ABSTRACT. River channel networks expand and contract dynamically depending on climatic conditions and the observational scale used to delineate channels from topographic data. A well-known but insufficiently explained phenomenon is that the total area of source basins often remains nearly invariant over a wide range of channel initiation thresholds. This study aims to provide a mechanistic explanation for this scale-invariant behavior and to identify geomorphologic regime shifts associated with ephemeral and perennial channel networks. The analysis is conducted in the Seolma Creek experimental basin (8.5 km²), a small mountainous catchment in central Korea. A 20-m resolution digital elevation model is processed using D8 flow routing, and channel networks are extracted for 889 threshold areas ranging from a single pixel to the entire basin. For each realization, channel sources are identified and corresponding source basins are delineated. The total source basin area is explicitly decomposed into the product of the number of sources and the mean source basin area, both expressed as power-law functions of the threshold area. To objectively identify scaling regimes, the Bai–Perron multiple structural break procedure is applied to log–log relationships without predefining breakpoint locations. Results show that in the low-threshold domain, corresponding to dense and highly dynamic ephemeral channel networks, the scaling exponents of source number and mean source basin area compensate for each other, resulting in an approximately invariant total source basin area. In contrast, in the high-threshold domain associated with perennial channel networks, this compensation breaks down and clear deviations from invariance emerge. The statistically detected breakpoints closely coincide with the spatial extent of perennial channels independently identified from topographic information. These findings demonstrate that scale invariance of total source basin area is not a trivial geometric property, but an emergent outcome of compensatory scaling between network density and drainage partitioning. The total source basin area therefore provides a robust, physically interpretable indicator for detecting regime shifts in channel network organization, with implications for understanding hydrologic resilience and landscape response under changing climatic conditions.

ABSTRACT. As climate change intensifies worldwide, the hydrological cycle strengthens, leading to shifts in precipitation, more frequent extreme events, and more severe water-related disasters. These changes pose significant challenges for assessing future climate-related disaster risks, particularly at regional scales. Consequently, reliable climate projections are critical for understanding climate change and supporting effective adaptation strategies. Although Global Climate Models (GCMs) simulate the Earth’s climate system to assess past climate variability and project future changes under different emission scenarios, their coarse spatial resolution remains a key limitation. In the Amazon River Basin, climate models continue to struggle to accurately represent rainfall, exhibiting persistent overestimations and underestimations across the basin due to uncertainties in large-scale circulation, convection, and sparse observational data. To address these limitations, this study evaluates the applicability of a deep-learning-based super-resolution framework (SR4DS) to downscale climate model precipitation to 0.05° (~5 km) resolution while simultaneously reducing systematic biases. Four GCMs from the NEX-GDDP-CMIP6 dataset were selected to assess historical and future precipitation under two emission scenarios, SSP2-4.5 and SSP5-8.5, in the Amazon River Basin. The results showed that SR4DS substantially improved the representation of basin-averaged precipitation across annual, monthly, and daily timescales, reducing systematic biases and better capturing observed variability compared with the original NEX-GDDP-CMIP6 products. In addition, SR4DS improved the representation of spatial rainfall patterns at annual and monthly scales, while performance at the daily scale remained more limited. Overall, the results demonstrate that SR4DS effectively improved precipitation representation across multiple temporal scales in the Amazon River Basin, supporting its application in regional climate and hydrological studies.

Funding This work was supported by the Technology Innovation Program (RS202400398858, Development of AI-based urban flood damage risk prediction and evaluation technology for practical use) funded by the Ministry of the Interior and Safety (MOIS, Korea).

ABSTRACT. Urban flooding, exacerbated by rapid urbanization and climate change, presents a significant challenge for cities worldwide. While Low Impact Development (LID) practices are recognized as a practical approach for flood mitigation, optimizing the combinations of various LID facilities across diverse urban land uses remains a computationally complex task. This study proposes a comprehensive framework to identify optimal, land use-informed LID combinations that maximize urban flood mitigation performance. The proposed framework integrates a physical rainfall-runoff model (PCSWMM), a machine learning-based surrogate model, and a Genetic Algorithm (GA) for optimization. To overcome the extensive computational requirements of physical model during the iterative optimization process, the machine learning-based surrogate model was trained to rapidly and accurately predict runoff. This methodology significantly enhances the efficiency of the optimization process while maintaining high predictive accuracy, making the framework practical for application. The framework was applied to a highly urbanized region in Gwangju Metropolitan City, South Korea, to determine the optimal combination of five LID types across five impervious land use categories under a wide range of scenarios, including varying rainfall return periods (5 to 500 years), durations (1 to 12 hours), and LID implementation areas (3 to 9% of impervious area). Results revealed that optimal LID combinations were more sensitive to rainfall duration and LID implementation area than to the rainfall return period. Detention-type facilities were found to be more effective for short-duration rainfall, whereas the proportion of infiltration-type facilities increased for long-duration events. Furthermore, as the LID implementation area increased, the detention function of LID facilities became more dominant in runoff reduction. The optimized LID combinations consistently outperformed optimal single-facility scenarios in runoff reduction. Notably, the optimized combinations significantly reduced inundation area and maximum inundation depth even under extreme storm events corresponding to a 500-year return period. The proposed framework serves as an efficient tool for designing tailored LID strategies, enhancing urban resilience to flooding.

ABSTRACT. In recent years, the increasing intensity and frequency of rainfall events due to climate change, combined with rapid urban expansion, have escalated flood risks in Japan, necessitating a transition toward "Basin-wide Flood Control." Although integrated models are essential for evaluating the effectiveness of diverse measures across catchments and floodplains, a significant challenge remains: while conventional lumped models can evaluate the effects of river channel improvements, they are unable to represent the impacts of catchment-scale measures, such as "paddy field dams," or the detailed processes of inland flooding. It remains unclear how the degree of drainage network representation in a distributed model influences the accuracy of simulating combined fluvial and inland flooding phenomena. This study investigates the impact of modeling detail on inundation reproduction, focusing on the flood events caused by Typhoon in August 2024 in the Seino region, Gifu Prefecture. The study targeted two distinct basins with different characteristics: the Kuise River basin, which consists of mountainous headwaters and downstream paddy fields, and the Suimon River basin, a developed plain area with advancing urbanization. Using the distributed integrated model "DioVISTA Flood," three scenarios were analyzed. Case 1 focused on main river channels, while Case 2 incorporated major drainage canals equipped with virtual sluice gates to prevent backflow. Case 3 investigated whether coarsening the computational mesh improves the reproducibility of the simulation by increasing the calculation grid from 25m to 50m. For all cases, rainfall data from five stations were distributed using Thiessen polygons. The model’s performance was validated against actual inundation records from Ogaki City (depths ≥ 0.05 m) and observed hydrographs at the Ichihashi and Hayashimachi gauging stations. Quantitative evaluation was conducted using the Intersection Over Union (IOU) metric, along with overestimation and underestimation rates. The quantitative evaluation indicated that Case 2 tended to achieve higher inundation reproducibility and lower overestimation rates compared to Case 1. These preliminary results suggest that without explicit drainage networks, rainfall may remain stagnant on the surface instead of being properly discharged, leading to potential overestimation in low-lying areas. Furthermore, comparisons at the Ichihashi and Hayashimachi generally showed consistency with observed peak water levels, though further refinement is needed.

ABSTRACT. As the second-longest river in China and the fifth-longest in the world,the ecological conservation and sustainable development of the Yellow River have become a significant national strategy. The floodplain area in its lower reaches is a crucial ecotone where aquatic and terrestrial ecosystems intertwine, playing a vital role in maintaining the ecological balance and supporting agricultural production in northern China. Research on the spatiotemporal changes and scenario simulation of ecological resilience in this region helps to analyze the differences in its capacity to mitigate ecological risks over time and space, providing a scientific basis for zonal ecological management and risk prevention. This study constructed an ecological resilience assessment model for the lower Yellow River floodplain based on three dimensions: resistance, adaptability, and recovery capacity. It evaluated the spatiotemporal evolution of ecological resilience from 1990 to 2025 and explored resilience zoning patterns at the county level using spatial autocorrelation analysis.Furthermore,based on the FLUS model, the future land use patterns under two scenarios for 2030—a baseline scenario and an ecological conservation-oriented scenario—were simulated. The spatial distribution of ecological resilience levels under these different development pathways was then assessed. The main findings are as follows: (1) From 1990 to 2010,low-resilience areas expanded from both banks toward the river channel, while high-resilience areas significantly decreased. (2) From 2010 to 2025,high-resilience areas continued to decline in most regions, but resilience improved substantially in the Yellow River estuary zone. (3) For 2030, under the ecological conservation-oriented scenario, both resistance and adaptability levels are projected to be significantly higher than under the baseline scenario. These findings provide a reliable theoretical basis for understanding the dynamics of ecological resilience in the lower Yellow River floodplain and for guiding its future sustainable planning and development.

ABSTRACT. Hydrological forecasting in poorly gauged basins remains a major challenge due to limited observations, complex hydrological processes, and uncertainties in model structure and forcing. This study presents a comprehensive evaluation of a spatially distributed hydrological data assimilation framework, LISFLOOD-HDAF, developed to systematically investigate how assimilation frequency and observation network design jointly influence streamflow prediction performance. The framework couples the process-based LISFLOOD model with an Ensemble Kalman Filter (EnKF) and is assessed through controlled synthetic twin experiments conducted in the upper Po River Basin, Italy. A factorial experimental design was implemented using six representative wet and dry hydrological events, three assimilation frequencies (6, 12, and 24 h), and eight contrasting observation network configurations that varied in gauge density and spatial arrangement. This design enabled a robust assessment of both temporal and spatial controls on assimilation efficiency. Results demonstrate that EnKF-based assimilation consistently outperforms open-loop simulations across all evaluation sites. Performance gains were most pronounced during wet periods and at downstream locations, where normalized Kling–Gupta efficiency values reached 0.80–0.90 and root mean square error reductions exceeded 50%. Assimilation frequency exhibited a non-monotonic and regime-dependent influence. Twelve-hour updates produced the most consistent improvements across midstream and downstream locations, while six-hour updates were more effective in responsive headwaters during wet events. In contrast, 24-hour updates performed better under dry conditions, reflecting slower hydrological response times. Observation network design strongly controlled assimilation skill: dense or strategically distributed networks provided the highest gains, yet performance gains saturated beyond intermediate densities. Notably, five well-placed gauges combined with 12-hour assimilation achieved near-optimal performance during wet periods, whereas only three to four gauges were sufficient under dry conditions. Overall, LISFLOOD-HDAF provides a robust and transferable framework for evaluating trade-offs between observational density, updating frequency, and forecasting skill. The results offer practical guidance for designing cost-effective and scalable monitoring strategies, particularly in poorly gauged basins, and support the development of operational flood forecasting systems that balance predictive performance with resource constraints.

ABSTRACT. Soil moisture (SM) dynamics prior to and during rainfall events play a critical role in controlling infiltration capacity, surface runoff generation, and ultimately flood occurrence, particularly in large and hydrologically complex river basins. This study focuses on investigating the spatiotemporal dynamics of SM within the Nakdong River Basin (NRB), the longest river basin in South Korea. Our analysis covers the high-rainfall period from June to September 2022, during which several significant flood events impacted the Korean Peninsula. To capture these dynamics, we employed two satellite datasets with contrasting resolutions and sensing mechanisms, Sentinel-1 Synthetic Aperture Radar (SAR) imagery and the Soil Moisture Active Passive (SMAP) product. SM estimates from both satellite sources were systematically compared against precipitation data to evaluate the sensitivity and distinct response patterns of each satellite product to flood-generating rainfall events. The primary objective is to quantitatively assess and compare the capability of high-resolution Sentinel-1 SAR data (20–22 m) versus coarse-resolution passive microwave SMAP data (10 km) in capturing SM variations critical to pre-flood conditions. Ultimately, this research aims to establish the potential of SM variability as an effective, dual-scale indicator for both pre-flood assessment and operational flood monitoring. The results indicate that Sentinel-1 effectively captures highly heterogeneous spatial variations in soil moisture and is particularly sensitive to local conditions, including soil type differences and uneven rainfall distribution. In contrast, SMAP reveals more temporally stable and consistent soil moisture patterns, showing a stronger correlation with regional-scale precipitation variability. These contrasting response characteristics highlight the complementary nature of Sentinel-1 and SMAP in monitoring soil moisture dynamics. Therefore, the integration of both satellite datasets, combined with precipitation information, provides a robust foundation for pre-flood condition monitoring and the development of flood early warning systems in large river basins.

ABSTRACT. Approximately 70% of South Korea’s land area is characterized by steep mountainous terrain, where surface runoff predominates over infiltration during precipitation due to specific topographical constraints. This vulnerability is exacerbated by the seasonal concentration of rainfall with nearly 60% of annual precipitation occurring during the summer and climate change induced increases in rainfall intensity and frequency. These factors contribute to an escalating trend in runoff and sediment yield, posing significant threats to human safety and infrastructure.To mitigate these risks, this study evaluates the effectiveness of land cover conversion, a prominent Nature-based Solution (NbS) that integrates ecological functions to address environmental challenges. Land cover conversion scenarios were developed by considering soil erosion vulnerability, and their impacts on peak and total runoff, as well as sediment yield, were analyzed. A physics based distributed model was employed to simulate three short term rainfall events, comparing the NbS scenarios against a baseline of existing land cover conditions. To ensure the model’s predictive reliability, 3,000 automatic calibrations were performed for each event, utilizing the Nash-Sutcliffe Efficiency (NSE) as the primary objective function. The simulation accuracy was validated using the statistical metrics of NSE, R², and PBIAS (%). The validation results for runoff showed an NSE of 0.88–0.96, an R² of 0.89–0.97, and a PBIAS of -8.09–5.07%. For sediment yield, the metrics ranged from 0.90–0.99 for NSE, 0.92–0.99 for R², and 1.47–15.51% for PBIAS, indicating highly stable and accurate simulation performance. The comparative analysis revealed that the proposed land cover conversion reduced peak runoff by 3.34–13.62% and total runoff by 3.39–9.03%, demonstrating substantial flood mitigation potential. Notably, the impact on sediment management was even more pronounced, with peak sediment yield decreasing by 3.35–26.22% and total volume by 6.74–19.50%. These findings confirm that NbS based land cover conversion is a highly effective strategy for concurrent runoff and sediment control. This research provides a critical empirical foundation for establishing sustainable, watershed-level sediment management strategies in response to intensifying climatic variability.

Funding This work was supported by Korea Environmental Industry&Technology Institute(KEITI) through R&D Program for Innovative Flood Protection Technologies against Climate Crisis Project, funded by Korea Ministry of Environment(MOE)(2022003460002)

ABSTRACT. Agricultural reservoirs have traditionally been designed and operated primarily for water supply purposes; however, the increasing frequency and intensity of extreme rainfall events have heightened concerns regarding their flood safety performance. Even under identical Probable Maximum Precipitation (PMP) conditions, variations in watershed runoff generation characteristics can significantly alter the shape of the inflow hydrograph and peak inflow discharge. Such differences may influence the initiation and progression of embankment failure and subsequently modify downstream flood inundation patterns. This study aims to quantitatively evaluate how variations in runoff characteristics under identical PMP conditions affect dam-break outflow behavior and downstream flood inundation sensitivity. Runoff characteristic scenarios were constructed by systematically varying infiltration losses and watershed response time while maintaining a constant PMP input. The resulting Probable Maximum Flood (PMF) inflow hydrographs were used as inputs to the DAMBRK model to simulate dam-break outflow hydrographs, which were subsequently coupled with a two-dimensional HEC-RAS model for downstream flood inundation analysis. Flood characteristics were quantified using spatially distributed indicators, including inundation area, maximum water depth, and arrival time. Unsupervised learning (k-means clustering) was applied to classify spatial flood response types, and supervised learning (Random Forest) was employed to identify dominant explanatory variables controlling flood sensitivity. Results indicate that changes in runoff characteristics produce substantial variations in peak inflow magnitude, peak outflow discharge, and hydrograph rising rate. In particular, shorter time to peak and steeper rising limbs were associated with nonlinear increases in downstream inundation area and maximum water depth. Flood response patterns were broadly classified into (1) high-depth, short-duration concentrated flooding and (2) shallow, wide-spread inundation types. Variable importance analysis revealed that peak outflow discharge and breach duration were the primary controlling factors governing flood magnitude under identical PMP conditions. The findings highlight the limitations of conventional single-design-flood-based safety assessments and underscore the necessity of incorporating runoff characteristic uncertainty into dam-break flood risk evaluation. This framework provides a scientific basis for identifying vulnerable downstream reaches, establishing risk-informed operational water levels, and enhancing climate-resilient flood management strategies for agricultural reservoirs.

ABSTRACT. In response to the increasing frequency and severity of flood events driven by climate change, reliable and efficient quantitative flood damage assessment has become a critical component of water resources management and disaster risk reduction in Korea. To facilitate the practical and operational implementation of the K-FRM(Korean-Flood Risk assessment Model), this study develops an intuitive and user-oriented plugin within the open-source Quantum Geographic Information System (QGIS) environment. The proposed K-FRM plugin provides an integrated and automated workflow encompassing four core modules: Hazard, Inventory, Vulnerability, and Loss. The Hazard module enables users to input flood inundation maps in raster or vector formats and perform essential preprocessing to delineate hazard zones. The Inventory module identifies exposed assets within inundated areas and retrieves standardized inventory databases such as buildings, agricultural land, population, and vehicles from central K-CDMS server, while supporting asset valuation and parameter configuration. The Vulnerability module applies depth–damage functions stored in the library to define damage ratios by asset type and inundation depth. The Loss module integrates the outputs from the preceding modules to estimate direct physical damage and corresponding economic losses, and generates spatial damage maps, statistical summaries, and structured result reports. By streamlining spatial data processing, database integration, GIS analysis, and computational procedures, the developed plugin substantially enhances the operational usability and reproducibility of K-FRM-based flood damage assessments. The proposed tool is expected to support a wide range of water resources and disaster management applications, including flood risk evaluation, formulation of damage mitigation strategies, and prioritization of preventive investment planning.

ACKNOWLEDGMENTS: Research for this paper was carried out under the KICT Research Program (Development of IWRM-Korea Technical Convergence Platform Based on Digital New Deal) funded by the Ministry of Science and ICT.

ABSTRACT. In recent years, global climate change has significantly increased the frequency and intensity of extreme flooding events, often exceeding the design capacity of conventional river infrastructure. This shift necessitates a transition toward more resilient flood control strategies that can safely accommodate occasional overtopping and intentional inundation. Traditional Japanese river engineering techniques, developed through centuries of experience, offer a promising blueprint for such adaptive management. However, to integrate these historical methods into modern river systems, their hydraulic functions must be rigorously quantified and evaluated using contemporary scientific standards.

This study focuses on the "Kutsuwa-domo" structure, a traditional hydraulic feature located at the Shimada site on the Hamato River in Japan. The research aims to elucidate the specific hydraulic effects of overflow levees integrated within this system. To achieve this, a series of high-resolution numerical simulations was conducted. These simulations modelled various structural configurations to assess how the placement of the Kutsuwa-domo and its levees influences local flow dynamics, water level distribution, and the behavior of suspended particles during high-water events.

The simulation results provided critical insights into the spatial optimization of traditional structures. Relocating the overflow levee slightly upstream was found to reduce the overall water depth across the study reach, suggesting enhanced efficiency in flow regulation. Within the interior of the Kutsuwa-domo, water levels were consistently higher in the downstream section; notably, the spatial extent of this high-pressure zone diminished as the total inflow depth decreased.

Furthermore, the study revealed a significant correlation between levee positioning and particle transport dynamics. When the overflow levee was situated in an upstream position, particles entering from the floodplain (high-water channel) were more effectively discharged over the structure. Conversely, a downstream levee position promoted the transport of particles originating from the main channel (low-water channel). This suggests that the spatial configuration of the levee can be strategically designed to manage sediment and debris transport depending on the specific needs of the river reach.

These findings offer quantitative evidence of the functional versatility of traditional river engineering. By demonstrating how structural placement dictates flow behavior and particle movement, this research provides a scientific foundation for incorporating heritage-inspired designs into modern, resilient flood management frameworks to mitigate the impacts of climate-induced disasters.

ABSTRACT. Agricultural reservoirs play a critical role in flood mitigation and water security in rural regions, yet many existing reservoirs are increasingly vulnerable to extreme rainfall events driven by climate change. While recent studies have emphasized improving forecast rainfall accuracy, operational reservoir safety during extreme events also depends on how effectively real-time observations and hydrological responses are integrated into short-term decision-making frameworks. This study aims to develop an observation-driven, short-term reservoir risk prediction framework that explicitly links inflow dynamics and storage responses under extreme rainfall conditions. The key challenge addressed in this study is the limited reliability of direct water-level forecasting during rapid hydrological transitions, particularly when observational data contain gaps, noise, or delayed responses. To overcome this limitation, we propose a reservoir water budget–based prediction strategy that emphasizes modeling inflow dynamics and their subsequent influence on reservoir storage evolution, rather than relying solely on direct water-level prediction. This approach enhances physical consistency and improves robustness during high-flow events. The methodology integrates multi-source real-time monitoring data, including inflow stream water levels, reservoir water levels, and on-site rainfall measurements collected from agricultural reservoir monitoring systems. Missing and inconsistent inflow observations are systematically reconstructed using a multivariate imputation framework to preserve temporal continuity. Machine learning models are then trained to capture the nonlinear relationship between reconstructed inflow dynamics, recent hydrometeorological conditions, and short-term reservoir storage changes at multiple lead times. The proposed framework was applied to an agricultural reservoir in Korea and evaluated for short-term prediction horizons relevant to emergency response. Results indicate that inflow-based prediction substantially improves the stability and responsiveness of reservoir storage forecasts during intense rainfall events, compared to conventional direct water-level prediction approaches. The framework demonstrates strong potential for supporting real-time reservoir operation and early warning decisions by reducing prediction lag and enhancing situational awareness during extreme hydrological conditions. This study highlights the importance of integrating physically meaningful hydrological processes with data-driven models to improve reservoir resilience under increasing climate extremes.

ABSTRACT. Abstract Purpose This study investigates how canopy elevation regulates throughfall, rainfall interception, and soil erosion, testing whether trees of the same species but with different crown heights are low, medium, high exert distinct hydrological and sedimentological effects under natural rainfall. Methods A year-long field experiment was conducted in woodland at Ritsumeikan University. Gross rainfall was recorded by a rain gauge installed in a nearby open area. Under each of three canopy elevation classes, three experimental soil boxes 37 cm × 25 cm were installed on a 20° slope and subdivided into compartments filled with decomposed granite and silica sand. Nineteen rainfall events were monitored in total. From Event 10 through Event 19, one rain gauge and three bottle collectors were deployed beneath each canopy class to measure throughfall. Soil displacement was measured within a designated 6 cm × 15 cm reference area in each box. This design enabled quantification of gross rainfall, throughfall partitioning, interception losses, and soil detachment across canopy elevations. Results Throughfall ratios showed clear elevation dependence: high canopies transmitted the largest proportion of gross rainfall, whereas low canopies retained more water and produced less throughfall. Correspondingly, soil erosion rates were highest beneath high canopies and lowest beneath low canopies. Across all canopy classes, decomposed granite produced greater erosion than silica sand, consistent with its finer texture, lower cohesion, and higher susceptibility to raindrop impact. Conclusion Canopy elevation strongly modulated rainfall redistribution and erosive energy. High canopies enhanced throughfall and sediment yield, while low canopies buffered rainfall energy and reduced soil loss. These results highlight canopy structural variation as a key control on forest hydrology and sediment dynamics, with implications for erosion management and sustainable woodland design.

ABSTRACT. Debris-flow runout mapping is essential for risk assessment, and recent studies increasingly rely on physics-based models (e.g., Nays2D, Deb2D) that require extensive geotechnical and topographic inputs. In contrast, the Random Walk Model (RWM) provides a probabilistic, computationally efficient alternative for rapid debris-flow propagation simulation using a small set of parameters. RWM requires four parameters: total released sediment volume (VT), sediment volume per step (VS), gravity weight (IW), and stopping slope (SS). While VT can be inferred from pre-/post-event imagery, VS, IW, and SS have typically been tuned manually, which is time-consuming and prone to locally optimal solutions. This study aims to automate and globalize parameter calibration for RWM by integrating a global optimization algorithm.

Three core datasets were used: a 5 m resolution DEM, debris-flow initiation and observed runout extent data, and a binary observation map for objective-function evaluation. The RWM execution code was modified to couple the model with Differential Evolution (DE) to estimate optimal values of VS, IW, and SS. Model performance was evaluated by pixel-wise agreement between simulated and observed runout.

DE-based calibration successfully identified optimal parameters. The resulting simulation matched 10,884 out of 11,801 pixels, achieving an accuracy of 92.23% in reproducing debris-flow propagation. The results demonstrate that automatic global optimization can replace manual tuning in RWM applications and can yield high-fidelity runout simulations with substantially reduced effort. These findings suggest that probabilistic, low-input models coupled with robust optimization provide a practical pathway for efficient debris-flow hazard mapping.

Funding This work was supported by the Technology Innovation Program (RS202400398858, Development of AI-based urban flood damage risk prediction and evaluation technology for practical use) funded By the Ministry of the Interior and Safety(MOIS, Korea)

ABSTRACT. Since initiating its first training program for participants from developing countries under the United States Agency for International Development (USAID) assistance framework in 1963, the Republic of Korea has evolved into a major donor country actively supporting international development cooperation. In this context, this study presents the development of an Integrated Flood Management System as part of a Korea International Cooperation Agency (KOICA) Official Development Assistance (ODA) project targeting the Greater Metro Manila area in the Philippines. The Philippines is highly vulnerable to hydrometeorological disasters, with more than 20 tropical cyclones affecting the country annually, resulting in significant loss of life and property, particularly in densely populated urban regions such as Metro Manila. The primary objective of this project is to establish an integrated flood forecasting, early warning, and monitoring system that consolidates hydrological and meteorological data, flood modeling, and real-time information dissemination to enhance preparedness and response capabilities. The proposed system aims to improve flood early warning accuracy and response efficiency by integrating real-time rainfall and water level observations, flood simulation models, and decision-support tools into a unified operational framework. Furthermore, the system is designed to strengthen coordination between national and local government agencies during flood events, thereby enhancing institutional disaster response capacity. Through the implementation of this integrated flood forecasting and early warning system, the project is expected to contribute to strengthening the disaster risk management capacity of the Philippine government while simultaneously promoting the international dissemination and recognition of Korea’s advanced flood management technologies. Ultimately, this study demonstrates the role of technology-driven ODA projects in supporting sustainable disaster resilience and urban flood risk reduction in developing countries.

ABSTRACT. Traditional assessment of Flood Control Level of Service (LOS) for dams has relied on deterministic hydrological frequency analysis, with a primary focus on inflow exceedance probabilities. However, actual dam failures or downstream flooding events frequently occur through compound failure mechanisms, in which extreme hydrological events interact with mechanical malfunctions such as spillway gate failures and human operational errors. Neglecting these non-hydrological factors can result in a significant underestimation of flood risk. To address this limitation, a comprehensive system reliability analysis framework based on Bayesian Networks (BN) is proposed. A causal network was constructed to integrate stochastic hydrological forcings (rainfall and inflow), state variables (initial reservoir level), and system performance nodes (gate availability, release decision-making, and downstream levee integrity). The BN structure enables quantification of the Conditional Probability of Failure (CPF) for the entire flood control system, explicitly accounting for interdependencies among variables. The framework was applied to the Soyang Dam watershed in Korea. The model simulated scenarios ranging from design floods to Probable Maximum Flood (PMF) events, evaluating the influence of mechanical reliability and operational protocols on overall system safety. The results indicate that the Risk-Informed LOS, derived from the BN, offers a more realistic measure of safety than the traditional return-period-based LOS. Sensitivity analysis further identified that operational failures, such as delayed gate opening, contribute substantially to the probability of overtopping during extreme events. These findings support a transition from static hydrological design to dynamic system reliability assessment as essential for resilient flood risk management under climate uncertainty.

[Acknowledgment] This work was supported by the Korea Environmental Industry & Technology Institute (KEITI) through Climate Resilient R&D Project for Water-Related Disaster Management, funded by the Korea Ministry of Climate, Energy and Environment (MCEE)(RS-2022-KE002032).

ABSTRACT. The estimation of Probable Maximum Precipitation (PMP) is essential for the hydrologic design and safety assessment of critical infrastructure, including nuclear power plants and dams. Conventional PMP derivation methods, which rely primarily on sparse point-gauge networks, often do not capture the high spatiotemporal variability of Local Intense Precipitation (LIP). This limitation is particularly significant for small catchments (e.g., less than 2.5 km2) where short-duration extreme events predominate. This study introduces an advanced framework for estimating Site-Specific Probable Maximum Precipitation (SSPMP) by integrating high-resolution weather radar data with Stochastic Storm Transposition (SST) techniques. In contrast to traditional approaches that utilize static isohyetal maps, the proposed methodology employs spatially continuous radar Quantitative Precipitation Estimation (QPE) data to identify and track extreme storm structures. A rigorous bias-correction process is applied to the radar data, using ground gauge observations to minimize mean-field bias while preserving the spatial heterogeneity of rainfall fields. The SST method is then used to resample and probabilistically transpose extreme storm catalogs over the target basin, accounting for the specific topographic and meteorological characteristics of the site. Preliminary results indicate that the radar-based SSPMP approach provides a more realistic and physically consistent upper limit of precipitation than traditional gauge-based extrapolation, particularly in capturing the spatial peak of storm cores in unmonitored areas. This framework offers a robust alternative for revisiting regulatory standards related to LIP and improves the reliability of hydrologic risk assessments for critical facilities.

[Acknowledgment] This work was supported by the Korea Environmental Industry & Technology Institute (KEITI) through Climate Resilient R&D Project for Water-Related Disaster Management, funded by the Korea Ministry of Climate, Energy and Environment (MCEE)(RS-2022-KE002032).

ABSTRACT. Reliable characterization of drought impacts requires quantifying both the delay from meteorological drought to agricultural and hydrological responses and the lag in recovery after meteorological conditions improve. This study investigates drought propagation and recovery behaviors in the Yeongsan-gang and Seomjin-gang River basins using WRF-Hydro simulations for 2013–2023 and a severity-controlled, event-based framework. Daily basin-averaged indices were computed for meteorological drought (SPEI), agricultural drought (SSMI from modeled soil moisture), and hydrological drought (SSI from modeled outlet streamflow). Propagation combinations were constructed using drought-index intensity thresholds: meteorological drought was defined as SPEI ≤ −1.0 (M2) and ≤ −1.5 (M3), and impact droughts were defined as SSMI/SSI ≤ −0.5, ≤ −1.0, and ≤ −1.5. Events were detected for each combination and accumulation scale (30, 90, 180 days), paired across drought types, and used to estimate Propagation time (TP) and Recovery lagtime (TR). The strength of TP–TR relationships was evaluated using linear regressions. To distinguish coupling from timing synchronization, we also computed TP-aligned lag correlation and maximum event-based lag correlation. We found that TP–TR relationships depend strongly on drought time scale and intensity thresholds. At the 30-day scale, both agricultural and hydrological droughts exhibit well-defined TP–TR relationships under the same moderate-to-severe impact thresholds: for agricultural drought, M2A2 (R²=0.473), M2A3 (R²=0.706), and M3A3 (R²=0.778); for hydrological drought, M2H2 (R²=0.387), M2H3 (R²=0.687), and M3H3 (R²=0.759). In contrast, weaker impact thresholds show limited timing coherence at 30 days. As the accumulation scale increases to 90 and 180 days, TP–TR relationships weaken substantially and, for several hydrological combinations, become statistically indistinguishable from no relationship (e.g., 90-day M2H1 R²=0.024; 180-day M2H3 R²=0.013). Lag diagnostics show that maximum event-based correlations remain high across pathways and scales, whereas TP-aligned correlations diverge between SSMI and SSI. This indicates that strong meteorological coupling can persist even when streamflow deficits are not synchronized to a single Propagation time. Overall, the results show that TP provides meaningful recovery timing structure for both drought types only at short accumulation scales under moderate-to-severe thresholds, while longer accumulation scales require lag-based characterization of drought propagation and recovery to separate synchronization from coupling. This work was supported by Korea Environmental Industry & Technology Institute (KEITI) through Climate Resilient R&D Project for Water-Related Disaster Management, funded by the Ministry of Climate, Energy and Environment (MCEE) (RS-2023-00231944), and was also supported by the National Research Foundation of Korea, which was funded by the Ministry of Science, ICT and Future Planning (RS-2024-00456724).

ABSTRACT. Urban flooding, marked by rapid onset, pronounced spatial heterogeneity, and short-duration inundation, has emerged as a critical challenge amid the climate change–driven intensification of extreme rainfall. Despite advances in numerical modeling and remote sensing, a fundamental limitation remains the lack of high-resolution observational data that capture the microscopic dynamics of urban inundation. This study presents an automated framework for generating microscopic time-series urban inundation data from CCTV imagery using fixed visual references. The proposed framework analyzes real-time and archived CCTV footage from flood-prone locations and estimates inundation depth by referencing fixed objects such as vehicle components, curbs, and roadside structures. Flood progression is automatically classified into three stages—onset, intensification, and recession—and a rule-based conversion scheme converts visually inferred depths into continuous time-series numerical data. The results demonstrate that the proposed method can reconstruct detailed temporal variations in inundation depth at a microscopic spatial scale. The generated data provide physically interpretable flood information that is difficult to obtain using traditional sensors and offer high-quality training and validation datasets for next-generation multimodal AI-based urban flood prediction models.

ABSTRACT. In cold, snowy regions such as Hokkaido, where increased hydropower generation at dams is required, snowmelt serves as a vital water resource but also poses flood risks, making accurate runoff estimation during the melt season essential. This study develops and evaluates a runoff calculation method that uses publicly available meteorological data—analyzed precipitation and snow depth—without requiring on-site observations, enabling several areas applicability. The method was applied to four dam basins in Hokkaido, Japan, covering both summer and snowmelt seasons. To improve accuracy, correction factors for rainfall and snowfall water equivalent were introduced based on long-term water balance analysis. Results indicate that using analyzed snow depth alone often led to significant bias: overestimation in Jozankei and underestimation in Hoheikyo, primarily due to the coarse spatial resolution of 5 km grids, which cannot capture elevation-dependent snow distribution. To address this, an alternative approach estimating snow water equivalent from analyzed precipitation was proposed. When combined with correction factors, this method substantially improved agreement between calculated and observed runoff. Seasonal hydrograph comparisons confirmed that corrected inputs reduced discrepancies and reproduced flow variations more accurately. Additionally, snowmelt runoff forecasting was conducted for April–June 2024, targeting 24-hour ahead predictions using MSM forecast data. The results demonstrated that applying snowfall water equivalent corrections significantly enhanced prediction accuracy for 24-hour average discharge compared to uncorrected cases, which consistently underestimated flows. This improvement underscores the critical role of accurate snow water equivalent estimation in operational dam management, particularly for optimizing reservoir operations and meeting hydropower generation targets during the melt season.

ABSTRACT. This research presents a bivariate copula model to simulate future precipitation patterns in the Banas River basin, an ephemeral river system in India. Utilizing data from the Central Water Commission (CWC) and temperature projections from the CMIP6 Copernicus dataset, the study addresses the challenges of modeling the complex, non-linear relationship between temperature and precipitation in ephemeral river basins. For the modeling purpose, five months are selected which fall into the monsoon period during which precipitation is observed. A time series analysis of the time ranges between the years 1979 and 2021 is performed using the Gaussian copula model. The proposed copula model effectively captures the joint distribution of temperature and precipitation, incorporating their interdependencies to produce more accurate precipitation forecasts under changing climate conditions. The R2 error, Pearson correlation coefficient, and Root Mean Squared Error (RMSE) values for the monsoon season are calculated. The R2 error values for June, July, August, September, and October are found to be 0.829. 0.811, 0.828, 0.764, and 0.875 respectively. Also, all these months demonstrate a strong Pearson correlation coefficient apart from September which shows a value of 0.874. The RMSE values for June, July, August, September, and October are 6.65, 7.72, 7.162, 3.774, and 4.468 respectively. By applying this model to historical data and future projections, we demonstrate its capability to enhance the precision of precipitation simulations, which is vital for water resource management and flood risk assessment in ephemeral river systems. This approach significantly advances climate modeling for regions affected by variable and intermittent river flows, contributing valuable insights for sustainable water management and climate adaptation strategies.

ABSTRACT. As the fundamental protective structure for bank and bed stabilization in navigation channel regulation projects, the flexible mattress is critical to ensuring the long-term stability of river channels. Its structural integrity is closely linked to navigation safety, hydraulic performance, and the sustainable economic development of river basins. Consequently, real-time and high-precision monitoring of the deformation behavior and service state of flexible mattresses is essential for enhancing their capacity to withstand potential risks under complex and dynamic hydrodynamic conditions. Furthermore, the continuous refinement of safety early-warning systems for flexible mattresses can provide robust technical support for the early identification of hazardous states, accurate localization of risk-prone zones, and scientifically informed decision-making throughout engineering operation and maintenance.This paper presents a systematic review of the current research progress on deformation monitoring technologies and early-warning methods for flexible mattresses. Particular attention is given to the application scenarios, technical characteristics, applicability limitations, and recent advancements of various deformation monitoring techniques, including acoustic monitoring, optical fiber sensing, pore water pressure monitoring, remote sensing-based monitoring, as well as settlement gauges and inclinometer monitoring. In addition, the fundamental principles and engineering applications of different stability evaluation approaches for flexible mattresses are comprehensively discussed, encompassing comprehensive evaluation methods, structural reliability-based assessment techniques, empirical formula-based approaches, and machine learning–driven evaluation systems.The review results indicate that a fully integrated system for deformation monitoring, service-state evaluation, and early warning specifically tailored to the operational conditions of flexible mattresses has not yet been established. Future research should therefore focus on elucidating the mechanisms governing flexible mattress deformation and sub-bank instability, developing real-time monitoring technologies that are compatible with the evolutionary characteristics of flexible mattress service states, and improving multi-factor coupled early-warning frameworks.The findings of this study provide valuable technical support and methodological references for the further development and optimization of monitoring and early-warning technologies aimed at assessing and managing the service states of flexible mattresses.

