow components (snowmelt, ice melt, and rainfall-runoff), but overall uncertainty dampens through precipitation-to-streamflow transformation. Temperature exerts a significant additional source of uncertainty in hydrologic simulations of such environments. GDM was found to be more sensitive to temperature variations, with &gt;50% increase in total flow for 20% increase in actual temperature, emphasizing that models that rely on lapse rates for the spatial distribution of temperature have much higher sensitivity. Results from this study provide critical insight into the challenges of utilizing satellite and reanalysis products for simulating streamflow in glacierized catchments. Significance Statement This work investigates the uncertainty of streamflow simulations due to climate forcing and model parameter/structure uncertainty and quantifies the relative importance of each source of uncertainty and its impact on simulating different streamflow components in glacierized catchments of High Mountain Asia. Results highlight that in high mountain regions, temperature uncertainty exerts a major control on hydrologic simulations and models that do not adequately represent the spatial variability of temperature are more sensitive to bias in the forcing data. These findings provide guidance on important aspects to be considered when modeling glacio-hydrological response of catchments in such areas and are thus expected to impact both research and operation practice related to hydrologic modeling of glacierized catchments.

Exploring the applicability of the experiment-based ANN and LSTM models for streamflow estimation

The Yeşilırmak River Basin in northern Türkiye is crucial for the region’s water supply, agriculture, hydroelectric power generation, and clean drinking water. The primary goal of this study is to determine which modeling approach is most appropriate for various locations within the basin and how well meteorological data can predict river flow rates. Hydrological and meteorological forecasting both depend on the prediction of river flow rates. An artificial neural network (ANN), Univariate and Multivariate Long Short-Term Memory (LSTM) models have been utilized for streamflow forecasting. This research aims to determine the best model for several provinces in the basin area and give decision-makers a tool for reliable river flow rate estimates by combining LSTM and ANN models. According to research findings, the supervised multivariate LSTM model performed better than the unsupervised model in accuracy and precision. The sliding window methodology is suitable for estimating river flow based on meteorological datasets because it offers a primary method for reinterpreting time-series data in a supervised learning style. Compared to LSTM models, the ANN model that has been statistically optimized through experiments (DoE) design performs better in forecasting the river flow rate in the Yeşilırmak River basin (R2 = 0.98, RMSE = 0.18). The study’s findings provided prospective cognitive models for the strategic management of water resources by forecasting future data from flow monitoring stations.

Optimal combinations of global evapotranspiration and terrestrial water storage products for catchment water balance

ABSTRACT During the last two decades, a great number of global products about evapotranspiration ( E ) and terrestrial water storage ( S ) have been released. This has led to numerous combinations for describing catchment water balance, and hence determining an appropriate combination has become an important issue. The main objective of this study is to evaluate various global products of E and S at the catchment scale and determine the most appropriate combination, taking two Korean catchments as examples. For E , we evaluated global evapotranspiration (Global ET), Global Land Evaporation Amsterdam Model (GLEAM), and Penman–Monteith–Leuning (PML). For S , six individual datasets of Gravity Recovery and Climate Experiment (GRACE) were evaluated, with Global Land Data Assimilation System (GLDAS) products used for comparison. We found that the performance is sensitive to the choice of E product while various S products displayed minimal differences in results. Based on four evaluation criteria, the combination of PML and GRACE-SH-GFZ is suggested as the most suitable pair. For the period between 2003 and 2013, these data exhibit noteworthy trends of increasing E , offset by decreasing S . Warming climate is suspected to be behind these trends. Our research presents an approach that allows for the estimation of monthly streamflow exclusively with global data products, which is an advancement in hydrological analysis and particularly useful for regions that lack in-situ data networks. This approach provides a new perspective in the application of global datasets for the assessment of water balance and could significantly improve predictions in ungauged basins.

Modified calibration strategies and parameter regionalization potential for streamflow estimation using a hydrological model

ABSTRACT Calibration and regionalization are two key tasks in hydrological modelling, and the availability of increased data and computing resources has enhanced our ability to complete these tasks. This paper addresses both tasks, i.e. (i) it proposes novel model structure-based calibration strategies with the goal to improve estimation of high streamflows, and (ii) it explores the association of soil hydraulic properties with optimal parameters to estimate streamflows in ungauged catchments. Both tasks are demonstrated by employing a conceptual hydrological model, Sacramento Soil Moisture Accounting (SAC-SMA), on multiple catchments in the Narmada River Basin. The initial calibrated model exhibited good model performance, with high flows being underestimated. Seven calibration strategies via revising parameter space are explored, and improvement in high flow performance is observed at nine out of 12 catchments. Estimated streamflows based on soil hydraulic parameters demonstrate regionalization potential. The results indicate successful application of the proposed methods which are transferable to other basins.

