The present study proposes a method to maximize the stiffness of a structure considering the uniform distribution of reaction forces. The variance of the reaction forces at the Dirichlet boundary is introduced to evaluate the uniformity of the reaction forces. Several formulations for optimal topology design are provided by introducing the constraints of variance of reaction forces. Multi-objective optimization considering mean compliance minimization and uniform distribution of force transmission is proposed by dimensionless quantity description. Sensitivity analysis is then derived based on the adjoint variable method and verified by the finite differentiation approximation(FDA). Several numerical examples verify the effectiveness of the proposed method.

 Unconfined compression test is generally used in the field of geotechnical engineering to evaluate the strength and deformation characteristics of cement-mixed soil in addition to rock and soil. However, there are several influential factors on the deformation and strength characteristics of cement-mixed soil, such as the specimen size and the spatial variations of cement content and water content caused by mixing quality, curing time, etc. This study develops random finite element method to evaluate the unconfined compressive strength of cylindrical specimens, compares the results of two-dimensional analysis and three-dimensional analysis, and discusses the effects of the specimen size, strength parameters, and spatial variability parameters on the unconfined compressive strength. It can be characterized that the unconfined compressive strength decreases with increasing spatial variability, which is more pronounced for larger specimens. To improve computational efficiency, a prediction model for unconfined compressive strength is constructed successfully by combining random finite element method with Convolutional Neural Network.

 The rockfall trajectory is affected by a number of parameters such as the material properties of rock and slope, slope geometry, rock initial condition, etc. The rock's initial condition such as initial angle and rock shape are unknown parameters and spatially variable parameters. This study investigates the influence of rock shape and initial angle variation on the rockfall trajectory. Under a slope model with the height of H = 15 m and an angle of 50°, the rockfall trajectory of nine rock shapes reflecting the flatness and elongation is released under different initial angles under Monte Carlo Simulation (MCS) using the Discrete Element Method (DEM). The 100 MCSs can ensure the error of reach probability within 0.001 considering rock shape and initial angle variation. The simulation results indicate that cuboidal compact rock shape tends to have concentrated and longitudinal trajectory distributions, whereas elongated or flattened shapes result in a wider and shorter trajectory distribution. Additionally, through mode analysis, a main region below the rock source is identified. This main region, with a width of 1.3 𝐻 and a length of 1.9 𝐻, captures 98% of the trajectory distribution, irrespective of variations in flatness and elongation of rock shapes.

 This study explores the optimal sensor placement for slope failure real-time monitoring. The sensor placement is optimized based on mode analysis using Singular Value Decomposition (SVD) and QR decomposition with column pivoting (QR-CP) for a simple model slope with single soil layer. The optimal sensor placement is discussed in terms of the required modes for reconstruction, the model reconstruction error, and the positions of optimal sensor in slope with different types of soil under different rainfall intensities. Results show that variations in rainfall patterns and slope soil properties, such as the soil-water characteristic curve, lead to differences in the information captured by the main modes from the mode analysis. When evaluated by cumulative contribution, the difference can be as large as 24.1% in this study. Furthermore, the results indicate that increasing rainfall intensity causes the optimal sensor locations to shift from the slope surface to deeper positions within the slope. Slopes composed of different soil types exhibit distinct failure modes under rainfall, such as loamy sand slope, which can be highly permeated by rainfall and maintain pore water pressure of approximately 60 kPa above the groundwater table, leading to different distributions of optimal sensor locations.

 Pluvial flooding, a common and destructive event, occurs when natural or engineered drainage systems are overwhelmed by intense or prolonged rainfall. A significant factor contributing to inadequate drainage capacity is sediment accumulation in minor drainage systems, especially open channel drains in lowland areas. Conventional studies have mainly focused on experimental evaluations of drainage capacity loss due to sediment. This study employs the InfoWorks ICM model to innovatively assess the impact of sediment depth on pluvial flood severity in lowland drainage systems. Our findings provide a groundbreaking quantification and visualization of how sediment affects flood inundation, revealing the combined influence of increased rainfall and sediment depth on flood duration and extent.

