Pub Date : 2026-01-12DOI: 10.1016/j.jhydrol.2026.134945
Sijie Tang , Shuo Wang , Jiping Jiang , Yi Zheng
Flash droughts pose significant challenges to water resource management and agricultural sustainability, making it imperative to improve their predictability to mitigate potential risks. This study presents a novel deep learning framework that integrates a spatial–temporal causality-aware (STC) module into a CNN-LSTM hybrid architecture to enhance flash drought prediction in China’s Greater Bay Area (GBA). Ablation experiments demonstrate that the causality module enhances model generalization (GA = 0.90) and performance (NSE = 0.83), and substantially increases the accuracy of flash drought onset prediction (F1 score = 0.33) compared to baseline models. Explainable Artificial Intelligence (AI) analyses further reveal that incorporating causality strengthens the predictive contributions of key flash drought drivers, including soil moisture memory, downward longwave radiation, and precipitation. Especially, it reveals new insights into drought drivers: downward longwave radiation emerges as a critical yet previously underrecognized predictor of soil moisture variability in humid subtropical climates. Additionally, this study distinguishes the mechanisms underlying slow and flash droughts, highlighting the dominant role of initial soil moisture and persistent shortwave radiation in slow droughts, versus rapid energy imbalances and longwave radiation in flash droughts. Further findings suggest that anthropogenic activities in China’s GBA intensify the complexity of drought mechanisms, increasing both prediction difficulty and regional vulnerability to hydrological extremes. The proposed framework and insights provide a foundation for developing more effective flash drought risk management and adaptation strategies in humid subtropical regions.
{"title":"Improved flash drought forecasting and attribution: A spatial-temporal causality-aware deep learning approach","authors":"Sijie Tang , Shuo Wang , Jiping Jiang , Yi Zheng","doi":"10.1016/j.jhydrol.2026.134945","DOIUrl":"10.1016/j.jhydrol.2026.134945","url":null,"abstract":"<div><div>Flash droughts pose significant challenges to water resource management and agricultural sustainability, making it imperative to improve their predictability to mitigate potential risks. This study presents a novel deep learning framework that integrates a spatial–temporal causality-aware (STC) module into a CNN-LSTM hybrid architecture to enhance flash drought prediction in China’s Greater Bay Area (GBA). Ablation experiments demonstrate that the causality module enhances model generalization (GA = 0.90) and performance (NSE = 0.83), and substantially increases the accuracy of flash drought onset prediction (F1 score = 0.33) compared to baseline models. Explainable Artificial Intelligence (AI) analyses further reveal that incorporating causality strengthens the predictive contributions of key flash drought drivers, including soil moisture memory, downward longwave radiation, and precipitation. Especially, it reveals new insights into drought drivers: downward longwave radiation emerges as a critical yet previously underrecognized predictor of soil moisture variability in humid subtropical climates. Additionally, this study distinguishes the mechanisms underlying slow and flash droughts, highlighting the dominant role of initial soil moisture and persistent shortwave radiation in slow droughts, versus rapid energy imbalances and longwave radiation in flash droughts. Further findings suggest that anthropogenic activities in China’s GBA intensify the complexity of drought mechanisms, increasing both prediction difficulty and regional vulnerability to hydrological extremes. The proposed framework and insights provide a foundation for developing more effective flash drought risk management and adaptation strategies in humid subtropical regions.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"667 ","pages":"Article 134945"},"PeriodicalIF":6.3,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957162","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1016/j.jhydrol.2026.134937
Xinghan Xu, Lei Hu, Xingyi Miao, Peng Xiao, Xiaohui Yan, Jianwei Liu
Missing values in water quality data (WQD), caused by sensor malfunctions, communication failures, and environmental disturbances, undermine the reliability of conventional imputation methods. To address these challenges, we propose the Temporal Embedding-based Self-Attention t-distributed Variational Autoencoder (TE-SAVAE-St), a model designed for robust and temporally consistent data reconstruction. The model incorporates a Student’s-t prior to handle outliers, temporal embeddings (TE) to capture chronological patterns, and multi-head self-attention (MSA) to model long-term dependencies and inter-variable correlations. Extensive experiments on real-world WQD datasets show that TE-SAVAE-St outperforms ten baseline methods under various missing data scenarios, reducing RMSE by 11.8% and SMAPE by 23.6% compared to state-of-the-art models. Ablation studies confirm the complementary benefits of TE, MSA, and Student’s-t components. Additionally, time complexity analysis demonstrates that TE-SAVAE-St achieves an optimal balance between computational efficiency and imputation accuracy, making it suitable for real-time and large-scale monitoring applications. Overall, TE-SAVAE-St offers a domain-aware framework for the accurate reconstruction of incomplete WQD, supporting continuous environmental monitoring.