A multi-model evaluation of probabilistic streamflow predictions via residual error modelling

Probabilistic streamflow predictions are valuable tools for predictive uncertainty estimation, hydrologic risk management, and support for decision-making in water resources. Usually, predictive uncertainty quantification is developed and assessed using only a single hydrological model, making it difficult to generalize to other model configurations. To tackle this issue, we assess changes in the model performance ranking of diverse streamflow models by applying a residual error model post-processing approach to multiple basins and multiple models. This assessment employed 141 basins from the Great Lakes watershed covering the USA and Canada, and 13 diverse streamflow models, which are evaluated using deterministic and probabilistic performance metrics. As the first study to implement probabilistic methods to diverse streamflow models applied to a multitude of basins, the analysis here examines the dependence of probabilistic streamflow estimation quality on model quality. Our findings show that streamflow model choice influences the robustness of probabilistic predictions. It was found that moving from deterministic to probabilistic predictions using a post-processing approach does not change the streamflow model performance ranking for the best and worst deterministic models, but models of intermediate rank in deterministic evaluation do not have consistent ranking when evaluated in probabilistic mode. Post-processing residual errors of long short-term memory (LSTM) network models are consistently the best-performing model in terms of deterministic and probabilistic metrics. This study highlights the significance of combining deterministic streamflow model predictions with residual error models for improving the quality and increasing the value of hydrological predictions, quantifying uncertainty, and facilitating decision-making in operational water management. It also clarifies the degree to which probabilistic predictions depend upon good model performance and can compensate for poor model performance.

Uncertain Benefits of Using Remotely Sensed Evapotranspiration for Streamflow Estimation—Insights From a Randomized, Large-Sample Experiment

Remotely sensed evapotranspiration (ETRS) shows promise for enhancing hydrological models, especially in regions lacking in situ streamflow observations. However, model calibration studies showed conflicting results regarding the ability of ETRS products to improve streamflow simulation. Rather than relying on model calibration, here we produce the first randomized experiment that explores the full streamflow–ET skill distribution, and also the first probabilistic assessment of the value of different global ETRS products for streamflow simulation. Using 280,000 randomized SWAT (Soil and Water Assessment Tool) model runs across seven catchments and four ETRS products, we show that the relationship between ET and streamflow skills is complex, and simultaneous improvement in both skills is only possible in a limited range. Parameter sensitivity analysis indicates that the most sensitive parameters can have opposite contributions to ET and streamflow skills, leading to skill trade-offs. Conditional probability assessment reveals that models with good ET skills are likely to produce good streamflow skills, but not vice versa. We suggest that randomized experiments such as ours should be performed before model calibration to determine whether using ETRS is worthwhile, and to help in interpreting the calibration results.

Identification of Time-Varying Conceptual Hydrological Model Parameters with Differentiable Parameter Learning

The parameters of the GR4J-CemaNeige coupling model (GR4neige) are typically treated as constants. However, the maximum capacity of the production store (parX1) exhibits time-varying characteristics due to climate variability and vegetation coverage change. This study employed differentiable parameter learning (dPL) to identify the time-varying parX1 in the GR4neige across 671 catchments within the United States. We built two types of dPL, including static and dynamic parameter networks, to assess the advantages of the time-varying parameter. In the dynamic parameter network, we evaluated the impact of potential evapotranspiration (PET), precipitation (P), temperature (T), soil moisture (SM), and normalized difference vegetation index (NDVI) datasets on the performance of dPL. We then compared dPL with the empirical functional method (fm). The results demonstrated that the dynamic parameter network outperformed the static parameter network in streamflow estimation. There were differences in streamflow estimation among the dynamic parameter network driven by various input features. In humid catchments, simultaneously incorporating all five factors, including PET, P, T, SM, and the NDVI, achieved optimal streamflow simulation accuracy. In arid catchments, it was preferable to introduce PET, T, and the NDVI separately for improved performance. dPL significantly outperformed the empirical fm in estimating streamflow and uncalibrated intermediate variables, like evapotranspiration (ET). Both the derived parX1 from dPL and the empirical fm exhibited significant spatiotemporal variation across 671 catchments. Notably, compared to parX1 obtained through the empirical fm, parX1 derived from dPL exhibited a distinct spatial clustering pattern. This study highlights the potential of dPL in enhancing model accuracy and contributes to understanding the spatiotemporal variation characteristics of parX1 under the influence of climate factors, soil conditions, and vegetation change.

