Pub Date : 2026-01-27DOI: 10.1016/j.jhydrol.2026.135032
He Zhao , Xiaorong Liu , Xinzhong Du , Qiuliang Lei , Hongbin Liu
Excess phosphorus (P) loss has contributed to the eutrophication of freshwater. Determining critical source areas (CSAs) for P loss is an important prerequisite for the cost-effective control and management of regional water environmental pollution. The changes of distribution for P CSAs in response to seasonal hydrological conditions have been revealed, but the variations of P CSAs under different storm events remain understudied. Here, we investigated the impacts of storm events on P CSAs identification in an agricultural headwater watershed in southwest China based on the SWAT model integrated with the cumulative pollution load curve method. The results showed that TP loss dynamics in the watershed are predominantly event-driven. Storm-event-based analysis identified more subbasins as CSAs and potential risk areas compared to CSAs identification on the annual average scale. Additionally, as storm event levels increased, the number of CSAs, their TP load intensity, and their contribution to total pollution significantly rose. In addition to storm events, other key factors influencing P loss and the spatial distribution of CSAs include slope, soil type, and land use, among which cultivated lands with anthropogenic fertilizer inputs were primary contributors to P loss. The study confirms the dominant role of storm events in P export and underscores the importance of considering storm events to accurately identify the P CSAs at the watershed scale.
{"title":"Impacts of storm events on watershed phosphorus critical source area identification based on SWAT model","authors":"He Zhao , Xiaorong Liu , Xinzhong Du , Qiuliang Lei , Hongbin Liu","doi":"10.1016/j.jhydrol.2026.135032","DOIUrl":"10.1016/j.jhydrol.2026.135032","url":null,"abstract":"<div><div>Excess phosphorus (P) loss has contributed to the eutrophication of freshwater. Determining critical source areas (CSAs) for P loss is an important prerequisite for the cost-effective control and management of regional water environmental pollution. The changes of distribution for P CSAs in response to seasonal hydrological conditions have been revealed, but the variations of P CSAs under different storm events remain understudied. Here, we investigated the impacts of storm events on P CSAs identification in an agricultural headwater watershed in southwest China based on the SWAT model integrated with the cumulative pollution load curve method. The results showed that TP loss dynamics in the watershed are predominantly event-driven. Storm-event-based analysis identified more subbasins as CSAs and potential risk areas compared to CSAs identification on the annual average scale. Additionally, as storm event levels increased, the number of CSAs, their TP load intensity, and their contribution to total pollution significantly rose. In addition to storm events, other key factors influencing P loss and the spatial distribution of CSAs include slope, soil type, and land use, among which cultivated lands with anthropogenic fertilizer inputs were primary contributors to P loss. The study confirms the dominant role of storm events in P export and underscores the importance of considering storm events to accurately identify the P CSAs at the watershed scale.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135032"},"PeriodicalIF":6.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146056301","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 : 2026-01-27DOI: 10.1016/j.jhydrol.2026.135033
Taotao Zhang , Bin Luan , Chi Zhang , Yang Xiao , Chen Xu , Jingxiu Wu
Particle transport over urban road surfaces is a critical contributor to urban stormwater pollution. However, existing particle transport models predominantly focus on individual rainfall effect, rendering the coupled effects of raindrop erosion and runoff effect inadequately understood. This study proposed a novel physically based transport model by introducing a two-layer conceptual model that incorporates both raindrop erosion and runoff effect. The model assumed that particles in the static control layer can enter the upper runoff layer through raindrop erosion and runoff effect. A series of laboratory experiments were performed to validate the proposed model, taking into account the conditions of with and without raindrop erosion, as well as mixed particles. Results demonstrated that model predictions exhibited excellent agreement with the experimental observations under various conditions. Critical runoff energy exhibited strong positive correlation with median particle size, while the efficiency coefficient increased with decreasing particle size. Raindrop erosion increased the peak concentration and transport rate by 3 to 4 fold, and significantly raised the final wash-off fraction of large-sized particles. Sensitivity analysis revealed that Manning’s coefficient and control layer depth attenuate peak transport rates, whereas slope length and detachment coefficient enhance them. This study advanced the understanding of particle transport mechanisms in urban environments and provided a foundation for developing distributed catchment models.
