羅吉斯回歸與不安定指數應用於塊體崩壞敏感度評估比較—以台東縣海端鄉為例

Comparison of Mass Wasting Susceptibility Evaluation for Haiduan Township in Taitung County: Logistic Regression and Instability Index

洪政耀(Cheng-Yao Hung) ； 林雪美(Hsueh-Mei Lin)

《中國地理學會會刊》 49期 (2012/12) Pp. 77-103

https://doi.org/10.29972/BGSC.201212.0004

摘要

台灣山區地形多樣複雜、地質脆弱，再加上雨量多集中在梅雨與夏季颱風所帶來的豪雨，使台灣山區常見塊體崩壞現象發生。另外，台灣人口快速的增加，土地利用的急切需求，促使邊緣環境地帶的快速發展，造成近年來坡地災害逐漸上升。為預防坡地災害的發生與國土規劃利用，繪製塊體崩壞敏感度地圖評估環境最敏感地區，以及調查敏感之因，著手進行防護措施，才能降低災害的發生。一般塊體崩壞敏感度的分析方法主要分為二大類群，即是直接與間接技術。由於直接式的方法過於主觀的，導致各專家所分析出來的敏感度有很大的差異。故使用二元分類統計方法、多變量統計等間接式方法，包含：不安定指數法與羅吉斯回歸等，並結合地理資訊系統完成屬性資料與基本環境圖層之建立。接著，利用不安定指數與羅吉斯回歸統計分析方法，完成海端鄉塊體崩壞敏感度評估，並比較二者方法之間的差異。本研究選定坡度、坡向、高程、土地利用、地質、河流距離、道路距離等7項作為敏感度評估變數。研究結果顯示不安定指數法在全區塊體崩壞的整體正確率約為61.09%；羅吉斯回歸模式則是75.72%。另外，ROC曲線下之 AUC值，羅吉斯回歸模式的0.8836優於不安定指數的0.6673，以上顯示羅吉斯回歸模式在判釋正確率上優於不安定指數法。

關鍵字

羅吉斯回歸 ； 不安定指數 ； 塊體崩壞 ； 分類誤差矩陣 ； ROC曲線

並列摘要

The occurrence of mass wasting is common in the mountains of Taiwan Island, due to the typical tectonic structure trigger earthquake and heavy precipitation from the seasonal plumrains and typhoons. The rapid increase of population and demand of more usable land also result in development expanding into marginal land, therefore posts further negative impact on the hillside environment. The purpose of this study is to establish a current mass wasting susceptibility map of which Haiduan township, which will provide important baseline data for future studies in landslide risk reduction and land-use planning. The analysis of mass wasting susceptibility can be conducted through direct or indirect approach. The direct approach is considered more subjective and less reliable. Therefore, this study takes the indirect approach ofbivariate and multivariate statistical analyses to evaluate mass wastingsusceptibility, specifically, the analysis of Instability Index and logistic regression model.Seven geographic factors are included in the analysis as covariates: lithology, slope angle, aspect, elevation, land use, distance to river, and distance to road. The results indicate an overall correctness rate of 61.09% for the instability index, and 75.72% for the logistic regression model. In the studies of receiver-operating characteristic (ROC) curve, theratio of area-under curve (AUC) is 0.6673 for the instability index and 0.8836 for the logistic regression model. These results suggest that the logistic regression model is a preferred method than instability index in the assessment of mass wasting susceptibility.

並列關鍵字

Logistic regression ； Instability Index ； Mass wasting ； classification error matrix ； ROC