{"title":"Imputation of continuous missing values in water quality data using a temporal embedding-based self-attention variational autoencoder","authors":"Xinghan Xu, Lei Hu, Xingyi Miao, Peng Xiao, Xiaohui Yan, Jianwei Liu","doi":"10.1016/j.jhydrol.2026.134937","DOIUrl":"https://doi.org/10.1016/j.jhydrol.2026.134937","url":null,"abstract":"Missing values in water quality data (WQD), caused by sensor malfunctions, communication failures, and environmental disturbances, undermine the reliability of conventional imputation methods. To address these challenges, we propose the Temporal Embedding-based Self-Attention t-distributed Variational Autoencoder (TE-SAVAE-St), a model designed for robust and temporally consistent data reconstruction. The model incorporates a Student’s-t prior to handle outliers, temporal embeddings (TE) to capture chronological patterns, and multi-head self-attention (MSA) to model long-term dependencies and inter-variable correlations. Extensive experiments on real-world WQD datasets show that TE-SAVAE-St outperforms ten baseline methods under various missing data scenarios, reducing RMSE by 11.8% and SMAPE by 23.6% compared to state-of-the-art models. Ablation studies confirm the complementary benefits of TE, MSA, and Student’s-t components. Additionally, time complexity analysis demonstrates that TE-SAVAE-St achieves an optimal balance between computational efficiency and imputation accuracy, making it suitable for real-time and large-scale monitoring applications. Overall, TE-SAVAE-St offers a domain-aware framework for the accurate reconstruction of incomplete WQD, supporting continuous environmental monitoring.","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"100S 1","pages":""},"PeriodicalIF":6.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957165","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Combined sewer overflows (CSO) pose risks to both water quality and public health. However, the impact of varying rainfall characteristics on CSO remains unclear. This study aims to assess and analyze the correlations between CSO and rainfall characteristics, specifically rainfall depth, duration, and the proportion of rainfall occurring during peak periods. The objective is to identify and elucidate the impacts of each main rainfall characteristic on the frequency of CSO. To conduct this analysis, rainfall data spanning 71 years were divided into 1,435 rainfall events and categorized into five representative rainfall patterns based on their temporal distribution. A hydraulic model of sewer network was employed to simulate the CSO results under different rainfall patterns. An indicator termed “rainfall-CSO contribution rate” was introduced to reflect the impact of rainfall on CSO. The neural network method was utilized to establish a relationship model between rainfall characteristics and rainfall-CSO contribution rate (R2 > 0.94). Sensitivity analysis and model visualization techniques were used to reveal the relationship between rainfall characteristics and rainfall-CSO contribution rate. Significant differences in contribution rates across rainfall patterns were observed (p = 0.012), indicating a strong association with rainfall characteristics. Specifically, rainfall patterns with a higher proportion of peak period precipitation correspond to greater CSO contribution rates. Within each rainfall pattern, rainfall depth was identified as the most critical factor affecting the CSO contribution rate, followed by rainfall duration, with average sensitivity indices of 0.580 and 0.274, respectively. The peak-period rainfall ratio had a minimal impact on the results, with an average sensitivity index of just 0.024. Furthermore, the study noted that, variations in CSO contribution rates across different patterns intensified with increasing rainfall depth, while the impact of rainfall duration diminished with longer durations. This research provides a methodical approach for quantitatively analyzing the relationship between rainfall characteristics and CSO contribution rates, facilitating rapid assessments of CSO conditions and informing urban planning and drainage management decisions.
