Regarding the adaptation strategies for climate change, non-structural measures take priority and revision of flood control plan carried out in foreign countries is under consideration in Japan.In this study, we confirmed the validity of estimating the flood volume after climate change progressing by using dynamical downscaling data of database for Policy Decision making for Future climate change in the Class A river located in Hokkaido.In addition, we suggested the method for making flood control plan appropriately by analyzing damage reduction using different types of countermeasures and evaluating their effects from an economic standpoint including excess floods caused by climate change.

 In order to evaluate risks of river disasters due to global warming, rainfall data of super-high resolution numerical simulation estimated by a regional climate model were given to the Rainfall-Runoff-Inundation (RRI) model simulations in Kinu and Kokai river basins. We evaluated the impact of climate change on the risks of river floods with the three viewpoints of water level, discharge, and inundation area. As a result, the excess frequency for the lower-risk water level significantly increased, whereas the increase for the higher-risk water level was unclear. Water discharge amount mostly increased, whereas drought-flow amount did not significantly change. Furthermore, the frequency with which innundation areas expand remarkably increased, suggesting an increased risk of floods. However, the 31-year-experiments in this study involve large uncertainties with the narrow target areas such as the Kinu and Kokai River basins. It is necessary to further increase the number of cases in the future.

 Climate change may causes not only increase of plan period- and basin-averaged rainfall but also change of spatiotemporal characteristics of heavy rainfall. Because the change of spatiotemporal characteristics of heavy rainfall may increase flood discharge and change flood damage pattern, we need to predict its characteristics in order to make efficient flood damage mitigation plan. In this study, we analyzed spatiotemporal characteristics of heavy rainfall over Tokachi river basin and Tokoro river basin in Hokkaido, Japan under historical and warmer climate from a large-ensemble dataset. We also evaluated influence of tropical cyclone which is difficult to analyze because of shortage of observation record. The results showed that spatiotemporal characteristics of annual maximum heavy rainfall under warmer climate is more concentrated and influence of tropical cyclone on two river basins has different tendency in terms of spatiotemporal concentration.

 We investigated the future change of Baiu heavy rainfall duration and its accumulated precipitation amount. We analyzed the Baiu heavy rainfall events represented in 5-km-mesh Non-Hydrostatic Regional Climate Model (RCM05) and the past real heavy rainfall events. As a result, the increasing trend of the accumulated precipitation amount per heavy rainfall duration can be found in the future. In addition, we verified the quantitation of RCM05 by comparing the past real heavy rainfall duration and its accumulated precipitation amount with these of RCM05’s present climate. Moreover, it also indicated that the Northern Kyusyu heavy rainfall in 2017 was a comparatively extreme event in terms of the heavy rainfall duration and its accumulated precipitation amount even in both of the past and the present of RCM05.

 In this study, we analyze the mechanism of the future change of a occurrence frequency of Guerrilla-heavy rainfall in the Kinki region in August by focusing on the future change of temperature lapse rate and water vapor inflow using a 5km-mesh regional climate model (RCM05).

 From the analysis, we show the frequency of days will increase when Showalter Stability Index (SSI), which expresses atmospheric stability, becomes lower in the Kinki region in late August although temperature lapse rate will decrease. Lower SSI means that atmosphere is unstable. Then we show that the reason of destabilization of SSI is the increase of water vapor in lower layer. Finally, we use Self-Organizing Map (SOM), which is one cluster classification method, to reveal the main reason of the increase of water vapor in the lower layer in late August. The results show the wind field which blows from the Pacific to the Kinki region is increasing, and this explains rich water vapor flux is supplied to the Kinki region.

 There is tight connection between weather condition and phenology, and a lot of studies to date indicated that air temperature in particular significantly affects flowering. It has been also reported that influence by the rising air temperature likely due to the global warming is now emerged as changes in phenology in recent years. Bloom date of a cherry blossom has become earlier every year probably due to warming of surface air temperature. The empirical equations using the observed data taken from 1960s to 1990s were suggested by Aono and others, and it was revealed that averaged temperature of early spring in particular significantly influenced flowering date of the cherry blossom.

 In this paper, we confirmed that the error by the empirical equations suggested by the previous study became larger for the newer periods, because of the global warming advancement than 50-60 years ago when observation of the cherry blossom flowering began in the meteorological observatory stations. Here we modified the coefficients of the equations, so that the accuracy of the new equations was improved.