ABSTRACT. This research aims to utilize digital twin technology to visualize flooding conditions in cities with underground spaces, thereby evoking a realistic perception of disaster realities and enabling individuals to internalize disaster prevention and mitigation measures as their own responsibility. In major Japanese cities experiencing increasingly severe flooding, underground shopping arcades, road tunnels, passageways, and subways are particularly vulnerable spaces. Floodwater entering through above-ground entrances of underground complexes can hinder evacuation, potentially increasing human casualties and paralyzing urban functions. To avoid this, measures that encourage voluntary and early evacuation are essential. Content that enables the unified visualization and recall of flood conditions during disasters is considered useful for this purpose. 　The study area selected is the middle reaches of the Toyohira River, which flows through central Sapporo City, Hokkaido, home to approximately 2 million residents. The Toyohira River is one of the world's fastest-flowing rivers, possessing a very high erosion risk. Furthermore, the middle reaches contain underground spaces such as underground shopping malls and subways, making it a flood-vulnerable area. This study integrated a 3D virtual space with river inundation simulations to create a digital twin. A game engine was used to create the 3D virtual space, employing a Digital Elevation Model (DEM) derived from base map information and the PLATEAU 3D city model. PLATEAU is a project aiming to create 3D models of cities nationwide in Japan and make them open source. For the inundation simulation, river flooding analysis was performed using iRIC Nays2DFlood based on maximum flow values under current climate conditions derived from large-scale ensemble climate prediction data. This calculated the time-series flood depth following a levee breach. By integrating this 3D virtual space with flood simulation, we achieved a digital twin that visualizes the conditions during the Toyohira River flood with objective spatial coordinates and high immersion. Such content provides an immersive experience close to an actual flood, conveying conditions similar to real flood disasters. It is considered useful for raising public awareness of disaster prevention and evacuation, thereby contributing to appropriate evacuation actions.

ABSTRACT. Current Korean design guidance for agricultural reservoirs commonly assumes flood-control storage as 20 percent of effective storage and recommends a specific flood-control storage, defined as flood-control storage divided by catchment area, of at least 100 millimeters. The guidance also defines flood-control target reservoirs using simple eligibility conditions such as catchment area of five square kilometers or more, flood travel time of one hour or longer, or a relatively large reservoir surface area compared with catchment area. This study tests the practical applicability of these design-based criteria using nationwide data for approximately 3,400 existing agricultural reservoirs managed by Korea Rural Community Corporation, under the assumption that the hydrologic and geomorphic characteristics of existing reservoirs can represent those of potential new projects. To ensure consistent attribute quality, we focus on reservoirs with storage capacity of 100,000 cubic meters or more and evaluate how many sites satisfy each eligibility condition and their combinations. Among these reservoirs, 173 sites meet the surface-area-to-catchment criterion together with a catchment area of 500 hectares or more, yet only 24 sites have a flood travel time of one hour or longer based on safety inspection reports. Next, we assess whether the recommended specific flood-control storage threshold of 100 millimeters is feasible when flood-control storage is set to 20 percent of storage. The results show that only 89 reservoirs satisfy the 100 millimeter threshold under the surface-area-to-catchment condition, only 15 satisfy it under the catchment-area condition, and the intersection of the key conditions yields merely 12 reservoirs—of which only two were confirmed to have an actual flood-control function. Furthermore, the 100 millimeter threshold often corresponds to an unrealistically large fraction of total storage, in some cases approaching or exceeding total storage, whereas the 20 percent storage assumption typically translates to roughly 20 to 60 millimeters in specific terms across regions. These findings indicate that the current criteria can be overly restrictive and, in practice, may not align with the physical and operational realities of agricultural reservoirs. The study suggests revising both the target-reservoir definition and the flood-control storage criterion to better reflect agricultural reservoir characteristics and feasible operational constraints. This work was supported by Korea Institute of Planning and Evaluation for Technology in Food, Agriculture and Forestry(IPET) through Intelligent Agricultural Infra Management for Climate Change Development Program, funded by Ministry of Agriculture, Food and Rural Affairs(MAFRA)(RS-2025-02215604).

ABSTRACT. Climate change is increasingly altering flood characteristics in South Korea, with changes in flood frequency, intensity, and duration occurring in complex and sometimes contrasting ways. These shifts have led to the expansion of multidimensional flood risk, which cannot be sufficiently captured by conventional assessments focused solely on peak discharge. In particular, the growing prevalence of extreme rainfall and localized heavy storms highlights the need for an integrated framework capable of representing diverse flood behaviors under future climate conditions. This study develops an Integrated Flood Risk Index (IFRI) to evaluate future flood risk across South Korea using IPCC AR6–based climate scenarios and nationwide hydrological simulations. Daily runoff was simulated for the period 1981–2100 using the Soil and Water Assessment Tool (SWAT), with 774 sub-basins across five major river basins adopted as spatial analysis units. From 20 regional climate model–shared socioeconomic pathway (RCM–SSP) combinations provided by the Korea Meteorological Administration, seven representative scenarios were selected based on a climate variability and extremity screening approach. Analyses were conducted for a historical baseline period (1991–2020) and two future periods: mid-century (2031–2060) and late-century (2061–2090). The IFRI integrates information on flood frequency and intensity derived from simulated runoff. A key component is the Standardized Flood Index (SFI), calculated by standardizing short-term accumulated runoff using a log-normal distribution. Flood events were identified when SFI values exceeded +1.0 or equivalent high-percentile thresholds. Based on changes in the IFRI, future flood regime responses were quantitatively classified into four types (Type 1–4), representing distinct patterns of flood risk evolution. Results indicate substantial spatial heterogeneity in future flood risk across South Korea. Many regions exhibit intensified flood hazards during the mid-century period, with partial moderation toward the late century while maintaining risk levels above the historical baseline. The shortening of flood return periods suggests an increased likelihood of more frequent and severe flood events. Overall, the proposed framework enables a systematic national-scale assessment of multidimensional flood risk and provides critical insights for climate-adaptive flood management and long-term water policy planning.

ABSTRACT. Developing countries are particularly vulnerable to climate change due to limited adaptive capacity and restricted access to high-resolution climate information. Although General Circulation Models (GCMs) provide valuable insights into future climate conditions, their coarse spatial resolution limits their applicability in local-scale hydrological and risk assessments. To address this gap, this study presents a novel deep learning–based super-resolution approach designed to enhance the spatial resolution of GCM precipitation outputs for more reliable hydrological modelling and assessments. The proposed approach shows strong agreement with observations, with correlation coefficients exceeding 0.9, alongside clear improvements in error metrics such as root mean square error (RMSE), Kling–Gupta Efficiency (KGE), and structural similarity (SSIM). When applied across Cambodia, the downscaled projections indicate statistically significant increases in rainfall intensity, frequency, duration, and extremes under both SSP2-4.5 and SSP5-8.5 scenarios for the near future (2026–2060) and far future (2061–2100), relative to the historical period (1985–2014). Seasonal analysis suggests an increasing contrast between hydrological regimes, with drier dry seasons and more intense rainfall during the monsoon, and notable spatial variability across the country. Changes in extreme rainfall characteristics indicate a growing frequency of short-duration, high-intensity events, which are closely linked to flood generation. The high-resolution precipitation datasets produced in this study preserve hydrologically meaningful spatial and temporal patterns and are therefore well suited for use in hydrological modelling. This study hypothesises that these projected rainfall changes will lead to higher streamflow and flood risk under future climate conditions. Therefore, these findings provide a robust foundation for future hydrological impact studies and support climate adaptation and disaster risk reduction planning in Cambodia.

Funding This work is financially supported by Korea Ministry of Climate, Energy, Environment(MCEE) as 「Research and Development on the Technology for Securing the Water Resources Stability in Response to Future Change(RS-2024-00332494)」.

ABSTRACT. The Saemangeum in South Korea is the world's longest sea-dike, and internal structural changes due to subsidence, cracking, and sinking at unspecified points are impossible to monitor with the naked eye. Since the completion of the seawall, there is a need for monitoring of the internal structural changes within the seawall over time and for analysis of the internal structural changes of the seawall using the non-destructive exploration in preparation for various causes of recent subsidence and sinking.

It is necessary to establish a comprehensive and preemptive seawall safety maintenance and management plan to prevent subsidence and collapse by conducting regular monitoring using ground penetration radar (GPR) exploration. To ensure the safety of internal fill materials and prevent subsidence or collapse, secure basic data on safety of breakwaters through continuous GPR monitoring. Confirmation of GPR abnormality points and establishment of countermeasures for vulnerable sections of breakwaters through non-destructive exploration.

Multichannel Analysis of Surface Waves (MASW), electric resistivity exploration, and dynamic cone penetration test (DCPT) were conducted to understand the ground characteristics of the lower part of the GPR exploration adverse reaction section and evaluate the inner safety of the seawall. MASW showed a partial enlarged section of the low-speed zone, and the electric resistivity exploration showed that the area of partial and low resistances increased as the sea level tidal increased, but the difference was insignificant. As a result of performing dynamic cone penetration tests in both the stable and unstable sections of the embarkment, the cone resistance, bearing capacity, and the corrected N-value were all relatively higher in the stable section compared to the unstable section. The shear wave velocity(Vs) from MASW showed a strong correlation with these results, allowing for the creation of cross-sectional profiles.

Contributing to the long-term maintenance and management of breakwaters by utilizing the monitoring method developed in this study. It is possible to prepare for and take countermeasures before the occurrence of abnormal areas (subsidence/depression). This monitoring method suitable for the characteristics of the breakwater suggested in the research results will be distributted to breakwater management practitioners and utilized as a method for systematic breakwater maintenance and management.

ABSTRACT. Increasing rainfall intensity under climate change has amplified flood risks, while conventional grey infrastructure is becoming difficult to maintain in regions facing population decline. This has increased interest in nature-based flood mitigation measures that utilize existing landforms. In Japan, valley-bottom lowlands (“Yatsu”) contain abandoned paddy fields with potential temporary storage capacity. Although previous studies demonstrated their flood mitigation potential, key operational factors—including inlet/outlet board height, timing of diversion, and spatial configuration—remain insufficiently examined. The hydraulic impacts of diverting river water into Yatsu areas and optimizing storage operation strategies are not yet fully understood. This study addresses these knowledge gaps.

The study was conducted in the upper Takasaki River basin within the Inbanuma catchment, Japan. Flood inundation simulations were performed using the Rainfall–Runoff–Inundation (RRI) model for four historical rainfall events (2013, 2019, 2022, and 2023). A 50-m resolution grid was constructed based on elevation, flow direction, and catchment area datasets. To clarify hydraulic responses and spatial characteristics of storage diversion, model parameters such as roughness, permeability, and channel geometry were simplified to generate controlled overflow conditions.

Since the RRI model does not explicitly represent inlet control boards, levee height was modified as a proxy to simulate diversion into valley-bottom lowlands. Levee heights were systematically varied from 0 m to 10 m at 0.3 m intervals, and peak water levels over a 48-hour simulation period were recorded for each scenario.

Four spatial configurations were examined: full-basin modification, tributary-scale modification, reach-scale modification, and localized modification. Comparative analyses of river and slope water levels were conducted to evaluate storage–floodplain balance and flood mitigation performance.

Simulation results indicate that levee height modification significantly influences peak water levels. In full-basin and tributary-scale scenarios, an optimal levee height of approximately 1.8 m reduced peak water levels by up to 0.7 m compared to the 0 m condition. Reach-scale and localized modifications produced smaller reductions of approximately 0.2 m and 0.03 m, respectively. Further increasing levee height beyond the optimal level led to rising water levels due to reduced overflow and storage exchange. These findings demonstrate that levee height and spatial deployment critically determine flood mitigation performance in valley-bottom lowlands.

ABSTRACT. Depth-Area-Frequency (DAF) curves are an essential tool in design rainfall estimation, as they represent variations in rainfall depth with increasing spatial extent. However, conventional DAF curves have primarily been developed using empirical or deterministic approaches based on historical observations, which limits their ability to explicitly account for the spatial dependence of rainfall across different area levels and to adequately describe the probabilistic variability of rainfall depth even for the same area. In particular, extreme rainfall events exhibit scale-dependent characteristics, and neglecting the dependence structure among different spatial scales may lead to overestimation or underestimation of design rainfall. Consequently, there is a strong need for a probabilistic DAF estimation approach that can simultaneously incorporate spatial dependence and uncertainty across area scales. In this study, probabilistic DAF curves corresponding to different return periods are derived by analyzing the statistical characteristics of area-based extreme rainfall and by quantifying the dependence structure between area levels using copula models. Annual maximum rainfall events are extracted from Automated Synoptic Observing System(ASOS) observational rainfall data, and area-averaged rainfall depths are calculated for multiple spatial scales for each event. Marginal probability distributions appropriate for each area level are then identified and fitted to characterize the frequency behavior of rainfall depth. Subsequently, copula models are applied to model the spatial dependence among rainfall depths at different area scales. Based on the conditional distributions derived from the fitted copula models, rainfall depths at larger spatial scales are probabilistically estimated given a specified return-period rainfall at a reference area, leading to the construction of probabilistic DAF curves. The proposed probabilistic DAF curves provide a quantitative representation of uncertainty in rainfall depth associated with spatial expansion, thereby overcoming key limitations of conventional empirical DAF curves. The results of this study are expected to contribute to more reliable design rainfall estimation and extreme rainfall analysis.

ABSTRACT. Recently, the urban flooding in South Korea has become more frequent and severe due to short-duration and high-intensity rainfall events. Such events typically develop within a few hours, leaving limited time for emergency response. While national-level flood forecasting systems mainly focus on river water levels, urban flooding is strongly influenced by local drainage conditions and is primarily managed by local governments. Therefore, a municipality-centered urban flood early warning system is essential to support timely and effective on-site decision-making. This study proposes a framework for a local government–based urban flood early warning system that integrates rainfall-based and inundation depth–based criteria. The system is designed to automatically identify high-risk areas and provide intuitive warning levels for municipal officers without requiring manual data interpretation. Short-term quantitative precipitation estimates (QPE) are first evaluated against impact-based rainfall thresholds that represent local vulnerability. In urbanized areas, high-resolution inundation depths simulated by an urban flood model are further used to assess spatially explicit flood risk. Early warning levels are classified into four stages—Attention, Caution, Alert, and Severe—based on inundation depths of 10 cm, 20 cm, and 50 cm, corresponding to impacts on pedestrians, traffic, and buildings. The proposed system was applied to Gunsan City, a coastal mid-sized city that has experienced repeated urban flooding. Flood events from 2022 to 2024 were analyzed to evaluate the applicability of the warning criteria. Results show that rainfall-based thresholds successfully issued Alert or Severe warnings approximately 1–2 hours before reported inundation events. Furthermore, inundation depth–based warnings provided detailed spatial differentiation within administrative districts, clearly identifying highly flooded grids that coincided with observed damage locations. In contrast to rainfall-only warnings, the inundation-based approach improved spatial consistency and supported more targeted evacuation and response planning. Overall, the results demonstrate that a high-resolution, forecast-based urban flood early warning system can enhance municipal preparedness by securing lead time and providing actionable, location-specific risk information. This approach offers a practical foundation for strengthening local flood resilience under increasing climate extremes.

[Acknowledgements] This work was supported by Korea Environment Industry & Technology Institute (KEITI) through Climate Change R&D Project for New Climate Regime, funded by Korea Ministry of Environment (MOE) (RS-2022-KE002152).

ABSTRACT. Coastal environments are characterized by high variability and complex topography, posing significant threats of hidden physical hazards to personnel engaged in waterfront operations. Existing hazard management approaches primarily rely on pre-operational safety education and Personal Protective Equipment. However, these methods are largely passive and lack real-time risk warning mechanisms for small hazardous objects like broken glass or medical needles. To address this limitation, this research proposes a risk identification workflow based on Edge Computing (Edge AI), aiming to explore the feasibility of enhancing personnel’s hazard awareness through proactive image detection technology. The study plans to utilize lightweight object detection models for real-time inference, which is expected to enable the preliminary identification and labeling of specific hazardous objects on mobile terminals. Furthermore, the research proposes to trigger real-time visual alerts upon detecting potential risks. This mechanism is expected to improve the efficiency of hazard identification and serve as a planning reference for the future development of coastal safety management technologies, thereby enhancing the effectiveness of safety management for waterfront operations.

ABSTRACT. Recent extreme rainfall events caused by climate change have increased in frequency and intensity, leading to a higher risk of damage to populations and buildings located downstream of reservoirs. The current quantitative downstream damage assessments in Korea are based on integrated analysis models that consider various factors such as flood inundation areas, exposed populations and assets, damage functions, and economic damage estimations. These assessment systems are generally designed based on flood risk analysis for river flows. However, in environments like the downstream areas of agricultural reservoirs, which have distinct structural and topographical characteristics, the current system cannot fully reflect the specific damage characteristics. Therefore, this study proposes the necessity for improving the quantitative downstream damage assessment system to reflect the unique characteristics of agricultural reservoir downstream areas. This research calculates water depth and velocity for each section through flood wave analysis, and accurately assesses the population and building damage in risk zones by incorporating grid-based population and building distribution into the flood inundation area. This study emphasizes the need to establish a more accurate disaster risk assessment system by considering the special risk factors of agricultural reservoirs downstream that are often overlooked by the existing evaluation systems. Moreover, the results of this study are expected to contribute to enhancing the disaster prevention and response capabilities for agricultural reservoirs, improving design standards, and formulating downstream disaster response strategies.

Acknowledgments : This work was supported by Korea Institute of Planning and Evaluation for Technology in Food, Agriculture and Forestry(IPET) through Intelligent Agricultural Infra Management for Climate Change Development Program, funded by Ministry of Agriculture, Food and Rural Affairs(MAFRA)(RS-2025-02215604)

ABSTRACT. Under the intensifying threat of climate change, river sluice gates serve as critical infrastructure for flood defense. Maintenance prioritization often relies on a comprehensive grading system that weighs structural stability over mechanical functionality. This approach leads to a "Safety Bias," where operational risks of mechanical components are masked by sound structural conditions. This study aims to quantify this bias and proposes a Dual-Track Risk Assessment Model to ensure reliability. We analyzed precision safety diagnosis data for 51 sluice gates in a single river basin. The analytical framework includes identifying safety bias, predicting physical residual life using carbonation models, and conducting scenario-based sensitivity analysis. By varying the mechanical weight from 0.15 to 0.90, we identified the threshold for "Rank Reversal," where mechanical risks begin to drive maintenance priorities.

The results revealed that 16% of facilities were rated as "Good" overall despite "Poor" mechanical conditions, indicating a critical underestimation of functional risks. Additionally, some relatively new facilities showed severe carbonation exceeding cover thickness, challenging the assumption that age equates to risk. The sensitivity analysis identified a Rank Reversal at a mechanical weight of 0.45. Below this threshold, structure-aging facilities dominated priorities, but above it, mechanism-aging facilities rose to the top, unmasking hidden risks. To ensure reliability, maintenance strategies must shift from structure-preservation to function-assurance. We suggest adopting a flexible risk assessment model with a mechanical weighting factor of at least 0.5 to prevent Safety Bias and ensure timely gate operation.

Acknowledgement: This study was supported by KEITI through 'development of integrated asset management technology for water resources infrastructures to be incorporated in digital twins'(RS-2024-00337673), funded by MCEE

ABSTRACT. Climate change is progressively altering aquatic environments, leading to an earlier onset and increased frequency of algal blooms in rivers and reservoirs. In particular, the proliferation of Cyanobacteria, which causes harmful algal blooms (HABs), produces toxins such as microcystin, thereby exerting detrimental effects on both aquatic ecosystems and public health. To mitigate these issues, algae removal substances certified by the Ministry of Environment are widely utilized. However, the existing certification process predominantly focuses on removal efficiency and acute ecotoxicity, leaving a significant research gap regarding the broader physicochemical changes in water quality following their application.

This study aimed to evaluate the impact of these substances on the aquatic environment through controlled mesocosm experiments. The substances were classified based on their primary mechanisms: algaecidal (direct killing) and flocculating (aggregation). We analyzed the interaction characteristics between these substances and various water quality parameters using a scatter matrix approach. The results demonstrated distinct functional differences between the two types. Algaecidal agents exhibited a strong negative correlation with Chlorophyll-a (r = -0.737, p < 0.05), indicating high efficacy in direct biomass reduction. Conversely, flocculating agents showed significant negative correlations with Total Organic Carbon (TOC, r = -0.648, p < 0.01) and Total Nitrogen (T-N, r = -0.758, p < 0.001).

These findings imply strategic distinctions in application. Flocculation-based substances possess a comparative advantage in long-term water quality improvement and preventive management by effectively reducing nutrient salts, which are key drivers of eutrophication. In contrast, algaecide-based substances are identified as superior emergency control measures for rapid response during high-concentration algal outbreaks. Although this study is limited by the sample size inherent to mesocosm conditions and requires further validation through field-scale research, the results provide critical foundational data. This research offers a scientific basis for refining water quality management strategies and improving the regulatory framework for selecting appropriate algae control agents based on site-specific environmental objectives.

ABSTRACT. Digital twins enable real-time monitoring, analysis, and prediction by synchronizing physical systems with virtual representations. Despite their potential, conventional digital twin implementations rely on complex physics-based modeling, high-performance computing, and large-scale data infrastructures, which hinder rapid deployment and operational scalability in field-level disaster management. These limitations are particularly restrictive for urban flood response, where low latency, robustness, and ease of operation are critical. This study proposes a Lightweight Digital Twin (LDT) framework tailored for urban flood risk management, designed to minimize computational complexity while preserving essential decision-support capabilities. The proposed LDT adopts a modular architecture that prioritizes core hydrologic–hydraulic indicators, AI-driven inference, and event-based data assimilation, enabling near–real-time operation under constrained computing environments. By reducing model dimensionality and data dependency, the framework supports fast initialization, adaptive updates, and reliable performance across heterogeneous urban settings. Building on this framework, we develop a field-deployable decision-support platform that integrates AI-based flood prediction, urban flood risk indexing, and early warning functions within a lightweight digital twin environment. To ensure operational validity and user-centered optimization, the platform is evaluated through an AI-enabled Living Lab, where local government officials and domain experts participate in iterative testing under real urban conditions. The Living Lab facilitates continuous calibration of model outputs, interface design, and operational workflows, enhancing both technical robustness and institutional acceptance. A facilitator-led coordination mechanism supports stakeholder communication, scenario-based evaluation, and conflict resolution during the validation process. In parallel, structured training modules are provided to improve system interpretability and on-site usability. Feedback from participating municipalities is systematically incorporated to refine model parameters, data pipelines, and alert thresholds. The proposed approach demonstrates how lightweight digital twins, when combined with AI analytics and participatory validation, can deliver practical, scalable, and transferable solutions for urban flood risk management. The resulting framework is expected to serve as a technical and policy reference model that can be adapted to other regions, contributing to improved resilience against flood hazards under climate change.

ABSTRACT. This study conducts a comparative evaluation of rainfall-runoff modeling approaches to enhance the accuracy of reservoir inflow simulation during spring season in major river basins of Korea, including the Yeongsan, Seomjin, and Geum Rivers. The inflows are influenced not only by rainfall but also by snow accumulation and subsequent snowmelt processes, particularly in mountainous and mid-altitude catchments. A conventional rainfall-driven GR4J model was compared with a reconfigured modeling framework that explicitly incorporates a temperature-index-based snow accumulation and melt module. Both models were forced with identical daily basin-averaged precipitation, temperature, and reference evapotranspiration data. Model parameters were calibrated and validated using long-term hydrometeorological observations, applying the SCE-UA global optimization algorithm with the Kling-Gupta Efficiency as the objective function. The comparative results demonstrate that the snowmelt-integrated model consistently outperforms the rainfall-only model across all study basins. Improvements are observed in terms of higher KGE and coefficient of determination values, along with reduced root mean square error. Performance gains are particularly pronounced in mountainous catchments where snowmelt contributes substantially to the inflow variability. These findings indicate that explicit representation of snowmelt processes significantly improves the reliability of spring inflow simulations and provides a more robust basis for reservoir operation, seasonal water supply planning, and future climate variability assessments.

ACKNOWLEDGEMENTS

This work was supported by Korea Environmental Industry & Technology Institute(KEITI) through Water Management Program for Drought, funded by Korea Ministry of Climate, Energy and Environment(MCEE)(RS-2023-00231944)

ABSTRACT. The increasing frequency and intensity of extreme rainfall events under climate change have raised concerns regarding the potential failure of agricultural reservoirs and the resulting downstream inundation hazards. This study aims to quantitatively assess downstream inundation characteristics following dam break events of an agricultural reservoir under both extreme flood and climate change scenarios in Korea. A case study was conducted for an agricultural reservoir located in the Geum River basin. Dam break scenarios were established under Probable Maximum Flood (PMF) inflow conditions by considering different combinations of breach width and breach duration. The DAMBRK model was applied to derive breach outflow hydrographs for each scenario. To account for future climate change conditions, additional inflow scenarios were constructed using climate projections from the CMIP6 SSP5-8.5 scenario. Among 18 Global Climate Models (GCMs), the CanESM5 model, which exhibited the largest mean annual precipitation, was selected to represent a severe climate change condition. The resulting breach outflow hydrographs were then incorporated as upstream boundary conditions in a two-dimensional HEC-RAS model to simulate downstream inundation and flood routing processes. Key flood characteristics, including flood wave arrival time, peak water level, and peak discharge, were analyzed at major downstream locations. The results indicate that, even under identical breach conditions, the application of climate change scenarios leads to significantly larger peak breach outflows compared to historical conditions, with increases exceeding approximately 40% in some cases. Downstream simulations further show increased peak discharges and shortened flood wave arrival times along the river reach. The inundation extent propagated up to approximately 5 km downstream from the dam site, highlighting the potential amplification of flood hazards under future climate conditions. The results of this study provide valuable insights into the impacts of climate change on dam break induced inundation and can support risk assessment and emergency planning for agricultural reservoirs in Korea.

ABSTRACT. River water resource facilities, including rainwater pumping stations, levees, and weirs, serve the primary purpose of preventing river flooding and protecting residential and agricultural areas from inundation. Ensuring the seamless operation of these facilities is critical to minimizing damage from flooding. To facilitate early detection of anomalies and enable rapid response, this study develops a real-time monitoring-based Decision Support System (DSS) for water resource facilities. The system development process encompassed requirements analysis, business process design, data collection and network architecture, software and functional design, and database and UI/UX design. Core requirements and functions were defined as sensor data collection, real-time status monitoring, weather information integration, early disaster detection and warning, operator decision support, and facility anomaly detection/accident prediction. The business process is designed to store collected data in a central database and utilize AI modules (developed by collaborating research institutions) to generate early disaster detection analytics. The system then transmits alerts to users and provides actionable insights for decision-making. For data acquisition, collection factors for both physical and virtual sensors were defined, and collection programs were developed. Data from pilot facilities are collected and stored in the database in real time. Meteorological data is collected via the Korea Meteorological Administration’s OpenAPI. Based on UI/UX design specifications, prototypes for monitoring and decision-support functions were developed and tested. Future research will focus on identifying improvements to finalize the system for full-scale application at demonstration sites. Through pilot operations, the stability and performance of the system will be validated to ensure its effectiveness in practical water resource management.

ABSTRACT. In this study, water resource management strategies were established based on the results of cloud seeding experiments. First, hourly precipitation time-series data for the past 20 years were collected from representative meteorological stations within the target dam basins to ensure a long-term climatic analysis. Next, the Inter-Event Time Definition (IETD) suitable for the specific characteristics of each site was calculated using an exponential distribution to separate independent rainfall events. During this process, a rainfall threshold (>=1.0 mm) was applied to filter only those events where cloud seeding experiments are feasible. For the separated independent rainfall events, the SCS Curve Number (SCS-CN) method was employed to estimate effective rainfall. Specifically, the potential maximum retention was calculated by applying the Curve Number (CN) a characteristic value determined by reflecting land cover and hydrologic soil groups and the initial abstraction was subsequently defined. The study areas focused on the Boryeong Dam and Yongdam Dam basins. The results indicated that the average number of rainfall events suitable for artificial seeding in both basins was approximately twice per month. Based on the SCS-CN method, the estimated amount of water resources secured per experiment was 0.13 million m3 for Boryeong Dam and 0.68 million m3 for Yongdam Dam. Furthermore, an evaluation of the effects of cloud seeding during past drought cases revealed that up to 14 experiments for Boryeong Dam and 18 experiments for Yongdam Dam would be required per event. These quantitative findings can serve as a fundamental basis for establishing future drought mitigation policies and sustainable water management plans.

ABSTRACT. During July 2020, an anomalously persistent Baiu front induced extreme precipitation over large parts of Japan, resulting in extensive flood damage and loss of life. In Kyushu, rainfall from 4–8 July produced unprecedented 24-, 48-, and 72-hour accumulations at multiple gauging stations. A key characteristic of the event was the high proportion of fatalities associated with flood inundation processes. The Hitoyoshi Basin, located along the middle reach of the Kuma River, experienced severe overbank flooding on 4 July, causing widespread inundation and substantial damage to the built environment. Flood-related human impacts are governed not only by evacuation timing and decision-making but also by hydraulic conditions that directly constrain evacuation feasibility, including rapid stage rise, high depth–velocity combinations, and the structural performance of residential buildings subjected to hydrodynamic loading. Despite these risks, flood actions on buildings are not explicitly addressed in the current Japanese Building Standard Law, and quantitative assessment frameworks for residential safety under inundation remain limited. This study aims to provide technical evidence to support the selection of appropriate horizontal versus vertical evacuation strategies by conducting high-resolution inundation simulations for Hitoyoshi City, Kumamoto Prefecture, and evaluating hazards associated with vehicle operation and residential exposure. Simulation results indicate that locally high flow velocities and short times to inundation onset significantly reduced the available evacuation window, thereby limiting mobility-based evacuation and contributing to increased mortality. Hazard mapping further suggests that areas exhibiting large unit discharge and elevated depth–velocity products correspond to heightened risk for structural failure and washout of housing. While the spatial pattern of estimated collapse hazard is qualitatively consistent with observed damage, the present structural risk index is derived solely from drag force estimations. For robust engineering risk assessment, further developments should incorporate additional loading and failure mechanisms such as buoyancy-induced uplift, hydrostatic pressure differentials due to indoor flooding, and coupled effects of debris impact and progressive structural degradation.

ABSTRACT. Recent climate change has led to rising temperatures and altered precipitation patterns, thereby increasing the frequency and variability of extreme events such as droughts and floods. Unlike other natural disasters, drought is an extreme and severe hazard that causes widespread damage over a long period. Korea’s specific climatic regime, characterized by summer-intensive precipitation, results in recurrent winter and spring water shortages and periodic severe drought events. Droughts are classified into meteorological, agricultural, hydrological, and socioeconomic categories; drought propagation describes the transmission of drought from one type to another over time. Propagated drought events cause more severe damage than non-propagated events due to their substantial impact on water supply systems. In this study, propagation characteristics (i.e., attenuation, lag, and lengthening) were analyzed using sensor-based drought indices derived from observed precipitation and reservoir storage. The study area was classified into 29 clusters based on the primary water source. The results showed that the average number of meteorological and hydrological drought occurrences was 85 and 18, respectively. The average propagation rate from meteorological to hydrological drought was approximately 14%. Regarding propagation characteristics, the average lag time was 5.04 weeks, and the average attenuation was 0.06. Among the propagated events, 92% exhibited lengthening characteristics, with an average lengthening time of 21.33 weeks. For the Chungju Dam basin (C1 cluster), the meteorological drought in 2014 propagated into a hydrological drought with lag and lengthening characteristics, affecting 1,448 people. In the same year, the adjacent Soyanggang Dam basin (C3 cluster) experienced a propagated drought with a longer lag time compared to the C1 cluster, resulting in more severe damage affecting 5,507 people.

ABSTRACT. The Ile Alatau Range, forming part of the Northern Tian Shan, is one of the most climate-sensitive mountain regions of Central Asia, where recent climate change, cryospheric degradation, and changes in precipitation regimes have significantly increased debris-flow hazards. Complex geological and geomorphological conditions, widespread glaciation, extensive moraine complexes, and active erosion processes make the region highly prone to both glacially induced and rainfall-triggered debris flows. Debris-flow events that occurred on 21 July 2023 in the river basins of the Ile Alatau were characterized by a pronounced spatial extent along the latitudinal direction and across multiple altitudinal zones of the mountain range. The events demonstrated a wide diversity of debris-flow genesis, formation mechanisms, flow types, and hydrological characteristics, as well as synchronous activation of debris-flow processes over a large part of the Northern Tian Shan. This regional-scale activation highlights the importance of integrated basin-wide analysis of debris-flow hazards and emphasizes the critical role of systematic monitoring of debris-flow-forming factors. On 21 July 2023, debris flows simultaneously occurred in numerous basins, extending from the Uzyn-Kargaly basin in the west to the Shelek basin in the east, covering a substantial part of the northern macroslope of the Tian Shan. Depending on basin-specific conditions, debris-flow formation involved all altitudinal zones of the range, from glaciated high-mountain areas to low-mountain zones. In some basins, debris-flow generation and transformation processes affected the entire altitudinal profile, whereas in others they were limited to high- and mid-mountain zones or exclusively to low-mountain areas. Peak discharges varied significantly depending on the altitude of debris-flow formation, reaching up to 100 m³/s in high-mountain basins, while in low-mountain zones they were as low as 0.1 m³/s or less. The observed flows included debris-rich, water–sediment, and mud flows generated through erosion–sliding and erosion–transport processes. According to the source of the water component, debris flows were classified as glacial and rainfall-induced, and in several basins different debris-flow types were observed simultaneously. The primary triggering factors were very intense rainfall combined with anomalously high air temperatures and favorable antecedent meteorological conditions, reflecting ongoing climate-change trends in the Northern Tian Shan. The identification of hazardous conditions and the occurrence of debris flows were made possible mainly due to the debris-flow monitoring system operated by Kazselezashchita. The event demonstrates that improving debris-flow forecasting in the Tian Shan strongly depends on advances in quantitative precipitation forecasting and the integration of monitoring data into predictive models.

ABSTRACT. Topographic data are basic datasets required in flood simulation processes. High-resolution elevation data from light detection and ranging (LiDAR) or ground surveys offer the required accuracy (Kumar et al, 2023), but are always impractical due to the excessively high cost involved, especially in inaccessible rural environments. Remote sensing-based Digital Terrain Models (RS-DTMs), derived from satellite data, provide a lighter way out in this gap in understanding. The study aimed to evaluate the topographic fidelity and channel reproducibility of RS-DTM in steep-slope agricultural areas. A comparison study among RS-DTMs, publicly accessible LiDAR data available with the Geospatial Information Authority of Japan (GSI), and precisely surveyed UAV-based LiDAR data, considered as the reference data, was carried out in this study. Based on the results, RS-DTM showed comparable performance to LiDAR for narrow channels (≤3 m) with R² and NSE > 0.9, although MAE and RMSE increased locally due to vegetation interference, mixed pixels, and a few outlier elevations. For wider or deeper reaches, RS-DTM captured subsurface bathymetric detail beyond the water-surface-only LiDAR products, explaining some apparent discrepancies. In conclusion, the findings herein prove that RS-DTM is a capable and cost-effective alternative for the creation of high-resolution elevation data with less dependence on extensive field surveys, for improving flood and hydrologic modeling in regions with limited or unavailable LiDAR data worldwide. The ability of RS-DTM to reproduce LiDAR-comparable topographic patterns, while incorporating bathymetric information critical for channel conveyance, supports its application in reliable hydraulic simulations and scalable flood modeling frameworks, particularly in data-scarce and steep-slope agricultural regions.

ABSTRACT. This study deals with the creation of a highly realistic digital twin of tsunami inundation to enhance residents' evacuation and disaster prevention awareness. Located in southwestern Hokkaido, Muroran City is surrounded by sea on three sides, making it extremely vulnerable to tsunami hazards from large offshore earthquakes. Improving residents' evacuation actions and disaster awareness is a critical challenge in this region. During the Hokkaido Pacific Coast Tsunami Warning issued following the July 30, 2025 earthquake off the Kamchatka Peninsula, a survey in Muroran City revealed that approximately 40% of residents located within the predicted inundation zone did not evacuate. This behavior suggests a misperception of tsunami risk and a diminished sense of crisis due to “normalization bias,” highlighting the growing need for information delivery that effectively appeals to people's sense of urgency. Highly realistic tsunami simulations are effective for improving risk awareness, but their development typically requires substantial funding. To address this challenge, this study proposes a low-cost, high-resolution tsunami inundation digital twin that integrates free software, open data, and visualization techniques from game engines. Tsunami inundation simulations were conducted using the free hydraulic calculation software iRIC, reproducing inundation depths and arrival times along the Muroran City coastline. Terrain and buildings were displayed as a realistic 3D virtual space using the Digital Elevation Model (DEM) from the Basic Map Information, along with PLATEAU and Photorealistic 3DTiles. Subsequently, using Unreal Engine 5, the tsunami simulation results were overlaid onto the three-dimensional virtual space, dynamically visualizing the inundation area during the tsunami, temporal water level changes, water flow, and acoustic effects. Tsunami inundation simulations using iRIC showed good agreement with Muroran City's tsunami hazard map in terms of inundation depth and arrival time. The first tsunami wave reached the coast approximately 40 minutes after the earthquake, followed by rapid inundation. These results were integrated into the UE5 digital twin, visualizing the time-dependent tsunami inundation process. Furthermore, the digital twin utilizing Google 3D provides high realism when viewed from an overhead perspective. Therefore, it is clear that rapid evacuation actions within a limited timeframe are essential during a tsunami event. This study is expected to contribute to improving the accuracy of tsunami risk assessments and raising awareness of the risks. Moving forward, we aim to develop content that incorporates additional evacuation routes and information to promote a higher level of evacuation awareness.