 The Weakly Compressible Smoothed Particle Hydrodynamics (WCSPH) method which was previously applied to rainfall-runoff simulation on natural topography has been reformulated to enable simulation of the motion of discontinuous patches of surface water due to a rainfall. The reformulationn indicates the basic equations of the original WCSPH may be used but suggest more logical modellong of the dynamics as the interaction with neighbouring fluid, the ground solid bodies. Unlike depth-averaged flow analyses the water depth is not the main variable in the governing equation but can be obtained by a separate post processing that considers the relation of the real water depth and the distribution of lumps of water for which the governning equations are applied. The simulation of overland flow on an urban terrain with model buildings are correctly simulated and the method is found to represent three-dimensional flow correctly.

 Andean glaciers have been rapidly retreating, yet the hydrological connections between glaciers and glacial lakes remain inadequately understood. This study aims to elucidate these connections in the Cordillera Real, Bolivia, using Landsat-5 and Landsat-8 imagery from 1984 to 2021. Glacier and lake areas were mapped, and the hydrological linkages between these features were evaluated through catchment-scale analysis. Our results show that lakes in close proximity to glaciers exhibit a higher correlation with glacier retreat, indicating a direct impact of glacier melt on these water bodies. In contrast, lakes farther from glaciers were more influenced by non-glacial factors. The spatiotemporal analysis revealed significant warming, especially at higher elevations, and increased evaporation rates, with higher rates on the eastern side. Although overall precipitation increased, it was insufficient to sustain downstream lakes. This research highlights the impending reduction in water resources due to global warming, which is expected to lead to higher evaporation rates and increased loss of glacier meltwater during transport. The accelerated melting of glaciers, coupled with changes in precipitation patterns, threatens the stability of water supplies in the region. Our research underscores the urgent need for comprehensive monitoring and sustainable water management strategies to mitigate the adverse effects of climate change on glacial and hydrological systems in the Andean region.

 Reservoir sedimentation is a critical issue that impacts dam operations by reducing storage capacity and increasing management costs. This study evaluated the effectiveness of memetic programming (MP) in predicting the suspended sediment concentration (SSC) in the Miwa Reservoir, Japan. The Miwa Dam faces challenges due to its high sediment yield and rapid discharge, necessitating accurate SSC predictions for efficient sediment management and dam operation. Hourly SSC and inflow discharge data for two periods were collected ((A) June 29-August 7, 2020, and (B) June 1-July 5, 2023). Three input scenarios were examined to predict SSC. The scenarios incorporating current and previous values of the inflow rate and SSC exhibited the highest accuracy, with correlation coefficients (R) ranging from 0.88 to 0.99 and normalized Nash-Sutcliffe coefficient (NNSC) ranging from 0.77 to 0.98. The MP model successfully captured SSC dynamics during flood events, demonstrating its potential as a valuable tool for reservoir management. Accurate real-time predictions facilitate better operational decisions, reduce sediment-related issues, and optimize reservoir functionality. This research highlights the potential of advanced machine learning techniques in sediment management, offering significant insights for reservoir sedimentation management. Future work should explore integrating additional influencing parameters, applying this approach to other reservoirs with similar challenges, and exploring the future prediction of SSC shortly for several hours in advance.

