Comparison of landslide susceptibility maps using random forest and multivariate adaptive regression spline models in combination with catchment map units

被引:30
作者
Chu, Lei [1 ,2 ]
Wang, Liang-Jie [1 ,2 ]
Jiang, Jiang [1 ,2 ]
Liu, Xia [1 ,2 ]
Sawada, Kazuhide [3 ]
Zhang, Jinchi [1 ,2 ]
机构
[1] Nanjing Forestry Univ, Collaborat Innovat Ctr Sustainable Forestry South, Nanjing 210037, Jiangsu, Peoples R China
[2] Nanjing Forestry Univ, Jiangsu Prov Key Lab Soil Eros & Ecol Restorat, Nanjing 210037, Jiangsu, Peoples R China
[3] Gifu Univ, Grad Sch Engn, Gifu 5011193, Japan
基金
中国国家自然科学基金;
关键词
landslide susceptibility; multivariate adaptive regression spline (MARSpline) model; random forest (RF) model; catchment map units (CMUs); ARTIFICIAL NEURAL-NETWORKS; ANALYTICAL HIERARCHY PROCESS; SUPPORT VECTOR MACHINE; LOGISTIC-REGRESSION; FREQUENCY RATIO; SPATIAL EVIDENCE; DECISION-TREE; SAMPLING STRATEGIES; FUZZY-LOGIC; AREA;
D O I
10.1007/s12303-018-0038-8
中图分类号
P [天文学、地球科学];
学科分类号
07 ;
摘要
Landslide susceptibility mapping (LSM) is a critical tool for mitigating the damages caused by geologic disasters. The selection of map units and mathematical models greatly affects the efficiency of LSM. To obtain the most appropriate combination of map units and mathematical models, four scales of catchment map units (CMUs) were analyzed and random forest (RF) and multivariate adaptive regression spline (MARSpline) models were applied in Gero City, Japan. The percentage of correctly identified landslides and the areas under the relative operating characteristic (ROC) curve were used to evaluate the model performances. The results indicate that the RF model had higher prediction accuracy than the MARSpline model, especially when the size of the CMU was 0.09 km(2). A relatively high percentage of landslides fell into the high and very high landslide susceptibility classes (73%) and the lowest percentage of landslides fell into the very low landslide susceptibility classes (0.82%). The prediction-area (P-A) plots indicated that the prediction rates were higher for the RF model than the MARSpline model. The results of this study also suggest that the model accuracy can be increased if the appropriate CMU size is used. Therefore, the potential benefits of using the RF model in combination with the appropriate CMU size should be further explored using additional landslide-conditioning factors and other models.
引用
收藏
页码:341 / 355
页数:15
相关论文
共 53 条
  • [1] A novel ensemble bivariate statistical evidential belief function with knowledge-based analytical hierarchy process and multivariate statistical logistic regression for landslide susceptibility mapping
    Althuwaynee, Omar F.
    Pradhan, Biswajeet
    Park, Hyuck-Jin
    Lee, Jung Hyun
    [J]. CATENA, 2014, 114 : 21 - 36
  • [2] [Anonymous], 2009, R FDN STAT COMP, P409
  • [3] GIS-based logistic regression for landslide susceptibility mapping of the Zhongxian segment in the Three Gorges area, China
    Bai, Shi-Biao
    Wang, Jian
    Lue, Guo-Nian
    Zhou, Ping-Gen
    Hou, Sheng-Shan
    Xu, Su-Ning
    [J]. GEOMORPHOLOGY, 2010, 115 (1-2) : 23 - 31
  • [4] Random forests
    Breiman, L
    [J]. MACHINE LEARNING, 2001, 45 (01) : 5 - 32
  • [5] Application of frequency ratio, statistical index, and index of entropy models and their comparison in landslide susceptibility mapping for the Baozhong Region of Baoji, China
    Chen, Wei
    Li, Wenping
    Hou, Enke
    Bai, Hanying
    Chai, Huichan
    Wang, Danzhi
    Cui, Xueli
    Wang, Qiqing
    [J]. ARABIAN JOURNAL OF GEOSCIENCES, 2015, 8 (04) : 1829 - 1841
  • [6] Evaluation of prediction capability of the artificial neural networks for mapping landslide susceptibility in the Turbolo River catchment (northern Calabria, Italy)
    Conforti, Massimo
    Pascale, Stefania
    Robustelli, Gaetano
    Sdao, Francesco
    [J]. CATENA, 2014, 113 : 236 - 250
  • [7] Landslide susceptibility assessment in the Hoa Binh province of Vietnam: A comparison of the Levenberg-Marquardt and Bayesian regularized neural networks
    Dieu Tien Bui
    Pradhan, Biswajeet
    Lofman, Owe
    Revhaug, Inge
    Dick, Oystein B.
    [J]. GEOMORPHOLOGY, 2012, 171 : 12 - 29
  • [8] Mapping landslide susceptibility with logistic regression, multiple adaptive regression splines, classification and regression trees, and maximum entropy methods: a comparative study
    Felicisimo, Angel
    Cuartero, Aurora
    Remondo, Juan
    Quiros, Elia
    [J]. LANDSLIDES, 2013, 10 (02) : 175 - 189
  • [9] Comparing landslide inventory maps
    Galli, Mirco
    Ardizzone, Francesca
    Cardinali, Mauro
    Guzzetti, Fausto
    Reichenbach, Paola
    [J]. GEOMORPHOLOGY, 2008, 94 (3-4) : 268 - 289
  • [10] Variable Importance Assessment in Regression: Linear Regression versus Random Forest
    Groemping, Ulrike
    [J]. AMERICAN STATISTICIAN, 2009, 63 (04) : 308 - 319