juliensimon/satellite-conjunctions

Satellite Conjunctions (SOCRATES Near-Misses)

Credit: NASA

Part of a dataset collection on Hugging Face.

Dataset description

A growing log of close approaches ("conjunctions") between catalogued objects in Earth orbit, from CelesTrak's SOCRATES screening service. Updated daily, accumulating distinct near-miss events over time.

SOCRATES (Satellite Orbital Conjunction Reports Assessing Threatening Encounters in Space) propagates the public two-line element (TLE) catalog forward roughly a week and reports every pair of objects predicted to pass within a few kilometers of each other. Each report gives the time of closest approach, the predicted miss distance, the relative speed, and an upper-bound collision probability. This is the public situational-awareness tool for tracking the increasingly crowded LEO environment, where active payloads, spent rocket bodies, and debris fragments routinely thread past one another.

To keep this dataset focused and bounded, it does not store the full daily screen (roughly 150,000 sub-5-km pairings). Instead it records only the genuinely close or risky events -- those with a predicted miss distance of 1 km or less, or a maximum collision probability of 1e-4 or greater -- and deduplicates the same physical encounter as its prediction is refined from day to day. The result is a historical record of serious near-misses suitable for studying conjunction rates, the objects and orbital regimes most often involved, and the growth of collision risk over time. Because predictions are derived from TLEs rather than high-precision owner/operator ephemerides, the figures are a screening aid, not operational collision-avoidance guidance.

This dataset is suitable for tabular classification tasks.

Schema

Column	Type	Description	Sample	Null %
event_key	string	Stable identifier for a conjunction event: the two NORAD IDs (sorted) joined with the time of closest approach rounded to the hour. Used to deduplicate the same physical encounter as its prediction is refined across daily runs.	2826-41668-2026062221	0.0%
norad_id_1	Int64	NORAD catalog number of the first object in the conjunction pair.	2826	0.0%
object_name_1	string	Name of the first object (payload, rocket body, or debris) as listed in the satellite catalog.	OPS 5712 (P/L 160)	0.0%
days_since_epoch_1	float64	Age in days of the orbital element set (TLE) used for the first object at the time of screening. Larger values mean a staler TLE and a less reliable prediction.	2.442	0.0%
norad_id_2	Int64	NORAD catalog number of the second object in the conjunction pair.	41668	0.0%
object_name_2	string	Name of the second object in the conjunction pair.	NOAA 16 DEB	0.0%
days_since_epoch_2	float64	Age in days of the orbital element set (TLE) used for the second object. Larger values mean a less reliable prediction.	2.792	0.0%
tca	datetime64[us]	Time of Closest Approach (UTC): the predicted moment the two objects are nearest.	2026-06-22 21:00:32.819000	0.0%
min_range_km	float64	Predicted miss distance at closest approach, in kilometers. The headline figure of a conjunction screen.	0.739	0.0%
relative_speed_kms	float64	Relative velocity of the two objects at closest approach, in km/s. High closing speeds (10-15 km/s are typical for crossing orbits) mean a tiny timing error translates to a large position error.	4.278	0.0%
max_probability	float64	Maximum collision probability for the encounter (dimensionless, 0-1), computed by scaling the assumed position uncertainty to its worst case. A screening upper bound, not an operational probability.	7.792e-06	0.0%
dilution_km	float64	Dilution distance (km): the combined position-uncertainty scale at which the maximum collision probability occurs. Reflects how uncertain the TLE-based positions are.	0.181	0.0%

Quick stats

• 26,850 recorded close-approach events (through 2026-07-02)
• 22,367 with a predicted miss distance under 1 km
• Closest approach on record: 0.005 km

Usage

from datasets import load_dataset

ds = load_dataset("juliensimon/satellite-conjunctions", split="train")
df = ds.to_pandas()

from datasets import load_dataset
import matplotlib.pyplot as plt

ds = load_dataset("juliensimon/satellite-conjunctions", split="train")
df = ds.to_pandas()

# Distribution of miss distances for recorded near-misses
fig, ax = plt.subplots(figsize=(9, 4))
ax.hist(df["min_range_km"], bins=40)
ax.set_xlabel("Miss distance at closest approach (km)")
ax.set_ylabel("Number of conjunctions")
ax.set_title("Satellite Conjunction Miss Distances")
plt.tight_layout()
plt.show()

# The tightest predicted passes
print(df.nsmallest(10, "min_range_km")[
    ["object_name_1", "object_name_2", "tca", "min_range_km", "max_probability"]])

Data source

https://celestrak.org/SOCRATES/

Update schedule

Daily at 16:40 UTC

Related datasets

• juliensimon/orbital-fragmentation-events

• juliensimon/reentry-events

• juliensimon/gcat-satellite-catalog

• juliensimon/starlink-fleet-data

• juliensimon/starlink-tle-latest

If you find this dataset useful, please consider giving it a like on Hugging Face. It helps others discover it.

About the author

Created by Julien Simon — AI Operating Partner at Fortino Capital. Part of the Space Datasets collection.

Citation

@dataset{satellite_conjunctions,
  title = {Satellite Conjunctions (SOCRATES Near-Misses)},
  author = {juliensimon},
  year = {2026},
  url = {https://huggingface.co/datasets/juliensimon/satellite-conjunctions},
  publisher = {Hugging Face}
}

License

celestrak-usage-policy