nasa-ibm-ai4science/euv-spectra

EVE EUV Spectra Modeling Dataset

This dataset provides time-aligned Extreme Ultraviolet (EUV) irradiance spectra from NASA’s SDO/EVE (Extreme Ultraviolet Variability Experiment) instrument, to be paired with temporally matched image-based input features from HelioFM pipeline. It is designed for use in image to spectra regression modeling.

Dataset Details

Dataset Description

The dataset consists of three splits (70-15-15): train, val, and test, each containing:

• A timestamp
• A 1343-dimensional spectrum corresponding to EUV wavelengths ranging from approximately 6.5 to 33.3 nm, observed at a 1-minute cadence
• EUV measurements from both flare and quiet sun conditions

Curated by: Dr. Shah Mohammad Bahauddin, Laboratory for Atmospheric and Space Physics, University of Colorado Boulder.

Dataset Sources

The data is sourced from high-resolution EUV spectra spanning 6.5 to 106.5 nm provided by the SDO/EVE instrument. For the downstream task of image-to-spectra modeling, we specifically utilize data from the MEGS-A (Multiple EUV Grating Spectrograph A) channel of EVE, which was operational from 2010 to 2014 and provided spectral coverage from 6.5 to 33 nm. The native EVE data are available from 2010 to 2014 at 10-second cadence. For this study, we extract one spectrum per minute by integration and retain only those flagged as high quality based on header metadata. This process yields a curated dataset of approximately two million spectra from five years of EVE operations. For downstream training and evaluation, we align with HelioFM’s 12-minute cadence, resulting in a final dataset of approximately 190,000 quality-filtered spectra for model development.

Uses

Direct Use

• Regression modeling of solar EUV spectra from coronal image inputs and differential emission measures.
• Timeseries forecasting of coronal EUV spectra.

Dataset Structure

Each split contains:

• timestamp: A UTC datetime string
• intensity: 1343 columns of EUV spectral intensity values labeled by their physical wavelengths (rounded to 2 decimal places in nanometers)

The data is balancedly sourced from both quiet-Sun and flare conditions and split into train, val, and test using consistent time bins.

Dataset Creation

Curation Rationale

This dataset enables machine learning models to learn from and predict EUV spectral behavior driven by solar dynamics. It addresses the need for high-resolution, calibrated spectral data paired with physics-based contextual input.

Source Data

The primary source of the dataset is the SDO-EVE Level 2b data product which contains 60-second integrations of extracted solar EUV irradiance in selected lines spanning a wide range of solar temperature measured by the MEGS-A and MEGS-B detectors. This dataset particularly contains MEGS-A data which measures from 6 to 33.33 nm. The Level 2b irradiance data are adjusted to 1-AU.

• Instrument Data: https://lasp.colorado.edu/eve/data_access/eve_data/products/level2b/

Data Collection and Processing

Each MEGS-A irradiance observation is represented as a one-dimensional vector with 1343 bins. These measurements are strictly positive and span a dynamic range of 1e-9 to 1e-1 W/m^2/nm, depending on the feature and flare class. While MEGS-A originally offered 10-second cadence, we compute 1-minute averaged EVE spectra to reduce noise and facilitate temporal alignment with AIA image data. For the downstream application, we further temporally downsample our data to match the machine learning–ready SDO/AIA image cubes, which were sampled at 12-minute cadence. This choice follows the pretraining configuration of HelioFM, which was optimized for compatibility with the 12-minute cadence of SDO/HMI magnetograms, allowing future multi-modal tasks involving both intensity and magnetic field evolution. By adopting the same cadence for training and evaluation, we ensure internal consistency, reduce complexity in data handling, and maintain computational feasibility. The resulting dataset contains hundreds of thousands of temporally aligned AIA image cubes and EUV spectra, covering Solar Cycle 24 and parts of Solar Cycle 25, and includes both quiet-Sun and active-region conditions.

We apply metadata-based screening using the quality flags provided in the EVE Level 2 data products. The SC_FLAGS byte identifies potential obstructions or pointing issues during observation. Only spectra flagged as 0 (clear, unobstructed) are retained, while any entries marked with atmospheric (Earth, lunar, planetary) penumbra or umbra conditions, or spacecraft off-pointing (bit 32 set), are excluded due to their impact on irradiance accuracy. The FLAGS byte indicates the instrument status and data processing integrity; entries with MEGS-A missing (bit 0 set), or those affected by too many integrations due to spacecraft timing issues (bit 4 set), are removed to prevent inclusion of anomalous or incomplete data. Additionally, we inspect the BIN_FLAGS byte for each spectral bin and discard spectra where a significant portion of the wavelength range contains nonzero flags, as this typically reflects calibration artifacts or spectral corruption. This multi-level flag screening is essential for eliminating spurious outliers and ensuring that the training set reflects genuine physical variability in the solar EUV spectrum.

Who are the source data producers?

For access and data product issues of source EVE data, please contact Don.Woodraska at lasp.colorado.edu. For science issues of EVE data, please contact Frank.Eparvier at lasp.colorado.edu.

Bias, Risks, and Limitations

• The dataset reflects only the time period available in the EVE and HelioFM input sources
• May not generalize to other solar and stellar conditions (e.g., polar events, rare flaring periods) that are outliers to solar cycle 24 and 25.
• Assumes accurate time alignment within a few minutes tolerance between AIA and EVE.
• Does not contain the raw AIA or HMI images, only EUV spectral targets.

Recommendations

This dataset is suitable for regression, forecasting, or inverse modeling tasks within plasma physics, solar astrophysics and heliophysics. Users should verify that model outputs are physically interpretable and validate with domain-expert knowledge.

Citation

Woods, T. N., Eparvier, F. G., Hock, R., Jones, A. R., Woodraska, D., Judge, D., ... & Viereck, R. (2012). Extreme Ultraviolet Variability Experiment (EVE) on the Solar Dynamics Observatory (SDO): Overview of science objectives, instrument design, data products, and model developments. The solar dynamics observatory, 115-143, doi: http://doi.org/10.1007/s11207-009-9487-6

Woodraska, D., & Eparvier, F. G. (2024). Solar Dynamics Observatory (SDO) Extreme Ultraviolet Variability Experiment (EVE): Version 8 science data product Release Notes, LASP / University of Colorado Boulder Technical Document. https://lasp.colorado.edu/eve/data_access/eve_data/products/level2b/EVE_L2B_V8_README.pdf

Glossary

• EUV: Extreme Ultraviolet
• EVE: Extreme Ultraviolet Variability Experiment
• AIA: Atmospheric Imaging Assembly (onboard SDO)
• SDO: Solar Dynamics Observatory

Dataset Card Authors

Dr. Shah Mohammad Bahauddin, Laboratory for Atmospheric and Space Physics, University of Colorado Boulder.

Dataset Card Contact

Email: [email protected]