| 摘要: | 本研究建立適合台灣地區機率式地震危害度分析使用的強地動預估式，以中央氣象局的自由場強地動觀測網計畫之強震資料，篩選出地震矩規模大於4.0之150 個地殼地震，共計19,887筆強震資料進行分析。以混合效應模型(mixed-effect model)及最大概似法(maximum likelihood estimation)當作迴歸模型，透過非線性的迴歸分析來探討地動值與震源、路徑及場址間的關係，並以地震學理論為基礎建立預估式，考慮規模、地震波傳播距離、場址特性及震源機制等參數，完成適合台灣地區在各週期之地殼地震強地動預估式。
本研究利用變動視窗法(Spatial-correlation Mobile Window)、路徑玫瑰圖法(Pah Diagram)、半變異圖法(Semi-variogram)、CI法及震央距離法等方式進行事件間變異性與記錄間變異性的分析與拆解。在分析事件間變異性上，以變動視窗法最為合理，而紀錄間變異性的拆解則以路徑玫瑰圖有最好的結果，並且是唯一可以將路徑間變異性從紀錄間變異性中完整拆解出來的方法，也使的每筆強震紀錄都可以擁有各自的事件內自然隨機誤差。最後合併變動視窗法與路徑玫瑰圖法求得最小之單一路徑標準差σSP，可較總標準差σT降低約40%-55%。
另外也建立特定條件之強地動關係式，其中包含單站(single station)、單震源(single source)及單震源對陣列(single source to an array)的條件。結果發現，使用特定條件之預估式求得之單站標準差σSS、單一路徑標準差σSP及事件內的自然隨機變異性σ0皆比一般預估式的拆解更低。最後在機率式地震危害度分析中套用總標準差、單站標準差及單一路徑標準差，結果顯示只考慮自然隨機誤差的的危害度曲線(2475年再現期)，比過去只考慮總標準差評估出的強地動值降低約20%，再現期越長評估出的強地動值差異也越大。
;In this study, we use 19,887 records for 150 crustal earthquakes with moment magnitudes greater than 4.0 obtained from the Taiwan Strong-Motion Instrumentation Program network to build the Taiwan ground-motion prediction equations (GMPEs) for peak ground acceleration and spectral accelerations. The nonlinear regression analysis of ground-motion prediction model is the mixed-effect model with maximum likelihood method. Though this regression analysis to discuss the relationship of source, path, and site.
This paper describes the approaches for the presentation of the components of the error in ground-motion estimates for future earthquakes: (1) spatial-correlation mobile widow, (2)path diagram, (3) semi-variogram, (4) closeness index and (5) the distance of epicenter. Comparing the results with those obtained with the same data, but using the closeness index, semi-variogram and the distance of epicenter approaches, show that we get a lower path-to-path sigma with the combination of the spatial-correlation mobile window and the path diagram methods. For peak ground acceleration and spectral accelerations at periods of 0.3 s, 1.0 s, and 3.0 s, the path-to-path standard deviations obtained in the new approaches are 40%–55% smaller than the total standard deviation.
We also set up the ground-motion prediction equations for the single station, single source and single source to an array in this study. When we use these specific conditions GMPEs to analyze the variance, we can obtain the smaller single-station sigma, single-path sigma, and intra-event aleatory variability than general GMPEs. If we only use aleatory variability in PSHA, then the resultant hazard level would be 20% lower than the traditional one in 2475 year. |
