以反方位角法從事臺灣外海之快速地震定位

Abstract

當地震發生後，一個快速且可信之震源位置與規模估算對於地震預警系統發布警報至關重要。其中震源位置之估算於現行中央氣象署地震預警系統中的定位流程分為兩步驟：首先是利用線性反演法求出震央位置後再利用垂直向格點搜尋法找出最佳震源深度。然而線性反演法取得震央位置有其侷限性，咎因於：(1)震央位置若位於測站包覆較差的區域(如臺灣外海)則估算結果將無法迅速收斂(2)線性反演法仰賴準確的P波到時，較不確定的P波到時將解算出較低準確的震央位置(3)線性反演法需等待足夠數量之測站觸發後才可開始進行定位程序(4)若地震帶來之破壞使測站間訊號傳輸失效，將延誤震央的取得。 本研究旨在對利用線性反演法在臺灣外海取得震央提供一替代方案，觀察包含P波到時之3秒時間窗內之水平方向上粒子動態情形：(1)利用主成分分析(Principal Component Analysis, PCA)，取得單一測站於事件P波到後粒子動態情形之主成分，以得到P波來向即反方位角(Back Azimuth, BAZ)；(2)利用Nakamura（1988）方法(以下簡稱NA)計算單一測站紀錄中水平與垂直分量間之交相關量決定反方位角(3)將兩種方法得到之反方位角加權平均，定為該測站於事件中的反方位角。將至少兩個測站於接收到P波後估算之反方位角交疊，即能初步取得震央位置。 本研究以臺灣強地動觀測網(TSMIP)、臺灣寬頻地震測網(BATS)以及中央氣象署地震觀測網(CWASN)，分析2020年至2023年芮氏地震規模大於5.0之94筆事件，分析單一測站在不同事件中反方位角穩定性，建立測站反方位角標準差目錄，研究指出三大測網共504站整體測站反方位角標準差達41.1°，然而取表現較佳之前1/4測站共117站為揀選測站，反方位角標準差則可優化達15.93°。以中央氣象署現有反方位角目錄為參考基準比較，分析每一測站對估算反方位角之差異，設計出快速取得震央之程序。測試24筆外海事件，平均震央誤差為40.26公里，震央誤差標準差為35.87公里。震央誤差多集中於40公里以內，顯示揀選測站取得的反方位角對震央位置具有穩定性。經由本法取得震央位置後期望未來可結合其他定位方法，如等時差曲面、沃羅諾伊胞等，進一步提升地震預警系統中震源位置估算。 本研究所使用之方法可在測站包覆差的臺灣外海地震作用良好，單一測站接收到P波即可開始定位程序，無需考慮測站間訊號傳輸問題，因此本方法提供較簡單以及較低的門檻觸發條件即可對事件進行震央位置計算，期望貢獻降低定位程序算力與降低誤報發生機會，爭取縮短發布時間之效益。

After an earthquake occurs, rapid and reliable estimation of source location and magnitude is critical for the issuance of alerts in earthquake early warning systems. In the current earthquake early warning system operated by the Central Weather Administration, the source location procedure consists of two steps: first, the epicenter is estimated using a linear inversion method, and then the optimal focal depth is determined through a vertical grid search. However, epicenter estimation using linear inversion has several limitations: (1) when the epicenter is located in regions with poor station coverage (such as offshore Taiwan), the estimation results cannot converge rapidly; (2) the linear inversion method relies on accurate P-wave arrival times, and uncertainties in P-wave arrivals will lead to lower accuracy in epicenter estimation; (3) the linear inversion method requires a sufficient number of stations to be triggered before the location procedure can be initiated; and (4) if earthquakes cause damage that disrupts signal transmission between stations, epicenter determination will be delayed. This study aims to provide an alternative approach to epicenter estimation for offshore earthquakes in Taiwan. The horizontal particle motion within a 3-second time window containing the P-wave arrival is analyzed: (1) Principal Component Analysis (PCA) is used to obtain the principal component of particle motion at a single station after the P-wave arrival, in order to determine the P-wave incoming direction, i.e., the back azimuth (BAZ); (2) Nakamura (1988) method is used to calculate the cross-correlation function between the horizontal and vertical components recorded at a single station to determine the back azimuth; (3) the back azimuths obtained from the two methods are combined using a weighted average and defined as the back azimuth for that station during the event. By overlapping the back azimuths estimated from at least two stations after receiving the P wave, a preliminary epicenter can be obtained. This study analyzes 94 earthquakes with local magnitudes greater than 5.0 that occurred between 2020 and 2023 using data from the Taiwan Strong Motion Instrumentation Program (TSMIP), the Broadband Array in Taiwan for Seismology (BATS), and the Central Weather Administration Seismic Network (CWASN). The stability of back azimuths at individual stations across different events is analyzed, and a catalog of station back azimuth standard deviations is established. The results show that the overall back azimuth standard deviation of 504 stations across the three networks is 41.1°. However, when selecting the top-performing one-quarter of stations (117 stations) as selected stations, the back azimuth standard deviation can be improved to 15.93°. Using the existing back azimuth catalog of the Central Weather Administration as a reference, differences in back azimuth estimation at each station are analyzed to design a rapid epicenter estimation procedure. Testing on 24 offshore events shows that the average epicentral error is 40.26 km, with a standard deviation of 35.87 km. Most epicentral errors are concentrated within 40 km, indicating that back azimuths obtained from selected stations provide stability for epicenter location. The method proposed in this study performs well for offshore earthquakes with poor station coverage. A single station can initiate the location procedure after receiving the P wave, without considering signal transmission between stations. Therefore, this method provides a simpler approach with lower triggering thresholds for epicenter estimation, and is expected to contribute to reducing computational load and decreasing the occurrence of false alarms, thereby shortening alert issuance time. After obtaining the epicenter location using this method, future work may combine it with other location methods, such as hyperbolic surface methods and Voronoi cells, to further improve source location estimation in earthquake early warning systems.