Urban floods pose a serious threat to human life, property safety, and socio-economic development. Urban flood modeling plays a vital role in disaster prevention and resilient infrastructure planning. Traditionally, one-dimensional (1D) drainage network and two-dimensional (2D) surface flow coupling models based on physical processes have been widely used to simulate urban flooding. However, such models require substantial computational resources and long execution times, making them unsuitable for real-time flood forecasting over large areas. This study proposes a CNN–Transformer model as an efficient surrogate for physical models in rapid urban flood forecasting. The model is trained using high-resolution gridded precipitation data and inundation simulation outputs generated by a calibrated 1D/2D hydrodynamic model, covering 5,475,655 grid cells within the study area. By integrating the spatial features of rainfall and the temporal dynamics of flood evolution, the model enables fast and reliable predictions. The study shows that the CNN-Transformer model closely matches the physical model in predicting flood depth, inundation extent, and flood volume, achieving a mean relative error below 10%, NSE ≥ 0.956, and R2 > 0.958. Additionally, the model reduces computation time by approximately 287.4–308.8 times compared to the physical model, meeting the real-time requirements of flood emergency response. These results indicate that the proposed method significantly improves computational efficiency while maintaining high predictive accuracy. It contributes to the development of smarter and more sustainable urban decision-making systems and holds strong practical value for real-time flood risk management, infrastructure planning, and climate resilience enhancement.

Rapid flood forecasting and early warning represent key non-structural measures currently employed to mitigate flood disasters induced by short-term intense rainfall. Enhancing the accuracy, lead time, and generalization capability of flood forecasting in small watersheds remains a critical challenge. In this study, we propose a novel rapid flood forecasting model that integrates Graph Neural Networks (GNN) with Transformer architectures, leveraging deep learning techniques. The model incorporates static physical attributes of surface grids and utilizes hydrodynamic model simulation results to guide learning. Specifically, the GNN component captures spatial relationship weights among grids and their neighboring cells, while the Transformer component models rainfall time series to dynamically track flood evolution processes. Experimental results demonstrate that the proposed GNN-Transformer-based watershed flood forecasting model outperforms traditional numerical models. The predicted results exhibit a mean relative error of no more than 15

High-arsenic groundwater in the middle-lower Yellow River Basin, widely used for agriculture, poses significant carcinogenic risks, yet its environmental characteristics and sources remain unclear. This study characterized shallow high-arsenic groundwater (As > 10 µg·L− 1), quantified population-specific health risks, and identified sources using PCA-PMF. Elevated As concentrations were observed in alluvial fan frontal swales and Yellow River burst fan sediments (20–45 m depth), reaching 128 µg·L− 1, with 40.8 Weakly alkaline reducing environment and high HCO3- environment are conducive to the formation of high-arsenic groundwater, accompanied by the release of Fe2+ and NH4+. A comprehensive assessment of the health risks of arsenic in different geomorphological features to humans was conducted. Utilized PCA-PMF for rigorous quantification of arsenic sources in shallow aquifers. It reveals the contribution process of environmental factors to health risk tiers.

Abstract Meningeal lymphatic vessels (mLVs) are vital for brain waste clearance, making them a promising therapeutic target. However, effective modulation strategies for mLVs with translational potential remain underdeveloped. Here, we develop a low-intensity focused ultrasound (LIFU) strategy that precisely targets the vault cranial meninges to non-invasively facilitate mLVs drainage. Using models of Alzheimer’s disease (AD) and aging, we demonstrate that this approach promotes CSF drainage, prevents cognitive decline, and reduces pathological biomarkers. Mechanistically, RNA sequencing combined with calcium imaging in vitro reveals that LIFU activates the Piezo1 ion channel in lymphatic endothelial cells, whereas pharmacological inhibition of Piezo1 abolishes LIFU’s therapeutic effects. Compliant with FDA safety guidelines, this LIFU protocol demonstrates strong clinical translatability. If its efficacy is clinically confirmed, LIFU offers a promising therapy for neurodegenerative diseases triggered by waste accumulation.