{"title":"A physically-based model for particle transport over urban road surface with consideration of raindrop erosion and runoff effect","authors":"Taotao Zhang , Bin Luan , Chi Zhang , Yang Xiao , Chen Xu , Jingxiu Wu","doi":"10.1016/j.jhydrol.2026.135033","DOIUrl":"10.1016/j.jhydrol.2026.135033","url":null,"abstract":"<div><div>Particle transport over urban road surfaces is a critical contributor to urban stormwater pollution. However, existing particle transport models predominantly focus on individual rainfall effect, rendering the coupled effects of raindrop erosion and runoff effect inadequately understood. This study proposed a novel physically based transport model by introducing a two-layer conceptual model that incorporates both raindrop erosion and runoff effect. The model assumed that particles in the static control layer can enter the upper runoff layer through raindrop erosion and runoff effect. A series of laboratory experiments were performed to validate the proposed model, taking into account the conditions of with and without raindrop erosion, as well as mixed particles. Results demonstrated that model predictions exhibited excellent agreement with the experimental observations under various conditions. Critical runoff energy exhibited strong positive correlation with median particle size, while the efficiency coefficient increased with decreasing particle size. Raindrop erosion increased the peak concentration and transport rate by 3 to 4 fold, and significantly raised the final wash-off fraction of large-sized particles. Sensitivity analysis revealed that Manning’s coefficient and control layer depth attenuate peak transport rates, whereas slope length and detachment coefficient enhance them. This study advanced the understanding of particle transport mechanisms in urban environments and provided a foundation for developing distributed catchment models.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135033"},"PeriodicalIF":6.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072672","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 : 2026-01-27DOI: 10.1016/j.jhydrol.2026.135046
Qiutong Yu, Bryan Tolson
Rainfall-runoff models based on Long Short-Term Memory (LSTM) networks typically require extensive datasets for training. While increasing the number of watersheds generally improves model performance, the appropriate temporal scope of training data—especially under potential non-stationarity from climate change—remains unclear, and the importance of data recency remains underexplored. This study investigates whether decades of historical data are necessary when training with large numbers of watersheds, and whether shorter, more recent training periods can match the performance of extended records. Using hydrometeorological data from 1374 North American watersheds (spanning 1950–2023), we systematically evaluate the effect of data recency on LSTM performance through three complementary experimental designs: (1) backward-expanding training periods (fixed recent data + progressively older blocks), (2) forward-expanding periods (fixed old data + progressively newer blocks), and (3) sliding-window periods (fixed-length windows moving forward). Results reveal that newer data universally improves predictions, while older data contributes marginally or negatively to model performance. Notably, the benefit of increasing the number of watersheds is conditional on data recency, as Prediction in Ungauged Basins (PUB) performance improves most significantly with recent data. These findings provide the empirical evidence for the importance of data recency, demonstrating that data recency—not just data volume—is critical for LSTM-based streamflow prediction, underscoring the need for recency-aware training strategies.
{"title":"How much historical data do we need? The role of data recency and training period length in LSTM-based rainfall-runoff modeling","authors":"Qiutong Yu, Bryan Tolson","doi":"10.1016/j.jhydrol.2026.135046","DOIUrl":"10.1016/j.jhydrol.2026.135046","url":null,"abstract":"<div><div>Rainfall-runoff models based on Long Short-Term Memory (LSTM) networks typically require extensive datasets for training. While increasing the number of watersheds generally improves model performance, the appropriate temporal scope of training data—especially under potential non-stationarity from climate change—remains unclear, and the importance of data recency remains underexplored. This study investigates whether decades of historical data are necessary when training with large numbers of watersheds, and whether shorter, more recent training periods can match the performance of extended records. Using hydrometeorological data from 1374 North American watersheds (spanning 1950–2023), we systematically evaluate the effect of data recency on LSTM performance through three complementary experimental designs: (1) backward-expanding training periods (fixed recent data + progressively older blocks), (2) forward-expanding periods (fixed old data + progressively newer blocks), and (3) sliding-window periods (fixed-length windows moving forward). Results reveal that newer data universally improves predictions, while older data contributes marginally or negatively to model performance. Notably, the benefit of increasing the number of watersheds is conditional on data recency, as Prediction in Ungauged Basins (PUB) performance improves most significantly with recent data. These findings provide the empirical evidence for the importance of data recency, demonstrating that data recency—not just data volume—is critical for LSTM-based streamflow prediction, underscoring the need for recency-aware training strategies.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135046"},"PeriodicalIF":6.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072673","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 : 2026-01-27DOI: 10.1016/j.jhydrol.2026.135037
