The utilization of satellite microwave (MW) observation is essential to understand the long-term and extensive cloud water content (CWC) distribution. Over highly heterogenous land, understanding of radiative transfer characteristics between land and atmosphere including clouds is important, but those are not elucidated. In this study, based on radiative transfer theory which covers land and atmosphere, we first clarify the important variables for radiative transfer between land and atmosphere under existence of clouds. Then, as base information for estimation of CWC over land using MW, we utilized in-situ observation and simulation to analyze the land emission, atmospheric downward radiation, and land emissivity. As a result, contrary to the commonly held image of being very complicated, we revealed that with downward radiation from thick clouds, especially exceeding 2kg/m2, the heterogeneity of emissivity can be reasonably negrected, because downward radiation becomes close to the physical temperature of land surface.

 The methodology for evaluating the amount and the permittivity of bound soil water was examined through comparing several dielectric mixing models for moist soils. The possibility and methodology for introducing pedotransfer function as well as water retention function into dielectric mixing models were then examined for effective representation of different soil-water characteristics over the world.

 A new wind velocimetry was proposed using clouds images based on stereo vision theory (Cloud Image Velocimetry : CIV). Cloud patterns taken by multiple ordinary digital cameras placed at different locations are processed. Three dimensional coordinates of cloud patterns in each image pair are to be reconstructed with stereo vision theory, and wind speed is calculated from their temporal displacement. CIV yielded wind velocity at cloud-bottom altitude. Wind direction by CIV corresponded with that estimated Himawari-8 satellite images. Wind speed by CIV was different from that obtained by radiosonde released Tateno, but this difference possibly reflecting the difference of observation time and location between CIV and the radiosonde.

 The purpose of this study was to identify the uncertainty of the probable maximum precipitation (PMP) estimation method using surface dew point data under the pseudo-adiabatic assumption. This PMP estimation method was proposed by the World Meteorological Organization (WMO) and widely utilized for many practical purposes, and precipitable water (PW) is the most important parameter to estimate PMP. We evaluated the pseudo-adiabatic assumption by utilizing the reanalysis data, JRA-55, to estimate the uncertainty in the PW estimation using the pseudo-adiabatic assumption. The JRA-55 data was verified with radiosonde data at 10 points across Japan, and it shows good reliability for the dew point and PW data. Our PW verification results show that the use of the pseudo-adiabatic assumption is suitable for those areas with frequent heavy rainfall events and where the dew point is relatively high. However, the PW estimated from the assumption shows noticeable differences from the PW measured from JRA-55 data in those areas where heavy rainfalls are not frequent, and thus the estimated PMP could have large uncertainty in those areas.

 In this study, we used a radiation model called the solar and longwave environmental irradiance geometory(SOLWEIG) to reproduce the thermal environment within an urban district, and compared with the data from mobile observation in terms of globe-temperature (Tg) which is calculated from Mean Radiation Temperature (MRT) and Sky View Factor (SVF) in the model. We confirmed poor reproducibility of Tg in spite of accurate calculation of SVF in the model, if a constant air temperature and global radiation are used for entire domain. Instead, we replaced these data with those obtained from observation. It resulted the model value almost caught the variation of observational result for Tg.

 Spatial variability of thermal comfort for pedestrians in a low-rise residential district was evaluated based on a mobile observation. The observation was conducted in August within a residential district near a tidal flad in Narashino, Chiba. The observation was taken place totally for 12 times among 2 days to obtain their ensemble characteristics to extract the local effect.

 Spatial variation of globe temperature was observed to be more than 3.5 degree, which is attributed to the effects from wet area, variability of vegetation distribution and geometry of the district. Wet flow from the wet area penetrated around 100m inland in case the major streets are perpendicular to the flow although it exceeded 300m in case the major streets are parallel to the flow.

 The urban heat island (UHI) is caused by the progress of urbanization, but is considered to be influenced by various factors that differ depending on the cities, such as the geography and the background climate. However, the comprehensive mechanism is not clear yet. In this study, we tried to clarify the influence of not only urbanization, but geography and background climate by comparing the heat island phenomena of multiple cities in the world using numerical simulation. The wind speed is the main contributor to the urban heat island intensity; the heat of the city is transferred more effectively to the surrounding area as the wind speed is stronger. The wind speed is considered to be a comprehensive indicator reflecting the effect of geography around the city, and the roughness of the buildings.

