| 摘要: | 由於地震造成的破裂與異常前兆的產生機制，目前尚未釐清，因此地震前兆相關的研究一直以來備受爭議。本論文的目的，在於驗證地電場的震前異常與地震的關聯性。研究主要從兩方面著手進行，以檢驗與解釋地電地震前兆的可能性。一是從台灣地電觀測網(GeoElectric Monitoring System, GEMS)的野外資料作有系統的分析，另一則是建立震電模型作解析與數值分析。
首先，從GEMS 野外資料分析開始，本論文建立一套包含預測模型以及二元分類模型的演算法，以檢查地電場的震前異常與地震的關聯性。此演算法稱作GEMS’ Times of Increased Probability (GEMSTIP)演算法，一開始是由Chen and Chen [Nat. Hazards., 84, 2016]提出。本論文會(1)減少原始版本的模型參數，以及(2)引入聯合測站法作改進，以便得到穩健的分析結果。這個新版的GEMSTIP 方法結合莫爾昌誤差圖表法(Molchan Error Diagram, MED)，來評估每組模型參數的預測表現值。另外，利用GEMSTIP 方法，
分析不同截止頻率的高通與低通濾波的地電場資料，還能夠決定出高訊噪比、與地震相關的地電場訊號頻段。除此之外，根據MED 推導出的統計顯著性檢驗的分析指出，潛藏的震電關係是客觀存在的，而且利用這個震電關係可以建立出基於地電異常的地震機率預報。
第二部分，則從室內岩石破壞實驗的物理角度出發，本論文提出一個完整、有自洽性的物理數學模型，稱作Chen-Ouillon-Sornette (COS)震電模型。這個COS 模型本質上結合伯里奇-諾波夫(Burridge-Knopoff，簡稱BK)彈簧塊體的力學系統與RLC 電路組成的電動系統。而在力學與電學之間的耦合，則是以應力激發電壓(Stress-Induced Voltage)的形式表現。BK 滑塊系統可以模擬每一個塊體的應力狀態與破裂事件的滑移量，而應力則會產生電壓變化，使得RLC 電路能進一步模擬電荷的產生與傳遞。COS 模型提供一般化的理論框架，以便模擬與分析地電與地震之間的關係。尤其是，它能夠再現出，震前觀測到許多次的電磁訊號的單峰脈衝(Unipolar Pulses)，也可以解釋在地電場訊號觀測到的統計量的異常與相變，例如Chen and Chen [Nat. Hazards, 84, 2016]與Chen et al. [Terr. Atmos. Ocean. Sci., 28, 2017]提出在震前的地電場偏度與峰度異常，以及埃夫塔克西亞斯研究團隊[Eftaxias et al., Nat. Hazards Earth Syst. Sci., 3, 2003]提出在震前的地電場功率譜的冪律指數變化。總結來說，根據台灣地電觀測網的野外資料分析，以及理論震電模型的解析與數值分析，本論文有力地支持地電地震前兆的存在，並認為以地電異常作地震機率預報是有可行性的，如此本論文將為地震前兆研究奠定重要的基石。;Due to the lack of the physical mechanisms of rupture precursors, studies of earthquake precursors are still debated and skeptical. The purpose of this thesis is to verify correlations between pre-seismic anomalies of geoelectric fields and earthquakes. The thesis includes two
main components to examine and explain the geoelectric precursors to large earthquakes: the first is the field data analyses observed from Taiwan Geoelectric Monitoring System (GEMS), and the second is the analytical and numerical analyses of a seismo-electric model.
Beginning with field data analysis, we examine the precursory behavior of geoelectric data with respect to large earthquakes by means of an algorithm including a predictive model and binary classification. This algorithm is dubbed as GEMS’ Times of Increased Probability (GEMSTIP), introduced originally by Chen and Chen [Nat. Hazards., 84, 2016]. In the thesis, we improve the GEMSTIP model’s robustness (i) by removing a time parameter of coarsegraining in the foregoing paper, and (ii) by introducing joint stations method instead of single
station method. Moreover, the GEMSTIP algorithm includes Molchan Error Diagram (MED) to evaluate the performance of a model parameter set in a forecasting dataset. This improved GEMSTIP algorithm also analyzes a large number of high- and low-pass filtered datasets with different cutoff frequencies, determining the frequency bands, which were indefinite in previous works, of the earthquake-related signals with high signal-to-noise ratio for the geoelectric data. Based on significance tests derived from MED, the underlying pattern of seismo-electric
relationship is objectively thought to exist. It is therefore appropriate for machine learning to extract this underlying relationship to establish earthquake probability forecasts.
In the second part, according to the observed physics from indoor experiements of rock fracturing tests, we introduce the first fully self-consistent model combining the seismic micro-ruptures occurring within a generalized Burridge-Knopoff spring-block model with the nucleation and propagation of electric charge pulses within a coupled electrokinetic system (an RLC circuit model). This model, coined as Chen-Ouillon-Sornette (COS) model, provides a
general theoretical framework for modeling and analyzing the relationships between geoelectric signals and earthquakes. In particular, it is able to reproduce the unipolar pulses that have often been reported before large seismic events, as well as various observed anomalies of the ambient electric field, such as pre-seismic skewness and kurtosis anomalies [Chen and Chen, Nat. Hazards, 84, 2016; Chen et al., Terr. Atmos. Ocean. Sci., 28, 2017], and pre-seismic power-law exponent variations of power spectral densities of geoelectric fields [Eftaxias et al., Nat. Hazards Earth Syst. Sci., 3, 2003]. In consequence, this thesis strongly supports the theory of seismo-electric precursors, and lays the foundations for earthquake probability forecasts. |
