SPOT影像與航照資料應用於崩塌地辨識之探討

Landslides Identification Using SPOT Imageries and Aerial Photographs

蕭國鑫(Kuo-Hsin Hsiao) ； 尹承遠(Chen-Yuan Yin) ； 劉進金(Jin-King Liu) ； 遊明芳(Ming-Fang Yu) ； 王晉倫(Chin-Lun Wang)

《航測及遙測學刊》 8卷4期 (2003/12) Pp. 29-42

https://doi.org/10.6574/JPRS.2003.8(4).3

摘要

本研究利用SPOT資料，結合影像自動分類、綠度與亮度植生指數，初步研判台灣中、北部的崩塌地區；再疊合3D透視影像與地形圖資訊所輔助螢幕判釋及數化的向量資料，以確定是否為崩塌地及修正辨識範圍，再與航照判釋成果比對。結果顯示 SPOT 影像受限於空間解像力及混合像元效應影響，若崩塌面積小於3 個像元，或在陰影區與靠近農作區的小型崩塌、崩塌後尚有植被或草類生長地區就不容易辨識，所以辨識崩塌個數及面積遠低於航照判釋結果；但對於較大型崩塌或需要快速得知災後崩塌資訊者，則可提供最迅速的崩塌資料。另由航照判釋崩塌資訊顯示，台灣中、北部地區崩塌地數量以小於1公頃者最多，崩塌面積則大部份介於1~50公頃之間；而發生崩塌的坡度以15°~30°的坡地居多，其次為30°~ 45°的坡度範圍，坡度在5°以下或75°以上者，甚少有崩塌發生。

關鍵字

崩塌 ； 纓帽轉換 ； 影像分類

並列摘要

SPOT images of Central and Northern Taiwan were used for identifying landslides and obtaining a distribution map of landslides. Three phases were adopted in this study, one is combining an image classification and a threshold of indices of greenness, and brightness values. Second is a modification of the above-result by on-screen interpretation of the classified image in a 3D perspective. Last is a further comparison with those interpreted from aerial photographs. It is concluded that a minimum identifiable unit of landslides on SPOT image would be larger than 3 pixels. Landslides in an area of near residential or farmlands, or landslides which are partially-vegetated, are usually overlooked because they are not be readily distinguishable with the environments. Thus, stereo-interpretation of conventional aerial photographs identifies a lot more landslides in these areas. Nevertheless, the advantage of large coverage of 60 km of a SPOT image as compared to 1~3 km coverage of an aerial photograph would intrigue the applications of emergent enumeration of hazards, However, a scheme combining both advantages of SPOT and aerial photographs is a sound solution for providing a single source of standardized information about landslides in Taiwan. The statistics of landslides obtained in this study by air-photos shows that the majority of the landslides are smaller than 1 hectare; whereas, most of the landslides are in an area size between 1 to50 hectares and most of them are in a gradient of 15 to 30 degrees.

並列關鍵字

Landslide ； Tasselled-Cap Transformation ； Image Classification