本研究建立極端雨量下廣域山崩即時預報系統，運用美國地質調查所開發之TRIGRS定值法山崩潛勢分析程式，整合相關輸入參數以分析降雨入滲時邊坡土層即時孔隙水壓激發狀態及對應之安全係數，並透過即時動態展示圖形介面，明瞭暴雨事件過境期間所處關心區域之即時山崩潛勢及警戒指標動態情形，協助防救災決策之用。整體山崩預報核心為TRIGRS山崩潛勢分析結果，其版本與邊界模式之使用正確性將影響山崩預報可信度，故須釐清TRIGRS版本差異、參數設定，與模式適用情形及準確率。在水文地質輸入參數部分，基於土層地質隨機分布性及地質鑽探不足等因素，致使難以透過地質鑽探或現地試驗來獲取具有代表性之水文地質參數。因而整合遺傳演算法之最佳化技術耦合TRIGRS模式進行分析參數反算，並進行逆分析參數研究，試圖尋找較合適參數組合，供下一場颱風事件來臨時正算預測使用。 經重新推導與勘誤TRIGRS解析解，釐清理論對應不同版本程式差異，系統採取TRIGRS 2008飽和有限邊界模式作為山崩潛勢預報核心。最佳化參數反算分析之參數研究中，在桃芝颱風率定與下一場敏督利颱風事件驗證下，最佳化反算分區設定為地質分區，引入擴散係數改良與反算後凝聚力修正之最佳化模組可產製具防災實務能力之水文地質參數圖層。本研究成功建置廣域即時山崩預報系統，整合上述研究結果並可運行即時雨量與莫拉克颱風事件下小林村研究區域之自動山崩潛勢動態分析展示，可初步供防災機構參考使用。 ;This study establishes a real-time forecasting system for regional landslide hazard assessment under extremely rainfall events. The system uses TRIGRS to evaluate the factor of safety of regional slopes in rainfall processes and to show variations of real-time susceptibility through ArcGIS platform. It can help researchers to realize susceptibility and alert indexes of the assigned slope and to make decisions of disaster prevention and mitigation. TRIGRS model is the core of the landslide prediction system. Different versions and boundary settings of TRIGRS influence the credibility of forecast. It’s necessary to clarify the difference of versions, parameter settings, applicable conditions, and precision. The hydrologic and mechanical parameters of landslide model are difficult to evaluate from the existing experiment due to random distribution of soil composition and data deficiencies. This research developes the reverse optimization analysis model to acquire reasonable parameters for landslide forecast under next typhoon. The optimization model combines Real-coded Genetic Algorithm with TRIGRS. After rederiving the analytical solution of TRIGRS and clarifying the relation between theory and program, the system uses TRIGRS 2008 with initial wet condition and finite boundary setting as the core of landslide forecast. The results of the reverse optimization analysis show that the practical parameters layers provided by the module with geology divisions, the modification of diffusivity coefficient, and final cohesion increase. The real-time forecasting system for regional landslide hazard assessment under extremely rainfall events has successfully established. It can operate completely automated landslide susceptibility analysis and visual display for Xiaolin village region under Morakot typhoon. It’s useful for disaster prevention unit.