{"title":"The impact of rainfall characteristics on combined sewer overflows in wet weather","authors":"Hao Wang, Zijan Wang, Pengfei Zeng, Zilong Liu, Bin Chen, Jinjun Zhou","doi":"10.1016/j.jhydrol.2026.134951","DOIUrl":"https://doi.org/10.1016/j.jhydrol.2026.134951","url":null,"abstract":"Combined sewer overflows (CSO) pose risks to both water quality and public health. However, the impact of varying rainfall characteristics on CSO remains unclear. This study aims to assess and analyze the correlations between CSO and rainfall characteristics, specifically rainfall depth, duration, and the proportion of rainfall occurring during peak periods. The objective is to identify and elucidate the impacts of each main rainfall characteristic on the frequency of CSO. To conduct this analysis, rainfall data spanning 71 years were divided into 1,435 rainfall events and categorized into five representative rainfall patterns based on their temporal distribution. A hydraulic model of sewer network was employed to simulate the CSO results under different rainfall patterns. An indicator termed “rainfall-CSO contribution rate” was introduced to reflect the impact of rainfall on CSO. The neural network method was utilized to establish a relationship model between rainfall characteristics and rainfall-CSO contribution rate (R<ce:sup loc=\"post\">2</ce:sup> > 0.94). Sensitivity analysis and model visualization techniques were used to reveal the relationship between rainfall characteristics and rainfall-CSO contribution rate. Significant differences in contribution rates across rainfall patterns were observed (p = 0.012), indicating a strong association with rainfall characteristics. Specifically, rainfall patterns with a higher proportion of peak period precipitation correspond to greater CSO contribution rates. Within each rainfall pattern, rainfall depth was identified as the most critical factor affecting the CSO contribution rate, followed by rainfall duration, with average sensitivity indices of 0.580 and 0.274, respectively. The peak-period rainfall ratio had a minimal impact on the results, with an average sensitivity index of just 0.024. Furthermore, the study noted that, variations in CSO contribution rates across different patterns intensified with increasing rainfall depth, while the impact of rainfall duration diminished with longer durations. This research provides a methodical approach for quantitatively analyzing the relationship between rainfall characteristics and CSO contribution rates, facilitating rapid assessments of CSO conditions and informing urban planning and drainage management decisions.","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"38 1","pages":""},"PeriodicalIF":6.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957292","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The two-layer fluid model is one of the most efficient methods for simulating the landslide into the water process. However, the landslide equations in the model overlook the viscosity coefficient and make certain assumptions in their derivation from the NS equations, which results in the simulator being less able to describe different flow-state disasters, or solid structures, and failing to simulate landslide-induced surges effectively while considering their interaction with the prevention structure within a unified fluid framework. Therefore, based on our previous study of the numerical origin of the inviscid defect in the SH model and the enhanced SH model proposed, from the perspective of fluid–fluid interaction, an improved flow-flow coupling model suitable for landslide surge and its prevention is proposed, which not only characterizes landslides with different flow states and describing solid control structures, but also efficiently realizes the analysis of landslide evolution, surge generation and its interaction with control structure piles under a unified fluid framework. In addition, for the cross-scale impact of landslide and surge on the pile and extensive calculation, the discrete solution of the new flow-flow model employs a limited volume method combined with the local mesh refinement technology. By setting multiple sets of examples, the study further carries out the simulation results of the new model. It improves the ability to calculate the interaction between landslides and pile-water bodies. It clarifies the preventive and treatment effects of different space layouts of pile groups on surges, proving that this technology is excellent for risk assessment. Finally, through the preview of the Sichuan Bageduzhai landslide-induced surge Incident, the study confirms that the improved model supports the reliability of disaster prediction and structural interactions, as well as the development of disaster computing and prevention technology.
{"title":"Efficient simulation of landslide-induced surges and control effects of different position piles based on an improved flow-flow model","authors":"Zhangqing Wang, Yong Wu, Yongjie Zhao, Yingpeng Wang, Siming He, Xinpo Li, Lei Zhu","doi":"10.1016/j.jhydrol.2026.134944","DOIUrl":"https://doi.org/10.1016/j.jhydrol.2026.134944","url":null,"abstract":"The two-layer fluid model is one of the most efficient methods for simulating the landslide into the water process. However, the landslide equations in the model overlook the viscosity coefficient and make certain assumptions in their derivation from the NS equations, which results in the simulator being less able to describe different flow-state disasters, or solid structures, and failing to simulate landslide-induced surges effectively while considering their interaction with the prevention structure within a unified fluid framework. Therefore, based on our previous study of the numerical origin of the inviscid defect in the SH model and the enhanced SH model proposed, from the perspective of fluid–fluid interaction, an improved flow-flow coupling model suitable for landslide surge and its prevention is proposed, which not only characterizes landslides with different flow states and describing solid control structures, but also efficiently realizes the analysis of landslide evolution, surge generation and its interaction with control structure piles under a unified fluid framework. In addition, for the cross-scale impact of landslide and surge on the pile and extensive calculation, the discrete solution of the new flow-flow model employs a limited volume method combined with the local mesh refinement technology. By setting multiple sets of examples, the study further carries out the simulation results of the new model. It improves the ability to calculate the interaction between landslides and pile-water bodies. It clarifies the preventive and treatment effects of different space layouts of pile groups on surges, proving that this technology is excellent for risk assessment. Finally, through the preview of the Sichuan Bageduzhai landslide-induced surge Incident, the study confirms that the improved model supports the reliability of disaster prediction and structural interactions, as well as the development of disaster computing and prevention technology.","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"10 1","pages":""},"PeriodicalIF":6.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957164","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1016/j.jhydrol.2026.134955