Song Miao , Heng Lyu , Huaiqing Liu , Yuan Li , Xiaoguang Ruan , Wenyu Liu
Terrestrial dissolved organic matter (tDOM), a key regulator of lake ecosystems and the global carbon cycle, significantly influences underwater light regimes, thermal structure, and biogeochemical cycles. However, quantifying the fluorescence intensity and spatiotemporal patterns of tDOM is a persistent challenge for traditional monitoring techniques, which creates a critical gap in regional carbon budget estimations and the understanding of large-scale ecosystem effects. To overcome this limitation, a novel remote sensing algorithm was developed to retrieve tDOM, which is characterized by the fluorescence intensity of fulvic acid-like and humic acid-like components, based on the absorption characteristics of colored dissolved organic matter (CDOM). This approach was subsequently applied to assess the spatiotemporal dynamics and long-term trends of tDOM across Yangtze River Delta (YRD) lakes from 1986 to 2024. The robustness of the algorithm was confirmed using both independent in-situ and satellite-ground match-up datasets (median absolute percentage error < 30%). The satellite-derived tDOM revealed significant spatiotemporal heterogeneity in the YRD. Trend analysis indicated that tDOM in the most lakes (63.46%) significantly increased, whereas it decreased in 17.31% and remained stable in 19.23% of the lakes. Furthermore, the random forest model results revealed that temperature and wind speed were the key drivers influencing the tDOM dynamics, acting synergistically with anthropogenic pressure which was associated with elevated tDOM levels. This study provides a feasible algorithm for large-scale tDOM monitoring and highlights the critical influences of climate change and anthropogenic activities on the estimation of lake carbon storage in rapidly developing regions.
{"title":"Four decades of terrestrial dissolved organic matter in lakes across the Yangtze River Delta: Spatiotemporal dynamics and driving factors","authors":"Song Miao , Heng Lyu , Huaiqing Liu , Yuan Li , Xiaoguang Ruan , Wenyu Liu","doi":"10.1016/j.jhydrol.2026.135037","DOIUrl":"10.1016/j.jhydrol.2026.135037","url":null,"abstract":"<div><div>Terrestrial dissolved organic matter (tDOM), a key regulator of lake ecosystems and the global carbon cycle, significantly influences underwater light regimes, thermal structure, and biogeochemical cycles. However, quantifying the fluorescence intensity and spatiotemporal patterns of tDOM is a persistent challenge for traditional monitoring techniques, which creates a critical gap in regional carbon budget estimations and the understanding of large-scale ecosystem effects. To overcome this limitation, a novel remote sensing algorithm was developed to retrieve tDOM, which is characterized by the fluorescence intensity of fulvic acid-like and humic acid-like components, based on the absorption characteristics of colored dissolved organic matter (CDOM). This approach was subsequently applied to assess the spatiotemporal dynamics and long-term trends of tDOM across Yangtze River Delta (YRD) lakes from 1986 to 2024. The robustness of the algorithm was confirmed using both independent in-situ and satellite-ground match-up datasets (median absolute percentage error < 30%). The satellite-derived tDOM revealed significant spatiotemporal heterogeneity in the YRD. Trend analysis indicated that tDOM in the most lakes (63.46%) significantly increased, whereas it decreased in 17.31% and remained stable in 19.23% of the lakes. Furthermore, the random forest model results revealed that temperature and wind speed were the key drivers influencing the tDOM dynamics, acting synergistically with anthropogenic pressure which was associated with elevated tDOM levels. This study provides a feasible algorithm for large-scale tDOM monitoring and highlights the critical influences of climate change and anthropogenic activities on the estimation of lake carbon storage in rapidly developing regions.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135037"},"PeriodicalIF":6.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146071694","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 : 2026-01-27DOI: 10.1016/j.jhydrol.2026.135026
Xiaofen Bai , Chen Lin , Shenmin Wang , Junfeng Xiong , Jinduo Xu , Kun Xue , DanHua Ma , Yijun Tong , Jianchun Chen , Wenzhuo Cui
The correlation between the spatial differentiation of pond-related ecosystem services value (PRESV) and geographic environments is essential for management decisions of ponds. However, the types of ponds and geographic environments are more diverse across a larger scale region, necessitating a systematic approach on PRESV assessment. Therefore, we conducted targeted research on PRESV across a large-scale region to reveal the factors influencing the spatial differentiation of PRESV. To address the lack of a systematic approach on PRESV, this study proposes an optimization method for the aquatic production (AP) through a refined index structure and differentiated yield calculations based on the InVEST model. Notably, the optimized AP evaluation method significantly achieved an R2 exceeding 0.6 and demonstrated a marked improvement over the existing method. Consequently, we generated spatial distribution data for the PRESV in China. The results indicate that: (1) The AP exhibits significant spatial differentiation corresponding to the surrounding geographic environments, summarized as “more in the south and less in the north, more in the east and less in the west”. (2) The differentiated surrounding water bodies, elevation and precipitation are potential drivers of the AP function to be disentangled in a way not possible at small watershed scale. (3) Since areas with high PRESV are all associated with high-density AP, ponds should be incorporated into watershed management.
