| 摘要: | 地震造成的損失除了直接導因於地震動外，地震動所引發的山崩也是生命財產損失之重要原因，因此，地震發生後快速且準確地知道哪裡可能已發生山崩，對災後救援或緊急應變對策研擬至關重要。
本研究目標為半自動化近即時繪製山崩潛感圖，盡量減少人工資料處理時間，本研究使用羅吉斯回歸計算潛感值，模型選擇了坡度、坡度粗糙度、地表粗糙度、總曲率、全坡高、愛氏震度、坡向、岩性。本研究依愛氏震度(Arias Intensity)進行資料分區，並根據兩種訓練資料取樣方法設計出兩種模型，以計算不同震度下的羅吉斯回歸模型潛感值，並評估不同震度下先天因子對潛感值之影響程度。
本研究撰寫了半自動化Python程式計算羅吉斯回歸模型潛感值，經過測試可以在1分鐘內建立完山崩潛感模型，並在15分鐘左右繪製出山崩潛感圖。未來希望能結合餘震預測的技術，當主震發生時，先利用已建立好的山崩潛感模型，建立第一版之山崩潛感圖，待取得遙測技術自動判釋取得山崩目錄以及強震站的地震訊號後，再以主震誘發山崩之山崩目錄建立新的模型，配合餘震預測技術，所得到餘震可能造成的震度空間分布，即可預測餘震可能造成的山崩分布，做為主震後救援或緊急應變之重要參考資料。
;The losses caused by earthquakes are not only directly attributable to the seismic shaking but also to the significant contribution of landslides triggered by the seismic activity. Therefore, it is crucial to quickly and accurately determine the locations where landslides may have occurred after an earthquake, in order to develop effective strategies for post-disaster rescue and emergency response.
The objective of this study is to generate semi-automatically near-real-time susceptibility maps for landslides, aiming to minimize manual data processing time. In this study, logistic regression is used to calculate the susceptibility. The model selected the following factors: slope, slope roughness, terrain roughness, total curvature, slope height, Arias Intensity, aspect, and lithology. The data is partitioned based on the Arias Intensity, and two models are designed using two different training data sampling methods to calculate the susceptibility values for different intensities. The study also evaluates the influence of inherent factors on susceptibility values under different intensities.
This study developed a semi-automated Python program to calculate susceptibility. Through testing, it was found that the program can establish a landslide susceptibility model within one minute and generate a susceptibility map in approximately 15 minutes. In the future, the aim is to integrate techniques for aftershock prediction. When a main shock occurs, the established landslide susceptibility model can be used to create an initial version of the susceptibility map. Then, with the automatic interpretation of remote sensing data for landslide inventories and seismic signals from strong motion stations, a new model can be constructed using the landslide catalog induced by the main shock. By combining this with aftershock prediction techniques and the spatial distribution of shaking intensity caused by aftershocks, it will be possible to predict the distribution of landslides that may occur as a result of aftershocks. This information can serve as an important reference material for rescue or emergency response after a main shock. |
