juliensimon/vlass-radio-sources

--- license: other license_name: vizier-scientific-use license_link: https://cds.unistra.fr/vizier-org/licences_vizier.html pretty_name: "VLASS Radio Sources (Epoch 1)" language: - en description: "Very Large Array Sky Survey (VLASS) Epoch 1 Quick Look component catalog with 3,381,277 radio source detections at 2-4 GHz (S-band)." task_categories: - tabular-classification - tabular-regression tags: - space - radio - vlass - vla - nrao - astronomy - open-data - tabular-data size_categories: - 1M<n<10M configs: - config_name: default data_files: - split: train path: data/vlass_radio_sources.parquet default: true --- # VLASS Radio Sources (Epoch 1) *Part of the [Astronomy Datasets](https://huggingface.co/collections/juliensimon/astronomy-datasets-69c24caf2f17e36128946743) collection on Hugging Face.* The Very Large Array Sky Survey (VLASS) Epoch 1 Quick Look component catalog from CIRADA, containing **3,381,277** radio source detections at S-band (2-4 GHz) with ~2.5 arcsecond resolution, covering the sky north of declination -40 degrees. VLASS is the modern successor to NVSS and FIRST, offering higher resolution and multi-epoch coverage. ## Dataset description VLASS is a synoptic all-sky radio survey using the Karl G. Jansky Very Large Array (VLA) in its B-configuration at S-band (2-4 GHz). The survey covers the entire sky visible to the VLA (declination > -40 deg, ~33,885 sq. deg.) in three epochs. This catalog contains Quick Look component detections from Epoch 1, processed by the Canadian Initiative for Radio Astronomy Data Analysis (CIRADA). Each row is a Gaussian component fitted to a radio detection using PyBDSF. Of the 3,381,277 total detections, **1,880,195** are in the curated main sample (duplicate-free, quality-filtered) and **1,826,680** are resolved sources. ## Key columns | Column | Type | Description | |--------|------|-------------| | `component_name` | string | IAU component name (VLASS1QLCIR JHHMMSS.ss+DDMMSS.s) | | `ra_deg` | float64 | Right ascension J2000 (degrees) | | `dec_deg` | float64 | Declination J2000 (degrees) | | `total_flux_mjy` | float64 | Integrated flux density at S-band (mJy) | | `peak_flux_mjy_beam` | float64 | Peak brightness at S-band (mJy/beam) | | `major_axis_arcsec` | float64 | Fitted major axis FWHM (arcsec) | | `minor_axis_arcsec` | float64 | Fitted minor axis FWHM (arcsec) | | `position_angle_deg` | float64 | Fitted position angle (degrees) | | `deconv_major_arcsec` | float64 | Deconvolved major axis (arcsec) | | `deconv_minor_arcsec` | float64 | Deconvolved minor axis (arcsec) | | `island_rms_mjy_beam` | float64 | Local rms noise (mJy/beam) | | `source_code` | string | Component type: S(ingle), C(omplex), M(ultiple), E(xtended) | | `nvss_distance_arcsec` | float64 | Angular separation from nearest NVSS source (arcsec) | | `first_distance_arcsec` | float64 | Angular separation from nearest FIRST source (arcsec) | | `duplicate_flag` | Int32 | Duplicate detection flag (0=unique) | | `quality_flag` | Int32 | Quality flag (0=good) | | `main_sample` | Int32 | Main sample membership (1=curated subset) | | `is_resolved` | bool | True if deconvolved major axis > 0 | Full schema includes 70 columns with uncertainties, beam properties, and image-plane measurements. ## Quick stats - **3,381,277** total component detections - **1,880,195** main sample sources (quality-filtered, duplicate-free) - **1,826,680** resolved sources (54.0%) - Median peak flux: 1238.00 mJy/beam - Declination range: -40.0 to 89.8 degrees - Frequency: S-band (2-4 GHz), ~2.5 arcsec resolution ## Usage ```python from datasets import load_dataset ds = load_dataset("juliensimon/vlass-radio-sources", split="train") df = ds.to_pandas() # Main sample only (quality-filtered) main = df[df["main_sample"] == 1] print(f"Main sample: {len(main):,} sources") # Flux distribution import matplotlib.pyplot as plt df["peak_flux_mjy_beam"].clip(upper=100).hist(bins=200, log=True) plt.xlabel("Peak flux (mJy/beam)") plt.ylabel("Count") plt.title("VLASS Source Flux Distribution") plt.show() # Sky density map plt.hexbin(df["ra_deg"], df["dec_deg"], gridsize=100, mincnt=1) plt.colorbar(label="Source count") plt.xlabel("RA (deg)") plt.ylabel("Dec (deg)") plt.title("VLASS Epoch 1 Sky Density") plt.show() # Cross-match proximity to NVSS/FIRST has_nvss = df["nvss_distance_arcsec"] < 10 print(f"Within 10 arcsec of NVSS source: {has_nvss.sum():,}") ``` ## Data source Gordon, Y.A., et al. (2021), *A Catalog of Very Large Array Sky Survey (VLASS) Epoch 1 Quick Look Components, Version 2.* Astrophysical Journal Supplement Series, 255, 30. Processed by CIRADA. Via VizieR CDS (J/ApJS/255/30). ## Related datasets - [NVSS Radio Source Catalog](https://huggingface.co/datasets/juliensimon/nvss-radio-catalog) — predecessor 1.4 GHz survey, 1.8M sources - [FIRST Radio Survey Catalog](https://huggingface.co/datasets/juliensimon/first-radio-catalog) — predecessor high-resolution 1.4 GHz survey ## Pipeline Source code: [juliensimon/space-datasets](https://github.com/juliensimon/space-datasets) ## Citation ```bibtex @dataset{vlass_radio_sources, author = {Simon, Julien}, title = {VLASS Radio Sources (Epoch 1)}, year = {2026}, publisher = {Hugging Face}, url = {https://huggingface.co/datasets/juliensimon/vlass-radio-sources}, note = {Based on Gordon et al. (2021) via VizieR CDS} } ``` ## License [VizieR Scientific-Use Terms](https://cds.unistra.fr/vizier-org/licences_vizier.html)Very Large Array Sky Survey (VLASS) Epoch 1 Quick Look component catalog with 3,381,277 radio source detections at 2-4 GHz (S-band).

