莫拉克颱風之豪雨誘發山崩，使小林村遭到滅村的慘劇。本研究針對小林村鄰近區域進行莫拉克颱風下之山崩分析，研究內容包括山崩特性統計、最佳化參數逆分析及即時動態分析介面開發。運用莫拉克颱風災前、災後之數值高程模型(Digital Elevation Model, DEM)、山崩目錄(Collapse Inventory)及各山崩潛勢因子圖層(地形因子、地質因子、水系因子及坡型因子)結合地理資訊系統進行空間分析、人工判釋及校正山崩目錄，了解各潛勢因子對小林村鄰近區域山崩之影響程度及自然邊坡特性。在最佳化參數逆分析之部分則採用遺傳演算法之最佳化技術結合暫態降雨入滲與網格式廣域邊坡穩定分析模式(Transient Rainfall Infiltration and Grid-Based Regional Slope-Stability Model, TRIGRS)進行最佳化參數逆分析。並參考山崩特性統計之成果、邊坡特性、水文參數關係及山崩臨界雨量觀念進行優化原始最佳化模組，使參數逆分析之結果較符合實際情形。最後考量TRIGRS成果圖層展示之不便性，開發即時動態分析介面，供防救災工作參考。山崩特性統計部分顯示水系為影響邊坡穩定最重要之因子；順向坡及地質構造線較可能觸發大規模之邊坡滑動。逐步改善最佳化分析模組之結果顯示，雖使山崩預測正確率從未改良前之95%下降至60~75%，但相關係數從0.238上升至0.998，且山崩歷程較符合實際之降雨情形，結果有一定之合理性，但未來仍有許多改善的空間。Xiaolin village was completely destroyed by the deep-seated landslide that resulted from the extreme rainfall of Morakot typhoon. To understand the landslides of the Xiaolin village area in Morakot typhoon, this research does the study included the statistic of landslide characteristic, the parameters of back analysis in optimization and development of real-time dynamic analysis interface. In the statistic of landslide characteristic, it uses the digital elevation models before and after the Morakot typhoon, the landslides inventory of Morackot typhoon and the layers of susceptibility factors. The susceptibility factors include topographical factors, geological factors, the stream factors and the factor slope type. The analysis of statistic and the calibration of landslides inventory are performed by spatial analysis and visual interpretation on GIS technique. This study in the statistic of landslide characteristic can realize the influence of susceptibility factors and the characteristic of natural slope in rain. The parameters of back analysis in optimization combined the RGA(Real-coded Genetic Algorithm, RGA) of optimal technique with TRIGRS (Transient Rainfall Infiltration and Grid-Based Regional Slope-Stability Model, TRIGRS) to evaluate the parameters in the Xiaolin village area. It also uses the results of the landslides characteristic statistics, the characteristics of natural slope, the relation between hydraulic conductivity and diffusive coefficient, and the critical cumulative rainfall of landslides to improve the origin of the optimal module. The improvement can make the results of optimization much better. To solve the inconvenience of the TRIGRS result exhibition, this research develops the real-time dynamic analysis interface. It can provide the information of slope stability. The preliminary results show that the stream factors are the most significant influencing factors on landslides. Large scale landslides often occurred in the area of dip slope and tectonic line. The results of the improvement module show the value of landslide’s accuracy from 95 % to 60~75 %, but the value of correlation coefficient from 0.238 to 0.998. The process of landslide occurrence is similar to real state of natural slope. Now the results of back analysis have rationality, but they also have much improvement in the future.
