地動值變異性的大小會對地震危害度分析的結果造成明顯影響。因此,如何獲得正確的地動預估式隨機誤差的變異性是目前面臨的重要課題。台灣地震資料相當豐富,是做地震相關研究最適合的地點之一。本研究更新、改進過去國內隱沒帶地震的地動預估式,建立隱沒帶板塊界面型及內部型地震的地動預估式。另外,選擇單站做分析的目的在於避開場址項對非地動隨機誤差的影響,期能有效掌握地動值的隨機變異性。 本研究資料取自氣象局自由場強地動觀測網計畫(TSMIP),從1991到2014年間地震矩規模大於4.0共計14,660筆強震資料。根據地震位置、深度及震源機制的判斷,將台灣東北部隱沒帶的地震分成界面型與內部型兩種,為了區別這兩種類型的地震之地動差異,本研究將這兩種類型的地震分開迴歸,建立各別的地動預估式。衰減模型參考Lin and Lee (2008)的衰減模型,增加規模平方項、連續型的場址參數及震源機制項,利用最大概似度法並搭配混合效應模型進行迴歸分析。 隱沒帶地震的地震波傳播時振幅衰減得慢,可以傳到較遠的距離。本研究分析結果顯示,相同規模、距離的情況下,板塊內部型地震的地動預估式估計之PGA值較板塊界面型地震地動預估式高,而地動預估式之標準差大小分別為0.502及0.567。選取記錄筆數大於50筆的44個測站做迴歸分析,建立個別之單站地動預估式,單站地動預估式的總標準差比全域地動預估式之總標準差約下降2%~37%,平均約下降22%。花蓮地區之單站地動預估式的總標準差相較全域地動預估式之總標準的降低程度較多,宜蘭地區最少,可能來自其它無法解釋的震源及路徑效應。;The sigma (standard deviation) of ground-motion prediction equation (GMPE) has great impact on probabilistic seismic hazard analysis (PSHA). Therefore, how to properly evaluate the sigma has been a crucial issue currently. There are abundant earthquake data in Taiwan, and it is one of the most earthquake-related research suitable sites. In this study, we update the ground-motion relationship for north eastern Taiwan subduction zone earthquakes by establishing new sets of interface and intraslab earthquakes GMPE. Moreover, single-station GMPEs are done for the purpose of reduction of the sigma due to elimination of the variability from site effect. In this study, ground-motion data of subduction zone for both interface and intraslab earthquakes are obtained from the Taiwan Strong-Motion Instrumentation Program (TSMIP). A total of 175 earthquakes and 14,660 records with moment magnitude greater than 4.0 are selected to establish PGA attenuation relationship. We chose the general usage of the functional forms by reviewing of previous studies. Each candidate term in the form was tested with Taiwan data set. The final form is generally similar to the form proposed by Lin and Lee (2008), besides a quadratic magnitude term, a VS30 term and a focal mechanism term were added. The coefficients of the equation are determined through non-linear regression analysis using maximum likelihood method (MLE) and mixed-effects model. Generally speaking, ground-motion from subduction zone earthquakes decay slower and may spread longer distances. The results show that intraslab earthquakes generally predict higher PGA than that of interface earthquakes. Both regional GMPE and 44 single-station GMPEs are done in this study. Comparing the sigma of regional GMPE and single-station GMPEs, the total sigmas of single-station GMPE are smaller than the regional sigma with a reduction rate from 1.5% to 37.0%, averaging 21.7%. In a comparison of the reduction rate between different regions, the evaluated from stations in Hualien have highest reduction rate, whereas that evaluated from stations in Ilan are the lowest. Such a difference might be due to different source and path effects.