由於科學家目前為止還不能成功預測地震，所以為了減少地震災害許多其他方法已經被提出，其中包括地震預警系統。雖然地震預警系統已經被認為是一個比較能有效減少地震災害的方法，但科學家也同時清楚沒有任何一套預警系統是完美的，意味著預警系統確實會產生誤報 (false alarm) 或應報未報 (missed alarm)的狀況。在本項研究計畫中，最主要的目的是對於地震預警系統的可靠度進行分析研究，進一步發展一套新的方法針對地震預警系統進行最佳化，藉由將兩種錯誤發生的機率降到最低。簡單來說，本計畫三個主要工作項目如下:1)根據機率統計的耦合理論(copula theory)去尋找對地震早期訊號(precursor)跟地震最大震動(peak motion)這兩個變數的聯合機率模型；2)根據此聯合機率模型去計算地震預警系統誤報及應報未報的機率；３)根據此機率進行系統優化並計算最佳化參數。 本研究案對於地震預警研究的貢獻可說是非常大的，尤其在於整個方法論的建立上依據統計學上的耦合理論。我們在建議書內提供兩個詳細算例；簡單來說，我們發現地震預警系統中使用的參數對於系統的可靠度影響相當大，凸顯了這些參數 ;It is a fact that we still cannot predict earthquakes with our current knowledge/technology as we speak, and therefore a variety of alternatives were proposed for seismic hazard mitigation, including earthquake early warning (EEW). Although EEW is considered a more practical and effective approach for seismic hazard mitigation, it is also a fact that not an early warning system is perfect without error, say, the occurrences of false alarm and missed alarm. In this research, we target on the reliability of on-site EEW, aiming to develop new decision-making criteria that can reduce its error as much as possible, and to optimize its performance with fewest missed and false alarms. The fundamental analytics used in this research are based on the copula theory in probability and statistics, which has been increasingly employed in engineering reliability assessments. As a result, the three key tasks/scopes of the research are as follows: i) calibrating copula-based joint probability functions for the precursor and peak motion of on-site EEW; ii) calculating the probability of missed alarm and false alarm based on the joint probability model, and iii) conducting optimization analyses in searching of the optimal warning threshold in association with the lowest probability/risk of missed alarm and false alarm combined. The contribution of the proposed research to EEW studies will be significant, establishing the new copula-based methodology for optimizing the performance of on-site EEW for the very first time. As the example calculation shown in the proposal, the reliability of an on-site EEW is fairly sensitive to the predetermined warning threshold, accentuating the importance of selecting a proper warning threshold in EEW, which can be resolved by the proposed method through the copula-based joint distribution model and the optimization computations.
