Statistical Modeling and Machine Learning in Heliophysics

Statistical Modeling and Machine Learning in Heliophysics#

This site is a practical collection of tutorials, worked examples, and references for statistical modeling and machine learning in heliophysics. The notebooks use a focused scientific Python stack. Native PyTorch is the main neural-network path, with concise Keras 3 alternatives running on the same Torch backend. NumPy, pandas, Matplotlib, scikit-learn, XGBoost, and other specialized tools appear where they fit the problem. The Software Toolkit gives a brief introduction to what each package does.

Why AI and ML matter here#

Heliophysics and space weather present an unusual data challenge: decades of observations coexist with sparse sampling across enormous regions and relatively few examples of the most extreme events. The recent Eos article “Vast Space, Sparse Data” discusses how AI and machine learning can complement physical understanding under these conditions. Community efforts such as LMAG25 bring together heliophysicists, geospace and space-weather researchers, forecasters, and machine-learning specialists to develop useful, interpretable, and carefully validated approaches. This collection offers a practical entry point into that broader effort.

Run the notebooks#

Notebook pages provide an Open in Colab button for running an example in the browser. For a reproducible local environment, install Conda, open a terminal in the repository, and run:

conda env create -f environment.yml
conda activate helio-data-methods
UV_PROJECT_ENVIRONMENT="$CONDA_PREFIX" uv sync --frozen --group notebooks
jupyter lab

The first command creates the environment; the remaining commands activate it, install the locked notebook dependencies, and open JupyterLab or JupyterNotebook.

Explore the material#

General ML#

Introductions to neural networks, convolutional models, tree models, transfer learning, tuning, and generative models. The notebooks include suggestions that can be explored interactively in Colab.

Statistical Modeling#

A space for future examples on classical statistical models, uncertainty, and time-series methods.

Heliophysics#

Applied examples from Heliophysics including Dst prediction, SEP occurrence, coronal loop reconstruction, and plasmasheet data modeling.

Resources#

A curated set of books, interactive references, lectures, and exercises for going further than simple demonstrations.