廣域崩塌潛勢模型的空間分析

Spatial Analysis for Regional Landslide Susceptibility Modeling

賴哲儇(Jhe-Syuan Lai) ； 蔡富安(Fuan Tsai) ； 姜壽浩(Shou-Hao Chiang)

《航測及遙測學刊》 22卷2期 (2017/06) Pp. 93-104

https://doi.org/10.6574/JPRS.2017.22(2).2

摘要

針對廣域崩塌潛勢模型空間分析，本研究於崩塌目錄中劃分崩塌源頭、拖曳帶與堆積區(後兩者稱之Run-out)等細緻類別，並利用隨機森林演算法和崩塌潛勢相關的空間資料建立廣域的崩塌潛勢模型。使用的空間資料圖層包含地形與植被等網格資料，以及斷層、水系、道路、地質和土壤等GIS 圖資。流程上，崩塌目錄的源頭與Run-out 區域轉換至像元型態後，與隨機取樣而得的非崩塌地樣本一起萃取對應的崩塌潛勢圖層，形成匯入隨機森林演算法的資料表。分析上，本研究探討不同樣本數、數值高程模型品質和演算法差異，對於崩塌潛勢模型的影響。藉由上述流程與分析，證明Run-out 獨立成單一類別的可行性，並作為Run-out 影響塌潛勢模型的先期成果。研究成果顯示，獨立Run-out 類別的整體精度可達85% ~95%，Kappa 值約介於0.8 至0.9 間；在樣本足夠的情形下，Run-out 類別的使用者精度及生產者精度甚至可高達0.9。本研究亦證實DEM 編修後配合隨機森林演算法，能得到較理想的模式化成果。藉由崩塌潛勢模型，本研究進而產生研究區域的崩塌潛勢圖，以期作為後續崩塌災害、風險等評估任務的基礎，並輔助崩塌災害相關的規劃和決策。

關鍵字

空間分析 ； 莫拉克颱風 ； 崩塌潛勢 ； 隨機森林

並列摘要

This study adopts a data mining-based algorithm, the random forests classifier, with geo-spatial data to construct a regional landslide susceptibility model on the basis of Typhoon Morakot in 2009. The developed model also takes account of landslide source, Run-out and non-occurrence signatures from the landslide inventory in order to increase the reliability of the susceptibility modeling. A total of ten causative factors were collected and used in this study, including topographic, vegetative, fault, geology, river, road, and soil data. This study transforms the landslide inventory and vector-based causative factors into the pixel-based format in order to overlay with other raster data for constructing the random forests based model. This study also uses different topographic data, numbers of samples and classifiers in the analysis to understand their impacts on the susceptibility modeling. Experimental results demonstrate that after identifying the Run-out signatures, the overall accuracy and Kappa coefficient have increased to be more than 85 % and 0.8, respectively. In addition, the user's accuracy and producer's accuracy of the Run-out class can reach 0.9 when the number of samples is sufficient to analyze. Correcting unreasonable topographic features of the digital terrain model and using the random forests algorithm also produce more reliable modeling results. According to the modeling results, a preliminary landslide susceptibility map of the study site is produced. Based on this map, future researches may further include other information to achieve landslide hazard and risk assessments as well as to assist land planning and policy marking.

並列關鍵字

Landslide Susceptibility ； Random Forests ； Spatial Analysis ； Typhoon Morakot