A Cluster-based Temporal Attention Approach for Predicting Cyclone-induced Compound Flood Dynamics

Environmental Modelling & Software, Journal Year: 2025, Volume and Issue: unknown, P. 106499 - 106499

Published: April 1, 2025

Language: Английский

---

Comprehensive evaluation of IMERG, ERA5-Land and their fusion products in the hydrological simulation of three karst catchments in Southwest China

Journal of Hydrology Regional Studies, Journal Year: 2024, Volume and Issue: 52, P. 101671 - 101671

Published: Jan. 20, 2024

The study was carried out in three karst catchments Southwest China. These catchments, Sancha, Liuzhou and Qianjiang, are located the middle reaches of Pearl River Basin with catchment area 17067 km², 46166 134137 respectively. Satellite or reanalysis precipitation data potential alternatives that can be used hydrological models for streamflow forecasting data-sparse catchments. This aims to investigate performance two widely datasets, IMERG ERA5-Land, as well their fusion rain gauge measurements by Geographically Weighted Regression method, southwest results indicate compared IMERG, ERA5-Land has a higher correlation coefficient better detection rate daily gauge-based data. However, overestimates annual all Merging further improve its rate, but does not effectively mitigate overestimation. Meanwhile, model calibration through parameter adjustment partly compensate error discharge simulation accuracy, it cannot fully cover overestimation Therefore, consistently perform badly simulation. In contrast, demonstrates good agreement shows comparable even slightly superior prediction Additionally, calibrated parameters driven closely resemble those suggest serve viable alternative simulations region.

Language: Английский

---

Uncertainty estimation of machine learning spatial precipitation predictions from satellite data

Machine Learning Science and Technology, Journal Year: 2024, Volume and Issue: 5(3), P. 035044 - 035044

Published: Aug. 13, 2024

Abstract Merging satellite and gauge data with machine learning produces high-resolution precipitation datasets, but uncertainty estimates are often missing. We addressed the gap of how to optimally provide such by benchmarking six algorithms, mostly novel even for more general task quantifying predictive in spatial prediction settings. On 15 years monthly from over contiguous United States, we compared quantile regression (QR), forests (QRF), generalized random (GRF), gradient boosting machines (GBM), light (LightGBM), neural networks (QRNN). Their ability issue quantiles at nine levels (0.025, 0.050, 0.100, 0.250, 0.500, 0.750, 0.900, 0.950, 0.975), approximating full probability distribution, was evaluated using scoring functions rule. Predictors a site were nearby values two retrievals, namely Precipitation Estimation Remotely Sensed Information Artificial Neural Networks (PERSIANN) Integrated Multi-satellitE Retrievals (IMERG), site’s elevation. The dependent variable mean precipitation. With respect QR, LightGBM showed improved performance terms rule 11.10%, also surpassing QRF (7.96%), GRF (7.44%), GBM (4.64%) QRNN (1.73%). Notably, outperformed all forest variants, current standard learning. To conclude, propose suite algorithms estimating prediction, supported formal evaluation framework based on rules.

Language: Английский

---

A novel error decomposition and fusion framework for daily precipitation estimation based on near-real-time satellite precipitation product and gauge observations

Journal of Hydrology, Journal Year: 2024, Volume and Issue: 640, P. 131715 - 131715

Published: July 23, 2024

Language: Английский

---

Streamflow Prediction in Human‐Regulated Catchments Using Multiscale Deep Learning Modeling With Anthropogenic Similarities

Water Resources Research, Journal Year: 2024, Volume and Issue: 60(9)

