Prediction of earth-fissure hazards: Unraveling the crucial roles of land use and groundwater fluctuations

Abstract

Understanding the occurrence of earth fissures in arid regions is crucial for informing land management practices and conservation strategies. In this study, we evaluate six innovative machine learning models for predicting earth-fissure hazards: the patient rule induction method, rotation forest, stochastic gradient boosting, sparse linear discriminant analysis, quadratic discriminant analysis with stepwise feature selection, and weighted subspace random forest (WSRF). By exploring the impact of various environmental factors on earth-fissure occurrence, we highlight the significant roles of land use and groundwater fluctuations in the development of earth fissures. Our findings demonstrate that afforested lands and declining groundwater levels are strongly associated with fissure occurrence. The WSRF model is the most effective in predicting diverse probabilities and providing a nuanced understanding of hazard levels. This study emphasizes the importance of considering environmental factors and selecting appropriate models for predicting earth-fissure hazards, ultimately promoting sustainable land management practices and mitigating potential risks associated with earth fissures.