 This study investigated the potential to use soil texture information for promoting the grid-based regionalization of a distributed rainfall-runoff model (1K-DHM) for flash-floods predictions, which aims at assigning representative model parameter sets to 30-second grid cells in ungauged basins. A previous study explored the parameterization based on land-use and soil types data, generating model parameter sets on soil-type classification (PS-STC). This study reclassified soil types with soil texture information, creating three new soil types based on soil grain composition: coarse-grain soil, fine-grain soil and gravelgrain soil. Representative model parameter sets of soil types for the grain size composition (PS-GSC) were identified through calibration in 27 donor catchments and a cross-verification within each donor group. Besides the final optimized parameter set, five additional parameter sets generated during the calibration processes were also selected and included in the cross-verification. The newly identified model parameter sets (PS-GSC) demonstrate different water flow characteristics in soils in terms of the depthdischarge relationship embedded in 1K-DHM. In the validation, the simulation results using the PS-GSC were compared with the ones using the PS-STC at 52 heterogenous catchments with 541 rainfall events, evaluated by the Nash-Sutcliffe efficiency (NSE) and peak discharge signatures. Results show that PS-GSC demonstrate a comparable performance to PS-STC in terms of NSE, while PS-GSC shows stronger performance in terms of peak discharge estimation and peak discharge time delay. These analyses reconfirm that the grid-based regionalization considering geospatial types can be used to achieve discharge forecast in ungauged basins and indicate that soil texture information has potential to enhance these forecasts by yielding more reliable peak discharges.

 Flood modeling is essential for effective flood risk management strategies, but the reliability of datasets has long been a significant obstacle to the accuracy of flood models, especially in ungauged or poorly gauged river basins. Advances in machine learning technology have provided solutions to many of these problems, particularly through the use of machine learning algorithms that significantly enhance data accuracy and reliability. This study aims to project flood conditions under climate and land use changes by leveraging machine learning-filtered data. The Global Land Cover and Land Use (GLCLU) maps and the Forest And Building Removed Copernicus Digital Elevation Model (FABDEM) were utilized to increase the accuracy of future flood projection. The future climate condition was applied by employing six global climate models (GCMs) under the Couple Model Intercomparison Project 6 (CMIP6). The study was conducted in Upper Citarum River Basin (UCRB) and analyzed a 2-years return period of rainfall event to represent the yearly flooding. The results show that the change in land cover and climate could increase the runoff volume by 2-10% until 2100. It could extend the affected area around 16-46%. This finding emphasizes the impact of land use and climate changes on flood conditions in UCRB. This study successfully projected future flood risk by leveraging machine learning filtered data. It increases the opportunity of future flood risk projection for poorly gauged river basins.

 In rapidly urbanizing regions, the proliferation of plastic debris in sedimentary environments poses multifaceted challenges, particularly in vulnerable urban slum areas where the deplorable state of waste management has led to considerable littering and dumping. In association with the general absence of adequate drainage systems, and appalling slum conditions, the presence, transportation, and accumulation of sediment plastic debris (SPD) significantly contribute to flood hazards and impacts in these settlements. This study investigated SPD production, transport mechanisms, drivers, and potential effects on floods in the urban slums of Yaounde, Cameroon. Through field survey, statistical analysis of field data, geolocalization and quantification of SPDs, and modeling of the flood event of February 26th, 2018, we present the distribution of “sediment plastic debris” at four locations within the flood plains of the Mfoundi River in Yaounde and outlines diverse ways through which this contributes to increased flood impacts. The findings showed tremendous open disposal of wastes, often occupying areas along major roads, drainage areas, and rivers. Geolocalized distribution of the SPD identified 333 major points located at distances ranging from the river course to 1 kilometer and beyond. Modeling of flood events revealed the potential transport of all floodplains’ SPDs, with considerable SPDs of 686 and 822 cubic meters determined at selected outlets. This research highlights the crucial role of sediment discharge in sediment transport and calls for integrated waste management strategies, urban planning, and community-based interventions to mitigate its impact on flood resilience.