 Flood fighting activities are one of the most important measure to mitigate flood damages in cooperation with flood protection measures. Although the activities have to be conducted effectively with considering appropriate human resource allocation, however, its quantitative evaluation has not been conducted because of the strong territorial characteristics so on. This study aims to develop a quantitative evaluation method for flood fighting activities. We developed a method which considered a needed work volume to protect levee, needed amount of activities to conduct flood fighting, and flood forecasting lead time. In addition, we applied the method to scenario simulation in the Kiso River basin, Japan.

 The population trends of the administerative districts including frequently flooded areas (FFA) in Japan was analyzed and compared between the districts with different geographical classification. As a result, we found that different trend could be seen between different geography. For example, districts with small rivers and wide plain tended to have high population density and less future population decrease. These differences were considered to reflect the social positioning of each FFA.

 To prepare for a large-scale disaster, self-help and mutual help are absolutely imperative. Promoting them further, it is important to raise consciousness of disaster prevention of not only those who have high consciousness but those who don’t. In this study, we randomly chose inhabitants from Basic Resident Register in Hamamatsu City, Shizuoka Prefecture, and asked them to participate the disaster prevention workshop ”Hamamatsu City Disaster Prevention Resident Council”. We analyzed how the random choosing contributes to those who have poor consciousness of disaster prevention to promote participation in the workshop.

 The purpose of this study is to obtain knowledge of relationships between landuse change and flood risk to prepared for the increase of flood disasters caused by climate change in the Kurokawa basin, Aso. We show the change in the number of houses from 1902 to 2003, and predict the number of inundated houses and the inundation depth induced by the 2012 northern Kyushu heavy rain its 1.4 times heavy rain. As a result, the number of inundated houses increase due to the increase of houses especially in the bottom plain and the lower terraces. In the case of the 1.4 times heavy rain, it is predicted that the inundation depth exceeding about 3.0 m would occur in the valley bottom plain and the lower terrace surface, which is not induced by the 2012 northern Kyushu heavy rain. It is predicted that exposure to these floods can be reduced or avoided in the valley bottom plain where the distance from the river is more than about 600 m, and in the natural embankments where the relative height is higher than about 1.8 m.

 The upstream-downstream balance of population and assets is widely used to understand flood risk in a river basin and minimize it. Although conventional flood control strategies have been developed based on the past balances, now it is required to consider the future balances to handle with flood risk because population declining must cause changes in population distribution patterns. This paper aims to estimate current and future flood risk (flood exposure) and clarify basin characteristics focusing on upstream-downstream balances. The result shows that metropolitan regions have a significant amount of flood exposure in the downstream and balance shifting will hardly occur there. In addition, basins with larger flood exposure in the up-midstream areas show greater balance shift towards downstream.

 In July 2018, Hiji river basin experienced a severe flood. Right after this flood, residents in the Hiji River basin and the government disputed over the operation rule of the Kanogawa dam and Nomura dam in the basin. This research aims to quantify the proper operation rules of the Kanogawa dam. Dam operation rules are quantitatively evaluated based on expected annual flood damage, which is calculated based on previous flood damages and the extreme value distribution of inflow into the dam. First, the previous and current rules are modelled, and the expected annual flood damage after the dam operation are calculated. In this simulation, the old rule proves to reduce the expected annual damage compared to the new rule. Second, the optimized dam operation rule, which can minimize the expected annual damage, is determined by changing the parameters of the dam operation. It is proven that the rule which aims to reduce the damage of extremely large floods is more optimal. Third, the expected annual damage of flexible operations, in which the operation rule is switched according to the predicted magnitude of inflow, is examined. The results shows that the expected annual damage can be decreased by switching rules, however, the results are almost the same as the expected annual damage of the rule which is optimal exclusively for large floods.

 As natural disaster hazard intensifies drastically, attributed to the effect of global warming, the capacity to prevent disaster in Japan has been weakened due to degrading infrastructure and an aging population. In order to reduce the damage caused by such disasters, the flood control adaptation method in the Kase River by using existing dams, Hokuzan Dam and Kasegawa Dam, were discussed. In this study, future climate conditions in the Kase River basin were set on the basis of d4PDF (: database for Policy Decision making for Future climate change).

 Computational results demonstrated that the storage capacity of Kasegawa Dam was enough large in order to control the flow discharge hydrograph due to the future precipitation. Therefore, if the existing dams can be fully utilized by carrying out prior discharge, the flood control will be possible in the Kase River basin even under the future extreme precipitation.

