.png){kind=icon}
.png){kind=icon}
Return
An uncertainty-aware framework with enhanced tree-based methods for CPT-based subsurface soil stratification toward geological disaster risk assessment
.png){kind=icon}
OA
AI
C
J
Y
DOI：10.1186/s40677-026-00396-2.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Reliable subsurface soil stratification is essential for geotechnical site characterization and geological disaster risk assessment, as the spatial distribution of soil layers strongly influences ground deformation, slope instability, and other geoenvironmental hazards. Cone penetration tests (CPTs) provide high-resolution information along the depth direction and have been increasingly used for subsurface stratification. However, conventional CPT-based stratification methods usually interpolate continuous cone penetration test parameters and then convert the interpolated results into soil behavior types based on the Robertson chart. When only sparse test data are available, this interpolation scheme may produce fuzzy soil-layer boundaries and uncertain classification results due to error propagation. This study proposes an uncertainty-aware framework with enhanced tree-based methods for CPT-based subsurface soil stratification. The proposed framework directly predicts the spatial distribution of categorical soil behavior types using a classification scheme rather than first interpolating continuous soil property fields. Tree-based ensemble methods are adopted to partition the subsurface space into different soil-type regions. Geotechnical distance fields are designed as informative inputs to enhance the spatial representation of sparse test data. The framework also integrates automatic soil-layer boundary detection using a gradient-based method and uncertainty estimation using the Gini impurity index with alpha-channel visualization. Synthetic case studies show that the classification scheme produces clearer soil-layer boundaries and higher prediction accuracy than the interpolation scheme. The geotechnical distance fields input also outperforms the conventional XY-coordinate input, and the random forest method with geotechnical distance fields achieves the best overall performance among the tree-based ensemble methods. In addition, the gradient-based boundary detection method successfully extracts soil-layer boundaries, while uncertainty estimation using the Gini impurity index highlights regions with relatively uncertain stratification results. A real case in New Zealand further shows that the predicted stratification results are generally consistent with borehole data while providing finer subsurface details. The proposed framework provides a data-driven and uncertainty-informed tool for constructing reliable subsurface stratification models from sparse cone penetration test data. It can support digital geotechnical site investigation and provide useful subsurface information for risk-informed geological disaster assessment.
Keywords:
Soil stratification
Geotechnical distance fields
Tree-based ensemble method
Uncertainty estimation
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
4
Papers:
264
Citations:
1.1K
Organization
Cited Papers