Bowen Li, Tingting Yang, Shibin Liu, Meng Yao, Jun Dong
The uneven distribution of colloidal Mg(OH)2 in heterogeneous porous media poses a significant challenge for its effective application in groundwater remediation. To address this issue, this study introduces a novel approach using xanthan gum (XG) as a viscosity modifier to enhance the migration of colloidal Mg(OH)2 into low permeability zone. Results indicate that XG is highly compatible with colloidal Mg(OH)2, viscosity modified colloidal Mg(OH)2 (VMC-Mg(OH)2) exhibits significant shear thinning properties. The increased viscosity effectively reduces the deposition of colloidal Mg(OH)2 and facilitates its return to groundwater. With the addition of XG to the system, the collision efficiency (η) between colloidal Mg(OH)2 and porous media decreased from 0.00865 to 0.00142, while the attachment efficiency (α) was reduced from 0.4858 to 0.1038. These variations notably enhance the migration performance of colloidal Mg(OH)2, with C/C0 increasing from 0.12 to 0.94. The incorporation of XG also leads to a substantial increase in colloidal Mg(OH)2 sweep efficiency in low permeability zone, rising from 53.6 % to 92.5 % as the XG concentration increased from 0 mg/L to 200 mg/L. Moreover, the simulation of collision efficiency (η) and attachment efficiency (α) accurately predicts the migration of VMC-Mg(OH)2 in heterogeneous porous media, with a maximum error of 5.39 %. These findings highlight the significant potential of VMC-Mg(OH)2 as a reactive reagent for remediating contamination in low permeability zone
{"title":"Migration of viscosity modified colloidal Mg(OH)2 in heterogeneous porous media: experiment and model simulation","authors":"Bowen Li, Tingting Yang, Shibin Liu, Meng Yao, Jun Dong","doi":"10.1016/j.jhydrol.2026.134955","DOIUrl":"https://doi.org/10.1016/j.jhydrol.2026.134955","url":null,"abstract":"The uneven distribution of colloidal Mg(OH)<ce:inf loc=\"post\">2</ce:inf> in heterogeneous porous media poses a significant challenge for its effective application in groundwater remediation. To address this issue, this study introduces a novel approach using xanthan gum (XG) as a viscosity modifier to enhance the migration of colloidal Mg(OH)<ce:inf loc=\"post\">2</ce:inf> into low permeability zone. Results indicate that XG is highly compatible with colloidal Mg(OH)<ce:inf loc=\"post\">2</ce:inf>, viscosity modified colloidal Mg(OH)<ce:inf loc=\"post\">2</ce:inf> (VMC-Mg(OH)<ce:inf loc=\"post\">2</ce:inf>) exhibits significant shear thinning properties. The increased viscosity effectively reduces the deposition of colloidal Mg(OH)<ce:inf loc=\"post\">2</ce:inf> and facilitates its return to groundwater. With the addition of XG to the system, the collision efficiency (<ce:italic>η</ce:italic>) between colloidal Mg(OH)<ce:inf loc=\"post\">2</ce:inf> and porous media decreased from 0.00865 to 0.00142, while the attachment efficiency (<ce:italic>α</ce:italic>) was reduced from 0.4858 to 0.1038. These variations notably enhance the migration performance of colloidal Mg(OH)<ce:inf loc=\"post\">2</ce:inf>, with <ce:italic>C/C<ce:inf loc=\"post\">0</ce:inf></ce:italic> increasing from 0.12 to 0.94. The incorporation of XG also leads to a substantial increase in colloidal Mg(OH)<ce:inf loc=\"post\">2</ce:inf> sweep efficiency in low permeability zone, rising from 53.6 % to 92.5 % as the XG concentration increased from 0 mg/L to 200 mg/L. Moreover, the simulation of collision efficiency (<ce:italic>η</ce:italic>) and attachment efficiency (<ce:italic>α</ce:italic>) accurately predicts the migration of VMC-Mg(OH)<ce:inf loc=\"post\">2</ce:inf> in heterogeneous porous media, with a maximum error of 5.39 %. These findings highlight the significant potential of VMC-Mg(OH)<ce:inf loc=\"post\">2</ce:inf> as a reactive reagent for remediating contamination in low permeability zone","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"82 1","pages":""},"PeriodicalIF":6.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957290","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1016/j.jhydrol.2026.134958
Chunye Hu, Fan Zhang, Jin Li, Xiaolei Liu, Fei Xing, Renzhi Li, Hao Wu, Heyu Yu, Ya Ping Wang
Subaqueous deltas worldwide are increasingly threatened by erosion, driven by the dual pressures of intensified storms and reduced fluvial sediment supply. An abandoned river delta, devoid of sediment input from its watershed, offers an ideal end-member case for investigating delta erosion processes. This study provides direct observational evidence of storm-driven sediment dynamics in such a sediment-starved delta, based on in situ measurements during both typical weather conditions and winter storms on the abandoned Yellow River Delta, China. During storms, fluid mud layers, wave-induced seabed liquefaction, and gravity flows were directly observed. Fluid mud developed through two mechanisms: wave-induced liquefaction combined with strong bed shear stress; and suspended sediment settling during slack water under weak waves. To enable a more systematic assessment of gravity flow dynamics, we refined a previous analytical model by incorporating additional transport processes. Using this model, we quantified, for the first time under storm conditions in a sediment-starved delta, that gravity flows contributed to ∼ 49% of the total sediment transport leaving the 10-m isobath region of the subaqueous delta, despite occurring during only ∼ 7% of the 18-day observation. These results highlight that storm-driven gravity flows can develop and play a pivotal role in controlling sediment balance even in sediment-starved subaqueous deltas. Our findings provide new insights into sediment dynamics of sediment-starved deltas under intensified storm forcing and offer a framework for understanding their long-term morphological evolution.
