由GPS速度場之群聚分析來探討台灣的活動構造分區

Using Cluster Analysis of GPS Velocities to Explore Active Tectonics in Taiwan

李依倢(Yi-Chieh Lee)

指導教授 ： 莊昀叡

國立臺灣大學/理學院/地理環境資源學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202001109

摘要

GPS觀測技術常用於研究地殼變形與活動構造，透過分析地表上的速度場資料，能進而推估地表深處的塊體運動並建立塊體模型。然而，這樣的模型建置仍需一定程度的仰賴學者們的專業知識與現有地質調查結果來做解釋，也使得結果多了些人為的主觀因素參雜其中。為了使模型建置能夠被更為客觀化的建置，本研究透過空間統計中的群聚分析，找到速度場運動特徵相似的GPS觀測站，並將以空間統計中所得到的分群結果與現有的塊體模型結果做比較，其中，分群所得到的群集可解釋為塊體，而群與群之間的邊界則有反映了不同構造塊體間的活動構造。
本研究採用K-medoids分群演算法分析台灣的GPS水平速度場，並以Elbow method、Average silhouette method與Gap statistic method來評估最佳塊體數。根據統計檢驗的方式，可以將台灣初步劃分為四個區塊，分別沿著縱谷斷層、西部麓山帶與歐亞大陸邊緣進行劃分。基於此結果後又持續地進行分群，最多可將台灣劃分至34個塊體。在分群過程中，仍可揭示出區域的活動構造及構造的活動性，此分群結果可作為尚具爭議性的活動構造的另一種佐證，也可部分的揭示出仍尚未被發現的活動構造。群聚分析提供了一個快速且視覺化的方式，可應用於地質資訊尚未明朗的地區，提供對於該地區構造的初步理解。

關鍵字

GPS速度場 ； 群聚分析 ； 活動構造 ； 分割式群聚法

並列摘要

GPS observations on the surface are commonly used for analyzing crustal deformation and active faults in such a way that surface velocities derived from GPS are often considered to be associated with fault movement at depths bounding rigid tectonic blocks. However, the choice of the tectonic blocks highly depends on either the knowledge of active fault geometries based on available field investigations and seismicity distribution or subjective determination from geodetic observations. To provide a more objective choice of blocks, cluster analysis is a useful tool, which groups a set of GPS velocities based on spatial statistics. Clusters can represent blocks and cluster boundaries and may reflect block-bounding faults. Under the arc-continent collision and mountain building, most of active faults in Taiwan are thrust faults or oblique faults. Therefore, it is crucial to examine how cluster analysis applied to the Taiwan region.
In order to examine the feasibility and characteristics of the cluster analysis for such thrust-dominant areas, this study used the K-medoids clustering algorithm for horizontal GPS velocities in Taiwan and estimate optimal block numbers by statistical methods of Elbow, Average silhouette, and gap statistic methods. The optimal cluster number is 4 and the analysis can estimate up to 34 clusters with discernible geologic features. Earlier determined clusters represent higher velocity gradient between the clusters, which may indicate active structures with higher slip rates. The cluster analysis can offer a simple and fast method to indicate the possible tectonic blocks and uncertain active structures.

並列關鍵字

Velocity Field ； Cluster Analysis ； Active Tectonic ； Partitional Clustering

參考文獻

Bertrand, E., Unsworth, M., Chiang, C.-W., Chen, C.-S., Chen, C.-C., Wu, F., Türkoğlu, E., Hsu, H.-L., and Hill, G., 2009, Magnetotelluric evidence for thick-skinned tectonics in central Taiwan: Geology, v. 37, no. 8, p. 711-714.

Bock, Y., and Melgar, D., 2016, Physical applications of GPS geodesy: a review: Rep Prog Phys, v. 79, no. 10, p. 106801.

Bonilla, M. G., 1975, A review of recently active faults in Taiwan, US Geological Survey, v. 41.

Bonilla, M. G., 1977, Summary of quaternary faulting and elevation change in Taiwan: Mem. Geol. Soc. China, v. 2, p. 43-55.

Bos, A. G., Spakman, W., and Nyst, M. C., 2003, Surface deformation and tectonic setting of Taiwan inferred from a GPS velocity field: Journal of Geophysical Research: Solid Earth, v. 108, no. B10.