{"title":"Ponds greatly influence watershed ecosystem services value: An optimized InVEST model for large-scale","authors":"Xiaofen Bai , Chen Lin , Shenmin Wang , Junfeng Xiong , Jinduo Xu , Kun Xue , DanHua Ma , Yijun Tong , Jianchun Chen , Wenzhuo Cui","doi":"10.1016/j.jhydrol.2026.135026","DOIUrl":"10.1016/j.jhydrol.2026.135026","url":null,"abstract":"<div><div>The correlation between the spatial differentiation of pond-related ecosystem services value (PRESV) and geographic environments is essential for management decisions of ponds. However, the types of ponds and geographic environments are more diverse across a larger scale region, necessitating a systematic approach on PRESV assessment. Therefore, we conducted targeted research on PRESV across a large-scale region to reveal the factors influencing the spatial differentiation of PRESV. To address the lack of a systematic approach on PRESV, this study proposes an optimization method for the aquatic production (AP) through a refined index structure and differentiated yield calculations based on the InVEST model. Notably, the optimized AP evaluation method significantly achieved an R<sup>2</sup> exceeding 0.6 and demonstrated a marked improvement over the existing method. Consequently, we generated spatial distribution data for the PRESV in China. The results indicate that: (1) The AP exhibits significant spatial differentiation corresponding to the surrounding geographic environments, summarized as “more in the south and less in the north, more in the east and less in the west”. (2) The differentiated surrounding water bodies, elevation and precipitation are potential drivers of the AP function to be disentangled in a way not possible at small watershed scale. (3) Since areas with high PRESV are all associated with high-density AP, ponds should be incorporated into watershed management.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135026"},"PeriodicalIF":6.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072674","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 : 2026-01-27DOI: 10.1016/j.jhydrol.2026.135038
Yanzi Cai , Xia Li , Guogui Chen , Yiliang Xie , Yujia Zhai , Tian Xie , Mingyu Wang , Fang Gao , Meili Feng , Ying Man , Baoshan Cui
Estuarine deltas are global biodiversity hotspots where hydrological connectivity plays a pivotal role in shaping ecosystem structure and functioning. However, hydrological connectivity in many estuaries has been profoundly altered by freshwater regulation and tidal intrusion, and its effects on community assembly remain poorly understood. Using the Pearl River Estuary as a model system, this study integrates flow-based functional and path-based structural connectivity to investigate how hydrological connectivity mediates phytoplankton community assembly under contrasting hydrographic regimes. Hydrological connectivity exhibited pronounced spatiotemporal heterogeneity, with functional connectivity showing clear spatial clustering during the wet season but declining sharply and becoming fragmented in the dry season. Structural connectivity presented a distinct spatial pattern, concentrated in the central and eastern delta. Phytoplankton communities exhibited marked seasonal contrasts: during the wet season, river-dominated conditions enhanced environmental filtering and species turnover, resulting in high spatial heterogeneity (βsor: 0.65 ± 0.11), while in the dry season, tidal intrusion and reduced freshwater inflow intensified environmental constraints and dispersal limitation, leading to stress-tolerant assemblages and elevated total beta diversity (βsor: 0.70 ± 0.15). Hydrological connectivity significantly modulated the balance between environmental and spatial processes, displaying season-dependent, nonlinear threshold-like responses along connectivity gradients. By explicitly distinguishing functional and structural connectivity, these findings advance a process-based framework for interpreting hydrological–ecological interactions in sluice-regulated distributary estuarine networks, with broader relevance for understanding biodiversity dynamics under anthropogenic pressures.
