2024-HL-GeoCL

FDL-X Heliolab 2024: Geoeffectiveness Continuous Learning

GEO-CLoak (GEOeffectiveness Continual Learning Or Active Knowledge) is an end-to-end ML pipeline for forecasting geomagnetic perturbations. It combines two models to produce predictions at different lead times:

• SHEATH: Predicts solar wind parameters (speed, density, temperature, magnetic field) from SDO solar imagery features. Uses a 3-layer MLP. Provides multi-day lead time, lower-fidelity forecasts.
• DAGGER-CL: Predicts ground-level geomagnetic field perturbations (dBe, dBn, dBz) at ~175 SuperMAG magnetometer stations. Uses a GRU encoder with continual learning (Elastic Weight Consolidation) to adapt to new data without forgetting. Provides ~30-minute lead time, higher-fidelity forecasts.

The pipeline flows: Sun (SDO imagery) → SHEATH → solar wind forecast → DAGGER-CL (with real-time L1 data) → station predictions → global maps → web app

Project Structure

geocloak/              Main Python package
  sheath2024/          SHEATH model, dataloader, training
  dagger-cl/           DAGGER-CL model, dataloaders, inference, continual learning
  gp/                  Gaussian process & spherical harmonic interpolation
  preprocess/          SDO image preprocessing (coronal hole/active region segmentation)
  datautilus/           Real-time data download from NOAA SWPC

public/                Standalone scripts for public use
  inference_sheath.py  Run SHEATH inference from the command line
  example_data/        Sample input files

feature_vector_extraction/   Feature engineering pipeline (InfluxDB → 26 SHEATH features)
updating_nrt_data/           Near-real-time data collection (ACE, DSCOVR, geomagnetic indices)
app_dev/                     Streamlit web application
scripts/                     Training scripts, utilities, and Jupyter notebooks
models/                      Trained model checkpoints and scaler files

Setup

This project requires Python 3.11+ (< 3.13) and uses uv for dependency management.

Install uv

If you don't have uv installed:

# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Or via Homebrew
brew install uv

Create environment and install dependencies

uv venv --python 3.11
uv sync

This creates a .venv/ virtual environment and installs all dependencies from pyproject.toml.

Activate the environment

source .venv/bin/activate

Or run commands directly without activating:

uv run python your_script.py

Install dev tools (optional)

To also install code formatting and linting tools (black, isort, flake8):

uv sync --extra dev

Contributing

The project uses black for code formatting, isort for import sorting, and flake8 for linting. Run these before submitting a pull request:

black .
isort --profile black .
flake8 --extend-ignore E203,E501

If you deem any lines of code correct and want to overwrite the PEP8 checks, add a # noqa: CODE at the end of the line (e.g. # noqa: W605).

Please fix any errors that are raised by these commands before submitting a pull request.

Further Development

1. Train SHEATH with NRT SDOML (Embeddings, AR&CH Features), using OMNI as targets.

Acknowledgements

This work is the research product of FDL-X Heliolab, a public/private partnership between NASA, Trillium Technologies Inc., and commercial AI partners Google Cloud, NVIDIA, and Pasteur Labs & ISI.

FDL-X Heliolab and its outputs have been designed, managed and delivered by Trillium Technologies Inc (trillium.tech). Trillium is a research and development company with a focus on intelligent systems and collaborative communities for planetary stewardship, space exploration, and human health. We express our gratitude to Google Cloud and NVIDIA for providing extensive computational resources.

The material is based upon work supported by NASA under award No(s) 80NSSC24M0122. Any opinions, findings, conclusions, or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Aeronautics and Space Administration.