In sudden river pollution incidents, rapid prediction of pollutant migration is critical for emergency response. This study proposes a seconds-level prediction model based on a feature-enhanced convolutional neural network-long short-term memory (CNN-LSTM) framework to forecast suspended-solids pollution in the Shizi River reach affected by combined sewer overflow (CSO) discharges. Training and validation datasets were generated using a coupled one-dimensional-two-dimensional hydrodynamic-water-quality model driven by overflow hydrographs derived from rainfall events with different return periods. The framework integrates multi-scale convolutional feature extraction with recursive temporal modeling to predict spatial suspended-solids concentration fields, affected areas, and event-scale pollutant loads. The model captures the nonlinear dependence of pollutant load on flow and source concentration across a wide concentration range, achieving coefficient of determination (R2) values consistently above 0.90. Spatial agreement remains high under low-to-moderate concentrations, and intersection over union (IoU) exceeds 0.98 in the late stage of high-concentration cases. Upon training completion, the model generates event-level predictions within seconds, providing rapid early warning and decision support for emergency management of sudden river suspended solids pollution.

Drought-flood abrupt alternation has become an important compound hydrological extreme, highlighting the need to understand urban runoff and inundation responses. This study develops a coupled 1D-2D urban inundation model that accounts for infiltration differences among underlying surfaces. Multiple drought and rainfall scenarios are designed to investigate runoff generation and inundation dynamics during the transition from drought to heavy rainfall. A threshold identification method for drainage and flood control is developed based on the "depth-duration" approach. Results show that mild to extreme drought generally reduces runoff and inundation, with diminishing marginal effects as drought intensity increases, whereas exceptional drought leads to a risk reversal from reduction to amplification; The influence of antecedent drought varies with rainfall return period: runoff is mainly controlled by initial soil moisture under small return periods but dominated by rainfall under large return periods; Inundation recession is controlled by a dual mechanism: return period governs the recession rate, and antecedent drought regulates drainage efficiency; Drainage and inundation control thresholds increase under mild to extreme drought but decrease under exceptional drought. These findings reveal nonlinear urban flood responses to drought-flood abrupt alternation and provide insights for flood risk assessment and drainage system resilience.

Frequent extreme rainfall events have intensified urban inundation, and the prolonged recession of accumulated water has become increasingly prominent, severely undermining urban flood resilience. Existing real-time assessment methods for urban inundation recession are inefficient and incapable of accurately supporting emergency drainage operations and post-disaster recovery. To address this gap, this study develops a dynamic assessment framework aimed at clarifying the inundation recession mechanisms, evaluating urban flood resilience, and providing scientific guidance for emergency dispatch strategies. The framework centers on multidimensional assessment indices and systematically considers the impact of rainfall structure. Using a typical flood-prone comprehensive urban area as a case study, this research quantitatively analyzes the spatiotemporal characteristics of the water recession process and the urban resilience. The results show that: (1) When the rainfall return period exceeds 50 years, the capacity of pipe network tends to saturate, and urban inundation recession relies more on surface runoff. Beyond a 100-year return period, the system’s recession response slows down nonlinearly, requiring emergency surface drainage pathways and measures to alleviate inundation. (2) Rainfall peak structure critically controls the recession rhythm. Rainfall with rear-peak and lower concentration significantly delay the high-pressure drainage window of the urban system. (3) The recession of pipe network generally lags behind that of area-wide surface by 0.3–0.5 h. Although spatial non-uniform rainfall mitigates overall drainage loads, it exacerbates local drainage bottlenecks. (4) Taking A3, A8 and A13 for examples, the highest node overflow densities occur at 1.15 h, 1.42 h, and 1.82 h, respectively, consistently between the corresponding recession start time of area-wide surface and pipe network. This spatiotemporal clustering precisely diagnoses the transition from surface-water recession to system-wide drainage recovery, while also pinpointing the pipe network nodes that hinder recession. It suggests that node overflow density can serve as an effective indicator for identifying critical recession points and resilience weaknesses within the urban system. These findings provide new quantitative evidence for urban drainage systems’ resilience assessment and intervention. A multidimensional and dynamic framework was proposed for recession identification and resilience assessment of urban inundation,based on a 1D-2D coupled urban inundation full-cycle model. Shifts the focus from static peak-state resistance to processoriented recovery resilience. Focusing on the urban inundation recession process as a novel perspective to investigate how urban flood responds to rainstorms characteristics. Discovered the temporal clustering of node overflow density between surface and pipe network recession phases, proposing PNNO as a key indicator of urban inundation recession and resilience, and demonstrating its practical effectiveness.

