| 摘要: | 臺灣受到板塊擠壓而成狹長島嶼，具有高山多及坡度陡的地形特徵，居民生活與山地環境的關係密不可分。由於全球暖化的影響，不只全球降雨的強度和頻率發生改變，進一步加速了侵蝕的過程，也使集水區的崩塌風險加劇，為維護與保育水土資源，將水庫集水區劃定為特定水土保持區，惟其劃定範圍與民眾的生活範圍重疊，使得區內的民生活動增加了對地層的擾動。在極端的降雨、劇烈的地表晃動，以及人為的交互作用下， 是造成坡地崩塌的主要因素。極端降雨最著名的例子即為2009年的莫拉克颱風，由於其主要降雨中心以嘉義與高屏山區為主，所帶來的降雨量超越了臺灣的年平均雨量，造成許多坡地災害，其中以臺南縣最為嚴重，位於臺南縣南化區的南化水庫集水區，也因為其特殊地質特性與人為的土地使用，該颱風事件後造成水庫大量的淤積。政府為其訂定特別條例與編列集水區保育經費以提升水庫集水區設施與植生復育，惟其條例與經費分別結束於2016年和2015。因此，後續對其分析崩塌潛勢，有助於面對未來的天然災害。
現今多以事件崩塌目錄評估集水區內崩塌潛勢區域，若能夠利用該地區未發生重大的災害事件下的崩塌地，分析並製作崩塌潛勢圖，詳細地瞭解該地區的基礎崩塌特性，並在天然災害發生之前提前準備，減少災害事件後的經濟損失與人員傷亡。為此本研究使用平日的崩塌目錄，以不安定指數法和羅吉斯迴歸分析南化水庫集水區內的基礎崩塌特性，於篩選後選定9項基礎地文因子(高程、坡度、坡向、地質、地形粗糙度、地形濕度指數、NDVI、距水系距離、距道路距離)，並增加促崩因子(0822豪雨事件與0611豪雨事件，2場降雨事件的總雨量因子)分析。經分析後，主要影響區內坡度穩定度的潛勢因子為坡度、NDVI與地形粗糙度，加入總雨量因子後，距道路距離因子上升幅度最為明顯。
依據上述地文因子與促崩因子以不安定指數法與羅吉斯迴歸，製作區內1個獨立崩塌潛勢模型和2個事件型崩塌潛勢模型。於不安定指數法繪製的模型中，可發現加入降雨條件後，區內的崩塌門檻由潛勢值5至6降至潛勢值4至5的區段間，而羅吉斯迴歸所繪製的模型中，變化較為明顯地則是低潛勢區域與中低潛勢區域。以分類誤差矩陣和ROC曲線評估兩個方法繪製的模型精度，預測崩塌的正確率皆達到50%以上，但模型總正確率則是以不安定指數繪製的模型精度較高，而ROC曲線下的面積(AUC面積)皆達到優良的鑑別力，代表可利用平日崩塌目錄評估區內的基礎崩塌特性，並用於預測南化水庫集水區的崩塌潛勢區域，也建議利用不安定指數法分析區內的潛勢區域，較能夠呈現出平日下的易崩塌區位。另為瞭解南化水庫集水區的植生復育成效，利用兩期分別是2009年與2019年衛星影像，以分析其上、中游的NDVI值，發現集水區經過10年後的NDVI值有顯著上升，顯見其植生復育的良好成效。
Taiwan is a long, narrow island by tectonic plates, it has many high mountains and steep terrain features, which makes the life of Taiwan residents and the mountain environment inseparable. Indirectly changing the intensity and frequency of rainfall in Taiwan due to the global warming, and accelerate the process of erosion in the watershed, in order to maintain and conserve water and soil resources, the reservoir watershed is designated as a Soil and Water Conservation Area,but this area overlaps with the living area of the people, which makes the livelihood activities in the watershed increase the disturbance to the stratum. Extreme rainfall, severe surface shaking, and human interaction are the major factors to cause landslide.In Taiwan, the most famous example of extreme rainfall is Typhoon Morakot in 2009, because its rainfall centers are mainly in Chiayi and Kaohsiung、Pingtung Mountains, the rainfall exceeds the annual average rainfall in Taiwan, causing many sloping land disasters, of which Tainan County is the most serious. The Nanhua Reservoir watershed , which in Tainan Nanhua District due to its special geological characteristics and land use, after typhoon Morakot caused a lot of siltation in the reservoir. The government has formulated special regulations for it and allocated funds for watershed conservation to upgrade reservoir watershed facilities and vegetation restoration, but the regulations and funding ended in 2016 and 2015, respectively. Therefore, susceptibility analysis of landslide will help to prepare future natural disasters.To predict susceptibility analysis of landslide is use the Event-Based Landslide Susceptibility Analysis,if we can use the normal day of landslide to predict susceptibility, we can more accurately understand the basic collapse characteristics of the area, and prepare in advance before natural disasters occur. economic losses and casualties.Therefore, we used the normal day of landslide to predict susceptibility，and use instability index method and logistic regression analysis and basic topograph index, elevation、slope、aspect、geology、topographic roughness index、topographic wetness index、normalized difference vegetation index(NDVI)、distance of drainage、distance of road, and two extremely heavy rain of accumulated precipitation.The major susceptibility landslide index is slope、NDVI and topographic roughness index,the distance of road is increase obviously after join rainfall.
According to the basic topographic factors and landslide-promoting factors, using instability index method and logistic regression analysis to draw one independent landslide susceptibility model and two event-base landslide susceptibility models. In the model which was drawn by the instability index method, it can be found that after adding rainfall conditions, the landslide threshold in the area decrease from the susceptibility value of 5 to 6 to the 4 to 5, and the model which was drawn by the logistic regression analysis, the most obvious changes are the low- susceptibility region and the medium-low susceptibility region. After evaluating the model accuracy with the classification error matrix and Receiver Operating Characteristic curve(ROC curve), the correct rate of predicting landslide is more than 50%, but the total accuracy of the model is higher than that of the model drawn by the instability index method, and the area under the ROC curve (AUC) had excellent discriminative. The representative uses the normal state model to evaluate the basic landslide characteristics in the area, which can be used to predict the landslide susceptibility area of the Nanhua Reservoir catchment area. And to understand the effect of vegetation restoration in the Nanhua Reservoir, two satellite images with a difference of ten years in 2009 and 2019 were used to analyze their NDVI values by cutting the upper and middle reaches. The NDVI value increased significantly after 10 years, indicating that the plant regeneration effect was good after ten years. |
