geoprior-v3 3.2.1

Physics-guided AI for geohazards

Physics-guided AI for Geohazards

GeoPrior-v3 is an open, reproducible research codebase for physics-guided machine learning in geohazard forecasting and risk assessment. Today, it focuses on land subsidence via GeoPriorSubsNet; next, it expands toward landslides and broader geohazard regimes.

---

✨ Why GeoPrior?

GeoPrior is built to answer practical scientific questions:

• Physics-consistent forecasting — Can predictions respect governing constraints (e.g., consolidation / groundwater dynamics)?
• Generalization across cities/regions — Can one workflow adapt to different regimes with a consistent packaging strategy?
• Uncertainty that supports decisions — How reliable are forecast intervals, and where do risks concentrate?
• Reproducibility — Can reviewers and researchers re-run results end-to-end (GitHub + Code Ocean)?

---

🔭 Scope & Roadmap

Current focus

• Land subsidence forecasting with GeoPriorSubsNet v3.2
• Physics residual monitoring + learned closures (e.g., effective parameters)

Planned

• Landslide forecasting modules (susceptibility + triggers + dynamics)
• Multi-hazard risk analytics (composable workflows and diagnostics)

---

📥 Installation

From PyPI (recommended)

pip install geoprior-v3

This installs GeoPrior and the scientific Python stack it depends on. Python 3.10+ is supported.

Development install (editable)

git clone https://github.com/earthai-tech/geoprior-v3.git
cd geoprior-v3
pip install -e .[dev]

The [dev] extra installs testing and tooling dependencies

(e.g., pytest, coverage, black, ruff).

---

⚡ Quick Start

1) Import + logging

import geoprior as gp
from geoprior.logging import get_logger, initialize_logging

initialize_logging(verbose=False)
log = get_logger(__name__)
log.info("GeoPrior logging is ready.")

2) Initialize a project config

GeoPrior now exposes a family-based CLI. The root entry point is geoprior, and dedicated shortcuts are also available for the run, build, and plot families.

geoprior-init --help
geoprior --help
geoprior-run --help
geoprior-build --help

To create nat.com/config.py interactively:

geoprior-init

3) Use the CLI families

The canonical root form is:

geoprior run <command> [args]
geoprior build <command> [args]
geoprior plot <command> [args]

Family-specific entry points avoid repeating the family name:

geoprior-run <command> [args]
geoprior-build <command> [args]
geoprior-plot <command> [args]

Examples:

geoprior run stage1-preprocess
geoprior-run stage2-train --help
geoprior build full-inputs-npz --help
geoprior-build physics-payload-npz --help
geoprior-run sm3-suite --preset tau50 --help

Compact CLI command table

Family	Command	Purpose
run	stage1-preprocess	Stage-1 preprocessing and artifact export.
run	stage2-train	Train the main GeoPrior model.
run	stage3-tune	Hyperparameter tuning workflow.
run	stage4-infer	Inference, evaluation, and export from trained runs.
run	stage5-transfer	Cross-city transfer evaluation.
run	sensitivity	Run physics or lambda sensitivity workflows.
run	sm3-identifiability	Run one SM3 synthetic identifiability experiment.
run	sm3-suite	Launch preset SM3 regime suites such as tau50 or both50.
run	offset-diagnostics	Run SM3 log-offset diagnostics.
build	full-inputs-npz	Merge Stage-1 split input NPZ files into one union NPZ.
build	physics-payload-npz	Export a physics payload NPZ from a trained model and inputs.
build	external-validation-metrics	Compute borehole or external validation metrics.
build	external-validation-fullcity	Build the full-city validation workflow end to end.
build	assign-boreholes	Assign validation boreholes to the nearest city cloud.
build	add-zsurf-from-coords	Merge elevation lookup tables and add z_surf to tabular data.
build	sm3-collect-summaries	Combine per-regime SM3 summary CSV files into one table.

Typical usage:

geoprior run stage1-preprocess
geoprior run stage2-train
geoprior build full-inputs-npz
geoprior-build external-validation-metrics --help
geoprior-run sm3-suite --preset tau50

4) Reproduce paper-style workflows

A typical end-to-end workflow is organized as:

• Stage-1 preprocessing
• Stage-2 training
• Stage-3 tuning
• Stage-4 inference
• Stage-5 transfer evaluation
• supplementary build and diagnostic commands

Example:

geoprior run stage1-preprocess
geoprior run stage2-train
geoprior run stage4-infer --help
geoprior build external-validation-metrics --help
geoprior build sm3-collect-summaries --help

---

🧪 Reproducibility

GeoPrior-v3 is designed for end-to-end reproducible science.

• Stable configs via nat.com/config.py and CLI overrides
• Deterministic pipeline stages exposed through geoprior run ...
• Reusable artifact builders exposed through geoprior build ...
• Capsule-friendly execution via codeocean/

If you publish results, please pin:

• the GeoPrior version (tag / release)
• config files used
• CLI commands invoked
• environment definition (e.g., environment.yml)

---

📚 Documentation

For detailed usage, API reference, workflow guides, and example galleries, please visit the official documentation:

geoprior-v3.readthedocs.io

---

💻 GeoPrior 3.0 Forecaster App

The GeoPrior research core is openly available to support reproducible scientific research.

In addition, GeoPrior includes the GeoPrior 3.0 Forecaster App, which offers a complete GUI-based application workflow built on top of the research framework. This application version is maintained separately from the public repository, which focuses on the core scientific and reproducible modules.

For access to the GeoPrior 3.0 Forecaster App, please contact the author directly.

• App tutorial: https://youtu.be/JtOpX5lv4iw
• Example simulation run: https://youtu.be/nCouLQQFpQg

---

🙌 Contributing

Contributions are welcome.

1. Check the Issues tracker for bugs or ideas.
2. Fork the repository.
3. Create a new branch for your feature/fix.
4. Add tests when applicable.
5. Open a Pull Request.

Please see CONTRIBUTING.md for details.

---

📜 License

GeoPrior-v3 is distributed under the Apache License 2.0. See LICENSE and NOTICE for details.

---

📞 Contact & Support

• Bug reports & feature requests: GitHub Issues
• Author: Laurent Kouadio — https://lkouadio.com/
• Email: [e-mail]