The sand and gravel which are materials of trapezoid CSG dams have quality fluctuation because they are uncontrolled raw materials. Therefore, it is necessary to monitor the material characteristics (grain distribution and surface water content) by laboratory tests. Then the authors had developed the automatic system that continuously monitors grain size of CSG materials by a digital image analysis technique and had verified its effectiveness by applying it to field crushed materials. On the other hand, since the surface water content of riverbed gravel is high due to many fine particles, and it is concerned that they effect the result of digital image analysis of grain distribution. Therefore, the authors newly developed automatic monitoring system for grain distribution and water content which combined digital image analysis and RI moisture meter measurement. Measurement of grain distribution is corrected by that of surface water quantity. Hence the system ensures the accuracy of grain distribution measurement for the material with high surface water quantity. This paper reports the result of basic experiment and field verification in applying the system to Sanru Dam (Kamikawa District, Hokkaido).

The strong rainfall which we have never assumed in the river design happens frequently and many water and sediment disasters happen. In order to minimize these damages, the dams which are able to control the flood damages are managed to maximize the reservoir utilization based on the regulations of Pre-discharge operation and Disaster prevention operation in the time of abnormal flood. But when we are going to aim at maximizing the reservoir utilization more effectively, it is indispensable to improve the accuracy of dam inflow forecasting. In this thesis, we will show that the ensemble machine learning method called Gradient Boosting is much more accuracy than the neural network method when we use them for the dam inflow forecasting scheme.