{"title":"Sediment transport mechanisms in sediment-starved subaqueous deltas: insights from storm-induced gravity flows","authors":"Chunye Hu, Fan Zhang, Jin Li, Xiaolei Liu, Fei Xing, Renzhi Li, Hao Wu, Heyu Yu, Ya Ping Wang","doi":"10.1016/j.jhydrol.2026.134958","DOIUrl":"https://doi.org/10.1016/j.jhydrol.2026.134958","url":null,"abstract":"Subaqueous deltas worldwide are increasingly threatened by erosion, driven by the dual pressures of intensified storms and reduced fluvial sediment supply. An abandoned river delta, devoid of sediment input from its watershed, offers an ideal end-member case for investigating delta erosion processes. This study provides direct observational evidence of storm-driven sediment dynamics in such a sediment-starved delta, based on in situ measurements during both typical weather conditions and winter storms on the abandoned Yellow River Delta, China. During storms, fluid mud layers, wave-induced seabed liquefaction, and gravity flows were directly observed. Fluid mud developed through two mechanisms: wave-induced liquefaction combined with strong bed shear stress; and suspended sediment settling during slack water under weak waves. To enable a more systematic assessment of gravity flow dynamics, we refined a previous analytical model by incorporating additional transport processes. Using this model, we quantified, for the first time under storm conditions in a sediment-starved delta, that gravity flows contributed to ∼ 49% of the total sediment transport leaving the 10-m isobath region of the subaqueous delta, despite occurring during only ∼ 7% of the 18-day observation. These results highlight that storm-driven gravity flows can develop and play a pivotal role in controlling sediment balance even in sediment-starved subaqueous deltas. Our findings provide new insights into sediment dynamics of sediment-starved deltas under intensified storm forcing and offer a framework for understanding their long-term morphological evolution.","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"34 1","pages":""},"PeriodicalIF":6.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957248","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1016/j.jhydrol.2026.134940
Xiaochao Wang , Yu Xiao , Chongli Di
Accurate measurement of river surface velocity is essential for hydrological research, hydraulic engineering, flood forecasting, and hydrological monitoring. Although non-contact imaging technologies offer promising alternatives to traditional contact-based methods, existing approaches often lack robustness under diverse environmental conditions, especially with unstable results under different flow velocities and varying densities of tracer features. To address these challenges, this study proposes a novel RemoteWaterNet, a lightweight deep learning framework for robust and efficient remote river surface velocimetry. The framework integrates simplified image preprocessing with a pre-trained optical flow model (SEA-RAFT) to extract initial flow features, followed by iterative refinement and unit conversion to estimate real-world flow velocities. Extensive training and fine-tuning across multiple datasets demonstrate that RemoteWaterNet achieves superior generalization under different environmental conditions. Experimental validation on eight field datasets shows that the newly proposed RemoteWaterNet improves accuracy by 26.33% compared to existing methods, with significant advantages in scenarios with diverse environments. Additionally, RemoteWaterNet reduces model parameters by 92.38%, making it highly suitable for real-time environmental monitoring. This study significantly advances the application of deep learning-based optical flow models in hydrological measurements and offers valuable new insights for the practical monitoring and management of river systems.