 This study reports the influence of road and embankment network on the flood hydrodynamics of river floodplains in the Haor basin, Bangladesh. The topological setting of this part of the country is unique, low-elevation floodplain (about 20,000 sq km) of a complex river system surrounded by the Meghalaya Mountain range from north and east. During the monsoon season, this floodplain stays inundated for 6-7 months and is exposed to flash floods almost every year during pre-monsoon and monsoon seasons. In the last decade, the intensity of flood level and duration has increased. Along with climate change, rapid urbanization, and land use change, the construction of high-elevation roads in the Haor floodplain is being criticized for degrading the flood situation. In this investigation, a hydrodynamic analysis has been conducted using a HEC-RAS 2D model for a selected river system of this region by reproducing the historical flood events. The model has been calibrated and validated for existing condition against the observed water level with NSE values of 0.974 and 0.943, respectively. A separate set of simulations has been conducted with natural topography, excluding the road and embankment network from model topology. Where the reduction in peak flood level is more than 25 cm for 38% and 42% area in the floodplain for the monsoon flood of 2021 and 2022 and about 30% for the 2022 pre-monsoon flood. However, after a week from the peak period, the difference in water level between existing and natural conditions increased, which indicates a shortening in the flood duration. It has been found that 34 km out of 58 km high elevation road of the model domain obstructs significant amount of floodplain flow, suggesting that more cross-drainage structures in those road alignment will improve the flood condition.

 Present study analyzed damages resulting from the 2014 flood as well as from the potential floods with the magnitudes of 10 years to 200 years return period. In particular, focusing on the 2014 flood, data necessary for formulating damage functions were collected by means of field surveys in addition to numerical simulation. Such surveys realized that the channel erosion caused loss of areas and agriculture areas covered by deposited sediment required some years to recover the normal situation before the event. Based on these results, the damage functions are formulated for houses and community areas, and for agricultural areas. Then, applying the formulated damage functions to the potential flood hazards which were obtained from the numerical simulations on different flood events with the magnitudes ranging 10 years to 200 years return periods, we evaluated the potential damages resulting from such floods events for houses, community areas and agricultural areas and found that the damage caused by land loss due to erosion was largest and the damage resulting from sediment deposition was second largest in all land uses.

 In Artificial Neural Network (ANN) models for rainfall prediction, the model performance relies heavily on the input data characteristics. Input variables of the model, such as meteorological parameters and their spatiotemporal variations, should be carefully selected. Due to the ability of ANN to detect non-linear and complicated patterns behind the data, a new input variable selection method called Zero Input Test (ZIT) is proposed in this study. A comparative study between ZIT and correlation coefficient (CC) is carried out considering the global data of atmospheric variables from lags of 1 to 12 months. Selections from the two approaches are evaluated by building two ANN models based on CC and ZIT selections to predict monthly rainfall of Upper Chao Phraya River Basin (UCPRB), Thailand. The selection results show that CC and ZIT identified different regions to be important variables. However, the rainfall prediction results of both models show similar accuracy and prediction trends. This suggests that regions outside those selected by CC also contribute to UCPRB rainfall, and there are many regions across time variation that influence the phenomenon. Additionally, the current inconsistency in ZIT selections indicates that there might be room for improvement of the method to further develop the input variable selection.

 The study was focused on developing a conceptualization for the preliminary prediction of gate operations (GOs) using the forecasted water levels (WLs) of upstream and downstream gauges of the underground regulating reservoir (URR) in the Kanda River Basin. The rainfall and WL minute-by-minute data of Zenpukuji and Upper Kanda watersheds were gathered from the last 13 years up to 2024. The deep learning model was built by considering preceding flood events for the input time steps, while the targeted time steps were organized with the upcoming WLs of seven gauges. The precision of WL forecasting was evaluated by adjusting the target lengths (TLs) from 10 minutes (TL10) to 90 minutes (TL90) with an increment of 10 minutes. The prediction accuracy worsened when moving from TL10 to TL90, and even the Nash–Sutcliffe model efficiency coefficient (NSE) remained at or above 0.96 until TL60, with root mean square (RMSE) consistently below 0.1 m. The GO influenced the accuracy of the longer TLs compared to shorter TLs, where almost all WL gauges achieved above 0.9 NSE for TLs from TL30 to TL60. The temporal exactness and peak alignments of forecasted hydrographs were examined to derive the longest TL for anticipating the initiation (Iɢᴏ) and termination (Tɢᴏ) of GOs in advance. The longest TL obtained Iɢᴏ and Tɢᴏ in advance was 40 minutes, while 60 minutes was the longest TL identified only the Iɢᴏ in advance.

