Author(s): Hao-Che Ho

Linked Author(s):

Keywords: Multi-phase approach; Early-warning system; ANNs; Typhoon events; Sediment flux

Abstract: The watershed sedimentation could be determined as the severe natural disaster source. The differences between the sediment yield estimation and the real situation are large due to the inadequacy of on-site measurement data. A robust early-warning approach that encloses both the physical mechanism and statistical analysis was proposed to predict the sedimentation and improve the deficit on the current methods. This three-phase early-warning system includes, Phase I (data collection), Phase II (data generation), and Phase III (AI prediction). In Phase I, HEC-HMS was applied to transform the measured precipitation data to the flow discharge from different sub-catchments. The empirical formulas of landslide volume and soil erosion were then adopted to provide reasonable boundary conditions to Phase II. A 2D model, SRH-2D, was verified with on-site data and conducted to obtain simulated data for the temporal variation of the sediment flux with different storm events. The simulated data can be the input for Phase III to train and test the artificial neural networks. Three prediction models presented a good performance from 1-h to 6-h lead time comparing to the observation. Even though the relatively poor performance was shown in the higher sediment discharge and the difference was gradually increasing with the longer forecast lead time, the prediction models still provide the proper prediction trend, especially the forecast lead time from 1-h to 3-h. This approach shows a significant improvement on the sediment prediction due to the sufficient simulated data, and can be well-applied to better manage sediment problems in advance.

DOI: https://doi.org/10.1007/978-981-97-6009-1_66

Year: 2022