ABSTRACT. Drought is a major natural disaster in China. For a long time, the lack of drought early warning indicators has led to issues such as inaccurate decision-making or excessive responses in drought emergency management. Focusing on drought risk early warning and regulation, this study, based on the theories of critical water resources regulation and risk hedging, proposes a fundamental theory for drought-limited water levels rooted in the emergency regulation of water resource systems. It reveals the evolution patterns of drought and regulatory thresholds, establishes a complete set of determination techniques for drought-limited water levels covering rivers, reservoirs, and lakes, and develops drought defense scheduling technologies along with an active prevention and control framework centered on these drought-limited water levels. The outcomes have been adopted by the Ministry of Water Resources and applied to determining drought-limited water levels at 1,278 sites across the country. They have been widely utilized in both national and local drought resistance practices.

ABSTRACT. Recent climate change has led to an increasing frequency of localized heavy rainfall events and increased rainfall intensity, thereby highlighting the importance of flood estimation in upstream small catchments for effective flood control and safe operation of agricultural reservoirs. However, in the current flood estimation framework, essential hydrological parameters such as the time of concentration and storage coefficient are often determined using empirical formulas. Many of these formulas were originally developed for relatively large watersheds, raising concerns regarding their applicability to small upstream catchments of agricultural reservoirs. In particular, the time of concentration (tc) and storage coefficient (K) have a direct influence on peak discharge, time to peak, and the overall shape of the flood hydrograph, necessitating parameter estimation and validation that appropriately reflect watershed characteristics. This study aims to estimate the time of concentration and storage coefficient of the Clark unit hydrograph for the upstream watershed of the Gwansan Reservoir through calibration based on rainfall–runoff simulation results and to evaluate their applicability. Rainfall–runoff simulations were conducted using the HEC-HMS model, and observed inflow hydrographs at the upstream reservoir were employed as reference data for model calibration. The time of concentration and storage coefficient were calibrated simultaneously, and the optimal parameter set was determined through a comprehensive evaluation of simulation performance in terms of peak discharge, time to peak, and hydrograph shape. The observation-based calibration procedure proposed in this study provides an empirical approach for estimating and validating Clark unit hydrograph parameters in small upstream catchments of agricultural reservoirs. The results are expected to improve the reliability of watershed runoff analysis for design flood estimation and flood response planning in agricultural reservoir management.

ABSTRACT. Anthropogenic emissions introduce nitrates, sulfates, and organic matter into receiving rivers, driving dynamic changes in the redox environment of water-sediment systems and inducing coupled biogeochemical cycles of elements. However, the concealed coupling mechanisms between sulfur and nitrogen cycles regulated by redox gradients within the critical interface of hyporheic zone remain poorly understood. This study focused on a typical hyporheic zone affected by combined agricultural and urban non-point source pollution. Using dual isotopes i.e., δ¹⁵N/δ¹⁸O-NO₃⁻, δ³⁴S/δ¹⁸O-SO₄²⁻, microbial molecular techniques i.e., 16S rRNA gene sequencing, metagenomic MAGs, network analysis, and machine learning models, we elucidated how redox gradients and seasonal dynamics jointly regulate microbial communities to drive coupled sulfur-nitrogen transformations. Results showed nitrate primarily derived from chemical fertilizers (52.1%–69.1%) and domestic sewage (19.7%–37.0%), with sewage contributions rising by ~15% in summer. Redox gradients mediated microbial community structure and functions: sulfur and nitrogen cycles were independent in aerobic zones; denitrification coupled with sulfur oxidation dominated anaerobic zones in winter, while iron reduction-driven DNRA coupled with sulfate reduction prevailed in summer. Concealed sulfur cycling was evidenced by significantly lower δ³⁴S values in sediment sulfate (3.75‰ in winter, 5.44‰ in summer) than in overlying water (12.24‰), and key anaerobic MAGs e.g., Acinetobacter spp.) harboring both gens related to sulfur reduction e.g., aprAB, dsrAB and oxidation e.g., sqr. This study clarifies the coupled mechanisms and ecological effects of concealed sulfur-nitrogen cycles in hyporheic zones under redox gradients, providing theoretical support for region- and season-specific pollution control strategies.

ABSTRACT. Localized heavy rainfall systems, particularly those exhibiting stationary characteristics, pose significant threats to hydro-environmental safety by causing flash floods and urban inundation. Traditional radar-based monitoring often relies on subjective analysis or simple reflectivity thresholds, failing to accurately distinguish the structural and dynamic characteristics of hazardous precipitation systems. This study proposes a novel Hazardous Weather Detection and Forecasting Method that integrates morphological classification with motion vector estimation to objectively identify and predict severe weather events. The proposed framework operates through two complementary processes. First, the morphological analysis utilizes 3-hour cumulative radar rainfall data to define Heavy Rainfall Areas (HRAs). By calculating parameters such as aspect ratio, area, and connectivity, precipitation systems are classified into specific types (e.g., linear, stationary, linear-stationary). Second, the dynamic analysis employs an Optical Flow algorithm to estimate the motion vectors of precipitation echoes. This process identifies "stationary risks" by detecting areas where high-intensity echoes exhibit movement speeds below a critical threshold. Furthermore, the system incorporates a Lagrangian persistence model to perform short-term forecasting (1 to 6 hours), extrapolating the trajectory and intensity of the identified systems. Application of this method to historical heavy rainfall events in Korea demonstrated that combining morphological features with motion vectors significantly improves the detection accuracy of stationary linear precipitation systems compared to conventional methods that rely solely on shape. This automated, quantitative approach is expected to enhance decision-making capabilities for flood defense and disaster response authorities.

Acknowledgements: The research for this paper was carried out under the KICT Research Program (Project no. 20250284–001, Development of Digital Urban Flood Control Technology for the Realization of Flood Safety City) funded by the Ministry of Science and ICT.

ABSTRACT. Purpose of the Work In August 2025, heavy rainfall associated with a linear precipitation zone triggered widespread flooding across Kyushu, Japan. In the Amikake River, Kagoshima Prefecture, this resulted in revetment collapse and scouring behind bridge abutments. This study aims to elucidate these phenomena from a mechanical perspective and provide a quantitative explanation of the contributing factors. Because the affected sites are located within a tidal channel, particular attention is given to how longitudinal water-surface profiles and flow conditions were influenced by tidal levels, leading to a complex failure process.

Key Issues or Problems Addressed The primary issue is that backwater effects specific to tidal channels can govern the external force field during flood events. The observed damage resulted from the simultaneous action of multiple processes: hydro-mechanical external forces (high-velocity zones, flow concentration due to channel curvature, and local acceleration around structures) and internal ground-related factors behind the revetment (increases in pore water pressure, delayed drainage, and reductions in effective stress). It is therefore necessary to consistently relate channel hydraulics to subsurface pressure responses based on a synchronized water-level history.

Methodology or Approach Used Numerical simulations incorporating flood discharge hydrographs and tidal variations were performed and validated against observed channel deformation. Based on these, the influence of tidal conditions on external forces—such as velocity distributions and shear stresses—was evaluated. Additionally, seepage flow analyses were conducted using rapid river-level fluctuations as boundary conditions. The spatio-temporal distributions of pore water pressure and hydraulic gradients were analyzed to assess the effects of drainage delay and prolonged excess pore water pressure on revetment stability.

Results or Conclusions The primary issue is that backwater effects specific to tidal channels can govern the external force field during flood events. The observed damage resulted from the simultaneous action of multiple processes: hydro-mechanical external forces (high-velocity zones, flow concentration due to channel curvature, and local acceleration around structures) and internal ground-related factors behind the revetment (increases in pore water pressure, delayed drainage, and reductions in effective stress). It is therefore necessary to consistently relate channel hydraulics to subsurface pressure responses based on a synchronized water-level history.

ABSTRACT. In recent years, climate change has intensified water-related disasters in Japan. Disaster management not only involves building structures like levees, but also requires initiatives for helping people to evacuate safely by using hazard maps. However, flood risk assessment remains challenging in ungauged basins, particularly those involving small and medium-sized rivers, due to the lack of hydrological observation data. Conventional flood analyses based on physical models require high-resolution input data on rainfall, topography, and geology, and wide-area, high-resolution simulations often entail substantial data acquisition efforts and high computational costs. In contrast, Digital Elevation Models (DEMs) are globally available, and flood risk assessment based solely on topographic information could facilitate rapid and transferable risk evaluation. Therefore, this study proposes a simplified flood risk assessment method using only topographic indices and evaluates its effectiveness by benchmarking against pluvial flood simulation results obtained using the DioVISTA flood analysis system for the Hidonodani River basin in Naruto City, Tokushima Prefecture, Japan. To characterize inundation risks caused not only by extreme events but also by relatively frequent rainfall, we applied design storms with return periods of 2, 5, and 10 years and calculated inundation distributions for each scenario. Using a 5 m resolution DEM, we computed the Topographic Wetness Index (TWI), indicating the potential for water accumulation, and the Topographic Position Index (TPI), reflecting relative topographic relief. We examined the relationship between simulated inundation areas and these topographic indices through GIS-based overlay analysis. Finally, based on these correlations, this study investigated the feasibility of identifying inundation-prone areas relying solely on topographic information.

ABSTRACT. Conventional embankment safety assessments have primarily relied on contact-based electrical resistivity exploration and visual inspections. These methods have significant limitations in steep terrain or difficult-to-access areas, and they also hinder rapid surveys immediately following disasters. In particular, areas downstream of embankments, adjacent water surfaces, and densely vegetated areas often remain blind spots, failing to adequately reflect structural vulnerabilities. Consequently, the need for new, non-contact, high-efficiency survey technologies is growing.

This study aimed to overcome these limitations by proposing a non-contact, high-precision survey technique that integrates drone-based LiDAR, optical imaging, and drone-based Ground-Source Semi-Airborne Transient Electromagnetic (SATEM). The core objective of this study was to develop an integrated drone-based exploration system capable of simultaneously diagnosing both external changes and internal structural abnormalities in reservoir embankments. The system's applicability and reliability were verified through field demonstrations on actual reservoirs.

To simultaneously diagnose both external changes and internal structural abnormalities in reservoir embankments, this study integrated a drone-based LiDAR/optical imaging system with a drone-based time-domain electromagnetic exploration system. For the external exploration, an industrial drone capable of RTK-based high-precision positioning control (Matrice 400 RTK, DJI) and a LiDAR sensor capable of acquiring high-density point clouds (Zenmuse L2, DJI) were used. For the internal exploration, a non-contact time-domain electromagnetic exploration system (Korea Institute of Geoscience and Mineral Resources) combining a ground-based transmitter and a drone-mounted receiver was utilized. This equipment configuration enabled stable data acquisition and internal structural analysis even on steep terrain, near-water areas, and slopes with limited access.

The drone-based precision exploration technique presented in this study complements the limitations of existing contact-based exploration and holds the technological potential to contribute to the advancement of reservoir embankment safety management systems. The non-contact exploration method enables safe and rapid surveys even on steep slopes, near-water areas, and inaccessible areas, which is expected to simultaneously improve survey efficiency and work safety.

ABSTRACT. The present study focuses on the occurrence of Senjo-Kousuitai events in order to examine the improvement in river water level prediction accuracy that has been achieved by incorporating water vapour content into the training data. In recent years, there has been growing concern over the intensification of flood damage due to increased torrential downpours associated with climate change. Consequently, there is an urgent requirement for higher-precision water level forecasting to ensure sufficient lead time for soft disaster prevention and mitigation measures. Whilst physical models representing rainfall-runoff processes are applied to rivers nationwide as a water level forecasting method, a key challenge is that long-term predictions cannot be made with high accuracy due to limitations in rainfall forecast precision. Concurrently, the utilisation of deep learning algorithms for flood forecasting, derived from hydrological observation data such as rainfall and water levels, has garnered significant attention. Numerous reports have attested to the efficacy of these forecasting models, demonstrating a notable degree of accuracy for prediction periods extending up to approximately 3 to 6 hours into the future. It is notable that the occurrence of Senjo-Kousuitai is cited as one factor contributing to significant flood damage in recent years; however, no research examples have focused specifically on water level prediction during Senjo-Kousuitai events. The occurrence of Senjo-Kousuitai is significantly related to the amount of water vapour entering the watershed. It is therefore considered effective, when predicting water levels using deep learning, to incorporate atmospheric water vapour content into the training data. This is predicted to improve prediction accuracy. The model under discussion was implemented as an LSTM (Long Short Term Memory) model using TensorFlow, which possesses libraries for deep learning. The model optimiser employed the Adam algorithm, with the loss function set to mean squared error (MSE). The hyperparameter search was conducted using Bayesian optimisation with Optuna. The incorporation of water vapour flux into the model resulted in an enhancement of the root mean square error (RMSE), the Nash-Sutcliffe efficiency (NSE), and the percentage of the potential energy (PEP%) for predictions of 12-hour-ahead water level at the Senoshita and Arase stations. However, a decline in accuracy was observed at the Katano-se and Obuchi stations. Secondly, while significant improvements in PEP(%) were observed when incorporating water vapour flux, no improvement in TEP was observed in this verification.

ABSTRACT. Recent advancements in AI technology have made it an essential tool across all fields. AI is actively being applied to flood forecasting and warning systems by predicting river water levels. Korea has developed the world's first LSTM-based AI flood forecasting platform, utilizing it for real-time predictions. However, improvements are needed in many areas to further improve prediction accuracy. This study proposes optimization techniques necessary to improve the accuracy of AI flood forecasting by applying various input observation data to an AI river water level prediction model, including data processing, hyperparameter tuning, prediction algorithms, and model architecture. The results are compared and analyzed to propose optimization techniques necessary for improving the accuracy of AI flood forecasting. Data processing methods include adjusting the ratio of flood events to normal events, converting observed data to accumulated values or stationary data, and so on. Hyperparameter tuning involves sensitivity analysis of key hyperparameters, including sequence length, hidden layer dimension, batch size, and dropout rate. In addition to the basic LSTM, the prediction algorithm applies GRU, Conv1d-LSTM, and Attention techniques. Through comparison of Multi-feature, which produces various prediction lead time results with a single model, and Single-feature, which produces a single prediction lead time result with each single model, we propose an optimized AI river water level prediction model.

ABSTRACT. Desalination is one of the most viable solutions for providing sustainable and clean drinking water in regions experiencing water scarcity. This research shows a thorough machine learning (ML) based approach for optimization and performance prediction for a Multi effect desalination (MED) system. To evaluate the system’s performance, MED experiments were conducted at different temperatures which was set at 40°C, 50°C and 60°C. I will make a comparative study analysis of four ML models, Artificial Neural Network (ANN), Support Vector Regression (SVR), Random Forest (RF), and Gradient Boosted Regression (GBR) for the prediction and optimization of distillate production. Significant variations in predictive accuracy and robustness under various operating conditions were revealed by a comparison of the developed ML models. The relative strengths of the modeling approaches were highlighted by evaluation using R², MAE, and RMSE, which also showcases the importance of selecting the right model for accurate MED performance prediction. The best-fit model was then further utilized for the input feature importance and optimization purposes. Feature importance analysis showed that hot water inlet temperature (Thot in) was the most influential parameter, followed by feed water temperature to the steam generator (Tfeed in SG). These two parameters account for roughly 60.81% of the cumulative importance, which indicates that optimizing and precisely controlling these variables should be given top priority to improve MED system performance. A maximum distillate production of 1.27 LPM was predicted by the ML-based optimization, which is a 223.52% increase over the average experimental distillate production of 0.39 LPM. This shows the great potential of data-driven optimization for significantly improving MED system performance.

ABSTRACT. A physics-informed neural network (PINN) model is applied to oscillating hydraulic jump phenomena that occur in a flow over a drop with a trench, and the results are compared with those obtained by a classical neural network (NN). Our simple PINN model gives less accurate predictions than the classical NN, inferring the necessity of balancing different terms in the composite loss function by assigning appropriate weights to the terms. On the other hand, the PINN shows an ability to predict the pressure distribution only from the governing-equation constraints, which is a promising feature for obtaining some variables that are difficult to measure.

ABSTRACT. This study investigates the application of a Physics-Informed Deep Operator Network (PI-DeepONet) for modeling free-surface flows governed by the Shallow Water Equations (SWEs). Unlike conventional data-driven surrogate models that require extensive labeled datasets, the proposed PI-DeepONet is trained using only physical constraints. The network is guided by a physics-informed loss function that enforces the SWE residuals, boundary conditions, and initial conditions, enabling the model to learn flow dynamics directly from governing equations without the need for numerical or experimental reference data.

To evaluate its performance, multiple flow problems are used, covering both steady and transient flow conditions. In each case, the PI-DeepONet is trained simultaneously with multiple sets of boundary and initial conditions to enhance its generalization capability. The trained model accurately predicts key hydrodynamic variables—including water depth and discharge—across a wide range of flow regimes. Quantitative evaluations show low root mean square errors compared with reference solutions, confirming that the PI-DeepONet effectively captures the underlying physics of free-surface flows.

Beyond accuracy, the proposed framework offers significant computational advantages. Once trained, the PI-DeepONet can rapidly predict flow variables at arbitrary spatial and temporal locations without solving the governing equations iteratively, achieving substantial speedups relative to traditional hydraulic solvers. This feature highlights its potential for real-time or large-scale hydrodynamic applications where rapid solution generation is critical.

Nonetheless, the study identifies current limitations in accurately resolving discontinuities and dynamically evolving wet–dry fronts, which remain challenging for operator-based neural architectures. Future work will focus on incorporating adaptive sampling strategies and local refinement mechanisms to improve model fidelity in these complex flow regions. Overall, the PI-DeepONet provides a promising and computationally efficient framework for physics-informed modeling of free-surface flows in hydraulic research.

ABSTRACT. As climate change increases the frequency and intensity of localized heavy rainfall, the vulnerability of small streams and small bridges in sub-basins where discharge and water level can change rapidly over short periods has become more pronounced. This study used field survey data and risk assessment records for small streams and small bridges to develop two modeling schemes: a baseline model that uses only the existing risk assessment items as inputs and an extended model that augments the same items with auxiliary indicators describing structural and environmental characteristics. In addition, we propose an analytical framework that enables a fair comparison of the high-risk facility screening performance of the two schemes under identical preprocessing, data splitting, and evaluation settings. The input variables consisted of item ratings from the risk assessment checklist together with the auxiliary indicators, while the target variable was defined as whether the local government designated the facility as high risk, coded as high risk or non high risk. Multiple classifiers, including logistic regression, random forests, XGBoost, and deep neural networks, were trained and evaluated using stratified five fold cross validation to compare models based on risk assessment items alone with those incorporating auxiliary indicators. The results indicate that the XGBoost model using both risk assessment items and auxiliary indicators achieved higher PR AUC and recall than a rule based criterion such as a risk score of 71 or higher, and it identified more ground truth high risk facilities under the same budget constraint in which only the top one to two percent of facilities could be prioritized for intervention.

Acknowledgements: This work was supported by Korea Environmental Industry & Technology Institute(KEITI) through Climate Resilient R&D Project for Water-Related Disaster Managements, funded by Ministry of Climate, Energy and Environment(MCEE)(RS-2024-00398012).

ABSTRACT. Fire flow reliability is a critical performance requirement in urban water distribution systems (WDS), directly influencing public safety and infrastructure resilience. Conventional WDS design and optimization approaches typically rely on steady-state hydraulic analyses, which may overlook transient pressure behaviors induced by operational events such as pump switching or valve operations. These pressure transients can significantly affect system reliability, pipe integrity, and the effective delivery of fire flows, yet they are rarely incorporated into optimization frameworks.

This study proposes a transient-aware multi-objective optimization framework for water distribution system design that explicitly accounts for both steady-state and transient hydraulic performance. The framework integrates extended period simulation for fire flow assessment with pressure transient analysis to evaluate system behavior under dynamic operating conditions. Two competing objectives are considered: minimization of capital investment costs and maximization of fire flow reliability, defined through pressure-based performance metrics under both normal and transient conditions.

A multi-objective evolutionary optimization algorithm is employed to explore trade-offs between cost and reliability while identifying Pareto-optimal design solutions. The proposed approach is demonstrated using a benchmark urban water distribution network, where pipe diameters and network configurations are optimized. Results indicate that designs optimized solely under steady-state assumptions may exhibit inadequate performance when transient effects are considered, potentially leading to underestimated risks. In contrast, the transient-aware optimization identifies solutions that provide improved pressure stability and fire flow reliability with only marginal increases in capital cost.

The findings highlight the importance of incorporating transient hydraulic behavior into WDS optimization and design processes, particularly for systems requiring high levels of reliability under emergency conditions. The proposed framework offers a practical and extensible methodology for utility engineers and decision-makers to enhance the robustness and resilience of urban water supply systems, supporting more informed infrastructure investment and risk-aware planning.

ABSTRACT. Levee breaches can lead to rapid and widespread flooding, requiring timely and reliable flood prediction to minimize impacts. While hydrodynamic models are widely used to simulate flood propagation, their high computational costs and time requirements limit real-time applications. To address this challenge, this study proposes a ConvLSTM-based deep learning surrogate model capable of predicting the spatiotemporal evolution of flood depth and extent in real-time following levee breach events. The model takes the breach location as an input feature, enabling dynamic predictions across various breach scenarios – in contrast to previous studies that assumed fixed the breach locations. We simulated a historical flood event in Korea using the Hydrologic Engineering Center’s River Analysis System (HEC-RAS) to generate a time-series flood depth extent and depth with 5-minute temporal and 10-meter spatial resolutions. It covers 17 distinct breach scenarios combined with varying rainfall intensities. The surrogate model is trained on this data to learn dynamics of flood propagation. In addition, final flood extents were compared with a Synthetic Aperture Radar (SAR) image to assess spatial agreement. The proposed model achieves a root mean square error (RMSE) of 0.03 m for flood depth and a nash-sutcliffe efficiency (NSE) of 0.75 and a critical success index (CSE) for spatiotemporal flood propagation, while reducing computational time by a factor of over 1600 compared to traditional hydrodynamic simulations. These results, validated through additional experiments in the Osong region of Korea, highlight the capability of the proposed surrogate model to enable real-time flood forecasting across diverse levee breach conditions, thereby supporting rapid emergency response and timely risk assessment. Acknowledgement: This work was supported by the National Research Foundation of Korea (NRF) grant, funded by the Korean government through the Ministry of Science and ICT (RS-2024-00457308, RS-2024-00456724), and by a grant from the Korea Environment Industry & Technology Institute (KEITI) through the R&D Program for Innovative Flood Protection Technologies against Climate Crisis, funded by the Korea Ministry of Climate, Energy and Environment(MCEE) (No. RS-2023-00218873).

ABSTRACT. Water distribution networks are the lifelines of urban communities, yet pipe bursts remain a critical global challenge—one that also demands attention in Hong Kong. According to official statistics from the Water Supplies Department, there were 27 pipe burst incidents citywide in 2024. Though the number has dropped sharply from historical levels, such incidents still lead to localized potable water loss, cause temporary traffic disruptions in affected areas, and result in property damage for nearby residents and businesses. These incidents are exacerbated by Hong Kong’s unique environment (high humidity and frequent temperature fluctuations) and the diversity of pipe materials in use (including elastic galvanized iron and viscoelastic polyethylene, which behave differently under stress). However, existing leak detection methods adopt a “one-size-fits-all” approach that overlooks material-specific burst mechanisms, limiting their overall reliability in early warning and prevention. To address this gap, our research develops a targeted solution for burst detection by integrating advanced mathematical modeling, experimental validation, and AI-driven analysis. We first derive analytical models to simulate burst-induced transient pressure signals, explicitly accounting for the distinct behaviors of elastic and viscoelastic pipes. These models are rigorously tested through numerical simulations and laboratory experiments using real pipeline setups at The Hong Kong Polytechnic University, ensuring alignment with real-world conditions. We then combine these physics-based models with Physics-Informed Neural Networks (PINNs)—a cutting-edge AI approach that embeds engineering principles into machine learning—to create a robust detection system. This hybrid framework eliminates the need for large-scale field data collection and enhances accuracy by leveraging both theoretical insights and data-driven learning. Our goal is to precisely identify burst locations and sizes in real time, even for Hong Kong’s complex, multi-material pipeline network. By overcoming the limitations of conventional methods, this research will provide water utilities with a cost-effective, reliable tool to reduce water loss, minimize disruptions, and protect critical infrastructure—strengthening the resilience of Hong Kong’s water supply system for the future.

ABSTRACT. The purpose of this study is to evaluate the operational potential of AI-driven hydrological modelling and its ability to complement traditional physics-based Global Hydrological Models (GHMs). While Long Short-Term Memory (LSTM) networks have shown promise in rainfall–runoff modelling, questions remain about their transferability to new regions and performance relative to well-calibrated GHMs. This work addresses the following critical challenges. It focuses on developing a robust, local workflow independent of cloud-based platforms to ensure scalability and ease of adaptation to new regions. It tackles the transferability of pretrained global or regional LSTM models to entirely different hydrological contexts while maintaining accuracy. The study benchmarks regionally fine-tuned LSTM models against a well-calibrated physics-based GHM developed at DHI to identify strengths and limitations. Using the Mekong Basin as a case study and leveraging a unique non-public dataset, the study implemented a scalable workflow for data preparation and model training. ERA5 meteorological data and HydroATLAS catchment attributes were processed following the CARAVAN standard for each of 60 sub-catchments. Three fine-tuning strategies were tested on a pretrained global LSTM model: (1) single-basin fine-tuning, (2) regional model training across all basins, and (3) hierarchical fine-tuning from regional to basin level. Each approach underwent hyperparameter optimization and was evaluated using KGE and NSE metrics. LSTM performance was systematically compared against the DHI-GHM to quantify strengths, limitations, and operational feasibility. Results show that fine-tuning LSTM models on both regional and local streamflow improved performance compared to local fine-tuning, with median KGE increasing from 0.65 to 0.72. This result does not match the overall accuracy of the DHI-GHM in the test period, which yields a median KGE of 0.75. However, the fine-tuned LSTM outperformed the physics-based model in all catchments with poorly described processes (e.g. irrigation abstraction and infiltration after overtopping), with even performance in all other well-calibrated catchments. The performance gap is even narrower when expressed in terms of NSE, as the LSTM model outperformed the DHI-GHM in terms of mean NSE but not median NSE. This study concludes that fine-tuned LSTM models can match or outperform physics-based GHMs in many cases, particularly where traditional models struggle with complex processes like irrigation and infiltration. By combining a scalable workflow, transfer learning strategies, and rigorous benchmarking, the research demonstrates the operational potential of neural hydrology for regionalization, paving the way for future hybrid and forecasting applications.

ABSTRACT. Flood control operation plans contain extensive rule-based knowledge that is essential for reservoir regulation under extreme hydrological conditions. However, their predominantly natural-language form makes them difficult to integrate into data-driven flood simulation and decision-support models. To address this challenge, this study proposes a plan-driven flood simulation framework, termed the Knowledge Graph–Physics Fusion Bidirectional Temporal Convolutional Network (KGPF-BiTCN), which integrates structured operational knowledge, deep learning, and hydrological physical constraints. First, operational elements and constraint conditions are automatically extracted from reservoir flood control plans and organized into a queryable knowledge graph. Based on this structured representation, a plan–scenario semantic mapping mechanism is developed to identify appropriate operational strategies under different inflow and forecast conditions, thereby generating corresponding release or restriction constraints. These constraints are embedded into a bidirectional temporal convolutional network enhanced with a spatio-temporal attention mechanism, enabling the model to capture complex temporal dependencies while maintaining consistency with operational rules. In addition, the water balance principle and the Muskingum flood-routing method are incorporated to ensure physical plausibility of the simulated flood processes. The proposed framework is applied to the Xiaolangdi–Huayuankou reach in the middle Yellow River using hydrological observations from 1969 to 2020. Compared with a baseline Temporal Convolutional Network, the proposed model shows substantial performance improvements, with the Nash–Sutcliffe efficiency increasing from 0.9157 to 0.9669 in the test set, while RMSE and MAE are reduced by 37.36% and 44.75%, respectively. The model also demonstrates superior capability in capturing rapid discharge variations during flood peaks. The results indicate that integrating structured operational knowledge with physics-informed deep learning can significantly enhance the accuracy, robustness, and operational consistency of flood simulations. The proposed approach provides a promising solution for intelligent reservoir flood-control operation and decision support aimed at improving resilience against water-related hazards in the context of climate change.

ABSTRACT. Accurate monthly river discharge prediction remains a major challenge in data-scarce regions, where limited hydrometric records constrain the applicability of conventional physically-based models and large-scale data-driven approaches. This study develops an interpretable, hydro-informed Random Forest (RF) framework to improve discharge forecasting under such conditions, explicitly integrating temporal memory, flow regime dynamics, and seasonal behavior. Three modeling scenarios were evaluated using a ten-year monthly discharge record (120 observations) with an 80%–20% chronological train–test split. Scenario 1 uses lagged discharges (Qₜ₋₁, Qₜ₋₂, Qₜ₋₃) to represent short-term hydrological memory. Scenario 2 augments Scenario 1 with sinusoidal seasonal encodings (sin/cos of month) to capture annual periodicity. Scenario 3, the most comprehensive, includes lagged discharges, monthly changes (ΔQ), rolling mean and standard deviation, and sinusoidal seasonal features, allowing the RF model to account simultaneously for short-term dynamics, regime transitions, and seasonal forcing, while maintaining physical interpretability.The RF models were built with 100–300 trees, maximum depth of 4–8, and minimum samples per split of 2–6, optimized via 5-fold cross-validation. Feature importance and SHAP analysis were used to interpret the contribution of each predictor. Results demonstrate that Scenario 3 substantially outperforms Scenario 2, achieving an R² of 0.93 compared to 0.84 and an RMSE of 87.8 m³/s compared to 145.6 m³/s on the independent test set. The improvement is attributable to the inclusion of hydrologically meaningful state and transition variables rather than increased model complexity. Residual analysis confirms better representation of both high-flow and low-flow conditions and reduced bias. These findings highlight that interpretable machine learning models, when informed by hydrological reasoning, can deliver high predictive skill even with limited data. The framework offers a practical and transparent solution for operational river flow forecasting, bridging the gap between conventional hydroinformatics, data science methodologies, and black-box AI models.

ABSTRACT. Artificial neural networks and intelligent technologies provide a novel methodological pathway to overcome the limitations of traditional models in water demand and runoff prediction, particularly their insufficient capacity to capture nonlinear processes and respond to complex driving mechanisms. This study focuses on the Jiziwan region of the Yellow River as a typical research area and develops a dual-case framework for water demand prediction and runoff simulation. For water demand prediction, daily water consumption data from a representative coal chemical enterprise from 2021 to 2024 were used to construct LSTM and GRU neural network models, forming an enterprise-scale intelligent water demand prediction system. Results indicate that the MA-GRU model at a daily scale significantly outperforms traditional weekly-scale models in both prediction accuracy and stability, effectively supporting refined water management and operational decision-making at the enterprise level. For runoff prediction, considering the complex hydrological processes influenced by urbanization, agricultural development, water resource management, and climate change, an LSTM rainfall–runoff model was developed to simulate and predict these processes with high precision. The model effectively captures the nonlinear relationships between rainfall and runoff and the long-term dependencies in time series, while revealing variations in runoff evolution under different climate scenarios. Overall, artificial intelligence and neural network techniques provide critical technical support for accurate water resource prediction, risk regulation, and optimized allocation in the Jiziwan region, offering significant theoretical and practical implications for sustainable regional water management.

ABSTRACT. Accurate water-body detection is a fundamental task in remote sensing, providing essential data for various applications including water resource management, and flood risk analysis. While multispectral satellite imagery provides rich spectral information for these tasks, its practical utility is often severely limited by cloud contamination, particularly during extreme weather events. This limitation severely constrains the reliability of conventional deep learning models, which often fail to account for cloud-occluded regions. Under cloud-covered observations, water-body detection becomes particularly challenging. In this paper, we propose Prithvi-CLOMA, a novel framework that integrates cloud-aware patch masking with a frozen Vision Transformer (ViT) foundation model for robust water-body detection under cloudy conditions. Our approach leverages NASA/IBM’s Prithvi-EO-2.0 as a pre-trained spatiotemporal encoder to extract rich multi-temporal features from Sentinel-2 imagery without task-specific fine-tuning of the backbone.

The proposed architecture introduces three key components: (1) a Cloud Patch Masker that identifies cloud-contaminated 16×16 patches using the Sentinel-2 cloud mask and replaces them with learnable mask tokens (similar to the masking strategy used in Prithvi), allowing the encoder to infer context-aware latent representations over cloud-contaminated regions through contextual attention; (2) a Temporal Difference Head that captures refined change features, effectively distinguishing flood-induced water expansion from seasonal or atmospheric noise; and (3) a lightweight convolutional decoder that fuses these representations to generate pixel-level segmentation maps at the original spatial resolution.

Notably, our framework can estimate cloud-robust flood-water extent at the target time (e.g., three previous point of time Sentinel-2 acquisitions), making it suitable for near-real-time disaster response when cloud-free observations are unavailable. Overall, prithvi-CLOMA bridges the gap between foundation-model representations and cloud-covered multispectral monitoring, enabling more dependable flood inundation mapping in practice.

ABSTRACT. In arid regions, the challenges posed by rainfall data availability, missing data, and limited historical records significantly affect hydrological modeling studies and climate change assessments. For various hydrology applications, it is essential to implement advanced techniques in order to obtain a complete dataset series. This study explores the implementation of multiple machine learning techniques to address the complexity of filling daily rainfall data for 88 rainfall stations in the Al-Batinah region of Oman, covering the period from 1993 to 2024. The machine learning models applied in this study include Multiple Linear Regression (MLR), Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Regression (SVR), and Gradient-Boosting Trees (GBT). A non-clustering approach is used as well as a clustering approach as part of the methodology. In the first method, rainfall stations are not clustered, while in the second method, optimal cluster numbers are calculated using K-means clustering. The target station utilizes the nearby rainfall station data located within a 50 km radius with the highest correlation coefficients. A novel Ensemble Fusion Model has been applied to improve the efficacy of multiple predictive models, including the RF Fusion Model (RF) and Multi-Model Super Ensemble Fusion Model (MMSE). The estimation approaches are further enhanced and evaluated by Bayesian optimization of hyperparameters, dataset imputation utilizing Multiple Imputation by Chained Equations (MICE), and Leave-One-Year-Out (LOYO) cross-validation. Based on the results, it can be concluded that the GBT model performs the best in both cluster and non-cluster approaches. A further benefit of applying Ensemble Fusion Models to rainfall gap-filling methods is that the coefficient of determination (R2) for clustering and non-clustering approaches increases to 22.5% and 22.2%, respectively.

ABSTRACT. The authors have developed a Artificial Neural Network(ANN) method to predict the water level or discharge of a river 24 hours ahead for the safety management of river construction. In this method, the inputs to ANN are rainfall distribution from the past 24 hours to the next 24 hours and the observed water levels from the past 6 hours. However, when there is insufficient training data due to a lack of observation for the target watershed, the prediction accuracy decreases. Then, the method to obtain the training data for the prediction by means of selecting similar basins was developed. This method follows these steps: 1) Identifying basin-specific characteristics such as basin area, total stream length, overall gradient, averaged annual cumulative rainfall, forestry coverage ratio, and area ratio of andosol. 2) Classifying the discharge characteristics of 126 dams into 10 classes using the K-means method. 3) Constructing a Decision Tree linking the basin-specific characteristics and the discharge characteristics. 4) Determining the class of the target basin by inputting the basin-specific characteristics into the Decision Tree. 5) Applying the data classified into the same class of the target site to the training data of that. Using this method, the inflow rates were predicted targeting 10 dams in Japan by using a dataset of 126 dams. Note that the inflow rate is defined as the inflow volume divided by the basin area. Focusing on the inflow events where inflow volume exceeded the certain threshold, the observations and predictions were compared. For the dam with the highest prediction accuracy, Nash coefficients ranged from 0.70 to 0.98 and RMSEs were below 1 mm/h. For most other dams, Nash coefficients ranged from 0.5 to 0.7, and RMSEs were between 1 and 3 mm/h. On the other hand, there were two dams where the predicted values did not increase even when the observed values increased, with Nash coefficient below zero and RMSE exceeding 5 mm/h. In these dams, the hydrographs of the selected basins were significantly different from that of the predicted basin, which was considered the reason for low prediction accuracy.

ABSTRACT. Increasing hydro-environmental risks under climate variability necessitate flood prediction approaches that remain effective beyond the confines of dense observational networks. Nevertheless, many established hydrological models continue to rely on intensive site-specific calibration, limiting their practical utility in ungauged basins and under emerging extreme rainfall regimes. Addressing this challenge, commonly recognized as the Prediction in Ungauged Basins (PUB) problem, this study proposes an innovative hybrid modeling framework designed to enhance flash flood predictability while supporting sustainable and transferable hydrological practices for unexperienced rainfall conditions. The proposed framework reformulates the rainfall–runoff process through a dual-structure representation that distinguishes overland flow propagation from effective rainfall generation. Surface runoff is predicted using a dynamic wave–based instantaneous unit hydrograph (DIUH), parameterized solely from globally accessible digital elevation and land-cover information. This physics-consistent configuration enables the explicit representation of nonlinear overland flow behavior and surface resistance while circumventing the prerequisite for local discharge observations. Complementing this component, effective rainfall is derived using a machine learning model trained on historical observed precipitation, catchment attributes, and discharge data. By explicitly estimating infiltration losses, this data-driven module converts total rainfall into effective precipitation, enabling robust cross-regional generalization of hydrological response characteristics. Simulations across independent test basins demonstrate that the framework delivers reliable predictive performance in data-limited environments. Importantly, the model exhibits stable and coherent responses during extreme flood events, indicating strong extrapolation capability under nonstationary rainfall-runoff conditions.