 Information of impact force is expected to assist in trauma severity evaluation for snow avalanche accidents. The study employed a coupled human-snow avalanche model to simulate the impact force on a movable human body in several snow avalanche conditions. The simulation results with the movable body had a smaller impact force than the fixed one, which might be due to a part of the impact energy from the snow avalanche transferring to the movement of the human body. The body motion differed in various flow conditions. It was categorized by flow depth into push, crash, and slip types, corresponding to deep, intermediate, and shallow depths. Their energy ratios were correlated with the Froude number, showing the linear relation in the push type, the independent relation in the crash type, and the exponential relation in the slip type. The empirical equation of the energy ratio for each motion type was proposed finally and expected to help assess the injury and vulnerability of snow avalanches for people in the backcountry.

 A regular and periodic three-dimensional bedform, called brick-pattern ripples, consisting of crests perpendicular to the flow direction and longitudinal bridges between their crests, appears under an oscillatory water flow in a certain condition. We made the characterization of brick-pattern ripples formed in an oscillatory water flow through systematic experimental results using a U-shaped oscillatory flow water tank. In addition, geometric properties and formation conditions of such ripples were examined on the basis of several dimensionless parameters. The formation diagram of brick-pattern ripples was experimentally found by using the parameters in comparisons with previous experimental and theoretical results.

 Heavy rainfall in arid regions has been increasing recently and can have devastating effects, especially in light of the climate change impacts. Therefore, predicting flash flood-prone areas is essential for proactive disaster management. This study aims to develop hazard maps over Saudi Arabia (KSA) based on historical precipitation dataset and projected future precipitation data of D4PDF. The spatial and temporal extreme rainfall patterns historically and future changes based on D4PDF climatic data were developed by investigating the present rainfall trends for the period of 2000–2011 and future rainfall trends for the period of 2031–2100. Frequency of the rainfall events exceeding different threshold (30, 40, 50mm) was analyzed. The trend analysis was conducted for the extreme events historically and in the future scenarios. The results shows that more extreme rainfall events are expected to occur in the central, eastern, and western regions of Saudi Arabia. In general, high variability trends have been observed in these regions from historical to future trends analysis. The results of hazard maps developed by using RRI model with maximum rainfall events in both history and future scenarios for three regions over KSA (Madinah, Central regions, and Jazan), showed that both Madinah and Central regions are no significant change from historical to future, however, Jazan show an extreme increasing in the hazard impacts more than 100%. The outcomes of this study can be used by decision maker in Saudi Arabia to implement mitigation measures for flash flood risk reduction.

 Present study examines responses of the longitudinal bed profiles to bed shear stress changes and associated changes of the bed load transport rates which are caused in straight open channels with different flow widths where the channel is connected by an abruptly widened approach or a narrowed one. To investigate the response of bed slope ratio to the change of bed-load transport rate, we employ two bed load formulas for discussions; one is the bed- load formula characterized by τ_∗^3/2 with τ_∗c, and the other is the bed load formula expressed by τ_∗^5/2. Besides, to see role of the bed load formula explicitly in the regime relation, the regime formulas for describing the bed slope ratio, specific bed elevation etc. were obtained, assuming that an equilibrium sediment transportation accomplished. The regime formula shows that the bed slope ratio, i₂/i₁ increases monotonically with the increase of the flow width ratio when τ_∗^5/2 type of the bed load formula is employed in the regime equation, where suffixes 1 and 2 show the quantities upstream and downstream. However, the regime formula shows that the bed slope ratio depends on the bed shear stress change when τ_∗^3/2 type bed load formula is employed. The present study topics are classic, and not new, but it is very important to study the response of the longitudinal bed profile to the bed load transport rate to look for a suitable form of the bed load equation.

