terrangraph 0.4.0

A graph-based framework for geotechnical site representation and subsurface intelligence

TerranGraph: A Graph-Based Framework for Geotechnical Site Representation

TerranGraph is a Python framework for geotechnical site representation, learning, and visualization. It integrates multi-source data processing, spatial graph construction, graph neural networks, and 3D visualization into a single workflow for subsurface modeling and prediction.

---

Overview

TerranGraph follows an end-to-end workflow:

Geotechnical Data → Spatial Graph → Representation Learning → Prediction → Visualization

Designed for subsurface modeling and site characterization, TerranGraph bridges geotechnical data and graph-based intelligence.

---

Features

• Multi-source data integration for Excel and CSV geotechnical datasets
• Spatial graph construction from borehole logs and spatial coordinates
• Graph neural networks for site representation learning and prediction
• 3D visualization with PyVista for subsurface modeling and cross-section analysis
• Interactive GUI based on PyQt5 for workflow exploration
• Scalable training with PyTorch and optional GPU acceleration
• Model export to standard 3D formats such as PLY, OBJ, and STL

---

Installation

pip install terrangraph

Quick Start

from terrangraph import DataHandler, GeoModel, GeoVisualizer

# Load geotechnical data
handler = DataHandler()
unique_types, num_classes = handler.load_data("your_dataset.xlsx")

# Preprocess and align coordinates
rect = handler.compute_mbr()
bounds = handler.align_coordinates(x_res=10, y_res=2, z_res=1)
params_dict = handler.create_param_dict()

# Construct spatial graph
coords, labels, edges = handler.generate_grid(bounds, r=2)

# Train graph neural network
model = GeoModel()
model.prepare_data(coords, labels, edges)
model.build_model(
    num_classes=num_classes,
    hidden_size=48,
    gcn_layers=3,
    mlp_layers=3,
    dropout=0.2
)
history = model.train(epochs=100, lr=3e-3)

# Prediction and visualization
predictions, probabilities = model.predict()
visualizer = GeoVisualizer()
mesh = visualizer.create_model(coords, predictions)
visualizer.show_model(mesh)

Input Data Format

The input dataset should include:

• X, Y, Z: spatial coordinates
• Soil/Rock Type: soil or rock type classification
• Borehole ID: borehole identifier (optional)
• Borehole Type: borehole category (optional)

Additional geotechnical parameters (e.g., strength, density, CPT measurements, etc.) can be incorporated for extended modeling tasks.

Use Cases

TerranGraph can be used for:

• 3D subsurface modeling
• Borehole-based site characterization
• Graph-based representation learning of geotechnical data
• Similarity-based site comparison and clustering
• Data-driven prediction of soil and rock properties

Project Structure

terrangraph/
├── src/
│   └── terrangraph/
│       ├── __init__.py              # Package exports
│       ├── cli.py                   # Command-line interface
│       └── core/
│           ├── __init__.py
│           ├── data_handler.py      # Data loading and preprocessing
│           ├── geo_model.py         # Graph neural network models
│           └── visualizer.py        # 3D visualization (PyVista)
│
├── assets/                          # Logo and banner images
├── docs/                            # Documentation files
├── CHANGELOG.md                     # Version history
├── LICENSE                          # License
├── pyproject.toml                   # Package configuration
└── README.md                        # Project documentation

Examples

See the examples/ directory:

• basic_usage.py: basic workflow
• advanced_visualization.py: advanced visualization
• batch_processing.py: batch processing
• custom_model.py: custom GNN architectures

Documentation

• Installation Guide
• API Reference
• User Tutorial
• Developer Guide

System Requirements

• Python 3.8+
• 8GB RAM minimum (16GB recommended)
• CUDA-compatible GPU (required for ml)
• OpenGL support for 3D visualization (required for viz)

Roadmap

Current Capabilities

• Graph-based geotechnical site representation
• Spatial graph construction from borehole data
• GNN-based prediction framework
• 3D subsurface visualization (PyVista)

Next Steps

• Integration of additional geotechnical parameters (e.g., CPT, strength, density)
• Integration of real-time monitoring data (e.g., tunneling operation data)
• Contrastive learning for site similarity analysis
• Support for large-scale geotechnical datasets
• Probabilistic modeling and uncertainty quantification

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Citation

If you use TerranGraph in your research, please cite:

@software{wang2026terrangraph,
  author = {Wang, Lai},
  title = {TerranGraph: A Graph-Based Framework for Geotechnical Site Representation},
  year = {2026},
  url = {https://github.com/wanglai25/terrangraph}
}

Support

• Email: [e-mail]
• Issues: GitHub Issues
• Discussions: GitHub Discussions