We have used a set of predictive rainfall and analytical models as main methods when predicting water levels or discharge levels. However, available flood forecasting technologies are inadequate and have not reached general utilization through practical operation. An important point is that fewer observation stations and less information exist for models in dam basins than for river management segments under dams. Therefore, predicting water levels or discharge levels based on physical models is difficult. As an alternative, the Internet of Things(IoT) is supported by the improvement and lessening costs of sensors and network technologies. Therefore, we set sensors artificially in the Digital Twin (A digital twin is a digital replica of a living or non-living physical entity. By bridging the physical and the virtual world, data is transmitted seamlessly allowing the virtual entity to exist simultaneously with the physical entity.(Saddik,2018)) basin produced using the distributed flow analysis model and using Long Short-Term Memory(LSTM), a sort of deep learning to confirm the accuracy of dam inflow forecasts obtained with different numbers and locations of sensors. Results demonstrate that the predictive precision fundamentally improves with increasing the number of sensors but that it doesn’t improve with sensors which generate biased data. Furthermore, we can infer a sensor number limit. Results show that adequate sensor location improves the predictive precision.

　Although the flow duration curve is important for hydropower development, observations of streamflow regime require considerable expenditures of cost and time. The stochastic flow regime prediction method is known to allow development of flow duration curves through use of limited observational data. Therefore, this method was applied to 30 basins in Japan. We validated it quantitatively. This method necessitates recession flow parameters, which were obtained in two ways: geomorphic recession flow model (GRFM) and geological recession approach (GRA). Results demonstrated that the flow duration curve fitted better for applying GRFM than GRA. This finding corresponded to the result found for the expressive ability for basin characteristics between GRFM and GRA. Actually, the basin gradients were steep because the basins were located in mountainous areas, which was consistent with the condition derived for GRFM. When applying GRA, we observed overestimation at the low flow regime. To mitigate this error, we recommend using water-storage consisting of baseflow component.