 In this paper, we report the result of comparing the water level prediction of the dam using several machine learning methods.Water level predictions using machine learning were calculated for Kanayama Dam and for Satsunaigawa Dam. The predictions were calculated from hydrological information for the basins of these two dams. The machine learning methods used for the predictions are the Random Forest, Fully Connected Neural Network (FCNN), Recurrent Neural Network methods, which has a structure for processing time series information, and the Elastic Net method, which is based on sparse modeling. The comparison found that FCNN and Elastic Net yield accurate results, with NS coefficients of 0.7 or greater. In Elastic Net, for cases other than those whose predicted rainfall has indeterminacy, the results were the most accurate, having NS coefficients of 0.7 or greater.

 It is required to evaluate climate change impacts on flood control and water utilization functions of dams after releasing ‘MLIT vision for dam upgrading under operation’. Furthermore, in the evaluation of the long-term dam function, the decrease of effective storage capacity by sedimentation progress is also an important factor. Up to now, however, there is no suitable evaluation methods to integrate these factors. In this research, we firstly predicted changes in flow regime curves and reservoir sedimentation under climate change impact. Secondly, several criteria have been proposed to select high priority existing dams for upgrading. Thirdly, we evaluated the appropriateness of the criteria for water utilization functions by examination at individual dams. Based on these evaluation criteria, we found that it is necessary to mitigate climate change impacts in order to maintain necessary functions at almost all dams.

 Heavy rainfall caused severe flood disasters across the western Japan in July 2018. In order to prevent frequent inundation in the downstream reaches where river improvement works have not been completed, a reservoir is often operated to deal with small or medium scale of floods by regulating its release discharge to a smaller values than that originally designed. This operation, however, can increase the risk of severe flood inundation in case of large floods, because water stored in the reservoir increases faster to the full storage volume, thus the flood control function is eliminated. This study aims at analyzing impacts of flood control operation rule for various scales of floods including large scale flood events using rainfall runoff inundation analysis in order to identify effective operation policy for flood risk reduction.

 The occurrence of earthquake and flood compound disaster in Japan was investigated by collecting the data of earthquake intensity and levee breach due to flood. The frequency of earthquake occurrence in 2000s was almost consistent with that in 2010s, while the frequency of levee breach due to flood in 2010s was greater than that in 2000s. In addition, the frequency which the yearly-maximum water level exceeded the design high-watar level has been increased since 2000s. Consequently, the time-interval between earthquake occurrence and levee breach in 2010s was shorter than that in 2000s, indicating that the possibility of occurrence of earthquake-flood compound disaster may increase.

 This study makes the simulation of a compound disaster involving flood and storm surge and evaluates the damage cost for the disaster in Japan. The tide level and daily rainfall, which cause the compound disasters, are calculated by means of frequency analysis for annual minimum atmospheric pressure. 2D non-uniform flow model expressing the inundation depth is carried out using the tide level data on coastline and daily rainfall distribution interpolated as input data. Damage cost is estimated using the inundation depth by each land use. The annual expected damage cost of flood in whole Japan is 4.04 trillion JPY, that of storm surge is 4.04 trillion JPY, and that of compound disaster is 3.89 trillion JPY. In 31 prefectures out of 46 prefectures except for Okinawa prefecture of Japan, the damage cost of only flood is larger than that of only storm surge and the compound disasters.

 In order to ensure safe evacuation of construction workers, heavy industrial machinery and machine parts at a river construction site during a flood, it is necessary to take measures several tens of hours before the flood.The person in charge of construction site refers to the results of multiple water level predictions and own experiences and decides the availability and timing of flood countermeasures. However, there are problems, 1) it takes time to prepare data such as the water level required when constructing a forecasting method, 2) the water level is different for each prediction method and the construction worker hesitates to make decisions.This article discusses the development of a neural network flood prediction technique which uses rain cloud images, which are easy to acquire. The technique was applied to Abukuma River and compared predicted values and measured values.Rain cloud images were divided to consider rain distribution with a small amount of data, and center of gravity and rainfall were used as learning data. As a result, the prediction accuracy rate increased by up to 60%.This technique can predict long-term flood by using weather forecast and can be applied in safety management of river construction.