{"title":"RemoteWaterNet: a lightweight and efficient algorithm for remote river surface velocimetry","authors":"Xiaochao Wang , Yu Xiao , Chongli Di","doi":"10.1016/j.jhydrol.2026.134940","DOIUrl":"10.1016/j.jhydrol.2026.134940","url":null,"abstract":"<div><div>Accurate measurement of river surface velocity is essential for hydrological research, hydraulic engineering, flood forecasting, and hydrological monitoring. Although non-contact imaging technologies offer promising alternatives to traditional contact-based methods, existing approaches often lack robustness under diverse environmental conditions, especially with unstable results under different flow velocities and varying densities of tracer features. To address these challenges, this study proposes a novel RemoteWaterNet, a lightweight deep learning framework for robust and efficient remote river surface velocimetry. The framework integrates simplified image preprocessing with a pre-trained optical flow model (SEA-RAFT) to extract initial flow features, followed by iterative refinement and unit conversion to estimate real-world flow velocities. Extensive training and fine-tuning across multiple datasets demonstrate that RemoteWaterNet achieves superior generalization under different environmental conditions. Experimental validation on eight field datasets shows that the newly proposed RemoteWaterNet improves accuracy by 26.33% compared to existing methods, with significant advantages in scenarios with diverse environments. Additionally, RemoteWaterNet reduces model parameters by 92.38%, making it highly suitable for real-time environmental monitoring. This study significantly advances the application of deep learning-based optical flow models in hydrological measurements and offers valuable new insights for the practical monitoring and management of river systems.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"667 ","pages":"Article 134940"},"PeriodicalIF":6.3,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957115","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1016/j.jhydrol.2026.134947
Yuankang Ye, Feng Gao, Shaoqing Zhang, Chang Liu
Radar-based precipitation nowcasting plays a vital role in short-term hydrometeorological forecasting and water resource management. Existing modeling methodology typically simplifies precipitation nowcasting to a task of spatiotemporal sequence prediction based on radar echo reflectivity data. However, the reliance on unimodal reflectivity data including intensity-only information restricts the model’s ability to characterize the phase evolution and dynamic process of hydrometeor particles, ultimately leading to insufficient extrapolation accuracy. This study breaks through the conventional unimodal data paradigm, aiming to capture the complex dynamic evolutionary features of hydrometeor particles. We integrate radar echo reflectivity and four additional physical parameters of hydrometeor particles into a deep learning framework and propose a novel Physics-Informed multimodal Echo Extrapolation neural network (PIEE). Furthermore, we systematically investigate the individual contributions of each physical parameter to the accuracy of radar echo extrapolation. Specifically, PIEE adopts a three-stage structure. First, a multimodal encoder with a dual-branch attention-based fusion strategy to capture diverse physical signals. Second, a novel gated spatiotemporal self-attention module is designed for deep feature extraction. Finally, the decoding stage generates the extrapolated radar echoes. Experimental results on a real multimodal radar echo dataset show that the proposed model demonstrates superior performance in two aspects. First, under a unimodal baseline architecture, the PIEE model clearly outperforms the comparison model. Second, after fusing multiple physical parameters, the PIEE achieves significant improvements in all the evaluated metrics, especially in the CSI and HSS metrics for the high echo intensity region ( 40 dBZ), with improvements of up to 24.2% and 20.3%, respectively. Furthermore, systematic ablation experiments on physical parameters quantify the effects of different combination methods on extrapolation accuracy, highlight the potential of physics-informed, multimodal deep learning approaches in improving short-term hydrological prediction accuracy, with implications for flood forecasting, early warning systems, and hydrometeorological risk management at catchment scales.
{"title":"Improving precipitation nowcasting via multiphysical parameter fusion in radar echo extrapolation","authors":"Yuankang Ye, Feng Gao, Shaoqing Zhang, Chang Liu","doi":"10.1016/j.jhydrol.2026.134947","DOIUrl":"https://doi.org/10.1016/j.jhydrol.2026.134947","url":null,"abstract":"Radar-based precipitation nowcasting plays a vital role in short-term hydrometeorological forecasting and water resource management. Existing modeling methodology typically simplifies precipitation nowcasting to a task of spatiotemporal sequence prediction based on radar echo reflectivity data. However, the reliance on unimodal reflectivity data including intensity-only information restricts the model’s ability to characterize the phase evolution and dynamic process of hydrometeor particles, ultimately leading to insufficient extrapolation accuracy. This study breaks through the conventional unimodal data paradigm, aiming to capture the complex dynamic evolutionary features of hydrometeor particles. We integrate radar echo reflectivity and four additional physical parameters of hydrometeor particles into a deep learning framework and propose a novel