| 摘要: | 崩塌發生主要由動態的誘發因子以及靜態的潛在因子所控制。降雨是誘發崩塌最主要的因子，過去研究常以降雨參數如降雨強度、降雨延時、累積雨量等作為誘因，但降雨型崩塌的發生機制更需要廣泛考慮坡地水文關係如降雨入滲後所造成的土壤含水量變化。根據許多研究指出，崩塌發生須達到足夠的土壤含水量以及含水量在土層中須滿足特定的分布情形，尤其是中層土壤(風化層或崩積層)含水量。
本研究應用三層筒狀模式(three-layer tank model)來納入降雨所計算出的土壤雨量指數(Soil Water Index, SWI)以及土壤雨量比(Soil Water Ratio, SWR)，用以概念性地表示土壤含水量以及其在土層中的分布情形。由水保局的重大土石災害報告篩選出SWI以及SWR2 (第二筒土壤雨量比)，並以颱風事件期間各自的最大值SWImax以及SWR2max為本研究的誘因代表；臨界降雨模式(critical rainfall model)考慮多種潛因所計算出的臨界降雨Qcr作為本研究的潛因代表，其可表示以力學為基礎所評估的邊坡穩定性。本研究應用邏輯斯迴歸整合SWImax、SWR2max以及Qcr，並以4場不同的颱風事件於臺灣南部的高屏溪流域所引發的新生崩塌地為例，包含敏督利颱風(2004)、海棠颱風(2005)、卡玫基颱風(2008)以及莫拉克颱風(2009)。整合模式在修正成功率(Modified Success Rate, MSR)皆表現良好，最高者為卡玫基颱風80.5%，最低者為莫拉克颱風76.2%，而非崩塌成功率優於崩塌成功率，因此土壤含水量的相關參數適合作為崩塌預測的誘因。由刀切法(Jack-knife method)分析可知，整合模式隨著事件的變化去調整誘因和潛因的相對貢獻程度，使其互補且平衡以達到最佳預測成果。然而相互驗證成果顯示不同事件與模式可能存在通用性問題，本研究結合其中3場事件所訓練整合模式獨立驗證於第4場事件，雖然MSR比起目標事件本身所訓練整合模式略差以及可能出現崩塌被高估或低估情形，但仍表現不錯，最高者為敏督利颱風76.4%，最低者為海棠颱風70.9%，表示利用多場已知事件的資料去預測未知事件的崩塌是具可行性的，亦可取得一較為通用的整合模式。綜上所述，本研究考慮土壤含水量以及邊坡穩定性並發展出一全新的崩塌潛勢預測模式，不僅預測成果良好，亦具有應用性。;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. |