 Railway operators enforce train operation control based on actual precipitation observed by rain gauges to ensure safe train operation in the times of heavy rainfall. Now by utilizing rainfall forecasting information they may make train operation safer. In this paper, we evaluate accuracy of predicting the time of exceeding the reference values of train operation control when using forecasting rainfall values by the translation model developed by Kyoto university and the high-resolution precipitation nowcast distributed by Japan Meteorological Agency. We evaluate the accuracy using the capturing ratio that indicate the proportion of cases where regulation is issued by forecasted values among cases where regulation is issued by actual values and the predictive ratio that indicate the proportion of cases where regulation is issued by actual values among cases where regulation is issued by forecasted values. As a result, it is found that, both of the capturing ratio and the predictive ratio of the translation model show higher values than those of the high-resolution precipitation nowcast.

 Rapid Development Cumulus Area (RDCA) indicates the developing process of cumulus clouds that are potentially expected to evolve into thunderstorm within one hour in around 10 km2 area. To express cloud developing process, Rapid Scan Observation (RSO) of Himawari-8 data is used to generate RDCA index ranging from 0.1 to 0.9 by adapting logistic regression model. As the RDCA index represents cloud development without information of updraft, we try to prove RDCA index can reflect updraft information by comparing RDCA index time series with parameter of differential reflectivity (Z_DR) and vertical wind velocity estimation obtained by multi Doppler analysis. As a result, based on three cases, we found all of them have a good temporal correlation between RDCA index time series and Z_DR, and only one case has a good temporal correlation between RDCA index and vertical wind velocity estimation.

 We used a large scale ensemble climate prediction database d4PDF composed of numerous ensemble experiment data. By analyzing d4PDF, we attempted to verify the probabilistic evaluation of future prediction and the change of extreme weather which is low-frequency and local-scale events. We analyzed the data for 2,500 years of past experimental data and the data for 5,400 years of the 4°C rise experiment data and the data for 3,240 years of the 2°C rise experiment data in Kyushu Island. We confirmed that the 1 hour precipitation has a model bias. For the probability evaluation of torrential rain in entire Kyushu, we used the general polar of distribution. The conclusion is the intensity and frequency of torrential rain will both increase. In addition, about the change in torrential rain, there are regional differences, the increase in precipitation is relatively small in the northern Kyushu, but it is large in the southern Kyushu.

 We have developed a data-driven Artificial Neural Network (ANN) model, capable of predicting water levels only with observed data sets, to support the drainage operations at the pump stations in low landareas. We tested two ANN models: a conventional model (Multiple Layer Perceptron, MLP) and an updated model (Long Short-Term Memory, LSTM), which effectively works on time-series predictions, in two lowland areas where different drainage systems exist. As to the simple darainage system in a small area, the LSTM is superior to the MLP with approximately 10% improvement of water level predictions within 2 hours. In addition, the LSTM predicted upto 3-hour water level, approximately 6% better than the MLP during the heavry rainfall event even in a larger, complex drainage system.

 This research is a development of our previous work, which was also based on artificial neural network usage for prediction of water levels at Hirakata station. This time we included rainfall information along with upstream water level data. The developed model uses a single hidden-layered feed-forward neural network. Prediction accuracy increased significantly compared to the previous research using a similar model. Regarding the whole validation period of three years, even though the model followed a certain pattern of decreasing performance with increasing lead time and increasing length of input data, the results still outperformed predictions obtained without using rainfall information. Some particular periods with peak events from the validation period were also checked. While all the results exhibited better outcomes compared to the previous model, this model did not follow a specific pattern corresponding with input data selection in any of these check periods.

 Flood is one of the major natural disasters causing serious damage around the world. Though many flood prediction systems have been developed recently, most of them do not consider reservoir operation in realtime flood forecast and alerting. This study aims to improve the accuracy of numerical flood prediction by implementing reservoir operation for flood control into our flood forecast system. The operation of each reservoir were formulated in the model based on the publicly-opened flood control procedures and reservoir parameters. The developed system was tested for the flash flood in Tone River basin in 2015, and we conducted the simulation using observed precipitation and the hindcast experiment using ensemble weather prediction. The observed-precipitation simulation confirmed that implementing reservoir operation improved the model performance, and the hindcast results show that the forecast skill of peak discharge was improved.

