Comparison of strategies for multistep-ahead lake water level forecasting using deep learning models

Journal of Cleaner Production, Journal Year: 2024, Volume and Issue: 444, P. 141228 - 141228

Published: Feb. 13, 2024

Language: Английский

---

Quantifying Predictive Uncertainty and Feature Selection in River Bed Load Estimation: A Multi-Model Machine Learning Approach with Particle Swarm Optimization

Water, Journal Year: 2024, Volume and Issue: 16(14), P. 1945 - 1945

Published: July 10, 2024

This study presents a comprehensive multi-model machine learning (ML) approach to predict river bed load, addressing the challenge of quantifying predictive uncertainty in fluvial geomorphology. Six ML models—random forest (RF), categorical boosting (CAT), extra tree regression (ETR), gradient (GBM), Bayesian model (BRM), and K-nearest neighbors (KNNs)—were thoroughly evaluated across several performance metrics like root mean square error (RMSE), correlation coefficient (R). To enhance training optimize performance, particle swarm optimization (PSO) was employed for hyperparameter tuning all models, leveraging its capability efficiently explore complex spaces. Our findings indicated that RF, GBM, CAT, ETR demonstrate superior (R score > 0.936), benefiting significantly from PSO. In contrast, BRM displayed lower (0.838), indicating challenges with approaches. The feature importance analysis, including permutation SHAP values, highlighted non-linear interdependencies between variables, discharge (Q), slope (S), flow width (W) being most influential. also examined specific impact individual variables on by adding excluding which is particularly meaningful when choosing input model, especially limited data conditions. Uncertainty quantification through Monte Carlo simulations enhanced predictability reliability models larger datasets. increased improved precision evident consistent rise R scores reduction standard deviations as sample size increased. research underscored potential advanced ensemble methods PSO mitigate limitations single-predictor exploit collective strengths, thereby improving predictions load estimation. insights this provide valuable framework future directions focused optimizing configurations hydro-dynamic modeling.

Language: Английский

---

A STUDY ON DEEP LEARNING FOR RIVER WATER LEVEL PREDICTION IN URBAN WATERSHEDS WITH A DISCUSSION ON UNDERGROUND REGULATING RESERVOIR GATE OPERATIONS

Journal of JSCE, Journal Year: 2025, Volume and Issue: 13(2), P. n/a - n/a

Published: Jan. 1, 2025

Language: Английский

---

Long-term prediction of Poyang Lake water level by combining multi-scale isometric convolution network with quantile regression

Journal of Hydrology Regional Studies, Journal Year: 2025, Volume and Issue: 59, P. 102365 - 102365

Published: April 17, 2025

Language: Английский

---

Towards an efficient streamflow forecasting method for event-scales in Ca River basin, Vietnam

Journal of Hydrology Regional Studies, Journal Year: 2023, Volume and Issue: 46, P. 101328 - 101328

Published: Feb. 1, 2023

The Ca River basin is located in the North Central Coast area of Vietnam This study aims to develop a deep learning framework that both effective and straightforward order forecast water levels advance multiple time steps for event scales. We have thoroughly studied assessed two models (DLMs), long-short term memory (LSTM) gated recurrent unit (GRU), their capacity levels, focusing on various aspects such as influence sequence length or impact hyperparameter selection. Besides, data scenarios were established using hydrological from eight severe floods between 2007 2019 examine effect input variables model performance. Water level was employed (S1 S2), whereas precipitation used only S2. cross-validation technique dynamically address issue limited data. inputs reformatted tensors then randomly divided into subsets. flexible tuning preserved sequential nature while enabling DLMs be trained efficiently. findings revealed exhibited equally excellent performances. NSE LSTM varies 0.999–0.971 compared 0.998–0.974 GRU model, corresponding cases one four-time ahead. indicated use multiple-input types (S2) contrary date type (S1) does not necessarily improve forecasting LSTM/GRU with hidden layer are adequate delivering high performance minimizing processing time.

