Three-phase data augmentation for the prediction of sediment flux in mountain basins during typhoon events

Abstract

Abstract Discrepancies between estimated sediment and actual yields are large. The sedimentation often causes severe damage due to a lack of on-site measurement data in the current disaster prevention system. This paper presents a robust early-warning system in which the statistical analysis used to predict sedimentation is conducted within the context of the underlying physical mechanisms. This three-phase early-warning system employs data collection, data generation, and AI (Artificial Intelligence) prediction. Data collection involves the use of HEC-HMS (Hydrologic Engineering Center – Hydrologic Modeling System) to transform measured precipitation data into flow discharge from various sub-catchments. Empirical formulas related to landslide volume and soil erosion are then used to establish suitable boundary conditions for data generation. Finally, a 2D model, SRH-2D (Sediment and River Hydraulic-Two Dimension model), is used to simulate data pertaining to temporal variations in sediment flux under various storm event scenarios. The simulated data are then used as input for the training and testing three artificial neural networks. The primary strength of this study is to improve the prediction ability in large-scale sedimentation issues in mountain rivers. These models returned good prediction results with 1–6 h lead times by interdisciplinary integration with the hydraulic and statistical fields.