天然壩之穩定性與堰塞湖壽命一向為堰塞湖防、減災關鍵議題。天然壩穩定性之評估已累積相當多研究成果，然而，堰塞湖壽命之預測討論較少。本研究利用統計模型，嘗試探討堰塞湖壽命與各影響因子間之關係。根據日本65筆堰塞湖天然壩的案例、區分成因(46個案例為地滑形成之堰塞湖、15個案例為土石流形成之堰塞湖)與誘因(豪雨誘發案例39筆、地震誘發案例26筆)之案例以及同集水區、同豪雨事件誘發(十津川地區案例21筆)之案例，分別進行因子與壽命相關性分析，以篩選出與堰塞湖壽命關聯性較高之因子，利用篩選出之因子進行多變量迴歸分析，建立預測堰塞湖壽命統計模型，並利用台灣草嶺及小林村堰塞湖案例驗證預測模型。相關性分析結果發現，影響堰塞湖壽命之主要因子包括：集水區面積、湖(壩)體積、湖(壩)高及壩寬。不同成因及誘因之堰塞湖案例，各因子對其壽命影響之重要性略有不同。基本上，豪雨誘發之堰塞湖，其集水區面積與壽命呈負相關；另外，不論區分成因與誘因與否，湖(壩)高及湖(壩)體積均為與堰塞湖壽命呈正相關之重要因子，顯示湖(壩)高越高，湖(壩)體積越大，堰塞湖蓄水時間越長，堰塞湖壽命即越長。然而壩高對天然壩穩定性之影響則呈負相關，此一差異顯示影響天然壩穩定性與堰塞湖壽命之因素可能不盡相同。值得注意的是，不論區分成因與誘因與否，排除因子相依共線性問題之統計模型其決定係數R2均不高，唯有採同集水區、同豪雨事件誘發案例之統計模型結果尚可接受(R2為0.706)，顯示堰塞湖統計模型建立時，應考慮降雨事件及地形、地質條件均質性。綜合而言，豪雨誘發之堰塞湖壽命與溢流時間有關，而溢流時間取決於降雨特性，然而此一效應不易納入統計模型。地震誘發之天然壩形成後，可造成潰壩的降雨事件週期更具有高度之不確定性，因此，其壽命預測困難度明顯更高。 The stability and longevity of landslide dams have been a central theme in hazard assessments. In this research, statistical model is used to establish the relationship between longevity of landslide dam and geometric parameters, according to 65 cases of landslide dams in Japan. The correlation analysis is used to screen out the significant variables. Multiple regression analysis is adopted to evaluate the longevity of landslide dam. The landslide dams oriented from sliding (46 cases) and debris flow (15 cases), as well as those oriented from rainfall (39 cases) and earthquake (26 cases), are analyzed separately. The two cases of Taiwan are used to validate the statistical models. The results show that the lake (dam) height and lake (dam) volume are the most relevant factors to the longevity of landslide dams. Multiple regression analysis results indicated low coefficient of determination (<0.3) of the statistical models. Only the prediction model from the landslide dams triggered in the same event and in the same catchment (21 cases) yields an acceptable coefficient of determination (0.706). The results indicate that to establish statistical model for predicting the longevity of landslide dam the similarity of triggering event and the geological characteristics should be consider. Finally, the triggering event, which could strongly dominate the longevity of rainfall-induced landslide dam, is difficult to be incorporated into the statistical model. Since the return period of the event to induce the overtopping and breach the landslide dam is difficult to be predicted, the prediction of the longevity of the earthquake-induced landslide dam from statistical approach is challenge.
