Earth fissure susceptibility mapping: Application of random subspace‐based novel ensemble approaches

M. Santosh,Alireza Arabameri,Aman Arora

DOI: https://doi.org/10.1002/gj.4932

2024-03-07

Geological Journal

Abstract:The aim of this research is to evaluate earth fissure susceptibility in a typical area of the Semnan province, a region in North‐Central Iran, by utilizing ensemble‐based machine learning approaches. The development of earth fissures, which are linear fractures with openings or offsets on the land surface, can severely affect landforms, especially in urban areas, in the form of earthquakes causing major concern on human lives as well as damage to infrastructures. Thus, an early warning map for lands susceptible to earth fissures can better equip planners for formulating mitigation strategies. In this study, we focus on the Damghan Plain in Iran for preparation of earth fissure susceptible maps using several topographical, hydrological, geological and environmental conditioning factors. In order to train these conditioning factors and preparation of earth fissure susceptibility maps, 124‐earth fissure field‐based samples, for training and validation purposes, were used by random subspace (RS) model based on four other machine learning ensemble methods such as RS‐Naïve‐Bayes Tree (NBTree), RS‐alternating decision tree (ADTree), RS‐Fisher's Linear Discriminant Function (FLDA) and RS‐Logistic model tree (LMT). From the validation technique, the receiver operating characteristic (ROC) curve performance test demonstrates that the RS‐NBTree model was the best suited with area under curve (AUC) = 0.974 followed by RS‐ADTree (AUC = 0.966), RS‐LMT (AUC = 0.954), RS‐FLDA (AUC = 0.948) and RS (AUC = 0.923). The results from our study can be useful for environmental management and risk reduction.

geosciences, multidisciplinary