Urban water environments are facing increasingly severe pollution challenges, and rigorous numerical models have become indispensable for mitigating urban water pollution. Building on the "Gridding + GPU acceleration + Dynamic Link Library (DLL)" approach, this study develops an advanced coupled model that integrates (i) twodimensional (2D) surface-water hydrodynamics and water-quality transport, (ii) 2D non-point-source pollutant (NPSP) build-up and wash-off, and (iii) one-dimensional (1D) pipe-network drainage and pollutant discharge, thereby enabling integrated simulation of the urban "Source-Plant-Network-River" (SPNR) system. The model employs high-resolution structured grids and a spatiotemporal flux scheme for multi-component pollutants in surface runoff, allowing accurate representation of NPSP wash-off and transport driven by coupled hydrologicalhydrodynamic processes. DLL-based bidirectional coupling is implemented to dynamically link 2D surface processes and 1D pipe network hydraulics and water quality processes while reducing distortions in parameter transfer across modules. GPU acceleration, together with optimized water-quality flux computations and removal of redundant operations, significantly improves computational efficiency. The model is applied to the main urban area of Changzhi City under three spatially distributed rainfall scenarios. Performance is evaluated against observations of inundation depth, zoned drainage/sewage discharge, and combined sewer overflow (CSO) flow and water quality. The results show that the Nash-Sutcliffe efficiency (NSE) exceeds 0.7 for inundation depths at four flood-prone locations and for flow and pollutant concentrations at three representative drainage-outfall zones and four CSO outfalls. On an RTX 3070 workstation, the optimized model completes a 7.22 h simulation on 8,484,785 uniform structured grids coupled with 32,982 pipe-network nodes in 8.15 h, reducing runtime by 12.6% compared with the pre-optimization model. The proposed modeling framework is robust and efficient, offering strong potential for high-precision integrated simulations of urban water environments from source to receiving waters, as well as for evaluation, forecasting, early warning, and comprehensive water-environment management from a watershed perspective.

In recent years, the increasing frequency of heavy rainfall events has intensified urban pluvial-flood risks, particularly in low-lying, infrastructure-deficient small and medium-sized cities. However, the spatiotemporal mitigation effects of drainage systems-especially within critical urban functional areas such as road networks-remain insufficiently explored. This study develops an integrated hydrodynamic framework combining the two-dimensional GAST model and the one-dimensional SWMM model to simulate 18 rainfall scenarios (return periods of 1a-100a) in Taocheng District, Hengshui, China. Two hydraulic conditions were assessed: with drainage (WN) and without drainage (NN), enabling a comparative analysis of surface inundation regulation, road inundation reduction, and flood-risk redistribution. Results show that under the WN scenario, the drainage system reduced peak surface and road inundation volumes by up to 50.92% and 69.81%, respectively, while advancing the road peak-inundation time by 24-36 min. This temporal decoupling from peak rainfall intensity lowers the likelihood of flow synchronization. In terms of road-flood risk, Level 4 (high-risk) grids decreased by over 65% during the 100a event, with reductions also observed across Levels 1-3-particularly at intersections, low-lying areas, and inlet-dense zones. These findings highlight the critical mitigation role of urban drainage systems during both moderate and extreme rainfall. They underscore the necessity of incorporating drainage infrastructure into urban flood models and risk assessments, supporting the development of climate-adaptive drainage strategies and resilient stormwater management for vulnerable cities.

