TY - JOUR
T1 - Landslide susceptibility analysis with logistic regression model based on FCM sampling strategy
AU - Wang, Liang Jie
AU - Sawada, Kazuhide
AU - Moriguchi, Shuji
N1 - Funding Information:
The authors gratefully acknowledge the support for this research from Young Researches Education Center for Innovation (Grant no. 50H2300031 ) and Gifu construction research center.
Copyright:
Copyright 2018 Elsevier B.V., All rights reserved.
PY - 2013/8
Y1 - 2013/8
N2 - Several mathematical models are used to predict the spatial distribution characteristics of landslides to mitigate damage caused by landslide disasters. Although some studies have achieved excellent results around the world, few studies take the inter-relationship of the selected points (training points) into account. In this paper, we present the Fuzzy c-means (FCM) algorithm as an optimal method for choosing the appropriate input landslide points as training data. Based on different combinations of the Fuzzy exponent (m) and the number of clusters (c), five groups of sampling points were derived from formal seed cells points and applied to analyze the landslide susceptibility in Mizunami City, Gifu Prefecture, Japan. A logistic regression model is applied to create the models of the relationships between landslide-conditioning factors and landslide occurrence. The pre-existing landslide bodies and the area under the relative operative characteristic (ROC) curve were used to evaluate the performance of all the models with different m and c. The results revealed that Model no. 4 (m=1.9, c=4) and Model no. 5 (m=1.9, c=5) have significantly high classification accuracies, i.e., 90.0%. Moreover, over 30% of the landslide bodies were grouped under the very high susceptibility zone. Otherwise, Model no. 4 and Model no. 5 had higher area under the ROC curve (AUC) values, which were 0.78 and 0.79, respectively. Therefore, Model no. 4 and Model no. 5 offer better model results for landslide susceptibility mapping. Maps derived from Model no. 4 and Model no. 5 would offer the local authorities crucial information for city planning and development.
AB - Several mathematical models are used to predict the spatial distribution characteristics of landslides to mitigate damage caused by landslide disasters. Although some studies have achieved excellent results around the world, few studies take the inter-relationship of the selected points (training points) into account. In this paper, we present the Fuzzy c-means (FCM) algorithm as an optimal method for choosing the appropriate input landslide points as training data. Based on different combinations of the Fuzzy exponent (m) and the number of clusters (c), five groups of sampling points were derived from formal seed cells points and applied to analyze the landslide susceptibility in Mizunami City, Gifu Prefecture, Japan. A logistic regression model is applied to create the models of the relationships between landslide-conditioning factors and landslide occurrence. The pre-existing landslide bodies and the area under the relative operative characteristic (ROC) curve were used to evaluate the performance of all the models with different m and c. The results revealed that Model no. 4 (m=1.9, c=4) and Model no. 5 (m=1.9, c=5) have significantly high classification accuracies, i.e., 90.0%. Moreover, over 30% of the landslide bodies were grouped under the very high susceptibility zone. Otherwise, Model no. 4 and Model no. 5 had higher area under the ROC curve (AUC) values, which were 0.78 and 0.79, respectively. Therefore, Model no. 4 and Model no. 5 offer better model results for landslide susceptibility mapping. Maps derived from Model no. 4 and Model no. 5 would offer the local authorities crucial information for city planning and development.
KW - FCM
KW - Landslide susceptibility
KW - Logistic regression model
UR - http://www.scopus.com/inward/record.url?scp=84877835260&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84877835260&partnerID=8YFLogxK
U2 - 10.1016/j.cageo.2013.04.006
DO - 10.1016/j.cageo.2013.04.006
M3 - Article
AN - SCOPUS:84877835260
VL - 57
SP - 81
EP - 92
JO - Computers and Geosciences
JF - Computers and Geosciences
SN - 0098-3004
ER -