搜集汇总

数据集介绍

构建方式

该数据集源自卡尔·央斯基甚大阵（VLA）开展的VLASS巡天项目，其构建依托于S波段（2–4 GHz）B构型观测所得Epoch 1 Quick Look图像。加拿大射电天文数据分析中心（CIRADA）采用PyBDSF软件对图像中的射电源进行高斯组分拟合，逐一提取流量、形态及位置参数，形成包含3,381,277个探测组分的原始目录。在此基础上，通过去重与质量筛选获得1,880,195个主样本源，其中1,826,680个为展源。整个流程将干涉测量成像与自动化源提取相结合，最终生成结构化的表格数据，并通过VizieR CDS发布。

特点

该数据集在规模与信息丰度上均具显著优势，涵盖赤纬−40度以北约33,885平方度的天区，总探测数超过三百万，主样本亦达一百八十余万。每一条记录包含70个字段，不仅提供积分流量、峰值亮度、拟合长短轴与位置角等基本参数，还给出退卷积尺寸、局部噪声均方根、源类型编码以及和NVSS、FIRST星表的交叉证认距离，并附有重复标志与质量标志。数据以单一Parquet文件按训练分划组织，体积紧凑而读取高效，适用于大规模射电天文学统计分析与机器学习建模。

使用方法

研究者可借助Hugging Face datasets库以一行代码加载该数据集并转换为Pandas数据框，进而依据main_sample字段筛选经过质量控制和去重的主样本，或利用quality_flag与duplicate_flag实施更为严格的自定义过滤。典型分析包括绘制峰值流量分布直方图、以六边形分箱呈现天区源密度、统计与NVSS或FIRST源角距离小于10角秒的交叉证认比例，以及基于is_resolved字段区分展源与点源。该数据亦可直接用于表格分类或回归任务，例如预测源的类型编码或流量属性，为射电巡天数据的挖掘提供便利入口。

背景与挑战

背景概述

射电天文巡天是揭示宇宙中星系、类星体及瞬变现象的核心手段。甚大阵天空巡天（VLASS）作为NVSS与FIRST的现代继任者，由美国国家射电天文台依托卡尔·G·詹斯基甚大阵于2017年启动，旨在以S波段（2-4 GHz）和约2.5角秒分辨率对北天区（赤纬>-40度）进行三轮观测。本数据集由加拿大射电天文数据分析倡议处理，基于Gordon等人2021年发布的Epoch 1快速视场分量星表，涵盖3,381,277个射电分量，为多波段交叉证认与射电形态分类研究提供了大规模基准。

当前挑战

该数据集所应对的领域问题在于从海量射电干涉图像中稳健提取源分量并实现自动分类与参数回归，这对传统方法在低信噪比、密集源混淆及复杂延展结构情形下的可靠性构成显著挑战。构建过程中，高斯分量拟合需克服旁瓣与噪声引起的虚假检测，同时须通过重复源标记和质量筛选保证主样本纯净性；跨巡天匹配NVSS与FIRST时还面临位置偏差和分辨率不匹配的困难。此外，将多波段观测统一至相同物理尺度以推断谱指数与源本征属性，仍需进一步解决系统误差与样本选择效应的难题。

常用场景

经典使用场景

在射电天文学的大尺度巡天研究中，该数据集最为经典的应用场景在于构建射电源的统计样本与天空密度分布图。研究者借助其涵盖北天赤纬-40度以上、约33,885平方度的S波段（2-4吉赫兹）探测记录，能够系统刻画射电源的通量分布、角尺寸特征及空间聚集模式。基于约338万个高斯拟合分量，学者可筛选出主样本以开展射电源计数与宇宙大尺度结构的交叉分析，进而检验射电星系演化模型。

解决学术问题

该数据集有效缓解了长期以来射电巡天在分辨率与灵敏度之间难以兼顾的困境。相较于NVSS与FIRST等前代巡天，VLASS以约2.5角秒的分辨率提供了更精细的源形态信息，并通过多历元观测设计支持时域射电天文学研究。其包含的去卷积轴长、位置角及局域噪声等参数，为区分致密源与延展源、识别复杂射电形态提供了关键依据，推动了射电源物理分类与虚假探测甄别等基础问题的解决。

衍生相关工作

围绕该数据集，已衍生出多项经典工作，包括CIRADA团队发布的VLASS Quick Look分量星表版本2，以及基于其开展的射电源计数与宇宙演化模型约束研究。该数据集亦催生了与NVSS、FIRST等星表的系统性交叉匹配工作，为构建统一射电源样本提供了范式。后续研究进一步将其与光学、红外巡天数据结合，用于活动星系核甄别与射电星系宿主星系性质探讨，形成了多波段协同研究的经典链条。