Published: Sept. 1, 2024

Abstract Accurate streamflow prediction in human‐regulated catchments remains a formidable challenge due to the complex disturbance of hydrological processes. To consider human modeling, this study introduces novel static attribute collection that combines river‐reach attributes with catchment attributes, referred as multiscale attributes. The is assembled into two deep learning (DL) methods, is, Long Short‐Term Memory (named Multiscale LSTM) and Differentiable Parameter Learning (DPL) model, performance evaluated across 95 United States (USA) 24 Yellow River Basin China. In USA, LSTM DPL models achieve similar median Kling‐Gupta Efficiency (KGE) 0.78 0.71, respectively. However, Basin, KGE values are 0.58 for 0.24 DPL. These results highlight DL models' ability leverage improved compared traditional predominantly influenced by river‐scale encompassing factors such connectivity status index (CSI), degree regulation (DOR), sediment trapping (SED), number dams. Additionally, satellite‐derived mean maximum river width (Width), slope water surface elevation (WSE) from Surface Water Ocean Topography Database (SWORD) contribute valuable insights anthropogenic influences. Moreover, our highlights significance selecting appropriate training data period, which emerges most dominant factor affecting model catchments. diversity during period enables capture broad spectrum signatures within these Consequently, emphasizes advantages underscores considering both natural enhance predictions environments.

Language: Английский

---

Enhanced rainfall nowcasting of tropical cyclone by an interpretable deep learning model and its application in real-time flood forecasting

Journal of Hydrology, Journal Year: 2024, Volume and Issue: unknown, P. 131993 - 131993

Published: Sept. 1, 2024

Language: Английский

---

Enhancing Streamflow Prediction in Ungauged Basins Using a Nonlinear Knowledge‐Based Framework and Deep Learning

Water Resources Research, Journal Year: 2024, Volume and Issue: 60(11)

Published: Oct. 30, 2024

Abstract In hydrology, a fundamental task involves enhancing the predictive power of model in ungagged basins by transferring information on physical attributes and hydroclimate dynamics from gauged basins. Introducing an integrated nonlinear clustering framework, this study aims to develop comprehensive framework that augments performance where direct measurements are sparse or absent. uniform manifold approximation projection (UMAP) is used as method extract essential features embedded hydro‐climatological properties. Then, Growing Neural Gas (GNG) find potentially share similar behaviors. Besides UMAP‐GNG, integration Principal Component Analysis (PCA) linear reduce dimensionality with common methods also assessed serve benchmarks. The results reveal combination algorithms PCA may lead loss while can more informative features. efficacy proposed across Contiguous United States (CONUS) training single Base Model using long short‐term memory (LSTM) for centroids all clusters then, fine‐tuning each cluster separately create regional model. indicate extracted UMAP‐GNG guide significantly improve accuracy most enhance median prediction within different 0.04 0.37 KGE ungauged

Language: Английский

---

Bayesian Model Averaging for Satellite Precipitation Data Fusion: From Accuracy Estimation to Runoff Simulation

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(7), P. 1154 - 1154

Published: March 25, 2025

Precipitation plays a vital role in the hydrological cycle, directly affecting water resource management and influencing flood drought risk prediction. This study proposes Bayesian Model Averaging (BMA) framework to integrate multiple precipitation datasets. The enhances estimation accuracy for simulations. BMA synthesizes four products—Climate Hazards Group Infrared with Station (CHIRPS), fifth-generation ECMWF Atmospheric Reanalysis (ERA5), Global Satellite Mapping of (GSMaP), Integrated Multi-satellitE Retrievals (IMERG)—over China’s Ganjiang River Basin from 2008 2020. We evaluated merged dataset’s performance against its constituent datasets Multi-Source Weighted-Ensemble (MSWEP) at daily, monthly, seasonal scales. Evaluation metrics included correlation coefficient (CC), root mean square error (RMSE), Kling–Gupta efficiency (KGE). Variable Infiltration Capacity (VIC) model was further applied assess how these affect runoff results indicate that BMA-merged dataset substantially improves when compared individual inputs. product achieved optimal daily (CC = 0.72, KGE 0.70) showed superior skill, notably reducing biases autumn winter. In applications, BMA-driven VIC effectively replicated observed patterns, demonstrating efficacy regional long-term predictions. highlights BMA’s potential optimizing inputs, providing critical insights sustainable reduction complex basins.

Language: Английский

---

Passive satellite hourly precipitation estimation over mainland China by combining cloud and meteorological parameters

Atmospheric Research, Journal Year: 2025, Volume and Issue: unknown, P. 108112 - 108112

Published: April 1, 2025

Language: Английский

---

Improving rainfall-runoff modelling using the fusion of satellite-based and gauge precipitation products in a data-sparse region

Hydrological Sciences Journal, Journal Year: 2025, Volume and Issue: unknown

Published: April 17, 2025

Language: Английский

---