phanerozoic/qiskit-calibration-drift

IBM Quantum Calibration Drift Dataset

Continuously-updated calibration data from IBM Quantum hardware with concurrent environmental measurements. Enables correlation analysis between qubit performance and atmospheric/space weather conditions.

Overview

Property	Value
Update frequency	Every 30 minutes
Backends	ibm_fez (156 qubits), ibm_torino (133 qubits), ibm_marrakesh (156 qubits)
Total qubits	445
Collection method	Automated polling via GitHub Actions
Calibration source	IBM Quantum Runtime API
Weather source	NWS API (NOAA)
Space weather source	SWPC (NOAA)

Schema

Field	Type	Description
backend	string	Backend identifier
qubit	int	Qubit index (0 to N-1), or -1 for two-qubit gate data
property	string	Calibration property name
value	float	Measured value
calibrated_time	string	IBM calibration timestamp (UTC)
observed_time	string	Collection timestamp (UTC ISO)
location	string	Data center location identifier
latitude	float	Data center latitude
longitude	float	Data center longitude
solar_zenith_deg	float	Solar zenith angle (>90° = night)
temperature_c	float	Local temperature (°C)
pressure_hpa	float	Barometric pressure (hPa)
humidity_pct	float	Relative humidity (%)
kp_index	float	Planetary K-index (0-9, geomagnetic activity)
solar_flux_sfu	float	10.7cm solar radio flux (SFU)
dst_nt	float	Dst index (nT, ring current strength)
bz_gsm_nt	float	IMF Bz component (nT, negative = geo-coupling)
neutron_flux	float	Cosmic ray proxy (Newark, DE monitor)

Important: Timestamp Interpretation

This dataset contains two timestamp fields with very different meanings:

• calibrated_time — When IBM last calibrated that specific property. These timestamps can span months because IBM does not recalibrate all properties simultaneously. Some properties (like T1/T2) may retain calibration timestamps from weeks or months ago.

• observed_time — When the poller actually collected the record. This reflects the true data collection period.

For time-series analysis, use observed_time. The calibrated_time span reflects IBM's stale cache, not the collection period.

Calibration Properties

Per-qubit:

• T1 — Relaxation time (seconds)
• T2 — Dephasing time (seconds)
• readout_error — Measurement error probability
• prob_meas0_prep1 — P(measure 0 | prepared 1)
• prob_meas1_prep0 — P(measure 1 | prepared 0)
• sx_error — SX gate error (native single-qubit gate)

Per-edge:

• cz_error_{i}_{j} — Two-qubit CZ gate error for edge (i, j)

Environmental Fields

Solar position:

• solar_zenith_deg — Sun angle from vertical (0°=overhead, 90°=horizon, >90°=night)

Weather (local to data center):

• temperature_c — Ambient temperature
• pressure_hpa — Barometric pressure (correlates with cosmic ray flux attenuation)
• humidity_pct — Relative humidity

Space weather (global):

• kp_index — Geomagnetic storm indicator (0=quiet, 9=severe storm)
• solar_flux_sfu — Solar activity proxy (higher = more active sun)
• dst_nt — Ring current strength (< -50 nT = storm, < -100 nT = severe)
• bz_gsm_nt — Interplanetary magnetic field z-component (negative = geomagnetic coupling)
• neutron_flux — Cosmic ray flux from Newark, DE monitor (pressure-corrected)

Data Center Locations

Location ID	Coordinates	Backends
yorktown_heights_ny	41.27°N, 73.78°W	ibm_torino, ibm_fez, ibm_marrakesh

Usage

from datasets import load_dataset

ds = load_dataset("phanerozoic/qiskit-calibration-drift", split="train")

# Filter by backend
torino = ds.filter(lambda x: x["backend"] == "ibm_torino")

# Correlation analysis: T1 vs pressure
t1_data = ds.filter(lambda x: x["property"] == "T1")
df = t1_data.to_pandas()
correlation = df["value"].corr(df["pressure_hpa"])

# Filter by space weather conditions
storm_data = ds.filter(lambda x: x["kp_index"] >= 5)

Research Applications

• Cosmic ray correlation: Barometric pressure modulates atmospheric shielding against cosmic rays, which cause quasiparticle poisoning in superconducting qubits.
• Geomagnetic storm effects: Kp index tracks magnetospheric disturbances that may correlate with qubit coherence.
• Solar cycle tracking: Dataset spans the declining phase of Solar Cycle 25 (peaked October 2024).
• Seasonal/diurnal patterns: Long-term collection enables detection of periodic environmental effects.

Note: Environmental correlation studies require weeks to months of continuous observed_time coverage. Check the current observation window before attempting such analyses.

Collection Method

Data is collected via GitHub Actions every 30 minutes:

1. Fetch space weather (NOAA SWPC)
2. Fetch local weather (NWS API)
3. Query IBM Quantum calibration data
4. Deduplicate by (backend, qubit, property, calibrated_time)
5. Append new records to dataset

Source: github.com/CharlesCNorton/qiskit-calibration-drift

Citation

@dataset{qiskit-calibration-drift,
  title={IBM Quantum Calibration Drift Dataset},
  author={Norton, Charles C.},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/phanerozoic/qiskit-calibration-drift}
}

Acknowledgments

We acknowledge the NMDB database (www.nmdb.eu), founded under the European Union's FP7 programme (contract no. 213007) for providing neutron monitor data.

License

CC-BY-4.0