ABSTRACT. Accurate streamflow prediction is fundamental for water resource management and disaster response. However, predicting streamflow with station-based meteorological observations faces challenges due to low spatial density. In contrast, gridded meteorological data provide spatially continuous information, leading to improved streamflow prediction accuracy. Deep learning (DL) models have been widely adopted in water management and mostly use precipitation as an input. Therefore, this study tests whether gridded precipitation improves the predictive accuracy of DL models for streamflow in the Miho River Watershed, South Korea. Modified Korean Parameter-elevation Regression on Independent Slopes Model (MK-PRISM) is used as gridded precipitation. The MK-PRISM data with 1 km spatial resolution consider elevation, topographic facet, and coastal proximity. This study utilizes six meteorological variables: precipitation, average temperature, maximum temperature, minimum temperature, wind speed, and relative humidity. Three DL models, Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Convolutional Neural Networks-LSTM (CNN-LSTM), are used in this study. Five experimental cases are developed for this study. Cases 1 through 4 utilize LSTM and Bi-LSTM, while Case 5 implements a CNN-LSTM. Case 1 uses station-averaged data across the watershed. Case 2 employs the average of MK-PRISM at the watershed level. Case 3 uses meteorological data from individual stations. Case 4 utilizes the average of MK-PRISM at the sub-basin level. Finally, Case 5 employs a CNN-LSTM to use the original format of MK-PRISM as input data. The results of this study will demonstrate the advantages of gridded precipitation to predict streamflow with DL models and propose a suitable format of gridded precipitation.

ABSTRACT. Abstract

This study proposes a hybrid model combining the Hydrological Simulation Program–FORTRAN (HSPF) with a Spatio-Temporal Graph Neural Network (Stem-GNN) to enhance 1-day-ahead Chl-a forecasting. Compared to HSPF–XGBoost and HSPF–Dual Input Neural Network models, the HSPF–Stem-GNN achieved the best validation performance (R² = 0.78, MAPE = 21.17%) and highest F1-score (0.97) for bloom classification. Permutation Feature Importance confirmed the key influence of temperature and simulated Chl-a on predictions. This framework offers a scalable, interpretable tool for real-time Chl-a prediction in stream systems.

1. Introduction

Chl-a serves as a key indicator of algal bloom risk, and short-term forecasting is crucial for effective water quality management. While physically based models like HSPF simulate hydrological and water quality processes, they often fall short in predicting biological variables due to structural and calibration limitations. We propose a hybrid approach that integrates HSPF with a Spatio-Temporal Graph Neural Network (Stem-GNN) to enhance 1-day-ahead Chl-a predictions by leveraging both physical simulations and data-driven learning.

2. Methodology

We used daily data from the Pungyeongjeon Watershed (South Korea) from 2013 to 2022, including observed Chl-a, meteorological data, and HSPF-simulated Chl-a. Three hybrid models were tested: HSPF–XGBoost, HSPF–Dual Input Neural Network (DI-NN), and HSPF–Stem-GNN. Simulated and meteorological features served as model inputs, with each architecture designed to predict Chl-a one day ahead. Model performance was assessed using R², MAPE, NSE, and PBIAS for regression, and Precision, Recall, and F1-score for algal bloom classification. Permutation Feature Importance was applied to interpret the influence of input variables.

3. Results

The HSPF–Stem-GNN model achieved the highest validation performance among the three hybrid models, with an R² of 0.78 and a MAPE of 21.17% for 1-day-ahead Chl-a prediction. It also demonstrated superior classification capability, achieving an F1-score of 0.97 in identifying high-risk bloom events. Feature importance analysis identified air temperature and HSPF-simulated Chl-a as the most influential drivers.

4. Conclusion

This study proposes a hybrid modeling framework that combines HSPF simulations with advanced machine learning to improve short-term Chl-a prediction. The HSPF–Stem-GNN model demonstrated superior performance in both regression and classification tasks, with strong potential for early-warning applications and practical use in water quality management.

This work was supported by the Management Technology for Groundwater Dams in Water Supply Vulnerable Areas Program of the Korea Environmental Industry & Technology Institute (KEITI), funded by the Ministry of Environment (MOE) (RS-2025-01842973).

ABSTRACT. The impacts of climate change on ecosystems and socio-economic systems are intensifying, increasing the importance of efficient water resources management and strategic planning. Accurate precipitation prediction is therefore a fundamental requirement for hydrological analysis and water-related decision making. Traditionally, physics-based numerical weather prediction models have been widely used for precipitation forecasting. More recently, artificial intelligence (AI)-based global weather prediction models have shown promising performance that can exceed conventional approaches. However, comprehensive evaluations of precipitation prediction skill and associated biases in these AI-based models is limited, and their suitability for hydrological and hydrometeorological applications remains insufficiently understood. To address this gap, this study quantitatively assesses the precipitation prediction performance and extreme precipitation representation of an AI-based global weather prediction model. This study focuses on the United States, using GraphCast as the AI-based model. As reference datasets, European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) and the observation-based gridded dataset Daymet are used, enabling a robust comparison among model simulations, reanalysis products, and observations. Precipitation prediction performance is evaluated using statistical metrics, and the ability of the model to represent the frequency and intensity of extreme precipitation events is analyzed. Furthermore, the spatial characteristics of predicted precipitation are analyzed to assess the consistency of precipitation patterns. The findings of this study provide a comprehensive understanding of the characteristics and practical applicability of AI-based global weather prediction models for precipitation forecasting and hydrometeorological applications. In addition, this study also offers insights to their potential role as reliable precipitation inputs for hydrological modelling.

Acknowledgment: This work was supported by Korea Environment Industry & Technology Institute(KEITI) through R&D Program for Innovative Flood Protection Technologies against Climate Crisis Program(or Project), funded by Korea Ministry of Environment(MOE)(RS-2023-00218873).

ABSTRACT. Urban flood forecasting is challenging due to the complex interplay of weather, hydrology, and city infrastructure. However, existing research often overlooks urban-scale urban flood prediction, has insufficiently analyzed the impact of road networks within sub-cities on flood prediction results, leading to inadequate urban flood prediction under the combined influence of rainfall, terrain spatial characteristics, and road distribution. To address these issues, we present TUF-TransUNet, a transformer-based deep learning model for city-scale flood prediction. This model integrates road network data with meteorological inputs, which can effectively enhance the understanding and prediction capabilities of complex temporal and spatial patterns in meteorological data. Model performance is illustrated through a case study in Hong Kong. Results show that TUF-TransUNet improves both forecast lead time and spatial accuracy, enabling the prediction of urban flood peaks up to six hours earlier than standard U-Net and transformer models without infrastructure data. Our results suggest that including road network data enables the model to accurately represent urban hydrological pathways, which are often dominated by transportation infrastructure rather than natural river channels. This is crucial for predicting flood propagation in highly urbanized areas, where conventional models that ignore such features may misrepresent flow routes and underestimate flood risk. This study highlights the critical role of transportation infrastructure in urban flood modeling and demonstrates that incorporating such data substantially improves both the accuracy and lead time of flood predictions, providing a practical tool for enhancing flood resilience in cities.

ABSTRACT. Accurate data is crucial for hydraulic infrastructure planning, water resource management, and hydrological disaster mitigation. Daily river-flow data are essential for this. However, this poses a serious challenge for developing countries where hydrological data measurement infrastructure is minimal, if not nonexistent. Furthermore, river flow data often exhibit limited recording duration, discontinuities, or incompleteness. The study area is located within the Serang–Lusi Watershed, Central Java, Indonesia. This study aims to develop an Artificial Neural Network (ANN) model using the backpropagation algorithm to predict daily river flow from available discharge measurements at river stations. The primary issue addressed in this study is the difficulty in representing nonlinear hydrological responses using traditional conceptual or empirical models when discharge data is limited. The model input comprises hydrological data for the watershed, including daily rainfall, evapotranspiration, and runoff coefficients. The target output is daily river-flow data recorded at observation points. The ANN was trained and evaluated in Python, with a systematic evaluation of the network architecture and training parameters to identify the optimal configuration across the entire dataset. The results showed that the model performed very well in capturing the rainfall-runoff relationship in the study area. The optimal model achieved a Mean Squared Error (MSE) of 0.032 during training and 0.047 during testing. Furthermore, the overall prediction accuracy reached approximately 98%, indicating that the proposed model is highly reliable for estimating daily streamflow. Therefore, this model can address the need for accurate streamflow data despite the limitations encountered in the field.

ABSTRACT. Compound drought and heatwaves (CDHWs) pose greater catastrophic impacts on the breadth of sectors in socio-environmental systems, including water management, agriculture, wildlife management, and heat-related illnesses or mortality, than individual extremes. Through the lens of land-atmospheric feedback, drought and heatwaves can act as a mutual attractor, amplifying the severity of each event or their compound impacts due to their coupling mechanisms. Thus, global warming, resulting in longer periods of intermittent rainfall and higher temperatures, is expected to increase the occurrence frequency of CDHWs and make their impacts more threatening. Under these increasingly challenging conditions, practitioners need early warning systems or long-term forecasting techniques for CDHWs to implement proactive mitigation measures. Korea has experienced more frequent and severe droughts and heatwaves, elevating the risk to sustainable water supply and increasing the likelihood of wildfire occurrences. Therefore, this study develops forecasting methods for early warning of CDHWs in South Korea using Long Short-Term Memory (LSTM) with the maximal overlap discrete wavelet transform (MODWT), hereafter called MODWT-LSTM, and Autoformer, which is the state-of-the-art forecasting deep learning method. Since the evolution of CDHWs is attributed to the interaction among hydrometeorological variables of different temporal scales, their temporal characteristics are too complex for deep learning models to learn essential temporal patterns for long-term forecasting. Therefore, the application of a signal decomposition method to make a raw time series simpler sub-series with a distinctive frequency range can facilitate forecasting CDHWs at an extended lead time. While MODWT-LSTM uses a decomposition method as a pre-process before the application of LSTM, Autoformer has a decomposition module inside the model. This study compares MODWT-LSTM and Autoformer and suggest the best method in forecasting CDHWs at an extended lead time for South Korea. To facilitate forecasting CDHWs and analyzing characteristics thereof, we develop a copula-based CDHWs Index (CDHWI) using precipitation and maximum temperature data from the Modified Korean Parameter-elevation Regressions on Independent Slopes Model (MK-PRISM) over the period 2000-2019 covering South Korea with a spatial resolution of 1km. Thus, MODWT-LSTM and Autoformer can be applied to the CDHWI for long-term forecasting with spatial seamlessness over the study area. By utilizing the proposed deep learning model to forecast CDHWI, practitioners can assess the times and regions with the highest risk of CDHWs, enabling preparedness for anticipated extreme dry-hot events.

ABSTRACT. Accelerating climate change has been accompanied by increases in the frequency and intensity of short-duration extreme rainfall, elevating the risk of urban water-related hazards such as pluvial flooding. In small urban catchments, inundation can occur within minutes and runoff response times are correspondingly short. Accordingly, high-temporal-resolution rainfall information is increasingly required for disaster preparedness and urban hydrologic analysis. In this study, an AI-based temporal downscaling method for rainfall time series is proposed using a diffusion-based conditional generative modeling framework. Within the proposed framework, hourly rainfall observations are used as conditioning information, while observed 10-minute rainfall series are used as target data. During inference, 10-minute rainfall series are generated to remain consistent with the given hourly totals. A conditional diffusion model is adopted, in that complex rainfall distributions can be represented through a probabilistic generation process in which samples are progressively refined from noise under the provided condition. This formulation is intended to capture within-hour temporal structure, including rainfall intermittency and short-lived high-intensity bursts, which are often challenging to represent using conventional mean-based temporal disaggregation approaches. To ensure physical consistency at the coarse temporal scale, a post-processing constraint is applied after generation. Mass conservation is enforced by matching the sum of generated 10-minute rainfall within each hour to the corresponding observed hourly totals. The ultimate objective is to establish a physically consistent rainfall generation scheme that better reflects rapid hydrologic responses in urban basins under extreme rainfall. The proposed approach is expected to provide improved forcing inputs for urban hydrologic modeling and flood-risk assessment, thereby supporting more reliable evaluation and decision-making for rapid-response catchments.

ABSTRACT. Climate change-driven increases in extreme rainfall events across Korea are amplifying soil erosion risks, making accurate estimation of rainfall erosivity (R-factor) essential for RUSLE modeling in hydro-environmental planning and disaster mitigation. While diverse computational tools exist, their relative performance in handling local monsoon data and computational demands has not been thoroughly compared. This study systematically reviewed five key R-factor calculators using Generative AI: RIST (USDA ARS tool for intensity summarization), WERM (web-based Korean erosivity module), WREC (SNU web decision support system), RainfallErosivityFactor (open-source R package), and simplified regression methods. Employing harmonized hourly and sub-hourly rainfall datasets from Korean Meteorological Administration stations, the programs were assessed on criteria including data input flexibility and coefficient of determination (R²). Methodologies encompassed dataset standardization, validation of the R-factor via EI30 indices, and classification of storm events. This evaluation sought to identify optimal tools for integration into AI-enhanced frameworks to bolster regional resilience against water-induced hazards in Korea.

ABSTRACT. Accurate and timely river water level prediction is essential for effective flood management and disaster mitigation. However, developing reliable prediction models remains challenging due to heterogeneous input variable configurations across monitoring stations and limited historical data availability at target locations. This study proposes a novel variable-agnostic masked transfer learning framework that enables robust river water level forecasting by leveraging knowledge transfer from multiple reference stations to data-sparse target stations. The proposed model consists of three key components: (1) Variable-Agnostic Processors that handle eight hydrological input types (water level, discharge, rainfall, cumulative rainfall, inflow, release, tide, and watershed-averaged rainfall) with automatic masking for unavailable variables; (2) Adaptive Gating Mechanism that dynamically learns variable importance based on hydrological context; and (3) Transformer encoder for capturing long-range temporal dependencies. The model employs a two-stage transfer learning strategy: pre-training on over 100 reference stations to learn general hydrological patterns, followed by fine-tuning with layer-specific learning rates for target station adaptation. The framework was applied to the Yeongsan River basin in South Korea, predicting water levels up to 6 hours ahead (36 lead times at 10-minute intervals) using 4-hour historical data. Comparative experiments were conducted against baseline models including LSTM, GRU, 1D Convolutional LSTM, and Transfer Learning-based LSTM. Results demonstrate that the proposed approach significantly outperforms baseline models, particularly during extreme water level events and for stations with incomplete variable availability. The adaptive gating mechanism provides interpretability by revealing the relative contribution of different input variables under varying hydrological conditions.

ABSTRACT. Recent advances in deep learning have enabled the widespread use of data-driven models, such as Long Short-Term Memory (LSTM), for real-time water level forecasting. Although these models often demonstrate high predictive accuracy under normal hydrological conditions, their reliability during extreme flood events remains uncertain. This study aims to identify the limitations of LSTM-based flood forecasting under extreme conditions and to examine whether scenario-augmented training can partially mitigate these limitations. A key challenge addressed in this study is the scarcity of extreme flood observations in historical datasets, which restricts the ability of purely data-driven models to generalize beyond previously observed flood magnitudes. Using water level data observed at monitoring stations, LSTM models were trained and evaluated across different flood magnitudes and lead times. Model performance was assessed with particular emphasis on peak water levels and rapidly rising flood stages, where prediction errors are most critical for operational flood warning. To investigate the impact of extreme-event data scarcity, synthetic extreme flood scenarios were generated using a physically consistent hydrological model and applied to monitoring stations. These synthetic scenarios were combined with observed datasets to construct scenario-augmented training samples. The performance of LSTM models trained with augmented datasets was compared against observation-only baselines to evaluate changes in predictive stability and extreme-event performance. Preliminary analysis indicates that scenario-augmented models exhibit improved predictive stability during extreme flood events compared to observation-only baselines.

ABSTRACT. Recent climate change has increased the frequency and intensity of droughts and floods worldwide, leading to heightened hydrological risks. Therefore, reliable streamflow prediction approaches are essential for efficient water resources management and strengthening drought response strategies. This study presents a methodological comparison of physics-based and data-driven modeling approaches for streamflow simulation in the Hoengseong Dam basin. A physics-based hydrological model, the Dynamic Water Resources Assessment Tool (DWAT), was implemented to represent watershed hydrological processes with explicit consideration of surface–groundwater interactions. A data-driven deep learning model based on a Gated Recurrent Unit (GRU) was constructed to model temporal patterns in streamflow. While both models were applied to the same catchment, they were developed using different types of input data, reflecting the fundamental differences between process-based and data-driven approaches. To examine the characteristics of the simulated streamflow, a set of hydrological analysis methods, including flow duration curve analysis, low-flow period inflow assessment, and drought-period time series analysis was adopted. The objective of this study is not to evaluate prediction performance, but to establish a comparative framework for examining how physics-based and data-driven models differ in structure, data requirements, and analytical behavior under various hydrological conditions. This framework is expected to support informed selection of streamflow modeling approaches in future studies, particularly for catchments with limited hydrological observations and for the development of hydrological datasets for drought response.

This work was supported by Korea Environment Industry & Technology Institute(KEITI) through Water Management Program for Drought Project, funded by Korea Ministry of Climate, Energy and Environment(MCEE).(2022003610002).

ABSTRACT. Water-level forecasting at a watershed scale is crucial for flood preparedness, reservoir operations, and watershed management, yet many watersheds lack long-term gauge records, with varying temporal coverages. We developed a clustering-based machine learning framework for one- and three-day-ahead water level prediction in the Saemangeum Watershed, South Korea. Daily water levels from 25 monitoring stations (2020–2024) were used to group stations into six hydrologically similar clusters via a k-means clustering algorithm using wavelet/entropy features. Within each cluster, we trained wavelet-transformed multilayer perceptron (WT–MLP) models under two representative-station strategies: (i) training at the cluster centroid station and (ii) training at the station with the longest continuous record. Model performance was evaluated with Nash–Sutcliffe efficiency (NSE), RMSE, and R². Predictive uncertainty was quantified using Monte Carlo input perturbation to derive 95% prediction intervals. Across clusters, the longest-record strategy consistently outperformed the centroid-based strategy, demonstrating the value of richer training histories when record lengths vary among stations. When the trained WT–MLP was transferred to all member stations, watershed-wide forecasting skill remained high: for one-day-ahead predictions, the longest-record strategy achieved mean NSE = 0.97 and RMSE = 0.06 across 25 stations; for three-day-ahead predictions, mean NSE = 0.83 and RMSE = 0.14. Uncertainty estimates were generally better calibrated for models trained on longer records, producing more stable coverage and narrower intervals than centroid-based models. By training a single WT–MLP per cluster and transferring it to stations with limited observations, the proposed framework reduces computational burden while maintaining strong predictive skill. This approach offers a practical, scalable solution for multi-station water level forecasting and early warning systems in data-scarce environments.

ABSTRACT. Reservoirs and dams are essential components of water resource management that significantly alter natural river flow regimes. Accurate prediction of streamflow in human-regulated catchments requires explicit consideration of reservoir operations. Although recent machine learning based hydrological models demonstrate high predictive accuracy, they often lack physical interpretability and do not explicitly represent anthropogenic flow regulation. In this study, we developed a physics-informed, differentiable hydrological modeling framework that explicitly integrates reservoir operations, and applied it to the Hangang river basin. A Target Storage and Release-Based (TSRB) reservoir operation scheme was transcribed into a differentiable, rule-based module within the differentiable parameter-learning Hydrologiska Byråns Vattenbalansavdelning (dPL-HBV) model. Model parameters were trained through end-to-end learning using a coupled LSTM network and station-observed streamflow data. A combined loss function, consisting of RMSE and RMSE computed on log-transformed streamflow, was used to balance performance across high- and low-flow regimes. Results showed that incorporating reservoir operations successfully improved the model’s ability to simulate streamflow in regulated, human-impacted catchments. By maintaining physical interpretability of anthropogenic flow regulation, the proposed framework provides a useful tool for supporting reservoir operation and dam management decision-making.

Acknowledgements : This work was supported by the National Research Foundation of Korea, which was funded by the Ministry of Science, ICT and Future Planning (RS-2024-00457308) and Korea Environment Industry & Technology Institute (KEITI) through ‘Wetland Ecosystem Value Evaluation and Carbon Absorption Value Promotion Technology Development Project’, funded by Korea Ministry of Climate, Energy and Environment (MCEE) (RS-2022-KE002030).

ABSTRACT. Monitoring algal growth in aquatic environments is essential for maintaining ecosystem health and for sustainable water supplement. Chlorophyll-a (Chl-a), a key indicator of algal biomass, has traditionally been monitored by in situ observations. However, these approaches are limited in spatial coverage and unsuitable for long-term monitoring. Satellite remote sensing provides an effective alternative for observing Chl-a over large areas. However, optical satellite imagery is frequently affected by cloud-induced data gaps, which impede the generation of spatiotemporally continuous Chl-a datasets. Consequently, gap-filling has become an essential methodological component in satellite-based Chl-a research.

Recent advances in foundation models pretrained on large-scale datasets offer new opportunities for gap-filling. The time-series foundation model MOMENT learns temporal continuity and variability from sequential observations to estimate data gaps along the time dimension. In contrast, the Earth observation foundation model Prithvi-EO learns spatial context from multispectral satellite imagery, enabling the reconstruction of missing regions using surrounding pixels. As these models are pretrained on fundamentally different information structures, they address missing data through complementary mechanisms and are both applicable to satellite-based Chl-a gap-filling.

In this study, we compare the gap-filling performance of MOMENT and Prithvi-EO by exploiting their respective temporal and spatial strengths. The study focuses on Daecheong Dam, South Korea, using Harmonized Landsat–Sentinel-2 (HLS) S30 data from 2016 to 2024. Cloud-induced Chl-a data gaps were reconstructed using each model. MOMENT leveraged fixed-length Chl-a time series (T = 512) to estimate missing observations over time, whereas Prithvi-EO reconstructed missing regions from single-date multispectral imagery using spatial structure. Both models were applied to the same observation period, and reconstructed Chl-a values were evaluated using standard evaluation metrics.

This study systematically evaluates the gap-filling characteristics and potential performance differences of these two foundation models. Specifically, we examine how temporal continuity–based modeling and spatial context–based reconstruction influence Chl-a gap-filling under cloud-induced missing conditions. By analyzing their respective strengths and limitations, this study provides insights into the applicability of foundation model–based approaches for satellite-derived Chl-a monitoring. Ultimately, our findings inform the design of spatiotemporally integrated gap-filling frameworks for long-term water quality monitoring and satellite-based water quality risk detection.

ABSTRACT. Artificial intelligence (AI) techniques have been extensively applied to predict lake eutrophication due to their advantages over traditional methods which eliminate the need to identify physicochemical and biological factors affecting algae growth, a complex process that involves time-consuming and labour-intensive experiments. The dynamic capability of AI applications allows timely responses to algal blooms, potentially mitigating their impact on aquatic ecosystems and water quality. The study utilized artificial intelligence (AI) techniques to develop a predictive tool for lake eutrophication. An Artificial Neural Network (ANN) approach was adopted to model lake eutrophication by predicting the lake trophic state index, Modified Lamparelli index (2025), a composite indicator derived from total phosphorus (TP) and chlorophyll a (Chl-a) concentrations developed under the study. The ANN model was trained using the Levenberg-Marquardt algorithm, with 11 years of water quality data as the input variable and the Modified Lamparelli index (2025) as the output variable. The analysis used various input variables to assess the relationship with the six target variable models. The model’s performance was evaluated using Pearson’s correlation coefficient (R), with values above 0.90 indicating excellent predictive capability. The study demonstrates that predictive models for lake eutrophication achieve optimal accuracy. The AI model can be further enhanced by continuously learning from new data, thereby refining management practices as conditions change over time.

ABSTRACT. AI-based river water level forecasting models are increasingly applied at flood alert stations to support flood warning and operational decision-making. Although these models often show satisfactory performance, prediction results may become unstable or less reliable depending on local hydrological conditions and flood situations. Such behavior indicates that AI-based forecasting models have inherent limitations that should be understood from an operational perspective. This study explores the limitations of AI-based river water level forecasting models and examines how prediction results can be interpreted differently according to operational purposes. Rather than focusing on model improvement or accuracy enhancement, this research adopts an exploratory approach to investigate situations in which prediction performance degrades and to assess the operational relevance of such predictions. LSTM-based river water level forecasting results were analyzed across multiple flood alert stations, with attention to prediction stability, lead-time availability, and behavior during flood events. The analysis suggests that prediction limitations frequently occur under specific hydrological and operational conditions, such as rapid water level fluctuations, complex regulation effects from hydraulic structures, and constrained observation environments. However, these limitations do not necessarily eliminate the operational value of the forecasts. In several cases, predictions with limited accuracy in continuous water level reproduction still provide useful information for early detection of threshold exceedance or situational awareness during flood events. The results highlight that the usefulness of AI-based flood forecasts depends on operational objectives and evaluation perspectives. This study emphasizes the need to interpret AI-based forecasting results by considering both model limitations and intended operational use. In addition, the findings suggest the potential value of explainable approaches to support the interpretation of AI-based forecasts under uncertain conditions. This study provides practical insights for the safer and more informed application of AI-based river water level forecasting models in real-world flood warning operations.

ABSTRACT. Integrated water resources management is becoming increasingly challenging due to climate change and socioeconomic development, with dams playing a key role in balancing the complex relation between environmental and socioeconomic demands. Projecting future dam operations is therefore essential for climate change adaptation and future water security. In this context, this study proposes an integrated methodology to assess future dam operation by combining traditional approaches with Machine Learning techniques. It integrates climate change impacts on inflows and water uses, as well as projections of hydroelectric production. The methodology is applied to the Conchas and Salas dams, located in the Spanish part of the transboundary Iberian Peninsula Lima River Basin. The expected operation of the two dams was assessed using the Long Short-Term Memory (LSTM) model, a Machine Learning technique. To such, expected inflows and hydroelectric production in each dam were considered as input variables. Inflows were simulated with the HEC-HMS hydrological model, considering a 20-model climate ensemble, RCP4.5 and RCP8.5 scenarios, and two future periods (2011–2040 and 2041–2070). The inflows were then subtracted of the expected upstream water demands, estimated considering the expected development of the different sectors and the impacts of climate change. The hydroelectric production was estimated through the definition of simplified scenarios of the evolution of electric sector in the Iberian Peninsula, focusing on the expected coordination between hydroelectricity and other renewable energy sources. The results showed that, in general, for both climate scenarios, between 2011 and 2070, it is projected an increase of the volume stored in the two reservoirs during autumn, winter, and spring. This will translate in a change of the current operating patterns, as volume storage will tend to be higher in the winter. Outflows are projected to increase in autumn and spring for the Conchas dam, while monthly outflows at the Salas dam are expected to increase under both RCPs. The proposed integrated methodology demonstrates a strong potential to be a decision-support tool for future dam management under climate change. By jointly incorporating climate-driven inflow projections, evolving water demands, and hydroelectric production scenarios within a Machine Learning framework, it enables a comprehensive and dynamic assessment of dam operation projections. It is also flexible and transferable, allowing its application to other reservoirs and river basins, representing a valuable contribution to improving the robustness and adaptability of water resources planning and dam operation in a changing climate and society.

ABSTRACT. River discharge is a fundamental hydrologic variable that underpins decision-making for water-resources operations (e.g., dams, weirs, and intake facilities) as well as flood forecasting/response and drought assessment. Discharge reflects basin-scale hydrologic conditions, and its rapid temporal variation during flood events driven by extreme rainfall is closely associated with flood impacts. Nevertheless, discharge observations in Korea rely largely on gauge-based measurements, which often suffer from observational gaps due to limited spatial coverage, maintenance costs, and reduced field accessibility during floods. In addition, site-specific stage–discharge relationships (rating curves) require periodic recalibration because they are sensitive to bed changes, channel modifications, and hydraulic structures (e.g., bridges, levees, and weirs), and their uncertainty can increase substantially during extreme floods. These limitations constrain the spatiotemporal applicability of gauge-based discharge information and introduce uncertainty in continuous monitoring and emergency response for ungauged reaches. To address these challenges, satellite-based wide-area observations have been increasingly used to monitor river water dynamics and hydrologic states. Sentinel-1 Synthetic aperture radar (SAR) enables reliable observations under cloudy and adverse weather conditions, including flood periods, whereas the optical sensor Geostationary Ocean Color Imager-II (GOCI-II) provides high temporal resolution for capturing continuous water-body changes. In this study, we propose a deep learning–based discharge estimation algorithm that fuses Sentinel-1 SAR and GOCI-II data, incorporating spatiotemporal learning to represent rising and recession limbs during flood events. We further integrate Height Above Nearest Drainage (HAND) and digital elevation model (DEM) to account for topographic constraints on inundation extent and include meteorological information to better represent the temporal variability of rainfall–runoff responses. Beyond direct comparison with observed discharge, we evaluate physical consistency by converting predicted discharge to water level using rating curves and examining its agreement with observed stage variations. Overall, the proposed framework indicates the potential of an all-weather, satellite-based discharge estimation system applicable during flood periods and suggests its utility as complementary information for basin operations and flood/drought warning systems.

ABSTRACT. Hydrologic forecasting at the subseasonal-to-annual scale in large transboundary basins presents challenges due to complex process interactions, extensive surface-water coverage, and large ungauged regions. These factors limit the predictive skill of traditional models and complicate operational water resource management. This study explores the potential of artificial intelligence to enhance hydrologic forecasting by comparing a data-driven Long Short-Term Memory (LSTM) neural network with the process-based Large Basin Runoff Model (LBRM) in the Laurentian Great Lakes basin. Both modeling frameworks were developed and calibrated using Climate Forecast System Reanalysis (CFSR) forcing data for 1979–2023 to simulate basin-wide runoff. Seasonal forecasting skill was then evaluated using archived 9-month Climate Forecast System (CFS) forecasts as climate inputs. Model performance was assessed at both sub-basin and regional scales, with particular emphasis on total runoff consistency from a water-balance perspective relevant to seasonal water supply forecasting. Results show contrasting strengths between the AI-driven and process-based approaches, with the LSTM capturing nonlinear climate–runoff relationships and the LBRM providing physically interpretable basin responses. The comparison highlights opportunities for hybrid modeling strategies that combine data-driven learning with physical process representation to improve forecast skill at seasonal to annual timescales. Using the Great Lakes as a testbed, this work advances AI-driven hydrologic modeling for large, complex basins and supports the development of more reliable long-lead forecasting tools for operational water management in transboundary systems worldwide.

ABSTRACT. Numerical weather prediction models are actively used for daily weather forecasting. However, predicting localized phenomena such as linear precipitation zones and guerrilla rainstorms remains a major challenge. Physics-based models require computationally expensive high-resolution grids to capture these details. Consequently, statistical approaches using AI have attracted attention as cost-effective alternatives. Although AI-based correction methods have been proposed, existing methods often lack sufficient accuracy. The objective of this research is to develop an improved correction algorithm based on Convolutional Neural Networks and to verify its accuracy and characteristics in correcting numerical weather prediction models. In this study, we trained a Convolutional Neural Network using forecast data from the numerical weather prediction model as input and actual observed values as ground truth. Consequently, the numerical weather prediction corrected by the CNN is obtained as the output. Specifically, forecast data with a 5-kilometer resolution were resized to match the 2.5-kilometer resolution of the observed radar data. We specifically employed a 3D-CNN architecture to enable learning in the temporal dimension, addressing the limitations of 2D methods that fail to capture historical context. In addition to precipitation, other variables, specifically wind speed components, were used as input to enhance physical consistency. Validation was conducted using data from October 2019. Through this research, we constructed a modified model based on CNN and successfully improved accuracy by strictly adjusting input parameters and the learning range. The results showed that the use of the 3D-CNN allowed the model to successfully reproduce the flow of rain clouds. Furthermore, adding wind speed as an input variable significantly improved prediction accuracy compared to using precipitation alone. The results demonstrate that the 3D-CNN output approaches the observed ground truth more closely than the original numerical forecasts. We created a model capable of reproducing temporally continuous phenomena using 3D-CNN.

ABSTRACT. Agricultural drought forecasting is critical for optimizing water resources and ensuring food security under increasing climate variability, yet accurate medium-term to long-term prediction remains challenging due to complex nonlinear interactions among hydro-climatic and vegetation processes. Although deep learning models represented by CNN-LSTM and ConvLSTM have achieved promising performance in spatiotemporal drought forecasting, they still face limitations in simultaneously capturing fine-scale spatial heterogeneity, long-term temporal dependencies, and complex multi-source environmental interactions. In particular, conventional ConvLSTM structures are prone to information attenuation during multi-step forecasting and lack mechanisms to dynamically emphasize drought-sensitive regions. To address these limitations, this study develops an attention-enhanced Convolutional Long Short-Term Memory framework (CNN_AttConvLSTM) that integrates convolutional layers for hierarchical spatial feature extraction, with ConvLSTM layers for modeling spatiotemporal evolution and the spatial attention mechanism to dynamically emphasize drought-prone areas. Multi-source remotely-sensed variables—including NDVI, land surface temperature (LST), soil moisture (SM), precipitation (Pr), NDWI, and VHI—were incorporated following standardized preprocessing, feature selection using Pearson correlation analysis, and normalization. The model was trained with five-fold cross-validation, and performance was evaluated using mean absolute error (MAE), root mean squared error (RMSE), coefficient of determination (R²), and structural similarity index (SSIM) for both single and multiple-step forecasts. Results indicate that the proposed model outperforms baseline methods (CNN, CNN-LSTM, and ConvLSTM), achieving 15.7% and 21.8% reductions in MAE and RMSE, respectively, for single-step prediction, with R² reaching 0.933. In seven-month forecasts, the model exhibits minimal error growth (~20% increase in MAE), demonstrating strong long-term stability and robustness. Spatial analyses further confirm reduced prediction errors and improved structural similarity in drought hotspots. These findings suggest that integrating spatial attention and multi-source data fusion substantially enhances drought forecasting performance, providing valuable technical support for early warning systems and adaptive water resource management in agricultural plains.

ABSTRACT. This study aims to enhance the effectiveness of integrated watershed flood management by developing a dam inflow estimation framework that combines bias-corrected rainfall forecasts, extensive in situ hydrological observations, and climate change–based scenarios. Based on the estimated inflows, an AI-based dam operation model was developed to optimize reservoir storage and minimize downstream impacts caused by flood releases, and its practical applicability was systematically validated. Using multiple plausible flood scenarios, the proposed model demonstrated that proactive reservoir operation based on forecast-corrected rainfall and scenario-driven inflows significantly reduces peak release rates and controls maximum reservoir water levels below the normal full supply level, compared to conventional real-time operation. This indicates a substantial improvement in dam operational safety. A case study applied to Yongdam Dam in South Korea showed that downstream releases could be maintained below the design flood discharge of the downstream river, while 9 out of 10 randomly selected scenarios successfully controlled reservoir water levels below the predefined risk threshold, achieving water level reductions exceeding 5%. These results suggest that stable reservoir operation and water supply can be ensured through predictive dam operation without excessively lowering reservoir storage during the flood season. The proposed AI-based dam operation framework is expected to effectively maintain dam safety and minimize downstream flood damage even under extreme rainfall events.

ABSTRACT. Accurate estimation of reservoir evaporation is critical for water resources management in tropical regions where evaporative losses are substantial. This study compares the following machine learning models: Least Squares Support Vector Regression (LS-SVR), Random Forest (RF), and XGBoost for monthly evaporation prediction at Hirakud Reservoir using 33 years of hydro-meteorological and operational data (1985–2017) for training and 7 years (2018–2024) for independent validation, with inputs including temperature, precipitation, humidity, wind speed, solar radiation, inflow, and release. Feature engineering incorporated nonlinear transformations and moving averages to represent temporal dependencies. Random Forest achieved the best performance (NSE 0.90 training, 0.76 testing; RMSE 8.27 MCM), outperforming XGBoost (NSE 0.69, RMSE 9.36 MCM) and LS-SVR (NSE 0.66, RMSE 9.80 MCM), and reliably captured seasonal and inter-annual variability without structural drift. Pre-monsoon months (March–May) exhibited the highest evaporation, averaging 72.65 MCM per month, 64 percent higher than winter losses. Solar radiation emerged as the dominant driver (40.1 percent relative importance), with reservoir releases significantly influencing evaporation during peak demand periods. The results demonstrate the robustness of ensemble tree-based models, particularly Random Forest, for operational reservoir evaporation forecasting and climate-adaptation planning in tropical reservoir systems.

ABSTRACT. Peatlands are highly sensitive to hydrological change, yet their representation in groundwater flow models remains challenging due to the strong vertical heterogeneity and compressibility of peat soils. One of the key sources of uncertainty is hydraulic conductivity (k), which varies not only with peat type but also with depth and water-table position, both of which are affected by drainage and climate-driven fluctuations. This study presents a workflow that links empirical depth–conductivity relationships derived from laboratory experiments and in situ measurements with data-driven regression and optimization techniques. Rather than replacing process understanding, these methods are used to support parameter calibration and to explore the transferability of depth-dependent k relationships to peat zones lacking direct measurements. The analysis is based on fen peat from the Biebrza Valley (northeastern Poland), a large near-natural peatland complex. Laboratory tests and field observations were used to quantify changes in vertical hydraulic conductivity resulting from peat desaturation and variations in groundwater levels. Empirical depth–k functions were then examined using supervised regression approaches to assess their potential to reduce calibration misfit and parameter uncertainty in groundwater flow modeling. The proposed approach was further applied to extrapolate depth-dependent hydraulic properties across peatland zones with limited data, allowing an evaluation of uncertainty associated with parameter transfer. The results indicate that combining empirical knowledge of peat hydraulic behavior with data-driven analysis can improve the consistency of model parameterization and support the representation of vertical flow gradients at the peat–aquifer interface. The study provides a methodological contribution relevant to the application of artificial intelligence in hydrological modeling, particularly in settings where data availability is limited and strong process constraints must be preserved. The framework may support future predictive simulations of peatland response to hydrological and climatic change.

