機率土壤液化潛勢圖之建構與應用

Abstract

2011年3月11日，日本發生東北地方太平洋近海地震，與其伴隨而來的海嘯與餘震所引發的大規模災害，官方將其稱為東日本大震災。其中，地震所引發的土壤液化為史上規模最大，甚至距離震央三、四百公里之遠的東京附近地區，至今仍有部分液化區仍在整建。臺北盆地是台灣的行政與經濟重心，且與東京均位於環太平洋地震帶，災害防範是十分重要的課題。 由於臺北盆地地質環境特殊，大半面積均存在土壤液化的可能，而土壤液化潛勢圖能夠直觀的反應可能的液化情形，且解讀容易與快速，因此廣泛的使用在都市與防災規劃中。 許多自然災害發生的可能性是一種不確定性的表現，土壤液化的發生受地震、土壤特性與評估模式不確定性的影響，本質是機率的，而非「會」與「不會」二元式簡單判定。土壤液化潛勢圖是由該地之地質鑽探資料、地震資料與地下水位資料計算而得，直觀上應該具有一定程度的隨機性。傳統上，土壤液化潛勢圖為定率分析，無法展現出其隨機性。因此，本研究著重在土壤液化潛勢圖的改良，先是決定地質鑽探資料之中，土壤參數的不確定性；再使用加噪多變數經驗模態分解法，處理地下水觀測資料之時間序列，獲得其本質模態函數與特徵時間尺度，進而使用人工神經網路預測其未來趨勢，是為NAMEMD-ANN預測模型，並針對特徵時間尺度評估其不確定性。最後，利用擾動差法，將定率土壤液化潛勢圖轉換為機率土壤液化潛勢圖，並計算出台北每一行政區具有液化風險之面積比例，與其正、負一標準差情況下之差異比例。 於實際應用上，本研究除了提供機率土壤液化潛勢圖的建立方法，更可以對其進行預測，得到不同時間、不同機率的土壤液化潛勢圖，期盼能為台灣土壤液化之防災規劃、政策決定與相關研究帶來貢獻。

The primary contribution of this paper involves the development of a probabilistic soil liquefaction potential mapping in Taipei area using uncertainty analysis and combining several different methods, i.e. the Hyperbolic Function Method (HBF), the Artificial Neural Network (ANN) model, the Noise-assisted Multivariate Empirical Mode Decomposition (NAMEMD) algorithm, and the Perturbance Moment Method (PMM). The Hyperbolic Function Method (HBF) is employed to evaluate the soil liquefaction potential as its equations are simple, it is constructed using the Taiwan database and widely adopted by the government in Taiwan. Moreover, the Artificial Neural Network (ANN) model coupled with the Noise-assisted Multivariate Empirical Mode Decomposition (NAMEMD) algorithm is proposed for analyzing and forecasting the ground water level. First, based on the previous research work and experiments, the geotechnical data and earthquake data are transformed into random variables. Then, the proposed NAMEMD-ANN model is applied to the groundwater level data to investigate the characteristic time scales, and the forecasting time and the analysis length can be determined by the characteristic time scales. Furthermore, the Pertubances Moment Method (PMM) and Monte Carlo methods (MC) are used to assess the statistical moments and probability of output, the Liquefaction Potential Index (LPI). Finally, the probabilistic soil liquefaction potential mapping can be plotted by Kriging. The proposed modeling framework was applied to the Taipei Basin and the results which include the LPI mapping, the contour of exceedance probability, the percentage of areas with a high probability to liquefy in different period are demonstrated and discussed in this paper.