近年山崩衍生之災害頻傳，使得山崩潛感分析成為熱門議題。山崩潛感分析可分為統計法與定率法，定率法較具物理意義，但要取得定率模型所需之力學與水力參數，則相當困難。透過逆分析，取得定率模型之參數則較為可行。然而，地質材料多具高度變異性，導致逆分析結果具不確定性。本研究以假設之案例( 共100個網格)，探討逆分析技術於山崩潛感分析定率模型參數取得之前景與限制。假設案例共使用三種參數空間分布型態(均質；各網格參數為分布相同；前述分布於各網格以隨機空間分佈型態存在)，以模擬各種不同參數分佈情形對逆分析之影響。藉由廣域邊坡穩定分析模型TRIGRS以及點估計法進行正分析，可獲得不同參數空間分佈假設下，各網格點之安全係數(或破壞機率)、山崩目錄以及水壓反應。逆分析時先假設全區為均質區，分別以山崩目錄、安全係數率定強度參數(凝聚力、摩擦角)與水力參數(水力傳導係數及水力擴散率)，而後再加入水壓反應資料，先率定水力參數，再利用山崩目錄或安全係數率定強度參數；其次考慮參數為常態或對數常態分布，但假設參數為定值，利用水壓反應與山崩目錄(或安全係數)率定水力與強度參數進行逆分析；最後，考慮參數異質性，利用水壓反應與山崩目錄(或破壞機率)率定水力與強度參數平均值與變異係數。結果顯示，若僅以山崩目錄，甚至利用安全係數或破壞機率進行參數率定，逆分析結果並不理想且非唯一。然而加入水壓反應先率定水力參數，再利用山崩目錄或安全係數進行強度參數逆分析，雖無法完全去除率定參數之非唯一性，但可有效地降低其非唯一性，且提升逆分析結果準確性。另外，若區域參數為一分布，但將其視為均質區進行逆分析，則逆分析所得參數與給定參數之平均值相距甚遠。最後，逆分析參數平均值與標準差之結果接近實際之參數分布，然非唯一性仍存在。由正算反應與各參數間相關性分析結果可知，山崩目錄及安全係數與水力參數之相關性偏低，此結果顯示僅使用山崩目錄(主要受強度參數與地形控制)欲反算水力參數並不合理。由相關性分析亦可發現水壓及山崩目錄與參數之相關性深受降雨強度與延時影響，故以定率法山崩潛感分析模式進行參數逆分析時，若能獲得不同時期(不同降雨強度與延時)之山崩目錄，則可有效地提升逆分析廣域參數之效能。 Landslide susceptibility analysis is crucial from hazard mitigation viewpoint. Statistical and deterministic approaches are frequently adopted for landslide susceptibility analysis. Based on physical models, deterministic approaches are superior to the statistical approaches for deterministic approaches fully taking the mechanical mechanisms into account. However, it is difficult to get the required hydraulic-mechanical parameters in a deterministic model. Back analysis is a promising way to calibrate the required parameters. Nevertheless, fewer researches really pay attention to discuss the accuracy of back analysis results. Therefore, this research uses hypothetical cases to evaluate the prospects and limitations of back analysis of regional hydraulic-mechanical parameters in a deterministic model. Three different spatial distribution types of hydraulic-mechanical parameters were assigned. Thereafter, landslide inventory, distribution of safety factor and failure probability, and pressure head of the hypothetical cases were calculated using a deterministic model, TRIGRS. These responses then used to calibrate the input parameters. The results show If we only use landslide inventory to calibrate (cohesion, friction angle, hydraulic conductivity and hydraulic diffusivity), the back calculation results which the best fit parameters are not unique and different from given parameters. The results also show if we can add hydrologic data to calibrate hydraulic parameters first, it can improve back analysis results and reduce non-uniqueness of calibrated parameters. From correlation analysis, we find the correlation coefficient between hydraulic parameters and landslide inventory is low and rainfall duration and intensity will affect it, so only use landslide inventory to calibrate hydraulic parameters is not reasonable. Calibration results can be further improved if we can obtain more event based landslide inventory maps (different intensity or duration) and combine to back analysis method.
