Machine learning reveals the width of fault damage zones in northeast Sichuan Basin, China

Abstract

Accurate understanding and identification of faults architecture is crucial in seismic data interpretation and earthquake analysis, where fault slip surfaces may interact with damage rocks, forming damage zones with a width larger than hundred meters. We use machine learning (ML) to show 10 kinds of seismic attributes from a seismic survey could be applied in identification and quantification of fault damage zone in northeast Sichuan Basin, China. The results indicate: (1) Six seismic attributes provide highest contribution to the fault characterization, including root mean square amplitude attributes, azimuth angle attributes, reverse attributes, original attributes, chaotic body attributes and ant body attributes; (2) The application of SHAP (SHapley Additive exPlanations) algorithm improves the model's accuracy, as the loss value (Mean Square Error , MSE) of the test data is restored from 17.86% to 16.03%; (3) Width estimation from the kernel density estimation algorithm (KDE) show the fault damage zone ranges from 0.3 to 1.2 km. Our work provides new insights into the interpretation of fault architecture in the subsurface, and we argue the geometrical parameters of the fault damage zone is significant for understanding the evolution of fault and earthquake simulations.

Keywords:  Fault damage zone; Seismic interpretation; Machine learning (ML); Geometrical parameters

Figure1.The seismic attributes of the actual work area entered into the model and the model calculation results: (A) Original attributes of the work area. (B) Variance attribute of the work area. (C) Results calculated by the ML model

Figure2. Thermal diagram presents the structure of the fault damage zone: (A) A vertical line perpendicular to the fault orientation correction; (B) indicates the fault range with a thermal index greater than 1.572; (C) indicates a fault range with a thermal index greater than 2.065; (D) indicates a thermal index greater than 2.401 fault range. The width of the damage zone could be estimated by these figures.

 

Figure3. Descriptive diagram of fault damage zone width. Fault_1 represents the direction of fault width with thermal index greater than 1.572; Fault_2 represents the direction of fault width with thermal index greater than 2.065; Fault_3 represents the fault width trend map with thermal index greater than 2.401