 We attempt to investigate the efficiency of applying a lithology factor with high-resolution XRAIN observation to accurate landslide hazard estimation. This study presents a model of landslide mapping using logistic regression with geological and high-resolution hydrometeorological factors, and analyzes hazardous conditions of landslide disasters occurred in Kure, Hiroshima during the heavy rainfall event in July of 2018. Being identical to the practical method of landslide early warning, the hydrometeorological factors are hourly cumulative rainfall and soil-water index. The lithology factor is derived from the seamless geological map. As a first trial, the model was simply calibrated using linear logistic regression on a recent landslide inventory composing of 646 events in Chugoku Region after 2012. 85% and 15% of events are used for training and accuracy test, and the calibrated model achieves a high accuracy of 91.8%. To verify, our model was applied to estimate landslide occurrence during the heavy rainfall in Kure, Hiroshima. The result verified our model can estimate highly accurate occurrence location.

 This study investigates flood detectability using global inundation maps derived from satellite images and floodplain mask. Annual total cumulative inundation extent (ATCIE: the accumulative total spatial extent of the flooded area) was derived from satellite-based daily surface water change based on 500 m Moderate Resolution Imaging Spectroradiometer (MODIS) for 2001–2015 overlayed onto newly developed floodplain mask from a global high-resolution Multi-Error-Removed Improved-Terrain (MERIT) digital elevation model. Flood detectability based on the ATCIE was then tested for 16 globally distributed historical flood events. Results indicated that standardized anomaly of ATCIE can successfully detect most of the anomalous inundation extent in historical extreme flood events. However, relatively small floods (return period < 100 year) were undetectable by ATCIE. Flood detectability of ATCIE has a correlation with the magnitude of the flood rather than basin size. Dense vegetation cover (> 40% in the basin), complexity and the intricate river basins (due to altering the flood signal by tributary rivers), or effect of cloud cover are additional potential sources of the undetectability.

 Woody debis sometimes has negative effect to our lives. For example, woody debris can damage the construction, and it can stop the gate of dams to move. Furthermore, it is said that the amount of woody debris will increase in the future because these days forest in Japan is getting dead and guerilla heavy rain happens increasingly. Theofore, it is important to understand the mechanism of woody debris and know the characterristics of wood accumulation and export in catchment scale. Komori et al. constructed the wood export model assuming that there were 2 types of wood export, flood flow-type and base flow-type. They assumed that flood flow-type export could happen as woods were recruited when it rained heavily, and base flow-type export could happen under the nomal condition. And they adapted the model to the dam catchments which belonged to Kitakami-river, in Iwate prefecture.

 The objectives of this study were : (1) to adapt the model to Terauchi dam catchment, where large amout of wood recruited because of the heavy rain in Northen Kyusyu in 2017 and understand the mechanism of wood export at that time ; (2) to compare the result in Terauchi dam catchment to that in the dam catchments belonging to Kitakami-river. As a result, calculated amount of wood export was almost accurate, so it meant that the characteristics of the wood accumulation and export in Terauchi dam catchment could be explained by flood flow-type and base flow-type. Moreover, it was revealed that the heavy rain event in 2012 might affect the large wood export in 2017. And also it seems that in Terauchi dam catchment, accumulated wood can be easily transported without heavy rain event, comparing with the catchments which belong to Kitakami-river.

 In this paper, we attempted to improve the evaluation method of the potential driftwoods generation by Yano et al. (2018) in the Shira-kawa River, in which a lot of volcanic soils are included. From the results by the modified evaluation method, it is clarified that i) the potential driftwoods generation can increase with a rise of 1, 3, and 6 hour precipitation; ii) the Kumamoto Earthquake in 2016 affected driftwoods generation, and more than 70% of the total driftwoods was generated from slopes which can collapse by only earthquake; and iii) the Tateno Dam, which is a stream type flood control dam, can trap more than 85% of total driftwoods generated in the overall Shira-kawa River basin.

 In Hiroshima Prefeture, many houses exist in alluvial fan area near mountain side, and many of them are designated as debris flow dangerous area. In the mountainous housing area, evacuation routs are limited so when heavy rainfall occurs, it might be difficult to move outside of the designated area. Therefore, it is important to understand the dangerous risk distribution inside the designated area to consider safe evacuation planning. In this study, we conducted debris flow simulations and focued on alluvial fan housing area’ s dangerous risk distribution. In simulations, sediment volume, discharge, and lanform conditions influence the debris flow flooding and deposition. And when debris flow occurs, many houses are destroyed and landform seems to change. Therefore, we applied DEM and DSM data to consider both conditions with houses and without houses. We considered a number of scenarios for sediment volume and discharges considering from Sabo Master plan’ s method, and overlay the scenarios results and consider the disaster risk’s distribution inside the designated area.

