juliensimon/space-track-tle-history

Space-Track TLE History

Complete archive of Two-Line Element (TLE) orbital data for every tracked object in Earth orbit, from 1959 to 2026. Sourced from Space-Track.org bulk exports.

Quick Start

from datasets import load_dataset

# Load a specific year
ds = load_dataset("juliensimon/space-track-tle-history", data_files="data/tle_2024.parquet")

# Load everything (238M rows — use streaming for large-scale analysis)
ds = load_dataset("juliensimon/space-track-tle-history", streaming=True)

Examples

Track the ISS across 27 years of orbital data

import duckdb

conn = duckdb.connect()

# ISS = NORAD 25544, launched 1998
iss = conn.sql("""
    SELECT YEAR(epoch) as year, COUNT(*) as updates,
           ROUND(AVG(altitude_km), 1) as avg_alt_km,
           ROUND(MIN(altitude_km), 1) as min_alt_km
    FROM read_parquet('hf://datasets/juliensimon/space-track-tle-history/data/tle_*.parquet')
    WHERE norad_id = 25544
    GROUP BY year ORDER BY year
""").df()
print(iss)
#  year  updates  avg_alt  min_alt  max_alt
#  1998      190    374.6    183.8    404.7   ← initial orbit after Zarya launch
#  2006     2348    342.2    331.3    349.6   ← low point before reboost campaign
#  2011     2109    369.2    345.2    388.5   ← gradual raising begins
#  2018     1829    409.5    408.1    411.4   ← stabilized at ~410 km
#  2025     1617    421.8    416.5    424.9   ← current altitude

Watch Skylab's reentry spiral (1973–1979)

# Skylab 1 = NORAD 6633, decayed July 11, 1979
skylab = conn.sql("""
    SELECT epoch, altitude_km, mean_motion, eccentricity
    FROM read_parquet('hf://datasets/juliensimon/space-track-tle-history/data/tle_197*.parquet')
    WHERE norad_id = 6633
    ORDER BY epoch
""").df()
# 838 TLE records tracking Skylab's descent from 430 km to reentry:
#  1973: avg 431.6 km  (just launched)
#  1976: avg 419.1 km  (slowly decaying)
#  1978: avg 386.5 km  (drag accelerating)
#  1979: avg 327.6 km, min 227.8 km  (final months before July 11 reentry)

Measure the Cosmos-Iridium collision debris cloud

# On Feb 10, 2009, Cosmos 2251 and Iridium 33 collided at 790 km altitude.
# Use the SATCAT dataset to identify debris fragments, then track them in TLE history.
debris = conn.sql("""
    WITH debris_ids AS (
        SELECT norad_id,
            CASE WHEN intl_designator LIKE '93036%' THEN 'Cosmos 2251'
                 WHEN intl_designator LIKE '97051%' THEN 'Iridium 33' END as source
        FROM read_parquet('hf://datasets/juliensimon/space-track-satcat/data/satcat.parquet')
        WHERE (intl_designator LIKE '93036%' OR intl_designator LIKE '97051%')
          AND object_type = 'DEB'
    )
    SELECT d.source, YEAR(t.epoch) as year,
           COUNT(DISTINCT t.norad_id) as tracked_fragments,
           ROUND(AVG(t.altitude_km), 0) as avg_alt_km
    FROM read_parquet('hf://datasets/juliensimon/space-track-tle-history/data/tle_*.parquet') t
    JOIN debris_ids d ON t.norad_id = d.norad_id
    GROUP BY d.source, year ORDER BY d.source, year
""").df()
print(debris)
# Cosmos 2251: 2 fragments pre-collision → 1,143 in 2009 → 670 in 2025 (decaying)
# Iridium 33:  0 pre-collision → 489 in 2009 → 131 in 2025
# Fragments slowly reenter: avg altitude drifts from 750 km down to 675 km

Dataset Description

238 million TLE records spanning 68 years (1959–2026), covering 50,000+ tracked objects including active satellites, rocket bodies, and debris. One zstd-compressed Parquet file per year, 11 GB total. Raw orbital elements with no filtering or derived classifications.

