運用永久散射體雷達干涉技術建立崩塌潛勢評估模型及邊坡變位門檻值－以布唐布納斯溪沿岸邊坡地區為例

臺灣為於板塊交界帶及副熱帶季風氣候區，導致臺灣的地形、地質及氣候狀況均非常複雜多變，這些原因造成臺灣山坡地頻繁有崩塌發生。因此，監測系統對臺灣而言是非常重要的。　　永久散射體雷達干涉技術 (PSInSAR) 是一種新興的地表變形測量技術，其具有許多優點，如覆蓋面積大、不受天氣影響、取樣間隔一致、高空間解析度及達公分甚至毫米等級的精度，令PSInSAR技術成為一種適合用於監測崩塌前地表變形之方法。　　本研究以高雄市桃源區布唐布納斯溪沿岸邊坡地區為研究區域，分析區域內的12項會影響邊坡崩塌發生可能性的潛感因子，透過相關性分析，找出與崩塌具顯著相關的顯著因子，並以這些顯著因子利用邏輯斯迴歸建立崩塌潛勢評估模型。　　除此之外，本研究使用Sentinel-1衛星之雷達影像，運用SNAP及StaMPS兩款軟體進行PSInSAR解算，以PSInSAR計算所得之地表變形計算地表變位速度，並透過優化分析，找出與崩塌最具相關性的地表變位速度之計算條件，並以此地表變位速度制定邊坡變位門檻值及修正崩塌潛勢評估模型。　　根據研究結果，證明邊坡是否會發生崩塌並非受單一因子所控制，而是眾多因子綜合影響的結果。其中，篩選出8項顯著因子-事件累積降雨量、坡度、坡向、地形粗糙度、剖面曲率、植生指標、水系距及岩體強度。並找出時間間距為六個月時之沿坡面方向的地表變位速度vslope與崩塌最具相關性，最能偵測到崩塌發生前之地表變形行為。並以地表變位速度vslope制定邊坡變位門檻值及建立潛感值修正模型，其中，注意值為3.8 mm/6-months，警戒值為23.0 mm/6-months；而潛感值修正模型利用vslope修正崩塌潛感值後，模型的整體預測準確率提升約3%，達到82%；並透過僅分析有PS點之斜坡單元來觀察vslope之貢獻，發現利用vslope修正崩塌潛感值後，模型的預測準確率提升約6%，高達85%，證明利用地表變位速度vslope修正崩塌潛感值可以有效提升崩塌潛勢評估模型之崩塌預測準確性。

關鍵字

潛感因子；永久散射體雷達干涉技術；邏輯斯迴歸；邊坡變位門檻值；崩塌潛勢評估模型

並列摘要

Taiwan is in plate junction zone and subtropical monsoon climate zone, resulting the terrain, geology and weather in Taiwan are very complex. These reasons cause common landslides happening in mountainous area. So the monitoring system is very important in Taiwan. 　　PSInSAR is a new technology to monitoring the ground deformation. It has many advantages, such like wide area coverage, not affected by the weather, good temporal sampling, high spatial resolution and millimetric or centimetric precision. So PSInSAR is a good method to detect the pre-landslide ground deformation. 　　The study area is located on Pu-Tun-Pu-Nas river area in Kaohsiung City. I choose 12 susceptibility factors which may affect the occurrence of landslides, and do the correlation analysis to get significant factors. Then using significant factors to build the landslide susceptibility assessment model by logistic regression. 　　Besides, I use Sentinel-1 SAR imagery to calculate PSInSAR by SNAP and StaMPS software. Then using the ground deformation calculated by PSInSAR to calculate the velocity and doing optimization analysis to get the velocity which has a higher correlation to landslides. Then finding the slope displacement threshold of velocity and using velocity to refine the landslide susceptibility model. 　　In the results, landslides are affected by many factors was confirmed. I find 8 significant factors – accumulated rainfall of the event, slope, aspect, terrain roughness, profile curvature, planting index, distance to river and rock strength and find the velocity along the slope surface vslope calculated by 6 months of deformation data has the highest correlation. Then I establish attention value of vslope is equal to 3.8 mm/6-months and warning value of vslope is equal to 23.0 mm/6-months and use vslope to build the susceptibility refinement model. Through using vslope to refine the landslide susceptibility, the predictive accuracy increases about 3% and the refined model’s predictive accuracy is 82%. If we just focus on the units which has PS points to see the contribution of vslope, the accuracy increment is about 6%, and the accuracy of the refined model is up to 85%. It shows that using vslope to refine the landslide susceptibility can increase the accuracy of the landslide susceptibility model.