透過您的圖書館登入
IP:44.201.97.0
  • Journals

結合時空因子與InSAR觀測資料之地表崩塌變位預測分析

PREDICTION OF LANDSLIDES USING MACHINE LEARNING TECHNIQUES BASED ON SPATIO-TEMPORAL FACTORS AND INSAR DATA

Abstracts


台灣山區陡峭地形和破碎地質,當颱風豪雨後坡地崩塌災害頻傳。複雜地理環境和極端氣候影響,如何有效運用空間觀測數據預測崩塌已是現行災害管理課題。隨著遙感探測技術發展,如衛星光譜影像和干涉合成孔徑雷達能以固定週期獲得地表屬性和幾何資訊。本研究運用新式遙測資訊,透過斜坡單元分析量化共十四種時空因子,由相關性分析偵測影響崩塌的因子,機器學習建構崩塌預測模型,依混淆矩陣評估預測品質。研究顯示崩塌預測成果優於80%正確率,並指出影響斜坡單元崩塌時空因素,為國土保育提供明確養護位址,並作工程選址、邊坡防護管理應用。

Parallel abstracts


Taiwan's mountainous areas feature steep terrain and broken geological environment. With the invasion of typhoons or heavy rain events, slope disasters such as landslide and debris flow can occur frequently. As a result, how to effectively apply multiple spatial data to predict landslide has become a forward-looking topic for current disaster management. The development of remote sensing detection technology, such as satellite spectral images and interferometric synthetic aperture radar (InSAR) collected in a fixed period is able to help obtain information on earth surface properties and geometric changes. Through the spatial analysis of the slop units, this research quantifies a total of 14 spatiotemporal factors. The correlation analysis is employed to detect factors that significantly cause landslides, and then artificial intelligence machine learning is used to construct a landslide prediction model. Finally, the confusion matrix verifies the prediction results and evaluates the quality. The research shows that the landslide prediction is better than 80% correct and points out the spatiotemporal factors that affect the slope unit collapse in the survey area. This result also suggests clearer conservation sites and potential assessment for land conservation. The overall study will be effectively used as the application for project site selection, slope protection, and disaster prevention management.

Read-around