基於TRIGRS之邊坡可靠度評估-以薩克亞金溪集水區為例

近年來極端降雨事件頻繁發生，在全球氣候變遷的影響之下，可預期邊坡崩塌發生頻率及崩塌規模將增加，所造成之坡地災害風險亦增加，為了能在有限的資源下辨識高風險地區，故建置廣域崩塌潛勢分析模型。傳統邊坡穩定性評估多假設參數為單一數值，採用定值法進行分析，以安全係數考量各參數對邊坡穩定性的影響；然而實際上因分析模式、資料量測之不確定性及自然環境之隨機性，使參數具有不確定性。故本研究將以機率模式取代定率模式，藉由可靠度分析降低安全係數無法反應系統不確定性之缺點，並辨別邊坡失效機率較高的區位、評估可能的崩塌規模，藉以了解整體邊坡穩定性的變化。在此研究中以石門水庫上游的薩克亞金溪集水區為例，基於美國地質調查所(USGS)開發之暫態降雨入滲邊坡網格穩定分析模式(TRIGRS)建立集水區崩塌潛勢模型，其中利用最佳化技術，以粒子群演算法(PSO)率定土壤參數。比較兩場降雨事件的模擬結果與實際崩塌資料後，山崩正確率皆大於70%、AUC值亦皆大於0.8，說明此模型的正確性符合工程接受範圍，且能良好預測降雨誘發之山崩區位。經由敏感度分析，確認土壤凝聚力、摩擦角及飽和水力傳導係數之不確定性具有關鍵影響，並以失效機率評估降雨誘發邊坡可靠度的變化。在此研究設計之兩個不同變異度的案例中，皆證實一階二次矩法(FOSM)中泰勒展開式以平均值為展開點且以0.1倍平均值為參數變動量計算偏導數之方法可應用於估算失效機率。一階二次矩法推估之失效機率與蒙地卡羅模擬(MCS)結果的相對誤差只有在高變異度的案例中坡度約25°處在降雨初期相對誤差可能增加至30%，其餘網格在降雨期間失效機率的相對誤差多在20%以內。由於FOSM分析效率高於MCS，故結合FOSM與TRIGRS評估降雨期間邊坡失效機率的變化。此外，定率模式中安全係數小於1視為邊坡破壞，但從模擬結果可知安全係數皆為1的網格可能會有不同的失效機率值，故機率模式相較於定率模式能較好地反應邊坡穩定性。研究中利用NASA全球衛星觀測降雨計畫(GPM)的降雨數據推估現況2、5、10、25、50、100年重現期降雨下的邊坡失效機率。薩克亞金溪集水區中，高崩塌潛勢區多為坡度30°~50°之邊坡，且當中越陡峭的邊坡越易受到降雨誘發而破壞，而較緩的邊坡則可忍受較高強度的降雨。此外，隨著重現期距加大，邊坡失效機率將增加，以低變異度現況為例，100年重現期失效機率大於0.6之累積面積比2年重現期增加約1.36%；雖然不同變異度推估之機率值不同，但皆可應用FOSM評估邊坡穩定性、判別相對高潛勢之區位。藉由GPM繁衍之基期及近未來氣象資料，比較兩者在各重現期下的失效機率差異，評估氣候變遷對薩克亞金溪集水區邊坡穩定性的影響。結果顯示高失效機率的邊坡區位與現況大致相符，且可預期邊坡崩塌有變嚴重之趨勢，但在短期內(2021至2040年)差異不算劇烈。此外，研究中以邊坡失效機率產製之崩塌潛勢圖可作為氣候變遷風險分析之危害因子，其結果可供氣候變遷風險評估及治山防災參考之用。

關鍵字

TRIGRS ；邊坡穩定；不確定性；一階二次矩法；失效機率；氣候變遷

並列摘要

In recent years, extreme rainfall events have occurred recurrently. Under the influence of global climate change, an escalation in frequency and magnitude of slope collapse could be foreseen, risking an enlargement of possibility for a disaster. In order to distinguish the regions with higher risks with limited resources, a regional landslide susceptibility analysis model is built. Considering the uncertainty of the parameters, we will replace the deterministic model with the probabilistic model in this study, and use the reliability assessment to locate the region of high failure probability and also to estimate the possible collapse magnitude to understand the changes in the overall slope stability. Taking the Sakayachin creek watershed in the upper reaches of Shimen Reservoir as an example, Transient Rainfall Infiltration and Grid-Based Regional Slope-Stability Analysis (TRIGRS) is adopted to establish a landslide susceptibility model. The Particle Swarm Optimization (PSO) is used to calibrate soil parameters. Comparing the simulated results with the recorded landslide data of two rainfall events, the landslide prediction accuracy are greater than 70% and the AUC value are greater than 0.8. This indicates that the accuracy of this model is in line with the engineering acceptance range and can therefore, is used to predict the location of rainfall-induced landslides. Through the sensitivity analysis, soil cohesion, friction angle, and saturated hydraulic conductivity are identified as random variables to assess rainfall-induced changes in slope reliability. In the two cases with different degrees of variability designed in this study, it is confirmed that when Taylor's expansion uses the mean value as the expansion point and 0.1 times the mean value as the parameter variation to calculate the partial derivative, the relative error between the first-order second-moment method (FOSM) and Monte Carlo simulation (MCS) is within 20%, and the analysis efficiency of FOSM is higher than that of MCS. In addition, the factor of safety less than 1 is considered as slope failure in the deterministic model, but the simulation results indicate that the grid with factor of safety equal to 1 may have different failure probability values, so the probabilistic model can better reflect the slope stability compared with the deterministic model. In the study, data from NASA's Global Precipitation Measurement (GPM) is used to estimate the slope failure probability for the current rainfall return periods of 2, 5, 10, 25, 50, and 100 years, respectively. The results show that the high landslide potential areas are mostly with slopes of 30°~50°. In addition, as the rainfall return period increases, the slope failure probability becomes greater. Taking the 2-year return period and 100-year return period of low variability as examples, the cumulative percentage of area with failure probability greater than 0.6 increases by about 1.36%.Although the estimated probability values are different, applying FOSM to evaluate the stability of the slope in two cases with different variances can identify the area of relatively high potential. Observing the impact of climate change on the slope stability in the Sakayachin creek watershed, the region of high failure probability is generally consistent with the basiline situation. The study shows that slope failures are expected to become more severe, but the changes are not significant in the short term (2021 to 2040). Moreover, the landslide susceptibility map generated by the slope failure probability in this study can be utilized as a hazard factor for climate change risk analysis, and the results can be used as a reference for climate change risk assessment and disaster prevention.