 As a basic study for formulating adaptation plans to climate change, meteorological and hydrological characteristics at the regional level in snowy cold regions were estimated by spatial interpolation method, using climate change data (MRI-NHRCM20) based on RCP emission scenario adopted by the IPCC’s Fifth Assessment Report. We quantitatively evaluated the future change of river water temperature, bias correcting and downscaling of climate change data, using the heat/water-balance model (LoHAS) and the tank model. The results of river water temperature simulation indicated that climate change is expected to raise river water temperature by approximately 6.6°C in May, also river water temperature area and period suitable for habitation of the landlocked masu salmon might decrease.

 LS3MIP (The Land Surface, Snow and Soil moisuture Model Intercomparison Project) has been approved as Coupled Model Intercomparison Project Phase 6 (CMIP6) endorsed MIP to investigate climate forcing and land feedback. Land modeling groups of international community are preparing their model experiments under CMIP6 corrdination. In this study, we describe the overview of the model experiment design of LS3MIP. Also, we introduce the results from our test drive using MATSIRO for checking whether the experiment configuration accords with the protocol. The results show that inter-annual trends of runoff, snow cover fraction, soil moisture and evapotranspiration in global scale are reproduced observational variability properly. Results suggest that water balance tend to decrease around Himalaya Mountains in the 20th century. This is probably due to negative trend of precipitation in the region.

 Climate reconstruction enables us to analyze long-term climate change. Understanding climate change is the key to improving climate prediction. In this study, we confirmed that annual variations of temperature for the past 1000 years can be reconstructed by data assimilation using oxygen isotopic data of ice cores, corals, and tree-ring cellulose, based on the previous study (Okazaki and Yoshimura, 2017). However, it’s difficult to reconstruct climate fields if proxies are sparse temporally and spatially. Even so, proxies’ impact on analysis does not always depend on the amount of proxies’ data. It is possible that about 20% proxies can reconstruct the past climate as same as all proxies used in this study, although observation impact is different spatiotemporally. Coral proxies in the central tropical Pacific appear to be the most effective for climate reconstruction by data assimilation.

 The earth climate has long glacial-interglacial cycles of ten to hundred thousands years. In this cycle, temperature increases rapidly and decreases slowly.

 This paper thus focus on a climate bistability problem originally suggested by Budyko and Sellers. In detail, an energy balance model associated with ocean and land process is used to simulate how the earth climate changes. The results suggest that the land and land ice effects promotes a climate bistability while the meridional heat transport and seasonality suppress a climate bistability.

 To solve global cycles of water, energy and carbon, most global models have applied radiative transfer scheme which expresses vegetation as a plate. To consider anthropogenic activities on forest and vegetation-climate feedback effect in boreal regions, it is necessary to express forest structure in radiative transfer scheme more precisely. In this study, through a sensitivity experiment by using high-resolution-three-dimensional radiative transfer model, we analysed the importance of five parameters related to forest structure for radiative transfer calculation.

 As a result, we show that the non-linear relationship between radiative transfer parameters and canopy coverage should be treated carefully and propose canopy length should be used as a new parameter. This study delivers important information to include data of forest structure obtained near future in global radiative transfer scheme.

 Nomarlized Differential Frequency Index (NDFI) is calculated by multiple microwave radiometers GMI and AMSR2 to produce a daily and 0.1-degree global surface water map for five years (2013-2017). Diurnal varation and bias between sensors in NDFI are adjusted. The basic performance of the surface water map is tested with global datasets. 5-year average NDFI and inundation ratio by Global Surface Water show very high correlation around Japan. Monthly variation of NDFI and inundation ratio by Yesterday’s Earth at EORC show positive correlation along some large river channels. NDFI shows a sudden increase at daily scale in some dry areas where daily variation of NDFI and antecedent precipitation index show positive correlation. In these cases, the increase in NDFI can be caused by inland water inundation rather than flood inundation.

 The purpose of this study is to clarify the temporal and spatial distribution of snowfall and snow-depth in Toyama prefecture for 43 years from 1973 to 2015. Daily mean air temperature, daily snowfall and daily snow-depth of 64 weather stations from 1 December to 31 March as snowfall season were used, and Mann-Kendall trend test was selected as one of non-parametric tests for this study.

 As a result of trend test, minimum air temperature of 45 weather stations had positive trend, 14 weather stations increased significantly (p value <0.05). Snowfall of 50 weather stations had negative trend, 25 weather stations decreased significantly (p value <0.05). Snow-depth of 47 weather stations had negative trend, 18 weather stations decreased significantly (p value <0.05). The decreasing trend of snowfall and snow-depth was remarkable in plain area. On the other hand, there was no significant trend of increase or decrease in mountainous area.

