Determining the Suitability of Two Different Statistical Techniques in Shallow Landslide (Debris Flow) Initiation Susceptibility Assessment in the Western Ghats

Abstract

In the present study, the Information Value (InfoVal) and the Multiple Logistic Regression (MLR) methods based on bivariate and multivariate statistical analysis have been applied for shallow landslide initiation susceptibility assessment in a selected subwatershed in the Western Ghats, Kerala, India, to determine the suitability of geographical information systems (GIS) assisted statistical landslide susceptibility assessment methods in the data constrained regions. The different landslide conditioning terrain variables considered in the analysis are geomorphology, land use/land cover, soil thickness, slope, aspect, relative relief, plan curvature, profile curvature, drainage density, the distance from drainages, lineament density and distance from lineaments. Landslide Susceptibility Index (LSI) maps were produced by integrating the weighted themes and divided into five landslide susceptibility zones (LSZ) by correlating the LSI with general terrain conditions. The predictive performances of the models were evaluated through success and prediction rate curves. The area under success rate curves (AUC) for InfoVal and MLR generated susceptibility maps shows 84.11% and 68.65%, respectively. The prediction rate curves show good to moderate correlation between the distribution of the validation group of landslides and LSZ maps with AUC values of 0.648 and 0.826 respectively for MLR and InfoVal produced LSZ maps. Considering the best fit and suitability of the models in the study area by quantitative prediction accuracy, LSZ map produced by the InfoVal technique shows higher accuracy, i.e. 82.60%, than the MLR model and is more realistic while compared in the field and is considered as the best suited model for the assessment of landslide susceptibility in areas similar to the study area. The LSZ map produced for the area can be utilised for regional planning and assessment process, by incorporating the generalised rainfall conditions in the area.

Spatial modelling of shallow landslide susceptibility: a study from the southern Western Ghats region of Kerala, India.

“…The results of this study are comparable with the other studies conducted in Ghats region as well as other parts of the world with similar environmental conditions. Krishnan et al (2015) studied landslide susceptibility of upland catchment of Meenachil River in southern Western Ghats using bivariate Information value method with MLR model and reported bivariate method gave higher accuracy over the MLR. Vijith et al (2014) used the weight of evidence model to assess the landslide susceptibility of southern Western Ghats with an accuracy of 89.2%.…”

“…slopes where zero landslides are selected and on the basis of expert judgement these slope pixels are converted to absence locations). The purpose of using MLR in landslide modelling is to find out the best fitting model to describe the relation between the presence and absence of landslides (Akgun, Kıncal, and Pradhan 2012;Krishnan et al 2015). The logistic regression model representing the maximum likelihood regression model can be expressed as Equation (8)…”

“…DSM is a modified form of theory of evidence and the main advantage is that the belief, disbelief, uncertainty and plausibility associated with landslide occurrence can be quantified statistically (Althuwaynee, Pradhan, and Lee 2012). Logistic regression is well known for its ability to predict the presence or absence of a phenomenon or outcome based on the values of the input predictor variables (Krishnan et al 2015). Hence, a robust comparison of these techniques will be useful in choosing the best method for future landslide susceptibility modelling.…”

“…Hence, a robust comparison of these techniques will be useful in choosing the best method for future landslide susceptibility modelling. The previous studies, in this regard (Pourghasemi, Mohammady, and Pradhan 2012;Pourghasemi et al 2013;Vijith et al 2014;Krishnan et al 2015;Chen et al 2015;Ding, Chen, and Hong 2016;Mondal and Mandal 2018;Bera, Guru, and Ramesh 2019), are mainly focused on explaining the overall accuracy of the models with receiver operating curve and area under curve value (ROC-AUC), which is difficult to interpret and unable to find the prediction accuracy of both landslide pixels and non-landslide pixels. We address the issue by using matrix-based sensitivity, specificity and classification accuracy methods and quantifying the prediction error through RMSE.…”

Quantitative Hazard Assessment of Landslides Using the Levenburg–Marquardt Back Propagation Neural Network Method in a Pipeline Area

“…It is important to note that landslides play a significant part in the history of geomorphic development, and landslides, at their essence, are a geomorphologic process on a slope. A slope with topographic profile curvature greater than 0.5 is classified as a convex slope, slopes with less than −0.5 are classified as concave slopes, and the rest of the slopes are classified linear slopes [38]. The curvature distribution of the slopes in the study area is complex (Figure 3e).…”