Reconstruction of missing streamflow series in human-regulated catchments using a data integration LSTM model

Study regionYellow River Basin in China, where streamflow dynamics were significantly impacted by human activities. Study focusWe introduced a deep learning-based method, i.e., Data Integration (DI) with Long Short-Term Memory (LSTM), which leverages Global Flood Awareness System (GloFAS) streamflow data. Multiscale (Catchment, River) attributes were incorporated into the DI LSTM to represent human disturbances on land surface. We employed this method to reconstruct daily streamflow series in 60 human-regulated catchments across the Yellow River Basin, and identified the sensitivity of the DI LSTM model to the multiscale attributes. New hydrological Insights for the RegionOur findings revealed that the DI LSTM model achieved favourable performance in streamflow estimation, with the highest Kling-Gupta efficiency (KGE) reaching up to 0.9, outperforming the Regular LSTM model, which was forced by meteorological variables. Multiscale attributes can enhance the DI model performance, particularly in large catchments with significant human activities. A two-step validation demonstrated the high accuracy of the reconstructed streamflow data across the Yellow River Basin, as the KGEs for streamflow estimation in 40 catchments are over 0.6. In summary, the DI LSTM model shows great potential for reconstructing streamflow in human-regulated catchments in arid regions. The reconstructed daily streamflow data contribute valuable insights for monitoring changing hydrological conditions, especially in regions lacking extensive streamflow monitoring networks.

Improving modelled streamflow using time-varying multivariate assimilation of remotely sensed soil moisture and in-situ streamflow observations

Hydrological models are widely used to estimate and forecast streamflow for various applications. Given the inherent uncertainties in these models, there is a pressing need to enhance the current state of the modelling strategies. Historically, hydrological models showed significant improvement through soil moisture (SM) assimilation. Particularly, when SM observations are combined with streamflow observations from the interior catchment locations during multivariate assimilation, models showed better performance. Recent studies also emphasized the importance of updating the parameters due to the transient nature of the catchment during the assimilation period. Additionally, it is crucial to determine whether it is necessary to assimilate all available observations. To address these issues, this study introduces five data assimilation (DA) scenarios ingesting advanced scatterometer (ASCAT) SM and in-situ streamflow observations into a conceptual hydrological model, aiming to enhance its performance. The findings reveal that while univariate assimilation improves both SM and streamflow estimates, but a substantial enhancement is observed in multivariate data assimilation (MVDA). Time-varying MVDA (TV-MVDA) substantially improves the model's performance compared to the open-loop scenario case. However, this improvement is only marginal when compared to the TV-MVDA scenario. Finally, the sensitivity-based TV-MVDA scenario exhibits comparable performance in streamflow estimation, while assimilating just 30 % of observations. These results suggest that Sensitivity based TV-MVDA can improve the model efficiently with minimal observation requirements.

Leveraging historic streamflow and weather data with deep learning for enhanced streamflow predictions

ABSTRACT Streamflow information is crucial for effectively managing water resources. The declining number of active gauging stations in many rivers is a global concern, necessitating the need for reliable streamflow estimates. Deep learning techniques offer potential solutions, but their application in southern Africa remains largely underexplored. To fill this gap, this study evaluated the predictive performance of gated recurrent unit (GRU) and long short-term memory (LSTM) networks using two headwater catchments of the Steelpoort River, South Africa, as case studies. The model inputs included rainfall, maximum, and minimum temperature, as well as past streamflow, which was utilized in an autoregressive sense. The inclusion of streamflow in this way allowed for the incorporation of simulated streamflow values into the look-back window for predicting the streamflow of the testing set. Two modifications were required to the GRU and LSTM architectures to ensure physically consistent predictions, including a change in the activation function of the GRU/LSTM cells in the final hidden layer, and a non-negative constraint that was used in the dense layer. Models trained using commercial weather station data produced reliable streamflow estimates, while moderately accurate predictions were obtained using freely available gridded weather data.