Earth fissure hazard prediction using machine learning models

Area of Science:

• Earth Science
• Hydrology
• Geology

Background:

• Earth fissures are surface cracks primarily in arid/semi-arid regions.
• Excessive groundwater withdrawal is a major cause of land subsidence and earth fissuring.
• Earth fissuring poses significant economic, social, and environmental risks, escalating into national disasters.

Purpose of the Study:

• To propose novel machine learning models for predicting earth fissure hazards.
• To identify vulnerable groundwater areas for informed water management and conservation.
• To enhance understanding of the complex factors contributing to earth fissure formation.

Main Methods:

• Simulated Annealing Feature Selection (SAFS) for identifying key predictive features.
• Application of Generalized Linear Model (GLM), Multivariate Adaptive Regression Splines (MARS), Classification and Regression Tree (CART), Random Forest (RF), and Support Vector Machine (SVM) for hazard prediction.
• Utilizing historical data on groundwater levels, withdrawal, well density, precipitation, and geological formations.

Main Results:

• All developed models demonstrated high accuracy (over 86%) and precision (over 81%) in predicting earth fissure hazards.
• The Random Forest (RF) model exhibited the highest performance among the tested machine learning algorithms.
• Generalized Linear Model (GLM) showed the lowest predictive performance.

Conclusions:

• Machine learning models, particularly Random Forest, offer a robust approach to predicting earth fissure hazards.
• Low elevations, high groundwater withdrawal, declining groundwater levels, high well and road density, low precipitation, and Quaternary sediments are key indicators of hazardous areas.
• Accurate hazard modeling is crucial for effective groundwater management and sustainable resource conservation planning.