 In railways, train operation control is implemented to ensure safety during rainfall. However, collapse caused by rainfall may occur after train operation restrictions are lifted. In considering disaster prevention measures for railways, it is important to clarify the causes and rainfall conditions of such delayed collapse. So in this research, rainfall conditions related to delayed collapse are considered, targeting five delayed-collapse incidents that occurred on railways. As a result of examination, incidents of collapse were classified into three types: concentration of water infiltration in surface soil, concentration of spring water, and rise of groundwater pressure in foundation ground. It was then shown that the storage height calculated from the storage function method and the standard tank model method was valid as a rainfall index in delayed collapse, considering the behavior of rainwater in the soil for each form.

 Real-time observation data of the river water level includes various anomalies. Such anomalies may cause fatal errors in judgments on disaster prevention activity and flood forecasting systems, but real-time anomaly detection has not been sufficiently implemented. In this study, we developed the model to detect anomalies in real-time for river water level data sent from observation stations every 10 minutes. By using machine learning, the water level at the current time of the objective observation station was estimated from the neighboring water level and rainfall. Then the anomaly score was calculated from the degree of deviation between the estimated water level and actual observation. Furthermore, the model was combined with the rule-based anomaly detection model. The proposed method was verified using actual observation data, and better performance was confirmed compared to the existing method.

 Water drained through sewer network installed in lower-lands shows complex behavior due to influence from pumping station and exhibits non-linearity in rainfall-runoff relation. Also validation of sewer network model is not sufficient due to lack of observed data. Present study validates urban inundation models by using water levels observed in storage pipes and applies databank-based data assimilation. Furthermore, effect of pumping operation is examined through sensitivity analysis. It was found that model prediction can be improved by the assimilation technique and also it was found that consideration of water levels in pumping stations are necessary.

 Real-time flood forecasting requires a hydraulic model that understands river channel characteristics and a hydrological model that ensures a high level of accuracy. In this study, we proposed a particle filter method capable of simultaneously estimating river-bed evolution at the water level gauging station and slope water depth, which is the initial conditions for the RRI model. The target river is the Seri River in Shiga Prefecture, Japan, in which sediment deposition is a concern. We conducted experiments using meteorological and hydrological data collected when Typhoon No. 18 hit the country in September 2013. As a result, we confirmed that the calculated water level at the present time can be assimilated to the observed water level, and that the estimated changes in river-bed evolution can be explained from a sediment-transport theory and river characteristics. In conclusion, the practicality and validity of the method we proposed in this study were verified.

 An inland flooding could be occerred frequently due to localized torrential rainfall enhanced by climate change. However, quantitative monitoring measurement has not yet been well-established for evaluating the damages and drainage management of inland flooding, as well as its threshold for evacuation from such complicated and qualitative information. In this study, for the purpose of grasping the risk of inland flooding, the reformed DIEX-Flood model, which interpolate and extrapolate water-level data into a streamwise direction with satisfying hydraulic principles, can calculate accurately with high-speed calibration performence. The improved model was verified by using the water level observation data of sewer pipe. The results indicate the proposed method can forcast streamwise distribution of water-level with high accuracy by using data assimilation. In data assimilation, the correction of forcasting is made with less influence from the rainfall prediction errors and other potential errors.

 To reduce the computational load and improve the robustness of a flood prediction system (DIEX-Flood), which can estimate the water level longitudinal distribution in the current and future, we corrected the equation of DIEX-Flood and improved the data assimikation algorism. As a result of applying this method to Kinu river flood, it can calculate multiple cases of floods. Furthermore, we proposed a new flood forecasting method combining DIEX-Flood and deep learning. This prediction method can forecast water level longitudinal distribution in the future, so it makes possible to know spatial and temporal distribution of flood risk. This method can be expected accuracy improvement by assimilating water level data such as crisis management type water level gauges.