Physics-Informed multimodal Echo Extrapolation neural network (PIEE). Furthermore, we systematically investigate the individual contributions of each physical parameter to the accuracy of radar echo extrapolation. Specifically, PIEE adopts a three-stage structure. First, a multimodal encoder with a dual-branch attention-based fusion strategy to capture diverse physical signals. Second, a novel gated spatiotemporal self-attention module is designed for deep feature extraction. Finally, the decoding stage generates the extrapolated radar echoes. Experimental results on a real multimodal radar echo dataset show that the proposed model demonstrates superior performance in two aspects. First, under a unimodal baseline architecture, the PIEE model clearly outperforms the comparison model. Second, after fusing multiple physical parameters, the PIEE achieves significant improvements in all the evaluated metrics, especially in the CSI and HSS metrics for the high echo intensity region (<ce:math altimg=\"si56.svg\"></ce:math> 40 dBZ), with improvements of up to 24.2% and 20.3%, respectively. Furthermore, systematic ablation experiments on physical parameters quantify the effects of different combination methods on extrapolation accuracy, highlight the potential of physics-informed, multimodal deep learning approaches in improving short-term hydrological prediction accuracy, with implications for flood forecasting, early warning systems, and hydrometeorological risk management at catchment scales.","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"34 1","pages":""},"PeriodicalIF":6.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957163","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Reservoirs are constructed by damming rivers to impound actual runoff (Ra) from upstream drainage areas, thereby securing water supply and buffering against hydrological extremes. A warming climate and intensifying human interventions are reshaping the water cycle, ultimately affecting the generation and distribution of Ra. Yet, Ra-related variations in reservoir drainage areas and their underlying drivers remain largely unknown, challenging sustainable reservoir management. Here, we combine the precise drainage boundaries of 913 large reservoirs in China with ISIMIP3a runoff simulations to bridge this gap. We analyze trends in the magnitude and variability of four Ra-related indicators, namely actual runoff volume (Qa), standardized runoff index (SRI-12), drought frequency (Df), and pluvial frequency (Pf). Interannual magnitude trends in Qa, SRI-12, Df, and Pf display consistent spatial patterns, with 60–70% of reservoirs exhibiting drying trends concentrated in the eastern belt of the Hu Line. In contrast, interannual variability trends display inconsistent patterns, with 20–50% of reservoirs exhibiting enhancing variability. Applying the Inter-Sectoral Impact Model Intercomparison Project Phase 3a (ISIMIP3a) attribution framework, we attribute these trends to anthropogenic climate change (ACC), natural climate variability (NCV), and human water and land management (HWLM). Attribution analyses reveal that ACC dominates the magnitude trends, with mean contribution rates of 75–85%. Conversely, NCV dominates variability trends in Qa and Df, HWLM primarily drives SRI-12 variability, and NCV and ACC jointly dominate Pf variability. Given the uncertainties and limitations in ISIMIP3a-based trend and attribution analyses, we advocate incorporating observational constraints to improve assessment accuracy, thereby informing adaptive reservoir management under changing environmental conditions.
水库是通过在河流上筑坝来截住上游流域的实际径流（Ra），从而确保供水和缓冲水文极端情况。气候变暖和人类干预的加剧正在重塑水循环，最终影响Ra的生成和分布。然而，水库排水区域的ra相关变化及其潜在驱动因素在很大程度上仍然未知，这对可持续的水库管理提出了挑战。在这里，我们将中国913个大型水库的精确排水边界与ISIMIP3a径流模拟相结合，以弥补这一差距。我们分析了实际径流量（Qa）、标准化径流指数（SRI-12）、干旱频率（Df）和降雨频率（Pf）这四个与降水相关的指标的幅度和变异趋势。Qa、SRI-12、Df、Pf的年际变化趋势具有一致的空间格局，60-70%的储层呈现干燥趋势，集中在胡线东段。相反，年际变化趋势不一致，20-50%的储层年际变化增强。应用ISIMIP3a （Inter-Sectoral Impact Model Intercomparison Project Phase 3a）归因框架，我们将这些趋势归因于人为气候变化（ACC）、自然气候变率（NCV）和人类水土管理（HWLM）。归因分析表明，ACC主导了震级趋势，平均贡献率为75 ~ 85%。相反，NCV主导Qa和Df的变异性趋势，HWLM主要驱动SRI-12变异性，NCV和ACC共同主导Pf变异性。考虑到基于isimip3的趋势和归因分析的不确定性和局限性，我们建议结合观测约束来提高评估精度，从而为变化环境条件下的适应性水库管理提供信息。
{"title":"Attribution of interannual runoff magnitude and variability in China’s large reservoir drainage areas using global hydrological Models","authors":"Xinyu Li, Kaiwen Wang, Qiuyu Luo, Guan Wang, Yu Lu, Haining Jiang, Jiamiao Yu, Changming Liu, Xiaomang Liu","doi":"10.1016/j.jhydrol.2026.134953","DOIUrl":"https://doi.org/10.1016/j.jhydrol.2026.134953","url":null,"abstract":"Reservoirs are constructed by damming rivers to impound actual runoff (<ce:italic>R<ce:inf loc=\"post\">a</ce:inf></ce:italic>) from upstream drainage areas, thereby securing water supply and buffering against hydrological extremes. A warming climate and intensifying human interventions are reshaping the water cycle, ultimately affecting the generation and distribution of <ce:italic>R<ce:inf loc=\"post\">a</ce:inf></ce:italic>. Yet, <ce:italic>R<ce:inf loc=\"post\">a</ce:inf></ce:italic>-related variations in reservoir drainage areas and their underlying drivers remain largely unknown, challenging sustainable reservoir management. Here, we combine the precise drainage boundaries of 913 large reservoirs in China with ISIMIP3a runoff simulations to bridge this gap. We analyze trends in the magnitude and variability of four <ce:italic>R<ce:inf loc=\"post\">a</ce:inf></ce:italic>-related indicators, namely actual runoff volume (<ce:italic>Q<ce:inf loc=\"post\">a</ce:inf></ce:italic>), standardized