This is the largest publicly available, ML-ready orbital element dataset. It covers the entire history of spaceflight from the earliest tracked objects to today's mega-constellations.

Schema

Column	Type	Description
norad_id	int32	NORAD catalog number (unique object ID)
epoch	timestamp[us, UTC]	TLE epoch — when the orbital elements were measured
inclination	float32	Orbital inclination (degrees)
raan	float32	Right ascension of ascending node (degrees)
eccentricity	float32	Orbital eccentricity
arg_perigee	float32	Argument of perigee (degrees)
mean_anomaly	float32	Mean anomaly (degrees)
mean_motion	float64	Mean motion (revolutions/day)
mean_motion_dot	float64	First derivative of mean motion (rev/day²) — drag indicator
bstar	float64	B* drag term — atmospheric drag coefficient
intl_designator	string	International designator (e.g., "98-067A" for ISS)
altitude_km	float32	Derived perigee altitude (km) from Kepler's third law

What's In Here

The dataset includes every object tracked by the US Space Surveillance Network:

• Active satellites: Starlink, OneWeb, GPS, Galileo, ISS, Hubble, and thousands more
• Rocket bodies: upper stages from launches dating back to the 1960s
• Debris: fragments from collisions (e.g., Cosmos-Iridium 2009), ASAT tests (e.g., Fengyun-1C 2007), and breakup events
• Decayed objects: historical records of objects that have since reentered the atmosphere

Use Cases

• Orbit prediction: train ML models to predict future orbital elements from historical trajectories
• Collision risk assessment: analyze conjunction events and close approaches
• Debris tracking: study the growth and evolution of orbital debris populations
• Atmospheric drag modeling: use B* and mean motion derivatives to study atmospheric density variations
• Constellation analysis: track the deployment and evolution of satellite constellations
• Reentry prediction: identify deorbiting objects from altitude decay patterns
• Space traffic management: analyze orbital shell congestion over time
• Astrodynamics research: benchmark orbit determination and propagation algorithms

Updates

This dataset is updated yearly when Space-Track publishes new bulk exports. The TLE archive is append-only — historical data does not change. See the pipeline repo for the build script.

Data Source

All orbital elements originate from the US Space Surveillance Network and are distributed by Space-Track.org, operated by the 18th Space Defense Squadron, United States Space Force.

The data was collected from Space-Track's yearly bulk TLE exports, which contain raw two-line element sets for all cataloged objects. Each TLE set represents a single orbital state observation at a specific epoch.

How to Propagate

TLE data is designed to be used with the SGP4/SDP4 propagator. To compute satellite positions:

from sgp4.api import Satrec, jday

# Reconstruct TLE lines from the dataset fields for SGP4 propagation
# Or use the orbital elements directly for analytical studies

Note: TLE accuracy degrades with time from epoch. For positions, propagate from the nearest epoch. For historical analysis, the orbital elements themselves (altitude, inclination, eccentricity) are directly usable without propagation.

File Sizes

Early years are small (few tracked objects); recent years are large (10,000+ active objects with frequent updates):

Decade	Typical Year Size	Objects
1960s	1–5 MB	~200
1980s	30–80 MB	~5,000
2000s	150–260 MB	~10,000
2020s	750–990 MB	~25,000+
2026	256 MB (partial, Jan–Mar)	~28,000

Related Datasets

• space-track-satcat — 68k objects with name, type, owner, launch/decay date. Join on norad_id.
• space-launch-log — 75k launches from 1942 to present
• starlink-fleet-data — Enriched Starlink data with status classification and ISL capability

Dataset pipelines: GitHub: juliensimon/space-datasets

License

CC-BY-4.0. Orbital element data is sourced from the US government (public domain) via Space-Track.org.

Citation

@dataset{space_track_tle_history,
  title={Space-Track TLE History: Complete Orbital Element Archive (1959–2026)},
  author={Julien Simon},
  year={2026},
  url={https://huggingface.co/datasets/juliensimon/space-track-tle-history},
  note={Orbital elements from Space-Track.org (18th Space Defense Squadron, USSF)}
}