Motor and cognitive dysfunctions are the most prevalent functional impairments in patients with stroke. The cerebellum is involved in regulating cognitive and motor functions in the human body, but the impact of cerebellar transcranial direct current stimulation (tDCS) on cognitive and motor functions in patients with stroke is currently unclear. This study aimed to evaluate the efficacy of cerebellar tDCS on cognitive and motor dysfunctions in patients with stroke. This randomized, single-blind, sham-controlled prospective study recruited patients with cerebral hemisphere stroke who had cognitive and motor deficits between November, 2018 and November, 2020. Participants were allocated into active tDCS (A-tDCS) or sham (S-tDCS) group. Both groups received standardized rehabilitation therapy and cognitive training along with cerebellar active tDCS or sham tDCS for 20 min, once a day, 5 days a week with 4-week treatments. Primary outcomes were the change from baseline to 4 weeks after randomization in the Montreal Cognitive Assessment (MoCA) and Berg Balance Scale (BBS). The clinically meaningful response rates in MoCA and BBS score, as well as score changes in Fugl-Meyer Assessment Scale for Lower Extremities (FMA-LE) and Barthel Index (BI) were also assessed. A total of 74 eligible participants (mean [SD] age, 56.4 [13.2] years; 16 female [21.6%]) were randomly assigned to A-tDCS (n = 37) or S-tDCS (n = 37) group and included in intention-to-treat analysis. At 4 weeks after randomization, compared to S-tDCS group, A-tDCS group showed significant higher MoCA score changes (6.78 [95% CI, 5.77–7.8] vs. 5.05 [4.03–6.08]; mean difference [95% CI], 1.73 [0.36–3.10]; P = 0.004) and BBS score changes (17.30 [13.59-21.00] vs. 11.86 [7.86–15.87]; 5.43 [0.23–10.64]; P = 0.037). In A-tDCS group, the clinically meaningful response rates in MoCA (78.4% vs. 54.1%; odds ratio [95% CI], 3.08 [1.12–8.50]; P = 0.015) and BBS (78.4% vs. 48.6%; 3.90 [1.37–11.09]; P = 0.011) scores were higher. The FMA-LE score changes showed no significant difference between the groups at 4 weeks after randomization (6.97 [5.13–8.81] vs. 6.57 [4.67–8.47]; 0.41 [-2.12-2.93]; P = 0.517). However, at 12 weeks after randomization, the FMA-LE score changes in the A-tDCS group were significantly higher than those in the S-tDCS group (12.03 [9.29–14.77] vs. 8.62 [6.40-10.84]; 3.41 [0.04–6.77]; P = 0.023). Four weeks of cerebellar tDCS in this trail improved cognitive and balance dysfunction after stroke. However, the improvement in lower limb motor function of hemiplegic patients was not significant, and only positive effects were observed during follow-up. These findings may present a novel targeted neuromodulation approach with potential for dual-domain functional enhancement in post-stroke rehabilitation. Registry Chictr.org.cn, ChiCTR2200061838, Registration date: 3 July 2022.

