結合地理資訊系統與機器學習法於淹水潛勢分析

近年來氣候災害所造成的危害程度，隨著人口的增加、社會經濟的發展及極端降雨事件日趨嚴重。台灣也因極端降雨事件，使得旱澇加劇越來越頻繁，其中颱風事件所帶來豪雨，使大量雨水在短時期內傾盆下降，致使平原地帶、較低地區氾濫成災，造成洪水災害。淹水的發生不僅造成經濟上的損失，也使人民性命受到威脅。本研究建立淹水潛勢評估模式，進而繪製淹水潛勢地圖，提供作為防災參考依據，以達到減災之目的。本研究選擇台灣東北部蘭陽溪流域為研究區域。蒐集2004年到2015年的颱風淹水事件及十二個淹水影響因子，包含了高程、坡度、坡向、曲率、平面曲率、剖面曲率、地形濕度指數、逕流強度指數、與河川的距離、岩性、土地使用、最大小時累積雨量。研究中選用了三種不同的機器學習法來建立淹水潛勢評估模式，分別為決策樹、裝袋決策樹及隨機森林；以交叉驗證的方式驗證各模式淹水判別成效，比較決策樹、裝袋決策樹及隨機森林這三種方法建立的淹水潛勢式評估模式；建立混淆矩陣來分析各模式準確率。另外，利用接收者操作特徵曲線與其曲線下面積，評鑑三種模式之識別力。各個模式在淹水潛勢判別上都有不錯的表現，其中又以隨機森林所建立之淹水潛勢評估模式表現最佳。隨機森林能準確模擬研究區域淹水分布之特性，並能展現當地地形因子特性。另外，以多次運行隨機森林模式，證明隨機森林有非常良好地強健性。最後，分析潛在淹水因子之貢獻度，貢獻度越大，代表對於淹水影響越大，最大小時累積降雨高達35%，其次為與河川的距離、地形濕度指數和逕流強度指數。未來可以根據本研究所發展之淹水潛勢評估模式，協助相關管理機關擬訂適當的防災策略。

關鍵字

淹水；淹水潛勢評估模式；決策樹；裝袋決策樹；隨機森林

並列摘要

In recent years, rapid urban growth and climate change have led to many environmental problems and increased risks of natural disasters, including flooding, landslide and drought. The typhoon rainfall is an important source of water resources. However, the heavy rainfall brought by typhoons frequently result in serious disasters. Floods have become more and more frequent in recent decades. Therefore, flood susceptibility models can efficiently mitigate the disasters are desired. The study area is the Lanyang river basin in northeastern Taiwan. All the flood-related factors and the historical flooding data are based on GIS technology. Flood events from 2004 to 2015 are collected. Moreover, twelve flood-related factors are used in the flood susceptibility analysis, namely, elevation, slope, aspect, curvature, plan curvature, profile curvature, topographic wetness index, stream power index, distance to river, lithology, land use and maximum hour rainfall. In this study, three methods are employed to construct flood susceptibility models for the study area. The three methods include the decision tree (DT), bagging tree (BT) and random forest (RF). The performances of three models are checked by the accuracy and the area under the receiver operating characteristic curve (AUC). For flood susceptibility maps, the DT-based model and the BT-based model underestimated the actual flooded area. However, the RF-based model has a good consistency between the correctly prediction and actual maps. The RF-based model outperforms over the DT-based model and the BT-based model in the study area. The RF-based model has efficiently improvement for flood susceptibility assessment. The maximum 1-hour rainfall, distance to river, topographic wetness index, and stream power index are the top four most important of the twelve flood-related factors. In conclusion, the proposed model is validated in this study, and it can be used for planning protective and mitigation measures.

並列關鍵字

Flood ； Flood susceptibility model ； Decision tree ； Bagging tree ； Random forest

參考文獻

1. Akhoondzadeh, M. (2016). "Decision Tree, Bagging and Random Forest methods detect TEC seismo-ionospheric anomalies around the time of the Chile, (Mw = 8.8) earthquake of 27 February 2010." Advances in Space Research 57(12): 2464-2469.

2. Alfieri, L., Salamon, P., Bianchi, A., Neal, J., Bates, P. and Feyen, L. (2014). "Advances in pan-European flood hazard mapping." Hydrological Processes 28(13): 4067-4077.

4. Breiman, L. (1996). "Bagging Predictors." Machine Learning 26(2): 123-140.

6. Bui, D.T., Ho, T.C., Pradhan, B., Pham, B.T., Nhu, V.H. and Revhaug, I. (2016). "GIS-based modeling of rainfall-induced landslides using data mining-based functional trees classifier with AdaBoost, Bagging, and MultiBoost ensemble frameworks." Environmental Earth Sciences 75(14): 1101.

7. Dimitriadis, P., Tegos A., Oikonomou A., Pagana, V., Koukouvinos, A., Mamassis, N., Koutsoyiannis, D. and Efstratiadis, A. (2016). "Comparative evaluation of 1D and quasi-2D hydraulic models based on benchmark and real-world applications for uncertainty assessment in flood mapping." Journal of Hydrology 534: 478-492.

國際替代計量

結合地理資訊系統與機器學習法於淹水潛勢分析

全文下載

主題瀏覽