 This study was carried out to grasp the actual condition of long term changes in the river water temperture in the Kyushu region and considered the relationship with factors. The water temperature rise rate (℃ / year) was obtained by using the water temperature data of the hydrological water quality database from 1971 to 2016. The average value of them was 0.034 ℃ / year, significance upward trend was detected at 35 of 132 sites, and the maximum value was 0.109℃/ year. The water temperature rise rate tended to be high in the water system with high population density or the water system with small annual average flow rate.In addition, the relation between the water temperature rise rate and the position of the weir was analyzed, it was shown that the weir within 2000m upstream from an observation point had an influence on the water temperature rise on the downstream side. At these points, the water temperature rise rate was particularly high in from May to November, and the average value in each month were 0.060 to 0.087℃ / year.

 A fundamental study was conducted to estimate effects of adaptation options for climate change based on impact assessment of river flow change on reservoir operation for water use. Current climate data (1979-2003) and future climate data (2075-2099) projected by MRI-AGCM3.2S were used as meteorological input. River discharges in both climates were respectively estimated from climate data by use of Hydro-BEAM, a cell concentrate type hydrological model, integrating operation of reservoirs for water supply and water intake in the target river basins. As a result of analysis for the Yoshino Rive basin, it was shown that the storage capacity of the Sameura Reservoir for water use would not be sufficient due to a decrease in river flow under the future climate. It was also shown thatn regulating water demands could be needed for adaptation in addition to reallocation of storage capacities of a multi-purpose reservoir.

 The change of flood damage due to climate change and depopulation was estimated in this study by combining flood simulation and multiple asset distribution change scenarios. It was found that each effect varies depending on basins and thus flood controls should be determined for each basin. Moreover, it was confirmed that asset concentration in urban areas can increase flood damage and the efficiency of flood measurement can be different according to the planned precipitation. These results indicate that it is necessary to consider flood control on city planning.

 In this study, we analyzed and evaluated the methods to understand the uncertainty of the precipitation prediction by climate change among different precipitation models. Representative Concentration Pathways 8.5 (2075-2099) future climate characteristics in Vietnam has been chosen with four best-scored CMIP5 General Circulation Models (GCMs) on DIAS and used in-situ temperature and rainfall data for statistical bias correction and downscaling. The 25-year average daily temperature results of four GCMs demonstrated small variation in past and future climates while 25-year average future rainfall demonstrated a large variability among models. It is shown that the northern component of the Northeast Monsoon in the future would dominate the impact and consequently trigger the increased rainfall in central part of Vietnam and the pressure field in the future has an impact on the rainfall in Northern part of Vietnam.

 The purpose of this research is to investigate the influence of selection of terrain model on the flood runoff analysis due to the spatial distribution and magnitude of rainfall when performing double-nested dynamical down scaling from regional climate model with horizontal resolution of 20 km. Based on JRA-55, two levels of nesting with horizontal grid spacing of 20 km and 5 km were performed. In the 5 km experiments, the Grid mean model and the Envelope Mountain model were employed as the terrain model. The flood runoff analysis in the Nagara River was conducted using results of these climate models as input conditions. In the Envelope Mountain model, it was confirmed that many precipitation concentrates at the edge of the mountainous area, and the precipitation amount in the inland area decreases. In the Nagara River basin, the precipitation amount and the flood peak flow rate were clearly overestimated. The importance of choice of terrain model in 5 km regional climate model on flood-runoff analysis was presented.

 This study aims to reveal the uncertainty of regional climate model (RCM) and impact assessment model in climate change impact assessment study in local scale. Therefore, we made an attempt to evaluate the uncertainty of RCM and impact assessment model by using two types of runoff model (distributed hydrological model (DHM) and storage function model (SFM)) and dynamical downscaling (DDS) experiments on selected past flood events. As the result, it is found that the difference of peak discharge between DHM and SFM is small except for DS5 km. In DS5 km, peak discharge of SFM overestimated DHM since heavy rainfall was occurred in around catchment and a part of catchment.

 In August 2016, three typhoons continuous attacked Hokkaido and heavy rainfall occurred in various places. As a result floods of rivers and sediment disasters occurred mainly in eastern of Hokkaido. Recently, there is concern that the impact of such climate change will become apparent. In this situation, flood control measures considering future climate change are required in Japan. The current flood control plan have set up safety standards based on observation data in the past, however it is essential to utilize climate prediction data to respond to events occurring under future climate change. In this study, we estimated annual maximum rainfall averaging per basin from output rainfall of climate model using d4PDF which is a massive ensemble climate predictions, and proposed methods to evaluate accuracy of this model. The results show that climate prediction data has high reproducibility of observation value, in addition its rainfall in subject basin will increase about 1.4 times in the future.