 The present study discusses sediment transport processes in the drainage basin of the Kelani River using the RSR model which is able to evaluate bedload and suspended load and their sediment size distribution at any cross-sections of stream channels. This model was applied to the 2016 and 2018 flood events to investigate temporal and spatial characteristics of sediment transportation. Simulated results show that the accumulated bedload volumes are comparable to the accumulated suspended load in these flood events. Besides, they show that the material of the suspended and bedload tend to mix in the peak flood stage, and the field data on the upstream and downstream sediment size distribution are similar to those of bedload and suspended sediment, respectively.

 This study investigates the impact of urbanization on climate in the Yangtze River Delta urban agglomeration from 1984 to 2022. Using high-resolution Landsat-derived land cover data and the Weather Research and Forecasting (WRF) model capable of considering distributed urban parameters and anthropogenic heating, we simulated climate with various landcover conditions for 1984, 2000, 2010, and 2022. Simulations of a June 2022 heatwave month were conducted while varying the urban extent according to the actual historical landcover. In the core areas, or areas that have been urban since 1984, the monthly near-surface air temperature (wind speeds) continually increased (decreased) with increasing urban cover. During the heatwave, urban cover change of 1984 to 2022 increased the core’s spatial average temperature by 0.52°C with windspeed decreasd by 0.13m/s. However, urban expansion areas farther away from the core tend to have lower air temperatures than the expansion areas closer to the core. The mechanism behind the heatwave-dependence on the urban agglomeration changes were explained using landscape metrics.

 In this study, a semi-implicit model is proposed for the bed friction and drag force terms of the shallow water equations (SWEs) by calculating bottom pressures to simulate dam-break flows with vegetation. The Manning formula is used for the bed friction term, while the drag force term is evaluated for non-equilibrium conditions considering the base component of drag force expression, water surface variations, and pressure gradient. Our model employs a cell-centered finite volume method with second-order spatial accuracy, while the Runge-Kutta second-order (RKSO) explicit time-stepping scheme is used for temporal discretization. Both the bed friction and drag force terms are treated in a semi-implicit manner, namely, such terms are included in each computing stage of the RKSO scheme with an implicitness factor to calculate depth-averaged (horizontal) velocities. Meanwhile, vertical velocities at water surface and bottom are calculated along with bottom pressures iteratively using the conjugate gradient method but only once for each time step. Henceforth, the bottom pressure values are used to update the horizontal velocities at the same time step. Our numerical model is validated against the experimental data and shows very good agreements in predicting the flow characteristics both inside and outside the vegetation zone.

 The Toyama coastal regions have experienced a significant impact from severe erosion. This degradation can be attributed to a combination of factors such as insufficient quantity of sediment supply from the Kurobe dam to the coastal area. In this study, we used Digital Shoreline Analysis System (DSAS) to identify historical shoreline changes with notable change rates. We calculated shoreline change rates from 1985-2024 by leveraging historical shoreline positions derived from google earth images. Shoreline recession was observed on many beaches, especially Nyuzen and Asahi beaches experienced significant erosion when compared to other beaches. However, from 2014 to 2024 shorelines have been regenerating by a gradual increase in sediment deposition along all beaches except Iwasehama beach. Furthermore, we employed Bruun Rule method to project future shoreline retreat projection based on Toyama beach characteristics, sediment size, historical wave characteristics and Regional Mean Sea Level Rise (RMSLR) projections. This study identified that under different scenarios of sea level rise, certain beaches in Toyama Prefecture could experience significant loss rates. Iwasehama beach could lose 14% of its beach under SSP1-2.6 scenario, while near future the Amaharashi beach in 2070 is 93% under SSP5-8.5, and for Himi beach in 2100 it is 46% under SSP2-4.5. The results show the beaches located at Himi and Amaharashi beaches are at risk of disappearing sooner than other beaches. These findings show that coastal areas are always changing, and it is important to keep monitoring them and adjusting management strategies accordingly. This research findings will help stakeholders identify the most effective measures to combat shoreline retreat caused by sea level rise.