{"title":"Seasonal shifts in phytoplankton community assembly mediated by hydrological connectivity in a sluice-regulated distributary estuarine network—as a case in the Pearl River Estuary","authors":"Yanzi Cai , Xia Li , Guogui Chen , Yiliang Xie , Yujia Zhai , Tian Xie , Mingyu Wang , Fang Gao , Meili Feng , Ying Man , Baoshan Cui","doi":"10.1016/j.jhydrol.2026.135038","DOIUrl":"10.1016/j.jhydrol.2026.135038","url":null,"abstract":"<div><div>Estuarine deltas are global biodiversity hotspots where hydrological connectivity plays a pivotal role in shaping ecosystem structure and functioning. However, hydrological connectivity in many estuaries has been profoundly altered by freshwater regulation and tidal intrusion, and its effects on community assembly remain poorly understood. Using the Pearl River Estuary as a model system, this study integrates flow-based functional and path-based structural connectivity to investigate how hydrological connectivity mediates phytoplankton community assembly under contrasting hydrographic regimes. Hydrological connectivity exhibited pronounced spatiotemporal heterogeneity, with functional connectivity showing clear spatial clustering during the wet season but declining sharply and becoming fragmented in the dry season. Structural connectivity presented a distinct spatial pattern, concentrated in the central and eastern delta. Phytoplankton communities exhibited marked seasonal contrasts: during the wet season, river-dominated conditions enhanced environmental filtering and species turnover, resulting in high spatial heterogeneity (βsor: 0.65 ± 0.11), while in the dry season, tidal intrusion and reduced freshwater inflow intensified environmental constraints and dispersal limitation, leading to stress-tolerant assemblages and elevated total beta diversity (βsor: 0.70 ± 0.15). Hydrological connectivity significantly modulated the balance between environmental and spatial processes, displaying season-dependent, nonlinear threshold-like responses along connectivity gradients. By explicitly distinguishing functional and structural connectivity, these findings advance a process-based framework for interpreting hydrological–ecological interactions in sluice-regulated distributary estuarine networks, with broader relevance for understanding biodiversity dynamics under anthropogenic pressures.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135038"},"PeriodicalIF":6.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072680","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 : 2026-01-27DOI: 10.1016/j.jhydrol.2026.135044
Wenhao Fu , Yuntian Chen , Zhongzheng Wang , Qiang Zheng , Dongxiao Zhang
Data assimilation of complex geological models in subsurface flow (e.g., channelized models) remains a challenging task due to the non-Gaussian characteristics and high dimensionality of model parameters. Effective parameterization is essential to preserving plausible geological features while reducing model complexity, which in turn improves the efficiency and accuracy of ensemble-based data assimilation workflow. Advances in generative modeling, particularly diffusion models, hold great promise for capturing complex geological structures, but their iterative sampling procedures are prohibitively time-consuming for data assimilation. To address this limitation, we propose principal component analysis-guided adversarial diffusion (PCA-GAD), a single-step diffusion-based generative framework that combines principal component analysis (PCA) with adversarial training. PCA-GAD first constructs a full-grid facies approximation from PCA coefficients to capture dominant channel geometries, and then refines this coarse seed in a single inference pass with a diffusion network trained using an adversarial loss. By leveraging PCA guidance, PCA-GAD preserves key geological structures while substantially reducing parameter dimensionality. Furthermore, in contrast to multi-step diffusion sampling, the adversarial accelerated framework requires only a single inference pass, dramatically shortening sampling time and thereby enabling much faster ensemble-based data assimilation. We demonstrate the effectiveness of the proposed method on two representative cases: a binary facies model and a bimodal permeability model. Compared with conventional parameterization techniques, PCA-GAD better reproduces complex geological structures and achieves more accurate data assimilation. This approach provides a reliable and effective way for uncertainty quantification in subsurface flow modeling under complex geological conditions.