 Recently, major disasters which seem to be the influence of climate change frequently occur in our country. Therefore, it is an urgent task to develop a flood control plan that takes into consideration the impact of climate change. In this study, we proposed a method to calculate probability rainfall considering the range of uncertainty from output rainfall of climate model using d4PDF which is massive ensemble climate prediction data. It also shows the change in the probability rainfall after climate change.

 A bias correction method that is appropriate for an huge ensemble dataset has been developed. The method was applied to the d4PDF, and validated usting AMeDAS dataset. The result indicates that the developed method has better reproducibility than previous methods. Especially the overestimation by the previous method is sucuesufully removed. The biascorrected dataset in the cathment scale shows it can reproduce historical extremes that is estimated using extreme value theory with small differences.

 This study proposes the simple correction formula of probability rainfall of 20-km d4PDF (database for Policy Decision making for Future climate change) for Gifu prefecture. Probability rainfalls by current climate simulation data of d4PDF and ground-based observation data by JMA are compared each by cumulative rainfall time. The relationships of the t-hour cumulative extreme rainfall of d4PDF and ground-based observation are proposed as follows: G1’_T =1.74 D1’_T, G168’_T =0.74 D168’_T., where, Dt’_T, and Gt’_T are T-year extreme probability t-hour cumulative rainfall of d4PDF and ground-based observation gage, respectively. RMSE for hourly rainfall is 18.1 mm, which is 22% of average value, and also RMSE for 168-hourly rainfall is 168mm, which is 30% of average value. The authors obtain the correction formula for short term rainfall with good accuracy and for long term rainfall with enough accuracy to predict landslide disaster.

 A real-time rainfall-runoff prediction system with a particle filter was developed using CommonMP and the effect of introducing multipoint discharge observation information was analyzed. The prediction system consists of nine storage function models with five discharge observation stations, and is applied to the Iwahana River basin (1221 km²) located at the upper part of the Tone River basin. Three data assimilation methods were compared in terms of prediction accuracy. The first method assimilates all storage variables of the storage function method using only the most downstream observed discharge; the second updates the storage variables using all observed discharge; and the third assimilates storage variables in each catchment using the just downstream observed discharge. The third method shows the best performance in terms of accuracy and computational amount.

 To seek accurate quantification of material transport in a watershed is an urgent issue in the examination of adaptation measures against climate change. A distributed hydrological model is a powerful tool for the issue, though there are many parameters and their values have wide ranges. The setting of parameters and validation take much effort. In this study, we developed an adjoint code of a distributed hydrological model and a method to estimate the spatial variability of parameters using the adjoint method. We applied this method to the Ibo River and assimilated river discharge, and estimated the parameters such as roughness coefficient, hydraulic conductivity, and storage coefficients. The results showed the adjoint model can estimate their spatial distributions suited to their characteristics. Especially, they were largely corrected from the middle area where the total precipitation was large to the assimilation point in the lower watershed. The simulations using the corrected parameters show the data assimilation can improve the peak discharge.

 Parameter uncertainty analysis of rainfall-runoff models is very important especially in urban watersheds due to the high flood risk in these areas. Among the different methods available for uncertainty analysis, bootstrap method gained popularity in view of its flexibility. Hence, this study aims to conduct the parameter uncertainty analysis of the urban storage function (USF) model, a storage function model specifically developed for the urban watersheds, using the model-based bootstrap method. We successfully evaluated the uncertainty of USF model parameters and the results exhibited that the 95% confidence interval of all parameters is wide compared with the search range during parameter estimation except for two parameters. Moreover, the parameters with the highest and least uncertainties were identified. Further, model simulation efficiency using the estimated parameters was found to be high with a Nash-Sutcliffe Efficiency value of 97%. Lastly, the effect of parameter uncertainty on model simulation uncertainty was analysed and found that the SCE-UA method along with the model-based bootstrap method can predict, on an average, 68% of observed data within the simulation uncertainty range of USF model.

 Data assimilation techniques are becoming popular in estimating hydraulic variables in ungauged basins with the recent advancements in the satellite technology. The Local Ensemble Transformation Kalman Filter (LETKF), which limits the assimilation domain by a “local patch”, is an efficient method for a global-scale data assimilation, but the optimization of the size and weighting function of the local patch is still challenging especially for river hydrodynamic models. Here we propose a method to estimate a reasonable local patch parameters, by fitting a Gaussian semi-variogram to the transformed Water Surface Elevation (WSE) data and defining the autocorrelation length for each river pixel. WSE simulated by CaMa-Flood hydrodynamic model was de-trended, seasonality removed and standardized to make the data suitable for semi-variogram analysis. A case study over the Amazon mainstem suggested that the auto-correlation lengths for upstream and downstream of Obidos GRDC location were derived respectively as 1886.69 km and 688.66 km. The semi-variogram analysis indicated that the river pixels of entire mainstream of the Amazon are correlated together. The estimated auto-correlation length and weighing function could be useful to determine the optimum parameters of the LETKF local patch.

