融合不同SAR演算法取得三維同震位移場之探討-以2018花蓮地震為例

Abstract

衛載合成孔徑雷達(Space-borne SAR)擁有高測量精度和大範圍量測地表位 移的特性，其產製出的地表位移場可以再進一步利用不同角度的位移資料拆解成 三維地表位移場，此三維地表位移場可以幫助我們了解和約制多種地殼活動像是 火山爆發、地震、地層下陷或山崩等。 合成孔徑雷達(SAR)在雷達波傳遞的過程會受到各種誤差來源，而其中對流層 和電離層是兩個主要影響 SAR 精準測量的誤差源，此外，不同獲取地表位移的演 算法有不同的雜訊程度和適用的地區，而三維地表位移場的反演往往需要兩種以 上的演算法的組合。因此，本研究檢驗了不同的演算法組合和評估了對流層和電 離層在獲取地表位移時的影響程度。總共有三種版本的三維地表位移場被產製出， 而其中最好的版本(Version 3)是根據空間解析度、精度和變形趨勢決定出，此版 本是由差分干涉合成孔徑雷達(DInSAR)和像素位移追蹤法(POT)跟多孔徑干涉法 (MAI)的融合製作出，顯示出 POT 和 MAI 互相彌補了彼此的弱項，即同時保有空間 解析度又在低相關性的地區有可靠的位移資料。另一項版本(Version 4)利用和版 本 3 相同的資料但在反演三維位移場先進行了去噪，結果顯示先前的去噪並未提 升整體表現。最後，2018 花蓮地震的損害代理地圖(Damage proxy map)和三維同 震地表位移場的比較顯示損害地圖上的破壞和地表位移並沒有很好的相關性，此 特徵可以進一步推論這些被偵測出的破壞應該和都市化程度、斷層破裂和地表運 動有關，而地表位移的貢獻應該小於上述三者。

Space-borne SAR vehicles demonstrated the capability of measuring surface motions with high accuracy, spatial resolution and vast spatial coverage. Such displacement field can be further inverted to a 3D displacement field with the aid of displacements in different directions. A 3D displacement field helps better visualize and constrain various crustal activities including volcanic eruptions, earthquakes, land subsidence and landslides etc. SAR is affected by various error sources along the radar wave travel path. The troposphere and the ionosphere are the two main disturbances that influence accurate measurements. Additionally, different displacement-recovering algorithms have different noise levels and credibility due to varied surface properties. Also, it often requires more than two algorithms to sufficiently resolve the 3D displacement field. Therefore, this study examined different combined algorithms for recovering a 3D displacement field along with assessing the errors contributed by the ionosphere and the troposphere. Three versions of 3D displacement fields were proposed and the best one among all was Version 3 which was inverted from DInSAR and fusion of Pixel-offset Tracking (POT) and Multiple Aperture Interferometry (MAI) based on spatial resolution, accuracy and deformation patterns. It showed that MAI and POT can both compensate the weaknesses of either one having good spatial resolution and reliable displacement results in low coherence area. An additional trial (Version 4) was made by the same inputs as Version 3 with prior denoising process before inversion. It didn’t improve the overall performance thus was inferred redundant. Lastly, the damage proxy map (DPM) of the 2018 Hualien earthquake was produced and compared with the 3D displacement fields. It showed that the inferred damages did not correlate with the offset resulted from the earthquake. It was further speculated that the detected seismic damages should be a function of urbanization, fault rupture and ground motion and the surface displacement might contribute less.