Comparing a long short-term memory (LSTM) neural network with a physically-based hydrological model for streamflow forecasting over a Canadian catchment

Journal of Hydrology, Journal Year: 2023, Volume and Issue: 627, P. 130380 - 130380

Published: Oct. 21, 2023

Streamflow forecasting is crucial in water planning and management. Physically-based hydrological models have been used for a long time these fields, but improving forecast quality still an active area of research. Recently, some artificial neural networks found to be effective simulating predicting short-term streamflow. In this study, we examine the reliability Long Short-Term Memory (LSTM) deep learning model streamflow lead times up ten days over Canadian catchment. The performance LSTM compared that process-based distributed model, with both using same weather ensemble forecasts. Furthermore, LSTM’s ability integrate observed on issue date data assimilation process required reduce initial state biases. Results indicate forecasted streamflows are more reliable accurate lead-times 7 9 days, respectively. Additionally, it shown recent flows as predictor can smaller errors first without requiring explicit step, generating median value mean absolute error (MAE) day lead-time across all dates 25 m3/s 115 assimilated model.

Language: Английский

---

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

Journal of Hydrology Regional Studies, Journal Year: 2024, Volume and Issue: 52, P. 101744 - 101744

Published: March 15, 2024

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

Language: Английский

---

Enhancing long short-term memory (LSTM)-based streamflow prediction with a spatially distributed approach

Hydrology and earth system sciences, Journal Year: 2024, Volume and Issue: 28(9), P. 2107 - 2122

Published: May 14, 2024

Abstract. Deep learning (DL) algorithms have previously demonstrated their effectiveness in streamflow prediction. However, hydrological time series modelling, the performance of existing DL methods is often bound by limited spatial information, as these data-driven models are typically trained with lumped (spatially aggregated) input data. In this study, we propose a hybrid approach, namely Spatially Recursive (SR) model, that integrates long short-term memory (LSTM) network seamlessly physics-based routing simulation for enhanced The LSTM was on basin-averaged meteorological and variables derived from 141 gauged basins located Great Lakes region North America. SR model involves applying at subbasin scale local predictions which then translated to basin outlet model. We evaluated efficacy respect predicting 224 stations across compared its standalone results indicate achieved levels par used training LSTM. Additionally, able predict more accurately large (e.g., drainage area greater than 2000 km2), underscoring substantial information loss associated basin-wise feature aggregation. Furthermore, outperformed when applied were not part (i.e., pseudo-ungauged basins). implication study predictions, especially ungauged basins, can be reliably improved considering heterogeneity finer resolution via

Language: Английский

---

Streamflow prediction in ungauged catchments through use of catchment classification and deep learning

Journal of Hydrology, Journal Year: 2024, Volume and Issue: 639, P. 131638 - 131638

Published: July 3, 2024

Language: Английский

---

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

Hydrology and earth system sciences, Journal Year: 2024, Volume and Issue: 28(13), P. 2871 - 2893

Published: July 4, 2024

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

Language: Английский

---

Boruta extra tree-bidirectional long short-term memory model development for Pan evaporation forecasting: Investigation of arid climate condition

Alexandria Engineering Journal, Journal Year: 2023, Volume and Issue: 86, P. 425 - 442

Published: Dec. 7, 2023

In this study, two deep learning approaches, bidirectional long short-term memory (BiLSTM) and (LSTM), were used along with adaptive boosting general regression neural network to forecast multi-step-ahead pan evaporation in arid climate stations Iran (Ahvaz Yazd). Lagged time series of meteorological data input the machine models. Two feature selection methods, i.e., Boruta extra tree XGBoost, select significant inputs reduce number model complexity. Different statistical metrics investigate performance. The results demonstrated that Boruta-extra-tree-based models more accurate than XGBoost-based Compared techniques, combination BiLSTM enabled one-day-ahead forecasting for both sites (Root Mean Square Error (RMSE) = 1.6857, Ahvaz station, RMSE 1.3996 Yazd station). proposed was up 30 days ahead stations. showed Boruta-BiLSTM could accurately

Language: Английский

---

Data reformation – A novel data processing technique enhancing machine learning applicability for predicting streamflow extremes

Advances in Water Resources, Journal Year: 2023, Volume and Issue: 182, P. 104569 - 104569

Published: Nov. 3, 2023

Language: Английский

---

Streamflow regime-based classification and hydrologic similarity analysis of catchment behavior using differentiable modeling with multiphysics outputs

Journal of Hydrology, Journal Year: 2025, Volume and Issue: 653, P. 132766 - 132766

Published: Jan. 29, 2025

Language: Английский

---

Improving trans-regional hydrological modelling by combining LSTM with big hydrological data

Journal of Hydrology Regional Studies, Journal Year: 2025, Volume and Issue: 58, P. 102257 - 102257

Published: Feb. 19, 2025

Language: Английский

---

CAMELS-DK: hydrometeorological time series and landscape attributes for 3330 Danish catchments with streamflow observations from 304 gauged stations

Earth system science data, Journal Year: 2025, Volume and Issue: 17(4), P. 1551 - 1572

Published: April 14, 2025

Abstract. Large samples of hydrometeorological time series and catchment attributes are critical for improving the understanding complex hydrological processes, model development, performance benchmarking. CAMELS (Catchment Attributes Meteorology Large-sample Studies) datasets have been developed in several countries regions around world, providing valuable data sources test beds analysis new frontiers data-driven modeling. Regarding lack from lowland, groundwater-dominated, small-sized catchments, we develop an extensive repository a CAMELS-style dataset Denmark (CAMELS-DK). This addition is first containing both gauged ungauged catchments as well detailed groundwater information. The provides dynamic static variables 3330 covering all various hydrogeological datasets, meteorological observations, well-established national-scale model. For 304 those streamflow observations provided, whereas simulated provided catchments. contains spanning 30 years (1989–2019) with daily step, will be updated once simulations become available. dense full spatial coverage instead only together simulation distributed, process-based model, enhances applicability such data, example, development hybrid physically informed modeling frameworks or other cases where consistent required. We also provide quantities related to human impact on system Denmark, abstraction irrigation. CAMELS-DK freely available at https://doi.org/10.22008/FK2/AZXSYP (Koch et al., 2024).

Language: Английский

---