| 摘要: | 全球定位系統（GPS）已成為研究地殼變動及地體動力學的利器，當有較大地震發生時，GPS資料可以有效的用於震前、同震和震後變形的相關研究，配合近斷層的密集觀測結果，更可偵測活動中盲斷層位置及其變形，估算斷層滑移率，作為地震潛勢評估的重要依據。台灣西南部地區為目前正在活動的構造帶，近年來頻繁的地震活動說明此區正處於一個較為劇烈活動的時期。本研究將藉由GPS時間序列分析修正1993年?2002年嘉南地區的GPS速度場，根據該區的地質、地震和震測資料做模型假設，利用彈性半無限空間模型，推求台灣西南部地區可能的斷層幾何形貌和滑移速率，以期對此區的活動構造能有更進一步的瞭解。 疊加本研究中連續測站資料之功率譜結果顯示，GPS時間序列中的雜訊為「全頻等幅雜波+閃變雜波」的模式，所以本研究中使用此模式進行GPS時間序列分析。長週期的觀測資料之模型參數可能有下列幾項：線性速率、年週期及半年週期的變化量、因地震或更換天線所導致的不連續、地震後線性速率的變化及指數衰減的現象。利用最大可能估計法（MLE）代入模型中，重複疊代估計各參數直到殘差值在時間域中呈隨機分佈，表示修正完全。重新評估模型參數的結果顯示，誤差放大為三?四倍，東西、南北及垂直方向殘差之平均RMS值分別為3.2 mm、2.2 mm、9.7 mm，此值也反應了觀測之精度。而後利用時間序列分析所建立的方法，修正嘉南地區速度場從事進一步研究。水平地殼運動速度場顯示，以九芎坑－木屐寮－六甲斷層的線型位置為分界，由東到西有明顯速度變小的趨勢，在嘉南沿海一帶，其相對於澎湖白沙島的速度幾乎為零。左鎮斷層以南，速度場開始有向南偏轉並且明顯變大的現象；到高雄－屏東一帶速度已達到50 mm/yr以上，速度場的變化顯示了台灣西南部地區的構造的複雜性。 根據修正後的GPS速度場，假設兩種斷層模型進行逆推。第一種模型假設地下某深度存在一滑脫面，第二種模型沿滑脫面向上延伸另一個斷層至地表C利用網格搜尋及Bootstrap方法所得結果顯示，滑脫面上延伸出去的斷層出露位置約在九芎坑－木屐寮－六甲斷層的線型位置，但結果顯示此斷層面的滑移量很小，沒有太大的貢獻；表示單一斷層模型即可符合地表觀測結果。當滑脫面頂部位於觸口斷層下方時，滑脫面上平均滑移速率為44.6 mm/yr，最佳深度為5 km，且斷層位置和深度兩參數間相關性高，隨斷層位置東移，最佳深度會逐漸加深至15 km；最佳傾角約為0°，為一水平的滑脫面。根據Hsu et al.（2003）利用1993?1999年GPS速度場，以二維彈性半無限空間錯位模型推求台灣造山帶間震期之地殼變形結果顯示，台灣南部構造可能存在近乎水平的滑脫面，約位於深度10 km處，滑脫面上的滑移速率為40 mm/yr，與本研究中的逆推結果相似。 Global Position System (GPS) has become an efficient tool for studying crustal deformation and geodynamics. In the event of a major earthquake, continuous GPS data are useful for researches on interseismic, coseismic and postseismic deformations. Combining with results from near-fault dense-spacing GPS surveys, we may detect locations and deformations of blind faults and estimate the fault-slip rates. These important information can be used for earthquakes potential evaluation. The southwestern Taiwan is an active tectonic area. Frequently occurred earthquakes indicate that this area now is in a period of highly seismic activities. In this study, we use GPS time series analysis to correct 1993-2002 velocity field in southwestern Taiwan. In order to understand the active tectonic structures in this area, we assume an appropriate fault model in elastic half-space based on geological, earthquake, and seismic reflection data, to invert for optimal fault geometry and slip rate. Stacking of power spectral densities from continuous GPS data in southwestern Taiwan, we found the slopes of spectra (spectral index) are -0.61, -0.62, and -0.55, for the E, N, U components, respectively. This result and former studies indicate the continuous GPS data are best described as a combination of white noise and flicker noise. Hence we use 「white noise + flicker noise model」as noise model for GPS time series analysis in this study. The observed motion of long-period recording continuous GPS station can be described by some model parameters such as linear rate, annual periodic and semi-annual periodic motions, coseismic offsets, postseismic rate change, and exponential decay after earthquakes. The maximum likelihood estimator (MLE) is utilized to estimate the amplitudes of white and flicker noise in the postfit residuals of each time series. The full covariance matrix is the sum of the white and flicker noise covariance matrices; it can be employed to model time series data again. By iterating foregoing steps, we can obtain the cleaned time series data and final model parameters. When the postfit residuals show a random distribution, it means the corrections are essentially complete. The mean RMS values of resdiuals which represent for precision of continuous GPS observations are 3.2mm, 2.2mm, and 9.7mm in the east, north and up components, respectively. We use the results from GPS time series analysis to estimate the 1993-2002 velocity field of southwestern Taiwan. With respect to Paisha (S01R) of Penghu , the horizontal velocity field shows obviously decrease from west to the east of Jiuchiunken－Muchiliao－Liuchia fault, and even decrease to nearly zero at the Chianan coastal area. On the other hand, the velocities increase southward and their directions become southwesterly to the south of Tsochen fault. It reaches to 50 mm/yr in the Kaohsiung－Pingtung area. This implies the complexity of structures in southwestern Taiwan. Based on the corrected velocity field, we assume two fault models in the inversion. The basis of model fault geometry is the thin-skinned theory that proposed a decollment underneath the fold-thrust belt in western Taiwan. The difference between the two models is one with a decollment only, and the other has a thrust fault extending to the surface from decollment. The results of inversion by grid search and bootstrap method indicate the surface traces of the thrust fault is near the Jiuchiunken－Muchiliao－Liuchia fault, but the slip rates on the extending fault plane are relatively small. In other words, the model without extending fault is enough to describe the GPS observations on the surface. When the surface projection for the western upper edge of decollment is very close to Chuko fault, average slip rate on decollment is 44.6 mm/yr. The optimal depth and dip are 5 km and 0°, respectively. It is a sub-horizontal decollment. However, there is a high correlation between two model parameters, position and depth. Down to 15 km, the depth of decollment becomes larger when the western edge of decollment moves further east. Hsu et al.（2003） used a GPS surface velocity field of Taiwan for the time period between 1993 and 1999 to infer interseismic slip rates on subsurface faults, their result show there exists a nearly horizontal decollment at a depth of 10 km and the slip rate on decollment is 40 mm/yr, consistent with our conclusion from this study. |
