|Abstract: ||本研究分為兩個主要重點，第一部分為探討強地動特徵，對於2011年日本宮城地震之加速度強地動記錄，運用基於經驗模態分解（EMD，Huang et al., 1998）法所建構之基線修正新方法，還原同震位移時間序列，經由與高頻GPS記錄之比對發現，使用此新方法時需至少以附近區域之同震變形量為參考，改善修正之精確度。接著運用2010年紐西蘭達菲地震序列及1999年臺灣集集地震之強地動記錄，了解非線性現象之時間回復性（Aguirre and Irikura, 1997）、計算非線性度（DNL）以定量探討非線性現象之強弱，結果顯示0.5至10 Hz較適合計算DNL，並應用此二地震之經驗於2008年中國汶川地震，進行DNL與場址及液化區域之比對。|
第二部分本研究運用隨機式點震源（Stochastic Point Source，Boore, 1983; Boore, 2003a）模擬，由臺灣地區強地動觀測計劃（TSMIP）之淺源、小規模觀測記錄，建構臺北盆地及桃園、新竹、苗栗、台中及南投地區（TAP及TCU地區）之經驗轉換函數（ETF），接著對於同樣淺源之地震完成良好之場址修正，降低高頻模擬之時間域及頻率域誤差，並與經過場址修正之地動預估模式（GMPE，Jean et al., 2006; 張毓文等，2010; 林柏伸，2009）達到相同水準。對於大規模地震如1999年集集地震，ETF之場址修正對於隨機式有限斷層（Stochastic Finite Fault，Beresnev and Atkinson, 1998; Motezedian and Atkinson, 2005; Boore, 2009）模擬能有效降低模擬之誤差，提供可信之頻譜，PGA之結果則略優於經過場址修正之GMPE預估。而經由本研究對於近斷層效應之討論發現，若震源區能根據地質調查、歷史地震調查合理給定斷層長寬時，使用隨機滑移量分佈之平均能提供可信之參考結果，而隨機式模擬結果未來仍需對於上盤效應、破裂方向性效應進行額外修正，以降低模擬方法之誤差。
;The important point of this study included two parts. Characteristics of strong motions were discussed in the first part. The Empirical Mode Decomposition (EMD) based new baseline correction scheme was constructed. Acceleration strong motion records of 2011 great Tohoku Japan earthquake were corrected to reproduce coseismic displacement time histories. When applying this new scheme we found that the correction results at least need coseismic deformation value in the nearby region as constrain to improve the accuracy after comparing with high frequency GPS records. Than the ground motions of 2010 Darfield New Zealand earthquake sequence and 1999 ChiChi Taiwan earthquake were analyzed to identify characteristics of time recuperated nonlinear site response (Aguirre and Irikura, 1997), and using Degree of Nonlinearity (DNL) to quantitatively discuss nonlinear site response. The DNL results showed the 0.5 to 10 Hz frequency band was suitable to calculate its DNL value. And this experience also applied to the 2008 Wenchuan China earthquake to discuss the relations between DNL and site class and liquefaction area.
Second part, stochastic point source simulation (Boore, 1983; Boore, 2003a) was first applied for strong motion stations of the Taiwan Strong Motion Instrumentation Program (TSMIP) in TAP and TCU region. This study selected shallow, small magnitude earthquake as database to construct Empirical Transfer Function (ETF) for each strong motion stations. Afterwards, ETFs were used to do the site corrections for stochastic simulation of the target earthquakes. Results showed site correction works well that it reduced errors in PGA and frequency band for high frequency simulations of shallow earthquakes. The uncertainty of PGA could reach the same level with that from Ground Motion Prediction Equations (GMPEs, Jean et al., 2006; Chang et al., 2010; Lin，2009) which had considered site correction. For stochastic finite fault (Beresnev and Atkinson，1998; Motezedian and Atkinson，2005; Boore，2009) simulation, ETF also works well and reducing errors. The correction results provide believable frequency spectrum and PGA slightly better than prediction from GMPE. On the other hand, after the discussion of near source effect for stochastic simulation in this study, the results showed if fault region including length and width could reasonably decided following geological survey or historical earthquake investigation, average of many random asperity distribution models could provide believable simulation. In the future, stochastic simulations still need to consider hanging wall effect, directivity effect to reduce errors for simulations.
Finally, four kinds of common transfer functions for site effect study including H/V ratio of S-wave (HV), spectral ratio method between soil to rock station pair (HH), microtremor H/V (MicroHV) and ETF were compared in the Taipei basin. The results indicated transfer function calculated from single station method (H/V for S-wave and microtremor) had good agreement with each other. Single station method reacted between basement A and surface for B class stations but reacted between basement B and surface for D and E classes. Finally, microtremor soil to rock spectral ratio (MHH) was tested to considered alternative transfer functions for those sites who lack of strong motion observation region in this study. After comparing with HH, it indicated the distance to reference rock should less than 10 to 15 km, and its reliable frequency band up to 2 Hz.