ABSTRACT. Floods are among the most destructive natural hazards, posing serious threats to human life, property, and critical infrastructure. Accurate and timely flood forecasting is therefore essential for risk mitigation, enabling authorities to implement preventive measures, optimize resource allocation, and issue early warnings. Conventional flood forecasting approaches, including hydrological modeling and numerical simulation, are primarily based on physical principles and require detailed parameterization of complex watershed processes. These methods often demand extensive calibration efforts and are limited by the availability and quality of input data. In contrast, recent advances in artificial intelligence and machine learning have facilitated the development of data-driven approaches, such as Long Short-Term Memory (LSTM) models, which have shown considerable promise for flood forecasting applications. A PC-based desktop application implementing flood forecasting using the aforementioned LSTM algorithm was successfully developed under the Annual Operation Plan (AOP) of Working Group of Hydrology in Typhoon Committee (TC) for improving flood forecasting modeling through AI technology. The system serves as one of the initiatives aimed at supporting authorities and practitioners in selected TC member countries: Lao PDR, Malaysia, the Philippines, and Thailand in transitioning from manpower-dependent practices to a more digital and automated approach to hydrological data management. The developed system consists of four main modules. Module 1 focuses on data preprocessing, including feature selection and preparation of input and output variables, and also provides cross-correlation analysis to examine lagged relationships between input features and target variables. Module 2 implements automated hyper-parameter optimization using a Bayesian search algorithm. Module 3 covers LSTM model training and testing, while Module 4 performs flood forecasting simulations based on the trained model. In addition, the automated search for optimal LSTM hyper-parameters using the Bayesian search algorithm significantly enhances system usability, enabling effective application by users with minimal background in artificial intelligence or programming. The system is currently undergoing final inspection prior to its release to the selected TC member countries for flood forecasting application.

ABSTRACT. Floods in river basins can occur rapidly, causing substantial damage and leaving limited time for an effective response. Accurate streamflow forecasting at short lead times is therefore essential for reducing flood impacts, particularly in densely populated or flood-prone regions. This study aims to investigate the role of river network information in streamflow prediction using AI-driven models, specifically Graph WaveNet (GWNET) and the entity-aware long short-term memory (EA-LSTM) model. Spatial dependencies are represented using four adjacency matrices constructed based on Euclidean or hydrological distance, with and without downstream directionality. One-hour-ahead streamflow predictions were conducted based on hourly precipitation, streamflow, and water level data from 20 stations in the Yeongsan and Seomjin River basins, South Korea, from 2016 to 2023. EA-LSTM achieved a higher prediction accuracy for upstream stations, while GWNET was more effective for downstream predictions, possibly due to its ability to capture upstream flow propagation. Incorporating downstream directionality improved the performance of GWNET further, suggesting that directional connectivity plays an important role in streamflow predictions. In areas where the hydrological-to-Euclidean distance ratio exceeded 1.3, the inclusion of hydrological distance in the adjacency matrix tended to produce more accurate predictions, likely because it offered a better representation of the actual river pathways. This study demonstrates the importance of incorporating river network structure into data-driven models and provides insights into how spatial configuration and directional connectivity influence streamflow forecasting, thereby contributing to the development of more effective early-warning systems.

Acknowledgements This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Ministry of Science and ICT (RS-2024-00457308) and the Korea Environment Industry & Technology Institute (KEITI) through the Climate Resilient R&D Project for Water-Related Disaster Management, funded by the Korea Ministry of Climate, Energy and Environment (MCEE) (RS-2023-00218873).

ABSTRACT. Water temperature is a key controlling factor in river ecosystems. Owing to data scarcity and methodological challenges, the cumulative thermal effects of cascade reservoirs remain difficult to quantify. This study develops a deep learning–based framework to reconstruct natural river water temperatures and to assess cumulative thermal impacts under joint multi-reservoir regulation. First, meteorological forcing data are used as inputs to evaluate the capability of four deep learning models (MLP, KAN, LSTM, and TFT) in modeling river water temperature. Second, the model with the most stable overall performance is adopted to reconstruct the river’s natural thermal state. Finally, we assess how observed water temperatures deviate from the natural baseline across different stages of cascade development. Results indicate that, for water temperature prediction, deep learning models that represent nonlinear processes exhibit more stable overall performance. The proposed approach can provide methodological support for ecological management in reservoir-impacted river reaches.

ABSTRACT. This study investigates the application of artificial intelligence for automated river space utilization surveys (RSUS), addressing the limitations of traditional manual patrols, particularly during night-time and for detailed user analysis. Leveraging the extensive network of CCTV (Closed Circuit Television) cameras installed by Japan's Ministry of Land, Infrastructure, Transport and Tourism (MLIT), we employ a multimodal approach combining the object detection model YOLOv8 (You Only Look Once version 8) and the large language-and-vision assistant (LLaVA). Our research evaluates the performance of these models in detecting humans and recognizing their actions (e.g., walking, cycling, photography) and attributes (gender, age) in riverine environments under varying conditions. Key findings indicate that detection accuracy for both models is highly dependent on image brightness, with optimal performance observed at a brightness value between 0.5 and 0.6, while recognition significantly degrades in low-light scenarios. Furthermore, while LLaVA demonstrates strong capability in gender recognition and certain actions, it shows limitations in accurately discerning age groups (e.g., misclassifying elderly individuals) and can misinterpret actions in low-light conditions (e.g., confusing smartphone use with golfing). The study also confirms that image resolution is a critical factor, with 4K video providing more reliable detection and recognition at longer distances compared to standard Full-HD. The results validate the potential of integrating object detection and multimodal AI models to supplement conventional RSUS, enabling near real-time, 24/7 monitoring and detailed analysis of river space usage, thereby offering a scalable solution for effective river management and infrastructure evaluation. This approach can support MLIT to evaluate the usage situation of the public riverine area, which will lead to necessary improvement or change in this area.

ABSTRACT. Predicting streamflow in river basins heavily modified by human infrastructure remains a major challenge for hydrologic modeling because dam and weir operations introduce strong nonlinearities into flow responses. Standard models often struggle to represent these regulated dynamics, particularly during extreme rainfall events when operational decisions critically influence flood peaks and timing. This study proposes an ensemble-based data assimilation (DA) framework for the highly regulated Nakdong River Basin in South Korea, which is governed by a dense network of dams and multipurpose weirs. We couple WRF-Hydro with the Data Assimilation Research Testbed (DART) to update both natural hydrologic states and managed reservoir states in an operationally controlled environment. The framework is applied to the August 2022 extreme rainfall event by assimilating streamflow and reservoir water-level observations to reduce errors in peak timing and flood magnitude associated with human operations. Beyond hydraulic variables, we also explore the integration of multivariate datasets such as satellite-based soil moisture to better constrain antecedent land-surface conditions, which can be particularly important in ungauged upstream areas where runoff generation is sensitive to initial wetness states. We assess both the limitations and benefits of these DA strategies, examining how the combination of reservoir operations and land-surface constraints can improve flood prediction skill in highly regulated basins.

ABSTRACT. Land cover change is a key driver of future groundwater recharge and hydrologic-cycle projections because it directly controls the partitioning of precipitation into infiltration, runoff, and evapotranspiration. To support AI-powered hydroinformatics applications, this study develops an AI-driven land cover change prediction workflow that produces consistent future land cover maps as scenario-ready spatial inputs. Using land cover maps from 1990, 2000, and 2010, the proposed approach learns transition behavior and spatial expansion patterns and predicts land cover change for 2020–2050 while integrating class-level areal demand constraints to maintain quantity consistency. Input variables include pixel-based topographic and accessibility factors (elevation, slope, and distance to roads and urban areas), neighborhood pattern metrics describing surrounding land cover composition, and temporal information represented by the target year. Transition potential is estimated with a cellular automata coupled multi-layer perceptron model (CA-MLP), and future land cover maps are allocated according to the learned transition relationships. Class-wise areal adjustments are then applied so that predicted areas match observed or planned areal demands, ensuring both spatial realism and total-area consistency. Model performance is evaluated through hindcasting (training on 1990–2000 and reproducing 2010) using confusion-matrix-based accuracy measures and spatial agreement metrics. The resulting future land cover maps provide practical, scenario-ready inputs for subsequent hydrologic and groundwater studies and demonstrate the value of AI-powered spatial modeling for hydroinformatics workflows.

Funding: The Research for this paper was carried out under the "2026 Groundwater Basic Survey Project", funded by the Ministry of Climate, Energy and Environment.

ABSTRACT. Per- and polyfluoroalkyl substances (PFAS) represent a major environmental and public health concern because of their extreme persistence, high mobility, bioaccumulative behavior, and associated toxicity. Identifying PFAS sources is therefore critical for environmental forensics and remediation planning. However, many existing source-allocation approaches rely on a large number of target compounds and often fail to distinguish between environmental matrices, such as soil and water. These limitations increase analytical costs and may introduce uncertainty into source attribution.

In this study, matrix-specific machine-learning classification models were developed to distinguish PFAS contamination derived from aqueous film-forming foam (AFFF) from non-AFFF sources in both soil and water environments. A comprehensive dataset was assembled from peer-reviewed studies published between 2012 and 2024. The dataset included 673 soil samples (437 AFFF-impacted and 263 non-AFFF-impacted), 520 water samples (379 AFFF-impacted and 141 non-AFFF-impacted), and compositional data from 111 original AFFF formulations. Concentration data for 12 legacy PFAS compounds were log-transformed to address skewed distributions, and non-detect values were treated using appropriate imputation methods.

A total of fifteen supervised classification algorithms were trained using the H2O AutoML framework, which incorporates automated hyperparameter optimization and ten-fold cross-validation. Model performance was evaluated using multiple metrics, including area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. For water samples, the optimal classifier was a Gradient Boosting Machine, which achieved an AUC of 0.9864 and an overall accuracy of 0.8929, with high sensitivity (0.9286) and specificity (0.8571). For soil samples, a Distributed Random Forest model performed best, yielding an AUC of 0.9936 and an accuracy of 0.9787, with both sensitivity and specificity exceeding 0.97.

Feature-importance analysis indicated that PFOS, PFHxS, and PFPeS were the most influential indicators for water-based classification, whereas PFHxS, PFPeA, and PFOS were most important for soil-based models. Further stepwise variable-reduction analysis demonstrated that robust classification performance (accuracy > 0.92) could be maintained using a reduced set of compounds—nine PFAS for water and six for soil—supporting the feasibility of a cost-effective sentinel-compound approach.

Overall, the proposed matrix-specific machine-learning framework substantially reduces analytical complexity while preserving high classification accuracy. This approach offers a practical tool for PFAS source tracking and can be readily extended to emerging PFAS and additional environmental media, thereby supporting more targeted and efficient remediation strategies.

ABSTRACT. River waste, particularly plastic debris such as PET bottles, not only detracts from the aesthetic value of rivers and coastlines but also poses a serious threat to ecosystems by directly causing mortality in aquatic organisms and by promoting bioaccumulation through the food chain. In addition, large accumulations of riverine waste can obstruct water flow, thereby increasing concerns regarding local flood control and river management. In Japan, visual inspections by river management personnel are routinely conducted during normal water stages to assess the conditions of river floodplains, embankments, and water gates. These inspections also collect information on river usage and the surrounding natural environment, as well as detect illegal waste dumping. However, the areas that can be visually inspected by personnel are limited because of the extensive spatial scale of river systems. Consequently, recent studies have investigated the use of UAV-acquired imagery for river condition assessment. Furthermore, approaches incorporating high-precision artificial intelligence (AI) techniques are increasingly being considered to improve the efficiency of illegal dumping detection. This study compares the detection performance of the YOLOv5 model using two generative AI-based data augmentation approaches: InstantMesh, a 3D model–based method, and Stable Diffusion, an image-generation–based method. These approaches were evaluated for detecting objects such as cans, PET bottles, and plastic bags in aerial imagery captured at an altitude of 60 m. Results showed that the 3D modeling approach (InstantMesh) achieved significantly higher overall detection rates (76–92%), particularly excelling in the detection of small objects and yellow plastic bags. However, it was prone to false positives, such as misidentifying grass as litter. In contrast, Stable Diffusion demonstrated strengths in detecting transparent plastic bags but showed lower overall accuracy (58–71%) and struggled with small-object detection. These results indicate that the controllable, 3D-based InstantMesh approach provides a more effective and reliable solution for high-precision aerial litter detection than a Stable Diffusion–based approach, particularly when object features are well defined in advance.

ABSTRACT. Snow Water Equivalent (SWE) represents the water volume released during snowpack melt and thus indicates the volume of water that will flow into rivers and streams as the snow melts. SWE can be estimated using various approaches, including in-situ measurements, remote sensing, and physically based models. Each method has certain limitations, such as high costs, low spatiotemporal resolution, and model representation and parameter calibration. In this study, we introduce a machine learning-based daily global gridded SWE (SWEML) product with 0.25° (~25 km) resolution for 1980–2020 to address these challenges. First, given the significant impact of topography and regional climatic variables on SWE, we used k-means clustering to regionalize the data into fourteen clusters. We, then, the Random Forest model is used to relate 11,687 daily in-situ SWE measurements from nine sources (as dependent data) to six meteorological forcing and three orography variables, time variables (year and month), vegetation type, digital elevation model data, snow observation mask and antecedent precipitation index (as predictor variables). Finally, we constructed SWEML, a global daily SWE product covering the entire global land area excluding Antarctica, using the RF model trained for each cluster, and we compared it with other global gridded datasets. SWEML is evaluated using a total of 10 reference datasets representing remote sensing, model-based, and reanalysis SWE products. SWEML demonstrates the highest accuracy, with a root mean square error (RMSE) of 10.33 mm and a bias of -7.13 mm. Notably, SWEML shows high accuracy in high-elevation regions, including the Rocky Mountains, achieving an RMSE of 7.30 mm and a correlation coefficient of 0.98. SWEML also shows strong agreement with Gamma airborne SWE over North America and the spatial patterns and peak SWE timing of the Andes Snow Reanalysis. Our new dataset provides to serve as the foundational dataset for study on snow-related water cycle and water resource management.

ABSTRACT. Accurate flood simulation in large basins requires parameter sets that are physically plausible and spatially transferable. Conventional single-site calibration often leads to overfitting, which degrades basin-wide consistency. This study develops an automated multi-site calibration workflow by coupling WRF-Hydro with the Parameter ESTimation tool (PEST++) for the Nakdong River Basin, South Korea. We calibrate key hydrologic and routing parameters using a historical extreme event in 2022, enforcing stream-order-dependent constraints to maintain realism from headwaters to the main stem. The framework is validated against an independent flood event to evaluate reliability and event-to-event transferability, focusing on the spatial trade-offs inherent in multi-objective calibration. To address the computational costs and data requirements of PEST++, we also introduce an AI emulation pathway that maps physiographic attributes to optimal parameter spaces identified through large-sample simulations. This emulator acts as a high-speed surrogate to extend calibrated knowledge to ungauged sub-basins and enable rapid ensemble experiments. By positioning AI as an extension of the physics-first framework, the proposed workflow offers a scalable solution for basin-wide flood forecasting and hydroinformatics.

ABSTRACT. Coastal areas in typhoon-prone regions are frequently affected by extreme wave conditions, which pose challenges to port operations, coastal safety, and disaster preparedness. Although short-term wave forecasts are routinely available, the resulting information is often delivered in numerical or technical forms that limit its practical use in day-to-day decision-making contexts. This study develops an integrated framework that combines short-term wave forecasting with digital twin–based visualisation, using Qijin Port in southern Taiwan as a case study. Historical and real-time wave observations are processed to estimate key wave parameters, such as significant wave height and wave period, over short forecasting horizons. The forecasting results are subsequently linked to a three-dimensional virtual representation of the port environment, where temporal changes in wave conditions are translated into scenario-based visual displays. The proposed framework does not aim to replace existing numerical wave models or high-resolution physical simulations. Instead, it focuses on improving the accessibility of wave information by connecting forecast outputs with intuitive spatial representations. Through visual and interactive exploration of short-term wave evolution, the framework supports a clearer understanding of wave-related hazards and their potential implications for port activities and surrounding coastal areas. The anticipated outcomes of this study include: (1) the development of a practical workflow for integrating wave forecasts with digital twin visualisation; (2) enhanced interpretability of short-term wave information for hazard awareness and preparedness; and (3) a modular framework that can be adapted to other coastal regions facing similar wave-related risks. These outcomes are expected to contribute to coastal hazard preparedness by facilitating more effective communication and use of wave forecasting information.

ABSTRACT. Timely and reliable information on where and how much it will rain is critical for managing urban floods and compound water-related hazards, especially under a changing climate. Recent deep learning studies using radar and rain gauges have improved high-resolution quantitative precipitation forecasts, but the outputs are still mainly static maps that are difficult for non-experts to interpret and explore interactively. This study proposes a multimodal AI framework that combines a radar–gauge-based vision backbone with a large language model to enable natural-language interaction with high-resolution rainfall forecasts over the Korean Peninsula. The vision component is built on composite weather radar fields with a 10-minute update cycle, augmented by surface rain-gauge observations. Past radar images and gridded gauge rainfall are fed into a U-Net-type convolutional network with a multi-head encoder and shared decoder, which predicts future rainfall fields at 10-minute intervals up to 12 hours ahead. The model will be trained on a multi-year archive of radar and gauge observations so that it can learn typical patterns of storm development and decay under diverse synoptic conditions. The resulting gridded forecasts are designed to provide spatially consistent rainfall fields that can be interpreted not only by experts but also by downstream AI components.On top of this vision backbone, the language module is intended to summarize and explain the forecast products in natural language, focusing on the timing, location and intensity of potentially hazardous rainfall for user-specified regions. In this way, users can query the system with simple questions about upcoming rainfall and receive answers that combine quantitative information with qualitative descriptions that are easier to understand. The study aims to demonstrate that the proposed multimodal system can (1) retain or improve upon state-of-the-art radar–gauge fusion performance, while (2) providing an intuitive, query-based interface for flood and disaster management. Ultimately, this framework is intended as a building block for interactive decision-support tools that integrate rainfall forecasts, impact indicators and expert knowledge in a single conversational environment.

ABSTRACT. Dam construction has profoundly modified sediment transport processes in river systems, leading to sediment accumulation within reservoirs and a consequent reduction in sediment supply to downstream reaches. These alterations have resulted in geomorphological and environmental issues in Japanese rivers, including riverbed degradation and deterioration of riverine environments. Accordingly, the quantitative assessment of sediment transport characteristics in the vicinity of dams has emerged as a critical research topic.

This study performs a simplified assessment of bedload sediment transport characteristics using a fixed-bed two-dimensional unsteady flow analysis in the upstream reach of the Funagira Dam in the Tenryu River system. Bedload fluxes generated under various discharge conditions are estimated based on a fixed-bed modeling framework, and both the magnitude and grain-size composition of sediment passing through the dam site are evaluated. These estimates are compared with bedload fluxes derived from movable-bed simulations to clarify the applicability and inherent limitations of fixed-bed analyses.

The study area extends approximately 10 km upstream from the Funagira Dam to the confluence with the Keta River, a tributary supplying sediment to the main channel. Two-dimensional unsteady flow simulations were conducted under fixed-bed conditions using the iRIC Nays2DH solver. For each discharge scenario, hydraulic variables were extracted after the flow field reached a quasi-steady state, and bedload fluxes were computed from bed shear stress at each computational grid cell.

Total bed shear stress was decomposed into streamwise and transverse components according to local flow direction to derive directional components of bedload transport. Longitudinal bedload transport rates were multiplied by grid width and integrated across cross sections to estimate sectional bedload fluxes, enabling examination of longitudinal variations and grain-size-specific transport characteristics at the dam site.

Comparison between fixed-bed and movable-bed analyses indicates that movable-bed simulations exhibit smoother longitudinal variations in bedload flux than fixed-bed analyses. While the fixed-bed approach provides reasonable estimates in relatively straight channel reaches with mild flow conditions, it tends to overestimate bedload flux in constricted reaches where morphological effects are pronounced. These findings suggest that fixed-bed analyses are useful for simplified evaluation of bedload transport characteristics; however, their applicability is limited where morphological effects exert dominant control. Consequently, inherent limitations exist in evaluating grain-size-specific sediment transport rates at the dam site based solely on fixed-bed analyses.

ABSTRACT. In recent years, the increasing frequency of extreme rainfall events has led to enhanced sediment production, making long-term assessment of sediment dynamics in river systems an important challenge. In the Takatoki River, which flows through the Kohoku region of Shiga Prefecture, Japan, a localized and short-duration heavy rainfall event occurred from August 4 to 5, 2022. This event caused river flooding, road inundation, and damage to residential areas. Since this flood disaster, prolonged turbidity has been repeatedly observed after subsequent flood events, a phenomenon referred to as the long-term turbidity problem of the Takatoki River. This phenomenon is considered to result from large volumes of sediment supplied from multiple upstream tributaries during the heavy rainfall, which were widely stored within channel reaches in mountainous sections of the river. Following the disaster, these sediments have been gradually transported downstream during subsequent floods, raising concerns regarding their impacts on downstream riverbed conditions and channel stability. However, the temporal and spatial characteristics of sediment migration and dispersion after such extreme events have not yet been sufficiently clarified. The ultimate objective of this study is to reproduce post-flood sediment dynamics in the Takatoki River over long time scales and across the river system using a one-dimensional (1D) bed variation model. As an initial step toward this objective, this study aims to identify which aspects of sediment behavior following large flood-induced sediment inputs can be represented by a 1D morphodynamic model and to clarify its limitations. First, the post-disaster behavior of sediment supplied during extreme rainfall events was reviewed and organized based on previous studies. Subsequently, bed variation simulations were conducted for the Takatoki River using Nays1D+, a one-dimensional morphodynamic solver within the iRIC framework. The simulation results were compared with observed riverbed changes and spatial characteristics of sediment distribution obtained from field observations. The results indicate that, in 1D bed variation modeling, bed aggradation and degradation are evaluated based on differences in sediment transport rates between adjacent cross-sections, resulting in uniform bed elevation changes within each computational reach. In contrast, field observations reveal spatially heterogeneous patterns of sediment deposition and erosion within cross-sections associated with meandering channel geometry, which are difficult to reproduce using a 1D modeling approach.

ABSTRACT. Traditional direct sampling methods for monitoring suspended sediment in rivers are limited by high operational costs, safety risks, and the inability to provide continuous real-time data. As an alternative, indirect measurement techniques utilizing acoustic backscatter intensity have gained attention. This study aims to comprehensively evaluate the performance of suspended sediment detection according to changes in ultrasonic frequencies. Controlled experiments were conducted in a large-scale flume at the Andong River Experiment Center of the Korea Institute of Civil Engineering and Building Technology (KICT). The experimental setup included single-frequency instruments, specifically the ADCP M9 and Riverpro (1200 kHz), alongside the multi-frequency Aquascat 1000R. These instruments were used to analyze acoustic response characteristics across a wide frequency spectrum ranging from 0.5 MHz to 5 MHz. To validate the accuracy of the acoustically estimated concentrations, the results were cross-referenced with particle size and concentration data from the laser-diffraction-based LISST-200x, as well as physical water samples. The results demonstrated that backscatter intensity and detection resolution varied significantly across different frequency bands even under identical sediment conditions. While certain lower frequency bands showed minimal scattering response, hindering accurate concentration estimation, the sensitivity to sediment passage improved markedly at higher frequencies. Based on acoustic scattering theory, this study highlights the necessity of optimizing frequency combinations for different sensor types and provides foundational data for establishing quantitative suspended sediment monitoring systems in river environments.

This research was funded by the Korea Environment Industry & Technology Institute (KEITI) through the Smart Water-supply Service Research Program, funded by the Korea Ministry of Climate, Energy, Environment (MCEE) (RS2024-00397970).

ABSTRACT. Local scour downstream of weirs has been primarily studied through straight channel experiments and field observations. Previous studies interpreted scour characteristics using Froude numbers or energy-based dimensionless relationships. However, curved sections in meandering rivers exhibit complex flow characteristics distinct from those in straight sections. A comprehensive evaluation is required before directly applying straight-section analysis methods to curved sections. This study compares newly acquired curved channel experimental data against existing straight channel data to analyze local scour behavior downstream of a weir. Through this analysis, it examines whether dimensionless parameter combinations originally proposed for straight channels maintain consistency in curved channel environments. For this research, experimental data were acquired from a curved flume with a weir installed immediately upstream of a bend. Dimensionless analysis was performed by normalizing scour depth with downstream and critical water depths, incorporating Froude numbers alongside hydraulic and geometric parameters. Results indicate that curved channel data generally follow the overall flow and scour trends reported in straight channel studies. Nevertheless, systematic deviations from straight channel observations were identified even under identical dimensionless conditions. Such discrepancies are attributed to flow structures unique to curved channels, including secondary flow development, asymmetric jet deflection downstream of the weir, and enhanced lateral redistribution of flow energy. These findings suggest that while straight channel data serve as useful references, additional research considering flow mechanisms driven by curvature is required for accurately understanding and predicting local scour in natural meandering rivers.

ABSTRACT. Aiming at the special operation mode of "high flow rate, high sediment concentration, and short duration" of the 2020 Yellow River water-sediment regulation, this study was conducted to accurately reveal its influence on the laws of sediment and salt diffusion. The study area is the coastal waters adjacent to the Yellow River Estuary. A three-dimensional coupled water-sediment-salt model was established based on the FVCOM model. Measured hydrological data from the Lijin Station were adopted as boundary conditions, and model validation was completed by integrating measured data from 3 offshore observation points in the estuary. This study focused on the spatiotemporal evolution of the flow field, sediment concentration field, and salinity field during the regulation period (July 1 to July 15). It coupled the synergistic mechanism of hyperpycnal flow-plume flow-shear front to quantify the effects of water-sediment regulation on sediment transport, deposition, and salinity diffusion. The results show that: During the 2020 water-sediment regulation, the maximum flood discharge at the Lijin Station reached 3280 m³/s, and the maximum sediment concentration was 215 kg/m³. Driven by strong water flow, sediment mainly diffused northeastward of the estuary. Its diffusion range expanded by approximately 1.8 times compared with the non-regulation period. Sediment deposition in the nearshore area (water depth 0~5 m) accounted for 35.2% of the total sediment transport volume. A large amount of fresh water brought by the regulation significantly altered the structure of the estuarine salinity field. The surface salinity was reduced by 8~12 psu compared with the non-regulation period. The salinity front migrated approximately 3.5 km from the nearshore to the open sea, exhibiting diurnal variations. These variations were characterized by "a larger diffusion range during the day and relative contraction at night". The coupling effect of tidal current and runoff is the core dynamic factor affecting sediment and salt diffusion. The estuarine shear front blocks the offshore diffusion of sediment. Meanwhile, hyperpycnal flow and plume flow form a two-layer sediment transport system. During the regulation period, the dominant role of runoff was enhanced, and the restrictive effect of tidal current was weakened. The research results fill the gap in the study of estuarine water-sediment coupling mechanisms under high-intensity operation modes. They can also provide a scientific basis for the ecological protection of the Yellow River Estuary, coastal zone restoration, and the optimization of water-sediment regulation schemes.

ABSTRACT. Mixing processes at river confluences play a critical role in controlling water-quality distributions and transport dynamics, yet field-based quantitative investigations remain limited due to observational challenges. When tributaries with different temperatures merge, density-driven flows may interact with complex velocity structures, resulting in non-uniform and transient mixing behavior.

This study investigates mixing characteristics at the Nakdong River confluence using boat-mounted field measurements. Three-dimensional velocity fields were obtained using an Acoustic Doppler Current Profiler (ADCP), while vertical profiles of temperature and electrical conductivity were measured across representative cross-sectional locations. To visualize spatial mixing patterns, water-quality data were interpolated using a 3D Kriging approach.

The results indicate that mixing behavior strongly depends on the relative momentum of incoming tributaries. Under high-momentum conditions, velocity structures dominate the flow, allowing warmer water masses to persist with limited vertical mixing. In contrast, under low-momentum conditions, temperature-induced density currents become more pronounced, with reduced influence from advective shear. In addition, a persistent dual helical motion was identified across the confluence, suggesting that secondary-flow structures can serve as an effective indicator of mixing development and decay.

These findings highlight the importance of combined density effects and secondary-flow dynamics in controlling mixing at river confluences and demonstrate the value of high-resolution field observations for improving process-based understanding of confluence hydraulics.

This research was funded by the Korea Environment Industry & Technology Institute (KEITI) through the Smart Water-supply Service Research Program, funded by the Korea Ministry of Climate, Energy, Environment (MCEE) (RS-2024-00397970).

ABSTRACT. In natural river bends, non-uniform bed change patterns are attributed to the complex interaction between centrifugal force and secondary flow. During this process, a sorting phenomenon occurs where particles are separated by size due to the flow. In this study, hydraulic experiments in a 90° curved channel were conducted to understand the spatial characteristics of bed material sorting within the bend section. The results showed a distinct difference in sediment transport patterns and bed topography changes between a single weir installed only downstream and series weirs placed both upstream and downstream. In particular, when a single weir was installed, bed changes were significant in both longitudinal and cross sections. Under maximum flow conditions, The concentrated sediment was deposited on the inner bank of the bend, specifically in the 30°–45° cross section, and a high sorting coefficient was observed. The main cause of this phenomenon is appears to be amplification of the secondary flow generated in the bend. The secondary flow accelerated the diagonal movement of sediment from the outer to the inner side, which ultimately increased local deposition. However, series weirs were applied, sorting index was homogeneous. It appears that the vortex formed immediately downstream due to the drop at the upstream weir effectively reduced the flow velocity. The lowered velocity minimized bed scouring and sediment suspension while blocking the amplification of the secondary flow, which reduced the transverse transport of sediment. Consequently, series weirs incorporating an upstream weir are considered more effective in maintaining bed stability and mitigating the sorting phenomenon than the single weir. Based on these results, the application of series weirs may be considered an effective method to ensure bed stability in sections with severe bed changes or a high risk of local scour.

ABSTRACT. Wonogiri Reservoir is a multifunctional reservoir located on the main stream of the upstream Bengawan Solo River. During the 44 years that the Wonogiri reservoir has been operating, there has been a significant reduction in storage capacity of 177 million m3. Integrated reservoir sedimentation management efforts are needed, starting from controlling erosion in the catchment area to handling sediment in the reservoir. Sedimentation management efforts will need to be reviewed/updated periodically in order to update strategies and integrate technical and ecological approaches in an integrated manner. In updating the sedimentation managemnt in the Wonogiri reservoir catchment area, it is necessary to analysing erosion and sedimentation conditions in the Wonogiri catchment area, evaluate the reduction in storage capacity based on the latest bathymetric data, assess the contribution of each sub-watershed to sedimentation and formulate recommendations for sustainable management. With this updating, a guideline for strategies for handling sedimentation in the Wonogiri reservoir catchment area and alternatives that are right on target (strategic planning) has been obtained as well as the availability of a division of roles for relevant stakeholders and community participation in integrated sedimentation handling and control activities in the Wonogiri. Based on erosion analysis, the Wonogiri catchment area have average erosion rate of 287.73 tons/ha/year. Bathymetric surveys indicate that the effective storage capacity has decreased from 388 million m³ in 1980 to 383 million m³ in 2024. Modeling results show that sedimentation rates reach 1,483,274 m³/year, with the largest contribution originating from the Keduang Sub-watershed (42%), followed by Tirtomoyo (14%), Temon (12%), and several smaller sub-watersheds. Based on the bathymetric data survey in 2024, there are a significant decrease of the total reservoir effective storage from 388 million m³ in 1980 to 383 million m³ in 2024 or around 68% from its initial capacity. This condition highlights the urgency of implementing integrated sediment management measures, both technical and non-technical, including dredging, construction of check dams, green belt revegetation, gully plug maintenance, and land-use regulation. This study aims to provide a comprehensive overview of erosion and sedimentation conditions within the Wonogiri catchment, assess the contribution of each tributary, and formulate sustainable management strategies. In addition, a five-year action plan has been prepared, consisting of three scenarios for sediment-handling capacity: an Optimistic scenario at 3,388,376.05 m³ (cost: IDR 292 billion), a Normal scenario at 3,091,876.05 m³ (cost: IDR 205 billion), and a Realistic scenario at 2,931,226.05 m³ (cost: IDR 182 billion).

ABSTRACT. Small and medium-sized mountain rivers, characterized by their steep gradients and confined valley settings, present unique challenges in flood risk management. During extreme weather events, these rivers not merely transport water, but also they become conduits for massive volumes of coarse-grained sediment, boulders, and large woody debris (LWD). As these materials move downstream, they often become lodged at narrow points or bends, leading to a significant reduction in the hydraulic conveyance. Despite the physical reality of this phenomenon, conventional practices for calculating design flood levels in river basic plans in Korea often overlook the effect of cross-sectional blockage, potentially leading to an underestimation of flood risks. To address this gap, a comprehensive study was conducted focusing on the impact of the 2022 Typhoon Hinnamnor, which brought severe rainfall to Pohang City. The precipitation exceeded a 200-year recurrence probability and providing a critical case study for extreme hydrological behavior. To ensure the accuracy of the analysis, related data was also collected including high-resolution drone aerial photography to capture post-flood geomorphic changes, cross-sectional field surveys, and hydrological data from neighboring stations to reconstruct the event's intensity. The research utilized the HEC-RAS model to simulate several blockage scenarios considering the ratio of cross-sectional area blocked by sediment and debris accumulation. The simulation results yielded a critical threshold: when more than 30% of a river's cross-section is choked by debris and sediment, inundation by flood flow occurs. Furthermore, the rise in flood water levels is proportional to the elevation of the riverbed caused by sediment deposition. The implications of these findings are vital for future disaster mitigation. By quantifying the relationship between physical blockage and water level rise, this study provides a scientific foundation for more resilient river management strategies. Moving forward, it is essential that flood defense measures transition from static hydraulic models to more dynamic approaches that account for the actual transport of material. Incorporating these "real-world" variables into river basic plans will allow for more effective infrastructure design and, ultimately, a significant reduction in flood-related damages in vulnerable mountainous regions.

ABSTRACT. Sediment transport in large regulated rivers is governed by flood-driven erosion, seasonal hydrological regimes, and engineering controls. In the middle reaches of the Yellow River Basin, daily runoff-sediment interactions exhibit strong intermittency, nonlinear threshold behavior, and spatial transitions associated with reservoir operation. However, existing approaches—such as linear regression models or single-copula frameworks—struggle to simultaneously represent the coupled complexity of high-dimensional temporal dependence, nonlinear sediment mobilization thresholds, and regulation-induced structural shifts in daily runoff-sediment systems. This study aims to quantify how sediment transport regimes and runoff-sediment coupling vary across river reaches and seasons. A hybrid Copula-Vector Autoregressive (VAR) stochastic framework is employed to represent temporal persistence and cross-variable dependence of daily runoff and sediment concentration at three representative stations: an upstream erosion-dominated reach, a reservoir-regulated reach, and a downstream mixed transport reach. Log-transformed variables are modeled using a VAR process, capturing short-term memory (optimal lag order p = 10). Residual innovations are fitted with a Student-t Copula, enabling representation of nonlinear and heavy-tailed dependence structures. Results indicate pronounced spatial and seasonal variability in hydro-sediment dependence. Upstream reaches exhibit strong synchronization of extreme runoff-sediment events during flood seasons, reflecting erosion-dominated transport. In contrast, joint extremes are substantially dampened at reservoir-regulated and downstream stations, revealing the influence of engineering controls and channel reworking. The fitted copula degrees of freedom (df ≈ 5) confirm heavy-tailed dependence, indicating stronger extreme synchronization than implied by Gaussian assumptions. The proposed framework provides a basin-scale stochastic perspective on sediment transport regime transitions, demonstrating that engineering regulation reduces sediment magnitude while fundamentally restructuring runoff-sediment dependence and weakening extreme-event synchronization.

ABSTRACT. In channelized rivers for flood control purposes, it aims to clarify how flow patterns within the pools change in response to varying discharges, by altering the longitudinal gradient between mounts, the interval between mounts, the pool depth, and the shape between mounts. The final goal is to clarify the optimal conditions under which unsteady gravels, which repeatedly undergo deposition and scouring within the pool, are most likely to form. In the experiment, four mounts with porosity formed by assembled boulders were installed. Parameters such as the vertical gradient connecting the mount tops, the spacing between mounts, and the pool depth were set. The bottom surface between mounts was made concave using crushed stones. A hydrograph was configured to verify the time series of flow velocity between mounts with porosity as discharge varied. Experiments were conducted under the scaled physical model to investigate the behavior of flow velocity between mounts within the configured hydrograph. The physical model was installed by keeping the longitudinal gradient of the mount installation section constant, constructing the mounts with assembled boulders were installed, filling the mount gaps with crushed stones, and varying the mount spacing and depth between mounts. In the experiments, under 1/100 gradient and two types of pool depth, the mount spacing to 1.5, 3.0, and 5.0 times the channel width was set. Analysis of the time-series changes in flow velocity near the pool bottom revealed that appropriately setting pool depth and mount spacing is crucial. Focusing on the time-series changes in flow velocity near the bottom at the deepest point, the change in ensemble-averaged flow velocity with respect to discharge variation was small, but the change in fluctuating flow velocity was large. The magnitude of the ensemble-averaged flow velocity varied with pool depth, and changing mount spacing caused the time-series changes recorded at the deepest point to differ along the longitudinal profile. When the mount spacing is 1.5 times the channel width, the flow velocity behavior differs even within the pool. Conversely, when the mount spacing is 5 times the channel width, the increased drop between mounts causes the flow velocity response to flow rate changes to differ from other cases. Further, variations in mount spacing and pool depth produce distinct differences in the time-series flow velocity changes. Consequently, in the installation of mounts with porosity in linearly channelized rivers, the optimize selection of mount spacing and pool depth is critically important.