 The rainfall spatial variability has not been considered in the Storage Function (SF) models so far even though there exist various SF models including the Urban SF (USF) model, a relatively new SF model mainly for urban watersheds. Therefore, in this study, we aim to propose a generalized USF (GUSF) model for the storm runoff analysis by considering the spatial rainfall distribution in the basin. This was achieved by the introduction of a new parameter named as rainfall distribution factor (𝛾) in the USF model. The GUSF and USF models were examined and the results revealed that the GUSF model with 𝛾 exhibited higher hydrograph reproducibility associated with the lowest error evaluation criteria which emphasize the effect of parameter 𝛾. Further, the Akaike information criterion (AIC) was used to establish the best model among two based on the number of optimized model parameters. The GUSF model received the lowest AIC score in calibration and validation which indicate that the inclusion of a single parameter, rainfall distribution factor, can substantially improve the performance of a model by representing the spatial rainfall distribution of basin in a better way.

 In recent years, urban floods have frequently occurred, and improving the accuracy of urban runoff prediction is an urgent issue. Urban runoff process is complicated, and it is difficult to construct a runoff model with high accuracy. The machine learning model can adjust the model parameters automatically if there are input data and output data, so it is possible to construct the model easily. So, in this paper, we aim to evaluate how much the urban runoff model can be emulated by the machine learning model, and the virtual rainfall and the virtual runoff （with the known true value already published by the authors）obtained from it. The artificial neural network model and deep learning model were constructed for quantity, and the reproducibility in learning flood and verification flood was compared and verified. We also evaluated the emulation performance when changing hyper parameters.

 The model to reproduce the runoff into low-lying lake is developed by using deep learning method. The model was composed of 4-layer feed forward network and as training algorithm Adam was applied. The target area is Toyanogata lake which is located at Niigata prefecture. Input data were hourly precipitation data and hourly data of quantity of drainage water from Toyanogata lake, and output data was hourly runoff to the lake. Mainly the examination was conducted about input data; how long previous data of precipitation and quantity of drainage water should be input to improve accuracy of the model. And the input length which stabilized accuracy of the model was found in both of condition that precipitation data was only input and that precipitation data and drainage data were input. In addition, comparison was made about relationship between construction of the model and accuracy model between 3-layer ANN model and deep learning model.

 A real-time rainfall-runoff prediction system was developed using the storage function method incorporating with a particle filter. The performance of the prediction system was examined at the upper part of the Yattajima River basin (5150 km²) in terms of several ways to incorporate observation data, state variables to give system noise (storage variables of basin, storage variables of river, and both of them), and time intervals for data assimilation. Our findings show that to assimilate state variables for each sub-basin provides better performance than to assimilate all state variables at the same time; to give system noise to catchment and river storage variables shows better performance for longer prediction time; and setting assimilation time intervals 10 minutes is better than 1 hour.

 In this study, polynomial chaos expansion (PCE) was utilized to choose the optimal model parameters of a distributed hydrological model (DHM) for the Ibo River watershed in Hyogo, Japan. Because of many model parameters, only parameters with high sensitivity in the simulation results were estimated by the PCE emulator. PCE provided straightforward orthogonal polynomials that effectively captured the behavior of the DHM with greatly reduced computational complexity. The parameter estimation with PCE reduced the misfit between observed and simulated discharges at three observation stations by 25.2%–39.8%. Based on the good agreement in this case study, PCE is recommended for application in other hydrological models.

 Most of hydrological models are developed for simulating river discharge in a temperate area with assumption of rainfall runoff process such as quick subsurface flow in a shallow soil layer in a temperate zone. However, there is not sufficient research on long-term rainfall runoff process considering a deep soil layer in a humid tropical area. This research clarifies the reasonable model structure for a humid tropical area and potential use for analyzing long-term rainfall runoff process and inundation process with limited available data in Sumatra island. Reflecting vertical infiltration process into deep soil layer and ground water to RRI model, it performs better for reproducing river discharge compared with the subsurface flow and following parameter setting for tropical region. The model shows flood-prone zone with more than 50 cm inundation depth at the downstream, which can occur at 80 % probability in one year.

 A river bankfull depth, one of the important topographic parameters in a river routing model, was estimated efficiently by using the CaMa-Flood model and a satellite altimetry dataset. We applied the developed method to the Amazon river basin, and, by an OSSE experiment, confirmed that river bankfull depth would be improved even if the uncertainty of runoff existed. In addition, by using a real satellite altimetry dataset, we estimated the river bankfull depth and confirmed this can make the inundation ratio in the basin estimated from the model closer to that from a SAR observation. Next challenges are to validate the method where cross-section data are available and to apply the method on a global scale.

 Saturated lateral flow on sub-grid scale is pointed out to have impact on global estimation of evapotranspiration and soil moisture. However, few global terrestrial models have represented saturated lateral flow explicitly.

