Kazuhiro Oda

Soil moisture condition inside the slope affects the occurrences of slope failures. The previous study revealed that the estimated regression model using a recurrent neural network was available to reproduce the field measurement data of volumetric water content at the time of heavy rain. However, the method of giving input data and the machine learning algorithms should be more discussed. This study discussed the method for estimating the regression model to reproduce the soil moisture condition by changing input data and machine learning algorithms. As a result, the estimated regression model by using antecedent precipi-tation indices as input data and RNN as the machine learning algorithm led to predict the variation of the volumetric water content with sufficient accuracy even on slopes with different soil types.

It is important to identify the model that can simulate the field measurement data of soil moisture conditions to predict the occurrence of landslide disasters due to heavy rain. This study verified the applicability of the recurrent neural network to predict the field measurement data of volumetric water content. The recurrent neural network model was estimated by using the training data at the time of weak rain, and the estimated model simulated the test data at the time of heavy rain with enough accuracy. The simulation results led to the conclusion that the recurrent neural network was an effective method to predict the field measurement data of volumetric water content.

<p> The authors have tried to estimate the seepage analysis models based on field measurement data of volumetric water content using the merging particle filter which is one of the data assimilation methods. In the previous studies, one-dimensional seepage analysis models were assumed, and unsaturated seepage behaviors were simulated by setting the free drainage boundary condition at the bottom of the one-dimensional models. However, in the case of heavy rain, because of the observation rising groundwater level at a field slope, the simulation setting the free drainage boundary condition could not reproduce the field measurement data. In this study, the authors proposed a new method estimating the seepage analysis model including the boundary condition at the bottom of the one-dimensional model using merging particle filter, and the applicability of the proposed method was verified. As a result, the estimated seepage analysis model by data assimilation could reproduce the measurement data of not only numerical experiments but also field slope. Therefore, the availability of the proposed method simulating from the unsaturated condition to the saturated condition was verified.</p>

The identification of soil hydraulic parameters is of importance on the simulation of rainwater infiltration. The authors have tried to identify the parameters based on the field measurement data, by using the Merging Particle Filter method. In this study, a new method of data assimilation of seepage analysis model including a boundary condition was proposed. The seepage analysis model was identified based on the measurement data of numerical experiments and in-situ slope. Thus, the availability of the proposed method was verified.

The accuracy of numerical simulation by infiltration model identified by a merging particle filter was discussed. The infiltration model identified by a merging particle filter could reproduce the behavior of volume water contents with high accuracy. The results of numerical simulation almost coincide with those of data assimilation although the soil moisture parameters were varied with the process of data assimilation.

The authors hypothesized that displacement on a model slope would not occur unless the initial quasi-saturated volumetric water content was exceeded. This prediction was based on a past experiment using a small model slope for the purpose of predicting slope failure beforehand. However, in previous experiments, the soil-layer thickness of the slope was as thin as 100 mm, so the behaviors of sensors installed at the upper, middle, and lower slopes were nearly the same. Also, the size of the model slope was small, and the characteristics of each point could not be grasped. Therefore, in this study, we verified whether the hypothesis obtained from past experimental results was valid even when the topsoil-layer thickness to 300mm and the model size are increased. The following findings were obtained. Displacement was confirmed to occur in the vicinity of the toe of the slope after the volumetric water content exceed the initial quasi-saturated volumetric water content. On the other hand, in the middle part of the slope, while still below the initial quasi-saturated volumetric water content, displacement was confirmed. Thus, depending on the installation position, the displacement may occur before the initial quasi-saturated volumetric water content is exceeded.

It is important to identify soil hydraulic parameters, in order to simulate rainwater infiltration process adequately. Recently, due to the development of sensing technology, field monitoring systems have been spread. In the systems, the volumetric water contents are measured, and large number of data are accumulated. The authors tried to identify soil hydraulic parameters based on the field measurement data, by using the Particle Filter method, which was one of the sequential data assimilation methods. In this study, the authors focused on the Merging Particle Filter, which was one of the Particle Filter algorisms, and the availavility of three algorisms of Particle Filter method was compared. As a result, it was found that the Merging Particle Filter was most available algorisms to identify soil hydraulic parameters based on field measurements.

近年の突発的かつ局所的な集中豪雨に伴う斜面災害に対し、現行の降雨規制基準の補助的指標として、初期擬似飽和体積含水率を提案した。また、その工学的意味を一連の模型実験を通じて明らかにした。

土壌水分特性ﾊﾟﾗﾒｰﾀをﾃﾞｰﾀ同化手法(粒子ﾌｨﾙﾀ)を使って決定する際,最も効率的なｱﾙｺﾞﾘｽﾞﾑについて考察している。SISよりもSIRの方が効率的であることを明らかにした。