Language: Английский

---

Deep neural network-based discharge prediction for upstream hydrological stations: a comparative study

Earth Science Informatics, Journal Year: 2023, Volume and Issue: 16(4), P. 3113 - 3124

Published: Aug. 21, 2023

Language: Английский

---

Unlocking precision in hydraulic engineering: machine learning insights into labyrinth sluice gate discharge coefficients

Journal of Hydroinformatics, Journal Year: 2024, Volume and Issue: 26(11), P. 2883 - 2901

Published: Nov. 1, 2024

ABSTRACT This study investigates the discharge coefficient (Cd) of labyrinth sluice gates, a modern gate design with complex flow characteristics. To accurately estimate Cd, regression techniques (linear and stepwise polynomial regression) machine learning methods (gene expression programming (GEP), decision table, KStar, M5Prime) were employed. A dataset 187 experimental results, incorporating dimensionless variables internal angle (θ), cycle number (N), water depth contraction ratio (H/G), was used to train evaluate models. The results demonstrate superiority GEP in predicting achieving determination (R2) 97.07% mean absolute percentage error 2.87%. assess relative importance each variable, sensitivity analysis conducted. revealed that H/G has most significant impact on followed by head (θ). (N) found have relatively insignificant effect. These findings offer valuable insights into operation contributing improved resource management flood control.

Language: Английский

---

A Novel Framework for Correcting Satellite-Based Precipitation Products for Watersheds with Discontinuous Observed Data, Case Study in Mekong River Basin

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(3), P. 630 - 630

Published: Jan. 20, 2023

Satellite-based precipitation (SP) data are gaining scientific interest due to their advantage in producing high-resolution products with quasi-global coverage. However, since the major reliance of is on distinctive geographical features each location, they remain at a considerable distance from station-based data. This paper examines effectiveness convolutional autoencoder (CAE) architecture pixel-by-pixel bias correction SP for Mekong River Basin (MRB). Two satellite-based (TRMM and PERSIANN-CDR) gauge-based product (APHRODITE) gridded rainfall mined this experiment. According estimated statistical criteria, CAE model was effective reducing gap between benchmark both terms spatial temporal correlations. The two corrected (CAE_TRMM CAE_CDR) performed competitively, TRMM appearing have slight over CDR, however, difference minor. study’s findings proved deep learning-based models (here CAE) products. We believe that technique will be feasible alternative delivering an up-to-current reliable dataset MRB studies, given sole available area has been out date long time.

Language: Английский

---

Performance Comparison of Bias-Corrected Satellite Precipitation Products by Various Deep Learning Schemes

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2023, Volume and Issue: 61, P. 1 - 12

Published: Jan. 1, 2023

Precipitation observations from a ground-based gauge provide reliable data source for hydrological and climatological studies. However, these are sparse in many regions of the world, particularly Mekong River Basin (MRB). Satellite-based precipitation products (SPPs) sole available with worldwide coverage. Despite this, there is mismatch between SPPs gauge-based observations, correct procedures should be utilized to minimize systematic bias SPPs. This study aimed benchmark efficacy four state-of-the-art bias-correcting deep learning models (DLMs) tropical rainfall measuring mission-based product named TRMM_3B42 (hereafter TRMM) over entire MRB. These were designed mainly based on convolutional neural network (CNN) encoder–decoder (ENDE) architectures, including ConvENDE, ConvUNET, ConvINCE, ConvLSTM. The bias-corrected dataset by DLMs was then confirmed against (Asian precipitation-highly resolved observational integration toward evaluation water resources, APHRODITE). From results obtained, all effectively minimized TRMM product. Among them, ConvENDE ConvUNET had higher consistency performance level compared ConvINCE Additionally, complexity did not enhance their efficiency, as case ConvLSTM, despite using computing resources. Given observed shortage MRB since 2016, application DLMs, such can serve improve reliability existing datasets valuable input various research purposes

Language: Английский

---

Determination of the surface roller length of hydraulic jumps in horizontal rectangular channels using the machine learning method

Stochastic Environmental Research and Risk Assessment, Journal Year: 2024, Volume and Issue: 38(7), P. 2539 - 2562

Published: March 29, 2024

Language: Английский

---