ABSTRACT. At the confluence of the Tone River and the Karasu River in Yattajima, Isesaki City, Gunma Prefecture, a water level gauge installed closely downstream of the confluence was buried due to flood events caused by Typhoons No. 12 and No. 15 in 2011. Since this water level gauge plays an important role in the flood monitoring and disaster prevention, the burial of gauge posed a serious problem for the continuous and reliable observation of water levels. Aerial photo analysis suggested that the burial was mainly caused by the elongation of the leftbank bar formed downstream of the confluence. The objective of this study is to clarify the mechanisms of sandbar behavior at the river confluence by using numerical simulations, with a particular focus on the influence of the flow rate ratio between the two rivers. Numerical analyses were conducted by using the two-dimensional flow and the movable bed model implemented in the iRIC software, named as “Nays2DH”. In this study, the flow rate ratio was defined as the flow rate of the Tone River divided by that of the Karasu River. A simple Y-shaped channel approximating the scale of the filed site was constructed, and a series of simulations with different flow rate ratios were conducted under steady-flow conditions to investigate their effects on sandbar morphology. Numerical results showed that the flow rate ratio had a significant influence on the shape and elongation of the sandbar. As the flow rate ratio increases, the sandbar tends to elongate further in the downstream direction. This behavior indicates that the inflow from the Tone River, which joined the Karasu River from the left bank side, enhances sediment transports and deposition patterns that promoted the downstream sandbar development. These findings suggest that changes in the discharge balance between the Tone River and the Karasu River played a crucial role in sandbar formation at the confluence and can directly affected the management of river monitoring facilities. The results of this study provide useful insights for the planning of water level gauge placement and river management at river confluences where sediment dynamics are strongly influenced by flood discharge conditions.

ABSTRACT. In fluvial and coastal environments, the resuspension of heterogeneous sediment leads to complex vertical variations in suspended sediment size distributions. While the upward diffusion by turbulence and downward settling by gravity are recognized as the primary physical mechanisms, a detailed understanding of how these factors influence the resulting grain size characteristics is essential for predicting sediment transport dynamics. This study presents a numerical investigation into the effects of turbulence intensity and particle settling velocity on the size distribution of fine sands through a novel modeling framework. The methodology involves the adaptation of a size distribution model originally developed for cohesive sediments. To account for the properties of noncohesive fine sand, the model was revised by setting the fractal dimension to a constant value of 3.0, thereby ensuring a fixed particle density equivalent to the primary sediment density. A stochastic approach was implemented using Monte Carlo simulations, where the Box-Muller method was utilized to generate random variables representing the inherent fluctuations in resuspension and deposition processes. This framework assumes equal mobility at the bed boundary, allowing for the simulation of heterogeneous bed material behavior. The size distribution model was coupled with a one-dimensional vertical transport model that solves momentum equations and utilizes a k-ε turbulence closure scheme to determine the eddy viscosity and flow velocity profiles. The integrated model was validated by replicating laboratory flume experiments. The numerical results showed high fidelity in reproducing vertical profiles of horizontal velocity and mass concentration, with correlation coefficients for the predicted size distributions ranging from 0.928 to 0.977. A critical finding is that both the mean grain size and the standard deviation of the suspended sediment distribution decrease as the elevation from the bed increases. Furthermore, we defined a dimensionless elevation, which identifies the location of the maximum volume fraction for each size class. This parameter was found to be directly proportional to the ratio between turbulence intensity and the settling velocity, highlighting the dominant role of this ratio in determining the vertical stratification and lognormal characteristics of suspended sands.

ABSTRACT. Vegetation has distinct effects on soil erosion control. Ground covers such as herbs and grasses are more effective than canopy covers such as shrubs and subtrees in reducing rainfall kinetic energy, minimizing raindrop impact, and decreasing splash erosion. In this study, surface runoff and sediment yield were measured in steep hillslops with different vegetation cover conditions, including herbs and grasses, as well as pine and maple subtrees. While surface runoff was not highly sensitive to vegetation cover, sediment yield showed significant differences depending on the presence or absence of rill development, which varied with ground and canopy cover conditions. In bare plots without vegetation, rills exhibited the greatest width and depth, and the highest sediment yields were recorded. Rills were rarely observed in herb–grass plots but frequently developed in subtree plots; however, their size and number were smaller than those observed in bare plots. The herbaceous layer not only reduced the kinetic energy of rainfall striking the soil surface but also effectively controlled the incision and expansion of rills caused by concentrated surface runoff. In contrast, raindrops falling through the leaves of subtrees still retained substantial kinetic energy, and tree trunks were not effective in controlling high-velocity concentrated flow. These results demonstrate that herbs and grasses play more significant hydrological roles than subtrees in controlling soil erosion and sediment yield.

ABSTRACT. River management in Japan has traditionally focused on improving flood conveyance capacity through channel excavation and vegetation removal, while also emphasizing environmental conservation. In recent years, increasing attention has been paid to the restoration of sediment dynamics and the maintenance and creation of diverse habitats associated with riverbed processes. However, riverbed variation is governed by complex interactions among channel geometry, sediment characteristics, and discharge conditions. Conventional two-dimensional morphodynamic simulations often require high computational costs and face difficulties in systematically evaluating the effects of grain size and a wide range of flow conditions.

This study aims to evaluate tendencies of riverbed erosion and deposition, as well as disturbance characteristics, by focusing on the spatial distribution of bed shear stress under multi-stage discharge conditions. A new indicator, termed bed variation potential, is proposed and constructed using hydraulic variables obtained from two-dimensional unsteady flow simulations with a fixed bed. The applicability of the proposed approach is examined through case studies of multiple rivers, including the Kizu River and the Nagara River, with particular emphasis on sediment transport processes and their contribution to habitat formation. In addition, relationships between in-channel landscapes and disturbance regimes are investigated through classification based on similarities in shear stress response patterns.

Two-dimensional unsteady flow simulations under multiple discharge stages were conducted using iRIC Nays2DH. Bedload transport rates for each grain size class were estimated from bed shear stress, and grid-based sediment flux balances were used to derive the bed variation rate, V_pa(m/h). The bed variation potential, H_pa(m), was then defined by integrating the bed variation rate weighted by discharge frequency derived from observed hourly discharge records over the analysis period. Furthermore, response patterns of hydraulic variables, including bed shear stress across multiple discharge stages, were organized as feature vectors, and k-means clustering was applied to classify spatial disturbance regimes within the channel.

The results indicate that no clear correlation was observed between actual bed elevation changes and the bed variation potential at individual grid cells. In contrast, when values were aggregated at the reach scale, spatial patterns and tendencies of erosion and deposition were reasonably reproduced. Although the absolute magnitudes of erosion and deposition tended to be overestimated, the proposed method effectively captured the relative spatial distribution of riverbed disturbance. Clustering based on shear stress responses further revealed spatially coherent regions exhibiting similar disturbance behaviors with increasing discharge, highlighting systematic relationships between disturbance regimes and in-channel geomorphic units.

ABSTRACT. This study investigates the effects of in-channel vegetation on flow dynamics and morphological responses under different channel planforms using two representative reaches located in the middle to lower Seomjin River. The reaches consist of a straight channel (Chimsil Wetland, Reach A) and a naturally meandering channel (Otter Habitat Ecological Conservation Area, Reach B), providing a framework for comparative analysis of vegetation–flow–morphology interactions. A two-dimensional numerical model, Nays2DH, was applied to simulate flow characteristics and bed evolution under steady and unsteady flow conditions, with emphasis on vegetation density and flood magnitude. Model sensitivity analyses showed that excessively coarse grid resolutions degraded predictive performance, whereas an optimal grid resolution achieved high accuracy (NSE = 0.911, RMSE = 0.074 m), with further refinement yielding limited gains at the expense of computational efficiency. Simulated water levels agreed well with a one-dimensional hydraulic model (HEC-RAS) for both 20- and 100-year flood scenarios, and validation against an observed extreme flood event (peak discharge of 8,756.31 m³/s) confirmed the robustness of the numerical framework. Increasing vegetation density significantly reduced channel conveyance capacity and induced flood-stage rise in both reaches, with maximum increases of up to 0.51 m and 0.57 m under the 2-year flood, and 1.03 m and 1.28 m under the 100-year flood in Reaches A and B, respectively. Morphodynamic simulations showed widespread erosion and deposition without vegetation, whereas higher vegetation density constrained bed changes and enhanced overall bed stability; however, persistent local scour was identified in the curved reach (Reach B). Bed Relief Index (BRI) analysis indicated a nonlinear response of bed morphology to discharge magnitude: vegetated conditions promoted bed flattening under low to moderate flows, whereas high-magnitude floods intensified bed undulation despite the presence of vegetation. Overall, this study provides quantitative insights into vegetation–flow–morphology interactions and supports vegetation-based river management and the application of nature-based solutions (NbS) for flood mitigation and river restoration.

This work is financially supported by Korea Ministry of Climate, Energy, Environment(MCEE) as 「Research and Development on the Technology for Securing the Water Resources Stability in Response to Future Change(RS-2024-00332494)」.

ABSTRACT. Japanese rivers have experienced substantial reductions in sediment dynamics as a result of extensive dam construction and intensive sand mining during the twentieth century. These interventions have led to widespread riverbed degradation and the simplification of riparian and in-channel habitats. In response, nature-based solutions (NbS), including traditional hydraulic structures such as Seigyu (crib spurs), have been reintroduced into contemporary river management strategies. Although Seigyu were widely employed until the 1950s, their long-term geomorphological effects under modern sediment-deficient conditions remain insufficiently quantified, particularly due to the lack of high-resolution, multi-temporal topographic analyses. This knowledge gap limits the generalization of design principles and the prediction of long-term morphological responses, thereby constraining the reliable transfer of Seigyu-based approaches to other river systems. Addressing this gap is essential to establish robust scientific evidence for the broader application of nature-based solutions in river restoration and management. This study aims to evaluate the geomorphological responses, sediment dynamics, and overall performance of Seigyu structures in the Kizu River over the period from the initial pilot implementation in 2017 to 2025. The methodology is based on multi-year, high-resolution datasets derived from UAV-based topographic surveys and aerial imagery. Digital Elevation Models (DEMs) and Digital Surface Models (DSMs) were generated and analyzed using a DEM of Difference (DoD) approach to quantify spatio-temporal patterns of erosion and deposition. Additional analyses of channel cross-sections, longitudinal profiles, and planform geometry were conducted to estimate volumetric sediment changes and sediment budgets across the study reach. The results demonstrate that Seigyu structures effectively promote sediment deposition at both upstream and downstream ends, while inducing localized bed erosion and scour along their lateral margins. Distinct spatial patterns were observed: bar heads are characterized by deeper scour and the formation of deep pools (wands), whereas bar tails exhibit enhanced sediment accumulation. Long-term monitoring indicates that the structures remain morphologically stable and hydraulically functional even following extreme events, such as the major flood in 2019, despite the persistence of an overall reach-scale degradation trend. These results demonstrate that Seigyu enhance ecohydraulic diversity by creating dynamic habitat features and exhibiting adaptive geomorphological behavior through natural flow–sediment interactions. As a nature-based solution, Seigyu strengthen river resilience and highlights the continued relevance of traditional river engineering for sustainable management under climate change, supporting a shift from fixed-bed control toward dynamic, moving-bed river management.

ABSTRACT. Effective sediment management and channel maintenance require clear identification of the discharge levels that activate sediment transport and induce meaningful bed disturbance. This study presents a practical diagnostic framework that integrates long-term flow variability with fixed-bed two-dimensional (2D) hydrodynamic simulations, enabling rapid assessment of sediment mobility without the computational burden of full morphodynamic modelling. The framework was applied to the Kizu River in the Yodo River system, Japan, using 15 years of hourly discharge data and high-resolution ALB/LP topography. A 28-stage discharge system derived from the cumulative hydrograph allowed representative simulation of normal, moderate, and peak flows. For each discharge stage, sediment-active areas, sediment fluxes and spatial patterns of erosion and deposition were quantified. The results highlight clear differences in geomorphic sensitivity along the river corridor. High flows transport large amounts of sediment yet often produce limited morphological adjustment, revealing low bed-change efficiency. In contrast, intermediate flows generate substantial bed disturbance in geomorphically sensitive reaches, particularly where shear-stress gradients are amplified by compound channels or planform confinement. These flow-dependent patterns provide actionable insights for river management, including the identification of discharge ranges suitable for environmental flow releases, sediment-augmentation operations and reach-scale rehabilitation planning. Because the method relies solely on fixed-bed hydrodynamics, it offers a cost-efficient and operationally feasible tool for evaluating alternative dam-release strategies, anticipating geomorphic consequences of management actions and supporting evidence-based sediment management in regulated rivers such as the Kizu River.

ABSTRACT. Local scour at the downstream end of fixed‑bed protection works can compromise the transition to a movable bed and threaten structural safety. However, many existing approaches emphasize equilibrium scour depth and rely on limited turbulence descriptors at the fixed–movable interface, leaving the transient scour‑hole geometry and sediment transport insufficiently quantified. This study combines laboratory measurements and morphodynamic modeling to characterize the spatio‑temporal evolution of local scour downstream of a fixed‑bed structure and to infer sediment flux using an Exner‑based procedure. Flume experiments were conducted with a fixed structure model connected to a uniform sand bed. For a representative run, centerline bed elevations were surveyed every 30 minutes for 19 h with 2–10 cm longitudinal spacing, while the maximum scour depth and scour length were monitored up to 24 h. The bed evolved through rapid incision at the interface, producing a steep upstream slope, followed by downstream extension with a milder gradient and the formation of a step‑like depositional front. The maximum scour depth increased rapidly during the early stage and approached an asymptote, reaching approximately 7.0 cm at 24 h, whereas the scour length continued to expand to about 220 cm. Exponential‑type regressions were derived to describe the bed changes. After the depositional front migrated beyond the fixed‑bed reach, the bed‑volume change between 7 and 19 h was used in the Exner framework to estimate a unit‑width sediment transport rate. This corresponds to sediment concentrations of approximately 11.4 ppm (volumetric) and 30.3 ppm (mass‑based). For morphodynamic prediction, the Exner equation was coupled with the Engelund–Hansen transport capacity; the computed bed profiles reproduced the measured profiles downstream of the maximum‑scour location with RMSE of 0.14–0.43 cm and mean relative errors of 0.38–1.81% over 9.5–19 h. Finally, a Buckingham‑π analysis is applied to express scour growth with dimensionless parameters, including a turbulent‑sediment number based on turbulent energy and sediment fall velocity, enabling consistent comparison across conditions. Acknowledgement This work was supported by Korea Environment Industry & Technology Institute(KEITI) through 「Research and Development on the Technology for Securing the Water Resouces Stability in Response to Future Change(RS-2024-00335281)」 Program, funded by Korea Ministry of Environment(MOE).

ABSTRACT. Floating debris in rivers is a major pathway through which land-based waste is transported to downstream and coastal environments. Under variable hydraulic conditions, conventional floating debris interception systems installed in rivers often face limitations in structural stability, interception efficiency, and applicability under real field conditions. These challenges highlight the need for hydraulically and structurally reliable debris interception facilities supported by comprehensive evaluation methods suitable for river environments. This study presents the hydraulic and structural design of a water-level-adaptive floating debris barrier system and examines its applicability to river environments through an integrated set of analytical, experimental, and field-based methods. To support site-specific design and layout, river characteristics were first examined using aerial imagery analysis, detailed land-cover classification, and two-dimensional hydraulic modeling. These analyses provided hydraulic and spatial information relevant to identifying potential debris source areas and establishing representative flow conditions for system design. Hydraulic performance of the proposed system was evaluated through physical model experiments conducted in a meandering open-channel flume. The experiments reproduced representative normal-flow and higher-flow conditions using Froude similarity. Interception efficiency was quantified using floating debris surrogates, while changes in upstream water level and velocity were measured to assess hydraulic impacts. The results indicated interception efficiencies exceeding 83% under normal flow and approximately 61% under higher flow conditions, with induced water-level variations remaining within 1%. To assess structural safety, hydrodynamic loads acting on the floating barrier were experimentally measured in a controlled test facility using acoustic Doppler current profilers and underwater tension meters. Load measurements were performed for multiple flow velocities and barrier configurations. The experimental results were used to calibrate numerical simulations, revealing that the effective drag coefficient of the barrier system exceeded theoretical values due to the presence of buoyant elements and flow disturbance effects. A drag coefficient of 4.0 provided the best agreement between measured and simulated loads and was adopted for design load estimation. Subsequent numerical analyses indicated design loads on the order of 5–8 tons for the proposed configuration. Field application in the Yugu River, Republic of Korea, confirmed stable operation of the system. The structure remained intact during a prolonged rainfall event with cumulative precipitation exceeding 500 mm, demonstrating its robustness under real river conditions. The results demonstrate that the proposed system and integrated evaluation framework can support the hydraulic and structural design of floating debris interception facilities intended for practical application in river environments.

ABSTRACT. In recent years, the risk of levee overtopping has increased due to extreme rainfall events and rising water levels caused by climate change, highlighting the need for high-performance and environmentally friendly reinforcement technologies that can effectively mitigate slope damage induced by overtopping. Conventional reinforcement methods exhibit limitations in terms of durability, environmental sustainability, and constructability, necessitating a systematic evaluation of the field applicability of bio-polymer-based high-strength products as alternative solutions. In this study, the structural performance and constructability of a bio-polymer-based high-strength reinforcement product were verified through full-scale levee overtopping experiments, while overtopping flow characteristics were precisely analyzed using slope velocity visualization based on image-based velocity measurement techniques. The experiments were conducted using a full-scale test section that replicated an actual levee cross-section, in which the bio-polymer-based reinforcement product was applied. Overtopping conditions were reproduced by gradually increasing the inflow discharge. During the pre-, during-, and post-overtopping stages, water level, discharge, slope displacement, and scour depth were measured. Simultaneously, an image-based velocity measurement system(STIV) was employed to non-intrusively capture the surface velocity distribution along the levee slope under overtopping conditions. Temporal and spatial variations in slope velocity were visualized, and the correlation between velocity concentration zones and scour development was analyzed. In addition, installation tests reflecting field construction conditions were conducted to comprehensively evaluate construction time, workability, installation quality, and field applicability. The experimental results demonstrated that the bio-polymer-based high-strength product maintained excellent scour resistance and structural stability under repeated overtopping conditions, while effectively reducing velocity concentration along the slope and suppressing scour progression. Image-based velocity measurements provided quantitative and visual insights into slope velocity distributions at different overtopping stages, offering valuable information for interpreting levee damage mechanisms and verifying reinforcement effectiveness. Furthermore, the proposed construction method exhibited superior constructability compared to conventional reinforcement techniques. This study experimentally validates the performance and constructability of a bio-polymer-based high-strength reinforcement product by integrating full-scale levee overtopping experiments with image-based velocity measurement techniques. The findings demonstrate the field applicability of environmentally friendly levee reinforcement technologies and provide fundamental data for advancing design criteria, establishing performance certification systems, and developing image-based levee safety assessment technologies for overtopping mitigation.

ABSTRACT. A tunnel spillway is a type of dam spillway capable of releasing water even at low reservoir levels, serving as a supplementary discharge structure that complements the release capacity of an ogee spillway. The tunnel spillway investigated in this study is structurally characterized by its position at mid-height of the dam body, where it penetrates through the ogee spillway; consequently, its discharge capacity may be influenced by the outflow from the ogee spillway. Because the tunnel spillway penetrates directly through the dam body, it is considerably shorter than conventional tunnel spillways, resulting in unique internal flow characteristics that have rarely been studied. This study therefore experimentally evaluated flow rates and internal pressures under varying conditions of air vent status, gate opening ratio, and reservoir water level to analyze the hydraulic behavior of the tunnel spillway. A 1:20 scale hydraulic model was constructed for a prototype tunnel spillway of 30 m in length, and Froude similarity was applied to analyze the flow characteristics. In the experiments, two gates were installed on the ogee spillway and one gate on the tunnel spillway to regulate discharge. Through model testing, flow rates were measured under various gate opening and reservoir water level conditions, flow characteristics were observed at three different gate openings of the air vent, and pressures were recorded at five points along the invert of the tunnel spillway. The internal flow was classified into two regimes depending on gate opening and water level: open channel flow with a free surface, and full pipe flow in which the entire cross section is occupied by water. Full pipe flow was further subdivided into partial full pipe flow, in which air pockets form due to air admitted through the air vent, and complete full pipe flow, in which the cross section is entirely filled with water. Under open channel flow conditions, positive pressure was observed regardless of water level, discharge rate, or gate opening. In contrast, under high water level full pipe flow with air pocket formation, strong negative pressure developed along the invert. Closing the air vent increased the discharge coefficient by an average of approximately 32% relative to the open condition, and the discharge coefficient under full pipe flow was found to be approximately 15% higher than that under open channel flow.

ABSTRACT. The presence of bridge piers within a river channel can generate localized flows, potentially leading to reduced pier stability due to localized scouring and factors contributing to disasters such as driftwood accumulation. Furthermore, considering high-sediment rivers like the Yellow River, mudflows, and the occurrence of flows containing large amounts of fine sediment during heavy rains, it is necessary to clarify the flow mechanisms of high-concentration flows involving bridge piers. Therefore, this study investigated the average flow characteristics and momentum transport characteristics in open-channel flows with oval-shaped bridge piers when viscosity was altered using a sodium polyacrylate (PSA) aqueous solution. Fluid viscosity was measured using a Fungilab Viscolead Advance L rotary viscometer. Eleven PSA solution concentrations ranging from 100 mg/l to 700 mg/l were tested. This revealed that high-concentration sediment flows induced by PSA solutions must be treated as pseudoplastic fluids exhibiting non-Newtonian behavior. Open-channel experiments were conducted using a circulating variable-gradient flume with a total length of 10 m, width B = 40 cm, and height 20 cm. To obtain fundamental insights, experiments were performed under smooth-slope conditions and also examined flow mechanisms with bridge piers installed. Flow velocity was measured using the PIV method with a 2W air-cooled visualization light source. In the smooth-slope condition, unlike clear water flow, the mainstream velocity was suppressed near the bottom and showed a gentle curve distribution at concentrations ranging from 250 mg/l to 700 mg/l. At concentrations between 125 mg/l and 250 mg/l, the mainstream velocity near the water surface became larger than in clean water. It was clarified that at 125 mg/l, 150 mg/l, and 200 mg/l, the velocity distribution approached the limit resistance reduction distribution. Thus, fundamental insights were gained regarding the vertical velocity distribution at various viscosities. For open channel flow with a single oval-shaped pier, it was clarified that as viscosity increases, the bypass flow near the pier tip and the flow toward the channel center near the downstream side of the pier are suppressed. Furthermore, regarding momentum transport characteristics, it was clarified that as viscosity increases, momentum transport due to turbulence is extremely suppressed compared to advection.

ABSTRACT. In this study, we evaluated the detention capacity of existing facilities using a database (DB) of constructed detention systems, with the aim of supporting future site selection and planning of detention facilities.

The facilities were classified into two groups: detention facilities in previously developed areas (Ministry of the Interior and Safety; 108 sites) and those in newly developed areas (Korea Land & Housing Corporation; 133 sites). We assessed detention performance relative to facility specifications by examining correlations between facility attributes (e.g., effective surface area and effective depth) and performance metrics (storage volume, reduction in direct runoff, and peak discharge reduction). The results indicated contrasting design tendencies: facilities in existing developed areas were generally constructed as “narrow and deep,” whereas those in newly developed areas were typically “wide and shallow.” Notably, a strong tendency was observed for facility areas to be limited to ≤10,000 m² in existing developed areas, and for effective depths to be constrained to ≤4 m in newly developed areas. These patterns are attributed to space limitations and differences in the primary purpose of detention facilities: in existing developed areas, they are installed to reduce flood risk in chronically inundated zones, while in newly developed areas, they are designed to minimize the increase in peak discharge between pre- and post-development conditions.

Among the facility variables, effective surface area exhibited the strongest correlation with detention performance, and a regression relationship between effective area and storage volume was derived. In terms of detention capacity expressed as equivalent direct-runoff reduction depth, the Ministry of the Interior and Safety facilities were evaluated at 5–25 mm (mean: 12 mm), while the Korea Land & Housing Corporation facilities showed 8–40 mm (mean: 22 mm). Overall, the findings provide practical criteria for pre-estimating site-specific feasible storage capacities when planning detention facilities in both existing and newly developed urban areas, and are expected to support the screening and selection of candidate sites for detention facility implementation.

This work was supported by Korea Planning & Evaluation Institute of Industrial Technology funded by the Ministry of the Interior and Safety (MOIS, Korea). [Development and Application of Advanced Technologies for Urban Runoff Storage Capability to Reduce the Urban Flood Damage / RS-2024-00415937]

ABSTRACT. At one of the hydroelectric power plants that we monitor, pitting corrosion has been observed in the steel lining of the headrace tunnel. Pitting corrosion causes localized wall thinning of steel components, resulting in potentially being one of factors affecting structural integrity of facility. Therefore, identification of factors contributing to pitting corrosion and consideration of effective countermeasures are important for long-term asset maintenance. In this study, water quality analysis and corrosion product analysis were conducted to estimate the mechanism of pitting corrosion. The performance, procurement, material cost, and workability of several candidate coatings with corrosion resistance were also compared to consider the countermeasure against pitting corrosion. In this performance comparison, onsite exposure test of the coated specimens was carried out to evaluate the corrosion resistance of the coatings. In addition, Taber abrasion test and pencil hardness test were conducted to evaluate the wear resistance and the scratch resistance of the coatings, as other indicators that may influence occurrence of pitting corrosion. The water quality analysis revealed that the reservoir water exhibited relatively high sulfate ion concentration compared with that of other typical rivers. The corrosion product analysis indicated the presence of iron-oxidizing bacteria and sulfate-reducing bacteria. Based on these results, the mechanism of pitting corrosion was estimated to be microbiologically influenced corrosion caused by iron-oxidizing bacteria and sulfate-reducing bacteria, initiated by the water inflow with high sulfate ion concentration. The onsite exposure test showed that the wear-resistant coatings with flake fillers exhibited less coating degradation after exposure than the coatings with corrosion resistance alone. The Taber abrasion test indicated that the wear-resistant coatings with particularly glass flake fillers showed smaller abrasion mass loss compared with the other coatings. The pencil hardness test demonstrated that the wear-resistant coatings with flake fillers had higher hardness than the coatings with only corrosion resistance. Based on these results, the application of one of the epoxy-based coatings with glass flake fillers was proposed as a potential countermeasure to suppress occurrence of pitting corrosion, as the coating was relatively superior to the other coatings in terms of resistance, procurement, material cost, and workability.

ABSTRACT. A movable weir is one of hydraulic structures for water supply, and support pillars are installed to open gates for flood control. Based on hydraulic design manual, concrete blocks are installed downstream of the weir as a measure against riverbed scouring. The commonly used flat-type riprap blocks with seepage block sheets, and there is no function to reduce flow velocity near the bottom. During flood stages, a deflecting flow is formed by the impingement of support pillars for the gates, and a localized scouring occurs downstream of the bed protection blocks. The local scouring might lead to block washouts due to suction effects at the block bases. The formation of the deflected flow causes the main flow to move downward near the riverbed. To counteract scouring downstream of movable weir, it is significant to mitigate the flow deflected by the pillars and raise the main flow toward the water surface. Yasuda and Suzuki proposed installing the consecutively assembled boulders downstream of concrete apron of the movable weir as a protection area. The velocity near the bottom can be reduced in the installation region, but the influence of the deflected flow formed by impact on the pillars persisted downstream of the protection area. The steady undulation continues far downstream by the formation of deflected flow, localized scouring near the side walls is noted. In this paper, the authors proposed the installation of a transition zone using crushed stones (in prototype, rocks with 0.2 to 0.3 m sizes) downstream of the assembled boulders to reduce flow velocity near the bed. An optimize length for the transition zone due to crushed stones was investigated experimentally. Experiments reveal that seepage flow inside crushed stones support the velocity near the bottom in the transition zone and may reduce the velocity of the deflected flow produced by the impingement to the support pillars. If the transition zone is not covered in the area where steady-state waves form, based on bottom profiles and flow velocity profiles, the velocity including turbulent intensity is not reduced to prevent local scouring. When the transition zone included the standing wave formation zone, the velocity suppression by seepage flow took effect, preventing scouring for extended periods (over 30 hours). This demonstrated that placing crushed stone in the optimal zone effectively reduces near-bottom velocities and prevents localized scouring downstream of the movable weir.

ABSTRACT. This study evaluates various scenarios to mitigate saltwater intrusion in the Seomjin River Estuary, which has been exacerbated by indiscriminate sand mining and riverbed lowering. Using the EFDC numerical model, we analyzed saltwater intrusion reduction effects under a fixed condition of Songjeong flow (10 cms) and Daap intake (3.912 cms). The scenarios consisted of 12 detailed cases across four categories: physical alternatives including Riverbed Flattening (Case E), Underwater Barrier (Case F), and Groyne installation (Case G), as well as an operational alternative, Flushing Discharge (Case H).

The simulation results indicated that the installation of artificial structures such as underwater barriers (Case F) and groynes (Case G), as well as temporary flushing discharge (Case H), had no effect on reducing the length of saltwater intrusion (based on 1.0 psu). In contrast, the Riverbed Flattening scenario (Case E), which restores the riverbed by 1 to 3 meters in the 1–17 km downstream section, shortened the saltwater intrusion length by 0.7 to 2.6 km, corresponding to a reduction rate of 3.1% to 11.5%. Furthermore, the salinity reduction effect was most pronounced in the riverbed restoration scenario (Case E), particularly during the neap tide period. Consequently, this study confirms that restoring the lowered riverbed is the most effective measure for reducing saltwater intrusion in the Seomjin River Estuary compared to the installation of artificial hydraulic structures.

ABSTRACT. In lakes with large water surface areas, wind waves cause shoreline erosion, leading to concerns about the disappearance of native aquatic plants along the water side and the decline or loss of habitat for aquatic animals such as small fishes, crustaceans, and benthic organisms. This is related to the fact that, following years of shoreline embankment improvements, nearly all lake shorelines now feature vertically concrete revetments. The installation of these vertical concrete revetments likely generates strong reflected waves, accelerating scouring near the shoreline, and local scouring along water side has been proceeded.　 To counteract shoreline erosion caused by wind waves, the authors proposed the installation of mount-shaped riprap due to large boulders, aiming to preserve habitats for aquatic animals and plants by creating a calm zone. A scaled model using boulders was set up targeting this mount-shaped riprap to experimentally investigate wave attenuation measures. Results showed that the temporal variation in flow velocity on the calm zone side was smaller than that on the wave-exposed side, with the standard deviation value indicating turbulence intensity being close to zero. This suggests that the generated breaking waves are dissipated by the formation of seepage flow inside boulders of the mount, demonstrating its wave-attenuation function. Furthermore, the formation of a calm zone by installing the mount-shaped riprap near the water side, creating a habitat for aquatic animals and plants, was confirmed at the field site where it was preliminary installed.　 Concrete breakwater blocks are commonly used for wave attenuation. To compare them with the mount-shaped riprap, concrete blocks were arranged in a mount-like configuration. Scale models were used to examine differences in flow conditions and time-series changes in flow velocity. Results showed that breakwater blocks exhibited greater fluctuations in time-series flow velocity changes on the calm zone side and higher standard deviation values indicating turbulence intensity. This indicates that wave-dissipating blocks have lower wave-dissipating capabilities compared to the mount structure due to boulders. Furthermore, installing wave-dissipating blocks at the water side may disturb the formation of calm zones, suggesting it would be difficult to preserve habitats for aquatic animals and aquatic plants.　 In other words, the installation of the mount-shaped riprap due to boulders demonstrated the potential for wave-dissipating functionality while also contributing to the environmental preservation of aquatic animals and aquatic plants.

ABSTRACT. With increasing interest in river environments and ecosystems, fishways are essential hydraulic structures for ensuring longitudinal ecological connectivity in rivers. However, a large proportion of fishways installed in Korea require rehabilitation, and many have been reported to show limited effectiveness in facilitating fish upstream migration. One of the most critical factors governing successful fish passage is the internal flow structure generated within the fishway. In this study, turbulent flow characteristics in a stepped fishway were investigated using a three-dimensional computational fluid dynamics (CFD) model. The effects of pool spacing ratio and overflow depth on the flow characteristics were systematically analyzed. The unsteady two-phase flow solver interFoam, provided by OpenFOAM, was employed, and the free surface was modeled using the Volume of Fluid (VOF) method. The results indicate that the mean velocity within the pool increased with increasing spacing ratio, whereas the mean turbulent kinetic energy exhibited no significant variation. The maximum velocity within the pool decreased with increasing spacing ratio at an overflow depth of 0.05 m. However, an opposite trend was observed as the overflow depth increased. Similarly, the maximum turbulent kinetic energy decreased with increasing spacing ratio at an overflow depth of 0.05 m. In contrast, at greater overflow depths, it decreased up to a spacing ratio of 0.35 and then remained nearly constant. This behavior suggests that high turbulent kinetic energy is generated under plunging flow, after which the maximum turbulent kinetic energy stabilizes as the flow regime transitions to streaming flow.

Acknowledgements This work was supported by Korea Environment Industry & Technology Institute (KEITI) through Research and Development on the Technology for Securing the Water Resources Stability in Response to Future Change, funded by Korea Ministry of Climate, Energy, Environment(MCEE)(RS-2024-00397820).

ABSTRACT. Urban roadway flooding has become more frequent and severe under intensified short-duration rainfall, particularly in densely developed areas where runoff is rapidly generated and conveyed along curb-and-gutter systems. In such settings, stormwater grate inlets serve as the primary interface that intercepts surface runoff and transfers it into subsurface drainage networks; therefore, inlet geometry, capture capacity, and spacing strongly govern roadway spread, traffic safety, and localized inundation. Although practical guidelines often rely on rule-of-thumb spacing, they may not adequately reflect roadway longitudinal and cross slopes, inlet configuration, and ultra-short-duration rainfall characteristics, resulting in insufficient capture or overly conservative installations. This study develops an integrated, hydraulics-based framework to (1) estimate gutter flow and roadway spread using curb-and-gutter hydraulics, (2) propose an ultra-short-duration rainfall intensity estimation approach that explicitly incorporates the roadway time of concentration, (3) derive shape-dependent inlet interception equations through laboratory experiments, and (4) establish a practical inlet spacing design procedure and implement it as a decision-support tool. The overall computation follows a sequential workflow widely adopted in roadway drainage practice: rainfall–runoff estimation, gutter flow and spread calculation, inlet interception evaluation, and bypass routing. Laboratory tests were conducted under controlled combinations of approach flow rate, longitudinal slope, cross slope, and inlet configuration. Using the experimental dataset, inlet interception efficiency and captured discharge were regressed against geometric parameters and dimensionless hydraulic indicators, yielding empirical capture formulas applicable to common roadway conditions. The resulting equations are consistent with prior evidence that grate inlet performance is sensitive to inlet dimensions, roadway slopes, local depression, and approach-flow characteristics, while providing calibration for conditions in which existing generalized equations may be overly conservative or insufficient. Finally, inlet spacing is determined by coupling cumulative contributing runoff with the experimentally derived capture relations and explicit design constraints. The proposed method produces spacing tables by roadway geometry and design frequency and supports scenario-based design for flood-prone segments such as sag points and low-lying corridors. Overall, this study offers a quantitative pathway to shift inlet placement from fixed spacing to performance-based design that accounts for ultra-short-duration rainfall, roadway hydraulics, and inlet configuration.

This research was supported by National Research Foundation of Korea(NRF) grant funded by the Korean government(MSIT)(No.2022R1A2C1012374).

ABSTRACT. Rapid urbanization and the increase in impervious surfaces have made urban drainage systems increasingly complex, with surcharged flows exceeding design frequencies frequently occurring during localized heavy rainfall events. These surcharged flows induce energy losses within conduits, significantly reducing drainage capacity and exacerbating urban flooding damages as a major cause. Particularly, among various junction types of combined manholes, 3-way (T-type) junction manholes exhibit pronounced asymmetric flows and energy losses due to lateral inflow, necessitating quantitative analysis of loss characteristics under surcharged conditions. This study investigated standard installation criteria for rectangular manholes commonly installed in stormwater conduit facilities through literature review and field surveys to select experimental conditions, and constructed 1/5-scale hydraulic models following sewer facility standards. Experiments were conducted by varying the selected experimental conditions: the ratio of manhole width (b) to connecting pipe diameter (d) (b/d), the ratio of lateral inflow (Qlat) to outflow (Qout) (Qlat/Qout), and experimental flow rates (1.0, 3.0, 5.0 L/s). All experiments were conducted under invert-free conditions to represent typical existing urban manhole structures. Experimental results showed that loss coefficients increased as the lateral flow ratio increased, while loss coefficients decreased as the ratio of manhole width to connecting pipe diameter increased. This indicates that energy losses in surcharged junction manholes are significantly dependent on lateral inflow rates and the ratio of manhole width to connecting pipe diameter. Loss coefficients according to lateral flow ratios were calculated as 0.4~1.72 for b/d=3.0 (lateral flow ratio 0.00~1.00), and 0.4~2.2 for b/d=2.5 (lateral flow ratio 0.00~1.00). From these results, manhole size-specific loss coefficient equations considering variations in lateral inflow were derived. The presented loss coefficients and equations account for manhole dimensions, providing fundamental data for rational estimation of loss coefficients for rectangular 3-way junction manholes in sewer design and urban flood analysis.

ABSTRACT. To reduce damage from driftwood caused by increasingly frequent torrential rains and lessen environmental impact, permeable structures such as slotted weirs and riprap weirs are increasingly being used for river crossing structures like intake weirs and erosion control dams. Especially in Japan, the Basic Guidelines for Creating Rivers with Natural Features were revised in 2024. These guidelines state that efforts to create rivers with natural features should be considered at all stages and processes—before improvement, during improvement, after improvement, and during disaster recovery—for all Classes rivers. Consequently, permeable structures such as riprap weirs and fascine works are being reevaluated. Even in such a situation, it is difficult to say that evaluation methods for the flow resistance of permeable structures such as riprap weirs and fascine works have been established. Therefore, this study proposes using the permeability resistance coefficient Cv, determined in relation to the structure compactness C, for calculating the flow resistance of permeable structures. The permeability resistance coefficient Cv is calculated using the formula Cv = 2gie/(u^2), based on the water depth and flow velocity determined by hydraulic experiments. The experiments were conducted in flow fields with Reynolds numbers ranging from 3000 to 10000 and Froude numbers from 0.16 to 0.56 by varying the initial flow velocity and initial water depth. A model of a riprap weir (compactness C = 63%) and a plastic porous body (C = 12%, 5%) were installed at lengths ls is from 0.02 m to 0.60 m. Water depth and flow velocity were measured at the channel center. The obtained permeability coefficient Cv was organized in relation to the structure compactness C, and its validity was verified through non-uniform flow calculations that considered the resistance of permeable structures. As a result, comparing experimental and calculated values for the backwater depth of the structure revealed a correlation coefficient of 0.99, indicating excellent reproducibility. This confirmed that the permeability resistance coefficient Cv is useful for evaluating the resistance of permeable structures.

