高屏溪流域林地脆弱度評估及崩塌地潛在植生恢復機制之研究

臺灣地處菲律賓海洋板塊及歐亞大陸板塊交接處，地質組成較複雜，當地震、颱風及暴雨侵襲時，容易引起崩塌及土石流等土砂災害。旗山、荖濃溪事業區屬高屏溪上游集水區之國有林班地，其為2009年莫拉克颱風重創之地區，林地崩塌災害嚴重，而崩塌災害前後之植生覆蓋變遷為集水區治理的重要工作。遙測衛星影像具有固定時間、大範圍取得地面資料之能力，可監測崩塌地分布情況及植生復育狀態。本研究收集1960年-2015年TCCIP降尺度降雨量資料，藉由Mann-Kendall趨勢檢定、Theil-Sen斜率推估、Cumulative Deviations檢定和Mann-Whitney-Pettitt檢定，探討長期降雨趨勢變化與變異點；利用2000年-2015年Landsat 系列衛星影像，以人工數化取得各年度崩塌地圖層，並以地理資訊系統將崩塌地分類成維持不變型、植生恢復型、新增型及擴崩型，分析歷年各類型崩塌地之分布與植生自然恢復狀況；再由遙測光譜資訊透過纓帽轉換獲取2000年、2009年與2015年三年間森林干擾程度變化；以地景指標探討不同類型崩塌之地景結構變異；以Kolmogorov-Smirnov常態檢定環境變數，不安定指數法與羅吉斯回歸法繪製崩塌潛勢模型，並以分類誤差矩陣和ROC曲線下面積(AUC)比較兩者準確性；利用主成分分析法縮減環境變數維度及分配各成分之權重，推估環境脆弱度之空間分布，此外歷年植生恢復之塊體萃取環境變數，利用羅吉斯回歸推估現有崩塌塊體植生恢復之可能性。本研究結果顯示，1960-2015年旗山事業區與荖濃溪事業區於四季、豐水期、枯水期和逐年降雨量均呈現顯著增加趨勢，且降雨趨勢變化之改變點落在1975年-2000年間。崩塌確實在旗山與荖濃溪事業區造成嚴重干擾，旗山於2000年、2009年與2015年之整體森林干擾指標值為2.128、2.751、2.670，代表2009年莫拉克颱風嚴重侵害，爾後於2015年有些許恢復，荖濃溪事業區各年度之整體干擾程度差異不大，以小樣區則可發現地區性干擾程度差異變動。由崩塌地之地景變遷分析可知，維持不變型崩塌地屬於破碎化地景，且地質結構不穩定，造成風災害後發生更多的擴崩地；新增型崩塌地主要受莫拉克颱風侵襲而生成，原因在於集中且強度降雨，使得地表土壤無法承受外來的重力，產生崩塌；植生恢復型在各年度恢復情形雖然良好，但恢復面積遠不及新崩塌面積。以不安定指數法推估旗山與荖濃溪之崩塌潛勢，其AUC分別為0.702與0.750；羅吉斯回歸法推估崩塌潛勢之AUC為0.819與0.808，整體評估模式優於不安定指數法。旗山與荖濃溪事業區之環境脆弱度，次嚴重與嚴重等級分別佔38%和31%，多分布於地勢陡峭之處，較不利於植被更新與恢復；羅吉斯回歸推估潛在植生恢復區域，其準確度分別旗山為72.59%，荖濃溪為67.76%，與脆弱度等級相比，無法恢復大多分布於次嚴重與嚴重之等級。面對極端氣候影響，森林集水區經營應順應生態經營之原則，考慮森林結構及生態功能，以維護生態系平衡。

關鍵字

降雨趨勢分析；森林干擾；地景生態學；崩塌潛勢；環境脆弱度；潛在植生恢復

並列摘要

Taiwan is located on the Pacific seismic belt where the geological structure is more complex, and cause frequent landslides, debris flows and other sediment disasters, especially when earthquakes, typhoons and heavy rains occurs. Chishan and Lennon river working circle lays in the national forest land of Kaoping river basin, and was greatly damaged by 2009 Morakot typhoon, thus, the understanding of land cover change before and after typhoon were important tasks in watershed management. However, satellite images enables the monitoring of large-scale landslide distributions. In this study, down-scale precipitation data was collected from TCCIP. The variation of long-term precipitation trend and change points was obtained by Mann-kendall trend verification, Theil-sen slope estimation, Cumulative Deviations verification and Mann-Whitney-Pettitt verification. Landsat images from the year 2000-2015 were acquires as materials and the digitizing of landslide distribution was completed by Geographic Information Systems (GIS). Accordingly, four types of landslide were classified, namely are unchanged, recovery, appending and expanding. The variation of forest disturbance degree from 2000-2015 was obtained through Tasseled Cap Transformation. Landscape metrics was applied to analyze the variation of landscape structure in different landslide types over the years. The environmental variables were examined by Kolmogorov-Smirnov test, and estimating landslide susceptibility via Instability index and Logistic regression, and the accuracy was reviewed by comparing Error Matrix and Area-under Curve. Estimating the distribution of environmental vulnerability through Principal Components Analysis, reduces the number of environmental variables and determines the weight of principal components. Moreover, the historical vegetation restoration data and environmental variables were used for evaluation of the vegetation restoration possibility in the existing landslide area by Logistic regression. The results shows that precipitation of season, wet period, dry period and annual precipitation at Chishan and Lennon river working circle has increased. Additionally, the change point of precipitation trend falls within 1975-2000. Landslides has indeed disturbed Chishan and Lennon river working circle. The IDI of Chishan were 2.128, 2.751 and 2.670, respectively, which represents that some areas were damaged by Morakot typhoon were partially recovered after 2015. There were no significant change in the annual IDI of Lennon river working circle. However, differences between each DI level of sample plot was found. The results of landscape metrics showed that the unchanged-type tend to be fragmented and geologically unstable, which leads to frequent natural disasters and caused existing landslide area to expand. The appending-type was mostly triggered by the Morakot typhoon, where heavy rainfalls hits the ground surface, and made surface objects unable to stand firmly. The recovery-type recovers gradually; however, the area of landslide formation is significantly faster than the area of recovery. The landslide susceptibility results indicates the receiver-operating characteristic (ROC) curve and the ratio of area-under curve (AUC) in Chishan and Lennon were 0.702 and 0.750 under the Instability index, 0.819 and 0.808 under the Logistic regression model, that the Logistic regression model is more appropriate than the Instability index. The environment vulnerability in Chishan and Lennon River were classified as heavy and extreme level of environmental vulnerability accounts for 38% and 31% of the area, mostly are located in relatively high altitude and steep terrain, where vegetation renewal and restoration is unlikely to happen. Estimating potential vegetation restoration ability in existing landslide area by Logistic regression, shows an overall correctness of 72.59% in Chishan working circle, and 67.76% in Lennon River working circle. Facing extreme climate challenges, the management of forest watershed must be based on the principle of ecological management, forest structure consideration and ecological function, in order to maintain the balance of the ecosystem.

並列關鍵字

Precipitation trend analysis ； Forest disturbance ； Landscape ecology ； Landslide susceptibility ； Environmental vulnerability ； Potential vegetation restoration