{"title":"Parameterization of complex geological models with PCA‑guided adversarial diffusion for ensemble data assimilation","authors":"Wenhao Fu , Yuntian Chen , Zhongzheng Wang , Qiang Zheng , Dongxiao Zhang","doi":"10.1016/j.jhydrol.2026.135044","DOIUrl":"10.1016/j.jhydrol.2026.135044","url":null,"abstract":"<div><div>Data assimilation of complex geological models in subsurface flow (e.g., channelized models) remains a challenging task due to the non-Gaussian characteristics and high dimensionality of model parameters. Effective parameterization is essential to preserving plausible geological features while reducing model complexity, which in turn improves the efficiency and accuracy of ensemble-based data assimilation workflow. Advances in generative modeling, particularly diffusion models, hold great promise for capturing complex geological structures, but their iterative sampling procedures are prohibitively time-consuming for data assimilation. To address this limitation, we propose principal component analysis-guided adversarial diffusion (PCA-GAD), a single-step diffusion-based generative framework that combines principal component analysis (PCA) with adversarial training. PCA-GAD first constructs a full-grid facies approximation from PCA coefficients to capture dominant channel geometries, and then refines this coarse seed in a single inference pass with a diffusion network trained using an adversarial loss. By leveraging PCA guidance, PCA-GAD preserves key geological structures while substantially reducing parameter dimensionality. Furthermore, in contrast to multi-step diffusion sampling, the adversarial accelerated framework requires only a single inference pass, dramatically shortening sampling time and thereby enabling much faster ensemble-based data assimilation. We demonstrate the effectiveness of the proposed method on two representative cases: a binary facies model and a bimodal permeability model. Compared with conventional parameterization techniques, PCA-GAD better reproduces complex geological structures and achieves more accurate data assimilation. This approach provides a reliable and effective way for uncertainty quantification in subsurface flow modeling under complex geological conditions.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135044"},"PeriodicalIF":6.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072677","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}
Real-time hydrodynamic prediction in plain river networks is essential for flood early warning and optimal water resources scheduling. However, traditional numerical models are computationally expensive, and purely data-driven approaches often lack generalization and physical consistency. Physics-informed neural networks (PINNs) embed physics but must be retrained when boundary conditions change, whereas operator learning methods depend heavily on large amounts of observational data that are often unavailable. To overcome these limitations, a physics-informed river operator network (PI-RONet) is proposed. In this framework, a configurable operator network (DeepONet/MIONet) serves as the encoder to flexibly handle dynamic and heterogeneous boundary conditions of varying complexity; a PINN acts as the decoder, embedding the Saint-Venant equations and junction equations into the loss function to ensure physical consistency and reduce dependence on dense observations. To support model training, an automated Python-RAS workflow is developed to generate large-scale high-fidelity datasets across multiple flow conditions. The model is validated through two cases of increasing complexity. Under 1000 unsteady test boundary conditions in Case 2, PI-RONet achieves high accuracy (), with 99.5% of conditions for discharge and 90.9% for water level. Compared with numerical models, PI-RONet reduces multi-scenario simulation time from hours to seconds, accelerating by nearly 5000 times. Ablation experiments demonstrate that MIONet improves water level prediction accuracy by reducing RMSE by 19.5% and shortens training time by 10.7%. Lastly, the trained model is deployed as an interactive web platform, RivONet, demonstrating its potential applications in real-time decision support and intelligent water management.