ABSTRACT. Abstract Many river facilities in South Korea were constructed decades ago and have been used for extended periods, raising concerns about performance deterioration and safety risks. Current maintenance practices rely primarily on periodic inspections and qualitative safety grades. However, this reactive approach typically initiates repairs only after visible defects are detected, which can increase restoration costs and limit the ability to anticipate sudden functional failures. Furthermore, the administrative lag between defect detection and the initiation of repairs can leave facilities in a degraded state for prolonged periods. This study proposed a method to predict performance degradation trends of sluice gates and culverts to support maintenance decision-making. We compiled historical records from precision safety inspections conducted from 1990 to 2025 for 3,529 facilities and applied a two-step framework combining Kaplan–Meier (KM) survival analysis with parametric distribution fitting. The transition to Safety Grade C was defined as the failure event indicating the need for maintenance, while facilities that had not reached Grade C by the last inspection were treated as right-censored. KM analysis estimated the median time to Grade C as 33.4 years. To overcome the discrete nature of the KM step function and enable continuous-time prediction, we fitted Weibull, GEV, Gumbel, and Gamma models to identify the stochastic model that best represents the deterioration process. Preliminary results suggest that the Weibull and Gumbel models provide the highest goodness-of-fit among the candidates, and their predicted median lives are consistent with the empirical KM median, supporting the reliability of the proposed approach. By generalizing inspection histories into continuous probabilistic models, this study enables quantitative prediction of remaining useful life and the probability of reaching Grade C within a planning horizon. The results are expected to help managers move beyond the limitation of defect-driven maintenance where interventions are initiated only after defects become apparent by identifying critical timing for Grade C transition and planning pre-emptive interventions accordingly.

Acknowledgement This work was supported by the Korea Environment Industry & Technology Institute (KEITI) through the Technology Development Project to optimize planning, operation, and maintenance of urban flood control facilities, funded by the Korea Ministry of Climate, Energy, Environment (MCEE) (RS-2024-00332378); and by the Korea Ministry of Climate, Energy, Environment (MCEE) as 「Graduate School specialized in Climate Change」

ABSTRACT. Mattress-type gabions, used to protect and green levees in Korea, are a revetment method combining vegetated mats and gabions. This eco-friendly revetment method was developed to address the low hydraulic stability of existing eco-friendly revetment methods. It consists of wrapping an open square wire mesh with geotextile, filling it with soil and stones, then covering it with seed-embedded geotextile and vegetated mats, and finally closing the wire mesh. This unique structure ensures stable resistance to flow even during the initial stages of construction through the dual structure of gabions and vegetated mats. Once vegetation is established, the root reinforcement of plant roots and the protective cover of stems and leaves enhance erosion resistance. However, limited experimental research on this type of revetment method has hindered its practical application in Korea. In this study, mattress-type gabions with established vegetation were installed on a 2 m × 10 m scale in a 2 m-wide steep channel with a slope of 13/100, which is operated by the River Experiment Center of the Korea Institute of Civil Engineering and Building Technology (KICT-REC). The erosion resistance experiments of mattress-type gabions were conducted with reference to the test method of ASTM D6460, which evaluates the erosion control performance of rolled erosion control products(RECPs). The experimental results showed that vegetated mattress-type gabions maintained acceptable erosion resistance up to approximately 260 N/m². This value is at least twice as high as the results of other previous eco-friendly revetment methods, demonstrating that mattress-type gabions can improve the low hydraulic stability of existing eco-friendly construction methods. These results are expected to be highly useful as reference data when applying mattress-type gabions or similar methods to river embankments in the future.

This research was supported by the Korea Institute of Civil Engineering and Building Technology (KICT) through the research project "Development of infrastructure disaster prevention technology based on satellites SAR" (Project No. 20260202-001).

ABSTRACT. Agricultural weirs are extensively installed in rural small streams to secure irrigation water; however, stagnant zones upstream of these weirs can cause water quality deterioration due to increased hydraulic retention time. Despite their widespread installation, quantitative studies comparing water quality between stagnant and flowing zones in rural small streams remain limited. This study aimed to analyze differences in hydraulic and water quality characteristics between stagnant and flowing zones under low-flow conditions at an agricultural weir in Geumseong Stream, Yeongsan River Basin. Monthly field monitoring was conducted from January to December 2024 (n=12). Velocity measurements and water quality sampling were performed under low-flow conditions, with simultaneous sampling 5 m upstream of the weir in both zones. The stagnant zone exhibited near-zero velocity, while the flowing zone maintained natural flow. Parameters measured included velocity, BOD, TOC, T-N, and T-P, analyzed following the Korean Ministry of Environment's Official Testing Methods. Statistical significance was evaluated using the Independent t-test for normally distributed parameters (T-P) and Mann-Whitney U test for non-normally distributed parameters (BOD, TOC, T-N) (α=0.05). The stagnant zone exhibited near-zero velocity, while the flowing zone showed a mean velocity of 1.24±0.31 m/s. BOD concentrations were significantly higher in the stagnant zone (3.78±3.06 mg/L) than the flowing zone (1.41±0.84 mg/L) (U=126.5, p<0.05). TOC concentrations were also significantly elevated in the stagnant zone (3.74±1.69 mg/L) compared to the flowing zone (2.50±1.26 mg/L) (U=111.5, p<0.05). Nutrients (T-N, T-P) showed higher trends in the stagnant zone but were not statistically significant (p>0.05). The stagnant zone exhibited significantly higher organic matter concentrations than the flowing zone, suggesting that increased retention time promotes organic matter accumulation. These findings indicate that hydraulic management strategies, including weir operation optimization, should be considered to mitigate water quality deterioration in agricultural small streams.

ABSTRACT. Indonesia has recently experienced tropical cyclones and extreme weather events that underscore the nation’s vulnerability to climate-related hazards. According to IPCC AR6, Indonesia is projected to warm by approximately 1.5–5°C over this century and face 45–110+ cm of sea level rise by 2100, depending on emission pathways. However, the increasing frequency and intensity of climate hazards have not been matched by integrated preparedness, particularly in the integration of climate risk into infrastructure planning, governance coordination, and long-term resilience strategies.

This study aims to evaluate risk exposure, structural vulnerability, and institutional readiness to climate hazards in Indonesia using multi-dimensional analysis, as well as to formulate resilience strategies. A PESTEL framework (Political, Economic, Social, Technological, Environmental, Legal) is applied to analyze climate risk exposure and identify actionable pathways. Political analysis considers governance capacity and coordination; Economic highlights recovery costs and infrastructure investment needs; Social emphasizes vulnerability and displacement; Technological reviews resilient infrastructure and early warning systems; Environmental underscores flooding, landslides, and ecosystem degradation; Legal examines regulatory enforcement, prioritization of sustainable land-use management, and alignment with international climate commitments.

The scope covers four representative case studies across diverse geographies. In November 2025, Cyclone Senyar, affecting Medan and Bener Meriah in Northern Sumatra, was developed under unusually warm ocean conditions, enabling the rare formation and landfall of a tropical cyclone near the equator. Medan, as a metropolitan center, suffered severe flooding due to rapid urbanization and inadequate drainage. Bener Meriah Regency experienced landslides and flash floods that disrupted rural connectivity and damaged coffee-producing areas, underscoring the vulnerability of highland communities. In early 2026, tropical depression conditions brought stronger winds and extreme rainfall in Kupang, exposing fragile infrastructure and limited adaptive capacity. Similarly, Semarang experienced prolonged extreme rainfall combined with rising coastal water levels that intensified tidal flooding, extending the duration and severity of inundation.

Findings reveal common challenges such as insufficient integration of climate risk into tangible disaster risk management and sustainable planning, persistent vulnerability from limited technological capacity, and gaps in legal enforcement. The study recommends a multi-level strategy combining strengthened governance, resilient infrastructure investment, community-based adaptation, innovation, and environmental improvement.

By synthesizing lessons learned across metropolitan, coastal, regional capital, and highland contexts, this research contributes to a holistic understanding of Indonesia’s climate resilience challenges. The PESTEL approach provides a structured adaptive pathway to translate vulnerability into resilience, ensuring preparedness, adaptability, and sustainability against future extreme weather events.

ABSTRACT. In recent years, flood hazards in Japan have become more frequent and severe due to climate change. It is therefore essential to urgently develop flood inundation maps. However, the development of such maps requires time and resources, and many areas remain unmapped.　 To address the challenge of unmapped regions, it is necessary to identify topographic factors related to inundation risk and to predict inundation risk using these factors, without performing numerical flood simulations. Based on this objective, this study conducted a logistic regression analysis within a generalized linear model to evaluate the relationship between flood risk and topographic indices using existing flood inundation area data from all first-class river systems across Japan. Two models were constructed, using binarized flood inundation areas under design rainfall and maximum expected rainfall scenarios as the response variables. Five topographic indices factors were selected as explanatory variables: elevation, slope angle, topographic wetness index, depression depth, and distance from the river. The standardized regression coefficients derived from the models showed that in the model with flood inundation area under design rainfall as the response variable, the coefficients varied significantly among basins, indicating that basin-specific topographical characteristics were dominant. Conversely, in the model with flood inundation area under maximum expected rainfall as the response variable, the coefficients showed less variation, suggesting that common topographical features contribute to flood risk regardless of the river system. Furthermore, as a result of the cluster analysis, the 109 first-class river systems in Japan were classified into six distinct clusters. These findings contribute to understanding the relationships between flood hazards and topographic factors and regional variations, and support flood risk assessment based on topographic information.

ABSTRACT. Ageing drainage infrastructure in Indian cities poses escalating risks to urban resilience, public health, and economic productivity. Rapid urbanization, the expansion of impervious surfaces, and climate- induced extreme rainfall have stressed the legacy systems designed for smaller populations and lower storm intensities. These outdated, often poorly maintained systems—many integrated with sewage networks—struggle to manage current rainfall volumes and runoff loads, making urban areas more vulnerable to flooding, contamination, and infrastructure breakdown. This paper aims at examining the phenomenon of ageing within urban drainage systems and develop a planning-oriented framework for assessing and managing their vulnerability. It adopts a structured review of academic, technical, and policy literature from 2000 to 2025, supplemented by case analyses from Indian and global cities such as Mumbai, Chennai, Delhi, London, and Tokyo, to trace the functional, structural, and temporal dimensions of ageing. Findings reveal that ageing is not a sole chronological factor but is driven by a combination of structural, hydraulic, institutional inefficiencies, etc. which are context specific. The paper delivers an assessment framework that will detect and measure ageing incorporating quantifiable parameters such as system age, material condition, maintenance frequency, water logging, etc. This tool enables city agencies to pinpoint the symptomatic areas and prioritize rehabilitation and renewal efforts. The proposed framework supporting sustainable, equitable, and resilient urban infrastructure management across Indian cities.

ABSTRACT. Recent flood management strategies in Japan emphasize basin-wide flood control, which requires not only evaluating measures implemented within individual tributary catchments but also understanding how the combined effects of multiple tributaries manifest in downstream reaches of the main river. However, quantitatively identifying the contributions of individual tributaries to flood hydrographs along the main river channel remains a challenge, particularly in basins where tributary inflows dominate flood dynamics. This study proposes a methodology to decompose flood hydrographs at any location along the main channel into contributions from upstream tributaries, thereby enabling a comprehensive understanding of flood runoff processes at the basin scale. The approach is applied to the upper Tenryu River basin during the July 2020 heavy rainfall event, which exhibited complex hydrological behavior due to heterogeneous rainfall distribution and strong tributary influences. The methodology combines Rainfall–Runoff–Inundation (RRI) model for the basin with two-dimensional unsteady flow simulation for the main river channel. Tributary inflow hydrographs and flow wave propagation velocities along the main channel are first computed through the coupled numerical simulation. Using the longitudinal distribution of wave speed derived from the two-dimensional model, travel times from each tributary confluence to arbitrary locations along the main river are estimated. Tributary inflows are then temporally shifted according to these travel times and superimposed to reconstruct the main river hydrograph, while simultaneously retaining the contribution of each tributary, and the reconstructed hydrograph is verified to be consistent with that derived from the two-dimensional flow simulation. For example, at a downstream location along the main river, the contribution of the Mibu River to the peak discharge was estimated to be approximately 14%, while that of the Koshibu River was approximately 16%. These results quantitatively demonstrate the substantial influence of tributary inflows on the peak discharge of the main river. To provide an overview of the contributions of all tributaries to discharge at each location along the main channel, a three-dimensional analytical representation was developed, enabling basin-wide interpretation of tributary contribution structures. Furthermore, the analysis reveals that tributary contributions vary markedly depending on flood stage. During the rising limb, smaller upstream catchments exert a relatively larger influence, whereas during the recession phase, larger tributaries become dominant. By organizing tributary contributions across all main-river cross sections, the proposed framework provides a basin-wide, stage-dependent perspective on flood runoff structure that can support more targeted flood management strategies.

ABSTRACT. In recent years, the increasing frequency and intensity of heavy rainfall events have significantly increased the risk of pluvial flooding, making conventional flood control measures insufficient. In response, flood management strategies in Japan have shifted toward a basin-wide approach that emphasizes the utilization of distributed storage and drainage functions at the basin scale. In the lower Chikugo River basin in Kyushu, a decentralized water management practice known as “Motase” has long been employed for pluvial flood mitigation. In this system, local residents collectively operate numerous sluice gates and overflow weirs distributed across an extensive network of agricultural creeks, thereby regulating runoff and flood impacts at the scale of the entire river basin. Through coordinated control of gates and weirs, Motase facilitates temporary storage of rainfall runoff and promotes gradual drainage within the canal network. Despite its long history and practical importance, the flood mitigation effectiveness of Motase has not yet been quantitatively evaluated at the basin scale. The objective of this study is to quantitatively evaluate the pluvial flood mitigation function of Motase by developing a numerical simulation model that explicitly represents its key functional characteristics. In the study area, the canal network is densely segmented by numerous water management facilities, forming in a highly compartmentalized system. To capture this structure, each segmented canal reach is modeled as an individual storage unit, allowing the representation of distributed retention and drainage processes associated with Motase to be incorporated into the numerical simulation. The model simulates canal storage dynamics and drainage pump station operations under multiple rainfall scenarios. Simulation results indicate that Motase substantially enhances both the storage capacity and natural drainage capacity of the canal network. As a result, the total inundation volume caused by pluvial flooding is reduced, and the required operation time of drainage pump stations is significantly shortened. These findings demonstrate that decentralized, community-based water management systems such as Motase can play a critical role in basin-wide pluvial flood mitigation. Furthermore, the results highlight the importance of incorporating local water management practices into numerical modeling frameworks to support effective and sustainable flood management strategies under increasing rainfall variability.

ABSTRACT. This study presents the development of a soft tentacle mechanism for a water-surface mobile rescue robot intended for safe physical interaction during aquatic rescue operations. Water-surface rescue scenarios often involve unstable environments, limited visibility, and the need for direct physical contact with victims, which requires grasping mechanisms that are both compliant and reliable. To address these challenges, the proposed system employs two independently actuated pneumatic soft-tentacle actuators designed to conform to the human body while minimizing localized contact stress. A dual-channel pneumatic actuation system and an onboard air-supply module are integrated into a compact platform, enabling autonomous operation and deployment on a small robotic vessel, while allowing flexible adaptation to water-surface disturbances and victim posture variations. Experimental evaluations are conducted using a human-body doll representing the chest and back to assess the interaction forces and contact pressures generated during grasping. The use of a human-body doll enables repeatable experiments under controlled conditions while approximating the geometric and compliance characteristics of the human torso. The measured values are analyzed with reference to established biomechanical safety limits relevant to human–robot physical contact in rescue contexts. The experimental results demonstrate that the proposed tentacle mechanism can achieve secure and compliant holding under low operating pressures, maintaining local tissue loads within safe ranges while providing sufficient grasping stability, even when contact conditions are not perfectly fixed. These findings indicate that pneumatic soft-tentacle actuators are well-suited for water-surface rescue applications requiring safe and adaptable human handling. This work provides a practical foundation for the future development of robotic rescue systems involving close and direct physical interaction with humans, and supports future investigations into hydrodynamic effects and grasping performance under flowing water conditions.

ABSTRACT. An effective understanding of sediment deposition and erosion within floodplains is essential from riparian space management, sustainable river planning and flood resilient land use. Floodplain sediment dynamics influence channel morphology, flood conveyance, and the stability of riparian zones. However, most related studies have been limited to hydraulic or hydrodynamic modeling approaches. Therefore, this study integrated Sentinel-1 differential interferometric synthetic aperture radar (DInSAR) deformation analysis was integrated with Sentinel-2 vegetation and soil moisture indices and one-dimensional hydraulic modeling to assess flood-induced sediment erosion and deposition in the Gamcheon River basin under non-flood, short flood, and long flood conditions. The results revealed a clear pattern of upstream erosion and downstream deposition during flood events, with pronounced depositional uplift during prolonged flooding and dominant surface lowering during non-flood conditions. These deformation patters closely matched with hydraulic model predictions and were supported by suspended sediment hysteresis behavior, strengthening confidence in the observed riparian geomorphic responses. The findings of this study demonstrate the effectiveness of the proposed integrated approach for quantifying floodplain sediment dynamics, supporting riparian space management by improving spatial characterization of sediment dynamics in flood-prone, data-scarce basins and informing adaptive sediment management strategies.

ABSTRACT. Landslide-generated tsunamis are relatively rare but can cause severe coastal damage. Previous experimental and numerical studies have investigated their generation mechanisms and provided empirical relationships to estimate scalar quantities such as maximum wave height or run-up. However, such scalar indicators alone are insufficient for practical, real-scale analysis of landslide-generated tsunamis, where the temporal evolution of water-surface elevation and flow velocity must be resolved rather than represented by peak values alone. In real-scale tsunami analysis, it is common to model the highly nonlinear generation process using three-dimensional numerical methods and then transfer the resulting near-field time series as boundary conditions to offshore propagation models based on nonlinear dispersive wave equations. While this hybrid framework provides high physical fidelity, performing the generation stage in three dimensions is computationally demanding and severely limits the number of scenarios that can be analyzed. To address this limitation, this study develops deep neural network (DNN) models that directly predict near-field water-surface elevation and horizontal velocity time series of landslide-generated tsunamis, reframing the near-field generation process as a data-driven time-series prediction problem rather than a purely physics-based simulation task. Two DNN models are constructed: one trained on experimentally measured free-surface elevation time series from 106 two-dimensional laboratory experiments, and another trained on horizontal velocity time series obtained from three-dimensional OpenFOAM simulations, as velocity measurements were unavailable in the experiments. Model performance is evaluated using leave-one-out cross-validation and benchmarked against laboratory observations, OpenFOAM results, and regression-based empirical formulas. The results show that the DNN models reproduce the temporal evolution of free-surface elevation with accuracy comparable to, and in many cases exceeding, that of OpenFOAM. Reliable reproduction of horizontal velocity time series is also achieved, particularly for partially submerged landslide cases, where the training dataset is largest and flow characteristics are more homogeneous. Scalar wave characteristics extracted from the predicted time series (first-wave crest and trough amplitudes and the first-wave period) also demonstrate predictive skill comparable to existing machine-learning approaches and empirical relationships. These findings indicate that DNN-based models can serve as efficient surrogates for near-field landslide-tsunami generation, enabling rapid production of physically consistent water-level and velocity boundary time series that cannot be obtained from scalar predictors alone. This capability provides a practical pathway for large-scale scenario-based and probabilistic hazard assessments, where repeated high-fidelity simulations are computationally prohibitive, and facilitates efficient integration with long-wave tsunami propagation models.

ABSTRACT. The impacts of climate change are increasingly evident worldwide, resulting in recurrent floods and droughts. In rivers, sediment load is expected to increase with increasing flood discharge. An increase in sediment load alters river environments, which in turn affects the safety of river infrastructure and the health of river ecosystems. Sediment Rating Curves (SRCs) obtained from field measurements are widely used in practice, but rarely have physical meanings. In contrast, total sediment load formulas, based on empiricism and experiments, do not adequately reflect localities.

This study conducted a sensitivity analysis of five widely used formulas for total sediment load. Yeoju Station on the Han-gang River is selected as the study site. The applied formulas include the Engelund–Hansen, Ackers–White, Yang, Brownlie, and Karim formulas. Discharge, median grain size of bed materials, and bed slope are selected as independent variables. The sensitivity analysis results indicate that the sediment load is most sensitive to discharge, followed by bed slope and median grain size. Sediment load responds to these independent variables consistently with Lane’s relationship. Performance tests of proposed SRCs and five total sediment load formulas are also conducted for the five major rivers in Korea using data measured in 2024, provided by the Korea Institute of Hydrological Survey. According to the performance test results, SRC performs best in the Han-gang River and Yeongsan-gang River. The Engelund–Hansen and Karim formulas show the best performance in predicting total sediment load in the Nakdong-gang River and Seomjin-gang River.

This work was supported by Korea Environment Industry & Technology Institute(KEITI) through 「Research and Development on the Technology for Securing the Water Resouces Stability in Response to Future Change(RS-2024-00335281)」 Program, funded by Korea Ministry of Climate, Energy, Environment(MCEE)

ABSTRACT. Sediment bypass tunnels (SBTs) are increasingly adopted as an effective measure to mitigate reservoir sedimentation; however, severe abrasion from high-velocity flow and gravel- and coarse-sediment transport remains a major challenge for tunnel durability. As a countermeasure, self-lining—sediment deposition between artificial roughness elements—has attracted attention as a passive and sustainable protection method. In this study, laboratory experiments were conducted to investigate how strip roughness geometry influences self-lining processes in shallow supercritical open-channel flows. Two types of roughness elements, triangular and square ribs, were installed on the channel bed with different relative spacings (λ/d). Flow structures under clear-water conditions were measured using particle image velocimetry (PIV), while sediment-feeding experiments were performed to evaluate deposition behavior in terms of coverage rate (CR) and volume rate (VR). The PIV measurements revealed that square ribs generate large and stable recirculation zones occupying most of the inter-rib region, whereas triangular ribs induce strong downward jets near reattachment points and pronounced surface fluctuations. During sediment feeding, complete filling between roughness elements was observed for both geometries. After the sediment supply stopped, however, clear differences emerged: square ribs retained a larger fraction of deposited sediment, while triangular ribs exhibited enhanced sediment loss associated with unstable flow structures. Narrow spacing (λ/d = 3) promoted continuous coverage regardless of geometry, whereas wide spacing (λ/d = 10) inhibited the formation of a stable self-lining layer. These results demonstrate that roughness geometry plays a critical role in controlling flow stability and sediment retention. In shallow-flow conditions with limited sediment supply, square strip roughness elements are more effective than triangular ones in maintaining self-lining layers, providing practical guidance for abrasion-resistant design of sediment bypass tunnels

ABSTRACT. Alternate sandbars induce flow meandering during floods and generate flow attack points, which can potentially cause erosion of riverbanks and levees. During Typhoon Hagibis in 2019, significant bank erosion occurred in the Chikuma River during the recession period after the flood peak. This erosion was attributed to changes in sandbar morphology and ultimately resulted in the collapse of a railway bridge of the Ueda Electric Railway. As the migration and development of alternate sandbars during floods modify the location and intensity of flow attack points, understanding sandbar behavior in response to flood hydrographs is essential for elucidating bank erosion risk and for improving river management. The objective of this study is to clarify the migration and development processes of alternate sandbars under unsteady flow conditions. Numerical simulations were conducted in which the rates of water level rise and fall were systematically varied to represent different flood hydrograph shapes. To verify the validity of the numerical approach, the simulation results were compared with laboratory experiments previously conducted under unsteady flow conditions, focusing on changes in sandbar morphology and downstream bar migration. The comparison demonstrates that key characteristics of alternate sandbars, including migration distance, bar height, and wavelength, obtained from the numerical simulations are in good agreement with those observed in the experiments. Subsequently, a series of numerical experiments was performed to investigate the influence of water level variation rates on sandbar behavior. The results indicate that differences in water surface slope and flow characteristics between the rising and falling stages of floods significantly affect the development and migration of alternate sandbars. In particular, the recession stage plays an important role in modifying bar geometry and bar height, which may enhance bank erosion potential. These findings highlight the necessity of considering unsteady flow effects, especially during the falling stages of flood hydrographs, in assessments of sandbar dynamics and riverbank stability.

ABSTRACT. Recent climate change has intensified riverine driftwood issues due to the increasing frequency of torrential rains and shifts in forest management. With escalating risks of infrastructure damage and flooding caused by channel obstruction, effective countermeasures are urgent. One of the causes of this is forest degradation. It is not easy to curb forest degradation and restore forests to a healthy state. One of realistic and economic counter measures of driftwood disaster is to construct driftwood capturing facility in rivers channels. However, designing these facilities is complex due to the intricate interactions between wood motions and flow streuctures. A bypass-type driftwood capture system is typically effective when installed on the outer bank of river bends by utilizing centrifugal force. Conversely, application in straight channels requires some additional devices, such as installing groynes on the opposite bank to divert driftwood into the capture zone. This study investigated the fundamental characteristics of such a system using hydraulic experiments and 3D numerical simulations. In the hydraulic experiments, we examined the influence of inlet angle, gryoynes and driftwood density on the capture ratio. The numerical analysis utilized NaysCUBE, a fully 3D solver capable of accounting for hydrodynamic pressure and reproducing complex phenomena like the secondary flow. Experiments varying the inlet opening angle (8°, 15°, 25°, 90°) revealed that 15° yielded the highest capture ratio, while 25° was the lowest. This is attributed to driftwood movement along the side wall and rebounding behavior at the inlet's straight section. The efficacy of opposite-bank groynes was clearly demonstrated. The study identified that the groyne’s effect consists of two mechanisms: advection path changes due to flow deflection and direct collision. Numerical analysis indicated that, under the tested conditions, these two effects are of a comparable magnitude. Regarding specific gravity (0.8, 1.2, 1.5), the capture rate was lowest at 1.2. This feature was attributed to the transverse velocity distribution near the mid-depth layer moving away from the bypass area. This characteristic velocity profile is likely driven by the combination effect with secondary flows induced by the groyne and inlet streamline curvature. Consequently, the study highlights the critical relationship between specific gravity and secondary flow structures. The computational results with the 3D numerical analysis model showed that it can reproduce the driftwood behavior and capture characteristics observed in the experiments to a certain extent. However, regarding quantitative prediction, discrepancies with the experimental values were observed, indicating the need for model improvement.

ABSTRACT. Suspended sediment transport models in turbulent free surface flows commonly assume a uniform particle size despite natural systems being commonly polydisperse, with size-dependent settling velocities and inter-particle interactions that can substantially alter concentration dynamics. In addition, existing non-uniform sediment studies in non-equilibrium state often adopt simplified and inconsistent bottom boundary conditions (e.g., equilibrium concentration, zero-flux, zero vertical diffusion gradient, or relaxation-type non-equilibrium conditions). The lack of a systematic comparison across these paradigms limits both mechanistic interpretation and predictive reliability under varying hydrodynamic conditions.

This study proposes an analytical framework to investigate steady-state, non-equilibrium transport of non-uniform sediment in turbulent free surface flows. A generalized bottom boundary condition is formulated and specialized into six commonly used forms, enabling a unified assessment of how boundary assumptions control the vertical concentration profile and the streamwise development of suspended sediment. The analysis further quantifies the roles of key parameters, including upstream sediment concentration, particle settling velocity, and deposition and entrainment rates, and examines how sediment non-uniformity modifies the concentration distribution.

A two-dimensional steady-state mathematical model is developed, yielding second-order linear partial differential equations with homogeneous and non-homogeneous boundary conditions. Solutions are obtained analytically using the principle of superposition combined with an integral transform method.

Results show that flux-equilibrium states can emerge when deposition and entrainment rates are imbalanced, producing near-bed concentrations that depart from bed-equilibrium values. Depending on the bottom boundary condition, the far-field vertical profiles may remain invariant or exhibit marked sediment depletion/stratification. The far-field concentration distribution is insensitive to upstream inflow concentration, whereas settling velocity is a primary determinant of the vertical structure. Deposition-related parameters strongly affect the non-equilibrium sediment concentration distribution. Finally, non-uniform sediment mixtures tend to produce more vertically homogeneous concentration profiles than uniform sediment due to hiding–hindering effects among particle size classes. These findings provide transferable guidance for selecting bottom boundary conditions and interpreting non-equilibrium transport predictions for polydisperse sediment in turbulence free surface flows, thereby directly supporting the H2-P1 project and its objective of resolving physically consistent vertical sediment distributions and the monitoring and assessment of complex eco-urban solutions going forward.

ABSTRACT. The Joganji River in Toyama Prefecture is one of the steepest rivers in Japan. In recent years, both bed material coarsening and bed degradation have been observed, and the riverbed is currently characterized by mixed sediment beds containing abundant fine particles. In mixed sediment beds, sediment sorting influences bed deformation and sandbar formation mechanisms differently from those observed in uniform sediment beds. Previous studies have suggested that sediment sorting tends to reduce bar height and bar wavelength. However, sandbar dynamics associated with long-term bed degradation and changes in grain-size distribution remain poorly understood. The purpose of this study is to clarify the differences in alternating bar formation and development under stepwise changes in sediment supply, including conditions of bed degradation and aggradation, by means of experimental flume tests using uniform and non-uniform sediment. The experimental flume was a straight rectangular channel with a length of 12 m, a width of 0.3 m, and a bed slope of 1/120. Three sediment types-uniform, non-uniform, and mountain sediment-were used, with their mean grain sizes adjusted to be nearly identical. A total of nine experimental cases were conducted. In uniform and non-uniform sediment cases, a series of continuous experiments was performed in which the sediment supply rate was varied stepwise to values of 1.0, 0.8, 1.0, and 1.2 relative to the equilibrium sediment supply rate. Regarding the bar wavelength, the uniform sediment cases consistently exhibited longer wavelengths than the non-uniform sediment cases. Similarly, bar height was larger in many uniform sediment cases and is consistent with previous studies. However, when comparing the bed elevation on bars between uniform and non-uniform sediment cases, the bar elevation in the non-uniform sediment cases was significantly higher than that in the uniform sediment cases. This difference is attributed to armoring of the low-flow channel, which suppresses local scour under non-uniform sediment conditions. The bar migration rate was higher in the non-uniform sediment cases than in the uniform sediment cases, which is attributed to the lower bar height under non-uniform sediment conditions. In addition, the response of bar characteristics to stepwise changes in sediment supply was more pronounced in the uniform sediment cases, whereas variations in bar height and wavelength were limited in the non-uniform sediment cases. This is because, under constant discharge conditions, bed armoring is less likely to be disrupted, resulting in a smaller effect of sediment supply on bar height than in the uniform sediment cases.

ABSTRACT. Accurate assessment of bedload transport remains a major challenge in fluvial sediment research. Many studies have attempted to improve prediction accuracy by developing enhanced bedload formulas based on various theoretical and empirical approaches. However, recent meta‐analyses have highlighted substantial data scatter in the relationship between Shields stress and the dimensionless bedload transport rate, indicating that even sophisticated formulas inherently contain unavoidable uncertainty. In this study, we propose a rescaling method for Shields stress aimed at reducing this scatter. First, contributing factors were examined to identify the key driver of variability. Grouping the dataset by channel bed slope revealed that increasing slope systematically leads to an overestimation of the Shields number. Based on this finding, we developed a new slope-dependent scaling expression for Shields stress. Application of this scaling significantly reduced the field data scatter by an average factor of approximately 15, and by up to several orders of magnitude in extreme cases. Furthermore, the rescaled Shields stress improved the performance of existing bedload formulas and enhanced the stability of newly fitted transport equations. These results suggest that channel-slope-based rescaling provides a promising and broadly applicable strategy for mitigating uncertainty in bedload estimation.

Acknowledgements: This research was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No. RS-2024-00336456).

ABSTRACT. This study applies a grid-based, explainable machine learning framework to identify key flood influencing factors by integrating historical flood trace maps with multi-layer spatial datasets. Four tree-based algorithms were employed to classify flooded and non-flooded grids at a 100 m resolution. Model performance was evaluated across subregions to examine spatial variability in predictive accuracy and the relative importance of influencing factors. The models achieved moderate-to-high predictive performance, with recall and F1-scores of up to 0.81 and 0.75, respectively. Feature-importance analyses consistently identified shortand long-duration rainfall (3-hour and 12-hour maximum), 5-day antecedent rainfall, maximum wind speed, groundwater level, and proximity to detention ponds and rivers as dominant predictors of flood occurrence. Their relative influence, however, varied across subregions. In inland and high-elevation basins, rainfall duration and groundwater dynamics were the primary controls, whereas coastal and urbanized zones were more influenced by drainage connectivity and proximity to infrastructure. These spatial contrasts reflect the heterogeneous hydro-geomorphological setting of Jeju Island and emphasize the importance of region-specific flood warning thresholds and adaptive management strategies. Overall, the findings highlight the potential of grid-based explainable machine learning for spatially adaptive flood risk assessment in volcanic island environments under changing climate conditions.

ABSTRACT. Since the early 2020s, artificial intelligence (AI) has become increasingly prominent in both academia and industry. In water resources engineering, AI-driven methods have been widely investigated to enhance the prediction of floods, droughts, water supply, and other hydrological processes. Recently, agentic AI—where large language models (LLMs) coordinate multiple analytical tools—has emerged as a promising paradigm for integrated analysis and operational decision support.

However, existing studies largely emphasize model development, with limited attention to sustainable operationalization through machine learning operations (MLOps). This limitation is critical in hydrological applications, which require continuous retraining, systematic evaluation, and reproducible workflows for real-world deployment.

We present NeuralRiverOps, an operational framework that integrates MLOps and agentic AI for multi-site flood forecasting in large river basins using long short-term memory (LSTM) networks. The framework supports sequential upstream-to-downstream modeling based on the neuralhydrology library and incorporates an MLflow-centered MLOps pipeline for model tracking, retraining, inference, and evaluation. PostgreSQL and MinIO manage structured time-series data and model artifacts, respectively. An agentic AI interface, implemented with Ollama and OpenWebUI, enables interactive orchestration of the pipeline. All components are containerized via Docker Compose to ensure reproducibility and scalability.

This framework advances the transition from isolated model development to robust, operational AI systems for disaster risk management.

ABSTRACT. Clogging conditions of storm drain inlets are a significant contributor to pluvial flood risk. Current mitigation strategies rely primarily on routine inlet maintenance, which provide limited responsiveness to increasingly frequent and unpredictable extreme rainfall events. This study proposes a novel framework to quickly evaluate inlet conditions in advance of anticipated storm events. Critical inlets are first identified through hydrological analysis as those that intercept flow from public roadways toward private properties and are located at elevations lower than the corresponding parcel entry points. The clogging conditions of these inlets are assessed using a computer vision model applied to street-level imagery acquired by a vehicle equipped with a 360° panoramic camera and global positioning system. A case study demonstrates that clogged critical inlets create anomalous flow patterns, diverting runoff toward downstream property parcels and increasing localized pluvial flood hazard. The proposed framework enables real-time identification of high-risk inlet conditions and provides a scalable decision-support tool for proactive pluvial flood risk management.

ABSTRACT. Since the early 2020s, the integration of artificial intelligence with conventional numerical modeling has emerged as an important paradigm in water resources management. In estuarine engineering, timely and accurate prediction of salinity intrusion and thermal stratification is essential for operational decision-making during barrage openings. However, multi-dimensional hydrodynamic models remain computationally intensive, limiting their applicability in real-time environments. To address this limitation, AI-based surrogate models have been investigated as efficient alternatives capable of approximating complex hydraulic simulations with substantially reduced computational cost.

The authors present an operational surrogate modeling framework for predicting two-dimensional vertical distributions of salinity and temperature in a Korean river estuary. The framework is trained using an extensive dataset derived from calibrated numerical simulations representing diverse hydrological and tidal conditions. Among the deep learning architectures examined, a Vector-to-Vector MLP-Mixer–based model demonstrated superior predictive performance with low estimation error. The model incorporates spatial grid dependencies by accounting for interactions among adjacent cells and supports sequential temporal analysis to represent evolving estuarine dynamics.

For operational applicability, the framework is designed to interpolate unobserved vertical layers and deliver near real-time predictions suitable for design assessment and adaptive management. This study contributes to the transition from computationally demanding numerical simulations toward reliable, real-time AI-based systems for integrated estuarine management.

ABSTRACT. Flood retention basins (FRBs) play a pivotal role in flood management strategies. The available flood storage capacity of an FRB dictates the maximum volume of floodwater it can store. It often varies due to the changing environment, which in turn impacts its flood control effectiveness. Based on the principle of allowing only specific areas to be inundated for flood control, we propose a systematic framework to quantify the variations in the available flood storage capacity of an FRB caused by the combined effects of human activities and climate change. Within this framework, the Variable Infiltration Capacity hydrological model is adopted to simulate runoff generation in FRBs and the occupied available flood storage capacity due to the local flood is estimated. To quantify the impacts of human activities, we develop a method for estimating the available flood storage capacity based on its water - level storage volume curves and the inundation principle. Then a Geodetector model is applied to determine the contributions of human activities and climate change. When applying this framework to 42 FRBs in the middle reaches of the Changjiang River Basin, the results revealed a significant 9.6% decrease in the total available flood storage capacity from 2000 to 2020. It was found that the reduction in available flood storage capacity due to human activities was much greater than the variations caused by climate change. The decrease in available flood storage capacities in the FRBs has led to an increase in the frequencies of flood inflows, which represent the maximum flood that the FRBs can prevent. These frequencies have increased from 0.111%, 0.167%, and 0.100% to 0.143%, 0.250%, and 0.167%, respectively. This decrease not only affects the protection targets of the local FRBs but also trigger cascading effects of excess floodwater for the downstream FRBs. Our study is expected to not only quantify variations in the available flood storage capacity but also offer valuable insights for the conservation of FRBs.

