wangdx25/goes_omni_electron_flux_forecasting

GOES–OMNI >2 MeV Electron Flux Forecasting

This dataset combines cross-calibrated NOAA GOES-14/GOES-16 >2 MeV electron flux with NASA/GSFC OMNI solar-wind and geomagnetic drivers on a uniform five-minute UTC grid.

It provides two configurations:

• ml-ready (default): scaled causal features, validity flags, unscaled 30-minute/6-hour/12-hour targets, and leakage-safe chronological splits.
• scientific-master: unscaled source measurements, instrument context, calibration factors, and explicit missingness.

Load the data

from datasets import load_dataset

ml = load_dataset(
    "snowsadh/goes-omni-electron-flux-forecasting",
    "ml-ready",
)
master = load_dataset(
    "snowsadh/goes-omni-electron-flux-forecasting",
    "scientific-master",
    split="full",
)

Size and coverage

Configuration / split	Rows	Coverage
Scientific master	1,326,459	2013-05-23 to 2025-12-31
ML train	978,265	2014-01-01 to 2023-06-30 11:55 UTC
ML validation	102,186	2023-07-01 to 2024-06-30 11:55 UTC
ML test	80,612	2024-07-01 to 2025-04-07 07:50 UTC

GOES flux ends on 2025-04-07; later scientific-master rows contain OMNI data only. Every ML target remains inside the same temporal split as its features.

Processing

GOES-14 one-minute observations are normalized to exact minutes and aggregated in log space. GOES-14 and GOES-16 are cross-calibrated in one-hour magnetic local-time bins over their 2017 overlap. The corrected series is combined with five-minute means of OMNI drivers. ML features use only causal information; missing inputs are carried forward for at most one hour and then replaced with a training-only median. RobustScaler is fitted on the training split only.

Full definitions are in DATA_DICTIONARY.md; preprocessing parameters are in preprocessing_meta.json; source-level provenance is in SOURCES.md.

Intended uses

• Forecasting GEO >2 MeV electron flux 30 minutes, 6 hours, or 12 hours ahead.
• Space-weather time-series analysis and storm case studies.
• Multi-instrument calibration and leakage-safe ML demonstrations.

Limitations

• The data form a GOES-14/GOES-16 time series, not a global radiation-belt map.
• Satellite longitude and detector geometry change across the 2017 splice.
• MLT-bin calibration cannot recover values below the GOES-14 EPEAD noise floor.
• Values below flux 10 are retained and explicitly marked.
• Imputed ML features must not be interpreted as physical observations.
• OMNI combines several upstream spacecraft and maps measurements toward Earth's bow-shock nose.
• No ISRO GRASP data are included, and the series is not validated at Indian GEO longitude.

License

The compiled data and documentation are licensed under CC BY 4.0 to the extent the creator holds applicable rights. Processing scripts are MIT-licensed. Upstream public-domain/CC0/source-specific terms remain in force; see LICENSE-DATA.md and SOURCES.md.

TsFile Conversion

• Original dataset: snowsadh/goes-omni-electron-flux-forecasting

• Modalities: Time-series

• Converted data files are listed in the YAML metadata above.

• Source README text and dataset-specific metadata are retained; the source Usage section is replaced with the executable TsFile Python SDK example below.

• Both source configurations are retained: ml-ready train/validation/test and scientific-master full.

• Source time is encoded as millisecond Time; satellite/instrument metadata and configuration labels are TAG columns where present.

• Numeric scientific and ML fields, including validity flags and forecasting targets, are retained.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("goes_omni_ml_ready_test_1.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())