runoff index (<ce:italic>SRI-12</ce:italic>), drought frequency (<ce:italic>D<ce:inf loc=\"post\">f</ce:inf></ce:italic>), and pluvial frequency (<ce:italic>P<ce:inf loc=\"post\">f</ce:inf></ce:italic>). Interannual magnitude trends in <ce:italic>Q<ce:inf loc=\"post\">a</ce:inf></ce:italic>, <ce:italic>SRI-12</ce:italic>, <ce:italic>D<ce:inf loc=\"post\">f</ce:inf></ce:italic>, and <ce:italic>P<ce:inf loc=\"post\">f</ce:inf></ce:italic> display consistent spatial patterns, with 60–70% of reservoirs exhibiting drying trends concentrated in the eastern belt of the Hu Line. In contrast, interannual variability trends display inconsistent patterns, with 20–50% of reservoirs exhibiting enhancing variability. Applying the Inter-Sectoral Impact Model Intercomparison Project Phase 3a (ISIMIP3a) attribution framework, we attribute these trends to anthropogenic climate change (ACC), natural climate variability (NCV), and human water and land management (HWLM). Attribution analyses reveal that ACC dominates the magnitude trends, with mean contribution rates of 75–85%. Conversely, NCV dominates variability trends in <ce:italic>Q<ce:inf loc=\"post\">a</ce:inf></ce:italic> and <ce:italic>D<ce:inf loc=\"post\">f</ce:inf></ce:italic>, HWLM primarily drives <ce:italic>SRI-12</ce:italic> variability, and NCV and ACC jointly dominate <ce:italic>P<ce:inf loc=\"post\">f</ce:inf></ce:italic> variability. Given the uncertainties and limitations in ISIMIP3a-based trend and attribution analyses, we advocate incorporating observational constraints to improve assessment accuracy, thereby informing adaptive reservoir management under changing environmental conditions.","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"96 1","pages":""},"PeriodicalIF":6.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957291","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1016/j.jhydrol.2026.134952
David A. Benson, Savannah Miller, Joel Barber
Previous studies found that best-fit parameters from variable-rate pumping tests show very high correlation between both aquifer and non-linear head loss parameters. This correlation implies non-uniqueness of parameter sets and potentially similar non-uniqueness of predicted well behavior. The resulting uncertainty may lead to a wide range of potential future pumping costs, making it difficult to evaluate how up-front investments to improve well efficiency will ultimately influence overall costs. To address this and assess how filter-pack material affects predictive pumping costs, we implement (in Python) a Bayesian sampling of the posterior distribution of aquifer and non-linear well loss parameters using Markov-chain Monte Carlo (MCMC) methods. We also implement a novel discharge correction to account for wellbore storage that significantly changes efficiency estimation. These methods are needed to analyze a number of single-well step-drawdown tests from the city of Castle Rock, Colorado, including wells constructed with both sand and glass-bead filter packs. Some wells were constructed with the more expensive glass-bead filter packs with the intent of saving pumping costs due to higher well efficiency. The resulting parameter distributions are highly correlated and non-Gaussian. Forward simulations using an ensemble of these parameter sets, along with a cost function to predict future pumping costs weighed against initial capital costs, show that there were no statistically significant improvements to well efficiency by using glass bead filter packs.
{"title":"Markov-chain Monte Carlo estimation of aquifer parameters, non-linear well losses, and probable costs of water extraction","authors":"David A. Benson, Savannah Miller, Joel Barber","doi":"10.1016/j.jhydrol.2026.134952","DOIUrl":"https://doi.org/10.1016/j.jhydrol.2026.134952","url":null,"abstract":"Previous studies found that best-fit parameters from variable-rate pumping tests show very high correlation between both aquifer and non-linear head loss parameters. This correlation implies non-uniqueness of parameter sets and potentially similar non-uniqueness of predicted well behavior. The resulting uncertainty may lead to a wide range of potential future pumping costs, making it difficult to evaluate how up-front investments to improve well efficiency will ultimately influence overall costs. To address this and assess how filter-pack material affects predictive pumping costs, we implement (in Python) a Bayesian sampling of the posterior distribution of aquifer and non-linear well loss parameters using Markov-chain Monte Carlo (MCMC) methods. We also implement a novel discharge correction to account for wellbore storage that significantly changes efficiency estimation. These methods are needed to analyze a number of single-well step-drawdown tests from the city of Castle Rock, Colorado, including wells constructed with both sand and glass-bead filter packs. Some wells were constructed with the more expensive glass-bead filter packs with the intent of saving pumping costs due to higher well efficiency. The resulting parameter distributions are highly correlated and non-Gaussian. Forward simulations using an ensemble of these parameter sets, along with a cost function to predict future pumping costs weighed against initial capital costs, show that there were no statistically significant improvements to well efficiency by using glass bead filter packs.","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"29 1","pages":""},"PeriodicalIF":6.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957293","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号