metadata

license: mit
size_categories:
  - n>1T

Solar Flare Forecasting Labels

Dataset Summary

This dataset provides binary labels for solar flare forecasting using a rolling window approach. Labels are generated based on both the maximum flare intensity and the cumulative flare intensity observed within each window. The data spans from 2010 to 2024 and is derived from NOAA GOES flare event listings.

Supported Tasks and Applications

binary-classification: Predict whether a given time window will contain a flare
- Maximum intensity: greater than or equals to M1.0.
- Cumulative intensity: greather than or equals to 10.
regression: Predict Maximum intensity or cumulative intensity within a given time window

Dataset Structure

Data Files

train.csv: Input data from January to June (6 months)
val.csv: Input data from July to September (3 months)
test.csv: Input data from October to December (3 months)

To create other train, validation, and test set, use these files

flare_cls_p1_1h.csv: Input data from January to March
flare_cls_p2_1h.csv: Input data from April to June
flare_cls_p3_1h.csv: Input data from July to September
flare_cls_p4_1h.csv: Input data from October to December

Features

Applies a rolling time window (24h window) to generate:
- max_goes_class: highest flare intensity/class in the 24-hour window
- cumulative_index: sum of sub-class values in the 24-hour window
- label_max: Binary label (max_goes_class ≥ M1.0)
- label_cum: Binary label (cumulative_index ≥ 10)
Splits dataset into four seasonal subsets (3-month each)
Supports optional filtering using an external index (e.g., image availability in SuryaBench)

Data Format

Each labeled record includes:

{
  "timestep": "2013-02-01T00:00:00",
  "max_goes_class": "M2.5",
  "cumulative_index": 38.5,
  "max_label": 1,
  "sum_label": 1
}

Dataset Details

Field	Description
Temporal Coverage	May 13, 2010 – Dec 31, 2024
Data Format	CSV (.csv), string-based schema
Data Size	Total 109,175 instances
Total File Size	~3.7MB