| 摘要: | 降雨所誘發的淺層崩塌不僅是重要的地形作用，更是嚴重的自然災害。這種崩塌的成因通常為降雨入滲至不飽和土壤，使得土壤的有效應力減少、剪力強度降低，而土石因受重力牽引，自斜坡上滑落，形成淺層崩塌。在淺層崩塌的潛勢分析中，物理機制模式 (physically based model) 是種常見的分析方法，其中最早提出之SHALSTAB模式結合了水文模式以及邊坡穩定模式，並在許多崩塌潛勢分析的研究中有著不錯的成果。SHALSTAB模式盡管已在許多研究中被採用，但其假設破壞面位於土壤與基岩的交界面，且地下水也累積於此交界面上。然而，這樣的假設可能會忽略土壤與基岩裂縫中流動的裂縫流 (fracture flow) 對於邊坡穩定所帶來的影響。因此，本研究假設 (1) 單層土壤發生崩塌 (2) 雙層結構，僅上層發生崩塌 (3) 雙層結構，上、下層皆發生崩塌，三種不同的破壞情境，並透過這三種不同的破壞情境建構崩塌模式。與此同時，八流向演算法 (Eight flow direction, D8) 、多流向演算法 (Multiple flow direction, MFD) 、無限流向演算法 (Infinity flow direction, D∞) 以及整合單流向與多流向演算法所發展出的阿爾法流向演算法 (Alpha flow direction, Dα)等四種不同的流向演算法，也將應用於研究之中。本次研究以台灣南部的荖濃溪之一子集水區為研究區域，此處於2009年遭受莫拉克颱風的襲擊，隨後發生的崩塌與土石流事件更是嚴重地影響當地的住戶。模式最後的模擬成果將與崩塌目錄進行比對，並透過修正成功率 (Modified Success Rate, MSR) 進行量化，藉此評估各模式的預測準確性。結果顯示，在三種情境中，情境二的預測結果最佳，情境一次之，情境三則較差，代表「雙層結構，僅上層發生崩塌」的假設，確實能改善分析成果。此外，在情境二的分析成果中，MSR值最高者為情境二搭配Dα法 (75.49 %)，D8法次之 (75.10 %)，MFD法第三 (74.99 %)，D∞法則最低 (74.51 %)，代表流向演算法也具有影響崩塌分析結果可能性。;Rainfall-induced shallow landslides are not only important geomorphic processes but also natural hazards. These landslides are commonly caused by the transient infiltration into initially unsaturated soils and can cause the debris flows mixed with subsurface flow. For shallow landslide susceptibility analysis, physically-based models have been developed, and SHALSTAB is one of the commonly used physically-based models, integrating a hydrologic model and a slope stability model to predict slopes prone to landslides. SHALSTAB assumes that the failure plane is located at the interface between soil and bedrock, and the groundwater is also accumulated at the soil-bedrock interface. However, this assumption may neglect the fracture flow within soil-bedrock fractures, which may also significantly influence the slope stability. Therefore, this study assumes three different failure scenarios: (1) single-layer model (2) double-layer model with middle failure plane (3) double-layer model with bottom failure plane. Meanwhile, four different flow-direction algorithms, including D8, MFD, D∞ and Dα have been applied in the landslide model. The study area is located at Laonung watershed in southern Taiwan, which was hit by Typhoon Morakot in 2009, and landslide and debris flow events severely affected the local settlements. Finally, the prediction results were compared with the landslide inventory, and this study assesses the model performance of different experiments using Modified Success Ratio analysis, to examine the validity of implementing the fracture flow in a physically-based landslide model in the study site. The results show that scenario 2 with applying flow algorithm Dα has the best prediction result, which indicates the validity of incorporating fracture flow in a physically-based model. |
