| 摘要: | 塊運動和地殼形變相關研究的資料常以全球衛星定位系統(Global Positioning System，簡稱GPS)為基礎，其中由GPS測站每日座標解所形成的時間序列(time series)除了可提供區域變形速度及應變率的估算外，也可偵測地表運動隨時間的演化。然而，GPS時間序列時常會在某些時間點上發生不同大小的非連續變化，其被稱為時間序列的錯移(offsets)。造成錯移的可能原因有地震、地滑、地層陷落、測站環境擾動、資料處理方式變化、儀器更換或韌體更新等，但大多數是不明因素。未被先驗提出的錯移將會影響時間序列的模型分析，進而在測站速度估算上引入不同程度的偏差。因此有效找出錯移會對時間序列的研究很有幫助。然而，隨著GPS連續測站資料的增加，資料處理時以人為方式逐條檢視時間序列中的不連續變化將變得不切實際。本研究使用自動偵測技術來解析GPS時間序列中錯移的發生時間與大小，並將結果以統計方式做系統性的展現及歸類。除了以此探索可能存在的地殼變形訊號外，也期望能將這些成果提供給相關的測網管理單位作為測站維護的紀錄與參考。本研究使用全台各地共495個GPS連續測站，並利用JUST(Ghaderpour et al. 2018)以及Hector(Bos et al. 2013)兩種方法對觀測長度大於4年的資料進行分析，以偵測時間序列中潛在的錯移，並與已知的地震事件以及測站儀器更換時間進行比對。初步結果顯示利用Hector找到的錯移有75%對應到地震事件、8%對應到儀器更換，而利用JUST找到的錯移有78%對應到地震事件、3%對應到儀器更換。在錯移量與震央-測站的距離的分析當中也發現到大部分的錯移量跟距離呈現反比，但是會根據地震對於各個地區造成的影響(如震度等)出現例外，對於可能是地震所造成的offset 需要如果只是根據發生時間判定會有誤，且自動偵測方法會因為資料缺失或者其他因素而將錯移發生時間定在錯誤的地方導致誤判錯移發生的原因。;GPS are essential for many geophysical applications such as crust deformation, plate tectonic, and so on. The GPS time series not only can calculate the strain rate of the area, but also can monitor the change of the surface. However, there are some abrupt changes in the GPS time series called offsets. These offsets may be caused by seismic events, equipment changes, creep events, environmental change, metadata change, or unknown reasons. Removing offsets can greatly improve GPS velocity solution and strain rate calculation. Therefore, finding these offsets can greatly improve the related research. However, manually detecting offsets is impracticable since that GPS stations are gradually increase. In this research, I use two automated methods to detect offsets, and systematically display and classify offsets’ results in statistical manner. In addition to exploring possible crustal deformation signals, I also expected that these results can benefit relevant survey network management units as records and references for station maintenance. With 479 GPS stations in Taiwan, I apply two methods JUST (Ghaderpour et al. 2018) and Hector (Bos et al. 2013) to analyze their three-component time series where the data lengths are more than 4 years. The reason I choose JUST and Hector is that JUST can resolve seasonal signals very well while Hector can analyze noises in the time series. Results show that in the offsets detected by Hector, 75% are caused by earthquakes, and 8% are due to equipment changes. For the offsets which are detected by JUST, 78% are caused by earthquakes, and 3% are due to equipment changes. |
