Assessment of statistical and machine learning methods for landslide susceptibility of Pithoragarh, Uttarakhand (India)

Abstract

Abstract Landslides are the most prevalent natural hazard in hilly regions of India. These can have a significant impact on the economy of a nation. This study examines the landslide susceptibility of the Pithoragarh, Uttarakhand, India, utilising various susceptibility methods, such as Frequency Ratio (FR), Information Value (IV), Weight of Evidence (WOE), and Certainty Factor (CF). The LSZ modelling was performed using fourteen landslide causative factors. Based on past landslide data, landslide locations were identified, which were further divided into a 70/30 ratio, with 70 representing training and 30 representing validation. Validation of the findings of the predicted maps of landslide susceptibility using Area under Curve (AUC) indicates that the predicted map using the FR approach has the highest prediction rate compared to other methods used for landslide susceptibility prediction. Also to check the feasibility of the machine learning method, we have considered logistic regression (LR) analysis using five out of fourteen factors. AUC revealed that LR has higher accuracy than Analytical Hierarchy Process (AHP) and Shannon Entropy (SE). Also, validation of all the models was done using Landslide Density Index (LDI) which shows the validity of all models. Thus, the results of all models can be used to predict landslide susceptibility in Pithoragarh.