運用隨機森林探討莫拉克颱風災區土石流發生因子關連性

本研究採用遙測影像、現地調查結果、農委會水保局2006～2010年土石流易致災因子調查資料及各項分析成果，萃取莫拉克颱風災區218條土石流潛勢溪流土石流發生三元素作為評估因子（地形、降雨及土砂來源），合計84個因子，以主成分分析與相關性分析，篩選莫拉克颱風引發土石流之18個顯著發生因子，採用美國貝爾實驗室開發的隨機森林演算法，針對多變量因子間關聯性進行研究與探討，發現莫拉克颱風時，土石流顯著發生因子分別是「有效累積雨量」影響最大，「時雨量」次之，之後依序「各潛勢溪流線50公尺環域範圍內之崩塌率」、「10度以上有效集水區高度平均值」等。本研究除期望釐清莫拉克颱風引致土石流發生因子特性外，亦可回饋分析結果於易致災因子調查，研修土石流易致災調查發生因子項目，主動於土石流潛勢溪流成災前進行防範與疏散，減少災害損失，防患於未然。

關鍵字

莫拉克颱風；土石流發生因子；主成分分析；隨機森林演算法

並列摘要

Typhoon Morakot lashed Taiwan during Aug. 7 to 9, 2009. It dumped heavy rainfall in southern Taiwan, especially around the Central Mountain Range in Pingtung, Chia-Yi, and Kaohsiung County. In view of this, comprehensive field investigation was carried out by government and private organizations after Typhoon Morakot, useful information of debris flow was gathered. Besides, after Typhoon Morakot, the debris flow-inducing factors become more challenging in Taiwan, many aspects had to be considered. The scope of this study was mainly discussed in debris flow-inducing factors in serious damaged areas which including Nantou, Chia-Yi, Tainan, Kaohsiung, Pingtung, Taitung County. Totally 218 torrents were included. Field investigation data and disaster records of Typhoon Morakot were utilized to analyze debris flow-inducing factors in three aspects: terrain, rainfall and sediment source. First, by using Principle Component Analysis (PCA) and Pearson Product Moment Correlation Analysis (CA) to select significant factors, 84 factors were reduced to the most important 18 factors. Then through descriptive statistics and scatter diagram were selected to discuss the correlation among ”average slope gradient of watershed”, ”landslide rate along the stream within 50 m buffer zone” as well as the ”rainfall intensity during Typhoon Morakot”. The above charts of characteristics analysis were used to summarize the range of factor value which tend to occur phenomenon of debris flow in Typhoon Morakot. Besides, Random Forest Algorithm (RF) was utilized to research the relationship toward multi-variables. The significant factors which tend to affect the debris flow-inducing factor were ”effective accumulated rainfall”, ”hourly rainfall”, ”landslide rate along the stream within 50 m buffer zone”, ”average elevation value of effective watershed which higher than 10 degree”, sequentially. By the results, the most significant factor is the rainfall factor during Typhoon Morakot. The results can be utilized in improving debris flow hazard management in the future.

並列關鍵字

Typhoon Morakot ； Debris Flow-inducing Factors ； Principal Component Analysis ； Random Forest Algorithm

參考文獻

Breimans, L.(2001).Random forests.Machine Learning.45(1),5-32.

Google Scholar

Chen, C. Y.,Yu, F. C.(2011).Morphometric analysis of debris flows and their source areas using GIS.Geomorphology.129,387-397.

Google Scholar

Chow, V. T.(ed.)(1964).Handbook of Applied Hydrology Section.New York:McGrall-Hill.

Google Scholar

Chung, C. J. F.,Fabbri, A. G.(1999).Probabilistic prediction models for landslide hazard mapping.Photogrammetric Engineering and Remote Sensing.65,1389-1399.