Wayanad study compares four models of landslide susceptibility

A study of landslide susceptibility in Wayanad District, Kerala, identified Sultan Bathery, Vythiri and Mananthavady among areas modeled as highly susceptible, with steep terrain and heavy rainfall patterns characteristic of these zones. The research compared four methods and found that VIKOR produced the strongest test performance, with an area under the receiver operating characteristic curve of 88%.

The district-wide analysis used a landslide inventory from 2018 and 17 conditioning factors representing topographical, hydrological, geological, geomorphological and anthropogenic conditions. The aim was to identify where mapped environmental and human-related conditions corresponded to greater landslide susceptibility. The study treated susceptibility as a modeled spatial pattern rather than as a forecast of when or where an individual landslide would occur.

Researchers generated four susceptibility maps using frequency ratio, logistic regression, TOPSIS and VIKOR. Frequency ratio and logistic regression were statistical methods, while TOPSIS and VIKOR belonged to the multi-criteria decision-making family. Applying these approaches to the same inventory and conditioning factors allowed the study to compare both the geographic patterns produced by different modeling choices and the models’ ability to discriminate landslide locations from non-landslide locations.

The 2018 inventory was split into training and testing subsets, with 70% of the data assigned to training and 30% to testing. Model performance was evaluated through receiver operating characteristic curves and area-under-the-curve analysis. In the held-out testing, VIKOR achieved an AUC of 88%, followed by TOPSIS at 84%, logistic regression at 82% and frequency ratio at 79%. These values measure discrimination within the study’s validation procedure; they do not show that any method can prevent landslides or determine the timing of a specific slope failure.

Cross-validation provided an additional assessment of the two statistical models across different subsets. Mean ROC–AUC values were 0.860 for logistic regression and 0.864 for frequency ratio, indicating good performance under that process. The close means show that the statistical approaches performed comparably in cross-validation, even though logistic regression had the higher individual testing AUC of 82%, compared with 79% for frequency ratio.

The high-susceptibility areas included Sultan Bathery, Vythiri and Mananthavady, which the researchers described as being mainly steep and associated with heavy rainfall patterns. On the maps, these areas represent combinations of the 17 factors that the models associated with greater landslide susceptibility. The analysis did not directly establish the number of landslides expected in each location, and its susceptibility classes should not be read as predictions of specific events.

The authors said the integrated framework can support disaster-risk reduction, sustainable land-use planning and infrastructure development in the ecologically sensitive Western Ghats. They also linked the evidence-based approach to Sustainable Development Goal 11 on sustainable cities and communities, Goal 13 on climate action and Goal 15 on life on land. These proposed uses are grounded in the models’ role as regional decision-support analyses rather than direct measurements of future landslide frequency.

The results remain bounded by the study’s geographic and data scope. The analysis covers Wayanad District and relies on a landslide inventory from 2018, so the mapped patterns are tied to the locations and conditions represented in that dataset. The supplied findings do not establish how the models would perform in other regions, during future rainfall seasons or under changed combinations of terrain, drainage, geology and human disturbance. Within these limits, all four approaches showed good measured discrimination, while VIKOR ranked highest in the held-out test.

Subject of Research: Comparison of statistical and multi-criteria models for mapping landslide susceptibility in Wayanad District, Kerala

Article Title: Landslide susceptibility mapping for disaster risk reduction using statistical and MCDM approaches in Wayanad District, Kerala, India

Article References: Mumtaz, A., Chakraborty, A., Paul, S., Das, S., Dhumal, R., & Singh, T. P. (2026). Landslide susceptibility mapping for disaster risk reduction using statistical and MCDM approaches in Wayanad District, Kerala, India. Discover Sustainability. https://doi.org/10.1007/s43621-026-04557-z

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04557-z

Keywords: landslide susceptibility, Wayanad District, Kerala, VIKOR, TOPSIS, logistic regression, frequency ratio, Western Ghats, disaster risk reduction, land-use planning, Landslide, susceptibility