license: mit task_categories:

• image-to-image
• time-series-forecasting tags:
• solar
• heliophysics
• astronomy
• AIA
• SDO
• EUV
• SXR
• GOES
• flare
• space-weather pretty_name: "FOXES: Framework for Operational X-ray Emission Synthesis" size_categories:
• 100K<n<1M language:
• en datasets:
• griffingoodwin04/FOXES-Data

FOXES: A Framework for Operational X-ray Emission Synthesis

FOXES predicts full-disk solar soft X-ray (SXR) flux and spatial patch contributions from seven-channel SDO/AIA extreme-ultraviolet images. This repository contains the Python API, data tools, training code, and evaluation workflows developed through Frontier Development Lab's Heliolab 2025.

Model demo · Dataset · Notebook walkthrough

Install

From a local checkout:

python -m pip install -e .

The package is named helio-foxes; import it as helio_foxes. For a fresh environment, editable tests, or notebook kernel setup, follow the development setup.

Predict one observation

The examples below use the same timestamps, checkpoints, and plotting settings as the saved outputs in the heliofoxes notebook. Use a registered JSOC email and download the checkpoints into the paths shown. Load the model once and reuse it for each observation.

from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
import helio_foxes

weights = Path("forecasting/trained_weights_and_normalization")
model = helio_foxes.load_model(
    weights / "updated_architecture_checkpoints" / "ComplexModel.ckpt",
    device="auto",
)
processor = helio_foxes.ObservationProcessor(
    email="you@example.com", harp_bbox_expansion=0.2,
)
observation = processor.fetch_observation("2023-07-11T01:30:00Z")
result = model.predict(observation)

print(f"Predicted XRS-B: {result.flux:.3e} W/m²")
if observation.sxr is not None:
    print(f"Measured XRS-B: {observation.sxr['xrsb_flux']:.3e} W/m²")

fig, ax = plt.subplots(figsize=(10, 7))
helio_foxes.plot_flux_map(
    result, ax=ax, show_harp_boxes=True, min_harp_flux=1e-7,
)
plt.show()

Saved notebook output for 2023-07-11 at 01:30 UTC:

Predicted XRS-B: 2.695e-06 W/m²
Measured XRS-B: 2.941e-06 W/m²

Inspect the AIA inputs

Plot the seven processed channels from the same observation:

fig, axes = plt.subplots(2, 4, figsize=(14, 7), constrained_layout=True)
for ax, wavelength, image in zip(
    axes.flat, observation.metadata.wavelengths, observation.data,
):
    low, high = np.percentile(image, [1, 99])
    ax.imshow(image, origin="lower", cmap="gray", vmin=low, vmax=high)
    ax.set_title(f"AIA {wavelength} Å")
    ax.axis("off")
for ax in axes.flat[len(observation.metadata.wavelengths):]:
    ax.axis("off")
plt.show()

Compare checkpoints

Reuse the same observation to compare the three notebook checkpoints:

checkpoints = {
    "Complex Gaussian": weights / "updated_architecture_checkpoints" / "ComplexModel.ckpt",
    "Lighter Gaussian": weights / "updated_architecture_checkpoints" / "SlightlyLighterModel.ckpt",
    "Deterministic": weights / "inverted-attention-mask.ckpt",
}
for name, checkpoint in checkpoints.items():
    predictor = helio_foxes.load_model(checkpoint, device="auto")
    prediction = predictor.predict(observation)
    print(f"{name}: {prediction.flux:.3e} W/m²")
    del predictor

The measured GOES XRS-B flux at the same time is 2.941 × 10⁻⁶ W/m², using GOES-16 and GOES-18. These are individual notebook examples.

Fetching downloads and preprocesses AIA data and obtains supplementary HARP regions and GOES measurements. Raw downloads are temporary. Missing HARP or GOES data do not discard a usable AIA observation.

Already have data? See loading your own processed arrays or local FITS files. Inputs must match the checkpoint's training preprocessing and wavelength order: [94, 131, 171, 193, 211, 304, 335] Å.

Predict a time range

observations = processor.iter_observations(
    "2024-03-23 00:45", "2024-03-23 02:45",
    cadence="1min", progress=True, save_dir="saved_observations",
    on_error="skip",
)
series = model.predict_series(
    observations, snapshot_time="2024-03-23 01:45", snapshot_wavelength=193,
)
ax = series.plot(
    show_snapshot=True, show_uncertainty=True, min_peak_flux=1e-7, legend="auto",
)
plt.show()

totals = series.to_dataframe()
harp_curves = series.harp_light_curves()  # {HARP number: DataFrame}

Using the Complex Gaussian model loaded above, the saved notebook run produced 121 predictions. The flux-map and AIA panels show the retained snapshot at 2024-03-23 01:45 UTC.

*Shading and regional ± values show the model's variance attribution; they are not calibrated yet.

The iterator downloads and preprocesses observations one at a time. Existing saved files are reused, so rerunning a range skips completed downloads and preprocessing. AIA record selection uses a bulk JSOC query.

snapshot_time keeps the nearest observation's AIA channel for the flux-map and image panels. Omit it to retain no image; use series.plot() for curves only. Gaussian models show uncertainty when available; deterministic models do not.

Reload saved observations without downloading or preprocessing again:

saved = helio_foxes.load_observations("saved_observations")
series = model.predict_series(saved)
series.plot()

See the API guide for timestamp formats, cadence, source toggles, snapshot selection, uncertainty, plot controls, and saved data.

Models and documentation

The repository includes the original deterministic models and newer local + global Gaussian checkpoints. The newer checkpoints have not yet been validated for prediction quality, spatial maps, or uncertainty calibration. Patch and HARP uncertainties attribute global model variance; they are not calibrated local error bars.

• Python API: observations, prediction, light curves, and plotting.
• Development and CLI workflows: setup, checkpoint paths, data preparation, inference, evaluation, training, and repository structure.
• Model background: original abstract and architecture discussion.

After editing installed source, restart an existing notebook kernel and rerun the setup cell so it loads the current API.

Citation

If you use this code or data in your work, please cite:

@software{FOXES,
    title           = {{FOXES: A Framework For Operational X-ray Emission Synthesis}},
    institution     = {Frontier Development Lab (FDL)},
    repository-code = {https://github.com/griffin-goodwin/FOXES},
    version         = {v1.0},
    year            = {2026}
}

Acknowledgement

This work is a research product of Heliolab (heliolab.ai), an initiative of the Frontier Development Lab (FDL.ai). FDL is a public–private partnership between NASA, Trillium Technologies (trillium.tech), and commercial AI partners including Google Cloud and NVIDIA. Heliolab was designed, delivered, and managed by Trillium Technologies Inc., a research and development company focused on intelligent systems and collaborative communities for Heliophysics, planetary stewardship and space exploration. We gratefully acknowledge Google Cloud for extensive computational resources and NVIDIA Corporation. This material is based upon work supported by NASA under award number No. 80GSFC23CA040. Any opinions, findings, conclusions or recommendations expressed are those of the authors and do not necessarily reflect the views of the National Aeronautics and Space Administration.

Large language models were used as brainstorming tools to discuss possible training strategies and methodological considerations. The authors retained full responsibility for all research decisions, interpretations, and conclusions presented in this work.