 We developed a new surface flow direction datasets at 1-sec (~30m) resolution for the entire Japan domain, using “Kiban Chizu Joho” digital elevation model and “Kokudo Suchi Joho” water body layers. The calculation of flow directions for a large domain used to be difficult due to errors in the input elevation data. We solved this problem by a new algorithm, which first calculate the initial-guess flow directions by a steepest slope method, and then ensure river connectivity by reversing the initial-guess flow directions when needed. The new flow direction data shows better consistency to the accrual river networks compared to the previous HydroSHEDS flow directions. We also generated supplementary data layers such as flow accumulation area, adjusted elevation, and river width. The new flow direction datasets will be published online, and considered to advance any geoscience studies which relies on flow direction data.

 A few of recent studies have applied deep learning technique for flood forecast. This study utilizes Recurrent Neural Network (RNN) with Gated Recurrent Units (GRUs) to hindcast the Kanto-Tohoku Flood in September 2015. Additionally, based on linear reservoir function, it applies Exponential Filtering (EF) as a preprocessor of input data to transform the statistical characteristics of input variable (i.e., rainfall) to of the target variable (i.e., river water stage). Compared with Feed Forward Network (FFN) model and RNN model without EF, the proposed model outperforms the flood events in Kinu river basin. In particular, it results reduced error for the highest water level which is 3.43 meter higher than of the highest level during the training period . Also, we investigate dependency of prediction skill on neural network structure and input data information density in Tone-river and Teshio-river basin, which shows critical number to predictability of water level and rainfall observation sites differs among target stations and basins.

 Synthesis probability method is often used to estimate design flood peak discharge. In this method, probability distribution of annual maximum total rainfall which is derived by hydrological frequency analysis is converted to that of annual maximum flood peak discharge which incorporates influence of rainfall’s spatio-temporal distribution under some assumptions. There is uncertainty in hydrological statistics corresponding to our recognition for total number of extreme value data. To evaluate uncertainty of hydrological statistics, we proposed hydrological frequency analysis method introducing confidence interval based on probability limit method test. In this research, we extend synthesis probability method by using this confidence interval. And, we propose the theoretical framework of uncertainty evaluation for T-year annual maximum flood peak discharge which is based on the extended synthesis probability method.

 In this study, we provide a contribution toward monitoring the unsteady behavior of discharge-stage hysteresis in a mountainous river. In order to monitor the unsteady patterns of river discharge hysteresis estimated by means of a novel fluvial acoustic tomography system (FATS), we therefore observed the corresponding river water surface slopes hysteresis during different rainfall events. We found that the temporal variations in water surface slope-stage during different scales of rainfall events resulted in different patterns of hysteresis loops which can be classified into three main categorizes: (i) no hysteresis, (ii) clockwise hysteresis, and (iii) counterclockwise hysteresis. In the case of low intensity rainfall events, no hysteresis behavior was detected. Meanwhile, for medium and high intensity rainfall events resulted in either clockwise or counter clockwise (WS-WL) hysteresis patterns. Finally, for medium and high intense hydrological events, water surface slope become almost constant after passing a certain level.

 Mount Sakurajima is one of the most active volcanoes in the world, where its frequent eruption causes the occurrence of debris flows every year. This paper discusses the fundamental aspect of spatial and temporal rainfall variability in Mt. Sakurajima by using X-band polarimetric (multi parameter) RAdar Information Network (XRAIN) related to debris flow occurrence. The analysis used XRAIN data from May 2015 to April 2016. The Meiyu-baiu rainfall phenomenon strongly affects the seasonal variability of rainfall in Mt. Sakurajima, which increases the precipitation in June to July. There are the possibilities of the beam blockage and the Vertical Profile Reflectivity problems to underestimate the XRAIN rainfall in Arimura and northern area of Sakurajima for a total of more than 1000 mm depth annually. However, there is no debris flow event reported in northern part during 2015 despite thicker accumulated volcanic ash, which may indicate that this area indeed receives less rain throughout a year. Most of the debris flows happen because of more than 9 hours long rainfall events, which validates the importance of rainfall on debris flow occurrence in Sakurajima.

