整合遙測特徵指標與自組織聚類神經網路模式於崩塌地萃取之研究―以莫拉克風災之阿里山溪集水區為例

Using Landslide Feature Indices and Self-Organizing Map for Landslide Extraction-A Case Study in Alisan Creek Watershed Struck by Morakot Typhoon

林文賜(Wen-Tzu Lin) ； 黃仕凱(Shih-Kai Huang) ； 黃碧慧(Pi-Hui Huang)

《水保技術》 6卷4期 (2011/12) Pp. 185-190

https://doi.org/10.30040/JSWCT.201112.0001

摘要

九十八年八月八日莫拉克颱風（又稱為八八水災）造成中南部及東部地區嚴重山崩、地滑、土石流、淹水及村毀等災情，本研究係以遭受莫拉克颱風災害之阿里山溪集水區為範圍，蒐集研究地區之集水區資料、空間數值資料及風災前後SPOT 衛星影像資料，發展萃取崩塌地影像之遙測光譜特徵指標及二層式自組織聚類神經網路萃取崩塌區位。分析結果顯示，第一層SOM模式為粗分類，將影像分類為精確崩塌區、模糊崩塌區及非崩塌區；再將模糊崩塌區透過第二層SOM模式（細分類），可分類為精確崩塌區及非崩塌區，合併二層之分類成果即可獲得全區崩塌地；研究地區經二層SOM模式萃取，其崩塌面積為619.91公頃，準確度評估之Kappa值為0.9766，顯示本研究發展模式可供崩塌地評估及坡地防災之參考。

關鍵字

崩塌地 ； 遙測特徵指標 ； 自組織聚類神經網路

並列摘要

Morakot typhoon, occurred on August 8, 2009, caused serious damages such as villages and bridges destroyed by landslides, debris flow and flood hazards in south-central and eastern Taiwan. The Alisan creek watershed, struck by Morakot typhoon, was chosen as studied area for assessing the landslide sites. In this study, watershed disasters, spatial digital data and SPOT images before and after typhoon were collected, and two landslide feature indices and two-layer self-organizing map neural network were developed to extract accurate landslides. The analyzed results indicate that the image after typhoon can be classified as precise landslide area (PLA), fuzzy landslide area (FLA) and non-landslide area (NLA) by the first-layer SOM extraction (called the coarse classification). The FLA can be isolated as PLA and NLA by the second-layer SOM extraction (called the fine classification). The completed accurate landslide sites can be obtained by merging first- and second-layer PLAs. By two-layer SOM extraction, there are 619.91 ha of landslide area with high Kappa value 0.9766 extracted at studied site. It shows the developed model can be used for landslide assessment and slopeland disaster prevention.

並列關鍵字

Landslide ； Landslide Feature Indices ； Self-Organizing Map