{"title":"Configurable physics-informed operator network for real-time multi-scenario hydrodynamics in river networks","authors":"Bingxi Huang , Huiming Zhang , Sheng Jiang , Hongwu Tang , Saiyu Yuan , Xiao Luo , Zhaohui Chen","doi":"10.1016/j.jhydrol.2026.135042","DOIUrl":"10.1016/j.jhydrol.2026.135042","url":null,"abstract":"<div><div>Real-time hydrodynamic prediction in plain river networks is essential for flood early warning and optimal water resources scheduling. However, traditional numerical models are computationally expensive, and purely data-driven approaches often lack generalization and physical consistency. Physics-informed neural networks (PINNs) embed physics but must be retrained when boundary conditions change, whereas operator learning methods depend heavily on large amounts of observational data that are often unavailable. To overcome these limitations, a physics-informed river operator network (PI-RONet) is proposed. In this framework, a configurable operator network (DeepONet/MIONet) serves as the encoder to flexibly handle dynamic and heterogeneous boundary conditions of varying complexity; a PINN acts as the decoder, embedding the Saint-Venant equations and junction equations into the loss function to ensure physical consistency and reduce dependence on dense observations. To support model training, an automated Python-RAS workflow is developed to generate large-scale high-fidelity datasets across multiple flow conditions. The model is validated through two cases of increasing complexity. Under 1000 unsteady test boundary conditions in Case 2, PI-RONet achieves high accuracy (<span><math><mrow><msup><mi>R</mi><mn>2</mn></msup><mo>≥</mo><mn>0.8</mn></mrow></math></span>), with 99.5% of conditions for discharge and 90.9% for water level. Compared with numerical models, PI-RONet reduces multi-scenario simulation time from hours to seconds, accelerating by nearly 5000 times. Ablation experiments demonstrate that MIONet improves water level prediction accuracy by reducing RMSE by 19.5% and shortens training time by 10.7%. Lastly, the trained model is deployed as an interactive web platform, RivONet, demonstrating its potential applications in real-time decision support and intelligent water management.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135042"},"PeriodicalIF":6.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072675","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 : 2026-01-27DOI: 10.1016/j.jhydrol.2026.135050
Femin C. Varghese, Sakila Saminathan, Subhasis Mitra
Compound drought–hot events (CDHEs) represent some of the most damaging climate extremes, yet their persistence, migration, and large-scale synchronization remain insufficiently understood. This study employs a three-dimensional (3D) event-based framework with the Blended Dry and Hot Index (BDHI) to detect and characterize global CDHEs for 1951–2022. Uncertainties associated with the choice of drought and compound indices are assessed across aridity zones, followed by an investigation of intercontinental synchronization patterns through complex network (CN) analysis and their underlying physical drivers. We identify 809 3D-CDHEs globally, with most events lasting 3–6 months and affecting 1–10 million km2. However, large and persistent events (>12 months) have become increasingly common in recent decades. Amazon, Central Africa, Northern North America, and parts of Russia and China, shows higher event frequency designating these regions as recurrent CDHE hotspots. Case studies of the 2015–2017 South American and 2022 Eurasian events provide new perspectives on their spatiotemporal evolution, including patterns of growth, contraction, and severity change. Uncertainty analysis demonstrates that CDHE detection is highly sensitive to index selection, with uncertainties amplifying under recent warming. CN analysis uncovers intercontinental synchronization hubs in the Amazon, West Africa, the Mediterranean, Southeast Asia, and North Asia, many historically associated with El Niño–Southern Oscillation (ENSO) episodes. Results further indicate a weakening ENSO–CDHE relationship and the growing dominance of anthropogenic warming as the primary driver of compound extremes. These findings offer novel insights into global CDHE dynamics and their changing drivers under a warming climate.
{"title":"Global compound drought–hot events: insights from a 3D-event based framework, intercontinental synchronization, and the evolving influence of climatic drivers","authors":"Femin C. Varghese, Sakila Saminathan, Subhasis Mitra","doi":"10.1016/j.jhydrol.2026.135050","DOIUrl":"10.1016/j.jhydrol.2026.135050","url":null,"abstract":"<div><div>Compound drought–hot events (CDHEs) represent some of the most damaging climate extremes, yet their persistence, migration, and large-scale synchronization remain insufficiently understood. This study employs a three-dimensional (3D) event-based framework with the Blended Dry and Hot Index (BDHI) to detect and characterize global CDHEs for 1951–2022. Uncertainties associated with the choice of drought and compound indices are assessed across aridity zones, followed by an investigation of intercontinental synchronization patterns through complex network (CN) analysis and their underlying physical drivers. We identify 809 3D-CDHEs globally, with most events lasting 3–6 months and affecting 1–10 million km<sup>2</sup>. However, large and persistent events (>12 months) have become increasingly