 Frequent flooding has been occurring in the central Phnom Penh City area, chronically affecting daily life, health and various socio-economical activities. Among them, the characteristics of the rainfall, especially for the shorter time scale ranging from hourly to daily, are poorly understood because of the limited measurement data of rainfall, thus it is still challenging to find a comprehensive and quantitative view of current and future rainfall characteristics in this region. The analysis indicated that the heavy rainfalls are frequent in recent years and they have more acute temporal variations than ever recognized by the intensity-duration curves. With the statistical bias correction and downscaling of GCM outputs, hourly rainfall intensity is projected to increase by 20 % to 80 % than current design rainfall, depending on GCM used for the analysis. The spatial variations are found to be very large, but no significant difference was found for the monthly totals. Because of the lack of reliable long-term hourly or sub-daily rainfall records, we evaluated the 3-hourly rainfall extremes from the MSWEP, and it was concluded it was difficult to use it for quantification of the rainfall for a short time.

 This study deals with trend analyses for the annual maximum data samples reinforced with historical information of the lake water level over 290 years at Lake Biwa, the flood water level over 328 years at Ohzu site in the Hiji River and the storm surge height and sea level height over 146 years at Venice, and an LSM model-based extreme value analysis for each of them excluding the sea level height data sample with a significant increasing trend. The following results are obtained. 1) In the t-distribution-based trend analysis for the slope of a straight line fitted to time series of each data sample no statistically-significant trend is detected. 2) Standard deviation of the probabilistic lake water level estimated from the data sample including historical information at Lake Biwa becomes significantly smaller than that estimated from the data sample with no historical information, which means an improvement of the statistical reliability. 3) Construction of the dam at an upstream location in the Hiji River may bring about a reduction of around 2 m in the estimate of the 100-year return flood water level at Ohzu site. 4) An estimate of the 500-year return storm surge height at Venice may be reasonable to some extent, considering the occurence frequency of extraordinary storm surges in the past 1200 years.

 Bayesian approach now becomes popular by using MCMC method. One of the advantages is that the assumption of the central limit theorem for the estimation errors is not required, thus the estimations of the parameters are distributed and they have the probability densities. But this will show troublesome aspects for engineers. We should choose a reference value for each parameter and our target return value. It is contrasted that maximum likelihood estimation gives us principally a point estimation with the accompanying errors secondary. This research shows the prediction distribution for coming extreme precipitaion in the condition that it exceeds the return level, and it gives also its related point estimation for parameters that will serve for flood design.

 The probabilistic precipitation may have a difference due to the accumulation of observation data, but it is not taken into account in the design of soil structures. In this study, the difference of the probabilistic precipitation in a certain return period was investigated, using the observation data of the regional meteorological observatory and the surface metrological observatory. From the statistical analysis using the observation data of the regional metrological observatory, the area where the probable precipitation increases can be estimated. Furthermore, statistical analysis using surface metrological observatory data revealed an increase in probabilistic precipitation at specific points.

 We tried to model hydrological data by using fat-tail distribution and fractal theory. First, we confirmed that annual maximum 1-hour precipitation is not fractal but white noises by spectrum analysys. We can generate the data by random numbers whose distribution is GEV or the Gumbel distribution. Lévy distribution’s tail is too fat. Secondly, we tried to model time series data of water quality by using fractal model like fBm or fLm. We found that not only Lévy distribution but also GEV and the Gumbel distribution are appropriate for distribution by which random numbers are generated when the spectral synthesis method is applied as fraclatl-generationg method.

 Brightband is known as the overestimate area of radar observation by melting of precipitation particles. In this study, correction of Brightband area by the linear interpolation was examined. This correction was applied to precipitation observation from November 2014 to April 2015, and overestimation of radar rain was successfully eliminated in the minutes identifying Brightband or in the closer area of the radar site. However most of overestimation was remained. In order to improve all the period of wintertime, the correction of conversion coefficient near 0 degree height should be considered.

 We analyzed the spatial distribution of convective clouds using Doppler radar for the Kanto and Tohoku torrential rains that caused a major flooding event in the Kinugawa River basin in September 2015. Specifically, based on observation data from multiple X-MP radars, a three-dimensional wind speed field was estimated by the variational method called MUSCAT method.

 Using the observed results, the convective clouds that are generated were moved momentarily. Then, the characteristics of the three-dimensional wind speeds inside and around them are arranged. Furthermore, when analyzing the temporal characteristics of the spatial interval of the convective cloud within the time zone in which the linear rainfall exists, lively convective clouds existed at intervals of about 10 km and about 50 km.

