本研究係整合遙測理論、GIS分析及電腦疊代演算法,發展崩塌地自動萃取模式,藉由多期SPOT衛星影像分析921震災後之崩塌區位。萃取模式可分爲粗分類及細分類兩階段,前者係以GIS-based視覺分類法,快速萃取震災後之崩塌地;後者則依粗分類結果,結合疊代Kappa分類法,萃取最佳精度之崩塌地。研究結果顯示,兩期影像崩塌地(1999/9/27及2003/7/20)之粗分類Kappa值分別爲0.662及0.681,細分類Kappa值分別爲0.825及0.721,雖粗分類之精度較細分類爲低,但其崩塌地面積之誤差率僅爲- 09%及10.52%,顯示粗分類之崩塌地已有甚佳精度。在自然復育情況下,地震後四年內之崩塌面積,已由822.969 公頃減少爲309.813公頃,約減少62.35%(513.156 公頃)。
The catastrophic earthquake, 7.3 on the Richter scale, occurred on September 21, 1999 and caused massive landslides in Central Taiwan. In order to effectively assess landslide change, a landslide auto-detection model, including coarse and fine classification, was developed to automatically extract optimal accuracy of landslide sites from multitemporal SPOT images by combining vegetation index, image differencing method, GIS technique with iterative algorithm. The coarse classification can roughly estimate the landslide by using GIS analysis and edge matching identification. The fine classification used kappa optimization method (KOM) to automatically calculate the optimal change threshold, kappa statistic and accurate landslide distribution. The optimal accuracy analysis shows that the landslide area with maximum kappa value on September 27, 1999 and July 20, 2003 were 822.969 hectares with 82.91% and 309.813 hectares with 71.22% respectively. Over about 4 years of natural restoration, the landslide area has decreased by 62.35% (513.156 hectares).