ObjectiveHydraulic control structures such as weirs, culverts, pump stations, and sluice gates play a crucial role in regulating river flow and mitigating flood risks. However, simulating these structures accurately in 1D river network models faces challenges including diverse structure types, complex flow conditions, and abrupt hydraulic parameter changes near the structures. Existing numerical methods often involve complex iterations or struggle to capture dynamic regulation processes, limiting the reliability of flood simulation results. This study aims to develop a straightforward and efficient simulation algorithm to address these issues, enabling accurate and stable modeling of various hydraulic control structures while ensuring computational efficiency and adaptability to real-world complex river networks.MethodsThe algorithm was developed based on the source term method and integrated into a 1D river network hydrodynamic model solved using the Godunov scheme. The 1D shallow water equations (Saint-Venant equations) were adopted as the control equations, with hydraulic control structures treated as source terms to account for their flow regulation effects. A novel artificial area method was proposed for junction processing, where each junction was modeled as a "reservoir" with a fixed base area calculated using adjacent river reach parameters. This eliminated the need for iterative calculations, simplifying junction water level updates through mass conservation equations. For specific structures: Weirs: Flow was computed using the weir flow formula incorporating discharge coefficient, weir crest width, and head over weir. Culverts: Different formulas were applied for free-flow and submerged flow conditions, considering culvert shape, height, and upstream-downstream head difference. Pump stations: Two processing modes were implemented: boundary adjustment based on pump curves for internal pump stations, and direct flow addition/subtraction for stations connected to external water bodies. Sluice gates: A virtual gate calculation zone was established between closed boundaries. Dynamic regulation was modeled through two control modes: manual operation with predefined opening sequences and automatic regulation based on upstream flow/water level. Flow through gates was determined by orifice flow (when K(t)/h < 0.65) or weir flow (when K(t)/h > 0.65) formulas, considering gate opening height and head difference. Ideal cases were constructed to validate the model against the well-established Storm Water Management Model (SWMM). Field application was conducted in the urban river network of Ganzhou City, Jiangxi Province, using two historical flood events (1998) and a 50-year return period flood for verification. Evaluation indices included the coefficient of determination (R²) and Nash-Sutcliffe Efficiency (NSE).Results and Discussions In ideal case validations, the proposed model showed excellent agreement with SWMM results: For weirs, R² values for flow and water level reached 0.990 and 0.960, respectively, with minor temporal lags attributed to different junction processing methods. For culverts, flow and water level R² values were 0.980 and 0.987, with slight discrepancies due to flow regime transition considerations in the proposed model. For pump stations, the model accurately captured flood diversion processes, achieving R² values of 0.992 (flow) and 0.998 (water level), maintaining stable internal river water levels after pump activation. For sluice gates, the model effectively simulated complex regulation processes (e.g., controlled discharge and free discharge), with a flow R² of 0.994. It successfully captured abrupt flow changes during gate opening/closing and accommodated multi-gate independent control. In the Ganzhou field application: Historical flood simulations yielded NSE values of 0.957 and 0.983, with peak water level differences of 0.08 m and 0.02 m compared to measured data. For the 50-year return period flood, controlled discharge simulations stabilized downstream flow at 3000 m³/s when upstream flow exceeded 2000 m³/s, and switched to free discharge at 4000 m³/s. Constant upstream water level control (102 m) was achieved through dynamic gate adjustments, maintaining stable water levels with continuous model convergence. Overall, the model demonstrated high accuracy (average R² > 0.97 for all structures) and stability. The artificial area method simplified junction calculations without compromising precision, while the source term approach efficiently integrated diverse hydraulic structures without requiring complex boundary redefinition. The gate regulation module, with detailed control rules and flow regime discrimination, outperformed SWMM in capturing real-world operational complexity.ConclusionsThe proposed source term-based algorithm provides a simple and reliable solution for simulating hydraulic control structures in 1D river networks. Key conclusions include: The algorithm accurately models weirs, culverts, pump stations, and sluice gates, with R² values ranging from 0.971 to 0.994 compared to SWMM, exceeding the threshold for strong correlation (R² > 0.7). The artificial area method enables efficient junction processing, eliminating iterative calculations and improving computational efficiency. The sluice gate simulation module effectively captures dynamic regulation processes, with control rules closely aligned to engineering practices. Field validation in Ganzhou confirms the model's applicability to real-world flood simulations, with NSE > 0.95 for historical floods and stable performance in complex regulation scenarios. This study advances river network hydrodynamic modeling by enhancing the accuracy and efficiency of hydraulic structure simulation. The model serves as a valuable tool for flood risk assessment and water resource management, particularly in urban river networks with dense hydraulic infrastructure.

