Temporal Fusion Transformers for streamflow Prediction: Value of combining attention with recurrence

Journal of Hydrology, Journal Year: 2024, Volume and Issue: 637, P. 131301 - 131301

Published: May 9, 2024

Language: Английский

---

A deep learning-based hybrid approach for multi-time-ahead streamflow prediction in an arid region of Northwest China

Hydrology Research, Journal Year: 2024, Volume and Issue: 55(2), P. 180 - 204

Published: Jan. 10, 2024

Abstract Accurate streamflow prediction is crucial for effective water resource management. However, reliable remains a considerable challenge because of the highly complex, non-stationary, and non-linear processes that contribute to at various spatial temporal scales. In this study, we utilized convolutional neural network (CNN)–Transformer–long short-term memory (LSTM) (CTL) model prediction, which replaced embedding layer with CNN extract partial hidden features, added an LSTM correlations on scale. The CTL incorporated Transformer's ability global information, CNN's LSTM's capture correlations. To validate its effectiveness, applied it in Shule River basin northwest China across 1-, 3-, 6-month horizons compared performance Transformer, CNN, LSTM, CNN–Transformer, Transformer–LSTM. results demonstrated outperformed all other models terms predictive accuracy Nash–Sutcliffe coefficient (NSE) values 0.964, 0.912, 0.856 ahead prediction. best among five comparative were 0.908, 0.824, 0.778, respectively. This indicated outstanding alternative technique where surface data are limited.

Language: Английский

---

Explaining the Mechanism of Multiscale Groundwater Drought Events: A New Perspective From Interpretable Deep Learning Model

Water Resources Research, Journal Year: 2024, Volume and Issue: 60(7)

Published: June 27, 2024

Abstract This study presents a new approach to understand the causes of groundwater drought events with interpretable deep learning (DL) models. As prerequisites, accurate long short‐term memory (LSTM) models for simulating are built 16 regions representing three types spatial scales in southeastern United States, and standardized index is applied identify 233 events. Two interpretation methods, expected gradients (EG) additive decomposition (AD), adopted decipher DL‐captured patterns inner workings LSTM networks. The EG results show that: (a) temperature‐related features were primary drivers large‐scale droughts, their importance increasing from 56.1% 63.1% as approached 6 months 15 days. Conversely, precipitation‐related found be dominant factors formation small‐scale catchments, overall ranging 59.8% 53.3%; (b) Seasonal variations inversely related between large small scales, being more significant summer larger winter catchments; (c) exhibited an “trigger effect” on causing studying areas. AD method unveiled how network behaved differently retaining discarding information when emulating different droughts. In summary, this provides perspective highlights potential prospect DL enhancing our understanding hydrological processes.

Language: Английский

---

A state-of-the-art review of long short-term memory models with applications in hydrology and water resources

Applied Soft Computing, Journal Year: 2024, Volume and Issue: unknown, P. 112352 - 112352

Published: Oct. 1, 2024

Language: Английский

---

Urban real-time rainfall-runoff prediction using adaptive SSA-decomposition with dual attention

Journal of Hydrology, Journal Year: 2025, Volume and Issue: unknown, P. 132701 - 132701

Published: Jan. 1, 2025

Language: Английский

---

A hybrid model coupling process-driven and data-driven models for improved real-time flood forecasting

Journal of Hydrology, Journal Year: 2024, Volume and Issue: 638, P. 131494 - 131494

Published: June 13, 2024

Language: Английский

---

Predicting the performance of green stormwater infrastructure using multivariate long short-term memory (LSTM) neural network

Journal of Hydrology, Journal Year: 2023, Volume and Issue: 625, P. 130076 - 130076

Published: Aug. 10, 2023

The expected performance of Green Stormwater Infrastructure (GSI) is typically quantified through numerical models based on hydrologic parameters and physics-based equations. With models, the choice a spatio-temporal discretization scheme for computational domain strenuous task that requires extensive calibration potentially lab-based experimentation. GSI has high temporal dynamics due to natural, anthropogenic, climatic processes are not well represented by traditional which calibrated against only few historical observations have user-defined constrained set outcomes. Deep learning-based predictive such as Long Short-Term Memory (LSTM) neural networks, offer an exciting opportunity quantify performance, accounting its highly dynamic constantly evolving nature leveraging advancements in observational data. A LSTM regression can overcome some limitations associated with hydrological aid development fully data-informed predictor. To demonstrate model outcomes, both methods were applied rain garden Villanova, PA, USA. Specifically, was used predict recession ponded water depth using five years observed eight predictors (i.e., precipitation, air temperature, soil moisture content at 10 cm, 35 65 cm 91 depth) target variable rate) considered training/testing model. comparative study USEPA Storm Water Management Model (SWMM) performed observe continuous rate time series specific storms. had score, Root Mean Square Error (RMSE), 0.081 series, outperforming SWMM score 2.173 when compared In case storm-specific prediction, also outperformed simulation four storms lower RMSE values application predicting crucial stride towards efficient real-time forecasting.

Language: Английский

---

Generative deep learning for probabilistic streamflow forecasting: Conditional variational auto-encoder

Journal of Hydrology, Journal Year: 2023, Volume and Issue: 629, P. 130498 - 130498

Published: Nov. 21, 2023

Language: Английский

---

Uncertainty Forecasting Model for Mountain Flood Based on Bayesian Deep Learning

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 47830 - 47841

Published: Jan. 1, 2024

Due to the characteristics of strong suddenness, high harmfulness, and frequent occurrence mountain flood disasters in small watersheds, accuracy reliability forecasting are insufficient watersheds. This paper studies key theories technologies, that is uncertainty model based on hydrologic physical mechanism. We design Bayesian Deep Learning (DL) models, it suitable for transfer spatiotemporal factors caused by floods disaster probability. The models include Linear Long Short-Term Memory (LSTM) model, we hope achieve an acceptable balance between (uncertainty confidence coverage) (confidence interval width). Meanwhile, extract effective information from multi-source multi-dimensional hazard factors' big data. experiment shows differences DL have long-term probability ability at both, but LSTM superior terms reliability, computational consumption.

Language: Английский

---

A process-driven deep learning hydrological model for daily rainfall-runoff simulation

Journal of Hydrology, Journal Year: 2024, Volume and Issue: 637, P. 131434 - 131434

Published: June 1, 2024

Language: Английский

---