 This study investigates the characteristics of suspended sediment transportation and associated morphological changes in the Meghna Estuary in Bangladesh, specifically focusing on sediment transport dynamics in the lower Meghna River and the sediment budget in the Noakhali islands area, which is located in the northeastern part of the Meghna Estuary. A depth-averaged 2-D model, spanning 350 km in length including the entire lower Meghna River and surrounding estuary area, is prepared for flow and sediment calculations, employing a uniform sediment size of 0.06 mm. The results of an entire year calculation, considering the wide seasonal variation, indicate that the Noakhali islands area experiences significant morphological changes during the ebb tide period of the spring tide time, particularly in the monsoon season. During this time, a strong current diverges from the main flow of the lower Meghna River, carrying a significant amount of suspended sediment, which leads to high sediment concentration particularly in the slack time, resulting in significant storage and deposition of fine sediment in this area. Therefore, this present study proposes a model which can evaluate the seasonal, daily, and hourly variation of sediment storage and deposition processes in an estuary dominated by fine sediment.

 A deep learning model using a convolutional neural network (CNN) and U-net is built to study the applicability in detecting surface weather fronts in Japan and surrounding sea. First, a CNN model is used to predict whether there are fronts (1) or not (0) in the region. If CNN outputs 1, frontal locations can be predicted by assembling binary classification of front/no front at each grid point within the region using a U-net. The predictability of CNN/U-net stems from their ability to match locations of outstanding horizontal gradients with observed frontal locations. Our study shows that CNN can achieve higher accuracy in filtering out no front cases by using fewer predictors than the U-net. Overall, the predictability of frontal locations varies with season and differs when stationary fronts are present or not, which suggests that dynamics related to seasonal frontogenesis can influence the spatial distribution of predictability of fronts.

 Accurate hydrological forecasting is essential for effective water resource management and flood prevention. This study investigates the impact of input variable selection and data quality on the performance of Artificial Neural Networks (ANNs) in predicting river discharge across three river basins: Hiyoshi Dam, Katsura, and Fukakusa. Utilizing a dataset of 50,000 training data, our findings reveal that, once we collect a relevant data pool, the choice of input variable selection methods—whether including all rainfall data, the nearest rainfall points, or variables with the highest correlation coefficients—has minimal effect on prediction accuracy when the ANN is adequately trained. However, with a limited training dataset of 5,000 data points, prediction accuracy becomes more sensitive to input selection, underscoring the importance of carefully choosing relevant variables. The study also demonstrates that the ANN model is resilient to noisy training data. Further analysis shows that the ANN can effectively disregard irrelevant data, maintaining robust performance. These insights provide valuable guidance for optimizing input variable selection and ensuring data quality in ANN-based hydrological forecasting, enhancing the reliability and effectiveness of predictive models in water resource management.

 As extreme rainfall and flood events occur more frequently in the past years, the need for predicting detailed flood extent in inundation areas has been increasing. A machine learning model was proposed in this study aiming for the rapid spatial explicit flood prediction of inundation areas. The training process was performed with simulation results from a well-established hydrodynamic model, and investigations of incorporating predictors of water depths from past time steps were conducted. Evaluation of the model was first performed by numerical simulations and a case study was carried out. The model successfully predicted the water depth of the region with an overall coefficient of determination (R²) values greater than 0.9. While the accuracy varies with respect to land use, the model was able to represent the spatial differences in hydraulic properties within the area.