Study Region: Urban road catchment areas with shallow-water drainage systems, where grate inlet clogging by floating debris frequently exacerbates flooding. Study Focus: This study employs a coupled Volume of Fluid-Discrete Element Method (VOF-DEM) model, validated by physical experiments, to simulate the transport and clogging behavior of three typical urban debris types-foam (300 kg/m3), leaves (1000 kg/m3), and plastic bags (1500 kg/m3)-under shallowwater drainage conditions. The objective is to investigate how material-specific properties (density and morphology) govern clogging dynamics and drainage performance. New Hydrological Insights for the Region: Results reveal that density governs movement patterns: lowdensity foam floats and migrates widely with minimal direct clogging; near-water-density leaves suspend and accumulate at grate gaps and corners; high-density plastic bags rapidly settle and cover the grate surface, reducing effective flow area. Morphology synergizes with density to enhance clogging stability. Clogging reduces drainage efficiency, with an average water level backwater of 3.2 mm and flow velocity reduction of 0-15.4%, showing spatial heterogeneity. This study establishes a multi-scale framework linking material properties to hydraulic response, providing quantitative references for anti-clogging grate design, targeted maintenance, and risk assessment of stormwater inlets in the study region.

Real-time forecasting of urban waterlogging is frequently hindered by the complex spatiotemporal mismatch between rainfall inputs and hydrological response, defined here as the Rainfall-Waterlogging Propagation Lag (RWPL). Although hydrodynamic modeling is widely employed, the physical mechanisms underlying the spatiotemporal heterogeneity of RWPL remain poorly understood, largely because conventional models lack the capability to track runoff sources and pathways. To bridge this gap, this study investigates the non-linear dynamics of RWPL using a high-performance coupled model (GAST-SWMM) integrated with a fully 1D-2D coupled Rainfall Time-Source Tracking (TST) module. Combining theoretical benchmarks established via idealized models with a case study of the Xi'an Moat Basin, we identified three distinct physical response patterns of urban waterlogging: (1) Topography-dominated pattern: characterized by a monotonic decrease in RWPL with increasing rainfall intensity, driven by the rapid concentration of overland flow waves. (2) Dynamic connectivity pattern: exhibiting a non-monotonic trend where RWPL initially increases then decreases with rainfall intensity. This reveals a fill-and-spill mechanism governed by explicit quantitative thresholds, where catchment expansion elongates flow paths and causes significant delays. (3) Pipe-buffering pattern: controlled by hydraulic blockage. The transition from free surface flow to pressurized pipe flow creates a buffering effect, decoupling surface inundation from rainfall peaks. Crucially, the coupled TST fingerprints provide robust quantitative evidence validating these mechanisms. These findings challenge the static catchment assumptions in traditional forecasting and provide a theoretical basis for a dynamic offline-online predictive framework, advancing the transition from rainfall monitoring to precise impact forecasting.

Sponge city source control facilities are increasingly recognized as essential for urban stormwater management, yet systematic quantification of their environmental and economic co-benefits remains limited. This study develops an integrated evaluation framework that links standardized engineering work-quantity norms with life cycle assessment (LCA), embedding construction quota logic to enhance reproducibility and policy applicability. Using the Ecoinvent v3.11 database and the ReCiPe 2016 method, the framework evaluates four source control facility types of permeable pavements, stormwater storage modules, vegetated swales, and bioretention in semiarid Northwest China under both baseline and dynamic conditions, including CMIP6 climate scenarios and longterm performance degradation. Results indicate that life cycle costs per functional unit are & YEN;343,400, & YEN;26,100, & YEN;706,600, and & YEN;772,300, respectively. Environmental payback periods remain within 8.63 years for most categories, while economic payback times range from 2.9 to 11.2 years. Construction materials and transportation dominate environmental burdens. The contributions of this study are threefold: it embeds engineering semantics into LCA for practical alignment, integrates climate dynamics and facility degradation into scenario modeling, and jointly evaluates environmental and economic paybacks. Together, these advances establish a transferable, quota-driven tool to support sustainable sponge city development in water-scarce and data-constrained regions.

