本研究係整合遙測技術、GIS分析及自我組織圖類神經網路(Self-Organizing Map, SOM)，發展崩塌地自動萃取模式，分析九九峰地區921震災初期之崩塌地。萃取模式可分為二個階段，第一階段為自我組織圖類神經網路之聚類分析，將研究地區之影像以非監督式學習分為指定數量之聚類；第二階段為依第一階段聚類結果，以GIS疊圖判釋方式，萃取最適合之崩塌區位。研究結果顯示，九九峰地區921震災初期之崩塌地面積為849.20 ha。本研究所建立之模式可迅速萃取崩塌區位，作為崩塌地治理之參考依據。

In this study, a landslide auto-detection model was developed to rapidly extract accuracy of landslide sites from multitemporal SPOT images by combining remote sensing, image differencing method with GIS technique. The model mainly consists of two stages. First, SOM clustering analysis is used to classify the differenced image as user-defined clusters using unsupervised learning. Second, GIS overlay interpretation, based on the first-stage SOM result, is used to rapidly extract the landslide sites. The analyzed results indicate that there were 849.20 ha of the landslide area extracted in the initial earthquake stage, and are useful for decision making and policy planning in the landslide area.