Probing relation between rainfall pattern and seismic detected water-and-sediment events

Southern Taiwan often experienced abundant monsoon seasons during seasonal transitions, and monsoons and typhoons controlled the rainfall patterns to be complex and varied, resulting the high intensity, prolonged duration, and high concentration. The aforementioned rainfall characteristics can increase the risk of water-and-sediment-related disasters.  To explore the correlation between rainfall patterns and water-and-sediment events, this study employs micro-seismic monitoring network, and the selected Putanpunuas River in southern Taiwan as a case study site. Frequent landslides in the middle and upper watershed supply the river with stable source of sediment materials. Consequently, during the periods with strong precipitation, our study site the shows high susceptibility of water-and-sediment events.  The seismic network comprises one station (BNAR) on the right bank and two stations (BNAL, BNAS) on the left bank downstream of the Putanpunuas River, and an additional station (BNAF) at the confluence of the Putanpunuas River and the Laonong River.  By conducting a series of spectrogram analysis, the average power spectral density (PSD) time series of each station can be computed. Then, we further quantified the seismic signal characteristic parameters for each water-and-sediment events.  This study initially employs various machine learning algorithms (Decision Tree, KNN, K-means, Auto-sklearn) to develop an optimized model for identifying water-and-sediment events, classifying different types of events, such as flooding (FD), debris flooding (DFD) and debris flow (DF), then providing a 4-year-length (2019~2023) catalog of water-and-sediment events.  Rainfall data including hourly precipitation and LiDAR estimated rainfall are collected from the rain gauge stations nearby study area. Using a certain definition (e.g., 4 mm/hr threshold for picking start time) of rain episodes, we calculated total number of episodes and established a rain episodes catalog.  The aforementioned datasets allow us to probe the relationship between rainfall patterns and water-and-sediment events, aiding in inferring the main rain episodes characteristics associated with water-and-sediment events . The  results of this study can be applied to predict potential water-and-sediment event types in Putanpunuas River using rainfall information as input. This can facilitate relevant early warning operations, reducing the societal impact of water-and-sediment disasters.

Key words : Rainfall Patterns, Rain Episode, Micro-seismic monitoring network, Putanpunuas River, Water-and-Sediment Events, Machine Learning