The increasing frequency of extreme rainfall events and rapid urbanization have amplified urban flood risks, necessitating advanced tools to assess infrastructure resilience. A dynamic framework integrating Graphics Processing Unit (GPU) Accelerated Surface Water Flow and Transport (GAST) model and system function curves is developed to evaluate road network flood resilience in Fengxi New City, China. High-resolution simulations under rainfall return periods (20–1000 years) quantified temporal degradation of road functionality, revealing nonlinear vulnerability patterns: segments exhibited 55.63% failure under 1000-year events, surpassing node losses (29.2%). Model validation using 2016 flood data showed high accuracy, with a 4.8% average error in inundation area predictions. The framework delineates four resilience phases—prevention, response, recovery, and adaptation—highlighting prolonged waterlogging at critical nodes as a key bottleneck for restoration. By mapping structural vulnerabilities and recovery dynamics, this approach identifies high-risk zones and prioritizes mitigation strategies. The physics-based methodology advances data-driven urban adaptation, offering actionable insights to optimize infrastructure planning under climate change.

In order to solve the problem of inaccurate simulation results caused by low-precision rainfall time data to hydrological and hydrodynamic models, the fractal interpolation of the iterative function system was used to interpolate the rainfall time data of the basin and the city based on the two-dimensional hydrodynamic model GAST. The simulation effects of the rainfall time data before and after interpolation were compared. The results show that: ①The two-dimensional hydrodynamic model constructed in this paper can accurately simulate the process of flooding. The Nash efficiency coefficient and mean absolute percentage error of the model are 0.784 and 7.81%, respectively, and the model validation effect is good. ② Fractal interpolation has a good effect in the simulation of the flood process. By using the interpolated rainfall data instead of the original data simulation, the Nash efficiency coefficient and the mean absolute percentage error of the model are increased to 0.168 and 1.43%, respectively, which further improves the accuracy of the model simulation. In summary, fractal interpolation can be applied to the two-dimensional hydrological and hydrodynamic model, which effectively improves the prediction accuracy of the model and provides more reliable decision support and technical guarantee for urban flood control and drainage and watershed flood management.

Flood inundation emulation models based on deep neural networks have been developed to address the high computational cost of traditional two-dimensional (2D) hydrodynamic models. However, challenges remain in urban areas with highly heterogeneous surfaces and complex artificial structures. In this study, we propose a novel emulation framework-Sparse-Point Learning and Interpolated Surface Reconstruction (SPIR)-to simulate high-resolution urban flood inundation with improved efficiency and accuracy. The framework consists of four main steps: (1) flood extent masks are generated through thresholding and morphological filtering to remove shallow water, rooftops, and isolated noise; (2) representative locations are selected within inundated regions based on frequency statistics and spatial clustering; (3) a lightweight fully connected-deconvolutional neural network (FC-DecMaskNet) is used to predict both flood extent masks and water depths at representative locations; and (4) the full water surface is reconstructed using interpolation techniques informed by hydrodynamic principles. Compared to conventional deep learning approaches, SPIR significantly reduces training complexity while improving prediction accuracy. It can generate a complete flood inundation map in approximately 6.3 s, achieving a nearly 2657-fold speed-up compared to the physics-based GAST(GPU Accelerated Surface Water Flow and Associated Transport) model. Evaluation results show that the SPIR framework achieves a POD over 99 % and a FAR below 0.78 %. The mean RMSE is only 0.0018 m, and the 99th percentile of gridwise errors (Q99) averages 0.0064 m.

