山崩引致之堰塞湖天然壩是常見的地質現象，因堰塞湖潰決時間通常很短，故如何快速分析堰塞湖的穩定性，從防災觀點而言相當重要。過去關於堰塞湖天然壩穩定性評估多利用壩高、壩體積、集水區面積、湖體積等單一因子或其組合所衍生之複合影響因子，進行單變量統計分析，以取得堰塞湖穩定性評估指標。然而，多因子間相互影響之效應即無法合理考慮。因此，本研究利用多變量統計方法篩選影響堰塞湖天然壩穩定性之重要因子，並建立預測堰塞湖穩定性之量化模式。 根據43筆日本堰塞湖資料，利用判別分析及羅吉斯迴歸兩種方法，建立包含尖峰流量、壩高、壩長及壩寬四個參數之統計模型，模式適合度達顯著，判釋能力相當高。另外，由於同時擁有尖峰流量、壩長與壩寬之堰塞湖驗證資料量相當不足，因此利用集水區面積替換尖峰流量，壩體積取代壩長及壩寬所建立的統計模型來預測84筆全球堰塞湖穩定性，判別分析預測正確率為78.6%，而羅吉斯迴歸預測正確率為66.7%。不論根據判別分析標準化函數係數或逐一惕除影響參數後觀察羅吉斯回歸模型ROC曲線下面積變化之方法，均得到堰塞湖尖峰流量(與集水區面積)為影響天然壩壩體穩定性之最重要參數。而最後利用兩組地形參數組合進行壽命的多變量迴歸，並對中國四川唐家山堰塞湖進行壽命預估，發現預估壽命為40天，與實際結果(30天之後)相去不遠。 Landslide damming a creek is a common phenomenon. According to the historical cases, the landslide dam usually breaks down in a short time after the formation of the lake. Therefore, evaluating the stability of the landslide dam quickly is essential from the hazard mitigation viewpoint. Dam height, dam volume, watershed area and lake volume are identified as important factors influencing the stability of landslide dam. However, systematic evaluating the dependency and significance of the influencing factors is required. This research utilized multivariate statistical method to screen out the significant variables, including peak flow, dam height, length and width, affected the stability of landslide dam. Based on discriminant analysis and logistic regression, the significance level of the proposed statistical model is acceptable. Since the watershed area is more easier to obtain than the peak flow rate and the lack of dam length and width, an alternative statistical model is proposed using watershed area, dam height, volume as influencing factors. We forecast 84 landslide dams worldwide and The forecast total accuracy rate of statistical model is 78.6% based on discriminant analysis and 66.7% based on logistic regression. Among the selected variables, the peak flow (or watershed area alternatively), is the most important geomorphological parameter for the stability of landslide dam. Finally, multiple regression is adopted to evaluate the longevity of landslide dam. Cases of landslide dams in Taiwan and China are used to verify the proposed model.
