phanerozoic/qiskit-calibration-drift

---
license: cc-by-4.0
task_categories:
- time-series-forecasting
- tabular-classification
tags:
- quantum-computing
- IBM-Quantum
- calibration
- hardware-characterization
- drift-analysis
- qiskit
- space-weather
- cosmic-rays
pretty_name: IBM Quantum Calibration Drift
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: backend
dtype: string
- name: qubit
dtype: int64
- name: property
dtype: string
- name: value
dtype: float64
- name: calibrated_time
dtype: string
- name: observed_time
dtype: string
- name: location
dtype: string
- name: latitude
dtype: float64
- name: longitude
dtype: float64
- name: solar_zenith_deg
dtype: float64
- name: temperature_c
dtype: float64
- name: pressure_hpa
dtype: float64
- name: humidity_pct
dtype: float64
- name: kp_index
dtype: float64
- name: solar_flux_sfu
dtype: float64
- name: dst_nt
dtype: float64
- name: bz_gsm_nt
dtype: float64
- name: neutron_flux
dtype: float64
splits:
- name: train
num_bytes: 421480585
num_examples: 1914304
download_size: 5303552
dataset_size: 421480585
---

# 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

```python
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](https://github.com/CharlesCNorton/qiskit-calibration-drift)

## Citation

```bibtex
@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

搜集汇总

数据集介绍

构建方式

在量子计算硬件性能监测领域，IBM Quantum校准漂移数据集通过自动化流程系统性地整合了多源时序数据。该数据集构建于GitHub Actions的定期轮询机制之上，每三十分钟同步采集IBM Quantum后端的校准参数、美国国家气象局的本地气象数据以及空间天气监测中心的空间天气指标。数据采集过程遵循去重策略，依据后端标识、量子比特索引、校准属性与校准时间戳的唯一组合来确保记录的新颖性，从而形成连续更新的校准与环境关联数据库。

特点

该数据集的核心特征在于其多维度的融合架构，不仅涵盖了超导量子比特的弛豫时间、退相干时间及门错误率等关键校准参数，还同步收录了数据中心的经纬度坐标、太阳天顶角、温湿度气压等局部环境变量，以及行星K指数、太阳射电流量与地磁扰动指数等空间天气指标。这种设计使得研究者能够深入探索量子硬件性能漂移与大气屏蔽效应、宇宙射线通量及地磁活动之间的潜在关联，为理解环境扰动对量子相干性的影响提供了丰富的数据基础。

使用方法

使用该数据集时，研究者可通过Hugging Face的datasets库直接加载训练分割，并利用过滤功能按后端、量子比特或校准属性进行数据子集提取。进行相关性分析时，建议以observed_time作为时序基准，避免校准时间戳可能存在的陈旧缓存误导。典型应用包括计算特定校准参数（如T1时间）与气压等环境变量的相关系数，或筛选高空间天气活动期（如Kp指数≥5）的数据以研究地磁暴对量子错误率的影响，从而支撑量子硬件可靠性与环境稳健性的实证研究。

背景与挑战

背景概述

量子计算硬件的性能稳定性是制约其实际应用的关键瓶颈之一，其中校准参数的时变漂移现象尤为突出。IBM Quantum Calibration Drift数据集由研究人员Charles C. Norton于2026年构建并发布，旨在系统探究环境因素与量子比特性能之间的关联机制。该数据集持续采集IBM Quantum平台上多个超导量子处理器（如ibm_torino、ibm_fez）的校准参数，并同步记录大气与空间天气数据，其核心研究问题在于揭示宇宙射线、地磁活动等外部扰动对量子比特相干时间、门保真度等关键指标的潜在影响。这一数据集为量子硬件表征与误差缓解研究提供了前所未有的多模态观测基础，推动了量子计算可靠性分析从孤立硬件测试向环境耦合建模的范式转变。

当前挑战

该数据集致力于解决量子硬件校准漂移的归因分析这一复杂问题，其核心挑战在于从高噪声、多尺度的观测数据中分离出环境信号与量子系统内在涨落的影响。具体而言，校准参数本身具有非均匀的时间戳特性，部分参数更新滞后数周，导致时间序列对齐与因果推断困难。在构建过程中，数据集成面临多源异构数据的实时同步难题，需协调量子硬件API、气象服务与空间天气监测网络的不同数据格式与更新频率。此外，环境变量与量子误差之间的耦合机制往往呈现非线性与滞后效应，且宇宙射线通量等关键驱动因子需通过中子监测站间接代理，这为建立精确的物理关联模型增添了不确定性。

常用场景

经典使用场景

在量子计算硬件性能监测领域，该数据集为研究量子比特校准参数的时变特性提供了关键数据支撑。其经典使用场景聚焦于时间序列分析与相关性建模，通过整合量子硬件校准数据与大气及空间天气观测变量，研究人员能够深入探索环境因素对超导量子比特相干时间、门保真度等核心性能指标的潜在影响。这种多模态数据融合为揭示量子系统在真实运行条件下的漂移规律奠定了实证基础。

衍生相关工作

基于该数据集的多变量时序特性，已衍生出若干探索环境-量子系统耦合机制的经典研究。相关工作重点发展了结合空间天气物理学与量子器件理论的跨学科分析框架，例如利用机器学习模型量化宇宙射线诱发准粒子中毒的概率关联。此外，数据集还催生了针对量子硬件鲁棒性优化的新型算法，包括基于环境感知的量子错误校正协议以及考虑地理分布的量子计算资源调度策略。

数据集最近研究

最新研究方向

在量子计算硬件表征领域，校准漂移现象是制约量子处理器性能稳定性的核心挑战之一。qiskit-calibration-drift数据集通过整合IBM量子硬件的持续校准数据与并发的环境监测指标，为探索外部环境因素与量子比特性能的关联性提供了前所未有的实证基础。当前研究前沿聚焦于利用该数据集的时间序列预测与表格分类能力，深入分析宇宙射线通量、地磁活动及大气压力等空间与局部天气变量对超导量子比特相干时间、门保真度等关键参数的动态影响。随着量子计算系统规模的扩大，理解并建模此类环境引起的性能漂移，对于开发自适应校准算法、提升硬件鲁棒性以及实现全天候可靠量子运算具有至关重要的科学意义与工程价值。