基於物理與統計模式氣候變遷下崩積層之邊坡風險分析

Abstract

許多研究指出在氣候變遷影響下，極端氣候事件發生的趨勢是增加的，這將使得坡地災害發生頻率增加，因鮮少研究以邊坡位移作為分析的指標，故本研究以新北市臺9甲線10.2K上邊坡之土石流潛勢溪流區域分析氣候變遷情境下邊坡風險。分析時依據過去青山工程顧問股份有限公司(青山顧問)監測的資料，SB-6的孔位位移甚大，故針對SB-6的孔位的邊坡來進行模擬與分析。2017年底青山顧問設置橫向排水工減輕坡地重量減少崩塌可能性，而在2018年6月以後地下水高降至最低，本研究將基於2018年6月至12月的資料分析氣候變遷之未來情境下的邊坡位移情況。首先利用ABAQUS物理模式進行分析，然而有限元素分析較為耗時費力，故使用倒傳遞類神經網路(Back Propagation Neural Network, BPNN)取代ABAQUS提高模擬的效率，並使用粒子群演算法(Particle Swarm Optimization, PSO)將土壤參數校正為最接近觀測的資料，後續則能利用校正後的土壤參數進行氣候變遷之未來情境的模擬分析。然而因為崩積層組成太過複雜，視為均質來校正誤差太大，發現將土壤層分成22個深度，其校正結果較為良好。另外，因為輸入因子對BPNN有所影響，所以設定3個BPNN輸入模式。3個模式中，第2種模式與第3種模式在氣候變遷之未來情境分析雨量與邊坡位移、地下水高、FS值的關聯性不錯，以及位移跟FS值的關係也合理，特別是第3種模式失效機率的計算也因為較符合觀測情形，較適合做氣候變遷之未來情境的預警模式，所以第3種模式是比較理想的模式。然而無論採用何種模式，氣候變遷之未來情境邊坡位移都比2018年高，但邊坡位移增加的量並不大，不如2017年下半的數值是大於行動值的，而是降至預警值，可推論氣候變遷之未來情境該處崩積層在自然狀況，發生立即嚴重崩塌災害的可能性不高。

Many studies discussed the effects of climate change on increasing extreme climate events, resulting in rising the frequency of slope disasters. Few studies used slope displacements as an indicators for evaluation under climate change. Thus, this study will focus on analyzing the impact of slope displacements under the climate change scenarios on a potential debris flow stream at provincial highway No. 9A at the 10.2-kilometer in New Taipei City. A few years ago, Land Engineering Consultants Co., Ltd. has monitored this region and found that the colluvium at the SB-6 position at the stream bank has large displacements. Therefore, the study will focus on simulating the displacement at SB-6 position in the period of June to December of 2018. ABAQUS, a physical model, is first used for analysis. Nevertheless, the analysis of finite elements is time-consuming, Back Propagation Neural Network (BPNN) is adopted to replace ABAQUS to improve the efficiency. Particle Swarm Optimization (PSO) is used to adjust soil parameters according to the observation. The adjusted soil model can be used to simulate the future scenarios under climate change. The colluvium is divided into 22 depths in simulation because assuming colluvium is homogeneous does not obtain good results. The prediction of BPNN relies on input factors, in light of this, three models are built using different input factors. In three models, the second model and third model can obtain better relationship between rainfall and slope displacements, rainfall and water-table height, and ranfall and FS value in the future scenarios from climate change, and the relationship between the FS value and slope displacements in the third model is more reasonable. Besides, the failure probability of the third model is suitable for early warning in future scenarios from climate change because the result from third model is the closest one from observation. To sum up, the third model is better than other models. However, no matter which models are the best, slope displacements in the future scenarios from climate change are higher than in 2018 averagely. The amount of slope displacements increasing is not large, and it will not be larger than displacements observed in 2017. It can be inferred that the chance of having serious slope failure at this colluvium in a natural condition of future scenario is quite low.