ABSTRACT
USING STEPWISE LOGISTIC REGRESSION TO SELECT IMPORTANT LANDSLIDE CAUSAL FACTORS IN LANDSLIDE
SUSCEPTIBILITY MAPPING
Norbert Simon1, Mairead de Roiste2, Michael Crozier2, Abdul Ghani Rafek1
Rodeano Roslee3
1School of Environment and Natural Resources Sciences, Faculty of Science and Technology, UniversitiKebangsaan Malaysia, 43600 UKM Bangi, Selangor, Malaysia
2School of Geography, Environment and Earth Sciences, Victoria University of Wellington, New Zealand
3School of Science & Technology, Universiti Malaysia Sabah, Jalan UMS, 88400 Kota Kinabalu, Sabah, Malaysia
The aim of this study is to identify landslide influencing factors that causes landslides in the west coast region of Sabah. Thirteen landslide factors were selected in this study. These factors are slope angle, slope curvature (plan, profile, & tangential), slope aspect, elevation, drainage density, lineament density, soil type, lithology, annual rainfall, road density, and land use. A total of 137 landslides were identified through aerial photographs interpretation and these landslides were used for analysis and also for validation. From the analysis, the model identified slope angle, elevation, road density and lithology as importantlandslide causal factors. These factors were later used in GIS to construct a landslide susceptibility map using the logistic regression method. The degree of fit method was used to validate the accuracy of the landslide susceptibility map which shows an accuracy of 70-79% indicating that the landslide susceptibility map has an acceptable accuracy for mapping landslides.
Introduction
Statistical approach has gained serious attention from many researchers over the years in landslide susceptibility mapping (e.g. Pradhan & Lee, 2007; Jaiswalet al., 2010; Lin et al., 2010; Mezughiet al., 2012).This study aims to show how a statistical model can help to determine important landslide causal factors by applying a stepwise procedure. The logistic regression model is used to achieve the study aim.
The study area
The west coast region of Sabah is selected for this study. The study area comprises the state capital Kota Kinabalu and several smaller towns such as Menggatal, Telipok and Tuaran. It covers an area of 387 km2 involving four adjacent 1:50,000 topograhic maps. The study area is bounded by the Crocker Range at the east while most of the central part is a flatland with several hilly area ranging from 20 to 384 m in elevation.
Selection of Data Layers
A landslide inventory map containing 137 landslides serves the most important purpose in this study where it was used together with the landslide causal factors to generate the landslide susceptibility map and also used for validation. Out of the 137 landslides, 69 landslides were used together with the landslide causal factors to construct the landslide susceptibility map and the other 68 landslides were used to validate the map. For modeling purpose, 81 stable units that are free of landslides were randomly selected and were also intersected with all 13 landslide causal factors. The 13 landslide causal factors used in this studyare slope angle, slope curvature (plan, profile, & tangential), slope aspect, elevation, drainage density, lineament density, soil type, lithology, annual rainfall, road density, and land use.The 69 landslides and 81 stable units were analysed using stepwise logistic regression to determine the important landslide causal factors.
Result & Discussion
Based on the analysis, three landslide factors were indicated as significant to landslide occurrences (Table 1). These factors are slope angle, road density and elevation. Although lithology is not indicated as significant, the lithology factor was included by the model to improve the model's performance. The landslide susceptibility map constructed from the four factors using logistic regression method is shown in Figure 1.
Table 1 The coefficient (B), standard error (S.E) and significance (p) values for each landslide factor class
Attributes / B / S.E / Sig (p) / Attributes / B / S.E / Sig (p)Lithology / 0.61 / Slope angle (°) / 0.133 / 0.030 / 0.000
Alluvium / -1.166 / 1.062
Interbedded sandstone and shale / 1.055 / 0.706 / Road density (m/40,000m2) / 0.004 / 0.001 / 0.002
Sandstone / 0.889 / 0.700
Shale / 0a / - / Elevation (m) / -0.010 / 0.002 / 0.000
Constant / -2.426
Figure 1 The landslide susceptibility map based on the logistic regression method
The map was validated using the training and testing datasets. The validation results showed that the training and testing datasets have an accuracy of 70% and 79% respectively (Figure 2). The testing validation result recorded higher accuracy because of higher percentage of landslides presence in the high susceptibility class.
Conclusion
This study has demonstrated the use of stepwise logistic regression in landslide susceptibility mapping. The model is also capable through stepwise statistical analysis to identify factors that significantly influence landslide occurrences in the study area.
References
Jaiswal, P., van Westen, C.J., & Jetten, V. (2010). Quantitative landslide hazard assessment along a transportation corridor in southern India. Engineering Geology, 116, 236–250.
Lin, Y.P., Chu, H.J., & Wu, C.F. (2010). Spatial pattern analysis of landslide using landscape metrics and logistic regression: a case study in Central Taiwan. Hydrology and Earth System Sciences Discussions, 7, 3423–3451.
Mezughi, T., Akhir, J.M., Rafek, A.G., & Abdullah, I. (2012). A Multi-class Weight of Evidence Approach for Landslide Susceptibility Mapping Applied to an Area Along the E-W Highway (Gerik – Jeli), Malaysia. EJGE, 16, 1259-1273.
Pradhan, B. & Lee, S. (2007). Utilization of Optical Remote Sensing Data and GIS Tools for Regional andslide Hazard Analysis Using an Artificial Neural Network Model. Earth Science Frontiers, 14 (6), 143-152.