| dc.description.abstract | Rainfall is the most common triggering factor for landslides in Taiwan, and thus previous studies have generally used rainfall variables (i.e. rainfall intensity) to assess slope stability. Theoretically, in addition to static factors, such as topography, geology, and soil-bedrock physics, the occurrence of rainfall-induced landslides requires a comprehensive consideration of hillslope hydrology, which can be associated with the development of soil water content during rainfall events.
In this study, for triggering factors, a three-layer tank model transfers the transient rainfall to the Soil Water Index (SWI) and Soil Water Ratio (SWR) adopted to simulate the conceptual soil water content and its distribution within a soil column. The SWI and SWR2 were selected as the proxy of triggering factors by Major Disaster Event Reports from Soil and Water Conservation Bureau (SWCB). A critical rainfall model is a process-based model adopted to assess slope stability and incorporates static factors to estimate the critical rainfall (Qcr) as the proxy of static factors. The SWImax, SWR2max, and Qcr were used to derive an effective logistic regression (logit) model for assessing landslide landslides and stable areas. Modeling historical landslides induced by Typhoon Mindulle (2004), Haitang (2005), Kalmaegi (2008) and Morakot (2009) in Kaoping Watershed, southern Taiwan, the proposed integrated model produced the highest MSR (Modified Success Rate) of 80.5% for Typhoon Kalmaegi and the lowest MSR of 76.2% for Typhoon Morakot. The results proved that applying soil water content could perform well in landslide prediction. Through the analysis of the Jack-knife method, the integrated model could equilibrate the relative contribution of triggering factors and static factors for the best MSR. However, due to the cross-validation, so this study combines three events to develop the model and tests the performance of the model using the fourth event. The multi-event model produced the highest MSR of 76.4% for Typhoon Mindulle and the lowest MSR of 70.9% for Typhoon Haitang. Although it had little worse MSR than above and may bring about over- or underestimating landslides, the results revealed that combining multi-event data can derive a universal integrated model. In summary, this study proposes a novel landslide prediction model that integrates soil water content and other static factors, showing excellent results and applicability. | en_US |