common in recent decades. Amazon, Central Africa, Northern North America, and parts of Russia and China, shows higher event frequency designating these regions as recurrent CDHE hotspots. Case studies of the 2015–2017 South American and 2022 Eurasian events provide new perspectives on their spatiotemporal evolution, including patterns of growth, contraction, and severity change. Uncertainty analysis demonstrates that CDHE detection is highly sensitive to index selection, with uncertainties amplifying under recent warming. CN analysis uncovers intercontinental synchronization hubs in the Amazon, West Africa, the Mediterranean, Southeast Asia, and North Asia, many historically associated with El Niño–Southern Oscillation (ENSO) episodes. Results further indicate a weakening ENSO–CDHE relationship and the growing dominance of anthropogenic warming as the primary driver of compound extremes. These findings offer novel insights into global CDHE dynamics and their changing drivers under a warming climate.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135050"},"PeriodicalIF":6.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072678","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 : 2026-01-26DOI: 10.1016/j.jhydrol.2026.135025
Leifang Li , Taihua Wang , Ge Li , Han Cheng , Shulei Zhang , Dawen Yang
Monitoring and managing water quality is challenged by pronounced short-term fluctuations in key indicators, driven by both hydro-meteorological variability and human activities. Although substantial research has focused on water-quality modeling, effectively characterizing week-scale dynamics using weekly data remains limited. To address this gap, this study develops weekly water-quality prediction models for ten sub-basins of the Yangtze River using seven machine-learning algorithms. The models demonstrate robust performance, with Nash–Sutcliffe efficiency (NSE) values ranging from 0.41 to 0.74, and are used to reconstruct weekly time series of four key indicators, pH, DO, CODMn, and NH3-N, from 2012 to 2018. Among the evaluated algorithms, Random Forest Regression performs best, achieving average NSE values of 0.56–0.81 despite a 45% missing-data rate. Shapley Additive Explanations (SHAP) analysis reveals that hydro-meteorological and anthropogenic drivers contribute comparably to weekly water-quality variability, with hydro-meteorological and agricultural sectors emerging as dominant controls in water quality. About half of the sub-basins show water quality exceedances mainly associated with hydro-meteorological conditions, whereas the remainder are influenced primarily by anthropogenic pressures, underscoring the need for differentiated and sector-targeted management strategies. By reconstructing missing records and quantifying driver importance, this study provides a practical framework for interpreting short-term water quality fluctuations and diagnosing weekly exceedance events. The proposed approach supports risk-informed water quality management in large river basins and is transferable to other monsoonal and temperate catchments.
{"title":"Hydro-meteorological factors and human activities contribute comparably to weekly water quality dynamics in the Yangtze River Basin","authors":"Leifang Li , Taihua Wang , Ge Li , Han Cheng , Shulei Zhang , Dawen Yang","doi":"10.1016/j.jhydrol.2026.135025","DOIUrl":"10.1016/j.jhydrol.2026.135025","url":null,"abstract":"<div><div>Monitoring and managing water quality is challenged by pronounced short-term fluctuations in key indicators, driven by both hydro-meteorological variability and human activities. Although substantial research has focused on water-quality modeling, effectively characterizing week-scale dynamics using weekly data remains limited. To address this gap, this study develops weekly water-quality prediction models for ten sub-basins of the Yangtze River using seven machine-learning algorithms. The models demonstrate robust performance, with Nash–Sutcliffe efficiency (NSE) values ranging from 0.41 to 0.74, and are used to reconstruct weekly time series of four key indicators, pH, DO, CODMn, and NH<sub>3</sub>-N, from 2012 to 2018. Among the evaluated algorithms, Random Forest Regression performs best, achieving average NSE values of 0.56–0.81 despite a 45% missing-data rate. Shapley Additive Explanations (SHAP) analysis reveals that hydro-meteorological and anthropogenic drivers contribute comparably to weekly water-quality variability, with hydro-meteorological and agricultural sectors emerging as dominant controls in water quality. About half of the sub-basins show water quality exceedances mainly associated with hydro-meteorological conditions, whereas the remainder are influenced primarily by anthropogenic pressures, underscoring the need for differentiated and sector-targeted management strategies. By reconstructing missing records and quantifying driver importance, this study provides a practical framework for interpreting short-term water quality fluctuations and diagnosing weekly exceedance events. The proposed approach supports risk-informed water quality management in large river basins and is transferable to other monsoonal and temperate catchments.</div></div>","PeriodicalId":362,"journal":{"name":"Journal of Hydrology","volume":"668 ","pages":"Article 135025"},"PeriodicalIF":6.3,"publicationDate":"2026-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146048582","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