This study delves into the effects of particle morphology (spherical, tetrahedral, and cubic) and size on the motion characteristics of particles during landslides and water erosion processes, using coupled Volume-of-Fluid and Discrete Element Method simulations. Simulation results reveal a positive correlation between particle sliding distance and its diameter, whereas angular velocity decreases as diameter increases. Spherical particles, due to their symmetrical shape, exhibit the highest mobility and readily form continuous flow layers. In contrast, tetrahedral and cubic particles display more complex motion patterns and are unevenly distributed within the landslide mass. Furthermore, as particle morphology shifts from spherical to tetrahedral and cubic, the critical initiation velocity for hydraulic transport gradually increases (from 9.6 to 13.6 m s(-1) and 15.5 m s(-1), respectively), highlighting the significant impact of morphology on particle initiation conditions. Different particle morphologies also lead to variations in hydrodynamic characteristics; spherical particles tend to form stable flow structures, whereas tetrahedral and cubic particles may generate more complex turbulence. Based on these findings, we propose a simplified empirical framework in which the critical initiation velocity scales as a power-law function of a quantified shape factor and particle size. This relationship provides a practical basis for estimating the hydraulic initiation thresholds of irregular particles in channel deposits and can support the assessment of debris-flow risks in steep mountainous regions.

As compound urban flooding grows more common, inland cities with high external inflow (HEI) encounter increased flood risks. This is due to the combined effects of such external inflow and local surface runoff. Current research has predominantly concentrated on local rainfall-runoff dynamics, with little emphasis on the compound flooding mechanisms and the corresponding risk progression in inland urban areas, influenced collectively by precipitation, external input, and fluctuations in river stages. A multi-module dynamic coupling framework was created to facilitate comprehensive modeling and risk assessment of compound urban flooding, which was implemented in Weinan City, China, a representative HEI city. The framework incorporated a semi-distributed hydrological module, a two-dimensional (2D) surface hydrodynamic module, a one-dimensional (1D) drainage network module, and a one-dimensional river hydrodynamic module, which were dynamically coupled through node-cell-section interactions to enable cross-module water and momentum exchange. In addition, a dualindicator risk assessment framework utilizing maximum water depth and the depth-velocity product was introduced to characterize inundation risk and hydrodynamic risk under compound forcing. The findings indicated that concentrated external inflow substantially altered surface runoff evolution, enhanced flow convergence and rapid conveyance along major road networks, while elevated river stages further restrict outlet discharge through backwater effects, thereby weakening drainage capacity and prolonging surface inundation. The dual-indicator assessment reveals a structural mismatch and partial spatial overlap between inundation risk and hydrodynamic risk. In the examined HEI city, inundation under a P = 100-year rainfall scenario was comparable to that under a P = 30-year rainfall scenario with external inflow, whereas external inflow primarily amplified hydrodynamic risk (659.39%) and river backwater predominantly increased inundation risk (73.06%). This study enhanced the modeling and risk evaluation of urban flooding caused by many compound factors, offering novel perspectives on optimizing drainage systems and managing floods resiliently in anticipation of future catastrophic events.

Urban flood disasters have become increasingly frequent due to extreme weather, posing significant risks to pedestrian safety. This study proposes a dynamic simulation framework for pedestrian evacuation by coupling a two-dimensional hydrodynamic model, a cellular automata (CA) model, and an Artificial Bee Colony (ABC) algorithm. The framework integrates flood evolution, pedestrian instability assessment, and path optimisation. Two case studies, a hypothetical shopping centre and a real urban area in Senigallia, Italy, are used for validation. Results show that the model can effectively reproduce pedestrian behaviours such as movement, detouring, and instability under dynamic flood conditions. Evacuation efficiency is strongly affected by exit configuration and crowd density. In the single-exit scenario, evacuation time increases with population and stabilises at about 15.6 min when exceeding 500 people. In contrast, the multi-exit scenario maintains a stable evacuation time of approximately 11.5 min with improved efficiency. The proportion of unstable pedestrians increases approximately linearly with crowd size. These findings highlight the importance of exit layout and provide support for flood emergency planning and decision-making.