 Radar-based Quantitative Precipitation Estimation (QPE) is primarily stated as techniques representing the best-fit rain rate measured at the ground. The current QPE method relies on either CAPPI maximum value or lowest radar tilt value. In fact, some discrepancies due to lag time and spatial variance remain found which cause the errors systematically propagated over a period. Furthermore, uncertainty factors on a void interspace due to the existence of a gap between available radar beams and ground have caused misleading in determining the actual rain rate. The real-time QPE model in this study is intended to improve the nowcasting rainfall predictor system. The multivariate projection model is used to predict the actual rain rate through the entanglement of physically precipitation factors such as wind shear, relative humidity, evaporation rate, and vertical moisture flux obtained from atmospheric sounding data. The vertical profiles of rainfall at various CAPPIs are collected and used as the response variable by the use of physical factors as a predictor variable to obtain the parameter coefficient value. This value interprets the signature of rainfall at the observed CAPPIs which then could be used for actual rainfall projection at the lower altitudes by real-time. Finally, the validation is taken through radar-gauge cross-correlation representing the actual rain rate. The model is performing well when it has a high correlation, least bias, and zero lag time.

 To calculate the return period of previous heavy rainfall events and to simulate rainfall data under a future climate, high-accuracy and high-resolution spatial rainfall data are required. While Precipitation Analysis (Radar-AMeDAS) by Japan Meteorological Agency is used in previous studies, X-band Multi-Parameter radar network (XRAIN) is used in this study to take the advantage of its higher-resolution. XRAIN has been pointed out to overestimate rainfall over mountainous area and around radar sites. To correct the overestimation, XRAIN measurements are matched up with Dual-frequency Precipitation Radar (DPR) on the core satellite of the Global Precipitation Measurement (GPM) mission, then correction methods for XRAIN are developed. And the correction result are evaluated by the rain gage data and the ground radar data provided by the Japan Meteorological Agency.

 To accurately estimate raindrop size distribution (DSD) and rain rate from multi-parameter radar, in this study, an advanced estimation algorithm was proposed basing on a conventional method proposed by Yamaguchi (2012) (Yamaguchi method). Yamaguchi method retrieves DSD parameters from radar reflectivity (Z_h) and specific differential phase shift (K_DP). However, Z_h is often underestimated due to rain attenuation, in the case that rain attenuation is strongly occurring. In this study, a new advanced method to retrieve accurate DSD information has been developed to solve the rain attenuation problem. DSD retrieved from K_DP and difference reflectivity (Z_DP) theoretically showed higher accuracy than that from K_DP and Z_h in large attenuation cases under pseudo observation experiments. This fact was also evaluated by observation data using an operational X-band multi-parameter radar and a raindrop disdrometer. These results are induced by the fact that Z_h is largely affected by intense rainfall attenuation compared with Z_DP. Therefore, it is concluded that the proposed new method is possible to retrieve DSD with higher accuracy than Yamaguchi method because rain attenuation is greatly affect retrieval accuracy of DSD.

 Accurate rainfall forecast is indispensable to realize non-structural counter measures for floods. On the other hand, there are still some difficulties in deterministic numerical weather forecasting, and the Japan Meteorological Agency makes test operations of ensemble weather forecast by using a meso-scale regional model in recent years. In this study, preliminary experimetns were conducted to develop a technique for reconstructing ensemble weather forecasting information by application of the particle filter which is one of sequential data assimilation techniques. The results of reconstruction experiments with different variance-covariance matrices in particle fileter showed that appropriate variance-covariance matrix is important to avoid degeneration of ensemble forecast. Observation at multiple times could improve results of reconstruction by partile fileter. Simultanueous filtering and reconstruction of ensemble forecast at multiple observation sites showed significant improvement of thread scores in target sites.

 For the purpose of the development of the monitoring and the understanding of the formation process of localized heavy rainfall, we explained the formation process of localized heavy rainfall in summer season using the localized heavy rainfall events on the Osaka Plain from the analyses of the multiple remote sensing observations data (Global Navigation Satellite System (GNSS), Himawari-8, X-band Multi-Parameter radar and X-band phased-array weather radar). In wet environment, clouds appeared on the Osaka Plain and moved to the plain from outside. In the situation, 12 isolated precipitating areas appeared with the appearance of new cumulonimbus clouds on the plain. The lifetime every precipitating area was 17 to 103 min; the maximum value of time accumulated rainfall near surface every precipitating area was 2.3 to 57.7 mm. The lifetime of the precipitating area seemed to have a relationship at temporal variation of precipitable water vapor derived by GNSS at the GNSS receiver near the precipitating area. We expect the estimation of the occurrence potential of localized heavy rainfall and the prior detection of the occurrence of rainfall at surface through the time variation of the multiple remote sensing observations data.

