nasa-ibm-ai4science/surya-bench-flare-forecasting

Full-disk Solar Flare Forecasting Dataset

Dataset Summary

This dataset provides labels for solar flare forecasting derived from NOAA GOES flare events from May 2010 to December 2024. Labels are constructed using a 24h rolling prediction window sampled at an hourly cadence. Each window is annotated with both GOES class (based on peak X-ray flux) and cumulative flare index.

Two derived binary labels are included for forecasting tasks:

• label_max: 1 if the maximum flare intensity in the window is ≥ M1.0.
• label_cum: 1 if the cumulative flare intensity in the window is ≥ 10.

For completeness, we also include (1) GOES class, which is determined from the peak X-ray flux of the most intense flare in the prediction window; and (2) cumulative index determined from all ≥C-class flares in the prediction window.

Supported Tasks and Applications

• binary-classification: Predict whether a time window will contain significant flaring activity.
  ordinal-classification: Predict flare-class of a given instance.
• regression: Predict cumulative flare index of a given instance.

Dataset Structure

Data Files

• train.csv: Instances from Feb 15 to Dec 31 in each year between 2010–2019
• validation.csv: Instances from Jan 15–28 of each year between 2010–2019
• test.csv: All instances from each year between 2020–2024
• leaky_validation.csv: Instances from Jan 1–14 and Jan 29–Feb 11 of each year between 2010–2019

Features

Each record includes four label fields:

• max_goes_class: Maximum GOES flare class (e.g., C5.2, M1.0, X3.2) in the prediction window, or FQ if no flares are present.
• cumulative_index: Weighted sum of flare subclasses ≥C-class in the prediction window.
  • C-class contributes weight ×1, M-class ×10, X-class ×100.
  • For example, an M2.0 flare adds 20, while an X3.5 flare adds 350.
• label_max: Binary label, 1 if goes_class ≥ M1.0, else 0.
• label_cum: Binary label, 1 if cumulative_index ≥ 10, else 0.

Example entry (in JSON format):

{
  "timestep": "2011-02-14 03:00:00",
  "goes_class": "X2.2",
  "cumulative_index": 297.1,
  "label_max": 1,
  "label_cum": 1
}

Dataset Details

Field	Description
Temporal Coverage	May 13, 2010 – Dec 31, 2024
Data Format	CSV (.csv), string-based schema
Data Shape	(1, 4) per instance
Data Size	Total 128,352 instances
Cadence	1 hour
Total File Size	~3.7MB