應用多期正射影像與點雲資料精進大規模崩塌之分區

近年來大規模崩塌成為備受關注之議題，於邊坡災害發生後首先需進行搶救災，並針對災害特性規劃合適之整治工程與長期監測方案。由於大規模崩塌影響範圍廣泛，因此可能同時涉及數種崩塌機制，邊坡分區分塊顯得尤為重要。然而，目前實務上多以單一時間之地形與地貌進行崩塌分區，此方法可能使分區與邊坡之運動行為和崩塌機制特徵不符，進而影響工程成效或使監測資料無法反映邊坡真實情況。因此本研究旨在基於多期正射影像與點雲資料精進現有崩塌分區之方法，透過分析崩塌不同時序之地形變化與三維位移場，進一步釐清崩塌機制並進行崩塌分區。本研究以光華崩塌地與台7線49.8k崩塌兩個不同類型之災害為例，兩案例之特點為其不穩定狀態皆持續半年以上，並且有多期正射影像與點雲資料記錄發生過程。研究採用M3C2演算法 (Multiscale Model-to-Model Cloud Comparison)分析點雲資料以了解災害前後之地形變遷，同時透過數位影像相關法 (Digital Image Correlation, DIC)分析平面位移分佈，並結合DSM資料加值2D DIC之結果獲得邊坡之三維運動變形行為，最後整合地表地質調查成果進行崩塌機制分區。由於光華崩塌地設有許多監測設備，因此本研究首先以該案例進行方法學驗證，建立不同位移特徵與破壞機制之關係，並應用至重建無監測資料記錄之時期以及其他崩塌案例。研究結果顯示光華邊坡與台7線49.8k邊坡之現生崩塌範圍應分別被劃分為四個與六個崩塌區塊，並釐清所涉及之破壞機制，包含岩體滑動、岩體傾倒、岩塊墜落與岩屑崩滑等。此外，研究亦通過影像處理方法與參數敏感度分析提升結果之品質，最終提出崩塌機制分區之流程與分析參數設定之建議。本研究之成果細緻化了目前崩塌分區方法，建立更為符合災害特性之分區標準，並為邊坡觀測之佈設以及後續更深入之邊坡穩定性分析提供參考之依據。

關鍵字

大規模崩塌；崩塌機制分區；邊坡三維運動行為；點雲資料分析；數位影像相關法 (DIC) ； Multiscale Model-to-Model Cloud Comparison (M3C2)

並列摘要

The stabilization of slopes is a crucial issue in Taiwan. The impacted area of large-scale landslide is extensive and may exhibit diverse characteristics so the primary task in engineering design is to zone the landslide. However, current zoning methods are based on the landscape and topography at a specific time, leading to inconsistency between the zonation and the actual condition of slopes. This research aims to zone the large-scale landslides by considering the failure mechanisms through the analysis of multi-temporal orthoimages and point cloud datasets. In this study, Multiscale model-to-model cloud comparison (M3C2) and Digital image correlation (DIC) were utilized to analyze the point cloud data and orthoimages, respectively. M3C2 can detect changing areas of the slope, addressing the limitation of DEMs of Differences (DOD) in steep terrain. DIC analysis was used to map the full-field planar kinematics and deformation behavior, and then incorporate the Digital surface model (DSM) to clarify the 3D displacements and infer the failure mechanisms. Finally, these results were combined with the field survey to perform the zonation. We selected the Guanghua landslide and the landslide at 49.8k mileage at Provincial Highway 7 as the case studies. Both cases were characterized by unstable conditions for more than half a year and were recorded the failures with multi-temporal datasets. Since many monitoring devices were installed in the Guanghua landslide, the methodology will first be validated in this case and subsequently applied to the time domain without monitoring data coverage and to another case. Additionally, image processing and parameter sensitivity analysis were also conducted to enhance the quality of the results. The results indicated the Guanghua area could be divided into four zones, involving the rockslides、toppling and shallow rock avalanches. The 49.8k landslide was divided into six zones, with failure mechanisms including the rockfall, rockslide, and rock avalanche. This research refined the lack of current landslide zonation and proposed the analysis process that provided a reference for planning the observation of slope and in-depth analysis of disasters.

並列關鍵字

large-scale landslide ； failure mechanism zonation ； 3D kinematics behavior ； point cloud datasets ； Digital Image Correlation (DIC) ； Multiscale Model-to-Model Cloud Comparison (M3C2)