 Recently, the localized torrential rainfall occur frequently in urban area, and it cause internal water flooding in a short time. In order to prevent such disaster, we studied the detection method for the localized torrential rainfall using vortex tubes observed by phased array weather radar installed in Osaka and Kobe. Phased array weather radar can observe three-dimensional rainfall and doppler velocity every 30 seconds and detect vortex tubes with vorticity of positive and negative paires inside the rapidly developing cumulonimbus cloud. In this study, we first created CAPPI data with 250m special resolution, and studied the method of detecting vortex tubes. Secondly, we proposed the spatiotemporal density of vortex tubes for detecting the organized cumulonimbus clouds with rainfall intensity of 50mm/h or more that maintain more than 10 minutes.

 High-resolution Precipitation Nowcasts (HN) provided by JMA (Japan Meteorological Agency) is a state-of-the-art operational short-term precipitation prediction. However, schemes covering uncertainties of prediction are not sufficiently considered in the single prediction. Also, modifications of predictions corresponding to the temporal change in prediction error are not sufficiently considered in the system. Thus, it is necessary to introduce ensemble prediction schemes. In addition, correcting schemes based on the systematic error of the prediction result are also expected. In this study, a precipitation prediction system is installed that the most accurate prediction is selected by the evaluation using scores of previously predicted precipitation among multiple predictions basing on precipitation motion extrapolation with various schemes. We also introduced a scheme to correct predictions based on information of errors in past predictions. The system showed the comparable scores to HN. For the line-shaped rain-band cases, the shapes and amounts of the maximum accumulated precipitation were more accurately predicted in the proposed system for the forecast time of 60 minutes. It showed high responsiveness to the occurrence of rainfall for cases with severe changes in precipitation. Further development of this system is expected in the future.

 Ensemble forcaset experiment and data assimilation experiment on a line-shaped rainband that occurred at northern part of Kyushu in 2017 have been conducted. In the ensemble forecast experiment, although the ensemble mean of forecasted rainfall was underpredicted than the observed rainfall, two members could well predicted the line-shaped rainband. A humid air mass that was lifted-up by Mt. Seburi was lifted-up by a wind convergence that was occurred by both northan wind and southan wind of Mt. Seburi. In the data assimilation experiment, X-band polarimetric radar data was assimilated to understand the developing mechanism of the parallel two line-shaped rainbands. The two rainbands caused the outer flow near surface, which made the other rainband strong.

 The objective of this study is to verify the capability of Rapid Scan Observation (RSO) of Himawari-8 data for a deeper understanding of the development of cumulus cloud, which can generate Guerilla-heavy rainfall (GHR). The first approach in this verification is utilizing the Rapid Development Cumulus Area (RDCA) index, which is generated by RSO, for estimating cumulus life stage. The estimation of cumulus life stage will be approached by comparing between RDCA index and cumulus life stage estimated by radar. Based on the first trial of the comparison, we found the possibility of RDCA index to estimate the cumulus life stage, as we discovered that there is a good correlation between RDCA index and cumulus life stage in our study case. In the second approach, we compare the radar-estimated hydrometeor type with cloud top conditions retrieved from RSO in the earliest stage of rain. The three types of brightness temperature difference (BTD) using band no. 11, 13 and 15 of RSO data are used to know the cloud condition by distinguishing water and ice phases in a cloud in the baby-rain stage. In this first trial of one case analysis, we found a potential of utilization of RSO to estimate cloud phase in the early stage of cumulus cloud since there is a good spatial correlation between them.

 A study of the variability of precipitation in space and time has been carried in southwestern Bangladesh. Quality controlled homogenous daily precipitation data of 9 stations for the period of 1981-2010 are used for the study. A total of 9 indices and monthly mean are examined. Mann−Kendall test in conjunction with Theil−Sen’s slope method has been used to reveal the significance of trends and quantify their magnitudes. A significant increasing trend in consecutive dry days is observed. An increasing trend of annual maximum consecutive 5-day precipitation is also observed. A projected trend in near future (2020-2050) is analysed using three general circulation models under two representative concentration pathway scenarios. In coming days, it is anticipated to have wetter monsoon and drier winter.

 A rainfall prediction model was proposed by applying Convolutional Neural Network to spatiotemporal two-dimensional data created from the time series information of meteorological observation at several sites. The rainfall threshold and the prediction lead time were set as the prediction setting. We used several meteorological variables as input data and studied a prediction model that does not use precipitation as input data. As the difference in the prediction accuracy in the prediction setting, it is confirmed that prediction accuracy decreases as the prediction lead time becomes longer, and rainfall prediction becomes more difficult as the threshold becomes higher. From the prediction model without precipitation, it turned out that the information of precipitation had an influence on the model accuracy, and it was suggested that other meteorological variables also contain information related to rainfall prediction.