ABSTRACT. Inter-basin water transfers (IBWTs) profoundly alter natural hydrological regimes, adding complexity to the trade-offs among different objectives. Understanding how multi-purpose reservoir systems respond to such external disturbances and identifying pathways to achieve multi-objective synergy are essential for advancing sustainable water resource management. This study investigates the response dynamics of multi-objective trade-offs in a reservoir group subjected to IBWT interventions and explores potential pathways for realizing synergistic benefits. A framework was applied to the Hanjiang River Basin (HRB), which serves as the water source area for the middle route of China’s South-to-North Water Transfer Project. The Variable Infiltration Capacity (VIC) model was calibrated and used to simulate runoff series across multiple spatial and temporal scales, capturing the basin’s hydrological response to varying climatic conditions. A modified Tennant method was then employed to establish differentiated ecological flow standards. A reservoir generalization model was developed to simulate reservoir group operations under scenarios combining varied inflows and water transfer schemes. The framework provides spatiotemporally explicit characterizations of the trade-offs and synergies among water supply, hydropower, and ecological objectives, and elucidates the perturbing effects of IBWTs on their interactions. The results indicate that water donation tends to reinforce negative feedbacks between water supply and hydropower, as well as between water supply and ecology, while weakening the positive feedback between hydropower and ecology. In contrast, water receiving appears to exert a positive regulatory influence on these feedbacks, potentially fostering synergies among the three objectives. These findings contribute to a systems-level understanding of reservoir responses to inter-basin water transfers and may offer useful insights for coordinating multi-objective trade-offs in complex water resource systems.

ABSTRACT. In the UK and many other developed regions, stormwater regulations require rainwater from impermeable surfaces at new developments to be captured attenuated and released in a controlled manner (NSSDS 2025). However, they are infrequently configured to maximise water reuse opportunities. A method is set out describing how existing rainwater attenuation tanks can potentially be 1) identified 2) evaluated and 3) converted into long-term rainwater storage (to enable reuse) rather than their current “flow-control” configuration. The procedures are conducted at case study catchment in the UK in Letchworth Garden City, a location where a Flood and Coastal Resilience Innovation Project “ResilienTogether” has installed a wide range of hydrometrical monitoring systems since 2022. The work supports the project’s nature-based-solutions workpackage, wherein new and existing stormwater assets are being re-designed to provide greater value in future climate scenarios. In this instance the project builds on concepts presented in (Melville-Shreeve et al, 2016) wherein rainwater harvesting systems were proposed as having the dual-purpose of stormwater management. In essence the project explores the possibility of converting existing single-use stormwater flow control tanks into dual-use water reuse assets that also manage rainwater during intense storms.

References: NSSDS - National standards for sustainable drainage systems (2025) https://www.gov.uk/government/publications/national-standards-for-sustainable-drainage-systems/national-standards-for-sustainable-drainage-systems-suds Accessed 19/12/25 Melville-Shreeve, P., Ward, S. and Butler, D. (2016). Rainwater Harvesting Typologies for UK Houses: A Multi Criteria Analysis of System Configurations, Water, 8(4), 129

ABSTRACT. Reconstructing high-frequency dissolved oxygen dynamics from low-frequency sampling is critical for accurate water quality assessment. This study introduces a novel modeling framework that integrates transfer learning for high-accuracy reconstruction with Structural Equation Modeling for process interpretation. Developed and evaluated using long-term monitoring data from a coastal system, the transfer learning model demonstrated superior performance over a baseline deep learning model. Analysis of the reconstructed high-frequency series revealed detailed spatiotemporal patterns and long-term trends not discernible from original samples. SEM identified temperature and sea level as dominant environmental drivers of DO variation. The framework validates that reconstructed data can uncover critical dynamics for management, providing interpretable scientific evidence to inform both water quality strategies and climate adaptation policies.

ABSTRACT. The Han River, 1577 km long, is the longest tributary of the Yangtze, with an average annual runoff of 45.8 billion m3. The Danjiangkou Reservoir, located in the upper Han River, has a total storage capacity of 31.95×109 m3. The reservoir is the water source project for the Middle Route of the South to North Water Diversion Project (SNWDP-MR). Over 75.5×109 m3 of water has been transported from the reservoir to northern China since 2014. Starting from January 2025, the inflow of the Han River was significantly lower than normal. The cumulative precipitation in upper Han River from April to July was only 300 mm, ranking the lowest since 1961. The natural inflow of Danjiangkou Reservoir is nearly 60% less than the same period over the years, which is the lowest since it was built in 1969. As a result, the water level of the reservoir dropped below 150 m on May 14th. To ensure the water supply for SNWDP-MR and the safety of water use downstream of the reservoir, emergency water regulation was carried out. The Danjiangkou and upstream reservoir group were finely regulated with "per m3/s of flow rate, per m3 of storage capacity, and per cm of water level". On the one hand, "increasing the water source", 19 reservoirs in the upper reach of the river were regulated to replenish the Danjiangkou Reservoir with nearly 1.8×109 m3 of water. On the other hand, "reducing water consumption", the water discharged from the Danjiangkou Reservoir and supplied for water diversion were reduced to the limit by 3.99×109 m3. By May 20th, the reservoir water level had stopped falling and risen to over 150 m by May 30th. During this period, joint operation of the Yangtze and Han River reservoir groups were also strengthened. The discharge flow of the Three Gorges Reservoir was regulated to be over 14000 m3/s, ensuring that the diversion of water from the Yangtze to the lower Han River is not less than 400 m3/s, and ensuring the safety of water supply in the middle and lower Han River. Furthermore, by dredging and extending the water supply network, the intake conditions of 20 water plants along the Han River were improved, ensuring the drinking water safety of 100,000 people and the irrigation water demand of 7.86×106 mu of farmland. This presentation will introduce the drought resistant operation of the reservoir group under extreme low water situation.

ABSTRACT. Malaysia is a country considered a water-rich country in Southeast Asia. Despite being blessed with abundance of water resource, Malaysia is not free from water supply issue. This paper discusses the potential of pond-based rainwater harvesting systems (RWHS) as a supplementary source for water reuse, complementing traditional rooftop catchments. Stormwater quality performance of wet ponds and detention ponds was evaluated between 2 monitoring periods (April to November 2003 and November2023 to January 2024). The findings reveal that wet pond and detention pond consistently achieved better water quality, particularly in lowering suspended solids, organic load, and nutrients. With proper treatment, detention ponds can function as effective on-site systems for water reuse. The findings are demonstrated through practical applications at the Bio-Ecological Drainage System (BIOECODS) facility within the USM Engineering Campus, supporting the Riverside Water Catchment Project (TAPS) and contributing to Malaysia’s water security.

ABSTRACT. Climate change has intensified global air temperatures, consequently increasing river water temperatures through direct heat transfer and altering hydrological patterns. Elevated river temperatures can significantly degrade aquatic ecosystems by reducing dissolved oxygen (DO), increasing biochemical oxygen demand (BOD), and disrupting thermal balance essential for native fish survival. This study aims to assess the impact of rising river temperatures on fish habitat suitability, with a focus on Hemibagrus nemurus, an endemic species sensitive to environmental changes. River hydrodynamics were simulated using a one-dimensional (1D) steady flow model in HEC-RAS, while water quality was analyzed using the General Constituent, Nutrient, and Water Temperature–Constant modules. Model calibration was performed using observed field data from July and August 2024. Habitat suitability was evaluated using a Suitability Index (SI) approach, incorporating key variables such as water temperature, DO, BOD, depth, and flow velocity. The results were further integrated with Weighted Usable Area (WUA) analysis to quantify the spatial extent of suitable habitat under thermal stress scenarios. In July 2024, under a 5°C temperature increase, the SI ranged from 0.385 to 0.763 but decreased significantly to 0.086–0.171 under a 9°C increase. In August 2024, the SI ranged from 0.305 to 0.743, with no optimal habitat area remaining under the highest temperature scenario. WUA values were recorded at 18.28% in July and increased to 36.48% in August, reflecting seasonal baseflow variations. These findings underscore the vulnerability of Hemibagrus nemurus habitats to climate-induced thermal stress. To enhance ecosystem resilience, preserving riparian vegetation and reducing surface runoff through improved land use management are recommended as key adaptation strategies for sustaining riverine biodiversity.

ABSTRACT. The hydrological performance of green roofs on high rises buildings is understudied. A collaborative research project between the University of Auckland and the Auckland Council established a green roof experiment to assess hydrological performance relating to substrate depth, plant species, substrate materials, and tray layouts. Seven experimental green roof trays with distinctive configurations representing one ballast stone tray, two Eco-pillow trays, two Daltons living roof mix trays, and two Daltons pine grow mix trays have been installed and monitored for over a year. The collected data account for seasonal variabilities in Auckland by assessing their performance in relation to local climatic conditions. In addition to substrate characteristics, the vegetation layer of the trays is comprised of local plant species selected by the local Iwi, Ngāti Whātua Ōrākei. Hydrology monitoring collects runoff and peak discharge reduction in rainfall events by comparing the actual precipitation depth with the runoff depth delivered by the configured green roof systems. The calibrated pressure transducers and customed orifice tubes estimate the runoff volume and peak discharge. This paper presents some of the results to date from the data collection. With the intensification of urbanisation and climate-induced shifts in precipitation patterns in New Zealand, green roofs have become a pivotal element within urban stormwater management. The insights from this study may foster technical innovations and inspire broader societal and policy transformations, thereby laying the groundwork for the development of more sustainable and resilient urban landscapes that harmonise human habitation with the natural environment.

ABSTRACT. Irrigation expansion has been a key adaptation strategy for sustaining agricultural production in water-limited North China, yet its net impact on terrestrial water storage (TWS) remains difficult to quantify because irrigation effects are spatially heterogeneous and often confounded by climate variability and land–atmosphere feedbacks. This study aims to quantify the impact of irrigation expansion on regional TWS over the North China Plain (NCP) during 2002–2020 and to diagnose where and when irrigation signals are most clearly detectable. We integrate three lines of evidence at a consistent spatial scale: (i) irrigation expansion (ΔIRR) derived from long-term irrigation maps, (ii) evapotranspiration (ET) and its components (total evaporation and transpiration) from satellite-based products, and (iii) monthly TWS changes (TWSC) from GRACE observations. All datasets are aggregated onto the GRACE mascon grid to enable cell-wise attribution. Temporal trends in ET and transpiration are quantified using the robust Theil–Sen estimator on monthly time series, complemented by seasonal (growing-season) analyses and monthly climatology diagnostics. To explicitly address spatial heterogeneity, we develop a regime-based classification that separates mascon cells with consistent irrigation–ET responses (ΔIRR>0 and transpiration/ET trend >0) from decoupled regimes (ΔIRR>0 but weak/negative ET response) and climate-dominated regimes (ET increase without irrigation expansion). Within the irrigation-consistent regime, a dose–response analysis (ΔIRR binning versus ET/transpiration trends) is used to quantify the intensity dependence of irrigation impacts. Results indicate pronounced spatial contrasts: only a subset of the NCP exhibits clear irrigation-consistent increases in transpiration/ET, while large areas show weak or absent ET responses despite irrigation expansion, implying the roles of climatic constraints, mixed land cover within mascons, and water-management practices. Seasonal diagnostics reveal that irrigation signals are most evident during the growing season rather than uniformly across the year. These findings highlight the necessity of regime-aware attribution for robustly linking irrigation expansion to consumptive water use and TWS changes, providing actionable insights for sustainable water management in intensively irrigated regions.

ABSTRACT. Nepal’s rural and mountainous communities face increasing risks from climate change, water scarcity, hydrological extremes, and ecosystem degradation, with direct implications for food security, public health, biodiversity, and sustainable development. Addressing these interconnected challenges requires integrated, inclusive, and locally driven approaches to water governance. This paper presents the experience of the Ramechhap UNESCO Peer Education Program – Nepal (RUPEP Nepal) in implementing community-led Integrated Water Resource Management (IWRM) to advance climate resilience, food security, and nature conservation in climate-vulnerable rural settings.

The initiative aligns with the 2030 Agenda for Sustainable Development, particularly SDGs 6 (Clean Water and Sanitation), 2 (Zero Hunger), 13 (Climate Action), and 15 (Life on Land), and operationalizes the water–energy–food–ecosystem nexus through participatory planning and nature-based solutions. Key challenges addressed include declining freshwater availability, flood and drought risks, water pollution, inadequate sanitation services, climate impacts on smallholder agriculture, and degradation of freshwater and terrestrial ecosystems.

Using a participatory and multi-stakeholder methodology, the program combined peer education, community-based adaptation, capacity-building, and climate-resilient infrastructure development. Core interventions included safe drinking water supply systems, eco-efficient water treatment technologies, sanitation and solid waste management initiatives, flood early warning systems, watershed restoration through reforestation, biodiversity protection measures, and the promotion of sustainable, water-efficient livelihoods linked to food security. Monitoring and evaluation drew on household surveys, health indicators, environmental assessments, and local government data.

Results indicate tangible development and resilience outcomes. Access to safely managed drinking water improved for over 2,000 households, while reported waterborne diseases declined by approximately 30% in targeted communities. Disaster preparedness and adaptive capacity increased through community-based early warning mechanisms. Watershed restoration and pollution control contributed to improved river health and ecosystem services, while capacity-building strengthened local institutions and enhanced the leadership of women and youth in water governance.

The findings demonstrate that community-driven IWRM, supported by peer education and nature-based solutions, offers a scalable and inclusive pathway for achieving integrated water security, sustainable food systems, and ecosystem resilience in least-developed and mountainous countries. The RUPEP Nepal experience provides policy-relevant lessons for UN Member States and development partners seeking to accelerate progress across interconnected SDG targets under the Water–Energy–Food–Nature nexus.

ABSTRACT. Realizing autonomous operation in urban water networks is often hindered by the computational burden of physical models and the limitations of reactive, rule-based logic. While model predictive control (MPC) offers a dynamic pathway toward intelligent management, its deployment is restricted by these computational barriers. This study introduces a tractable framework that integrates a deep learning surrogate into the MPC architecture, effectively enabling the real-time responsiveness required for autonomous control. A case study in the Baishi Chong catchment, Zhongshan City, China, demonstrates the efficacy of this framework. Results indicate that this autonomous approach significantly outperforms traditional rule-based control (RBC), particularly under frequent storm conditions. Beyond hydraulic safety, the system autonomously optimizes ecological benefits; at a critical monitoring point, it more than tripled the duration of ecological water level maintenance and eliminated alert periods prevalent under RBC. These findings suggest that integrating deep learning-driven MPC provides a robust foundation for autonomous water network management, significantly enhancing infrastructure performance without the need for substantial physical construction.

ABSTRACT. Climate change significantly impacts water security by altering key hydrological processes. This paper examines the intensification of the hydrological cycle under changing climatic conditions and provides an overview of major threats to water security. It highlights the role of modelling in quantifying climate-related impacts and reviews adaptation strategies to enhance resilience. Additionally, the paper explores the importance of effective water governance in managing these challenges. The aim is to advance understanding and support improved management of water security under climate change

ABSTRACT. Traditional stochastic models have long served as theoretical baselines for hydrological risk assessment but often struggle to resolve the non-stationarity of intensifying climate extremes. This study investigates the transition toward data-driven AI paradigms to address these limitations through an integrated spatiotemporal framework. In the temporal domain, we benchmarked emerging Time Series Foundation Models—specifically TimeGPT, TimesFM, Chronos, and Moirai—against conventional LSTM and SARIMA baselines using 40 years of hydrometeorological data from major cities in South Korea. Our analysis reveals that Foundation Models achieve zero-shot accuracy comparable to fully trained baselines while significantly reducing inference latency. Notably, the Moirai model demonstrated superior sensitivity to hydrological extremes, effectively mitigating the smoothing artifacts typical of recurrent architectures. Complementing this temporal analysis, we applied Generative Adversarial Networks (GANs) to downscale GPM IMERG satellite precipitation data for the complex terrain of Jeju Island. The GAN-based approach successfully reconstructed high-frequency orographic rainfall patterns that conventional CNN models failed to capture. These findings suggest that integrating Foundation Models and Generative AI provides a robust strategy for managing hydrological uncertainty, offering a scalable alternative to rigid probabilistic assumptions in an era of climate instability.

ABSTRACT. The impact by climate change is increasingly affects defense plans against floods, inland floods, landslides, storm surges, and high tides in Japan. The future plan was usually designed using past records on extreme precipitation and tide level. However, record-making precipitation and historical floods are increasing. Overtopping from river levee causes breaching in many locations even in the river reaches managed by Ministry of Land, Infrastructure, Transport and Tourism in Japan (MLIT). The plans are being revised considering future impacts of increase in rainfall and the rise of tide level using the target scenario of the Paris Agreement on Climate Change, global temperature rise Plan below 2°C. The MLIT issued the new strategy that the management of low and high waters which is now governed by river management section, agricultural, sanitary, forest management, etc. need to collaborate for increasing disaster resilience and sustainability, including urban design section for decreasing residents in flooding risk area, private companies and residents. The strategy is called ‘River Basin Disaster Resilience and Sustainability by All’. The policy takes comprehensive and multilayered actions, important three components for decreasing disaster as follows: 1) strengthening countermeasures for preventing floods, 2) reducing exposure on high flooding risk area, and 3) increasing resilience for floods. The present situation of the three strategies in Japan is reviewed and one of the adaptation methods is introduced for levee resilience on overflowing water and infiltration. Experiments were conducted based on the assumption that the installation of a vertical drainage layer, which can be constructed without placing a burden on existing embankments, would be an embankment improvement method that would prevent both overflow and seepage. The embedding drainage layer on the landside toe could significantly reduce the maximum scour depth compared to embankments without the drain at the toe. When comparing simple performance in terms of suppressing the expansion of scour hole due to overflow by the indices of reduction of scoured depth and length, the horizontal drainage layer from previous studies is more resistant to overtopping flow. However horizontal drain length is not enough in present design method. New idea is discussed based on the scoured phenomena by nappe flow from a levee.

ABSTRACT. The Han River, 1577 km long, is the longest tributary of the Yangtze, with an average annual runoff of 45.8×109 m3. The Danjiangkou Reservoir, located in the upper Han River, has a total storage capacity of 31.95×109 m3. The reservoir has comprehensive functions such as flood control, water supply, power generation, shipping, and ecology. It is a key project for the development and management of the Han River. Starting from January 2025, the inflow of the Han River was significantly lower than normal. The cumulative precipitation in upper Han River from April to July was only 300 mm, ranking the lowest since 1961. However, since August 25th, when the autumn rain season began, the rainfall in the upper Han River has sharply shifted from scarcity to excess. By mid-October, the cumulative precipitation in the upper Han River had reached 592 mm, which was 2.9 times the average for the same period over the years and ranked first in the same period since 1961. There has been a sharp shift in drought and flood. During the autumn flood season, the Han River experienced 7 consecutive numbered floods, making it the "most densely numbered flood" since the existence of measurements. The inflow of Danjiangkou Reservoir was 30.7×109 m3, ranking second since it was built in 1969. During the autumn flood season, a joint operation of the reservoir group in the upper Han River was carried out. A total of 17.4 ×109 m3 of floodwater was stored. Among them, the Danjiangkou Reservoir stored 12.7×109 m3 of floodwater, reducing the inflow peak of 19600 m3/s to 1450 m3/s for discharge, with a peak reduction rate of 93%. This lowered the flood peak water levels at the main control hydrological stations in the middle and lower Han River by 8.32-11.26 m. A total of 1.3×106 mu of farmland was spared from flooding, about 510,000 people were spared from relocation, and reducing economic losses by approximately 12.2×109 yuan. Meanwhile, in the late autumn flood season, flood control and water storage were coordinated, and the water storage process was continuously optimized. The Danjiangkou Reservoir achieved the goal of 170 m of full storage on October 18th, achieving a double victory in autumn flood defense and water storage at the end of the flood season. This presentation will introduce the operation of reservoir group under extreme autumn flood conditions in the Han River.

ABSTRACT. This paper examines flood resilience in New Zealand, emphasizing the influence of atmospheric rivers and cyclones in generating extreme rainfall and flooding. These large-scale weather systems transport significant moisture, often triggering severe storms that challenge existing infrastructure. Using a New Zealand case study, the paper presents a framework linking these atmospheric phenomena to natural hazards. It also explores how weather-driven extremes can be integrated into water infrastructure design to enhance resilience. Finally, the paper reviews resilience frameworks and indices for evaluation resilience in urban stormwater systems using New Zealand case studies.

ABSTRACT. Under the dual pressures of climate change and rapid urbanization, flood risks are becoming increasingly serious. Enhancing flood resilience of different subjects as well as systems is very important to adapt to climate change. This study assesses the housing flood resilience in the flood risk context. We consider two main drivers to establish a quantitative measure of flood resilience, i.e., the capability of individual housing to resist, and the capability to recover from flood damages. Accordingly, we develop a Housing Flood Resilience Index (H-FRI). The H-FRI comprises three parts: the flood hazard characteristics, the housing exposure and vulnerability, to represent flood resistance mapped over time; the ability for housing recovery from damages caused by inundation, considering the compensation; and the functional capacity, represented by the flood duration. This study takes Langouwa Flood Storage and Detention Area (FSDA) as the study area and sets five scenarios, including the actual operation of the “23·7” extreme flood in the Haihe River Basin, and four scenarios under different flood return periods, i.e., 20-year, 30-year, 50-year, and 100-year return periods. The results show that the larger the inundation area, the deeper the inundation depth, and the longer the duration of housing flooded, the smaller the compensation funds, the smaller the H-FRI, which means the lower the flood resilience of residential buildings. Based on the assessment results, measures for enhancing the housing flood resilience in FSDAs are proposed as follows. For instance, improving the submergence resistance capacity of existing housings to reduce the extent of flood-induced damage to the buildings during inundation, increasing the capacity of pumping stations for forced drainage to shorten the housing flooded time, and expanding the diversified approaches to securing flood compensation funds, e.g. flood insurance, so as to accelerate the post-disaster recovery and reconstruction process.

ABSTRACT. Coastal regions of the Asia-Pacific are increasingly exposed to disasters driven by tropical cyclones, storm surges, extreme rainfall, sea-level rise, and intensifying human interventions along the coast. The interaction of these drivers often results in compound flooding and cascading impacts on infrastructure, ecosystems, and coastal communities. Addressing such complex and interconnected risks requires a shift beyond conventional hazard-specific assessments toward integrated, predictive, and decision-oriented frameworks that can support both short-term disaster response and long-term resilience planning. This study explores how emerging technologies are advancing the prediction and mitigation of coastal disasters and supporting resilience building in data-scarce and hazard-prone coastal environments. Recent advances in Earth observation, including high-resolution optical and SAR satellite imagery, UAV-based coastal surveys, and nearshore bathymetric monitoring, have substantially improved the characterization of coastal morphology, inundation extent, and shoreline dynamics during extreme events. When integrated with in-situ observations and sensor-based monitoring networks for tides, waves, and rainfall, these datasets provide near-real-time inputs for dynamic coastal hazard assessment. Numerical modeling remains central to translating multi-source observations into physically consistent representations of coastal processes. Advances in high-resolution hydrodynamic and wave models, together with data assimilation and probabilistic frameworks, enable more reliable prediction of storm surge, coastal inundation, and compound flooding under extreme conditions. Data-driven methods, including machine learning as a subset of artificial intelligence, are increasingly used to complement process-based models by accelerating simulations, capturing non-linear system responses, and supporting rapid impact assessment. More advanced AI-based approaches, such as deep learning and hybrid physics–AI models, are emerging as powerful tools for emulating complex coastal dynamics, enabling rapid ensemble forecasting, and supporting real-time decision-making under computational and data constraints. Beyond hazard prediction, emerging technologies also play a critical role in mitigation and resilience planning. Digital twins, coastal vulnerability mapping, and integrated decision support systems facilitate the evaluation of adaptation strategies across multiple hazard and climate scenarios. Particular attention is given to the assessment of nature-based and hybrid solutions, such as mangroves and coastal wetlands, which provide effective risk reduction while sustaining critical ecosystem services. The discussion concludes by highlighting key challenges and opportunities in translating technological advances into operational tools that support early warning, adaptive planning, and long-term coastal resilience across the Asia-Pacific region.

ABSTRACT. Mediterranean territories are subject to increasingly unstable hydrological regimes, where long dry periods are regularly interrupted by short-duration, high-intensity rainfall events. In urban and peri-urban areas, the combination of impervious surfaces and steep slopes leads to faster runoff and strongly limits infiltration to the groundwater. Under these conditions, rainfall is often transferred out of the site within a short time frame, leaving little opportunity for local retention or reuse, even as demands on potable water resources continue to rise. The present study investigates opportunities for the local recovery, storage, and reuse of rainwater on the CNRS Géoazur laboratory site in Sophia Antipolis, southeastern France. The site, covering approximately 50,000 m², is characterized by mixed land use, including buildings, roadways, and vegetated areas, and extends across a sloping terrain draining toward a Mediterranean environment. A stormwater drainage network is in place and ensures runoff collection, although localized surface runoff events may still occur. The CNRS Géoazur pilot location​ was selected ​ to investigate the development of a vegetable garden operating under a net-zero water balance. The core objective is to determine how coupled hydrological and hydraulic modeling can be used to quantify mobilizable water volumes and to delineate areas suitable for temporary storage. The methodology is based on a ​numerical ​modeling approach combining detailed topographic analysis with surface runoff simulations. LiDAR data were used to generate digital terrain models (DTM) at 1m and 3m resolutions to identify flow paths and accumulation points. Land-use data were refined via GIS to quantify impervious surfaces and assign infiltration and roughness parameters. Runoff and drainage processes were then simulated using IBER+ modeling software under various rainfall scenarios. The results clearly demonstrate the dominant influence of both topography and soil sealing on the concentration and rapid routing of surface flows toward downstream outlets. Multiple locations with temporary storage potential were identified, together with substantial recoverable volumes originating from roof runoff and complementary sources such as air-conditioning condensates. Taken together, these results provide a quantitative basis for the design of storage solutions and illustrate the contribution of integrated runoff modeling to local water reuse strategies and climate change adaptation through reduced reliance on potable water.

ABSTRACT. Purpose of the work: Traditional gravity irrigation, recognized as part of France’s Intangible Cultural Heritage, serves as a vital hydraulic and social pillar in Mediterranean mountain regions. In the Alpes-Maritimes, these networks historically support local agriculture, promote groundwater recharge, and enhance landscape resilience against harsh climatic conditions. The objective of this study is to conduct an exhaustive inventory of these gravity networks to propose innovative, sustainable, and localized modernization methods that respect their heritage value. Key issues addressed: These ancestral systems currently face critical challenges, including the physical deterioration of structures, loss of traditional technical know-how, and increasing destruction caused by extreme weather events. Furthermore, their integration into modern water resource management tools remains insufficient, complicating real-time assessment and long-term planning. Methodology: Focusing on the Vésubie river basin, specifically the municipalities of Valdeblore and Roquebillière, the study employs a multidisciplinary approach. The methodology integrates Geographic Information System (GIS) tools and BD TOPO database analysis with field surveys and interviews with local stakeholders. Additionally, the project experiments with satellite remote sensing (Sentinel-2) to link canal functioning with agricultural and environmental dynamics, such as soil moisture retention and crop resilience. Conclusions: The study provides operational tools, including GIS-based channel profiles and technical indicators, to strengthen the resilience of traditional hydraulic systems. By combining hydrological analysis with an institutional framework study, the results offer concrete levers for local actors to adapt ancestral infrastructures to contemporary climate change. Ultimately, this transferable methodology contributes to the international debate on integrating traditional systems into regional water policies.

ABSTRACT. In a context of climate change and increasingly intense rainfall, sustainable stormwater management represents a major challenge for highly urbanized sites, particularly in Mediterranean environments. Soil de-impermeabilization and rainwater management at source appear to be essential levers for limiting runoff, reducing network overload, and improving the living environment. It is in this context that the Université Côte d'Azur has decided to de-impermeabilize the Trotabas law campus. The goal of this study is to propose a technically and financially viable solution for stormwater management for annual rainfall. The methodology is based on a comparison of two separate studies conducted by two consulting firms, as well as a third study that we are conducting ourselves, relying on numerical modelling approaches. The objective is to compare the two “standard” studies carried out with the study relying on hydrologic and hydraulic numerical modelling to determine the optimal de-impermeabilization solution best suited to the specific context of the site, which presents multiple constraints. The de-impermeabilization and disconnection areas are be analyzed in terms of topographical and geological criteria in order to assess their technical feasibility.. A multicriteria analysis l enables to select the most suitable areas in terms of slope, geology, and vehicle traffic for the implementation of de-impermeabilization or disconnection solutions. The study area has been divided into three zones: the library, the main building, and the parking lot. In order to determine the catchment areas for each zone and identify the associated active surface area, a simulation was carried out using IBER software, with artificial Manning coefficients to facilitate surface runoff. Knowing the active surface area and using Montana's formula, the volume of water to be absorbed by our installations is estimated. This appreachl enables us to design the associated de-impermeabilization structures. Using the IBER-SWMM coupled software, we then carried out two simulations to compare the volumes of water entering the pipes before and after the project. This comparison allows us to assess whether the proposed developments will enable annual rainfall to be managed in accordance with the specifications, and to quantify the volume of water that will not be sent into the stormwater network. Finally, once the design has been completed for annual rainfall, a cost analysis is carried out to verify that the project complies with the allocated budget of €800,000.

ABSTRACT. This project presents a hydrological and hydraulic study of the La Clapière landslide, located in the Upper Tinée Valley (Alpes-Maritimes, France). With an estimated volume of 50 million cubic meters, this site is among the largest mass movements in Europe. It is situated within fractured metamorphic gneiss, where fissure networks act as perched reservoirs. The slope's hydrological behavior is driven by an Alpine pluvio-nival climate, which also governs the discharge of the Tinée River flowing at the landslide toe. This immediate proximity between the unstable slope and valley floor heightens associated risks.

This study aims to improve the characterization of the landslide's hydrological functioning and to evaluate the hydraulic impacts of a failure scenario. The hydrological component adopts a threefold approach: (1) establishing a water balance for the instrumented spring at the slope toe; (2) developing a multiple linear regression model to relate spring discharge to atmospheric forcing; and (3) comparing electrical resistivity tomography (ERT) profiles between dry and wet periods to visualize subsurface water storage dynamics. The monitored spring exhibits stable discharge with low sensitivity to short-term meteorological variations, while ERT data reveal both persistent saturated zones and seasonal deep recharge. However, the landslide hosts multiple springs with potentially contrasting behaviors, limiting the generalization of these findings to the entire hydrological system.

In parallel, a numerical hydraulic model was developed using Iber to address the hydraulic risk. The landslide toe already encroaches upon the Tinée riverbed, creating a constant threat of obstruction. Simulations evaluated two critical scenarios: first, the formation of a natural dam causing upstream flooding toward Saint-Étienne-de-Tinée village, and second, the sudden breach releasing a downstream dam-break wave toward Isola village. This tool enables mapping of water depths and hazard zones to anticipate consequences of further slope destabilization.

The study concludes with a discussion of the results and a synthesis of the inherent limitations of this work. These stem first from the available data, whose spatio-temporal resolution limits the precise quantification of internal fluxes. They also relate to the difficulty of faithfully representing the heterogeneity of a dual-porosity massif, where slow matrix flow and rapid fracture flow coexist. Furthermore, the hydraulic modeling results remain contingent upon the hypothesized breach geometry, which introduce uncertainties in flood extent predictions. Despite these uncertainties, this work enhances the understanding of La Clapière's hydrological functioning and assesses the major hydraulic risks that its interaction with the Tinée River poses to the valley.

ABSTRACT. The European project EduWaterLab introduces an innovative pedagogical approach to teaching hydrology and hydraulics at the university level. Coordinated by UPC (Spain) in partnership with BTU (Germany), WUT (Poland), and UNICAL (Italy), the project shifts the focus from tool-oriented learning to a "back-to-basics" understanding of physical and environmental processes. By utilizing a web-based platform and shared laboratory resources, the consortium allows students to access specialized experiments regardless of their local institutional capacity. This paper presents one experiment of the project: the characterization of vertical infiltration in porous bituminous mixtures (Porous Asphalt - PA). Urban flooding remains a critical challenge in densified cities, and Sustainable Drainage Systems (SuDS) offer a resilient solution by reducing surface runoff. The experiment uses a sample of PA11 mixture from a pilot project in Barcelona (Jordi Girona St.) to demonstrate fundamental hydraulic principles. The methodology involves a falling-head laboratory test using a transparent graduated cylinder and video-tracking techniques to monitor the descent of the water column. Results indicate infiltration rates between 7,800 and 11,200 mm/h, which show high correlation with in-situ measurements performed using the LCS Permeameter (11,096 mm/h). Beyond the technical validation of the PA11 mixture, this communication highlights the experiment’s role as a "Shared Teaching Unit." Through the EduWaterLab platform, students across Europe can engage with the experimental videos, raw data, and protocols, fostering a collaborative learning environment that prepares the next generation of engineers to tackle the complexities of the modern water cycle.

ABSTRACT. Collective gravity-fed irrigation networks are vital to socio-hydraulic systems in Mediterranean mountain regions, ensuring agricultural productivity, equitable water distribution, and ecosystem services. In the Alpes-Maritimes territory (southeastern France), these traditional systems recently recognized as intangible cultural heritage face growing threats from aging infrastructure, limited monitoring, regulatory challenges, and increasing hydrological extremes linked to climate change. Notably, water intakes, as critical nodes, must ensure reliable diversion during low-flow periods and withstand high energy floods. This study establishes a baseline assessment of collective irrigation networks and develops pilot sites for technical, digital, and sustainable modernization. Research focuses on the Vésubie watershed, a steep catchment with rapid hydrological responses, and high exposure to both floods and droughts.

An integrated methodology combines field surveys, in situ monitoring, and numerical modeling. Catchment-scale hydrological processes are simulated with the semi-distributed rainfall–runoff model HEC-HMS to reproduce low-flow and flood conditions. Hydraulic performance of water intakes is analyzed using the two-dimensional model IBER, enabling detailed simulations of water depths, velocities, and forces on structures. In parallel, field-deployable sensors measuring water levels, precipitation, and soil moisture are deployed along canals to support model calibration and provide insights into flow dynamics and water losses assessment.

Preliminary results highlight the high sensitivity of existing intakes to extreme discharges and reveal structural vulnerabilities during floods. The combined use of HEC-HMS and IBER provides a robust framework for testing rehabilitation scenarios and supporting decision-making. Beyond technical findings, co-construction with local water user associations ensures that modernization strategies respect traditional practices and local constraints.

ABSTRACT. This study develops a 2D hydraulic modeling framework for the lower Var Valley (French Riviera) designed to update the existing hydraulic model in the AquaVar decision support system. The aim is to test and benchmark different numerical codes and assess possibilities offered by different functionalities embedded in theses codes (meshing, use of GPU, including hydraulic structures or bed evolution implementation).The following computational codes: Mike 21 FM, IBER+, and TELEMAC-2D are compared, each using distinct numerical methods to solve the shallow water equations: finite volume for Mike 21 FM and IBER+, and finite element for TELEMAC-2D. IBER+ flexibility with structured meshes enables a comparison with unstructured mesh approaches from the other two computational codes. Preliminary steady-flow simulations revealed differences in volume conservation errors and computational times. First, realistic initial conditions and consistent boundary constraints were necessary to compare numerical solving performance under steady state conditions. Then, flood events with 1- and 2-year return period are to be modeled with each code before being evaluated against a downstream flood hydrograph with an NSE index. Codes will also be compared based on computing time, mass conservation and observed numerical instabilities. Key challenges include ensuring consistency between topographic data and modeled events as well as guaranteeing the accuracy of downstream rating curves and upstream hydrometric data. An update of the current model would also integrate sediment transport modeling, leveraging IBER+ and Mike 21 FM capabilities to assess morphological changes of the riverbed during flood events.

ABSTRACT. Urban surface sealing currently covers 8 % of the French territory and increases runoff, particularly in Mediterranean regions exposed to heavy rainfall. De-impermeabilizing parking areas using Sustainable Urban Drainage Systems (SUDS) such as permeable pavements or porous concrete provides a potential solution, reconciling urban uses with sustainable stormwater management by alleviating the regular stormwater system. Yet, assessment of (i) the effectiveness of the disconnection, (i) potential risks of infiltration toward ground-water resource, or (iii) methods to monitor and numerically model theses SUDS performance are aspects where best practices are not consensual. This study evaluates (i) to (iii), focusing on the hydraulic performance of a de-impermeabilization system installed on the STAPS campus of Côte d’Azur University parking lot. The site, highly urbanized, exposed to flood risk, and located 8m (in average) over an exploited alluvial aquifer benefit from highly permeable alluvial soils that promote infiltration. A pilot installation of three parking spaces paved with Hydromedia® permeable concrete was instrumented to monitor the system’s behavior under real- scale conditions using multi-parameter instrumentation. The methodology combines in situ monitoring with multi-scale numerical modeling. Vertical infiltration and water transfer through the pavement structure were simulated using HYDRUS-1D. Surface runoff was modeled with the IBER hydraulic model to identify preferential flow paths and accumulation zones. The expected results will validate the effectiveness and performance of the Hydromedia® permeable pavement and help determine suitable locations for stormwater harvesting and storage for the irrigation of future green spaces. This project thus provides decision-support elements for optimizing local stormwater management and promoting de-impermeabilization solutions adapted to Mediterranean urban environments.