 Therefore, this study adapted an upscaling method to MATSIRO model, one of the land surface models, in order to develop global terrestrial model which deals with saturated lateral flow within a grid explicitly. 50-year simulation with saturated lateral flow demonstrated less soil moisture in most areas than the simulated results by original MATSIRO, which could be attributed to soil moisture’s concentration in lower hill slope in each grid which leads to increase evaporation and runoff. In addtion, the results suggested that considering sub-grid distribution of vegetation is also important to evaluate the impact of saturated lateral flow on global hydrological estimation, since the difference of water-energy balance between a valley and a peak occurs when vegetations are concentrated at the valley.

 This research investigated runoff characteristics generated by two LSMs: MRI-SiB, which was embedded in a regional climate model NHRCM, and SiBUC. The generated runoff were given as input to a flow routing model 1K-FRM for simulating the river discharge to analyze their impacts. The simulation was applied in the upper part of Ping River basin in Thailand. First, the runoff characteristics from both models were analyzed. Then, river discharge simulation was conducted by utilizing the runoff from both LSMs. The simulated river discharge from both LSMs were analyzed in terms of volume of discharge and timing of peak discharge. Some possible causes of different runoff estimation by both LSMs were investigated by analyzing the effect of soil parameter settings, water budget, and model structures.

 The Amu Darya delta is an irrigation intensive region located in an arid drought-prone part of the Aral Sea basin. Due to irrigation which is characterized by high water application rates, the groundwater level is high and has resulted in secondary salinization. This coupled with the frequent droughts causes the irrigation area to constantly change for farmers to adapt to cropping environmental problems and the changing annual climate. However, obtaining an accurate report of the actual distribution of the irrigated area has proven to be difficult even to the local government and local agricultural institutes. This study aims to assess the potential of using Land Surface Temperature (LST) from MODIS and a Land Surface Model (LSM) to detect annual changes in irrigated area. 3 indices were developed using LST by MODIS and LSM based on the concept of heat capacity difference between water and soil. The LSM provides LST for ideal conditions while MODIS provides the actual LST. A combination of the two enables the elimination of external influence on LST such as rainfall and geological variations which may impact on the LST. A distributed map of all the 3 indices shows the potential of LST in detecting drought. The irrigation fraction during a drought year was observed to be lower as compared to that of a normal year. This was true especially further away from the water source due to water scarcity. In addition, a comparison of the sum of all the meshes in the study area for each of the 3 indices with the volume of water released from the Tuyamuyun reservoir shows a similar trend. Tuyamuyun is indicative of water availability in this drought-prone region, therefore, the indices developed can be used to indicate irrigation activity here.

 After 2000, the northeastern Brazil has been suffering by serious droughts. Particularly, the reservoirs reduced to only 6 % of the total water storage in Ceará State caused by the severe drought in several years after 2012. In this study, the long-term agricultural drought (2003-2017) was simulated by the coupled land and vegetation data assimilation system (CLVDAS) over the northeastern Brazil. Simultaneously, we could recognize that the annual crop production and the irrigation water volume, which is necessary to obtain a target yield, are estimated by using LAI’s deviation from CLVDAS.

 Now that the numerous ensemble members datasets such as d4PDF are available, we can take nonparametric approaches to clear the statistical properties of even the extremes. It will be no longer a suffering issue from the extrapolation. However it doesn't mean that the extreme value theories are not needed. Simple counting the number of occurrences is insufficient but some average process will work for more reliable estimation. This paper shows the theoretical requisites for nonparametric statistics of joint extremes, and it demonstrates to quantify the simultaneous occurrence of extreme precipitations become less dependent as the locations are distant.

 This study aimed to verify the stationarity of historical and simulated precipitation data from the Database for Policy Decision-Making for Future Climate Change (d4PDF) by performing the nonparametric trend test. The test was conducted on summer precipitation data from seven locations in Japan. As a result of separately testing each ensemble member of d4PDF, it was found that the percentage with an increasing trend was significantly higher than the percentage with a decreasing trend. Results of testing all ensemble member together confirmed that the increasing trend was significant and thus also verified the nonstationarity of precipitation. After removing the trend from the combined d4PDF data, stationarity of the data was confirmed. 100-year precipitation data were calculated via the nonparametric method. As a result of comparing the observational data from the seven locations and the 100-year precipitation, we determined that 1-hour precipitation of d4PDF was underestimated.