崩塌地形與地表覆蓋變化對合成孔徑雷達強度時間序列之影響：以苗栗鹿場崩塌地為例

Abstract

坡地山崩是很常見且危害極高的災害，尤其台灣高山多雨的環境，土石崩塌的事件不斷發生，因此需要長期監控崩塌地之變化，掌握崩塌地狀況。傳統上，進行國土衛星監測，大多使用光學衛星影像繪製崩塌地範圍，但其劣勢在於易受到雲霧遮蔽無法進行高頻率時間分析，且僅能針對地表特徵物變遷做圈繪。合成孔徑雷達（Synthetic Aperture Radar, SAR）衛星由於波長較長可穿透雲層回傳訊號，且微波不僅能辨識地物還能夠偵測地形變化，故可彌補光學衛星影像的不足。然而過往研究中，因崩塌地位於植被和地形干擾的環境，導致同調性低，較難利用SAR的相位資訊解算位移量，但可藉由SAR強度資訊的變遷應用於快速圈繪崩塌地位置。然而，SAR強度資訊易受到環境影響而有差異變化，目前研究仍缺少探討SAR強度的變化量如何反映崩塌地與環境因子的關聯性。本研究目的在於發展一套雷達強度時間序列的計算流程，並進一步分析環境因子和雷達訊號之間的相關性。 本研究選定苗栗鹿場之大型崩塌地為研究區域，原因其一為此崩塌地位於Sentinel-1A/B升軌方向適合的觀測方位；其二，2018年4月13日發生超過十公頃之大規模崩塌，崩塌範圍較大易於觀測；其三，崩塌事件前後無降雨事件，可排除土壤濕度的影響，建立基礎假設為：無土壤濕度、裸露地均質、林相單一，僅探討四個環境因子：地形（坡向、坡度、區域入射角）地貌（地表覆蓋）。本研究方法首先建立事件前期SAR強度平均的量化結果，並分析四個環境因子的關係，接著比較崩塌事件中期的SAR強度變遷，最後利用前兩者的結果提出雷達強度時間序列的計算流程，並用光學影像和無人機空拍數值模型驗證。研究結果顯示，坡向因子主要控制了SAR強度的長期平均反射值高低，而坡度跟地表覆蓋物也交互被影響著，坡度因子能反映前坡縮短、疊置的狀況，而地表覆蓋物則有時間週期的變化趨勢。在事件中期，具有四種時間序列顯示了崩塌中不同的變化趨勢：強度增強、強度降低、強度不變和提前有變化，其變化趨勢符合事件前期作為背景值的坡向量化結果。最後透過前兩者結果的標準化和多重門檻值的方法流程判斷坡向改變，整體準確度為63%，坡頂和坡趾從坡向面東為主，崩成以西為主或南北方，流動部則較少坡向改變，並且在最大崩塌事件前也可偵測到部分崩塌事件以及河道堆積。本研究流程有助於判斷崩塌事件前後的坡向改變並進一步探討崩塌地的地形演育。

Landslides are a very common and extremely hazardous disaster, especially in Taiwan, where landslides are frequently triggered by earthquakes and heavy rain. Traditionally, optical satellite images are widely used to map landslides but are limited for high-frequency time series analysis because the images are susceptible to cloud and fog. In addition, the images are simply used for change detection based on surface features. Synthetic Aperture Radar (SAR) satellites can make up for the limitations since they have long wavelengths, which can penetrate clouds. Not only microwaves can identify ground objects but also detect terrain changes. In previous studies, most of the landslides limit coherence by dense vegetation and rugged topography that is difficult to calculate the displacement by SAR phase. More studies have pointed out that the intensity changes of SAR can quickly detect landslides. However, few studies explored how the temporal change of SAR intensity reflects the correlation between landslide and environmental factors. The study aims to develop a procedure for computing SAR intensity time series and further to analyze the relationships between the intensity signals and environmental factors. The study area is located in Luchang, Miaoli area according to the following reasons. First, it is a suitable observation position for the ascending of Sentinel-1A/B. Second, a large-scale landslide event occurred on April 13, 2018. Third, it was not caused by rainfall, so the influence of soil moisture can be excluded. In terms of research methods, first, we establish the long-term average quantitative results before the event and compare the relationships of four factors including aspect, slope, local incidence angle and landcover. Then we compare SAR intensity variation during the event. Finally, we set the procedure to identify landslide evolution, which is verified by optical image and DEM of drone aerial photography. The research results show that before the event, aspects mainly control the long-term average reflection of the SAR intensity in the mountain area, and slope and landcover are also influential. Slope can reflect the shortening and overlay. Landcover result in periodic variations in the time series. During the event, there are four time-series trends: intensity increase, intensity decrease, intensity unchanged, and previous change. Those trends are similar to the aspect result in the early period of the event. We identify the change of aspect by standardization and the multiple thresholds method, and the overall accuracy is 63%. In the scarp and toe areas, the aspect changes from the east to other directions, but the aspect remains unchanged in the main body. Moreover, we can detect minor landsliding prior to the main event and the formation of the debris dam. The procedure can provide preliminary identification of the aspect changes, which is helpful for discussing landslide evolution.